xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
/*
|
|
|
|
* Machine specific setup for xen
|
|
|
|
*
|
|
|
|
* Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/sched.h>
|
|
|
|
#include <linux/mm.h>
|
|
|
|
#include <linux/pm.h>
|
2010-08-25 14:39:17 -06:00
|
|
|
#include <linux/memblock.h>
|
2011-04-01 16:28:35 -06:00
|
|
|
#include <linux/cpuidle.h>
|
2012-03-13 18:06:57 -06:00
|
|
|
#include <linux/cpufreq.h>
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
|
|
|
|
#include <asm/elf.h>
|
2008-01-30 05:30:42 -07:00
|
|
|
#include <asm/vdso.h>
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
#include <asm/e820.h>
|
|
|
|
#include <asm/setup.h>
|
2008-06-16 15:54:49 -06:00
|
|
|
#include <asm/acpi.h>
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
#include <asm/xen/hypervisor.h>
|
|
|
|
#include <asm/xen/hypercall.h>
|
|
|
|
|
2010-10-25 17:32:29 -06:00
|
|
|
#include <xen/xen.h>
|
2008-05-26 16:31:19 -06:00
|
|
|
#include <xen/page.h>
|
2008-03-17 17:37:17 -06:00
|
|
|
#include <xen/interface/callback.h>
|
2009-02-06 20:09:48 -07:00
|
|
|
#include <xen/interface/memory.h>
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
#include <xen/interface/physdev.h>
|
|
|
|
#include <xen/features.h>
|
|
|
|
#include "xen-ops.h"
|
2007-07-20 01:31:43 -06:00
|
|
|
#include "vdso.h"
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
|
|
|
|
/* These are code, but not functions. Defined in entry.S */
|
|
|
|
extern const char xen_hypervisor_callback[];
|
|
|
|
extern const char xen_failsafe_callback[];
|
2008-12-16 12:56:06 -07:00
|
|
|
extern void xen_sysenter_target(void);
|
|
|
|
extern void xen_syscall_target(void);
|
|
|
|
extern void xen_syscall32_target(void);
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
|
2010-08-30 17:41:02 -06:00
|
|
|
/* Amount of extra memory space we add to the e820 ranges */
|
2011-09-28 10:46:34 -06:00
|
|
|
struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
|
2010-08-30 17:41:02 -06:00
|
|
|
|
2011-09-28 10:46:32 -06:00
|
|
|
/* Number of pages released from the initial allocation. */
|
|
|
|
unsigned long xen_released_pages;
|
|
|
|
|
2010-09-14 11:19:14 -06:00
|
|
|
/*
|
|
|
|
* The maximum amount of extra memory compared to the base size. The
|
|
|
|
* main scaling factor is the size of struct page. At extreme ratios
|
|
|
|
* of base:extra, all the base memory can be filled with page
|
|
|
|
* structures for the extra memory, leaving no space for anything
|
|
|
|
* else.
|
|
|
|
*
|
|
|
|
* 10x seems like a reasonable balance between scaling flexibility and
|
|
|
|
* leaving a practically usable system.
|
|
|
|
*/
|
|
|
|
#define EXTRA_MEM_RATIO (10)
|
|
|
|
|
2011-09-29 05:26:19 -06:00
|
|
|
static void __init xen_add_extra_mem(u64 start, u64 size)
|
2010-08-30 17:41:02 -06:00
|
|
|
{
|
2011-01-18 18:09:41 -07:00
|
|
|
unsigned long pfn;
|
2011-09-29 05:26:19 -06:00
|
|
|
int i;
|
2011-01-18 18:09:41 -07:00
|
|
|
|
2011-09-29 05:26:19 -06:00
|
|
|
for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
|
|
|
|
/* Add new region. */
|
|
|
|
if (xen_extra_mem[i].size == 0) {
|
|
|
|
xen_extra_mem[i].start = start;
|
|
|
|
xen_extra_mem[i].size = size;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
/* Append to existing region. */
|
|
|
|
if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
|
|
|
|
xen_extra_mem[i].size += size;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (i == XEN_EXTRA_MEM_MAX_REGIONS)
|
|
|
|
printk(KERN_WARNING "Warning: not enough extra memory regions\n");
|
2010-08-30 17:41:02 -06:00
|
|
|
|
2011-11-28 10:46:22 -07:00
|
|
|
memblock_reserve(start, size);
|
2010-09-15 14:32:49 -06:00
|
|
|
|
2011-09-29 05:26:19 -06:00
|
|
|
xen_max_p2m_pfn = PFN_DOWN(start + size);
|
2012-08-17 14:43:28 -06:00
|
|
|
for (pfn = PFN_DOWN(start); pfn < xen_max_p2m_pfn; pfn++) {
|
|
|
|
unsigned long mfn = pfn_to_mfn(pfn);
|
|
|
|
|
|
|
|
if (WARN(mfn == pfn, "Trying to over-write 1-1 mapping (pfn: %lx)\n", pfn))
|
|
|
|
continue;
|
|
|
|
