Patching a jump label involves patching a single instruction at a time,
swizzling between a branch and a NOP. The architecture treats these
instructions specially, so a concurrently executing CPU is guaranteed to
see either the NOP or the branch, rather than an amalgamation of the two
instruction encodings.
However, in order to guarantee that the new instruction is visible, it
is necessary to send an IPI to the concurrently executing CPU so that it
discards any previously fetched instructions from its pipeline. This
operation therefore cannot be completed from a context with IRQs
disabled, but this is exactly what happens on the jump label path where
the hotplug lock is held and irqs are subsequently disabled by
stop_machine_cpuslocked(). This results in a deadlock during boot on
Hikey-960.
Due to the architectural guarantees around patching NOPs and branches,
we don't actually need to stop_machine() at all on the jump label path,
so we can avoid the deadlock by using the "nosync" variant of our
instruction patching routine.
Fixes: 693350a799 ("arm64: insn: Don't fallback on nosync path for general insn patching")
Reported-by: Tuomas Tynkkynen <tuomas.tynkkynen@iki.fi>
Reported-by: John Stultz <john.stultz@linaro.org>
Tested-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Tuomas Tynkkynen <tuomas@tuxera.com>
Tested-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Return statements in functions returning bool should use true or false
instead of an integer value. This code was detected with the help of
Coccinelle.
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Now that we understand the deadlock arising from flush_icache_range()
on the kexec crash kernel path, add a comment to justify the use of
__flush_icache_range() here.
Reported-by: Dave Kleikamp <dave.kleikamp@oracle.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
The SDEI stack helper functions are only used by _on_sdei_stack() and
refer to symbols (e.g. sdei_stack_normal_ptr) that are only defined if
CONFIG_VMAP_STACK=y.
Mark these functions as static, so we don't run into errors at link-time
due to references to undefined symbols. Stick all the parameters onto
the same line whilst we're passing through.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Include KASLR offset in arm64 VMCOREINFO ELF notes to assist in
debugging. vmcore parsing in user-space already expects this value in
the notes and we are providing it for portability of those existing
tools with x86.
Ideally we would like core code to do this (so that way this
information won't be missed when an architecture adds KASLR support),
but mips has CONFIG_RANDOMIZE_BASE, and doesn't provide kaslr_offset(),
so I am not sure if this is needed for mips (and other such similar arch
cases in future). So, lets keep this architecture specific for now.
As an example of a user-space use-case, consider the
makedumpfile user-space utility which will need fixup to use this
KASLR offset to work with cases where we need to find a way to
translate symbol address from vmlinux to kernel run time address
in case of KASLR boot on arm64.
I have already submitted the makedumpfile user-space patch upstream
and the maintainer has suggested to wait for the kernel changes to be
included (see [0]).
I tested this on my qualcomm amberwing board both for KASLR and
non-KASLR boot cases:
Without this patch:
# cat > scrub.conf << EOF
[vmlinux]
erase jiffies
erase init_task.utime
for tsk in init_task.tasks.next within task_struct:tasks
erase tsk.utime
endfor
EOF
# makedumpfile --split -d 31 -x vmlinux --config scrub.conf vmcore dumpfile_{1,2,3}
readpage_elf: Attempt to read non-existent page at 0xffffa8a5bf180000.
readmem: type_addr: 1, addr:ffffa8a5bf180000, size:8
vaddr_to_paddr_arm64: Can't read pgd
readmem: Can't convert a virtual address(ffff0000092a542c) to physical
address.
readmem: type_addr: 0, addr:ffff0000092a542c, size:390
check_release: Can't get the address of system_utsname
After this patch check_release() is ok, and also we are able to erase
symbol from vmcore (I checked this with kernel 4.18.0-rc4+):
# makedumpfile --split -d 31 -x vmlinux --config scrub.conf vmcore dumpfile_{1,2,3}
The kernel version is not supported.
The makedumpfile operation may be incomplete.
Checking for memory holes : [100.0 %] \
Checking for memory holes : [100.0 %] |
Checking foExcluding unnecessary pages : [100.0 %]
\
Excluding unnecessary pages : [100.0 %] \
The dumpfiles are saved to dumpfile_1, dumpfile_2, and dumpfile_3.
makedumpfile Completed.
