2008-01-30 05:30:09 -07:00
|
|
|
/*
|
2005-04-16 16:20:36 -06:00
|
|
|
* Firmware replacement code.
|
2008-01-30 05:30:09 -07:00
|
|
|
*
|
2005-04-16 16:20:36 -06:00
|
|
|
* Work around broken BIOSes that don't set an aperture or only set the
|
2008-01-30 05:30:09 -07:00
|
|
|
* aperture in the AGP bridge.
|
|
|
|
* If all fails map the aperture over some low memory. This is cheaper than
|
|
|
|
* doing bounce buffering. The memory is lost. This is done at early boot
|
|
|
|
* because only the bootmem allocator can allocate 32+MB.
|
|
|
|
*
|
2005-04-16 16:20:36 -06:00
|
|
|
* Copyright 2002 Andi Kleen, SuSE Labs.
|
|
|
|
*/
|
|
|
|
#include <linux/kernel.h>
|
|
|
|
#include <linux/types.h>
|
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/bootmem.h>
|
|
|
|
#include <linux/mmzone.h>
|
|
|
|
#include <linux/pci_ids.h>
|
|
|
|
#include <linux/pci.h>
|
|
|
|
#include <linux/bitops.h>
|
2006-09-26 02:52:40 -06:00
|
|
|
#include <linux/ioport.h>
|
2008-03-13 16:05:41 -06:00
|
|
|
#include <linux/suspend.h>
|
2005-04-16 16:20:36 -06:00
|
|
|
#include <asm/e820.h>
|
|
|
|
#include <asm/io.h>
|
2007-10-24 04:49:47 -06:00
|
|
|
#include <asm/gart.h>
|
2005-04-16 16:20:36 -06:00
|
|
|
#include <asm/pci-direct.h>
|
2006-01-11 14:44:27 -07:00
|
|
|
#include <asm/dma.h>
|
2006-06-26 05:56:40 -06:00
|
|
|
#include <asm/k8.h>
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2007-10-24 04:49:50 -06:00
|
|
|
int gart_iommu_aperture;
|
2008-03-13 04:03:58 -06:00
|
|
|
int gart_iommu_aperture_disabled __initdata;
|
|
|
|
int gart_iommu_aperture_allowed __initdata;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
int fallback_aper_order __initdata = 1; /* 64MB */
|
2008-03-13 04:03:58 -06:00
|
|
|
int fallback_aper_force __initdata;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
int fix_aperture __initdata = 1;
|
|
|
|
|
2008-04-19 02:31:11 -06:00
|
|
|
struct bus_dev_range {
|
|
|
|
int bus;
|
|
|
|
int dev_base;
|
|
|
|
int dev_limit;
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct bus_dev_range bus_dev_ranges[] __initdata = {
|
|
|
|
{ 0x00, 0x18, 0x20},
|
|
|
|
{ 0xff, 0x00, 0x20},
|
|
|
|
{ 0xfe, 0x00, 0x20}
|
|
|
|
};
|
|
|
|
|
2006-09-26 02:52:40 -06:00
|
|
|
static struct resource gart_resource = {
|
|
|
|
.name = "GART",
|
|
|
|
.flags = IORESOURCE_MEM,
|
|
|
|
};
|
|
|
|
|
|
|
|
static void __init insert_aperture_resource(u32 aper_base, u32 aper_size)
|
|
|
|
{
|
|
|
|
gart_resource.start = aper_base;
|
|
|
|
gart_resource.end = aper_base + aper_size - 1;
|
|
|
|
insert_resource(&iomem_resource, &gart_resource);
|
|
|
|
}
|
|
|
|
|
2005-06-08 16:49:25 -06:00
|
|
|
/* This code runs before the PCI subsystem is initialized, so just
|
|
|
|
access the northbridge directly. */
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2008-01-30 05:30:09 -07:00
|
|
|
static u32 __init allocate_aperture(void)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
u32 aper_size;
|
2008-01-30 05:30:09 -07:00
|
|
|
void *p;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2008-04-14 21:40:37 -06:00
|
|
|
/* aper_size should <= 1G */
|
|
|
|
if (fallback_aper_order > 5)
|
|
|
|
fallback_aper_order = 5;
|
2008-01-30 05:30:09 -07:00
|
|
|
aper_size = (32 * 1024 * 1024) << fallback_aper_order;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2008-01-30 05:30:09 -07:00
|
|
|
/*
|
|
|
|
* Aperture has to be naturally aligned. This means a 2GB aperture
|
|
|
|
* won't have much chance of finding a place in the lower 4GB of
|
|
|
|
* memory. Unfortunately we cannot move it up because that would
|
|
|
|
* make the IOMMU useless.