WARN(mfn != INVALID_P2M_ENTRY, "Trying to remove %lx which has %lx mfn!\n",
|
|
|
|
pfn, mfn);
|
2011-01-18 18:09:41 -07:00
|
|
|
|
|
|
|
__set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
|
2012-08-17 14:43:28 -06:00
|
|
|
}
|
2010-08-30 17:41:02 -06:00
|
|
|
}
|
|
|
|
|
2012-04-06 14:10:20 -06:00
|
|
|
static unsigned long __init xen_do_chunk(unsigned long start,
|
|
|
|
unsigned long end, bool release)
|
2009-09-16 01:56:17 -06:00
|
|
|
{
|
|
|
|
struct xen_memory_reservation reservation = {
|
|
|
|
.address_bits = 0,
|
|
|
|
.extent_order = 0,
|
|
|
|
.domid = DOMID_SELF
|
|
|
|
};
|
2009-09-16 13:38:33 -06:00
|
|
|
unsigned long len = 0;
|
2009-09-16 01:56:17 -06:00
|
|
|
unsigned long pfn;
|
|
|
|
int ret;
|
|
|
|
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
for (pfn = start; pfn < end; pfn++) {
|
|
|
|
unsigned long frame;
|
2012-04-06 14:10:20 -06:00
|
|
|
unsigned long mfn = pfn_to_mfn(pfn);
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
|
2012-04-06 14:10:20 -06:00
|
|
|
if (release) {
|
|
|
|
/* Make sure pfn exists to start with */
|
|
|
|
if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
|
|
|
|
continue;
|
|
|
|
frame = mfn;
|
|
|
|
} else {
|
|
|
|
if (mfn != INVALID_P2M_ENTRY)
|
|
|
|
continue;
|
|
|
|
frame = pfn;
|
|
|
|
}
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
set_xen_guest_handle(reservation.extent_start, &frame);
|
|
|
|
reservation.nr_extents = 1;
|
|
|
|
|
2012-04-06 14:10:20 -06:00
|
|
|
ret = HYPERVISOR_memory_op(release ? XENMEM_decrease_reservation : XENMEM_populate_physmap,
|
|
|
|
&reservation);
|
|
|
|
WARN(ret != 1, "Failed to %s pfn %lx err=%d\n",
|
|
|
|
release ? "release" : "populate", pfn, ret);
|
|
|
|
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
if (ret == 1) {
|
2012-04-06 14:10:20 -06:00
|
|
|
if (!early_set_phys_to_machine(pfn, release ? INVALID_P2M_ENTRY : frame)) {
|
|
|
|
if (release)
|
|
|
|
break;
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
set_xen_guest_handle(reservation.extent_start, &frame);
|
|
|
|
reservation.nr_extents = 1;
|
|
|
|
ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
|
2012-04-06 14:10:20 -06:00
|
|
|
&reservation);
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
len++;
|
|
|
|
} else
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (len)
|
2012-04-06 14:10:20 -06:00
|
|
|
printk(KERN_INFO "%s %lx-%lx pfn range: %lu pages %s\n",
|
|
|
|
release ? "Freeing" : "Populating",
|
|
|
|
start, end, len,
|
|
|
|
release ? "freed" : "added");
|
|
|
|
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
return len;
|
|
|
|
}
|
2012-05-03 09:15:42 -06:00
|
|
|
|
|
|
|
static unsigned long __init xen_release_chunk(unsigned long start,
|
|
|
|
unsigned long end)
|
|
|
|
{
|
|
|
|
return xen_do_chunk(start, end, true);
|
|
|
|
}
|
|
|
|
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
static unsigned long __init xen_populate_chunk(
|
|
|
|
const struct e820entry *list, size_t map_size,
|
|
|
|
unsigned long max_pfn, unsigned long *last_pfn,
|
|
|
|
unsigned long credits_left)
|
|
|
|
{
|
|
|
|
const struct e820entry *entry;
|
|
|
|
unsigned int i;
|
|
|
|
unsigned long done = 0;
|
|
|
|
unsigned long dest_pfn;
|
|
|
|
|
|
|
|
for (i = 0, entry = list; i < map_size; i++, entry++) {
|
|
|
|
unsigned long s_pfn;
|
|
|
|
unsigned long e_pfn;
|
|
|
|
unsigned long pfns;
|
|
|
|
long capacity;
|
|
|
|
|
2012-07-17 23:06:39 -06:00
|
|
|
if (credits_left <= 0)
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
break;
|
|
|
|
|
|
|
|
if (entry->type != E820_RAM)
|
|
|
|
continue;
|
|
|
|
|
2012-07-17 23:06:39 -06:00
|
|
|
e_pfn = PFN_DOWN(entry->addr + entry->size);
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
|
|
|
|
/* We only care about E820 after the xen_start_info->nr_pages */
|
|
|
|
if (e_pfn <= max_pfn)
|
|
|
|
continue;
|
|
|
|
|
2012-07-17 23:06:39 -06:00
|
|
|
s_pfn = PFN_UP(entry->addr);
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
/* If the E820 falls within the nr_pages, we want to start
|
|
|
|
* at the nr_pages PFN.