[0] https://www.spinics.net/lists/kexec/msg21195.html
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
Acked-by: James Morse <james.morse@arm.com>
Signed-off-by: Bhupesh Sharma <bhsharma@redhat.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
It is useful to get the running time of a thread. Doing so in an
efficient manner can be important for performance of user applications.
Avoiding system calls in `clock_gettime` when handling
CLOCK_THREAD_CPUTIME_ID is important. Other clocks are handled in the
VDSO, but CLOCK_THREAD_CPUTIME_ID falls back on the system call.
CLOCK_THREAD_CPUTIME_ID is not handled in the VDSO since it would have
costs associated with maintaining updated user space accessible time
offsets. These offsets have to be updated everytime the a thread is
scheduled/descheduled. However, for programs regularly checking the
running time of a thread, this is a performance improvement.
This patch takes a middle ground, and adds support for cap_user_time an
optional feature of the perf_event API. This way costs are only
incurred when the perf_event api is enabled. This is done the same way
as it is in x86.
Ultimately this allows calculating the thread running time in userspace
on aarch64 as follows (adapted from perf_event_open manpage):
u32 seq, time_mult, time_shift;
u64 running, count, time_offset, quot, rem, delta;
struct perf_event_mmap_page *pc;
pc = buf; // buf is the perf event mmaped page as documented in the API.
if (pc->cap_usr_time) {
do {
seq = pc->lock;
barrier();
running = pc->time_running;
count = readCNTVCT_EL0(); // Read ARM hardware clock.
time_offset = pc->time_offset;
time_mult = pc->time_mult;
time_shift = pc->time_shift;
barrier();
} while (pc->lock != seq);
quot = (count >> time_shift);
rem = count & (((u64)1 << time_shift) - 1);
delta = time_offset + quot * time_mult +
((rem * time_mult) >> time_shift);
running += delta;
// running now has the current nanosecond level thread time.
}
Summary of changes in the patch:
For aarch64 systems, make arch_perf_update_userpage update the timing
information stored in the perf_event page. Requiring the following
calculations:
- Calculate the appropriate time_mult, and time_shift factors to convert
ticks to nano seconds for the current clock frequency.
- Adjust the mult and shift factors to avoid shift factors of 32 bits.
(possibly unnecessary)
- The time_offset userspace should apply when doing calculations:
negative the current sched time (now), because time_running and
time_enabled fields of the perf_event page have just been updated.
Toggle bits to appropriate values:
- Enable cap_user_time
Signed-off-by: Michael O'Farrell <micpof@gmail.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
When kernel mode NEON was first introduced to the arm64 kernel,
every call to kernel_neon_begin()/_end() stacked resp. unstacked
the entire NEON register file, making it worthwile to reduce the
number of used NEON registers to a bare minimum, and only stack
those. kernel_neon_begin_partial() was introduced for this purpose,
but after the refactoring for SVE and other changes, it no longer
exists and was simply #define'd to kernel_neon_begin() directly.
In the mean time, all users have been updated, so let's remove
the fallback macro.
Reviewed-by: Dave Martin <Dave.Martin@arm.com>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
machine_kexec flushes the reboot_code_buffer from the icache
after stopping the other cpus.
Commit 3b8c9f1cdf ("arm64: IPI each CPU after invalidating the I-cache
for kernel mappings") added an IPI call to flush_icache_range, which
causes a hang here, so replace the call with __flush_icache_range
Signed-off-by: Dave Kleikamp <dave.kleikamp@oracle.com>
Cc: AKASHI Takahiro <takahiro.akashi@linaro.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
We always run userspace with interrupts enabled, but with the recent
conversion of the syscall entry/exit code to C, we don't inform the
hardirq tracing code that interrupts are about to become enabled by
virtue of restoring the EL0 SPSR.
This patch ensures that trace_hardirqs_on() is called on the syscall
return path when we return to the assembly code with interrupts still
disabled.