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
2008-04-14 21:40:37 -06:00
|
|
|
/*
|
|
|
|
* using 512M as goal, in case kexec will load kernel_big
|
|
|
|
* that will do the on position decompress, and could overlap with
|
|
|
|
* that positon with gart that is used.
|
|
|
|
* sequende:
|
|
|
|
* kernel_small
|
|
|
|
* ==> kexec (with kdump trigger path or previous doesn't shutdown gart)
|
|
|
|
* ==> kernel_small(gart area become e820_reserved)
|
|
|
|
* ==> kexec (with kdump trigger path or previous doesn't shutdown gart)
|
|
|
|
* ==> kerne_big (uncompressed size will be big than 64M or 128M)
|
|
|
|
* so don't use 512M below as gart iommu, leave the space for kernel
|
|
|
|
* code for safe
|
|
|
|
*/
|
|
|
|
p = __alloc_bootmem_nopanic(aper_size, aper_size, 512ULL<<20);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (!p || __pa(p)+aper_size > 0xffffffff) {
|
2008-01-30 05:30:10 -07:00
|
|
|
printk(KERN_ERR
|
|
|
|
"Cannot allocate aperture memory hole (%p,%uK)\n",
|
|
|
|
p, aper_size>>10);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (p)
|
2007-05-02 11:27:13 -06:00
|
|
|
free_bootmem(__pa(p), aper_size);
|
2005-04-16 16:20:36 -06:00
|
|
|
return 0;
|
|
|
|
}
|
2008-01-30 05:30:10 -07:00
|
|
|
printk(KERN_INFO "Mapping aperture over %d KB of RAM @ %lx\n",
|
|
|
|
aper_size >> 10, __pa(p));
|
2006-09-26 02:52:40 -06:00
|
|
|
insert_aperture_resource((u32)__pa(p), aper_size);
|
2008-03-13 16:05:41 -06:00
|
|
|
register_nosave_region((u32)__pa(p) >> PAGE_SHIFT,
|
|
|
|
(u32)__pa(p+aper_size) >> PAGE_SHIFT);
|
2008-01-30 05:30:09 -07:00
|
|
|
|
|
|
|
return (u32)__pa(p);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-06-08 16:49:25 -06:00
|
|
|
/* Find a PCI capability */
|
2008-05-27 10:03:56 -06:00
|
|
|
static u32 __init find_cap(int bus, int slot, int func, int cap)
|
2008-01-30 05:30:09 -07:00
|
|
|
{
|
2005-04-16 16:20:36 -06:00
|
|
|
int bytes;
|
2008-01-30 05:30:09 -07:00
|
|
|
u8 pos;
|
|
|
|
|
2008-04-19 02:31:11 -06:00
|
|
|
if (!(read_pci_config_16(bus, slot, func, PCI_STATUS) &
|
2008-01-30 05:30:09 -07:00
|
|
|
PCI_STATUS_CAP_LIST))
|
2005-04-16 16:20:36 -06:00
|
|
|
return 0;
|
2008-01-30 05:30:09 -07:00
|
|
|
|
2008-04-19 02:31:11 -06:00
|
|
|
pos = read_pci_config_byte(bus, slot, func, PCI_CAPABILITY_LIST);
|
2008-01-30 05:30:09 -07:00
|
|
|
for (bytes = 0; bytes < 48 && pos >= 0x40; bytes++) {
|
2005-04-16 16:20:36 -06:00
|
|
|
u8 id;
|
2008-01-30 05:30:09 -07:00
|
|
|
|
|
|
|
pos &= ~3;
|
2008-04-19 02:31:11 -06:00
|
|
|
id = read_pci_config_byte(bus, slot, func, pos+PCI_CAP_LIST_ID);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (id == 0xff)
|
|
|
|
break;
|
2008-01-30 05:30:09 -07:00
|
|
|
if (id == cap)
|
|
|
|
return pos;
|
2008-04-19 02:31:11 -06:00
|
|
|
pos = read_pci_config_byte(bus, slot, func,
|
2008-01-30 05:30:09 -07:00
|
|
|
pos+PCI_CAP_LIST_NEXT);
|
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