|
|
|
|
* If that would mean going past the E820 entry, skip it
|
|
|
|
*/
|
|
|
|
if (s_pfn <= max_pfn) {
|
|
|
|
capacity = e_pfn - max_pfn;
|
|
|
|
dest_pfn = max_pfn;
|
|
|
|
} else {
|
|
|
|
capacity = e_pfn - s_pfn;
|
|
|
|
dest_pfn = s_pfn;
|
|
|
|
}
|
|
|
|
|
2012-07-17 23:06:39 -06:00
|
|
|
if (credits_left < capacity)
|
|
|
|
capacity = credits_left;
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
|
2012-07-17 23:06:39 -06:00
|
|
|
pfns = xen_do_chunk(dest_pfn, dest_pfn + capacity, false);
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
done += pfns;
|
|
|
|
*last_pfn = (dest_pfn + pfns);
|
2012-07-17 23:06:39 -06:00
|
|
|
if (pfns < capacity)
|
|
|
|
break;
|
|
|
|
credits_left -= pfns;
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
}
|
|
|
|
return done;
|
|
|
|
}
|
2012-05-03 09:15:42 -06:00
|
|
|
|
|
|
|
static void __init xen_set_identity_and_release_chunk(
|
|
|
|
unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages,
|
|
|
|
unsigned long *released, unsigned long *identity)
|
|
|
|
{
|
|
|
|
unsigned long pfn;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If the PFNs are currently mapped, the VA mapping also needs
|
|
|
|
* to be updated to be 1:1.
|
|
|
|
*/
|
|
|
|
for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++)
|
|
|
|
(void)HYPERVISOR_update_va_mapping(
|
|
|
|
(unsigned long)__va(pfn << PAGE_SHIFT),
|
|
|
|
mfn_pte(pfn, PAGE_KERNEL_IO), 0);
|
|
|
|
|
|
|
|
if (start_pfn < nr_pages)
|
|
|
|
*released += xen_release_chunk(
|
|
|
|
start_pfn, min(end_pfn, nr_pages));
|
|
|
|
|
|
|
|
*identity += set_phys_range_identity(start_pfn, end_pfn);
|
|
|
|
}
|
|
|
|
|
2011-09-28 10:46:36 -06:00
|
|
|
static unsigned long __init xen_set_identity_and_release(
|
|
|
|
const struct e820entry *list, size_t map_size, unsigned long nr_pages)
|
2009-09-16 01:56:17 -06:00
|
|
|
{
|
2011-09-28 10:46:36 -06:00
|
|
|
phys_addr_t start = 0;
|
2009-09-16 01:56:17 -06:00
|
|
|
unsigned long released = 0;
|
2011-02-01 15:15:30 -07:00
|
|
|
unsigned long identity = 0;
|
2011-09-28 10:46:36 -06:00
|
|
|
const struct e820entry *entry;
|
2011-02-01 15:15:30 -07:00
|
|
|
int i;
|
|
|
|
|
2011-09-28 10:46:36 -06:00
|
|
|
/*
|
|
|
|
* Combine non-RAM regions and gaps until a RAM region (or the
|
|
|
|
* end of the map) is reached, then set the 1:1 map and
|
|
|
|
* release the pages (if available) in those non-RAM regions.
|
|
|
|
*
|
|
|
|
* The combined non-RAM regions are rounded to a whole number
|
|
|
|
* of pages so any partial pages are accessible via the 1:1
|
|
|
|
* mapping. This is needed for some BIOSes that put (for
|
|
|
|
* example) the DMI tables in a reserved region that begins on
|
|
|
|
* a non-page boundary.
|
|
|
|
*/
|
2011-02-01 15:15:30 -07:00
|
|
|
for (i = 0, entry = list; i < map_size; i++, entry++) {
|
2011-09-28 10:46:36 -06:00
|
|
|
phys_addr_t end = entry->addr + entry->size;
|
|
|
|
if (entry->type == E820_RAM || i == map_size - 1) {
|
|
|
|
unsigned long start_pfn = PFN_DOWN(start);
|
|
|
|
unsigned long end_pfn = PFN_UP(end);
|
2011-02-01 15:15:30 -07:00
|
|
|
|
2011-09-28 10:46:36 -06:00
|
|
|
if (entry->type == E820_RAM)
|
|
|
|
end_pfn = PFN_UP(entry->addr);
|
2011-02-01 15:15:30 -07:00
|
|
|
|
2012-05-03 09:15:42 -06:00
|
|
|
if (start_pfn < end_pfn)
|
|
|
|
xen_set_identity_and_release_chunk(
|
|
|
|
start_pfn, end_pfn, nr_pages,
|
|
|
|
&released, &identity);
|
2011-02-01 15:15:30 -07:00
|
|
|
|
2011-09-28 10:46:36 -06:00
|
|
|
start = end;
|
2011-02-01 15:15:30 -07:00
|
|
|
}
|
|
|
|
}
|
2011-09-28 10:46:36 -06:00
|
|
|
|
2012-03-30 13:37:07 -06:00
|
|
|
if (released)
|
|
|
|
printk(KERN_INFO "Released %lu pages of unused memory\n", released);
|
|
|
|
if (identity)
|
|
|
|
printk(KERN_INFO "Set %ld page(s) to 1-1 mapping\n", identity);
|
2011-09-28 10:46:36 -06:00
|
|
|
|
|
|
|
return released;
|
2011-02-01 15:15:30 -07:00
|
|
|
}
|
2011-08-19 08:57:16 -06:00
|
|
|
|
|
|
|
static unsigned long __init xen_get_max_pages(void)
|
|
|
|
{
|
|
|
|
unsigned long max_pages = MAX_DOMAIN_PAGES;
|
|
|
|
domid_t domid = DOMID_SELF;
|
|
|
|
int ret;
|
|
|
|
|
2011-12-14 05:16:08 -07:00
|
|
|
/*
|
|
|
|
* For the initial domain we use the maximum reservation as
|
|
|
|
* the maximum page.