Fixes: f37099b699 ("arm64: convert syscall trace logic to C")
Reported-by: Julien Grall <julien.grall@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Pull in arm perf updates, including support for 64-bit (chained) event
counters and some non-critical fixes for some of the system PMU drivers.
Signed-off-by: Will Deacon <will.deacon@arm.com>
The xen-privcmd driver, which can be modular, calls set_pte_at()
which in turn may call __sync_icache_dcache().
The call to __sync_icache_dcache() may be optimised out because it is
conditional on !pte_special(), and xen-privcmd calls pte_mkspecial().
But it seems unwise to rely on this optimisation.
Fixes: 3ad0876554 ("xen/privcmd: add IOCTL_PRIVCMD_MMAP_RESOURCE")
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Will Deacon <will.deacon@arm.com>
This adds support for the STACKLEAK gcc plugin to arm64 by implementing
stackleak_check_alloca(), based heavily on the x86 version, and adding the
two helpers used by the stackleak common code: current_top_of_stack() and
on_thread_stack(). The stack erasure calls are made at syscall returns.
Additionally, this disables the plugin in hypervisor and EFI stub code,
which are out of scope for the protection.
Acked-by: Alexander Popov <alex.popov@linux.com>
Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Laura Abbott <labbott@redhat.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
In preparation for enabling the stackleak plugin on arm64,
we need a way to get the bounds of the current stack. Extend
on_accessible_stack to get this information.
Acked-by: Alexander Popov <alex.popov@linux.com>
Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Laura Abbott <labbott@redhat.com>
[will: folded in fix for allmodconfig build breakage w/ sdei]
Signed-off-by: Will Deacon <will.deacon@arm.com>
Kconfig reports a warning on x86 builds after the ARM64 dependency
was added.
drivers/acpi/Kconfig:6:error: recursive dependency detected!
drivers/acpi/Kconfig:6: symbol ACPI depends on EFI
This rephrases the dependency to keep the ARM64 details out of the
shared Kconfig file, so Kconfig no longer gets confused by it.
For consistency, all three architectures that support ACPI now
select ARCH_SUPPORTS_ACPI in exactly the configuration in which
they allow it. We still need the 'default x86', as each one
wants a different default: default-y on x86, default-n on arm64,
and always-y on ia64.
Fixes: 5bcd44083a ("drivers: acpi: add dependency of EFI for arm64")
Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Will Deacon <will.deacon@arm.com>
This is a fix against the issue that crash dump kernel may hang up
during booting, which can happen on any ACPI-based system with "ACPI
Reclaim Memory."
(kernel messages after panic kicked off kdump)
(snip...)
Bye!
(snip...)
ACPI: Core revision 20170728
pud=000000002e7d0003, *pmd=000000002e7c0003, *pte=00e8000039710707
Internal error: Oops: 96000021 [#1] SMP
Modules linked in:
CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.14.0-rc6 #1
task: ffff000008d05180 task.stack: ffff000008cc0000
PC is at acpi_ns_lookup+0x25c/0x3c0
LR is at acpi_ds_load1_begin_op+0xa4/0x294
(snip...)
Process swapper/0 (pid: 0, stack limit = 0xffff000008cc0000)
Call trace:
(snip...)
[<ffff0000084a6764>] acpi_ns_lookup+0x25c/0x3c0
[<ffff00000849b4f8>] acpi_ds_load1_begin_op+0xa4/0x294
[<ffff0000084ad4ac>] acpi_ps_build_named_op+0xc4/0x198
[<ffff0000084ad6cc>] acpi_ps_create_op+0x14c/0x270
[<ffff0000084acfa8>] acpi_ps_parse_loop+0x188/0x5c8
[<ffff0000084ae048>] acpi_ps_parse_aml+0xb0/0x2b8
[<ffff0000084a8e10>] acpi_ns_one_complete_parse+0x144/0x184
[<ffff0000084a8e98>] acpi_ns_parse_table+0x48/0x68
[<ffff0000084a82cc>] acpi_ns_load_table+0x4c/0xdc
[<ffff0000084b32f8>] acpi_tb_load_namespace+0xe4/0x264
[<ffff000008baf9b4>] acpi_load_tables+0x48/0xc0
[<ffff000008badc20>] acpi_early_init+0x9c/0xd0
[<ffff000008b70d50>] start_kernel+0x3b4/0x43c
Code: b9008fb9 2a000318 36380054 32190318 (b94002c0)
---[ end trace c46ed37f9651c58e ]---
Kernel panic - not syncing: Fatal exception
Rebooting in 10 seconds..