return 0;
|
2008-01-30 05:30:09 -07:00
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
/* Read a standard AGPv3 bridge header */
|
2008-05-27 10:03:56 -06:00
|
|
|
static u32 __init read_agp(int bus, int slot, int func, int cap, u32 *order)
|
2008-01-30 05:30:09 -07:00
|
|
|
{
|
2005-04-16 16:20:36 -06:00
|
|
|
u32 apsize;
|
|
|
|
u32 apsizereg;
|
|
|
|
int nbits;
|
|
|
|
u32 aper_low, aper_hi;
|
|
|
|
u64 aper;
|
2008-04-13 02:11:41 -06:00
|
|
|
u32 old_order;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2008-04-19 02:31:11 -06:00
|
|
|
printk(KERN_INFO "AGP bridge at %02x:%02x:%02x\n", bus, slot, func);
|
|
|
|
apsizereg = read_pci_config_16(bus, slot, func, cap + 0x14);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (apsizereg == 0xffffffff) {
|
2008-01-30 05:30:10 -07:00
|
|
|
printk(KERN_ERR "APSIZE in AGP bridge unreadable\n");
|
2005-04-16 16:20:36 -06:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-04-13 02:11:41 -06:00
|
|
|
/* old_order could be the value from NB gart setting */
|
|
|
|
old_order = *order;
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
apsize = apsizereg & 0xfff;
|
|
|
|
/* Some BIOS use weird encodings not in the AGPv3 table. */
|
2008-01-30 05:30:09 -07:00
|
|
|
if (apsize & 0xff)
|
|
|
|
apsize |= 0xf00;
|
2005-04-16 16:20:36 -06:00
|
|
|
nbits = hweight16(apsize);
|
|
|
|
*order = 7 - nbits;
|
|
|
|
if ((int)*order < 0) /* < 32MB */
|
|
|
|
*order = 0;
|
2008-01-30 05:30:09 -07:00
|
|
|
|
2008-04-19 02:31:11 -06:00
|
|
|
aper_low = read_pci_config(bus, slot, func, 0x10);
|
|
|
|
aper_hi = read_pci_config(bus, slot, func, 0x14);
|
2005-04-16 16:20:36 -06:00
|
|
|
aper = (aper_low & ~((1<<22)-1)) | ((u64)aper_hi << 32);
|
|
|
|
|
2008-04-13 02:11:41 -06:00
|
|
|
/*
|
|
|
|
* On some sick chips, APSIZE is 0. It means it wants 4G
|
|
|
|
* so let double check that order, and lets trust AMD NB settings:
|
|
|
|
*/
|
2008-04-13 19:42:31 -06:00
|
|
|
printk(KERN_INFO "Aperture from AGP @ %Lx old size %u MB\n",
|
|
|
|
aper, 32 << old_order);
|
|
|
|
if (aper + (32ULL<<(20 + *order)) > 0x100000000ULL) {
|
2008-04-13 02:11:41 -06:00
|
|
|
printk(KERN_INFO "Aperture size %u MB (APSIZE %x) is not right, using settings from NB\n",
|
|
|
|
32 << *order, apsizereg);
|
|
|
|
*order = old_order;
|
|
|
|
}
|
|
|
|
|
2008-01-30 05:30:10 -07:00
|
|
|
printk(KERN_INFO "Aperture from AGP @ %Lx size %u MB (APSIZE %x)\n",
|
|
|
|
aper, 32 << *order, apsizereg);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2008-04-13 19:42:31 -06:00
|
|
|
if (!aperture_valid(aper, (32*1024*1024) << *order, 32<<20))
|
2008-01-30 05:30:09 -07:00
|
|
|
return 0;
|
|
|
|
return (u32)aper;
|
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2008-01-30 05:30:09 -07:00
|
|
|
/*
|
|
|
|
* Look for an AGP bridge. Windows only expects the aperture in the
|
|
|
|
* AGP bridge and some BIOS forget to initialize the Northbridge too.