|
|
|
|
*
|
|
|
|
* For guest domains the current maximum reservation reflects
|
|
|
|
* the current maximum rather than the static maximum. In this
|
|
|
|
* case the e820 map provided to us will cover the static
|
|
|
|
* maximum region.
|
|
|
|
*/
|
|
|
|
if (xen_initial_domain()) {
|
|
|
|
ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
|
|
|
|
if (ret > 0)
|
|
|
|
max_pages = ret;
|
|
|
|
}
|
|
|
|
|
2011-08-19 08:57:16 -06:00
|
|
|
return min(max_pages, MAX_DOMAIN_PAGES);
|
|
|
|
}
|
|
|
|
|
2011-09-29 05:26:19 -06:00
|
|
|
static void xen_align_and_add_e820_region(u64 start, u64 size, int type)
|
|
|
|
{
|
|
|
|
u64 end = start + size;
|
|
|
|
|
|
|
|
/* Align RAM regions to page boundaries. */
|
|
|
|
if (type == E820_RAM) {
|
|
|
|
start = PAGE_ALIGN(start);
|
|
|
|
end &= ~((u64)PAGE_SIZE - 1);
|
|
|
|
}
|
|
|
|
|
|
|
|
e820_add_region(start, end - start, type);
|
|
|
|
}
|
|
|
|
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
/**
|
|
|
|
* machine_specific_memory_setup - Hook for machine specific memory setup.
|
|
|
|
**/
|
|
|
|
char * __init xen_memory_setup(void)
|
|
|
|
{
|
2009-02-06 20:09:48 -07:00
|
|
|
static struct e820entry map[E820MAX] __initdata;
|
|
|
|
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
unsigned long max_pfn = xen_start_info->nr_pages;
|
2009-02-06 20:09:48 -07:00
|
|
|
unsigned long long mem_end;
|
|
|
|
int rc;
|
|
|
|
struct xen_memory_map memmap;
|
2011-09-29 05:26:19 -06:00
|
|
|
unsigned long max_pages;
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
unsigned long last_pfn = 0;
|
2010-08-30 17:41:02 -06:00
|
|
|
unsigned long extra_pages = 0;
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
unsigned long populated;
|
2009-02-06 20:09:48 -07:00
|
|
|
int i;
|
2010-09-02 09:16:00 -06:00
|
|
|
int op;
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
|
2008-05-26 16:31:19 -06:00
|
|
|
max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
|
2009-02-06 20:09:48 -07:00
|
|
|
mem_end = PFN_PHYS(max_pfn);
|
|
|
|
|
|
|
|
memmap.nr_entries = E820MAX;
|
|
|
|
set_xen_guest_handle(memmap.buffer, map);
|
|
|
|
|
2010-09-02 09:16:00 -06:00
|
|
|
op = xen_initial_domain() ?
|
|
|
|
XENMEM_machine_memory_map :
|
|
|
|
XENMEM_memory_map;
|
|
|
|
rc = HYPERVISOR_memory_op(op, &memmap);
|
2009-02-06 20:09:48 -07:00
|
|
|
if (rc == -ENOSYS) {
|
2010-10-28 12:32:29 -06:00
|
|
|
BUG_ON(xen_initial_domain());
|
2009-02-06 20:09:48 -07:00
|
|
|
memmap.nr_entries = 1;
|
|
|
|
map[0].addr = 0ULL;
|
|
|
|
map[0].size = mem_end;
|
|
|
|
/* 8MB slack (to balance backend allocations). */
|
|
|
|
map[0].size += 8ULL << 20;
|
|
|
|
map[0].type = E820_RAM;
|
|
|
|
rc = 0;
|
|
|
|
}
|
|
|
|
BUG_ON(rc);
|
2008-05-26 16:31:19 -06:00
|
|
|
|
2011-09-29 05:26:19 -06:00
|
|
|
/* Make sure the Xen-supplied memory map is well-ordered. */
|
|
|
|
sanitize_e820_map(map, memmap.nr_entries, &memmap.nr_entries);
|
|
|
|
|
|
|
|
max_pages = xen_get_max_pages();
|
|
|
|
if (max_pages > max_pfn)
|
|
|
|
extra_pages += max_pages - max_pfn;
|
|
|
|
|
2011-09-28 10:46:36 -06:00
|
|
|
/*
|
|
|
|
* Set P2M for all non-RAM pages and E820 gaps to be identity
|
|
|
|
* type PFNs. Any RAM pages that would be made inaccesible by
|
|
|
|
* this are first released.