(diagnosis)
* This fault is a data abort, alignment fault (ESR=0x96000021)
during reading out ACPI table.
* Initial ACPI tables are normally stored in system ram and marked as
"ACPI Reclaim memory" by the firmware.
* After the commit f56ab9a5b7 ("efi/arm: Don't mark ACPI reclaim
memory as MEMBLOCK_NOMAP"), those regions are differently handled
as they are "memblock-reserved", without NOMAP bit.
* So they are now excluded from device tree's "usable-memory-range"
which kexec-tools determines based on a current view of /proc/iomem.
* When crash dump kernel boots up, it tries to accesses ACPI tables by
mapping them with ioremap(), not ioremap_cache(), in acpi_os_ioremap()
since they are no longer part of mapped system ram.
* Given that ACPI accessor/helper functions are compiled in without
unaligned access support (ACPI_MISALIGNMENT_NOT_SUPPORTED),
any unaligned access to ACPI tables can cause a fatal panic.
With this patch, acpi_os_ioremap() always honors memory attribute
information provided by the firmware (EFI) and retaining cacheability
allows the kernel safe access to ACPI tables.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Reviewed-by: James Morse <james.morse@arm.com>
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reported-by and Tested-by: Bhupesh Sharma <bhsharma@redhat.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
There has been some confusion around what is necessary to prevent kexec
overwriting important memory regions. memblock: reserve, or nomap?
Only memblock nomap regions are reported via /proc/iomem, kexec's
user-space doesn't know about memblock_reserve()d regions.
Until commit f56ab9a5b7 ("efi/arm: Don't mark ACPI reclaim memory
as MEMBLOCK_NOMAP") the ACPI tables were nomap, now they are reserved
and thus possible for kexec to overwrite with the new kernel or initrd.
But this was always broken, as the UEFI memory map is also reserved
and not marked as nomap.
Exporting both nomap and reserved memblock types is a nuisance as
they live in different memblock structures which we can't walk at
the same time.
Take a second walk over memblock.reserved and add new 'reserved'
subnodes for the memblock_reserved() regions that aren't already
described by the existing code. (e.g. Kernel Code)
We use reserve_region_with_split() to find the gaps in existing named
regions. This handles the gap between 'kernel code' and 'kernel data'
which is memblock_reserve()d, but already partially described by
request_standard_resources(). e.g.:
| 80000000-dfffffff : System RAM
| 80080000-80ffffff : Kernel code
| 81000000-8158ffff : reserved
| 81590000-8237efff : Kernel data
| a0000000-dfffffff : Crash kernel
| e00f0000-f949ffff : System RAM
reserve_region_with_split needs kzalloc() which isn't available when
request_standard_resources() is called, use an initcall.
Reported-by: Bhupesh Sharma <bhsharma@redhat.com>
Reported-by: Tyler Baicar <tbaicar@codeaurora.org>
Suggested-by: Akashi Takahiro <takahiro.akashi@linaro.org>
Signed-off-by: James Morse <james.morse@arm.com>
Fixes: d28f6df130 ("arm64/kexec: Add core kexec support")
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
CC: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Not all toolchains have the baremetal elf targets, RedHat/Fedora ones
in particular. So, probe for whether it's available and use the previous
(linux) targets if it isn't.
Reported-by: Laura Abbott <labbott@redhat.com>
Tested-by: Laura Abbott <labbott@redhat.com>
Acked-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Cc: Paul Kocialkowski <contact@paulk.fr>
Signed-off-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Will Deacon <will.deacon@arm.com>
It's possible for userspace to control idx. Sanitize idx when using it
as an array index, to inhibit the potential spectre-v1 write gadget.