|
|
|
|
* Work around this here.
|
|
|
|
*
|
|
|
|
* Do an PCI bus scan by hand because we're running before the PCI
|
|
|
|
* subsystem.
|
|
|
|
*
|
|
|
|
* All K8 AGP bridges are AGPv3 compliant, so we can do this scan
|
|
|
|
* generically. It's probably overkill to always scan all slots because
|
|
|
|
* the AGP bridges should be always an own bus on the HT hierarchy,
|
|
|
|
* but do it here for future safety.
|
|
|
|
*/
|
2008-05-27 10:03:56 -06:00
|
|
|
static u32 __init search_agp_bridge(u32 *order, int *valid_agp)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2008-04-19 02:31:11 -06:00
|
|
|
int bus, slot, func;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
/* Poor man's PCI discovery */
|
2008-04-19 02:31:11 -06:00
|
|
|
for (bus = 0; bus < 256; bus++) {
|
2008-01-30 05:30:09 -07:00
|
|
|
for (slot = 0; slot < 32; slot++) {
|
|
|
|
for (func = 0; func < 8; func++) {
|
2005-04-16 16:20:36 -06:00
|
|
|
u32 class, cap;
|
|
|
|
u8 type;
|
2008-04-19 02:31:11 -06:00
|
|
|
class = read_pci_config(bus, slot, func,
|
2005-04-16 16:20:36 -06:00
|
|
|
PCI_CLASS_REVISION);
|
|
|
|
if (class == 0xffffffff)
|
2008-01-30 05:30:09 -07:00
|
|
|
break;
|
|
|
|
|
|
|
|
switch (class >> 16) {
|
2005-04-16 16:20:36 -06:00
|
|
|
case PCI_CLASS_BRIDGE_HOST:
|
|
|
|
case PCI_CLASS_BRIDGE_OTHER: /* needed? */
|
|
|
|
/* AGP bridge? */
|
2008-04-19 02:31:11 -06:00
|
|
|
cap = find_cap(bus, slot, func,
|
2008-01-30 05:30:09 -07:00
|
|
|
PCI_CAP_ID_AGP);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (!cap)
|
|
|
|
break;
|
2008-01-30 05:30:09 -07:00
|
|
|
*valid_agp = 1;
|
2008-04-19 02:31:11 -06:00
|
|
|
return read_agp(bus, slot, func, cap,
|
2008-01-30 05:30:09 -07:00
|
|
|
order);
|
|
|
|
}
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
/* No multi-function device? */
|
2008-04-19 02:31:11 -06:00
|
|
|
type = read_pci_config_byte(bus, slot, func,
|
2005-04-16 16:20:36 -06:00
|
|
|
PCI_HEADER_TYPE);
|
|
|
|
if (!(type & 0x80))
|
|
|
|
break;
|
2008-01-30 05:30:09 -07:00
|
|
|
}
|
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
2008-01-30 05:30:10 -07:00
|
|
|
printk(KERN_INFO "No AGP bridge found\n");
|
2008-01-30 05:30:09 -07:00
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
x86: disable the GART early, 64-bit
For K8 system: 4G RAM with memory hole remapping enabled, or more than
4G RAM installed.
when try to use kexec second kernel, and the first doesn't include
gart_shutdown. the second kernel could have different aper position than
the first kernel. and second kernel could use that hole as RAM that is
still used by GART set by the first kernel. esp. when try to kexec
2.6.24 with sparse mem enable from previous kernel (from RHEL 5 or SLES
10). the new kernel will use aper by GART (set by first kernel) for
vmemmap. and after new kernel setting one new GART. the position will be
real RAM. the _mapcount set is lost.