|
|
|
|
*/
|
|
|
|
xen_released_pages = xen_set_identity_and_release(
|
|
|
|
map, memmap.nr_entries, max_pfn);
|
2011-09-29 05:26:19 -06:00
|
|
|
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
/*
|
|
|
|
* Populate back the non-RAM pages and E820 gaps that had been
|
|
|
|
* released. */
|
|
|
|
populated = xen_populate_chunk(map, memmap.nr_entries,
|
|
|
|
max_pfn, &last_pfn, xen_released_pages);
|
|
|
|
|
2012-05-29 10:36:43 -06:00
|
|
|
xen_released_pages -= populated;
|
|
|
|
extra_pages += xen_released_pages;
|
xen/setup: Populate freed MFNs from non-RAM E820 entries and gaps to E820 RAM
When the Xen hypervisor boots a PV kernel it hands it two pieces
of information: nr_pages and a made up E820 entry.
The nr_pages value defines the range from zero to nr_pages of PFNs
which have a valid Machine Frame Number (MFN) underneath it. The
E820 mirrors that (with the VGA hole):
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 00000000000a0000 (usable)
Xen: 00000000000a0000 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000080800000 (usable)
The fun comes when a PV guest that is run with a machine E820 - that
can either be the initial domain or a PCI PV guest, where the E820
looks like the normal thing:
BIOS-provided physical RAM map:
Xen: 0000000000000000 - 000000000009e000 (usable)
Xen: 000000000009ec00 - 0000000000100000 (reserved)
Xen: 0000000000100000 - 0000000020000000 (usable)
Xen: 0000000020000000 - 0000000020200000 (reserved)
Xen: 0000000020200000 - 0000000040000000 (usable)
Xen: 0000000040000000 - 0000000040200000 (reserved)
Xen: 0000000040200000 - 00000000bad80000 (usable)
Xen: 00000000bad80000 - 00000000badc9000 (ACPI NVS)
..
With that overlaying the nr_pages directly on the E820 does not
work as there are gaps and non-RAM regions that won't be used
by the memory allocator. The 'xen_release_chunk' helps with that
by punching holes in the P2M (PFN to MFN lookup tree) for those
regions and tells us that:
Freeing 20000-20200 pfn range: 512 pages freed
Freeing 40000-40200 pfn range: 512 pages freed
Freeing bad80-badf4 pfn range: 116 pages freed
Freeing badf6-bae7f pfn range: 137 pages freed
Freeing bb000-100000 pfn range: 282624 pages freed
Released 283999 pages of unused memory
Those 283999 pages are subtracted from the nr_pages and are returned
to the hypervisor. The end result is that the initial domain
boots with 1GB less memory as the nr_pages has been subtracted by
the amount of pages residing within the PCI hole. It can balloon up
to that if desired using 'xl mem-set 0 8092', but the balloon driver
is not always compiled in for the initial domain.
This patch, implements the populate hypercall (XENMEM_populate_physmap)
which increases the the domain with the same amount of pages that
were released.
The other solution (that did not work) was to transplant the MFN in
the P2M tree - the ones that were going to be freed were put in
the E820_RAM regions past the nr_pages. But the modifications to the
M2P array (the other side of creating PTEs) were not carried away.
As the hypervisor is the only one capable of modifying that and the
only two hypercalls that would do this are: the update_va_mapping
(which won't work, as during initial bootup only PFNs up to nr_pages
are mapped in the guest) or via the populate hypercall.
The end result is that the kernel can now boot with the
nr_pages without having to subtract the 283999 pages.
On a 8GB machine, with various dom0_mem= parameters this is what we get:
no dom0_mem
-Memory: 6485264k/9435136k available (5817k kernel code, 1136060k absent, 1813812k reserved, 2899k data, 696k init)
+Memory: 7619036k/9435136k available (5817k kernel code, 1136060k absent, 680040k reserved, 2899k data, 696k init)
dom0_mem=3G
-Memory: 2616536k/9435136k available (5817k kernel code, 1136060k absent, 5682540k reserved, 2899k data, 696k init)
+Memory: 2703776k/9435136k available (5817k kernel code, 1136060k absent, 5595300k reserved, 2899k data, 696k init)
dom0_mem=max:3G
-Memory: 2696732k/4281724k available (5817k kernel code, 1136060k absent, 448932k reserved, 2899k data, 696k init)
+Memory: 2702204k/4281724k available (5817k kernel code, 1136060k absent, 443460k reserved, 2899k data, 696k init)
And the 'xm list' or 'xl list' now reflect what the dom0_mem=
argument is.