Found by smatch.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
syscall_trace_{enter,exit} are only called from C code, so drop the
asmlinkage qualifier from their definitions.
Signed-off-by: Will Deacon <will.deacon@arm.com>
To minimize the risk of userspace-controlled values being used under
speculation, this patch adds pt_regs based syscall wrappers for arm64,
which pass the minimum set of required userspace values to syscall
implementations. For each syscall, a wrapper which takes a pt_regs
argument is automatically generated, and this extracts the arguments
before calling the "real" syscall implementation.
Each syscall has three functions generated:
* __do_<compat_>sys_<name> is the "real" syscall implementation, with
the expected prototype.
* __se_<compat_>sys_<name> is the sign-extension/narrowing wrapper,
inherited from common code. This takes a series of long parameters,
casting each to the requisite types required by the "real" syscall
implementation in __do_<compat_>sys_<name>.
This wrapper *may* not be necessary on arm64 given the AAPCS rules on
unused register bits, but it seemed safer to keep the wrapper for now.
* __arm64_<compat_>_sys_<name> takes a struct pt_regs pointer, and
extracts *only* the relevant register values, passing these on to the
__se_<compat_>sys_<name> wrapper.
The syscall invocation code is updated to handle the calling convention
required by __arm64_<compat_>_sys_<name>, and passes a single struct
pt_regs pointer.
The compiler can fold the syscall implementation and its wrappers, such
that the overhead of this approach is minimized.
Note that we play games with sys_ni_syscall(). It can't be defined with
SYSCALL_DEFINE0() because we must avoid the possibility of error
injection. Additionally, there are a couple of locations where we need
to call it from C code, and we don't (currently) have a
ksys_ni_syscall(). While it has no wrapper, passing in a redundant
pt_regs pointer is benign per the AAPCS.
When ARCH_HAS_SYSCALL_WRAPPER is selected, no prototype is defines for
sys_ni_syscall(). Since we need to treat it differently for in-kernel
calls and the syscall tables, the prototype is defined as-required.
The wrappers are largely the same as their x86 counterparts, but
simplified as we don't have a variety of compat calling conventions that
require separate stubs. Unlike x86, we have some zero-argument compat
syscalls, and must define COMPAT_SYSCALL_DEFINE0() to ensure that these
are also given an __arm64_compat_sys_ prefix.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Dominik Brodowski <linux@dominikbrodowski.net>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
In preparation for converting to pt_regs syscall wrappers, convert our
existing compat wrappers to C. This will allow the pt_regs wrappers to
be automatically generated, and will allow for the compat register
manipulation to be folded in with the pt_regs accesses.
To avoid confusion with the upcoming pt_regs wrappers and existing
compat wrappers provided by core code, the C wrappers are renamed to
compat_sys_aarch32_<syscall>.
With the assembly wrappers gone, we can get rid of entry32.S and the
associated boilerplate.
Note that these must call the ksys_* syscall entry points, as the usual
sys_* entry points will be modified to take a single pt_regs pointer
argument.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
We don't currently annotate our mmap implementation as a syscall, as we
need to do to use pt_regs syscall wrappers.
Let's mark it as a real syscall.
There should be no functional change as a result of this patch.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Dominik Brodowski <linux@dominikbrodowski.net>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
We don't currently annotate our various sigreturn functions as syscalls,
as we need to do to use pt_regs syscall wrappers.
Let's mark them as real syscalls.
For compat_sys_sigreturn and compat_sys_rt_sigreturn, this changes the
return type from int to long, matching the prototypes in sys32.c.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Dominik Brodowski <linux@dominikbrodowski.net>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
With pt_regs syscall wrappers, the calling convention for
sys_personality() will change. Use ksys_personality(), which is
functionally equivalent.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Our syscall tables are aligned to 4096 bytes, which allowed their
addresses to be generated with a single adrp in entry.S. This has the
unfortunate property of wasting space in .rodata for the necessary
padding.
Now that the address is generated by C code, we can rely on the compiler
to do the right thing, and drop the alignemnt.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
We can zero GPRs x0 - x29 upon entry from EL0 to make it harder for
userspace to control values consumed by speculative gadgets.