Bad page state in process 'swapper'
page:ffffe2000e600020 flags:0x0000000000000000 mapping:0000000000000000 mapcount:1 count:0
Trying to fix it up, but a reboot is needed
Backtrace:
Pid: 0, comm: swapper Not tainted 2.6.24-rc7-smp-gcdf71a10-dirty #13
Call Trace:
[<ffffffff8026401f>] bad_page+0x63/0x8d
[<ffffffff80264169>] __free_pages_ok+0x7c/0x2a5
[<ffffffff80ba75d1>] free_all_bootmem_core+0xd0/0x198
[<ffffffff80ba3a42>] numa_free_all_bootmem+0x3b/0x76
[<ffffffff80ba3461>] mem_init+0x3b/0x152
[<ffffffff80b959d3>] start_kernel+0x236/0x2c2
[<ffffffff80b9511a>] _sinittext+0x11a/0x121
and
[ffffe2000e600000-ffffe2000e7fffff] PMD ->ffff81001c200000 on node 0
phys addr is : 0x1c200000
RHEL 5.1 kernel -53 said:
PCI-DMA: aperture base @ 1c000000 size 65536 KB
new kernel said:
Mapping aperture over 65536 KB of RAM @ 3c000000
So could try to disable that GART if possible.
According to Ingo
> hm, i'm wondering, instead of modifying the GART, why dont we simply
> _detect_ whatever GART settings we have inherited, and propagate that
> into our e820 maps? I.e. if there's inconsistency, then punch that out
> from the memory maps and just dont use that memory.
>
> that way it would not matter whether the GART settings came from a [old
> or crashing] Linux kernel that has not called gart_iommu_shutdown(), or
> whether it's a BIOS that has set up an aperture hole inconsistent with
> the memory map it passed. (or the memory map we _think_ i tried to pass
> us)
>
> it would also be more robust to only read and do a memory map quirk
> based on that, than actively trying to change the GART so early in the
> bootup. Later on we have to re-enable the GART _anyway_ and have to
> punch a hole for it.
>
> and as a bonus, we would have shored up our defenses against crappy
> BIOSes as well.
add e820 modification for gart inconsistent setting.
gart_fix_e820=off could be used to disable e820 fix.
Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 05:33:09 -07:00
|
|
|
static int gart_fix_e820 __initdata = 1;
|
|
|
|
|
|
|
|
static int __init parse_gart_mem(char *p)
|
|
|
|
{
|
|
|
|
if (!p)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
if (!strncmp(p, "off", 3))
|
|
|
|
gart_fix_e820 = 0;
|
|
|
|
else if (!strncmp(p, "on", 2))
|
|
|
|
gart_fix_e820 = 1;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
early_param("gart_fix_e820", parse_gart_mem);
|
|
|
|
|
|
|
|
void __init early_gart_iommu_check(void)
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
* in case it is enabled before, esp for kexec/kdump,
|
|
|
|
* previous kernel already enable that. memset called
|
|
|
|
* by allocate_aperture/__alloc_bootmem_nopanic cause restart.
|
|
|
|
* or second kernel have different position for GART hole. and new
|
|
|
|
* kernel could use hole as RAM that is still used by GART set by
|
|
|
|
* first kernel
|
|
|
|
* or BIOS forget to put that in reserved.
|
|
|
|
* try to update e820 to make that region as reserved.
|
|
|
|
*/
|
aperture_64.c: duplicated code, buggy?
Hi!
void __init early_gart_iommu_check(void)
contains
for (num = 24; num < 32; num++) {
if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00)))
continue;
loop, with very similar loop duplicated in
void __init gart_iommu_hole_init(void)
. First copy of a loop seems to be buggy, too. It uses 0 as a "nothing
set" value, which may actually bite us in last_aper_enabled case
(because it may be often zero).