Acked-by: David Vrabel <david.vrabel@citrix.com>
[v2: Use populate hypercall]
[v3: Remove debug printks]
[v4: Simplify code]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
2012-04-06 08:07:11 -06:00
|
|
|
|
|
|
|
if (last_pfn > max_pfn) {
|
|
|
|
max_pfn = min(MAX_DOMAIN_PAGES, last_pfn);
|
|
|
|
mem_end = PFN_PHYS(max_pfn);
|
|
|
|
}
|
2011-09-29 05:26:19 -06:00
|
|
|
/*
|
|
|
|
* Clamp the amount of extra memory to a EXTRA_MEM_RATIO
|
|
|
|
* factor the base size. On non-highmem systems, the base
|
|
|
|
* size is the full initial memory allocation; on highmem it
|
|
|
|
* is limited to the max size of lowmem, so that it doesn't
|
|
|
|
* get completely filled.
|
|
|
|
*
|
|
|
|
* In principle there could be a problem in lowmem systems if
|
|
|
|
* the initial memory is also very large with respect to
|
|
|
|
* lowmem, but we won't try to deal with that here.
|
|
|
|
*/
|
|
|
|
extra_pages = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
|
|
|
|
extra_pages);
|
|
|
|
i = 0;
|
|
|
|
while (i < memmap.nr_entries) {
|
|
|
|
u64 addr = map[i].addr;
|
|
|
|
u64 size = map[i].size;
|
|
|
|
u32 type = map[i].type;
|
|
|
|
|
|
|
|
if (type == E820_RAM) {
|
|
|
|
if (addr < mem_end) {
|
|
|
|
size = min(size, mem_end - addr);
|
|
|
|
} else if (extra_pages) {
|
|
|
|
size = min(size, (u64)extra_pages * PAGE_SIZE);
|
|
|
|
extra_pages -= size / PAGE_SIZE;
|
|
|
|
xen_add_extra_mem(addr, size);
|
|
|
|
} else
|
|
|
|
type = E820_UNUSABLE;
|
2010-09-29 17:54:33 -06:00
|
|
|
}
|
|
|
|
|
2011-09-29 05:26:19 -06:00
|
|
|
xen_align_and_add_e820_region(addr, size, type);
|
2010-09-02 18:10:12 -06:00
|
|
|
|
2011-09-29 05:26:19 -06:00
|
|
|
map[i].addr += size;
|
|
|
|
map[i].size -= size;
|
|
|
|
if (map[i].size == 0)
|
|
|
|
i++;
|
2009-02-06 20:09:48 -07:00
|
|
|
}
|
2008-06-16 15:54:49 -06:00
|
|
|
|
|
|
|
/*
|
2010-10-28 12:32:29 -06:00
|
|
|
* In domU, the ISA region is normal, usable memory, but we
|
|
|
|
* reserve ISA memory anyway because too many things poke
|
2008-06-16 15:54:49 -06:00
|
|
|
* about in there.
|
|
|
|
*/
|
|
|
|
e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
|
|
|
|
E820_RESERVED);
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
|
2008-06-16 15:54:46 -06:00
|
|
|
/*
|
|
|
|
* Reserve Xen bits:
|
|
|
|
* - mfn_list
|
|
|
|
* - xen_start_info
|
|
|
|
* See comment above "struct start_info" in <xen/interface/xen.h>
|
|
|
|
*/
|
2011-07-12 03:16:06 -06:00
|
|
|
memblock_reserve(__pa(xen_start_info->mfn_list),
|
|
|
|
xen_start_info->pt_base - xen_start_info->mfn_list);
|
2008-06-16 15:54:46 -06:00
|
|
|
|
|
|
|
sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
|
|
|
|
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
return "Xen";
|
|
|
|
}
|
|
|
|
|
2007-07-20 01:31:43 -06:00
|
|
|
/*
|
|
|
|
* Set the bit indicating "nosegneg" library variants should be used.
|
2008-07-12 03:22:00 -06:00
|
|
|
* We only need to bother in pure 32-bit mode; compat 32-bit processes
|
|
|
|
* can have un-truncated segments, so wrapping around is allowed.