We don't blat x30, since this is stashed much later, and we'll blat it
before invoking C code.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Now that all of the syscall logic works on the saved pt_regs, apply_ssbd
can safely corrupt x0-x3 in the entry paths, and we no longer need to
restore them. So let's remove the logic doing so.
With that logic gone, we can fold the branch target into the macro, so
that callers need not deal with this. GAS provides \@, which provides a
unique value per macro invocation, which we can use to create a unique
label.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Now that syscalls are invoked with pt_regs, we no longer need to ensure
that the argument regsiters are live in the entry assembly, and it's
fine to not restore them after context_tracking_user_exit() has
corrupted them.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Now that the syscall invocation logic is in C, we can migrate the rest
of the syscall entry logic over, so that the entry assembly needn't look
at the register values at all.
The SVE reset across syscall logic now unconditionally clears TIF_SVE,
but sve_user_disable() will only write back to CPACR_EL1 when SVE is
actually enabled.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Dave Martin <dave.martin@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Currently syscall tracing is a tricky assembly state machine, which can
be rather difficult to follow, and even harder to modify. Before we
start fiddling with it for pt_regs syscalls, let's convert it to C.
This is not intended to have any functional change.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
As a first step towards invoking syscalls with a pt_regs argument,
convert the raw syscall invocation logic to C. We end up with a bit more
register shuffling, but the unified invocation logic means we can unify
the tracing paths, too.
Previously, assembly had to open-code calls to ni_sys() when the system
call number was out-of-bounds for the relevant syscall table. This case
is now handled by invoke_syscall(), and the assembly no longer need to
handle this case explicitly. This allows the tracing paths to be
simplified and unified, as we no longer need the __ni_sys_trace path and
the __sys_trace_return label.
This only converts the invocation of the syscall. The rest of the
syscall triage and tracing is left in assembly for now, and will be
converted in subsequent patches.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
In preparation for invoking arbitrary syscalls from C code, let's define
a type for an arbitrary syscall, matching the parameter passing rules of
the AAPCS.
There should be no functional change as a result of this patch.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
The arm64 sigreturn* syscall handlers are non-standard. Rather than
taking a number of user parameters in registers as per the AAPCS,
they expect the pt_regs as their sole argument.
To make this work, we override the syscall definitions to invoke
wrappers written in assembly, which mov the SP into x0, and branch to
their respective C functions.
On other architectures (such as x86), the sigreturn* functions take no
argument and instead use current_pt_regs() to acquire the user
registers. This requires less boilerplate code, and allows for other
features such as interposing C code in this path.
This patch takes the same approach for arm64.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Tentatively-reviewed-by: Dave Martin <dave.martin@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
In subsequent patches, we'll want to make use of sve_user_enable() and
sve_user_disable() outside of kernel/fpsimd.c. Let's move these to
<asm/fpsimd.h> where we can make use of them.
To avoid ifdeffery in sequences like:
if (system_supports_sve() && some_condition)
sve_user_disable();
... empty stubs are provided when support for SVE is not enabled. Note
that system_supports_sve() contains as IS_ENABLED(CONFIG_ARM64_SVE), so
the sve_user_disable() call should be optimized away entirely when
CONFIG_ARM64_SVE is not selected.
To ensure that this is the case, the stub definitions contain a
BUILD_BUG(), as we do for other stubs for which calls should always be
optimized away when the relevant config option is not selected.
At the same time, the include list of <asm/fpsimd.h> is sorted while
adding <asm/sysreg.h>.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Dave Martin <dave.martin@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Now that we have sysreg_clear_set(), we can use this instead of
change_cpacr().
Note that the order of the set and clear arguments differs between
change_cpacr() and sysreg_clear_set(), so these are flipped as part of
the conversion. Also, sve_user_enable() redundantly clears
CPACR_EL1_ZEN_EL0EN before setting it; this is removed for clarity.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Dave Martin <dave.martin@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Now that we have sysreg_clear_set(), we can consistently use this
instead of config_sctlr_el1().