|
2007-07-20 01:31:43 -06:00
|
|
|
*/
|
2008-01-30 05:33:25 -07:00
|
|
|
static void __init fiddle_vdso(void)
|
2007-07-20 01:31:43 -06:00
|
|
|
{
|
2008-07-12 03:22:00 -06:00
|
|
|
#ifdef CONFIG_X86_32
|
|
|
|
u32 *mask;
|
|
|
|
mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
|
|
|
|
*mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
|
|
|
|
mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
|
2007-07-20 01:31:43 -06:00
|
|
|
*mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
|
2008-07-08 16:07:14 -06:00
|
|
|
#endif
|
2007-07-20 01:31:43 -06:00
|
|
|
}
|
|
|
|
|
2011-05-04 12:17:21 -06:00
|
|
|
static int __cpuinit register_callback(unsigned type, const void *func)
|
2008-03-17 17:37:17 -06:00
|
|
|
{
|
2008-07-08 16:07:02 -06:00
|
|
|
struct callback_register callback = {
|
|
|
|
.type = type,
|
|
|
|
.address = XEN_CALLBACK(__KERNEL_CS, func),
|
2008-03-17 17:37:17 -06:00
|
|
|
.flags = CALLBACKF_mask_events,
|
|
|
|
};
|
|
|
|
|
2008-07-08 16:07:02 -06:00
|
|
|
return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
|
|
|
|
}
|
|
|
|
|
|
|
|
void __cpuinit xen_enable_sysenter(void)
|
|
|
|
{
|
2008-07-08 16:07:14 -06:00
|
|
|
int ret;
|
2008-07-10 17:24:08 -06:00
|
|
|
unsigned sysenter_feature;
|
2008-07-08 16:07:14 -06:00
|
|
|
|
|
|
|
#ifdef CONFIG_X86_32
|
2008-07-10 17:24:08 -06:00
|
|
|
sysenter_feature = X86_FEATURE_SEP;
|
2008-07-08 16:07:14 -06:00
|
|
|
#else
|
2008-07-10 17:24:08 -06:00
|
|
|
sysenter_feature = X86_FEATURE_SYSENTER32;
|
2008-07-08 16:07:14 -06:00
|
|
|
#endif
|
2008-07-08 16:07:02 -06:00
|
|
|
|
2008-07-10 17:24:08 -06:00
|
|
|
if (!boot_cpu_has(sysenter_feature))
|
|
|
|
return;
|
|
|
|
|
2008-07-08 16:07:14 -06:00
|
|
|
ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
|
2008-07-10 17:24:08 -06:00
|
|
|
if(ret != 0)
|
|
|
|
setup_clear_cpu_cap(sysenter_feature);
|
2008-03-17 17:37:17 -06:00
|
|
|
}
|
|
|
|
|
2008-07-08 16:07:14 -06:00
|
|
|
void __cpuinit xen_enable_syscall(void)
|
|
|
|
{
|
|
|
|
#ifdef CONFIG_X86_64
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
|
|
|
|
if (ret != 0) {
|
2008-07-12 03:22:06 -06:00
|
|
|
printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
|
2008-07-10 17:24:08 -06:00
|
|
|
/* Pretty fatal; 64-bit userspace has no other
|
|
|
|
mechanism for syscalls. */
|
|
|
|
}
|
|
|
|
|
|
|
|
if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
|
2008-07-08 16:07:14 -06:00
|
|
|
ret = register_callback(CALLBACKTYPE_syscall32,
|
|
|
|
xen_syscall32_target);
|
2008-07-12 03:22:06 -06:00
|
|
|
if (ret != 0)
|
2008-07-10 17:24:08 -06:00
|
|
|
setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
|
2008-07-08 16:07:14 -06:00
|
|
|
}
|
|
|
|
#endif /* CONFIG_X86_64 */
|
|
|
|
}
|
|
|
|
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
void __init xen_arch_setup(void)
|
|
|
|
{
|
2010-07-15 12:56:49 -06:00
|
|
|
xen_panic_handler_init();
|
|
|
|
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
|
|
|
|
HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
|
|
|
|
|
|
|
|
if (!xen_feature(XENFEAT_auto_translated_physmap))
|
2008-12-16 12:56:06 -07:00
|
|
|
HYPERVISOR_vm_assist(VMASST_CMD_enable,
|
|
|
|
VMASST_TYPE_pae_extended_cr3);
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
|
2008-07-08 16:07:02 -06:00
|
|
|
if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
|
|
|
|
register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
|
|
|
|
BUG();
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
|
2008-03-17 17:37:17 -06:00
|
|
|
xen_enable_sysenter();
|
2008-07-08 16:07:14 -06:00
|
|
|
xen_enable_syscall();
|
2008-03-17 17:37:17 -06:00
|
|
|
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
#ifdef CONFIG_ACPI
|
|
|
|
if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
|
|
|
|
printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
|
|
|
|
disable_acpi();
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
|
|
|
memcpy(boot_command_line, xen_start_info->cmd_line,
|
|
|
|
MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
|
|
|
|
COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
|
|
|
|
|
2010-11-22 18:17:50 -07:00
|
|
|
/* Set up idle, making sure it calls safe_halt() pvop */
|
|
|
|
#ifdef CONFIG_X86_32
|
|
|
|
boot_cpu_data.hlt_works_ok = 1;
|
|
|
|
#endif
|
2011-04-01 16:28:35 -06:00
|
|
|
disable_cpuidle();
|
2012-03-13 18:06:57 -06:00
|
|
|
disable_cpufreq();
|
2011-11-21 16:02:02 -07:00
|
|
|
WARN_ON(set_pm_idle_to_default());
|
2007-07-20 01:31:43 -06:00
|
|
|
fiddle_vdso();
|
xen: Core Xen implementation
This patch is a rollup of all the core pieces of the Xen
implementation, including:
- booting and setup
- pagetable setup
- privileged instructions
- segmentation
- interrupt flags
- upcalls
- multicall batching
BOOTING AND SETUP
The vmlinux image is decorated with ELF notes which tell the Xen
domain builder what the kernel's requirements are; the domain builder
then constructs the address space accordingly and starts the kernel.