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Dave Martin <dave.martin@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morse <james.morse@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Currently we assert that the SCTLR_EL{1,2}_{SET,CLEAR} bits are
self-consistent with an assertion in config_sctlr_el1(). This is a bit
unusual, since config_sctlr_el1() doesn't make use of these definitions,
and is far away from the definitions themselves.
We can use the CPP #error directive to have equivalent assertions in
<asm/sysreg.h>, next to the definitions of the set/clear bits, which is
a bit clearer and simpler.
At the same time, lets fill in the upper 32 bits for both registers in
their respective RES0 definitions. This could be a little nicer with
GENMASK_ULL(63, 32), but this currently lives in <linux/bitops.h>, which
cannot safely be included from assembly, as <asm/sysreg.h> can.
Note the when the preprocessor evaluates an expression for an #if
directive, all signed or unsigned values are treated as intmax_t or
uintmax_t respectively. To avoid ambiguity, we define explicitly define
the mask of all 64 bits.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Dave Martin <dave.martin@arm.com>
Cc: James Morse <james.morse@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
In do_notify_resume, we manipulate thread_flags as a 32-bit unsigned
int, whereas thread_info::flags is a 64-bit unsigned long, and elsewhere
(e.g. in the entry assembly) we manipulate the flags as a 64-bit
quantity.
For consistency, and to avoid problems if we end up with more than 32
flags, let's make do_notify_resume take the flags as a 64-bit unsigned
long.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Dave Martin <dave.martin@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
This reverts commit 7e7df71fd5.
When unwinding out of the IRQ stack and onto the interrupted EL1 stack,
we cannot rely on the frame pointer being strictly increasing, as this
could terminate the backtrace early depending on how the stacks have
been allocated.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Implement calls to rseq_signal_deliver, rseq_handle_notify_resume
and rseq_syscall so that we can select HAVE_RSEQ on arm64.
Acked-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Building without NUMA but with FLATMEM results in a link error
because mem_map[] is not available:
aarch64-linux-ld -EB -maarch64elfb --no-undefined -X -pie -shared -Bsymbolic --no-apply-dynamic-relocs --build-id -o .tmp_vmlinux1 -T ./arch/arm64/kernel/vmlinux.lds --whole-archive built-in.a --no-whole-archive --start-group arch/arm64/lib/lib.a lib/lib.a --end-group
init/do_mounts.o: In function `mount_block_root':
do_mounts.c:(.init.text+0x1e8): undefined reference to `mem_map'
arch/arm64/kernel/vdso.o: In function `vdso_init':
vdso.c:(.init.text+0xb4): undefined reference to `mem_map'
This uses the same trick as the other architectures, making flatmem
depend on !NUMA to avoid the broken configuration.
Fixes: e7d4bac428 ("arm64: add ARM64-specific support for flatmem")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Add support for 64bit event by using chained event counters
and 64bit cycle counters.
PMUv3 allows chaining a pair of adjacent 32-bit counters, effectively
forming a 64-bit counter. The low/even counter is programmed to count
the event of interest, and the high/odd counter is programmed to count
the CHAIN event, taken when the low/even counter overflows.
For CPU cycles, when 64bit mode is requested, the cycle counter
is used in 64bit mode. If the cycle counter is not available,
falls back to chaining.
Cc: Will Deacon <will.deacon@arm.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
The arm64 PMU updates the event counters and reprograms the
counters in the overflow IRQ handler without disabling the
PMU. This could potentially cause skews in for group counters,
where the overflowed counters may potentially loose some event
counts, while they are reprogrammed. To prevent this, disable
the PMU while we process the counter overflows and enable it
right back when we are done.
This patch also moves the PMU stop/start routines to avoid a
forward declaration.
Suggested-by: Mark Rutland <mark.rutland@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
armv8pmu_select_counter always returns the passed idx. So
let us make that void and get rid of the pointless checks.
Suggested-by: Mark Rutland <mark.rutland@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
The armpmu uses get_event_idx callback to allocate an event
counter for a given event, which marks the selected counter
as "used". Now, when we delete the counter, the arm_pmu goes
ahead and clears the "used" bit and then invokes the "clear_event_idx"
call back, which kind of splits the job between the core code
and the backend. To keep things tidy, mandate the implementation
of clear_event_idx() and add it for exisiting backends.
This will be useful for adding the chained event support, where
we leave the event idx maintenance to the backend.
Also, when an event is removed from the PMU, reset the hw.idx
to indicate that a counter is not allocated for this event,
to help the backends do better checks. This will be also used
for the chain counter support.
Cc: Will Deacon <will.deacon@arm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Julien Thierry <julien.thierry@arm.com>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Convert the {read/write}_counter APIs to handle 64bit values
to enable supporting chained event counters. The backends still
use 32bit values and we pass them 32bit values only. So in effect
there are no functional changes.
Cc: Will Deacon <will.deacon@arm.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Julien Thierry <julien.thierry@arm.com>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Each PMU defines their max_period of the counter as the maximum
value that can be counted. Since all the PMU backends support
32bit counters by default, let us remove the redundant field.
No functional changes.
Cc: Will Deacon <will.deacon@arm.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Julien Thierry <julien.thierry@arm.com>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Current ACPI ARM64 NUMA initialization code in
acpi_numa_gicc_affinity_init()
carries out NUMA nodes creation and cpu<->node mappings at the same time
in the arch backend so that a single SRAT walk is needed to parse both
pieces of information. This implies that the cpu<->node mappings must
be stashed in an array (sized NR_CPUS) so that SMP code can later use
the stashed values to avoid another SRAT table walk to set-up the early
cpu<->node mappings.
If the kernel is configured with a NR_CPUS value less than the actual
processor entries in the SRAT (and MADT), the logic in
acpi_numa_gicc_affinity_init() is broken in that the cpu<->node mapping
is only carried out (and stashed for future use) only for a number of
SRAT entries up to NR_CPUS, which do not necessarily correspond to the
possible cpus detected at SMP initialization in
acpi_map_gic_cpu_interface() (ie MADT and SRAT processor entries order
is not enforced), which leaves the kernel with broken cpu<->node
mappings.
Furthermore, given the current ACPI NUMA code parsing logic in
acpi_numa_gicc_affinity_init(), PXM domains for CPUs that are not parsed
because they exceed NR_CPUS entries are not mapped to NUMA nodes (ie the
PXM corresponding node is not created in the kernel) leaving the system
with a broken NUMA topology.
Rework the ACPI ARM64 NUMA initialization process so that the NUMA
nodes creation and cpu<->node mappings are decoupled. cpu<->node
mappings are moved to SMP initialization code (where they are needed),
at the cost of an extra SRAT walk so that ACPI NUMA mappings can be
batched before being applied, fixing current parsing pitfalls.
Acked-by: Hanjun Guo <hanjun.guo@linaro.org>
Tested-by: John Garry <john.garry@huawei.com>
Fixes: d8b47fca8c ("arm64, ACPI, NUMA: NUMA support based on SRAT and
SLIT")
Link: http://lkml.kernel.org/r/1527768879-88161-2-git-send-email-xiexiuqi@huawei.com
Reported-by: Xie XiuQi <xiexiuqi@huawei.com>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Cc: Punit Agrawal <punit.agrawal@arm.com>
Cc: Jonathan Cameron <jonathan.cameron@huawei.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Hanjun Guo <guohanjun@huawei.com>
Cc: Ganapatrao Kulkarni <gkulkarni@caviumnetworks.com>
Cc: Jeremy Linton <jeremy.linton@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Xie XiuQi <xiexiuqi@huawei.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Flatmem is useful in reducing kernel memory usage.
One usecase is in kdump kernel. We are able to save
~14M by moving to flatmem scheme.
Cc: xe-kernel@external.cisco.com
Cc: Nikunj Kela <nkela@cisco.com>
Signed-off-by: Nikunj Kela <nkela@cisco.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
lkdtm calls flush_icache_range(), which results in an out-of-line call
to __flush_icache_range(), which is not exported to modules.
Export the symbol to modules to fix this build breakage.
Fixes: 3b8c9f1cdf ("arm64: IPI each CPU after invalidating the I-cache for kernel mappings")
Signed-off-by: Will Deacon <will.deacon@arm.com>