Xen has its own entrypoint for the kernel (contained in an ELF note).
The ELF notes are set up by xen-head.S, which is included into head.S.
In principle it could be linked separately, but it seems to provoke
lots of binutils bugs.
Because the domain builder starts the kernel in a fairly sane state
(32-bit protected mode, paging enabled, flat segments set up), there's
not a lot of setup needed before starting the kernel proper. The main
steps are:
1. Install the Xen paravirt_ops, which is simply a matter of a
structure assignment.
2. Set init_mm to use the Xen-supplied pagetables (analogous to the
head.S generated pagetables in a native boot).
3. Reserve address space for Xen, since it takes a chunk at the top
of the address space for its own use.
4. Call start_kernel()
PAGETABLE SETUP
Once we hit the main kernel boot sequence, it will end up calling back
via paravirt_ops to set up various pieces of Xen specific state. One
of the critical things which requires a bit of extra care is the
construction of the initial init_mm pagetable. Because Xen places
tight constraints on pagetables (an active pagetable must always be
valid, and must always be mapped read-only to the guest domain), we
need to be careful when constructing the new pagetable to keep these
constraints in mind. It turns out that the easiest way to do this is
use the initial Xen-provided pagetable as a template, and then just
insert new mappings for memory where a mapping doesn't already exist.
This means that during pagetable setup, it uses a special version of
xen_set_pte which ignores any attempt to remap a read-only page as
read-write (since Xen will map its own initial pagetable as RO), but
lets other changes to the ptes happen, so that things like NX are set
properly.
PRIVILEGED INSTRUCTIONS AND SEGMENTATION
When the kernel runs under Xen, it runs in ring 1 rather than ring 0.
This means that it is more privileged than user-mode in ring 3, but it
still can't run privileged instructions directly. Non-performance
critical instructions are dealt with by taking a privilege exception
and trapping into the hypervisor and emulating the instruction, but
more performance-critical instructions have their own specific
paravirt_ops. In many cases we can avoid having to do any hypercalls
for these instructions, or the Xen implementation is quite different
from the normal native version.
The privileged instructions fall into the broad classes of:
Segmentation: setting up the GDT and the GDT entries, LDT,
TLS and so on. Xen doesn't allow the GDT to be directly
modified; all GDT updates are done via hypercalls where the new
entries can be validated. This is important because Xen uses
segment limits to prevent the guest kernel from damaging the
hypervisor itself.
Traps and exceptions: Xen uses a special format for trap entrypoints,
so when the kernel wants to set an IDT entry, it needs to be
converted to the form Xen expects. Xen sets int 0x80 up specially
so that the trap goes straight from userspace into the guest kernel
without going via the hypervisor. sysenter isn't supported.
Kernel stack: The esp0 entry is extracted from the tss and provided to
Xen.
TLB operations: the various TLB calls are mapped into corresponding
Xen hypercalls.
Control registers: all the control registers are privileged. The most
important is cr3, which points to the base of the current pagetable,
and we handle it specially.
Another instruction we treat specially is CPUID, even though its not
privileged. We want to control what CPU features are visible to the
rest of the kernel, and so CPUID ends up going into a paravirt_op.
Xen implements this mainly to disable the ACPI and APIC subsystems.
INTERRUPT FLAGS
Xen maintains its own separate flag for masking events, which is
contained within the per-cpu vcpu_info structure. Because the guest
kernel runs in ring 1 and not 0, the IF flag in EFLAGS is completely
ignored (and must be, because even if a guest domain disables
interrupts for itself, it can't disable them overall).
(A note on terminology: "events" and interrupts are effectively
synonymous. However, rather than using an "enable flag", Xen uses a
"mask flag", which blocks event delivery when it is non-zero.)
There are paravirt_ops for each of cli/sti/save_fl/restore_fl, which
are implemented to manage the Xen event mask state. The only thing
worth noting is that when events are unmasked, we need to explicitly
see if there's a pending event and call into the hypervisor to make
sure it gets delivered.
UPCALLS
Xen needs a couple of upcall (or callback) functions to be implemented
by each guest. One is the event upcalls, which is how events
(interrupts, effectively) are delivered to the guests. The other is
the failsafe callback, which is used to report errors in either
reloading a segment register, or caused by iret. These are
implemented in i386/kernel/entry.S so they can jump into the normal
iret_exc path when necessary.
MULTICALL BATCHING
Xen provides a multicall mechanism, which allows multiple hypercalls
to be issued at once in order to mitigate the cost of trapping into
the hypervisor. This is particularly useful for context switches,
since the 4-5 hypercalls they would normally need (reload cr3, update
TLS, maybe update LDT) can be reduced to one. This patch implements a
generic batching mechanism for hypercalls, which gets used in many
places in the Xen code.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Ian Pratt <ian.pratt@xensource.com>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Cc: Adrian Bunk <bunk@stusta.de>
2007-07-17 19:37:04 -06:00
|
|
|
}
|