(Beware, it is hard to test this patch, because this code has about
2^8 different code paths, depending on hardware and cmdline settings).
Plus, the second loop does not check for consistency of
aper_enabled. Should it?
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-05-26 12:40:47 -06:00
|
|
|
int i, fix, slot;
|
x86: disable the GART early, 64-bit
For K8 system: 4G RAM with memory hole remapping enabled, or more than
4G RAM installed.
when try to use kexec second kernel, and the first doesn't include
gart_shutdown. the second kernel could have different aper position than
the first kernel. and second kernel could use that hole as RAM that is
still used by GART set by the first kernel. esp. when try to kexec
2.6.24 with sparse mem enable from previous kernel (from RHEL 5 or SLES
10). the new kernel will use aper by GART (set by first kernel) for
vmemmap. and after new kernel setting one new GART. the position will be
real RAM. the _mapcount set is lost.
Bad page state in process 'swapper'
page:ffffe2000e600020 flags:0x0000000000000000 mapping:0000000000000000 mapcount:1 count:0
Trying to fix it up, but a reboot is needed
Backtrace:
Pid: 0, comm: swapper Not tainted 2.6.24-rc7-smp-gcdf71a10-dirty #13
Call Trace:
[<ffffffff8026401f>] bad_page+0x63/0x8d
[<ffffffff80264169>] __free_pages_ok+0x7c/0x2a5
[<ffffffff80ba75d1>] free_all_bootmem_core+0xd0/0x198
[<ffffffff80ba3a42>] numa_free_all_bootmem+0x3b/0x76
[<ffffffff80ba3461>] mem_init+0x3b/0x152
[<ffffffff80b959d3>] start_kernel+0x236/0x2c2
[<ffffffff80b9511a>] _sinittext+0x11a/0x121
and
[ffffe2000e600000-ffffe2000e7fffff] PMD ->ffff81001c200000 on node 0
phys addr is : 0x1c200000
RHEL 5.1 kernel -53 said:
PCI-DMA: aperture base @ 1c000000 size 65536 KB
new kernel said:
Mapping aperture over 65536 KB of RAM @ 3c000000
So could try to disable that GART if possible.
According to Ingo
> hm, i'm wondering, instead of modifying the GART, why dont we simply
> _detect_ whatever GART settings we have inherited, and propagate that
> into our e820 maps? I.e. if there's inconsistency, then punch that out
> from the memory maps and just dont use that memory.
>
> that way it would not matter whether the GART settings came from a [old
> or crashing] Linux kernel that has not called gart_iommu_shutdown(), or
> whether it's a BIOS that has set up an aperture hole inconsistent with
> the memory map it passed. (or the memory map we _think_ i tried to pass
> us)
>
> it would also be more robust to only read and do a memory map quirk
> based on that, than actively trying to change the GART so early in the
> bootup. Later on we have to re-enable the GART _anyway_ and have to
> punch a hole for it.
>
> and as a bonus, we would have shored up our defenses against crappy
> BIOSes as well.
add e820 modification for gart inconsistent setting.
gart_fix_e820=off could be used to disable e820 fix.
Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 05:33:09 -07:00
|
|
|
u32 ctl;
|
|
|
|
u32 aper_size = 0, aper_order = 0, last_aper_order = 0;
|
|
|
|
u64 aper_base = 0, last_aper_base = 0;
|
aperture_64.c: duplicated code, buggy?
Hi!
void __init early_gart_iommu_check(void)
contains
for (num = 24; num < 32; num++) {
if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00)))
continue;
loop, with very similar loop duplicated in
void __init gart_iommu_hole_init(void)
. First copy of a loop seems to be buggy, too. It uses 0 as a "nothing
set" value, which may actually bite us in last_aper_enabled case
(because it may be often zero).
(Beware, it is hard to test this patch, because this code has about
2^8 different code paths, depending on hardware and cmdline settings).
Plus, the second loop does not check for consistency of
aper_enabled. Should it?
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-05-26 12:40:47 -06:00
|
|
|
int aper_enabled = 0, last_aper_enabled = 0, last_valid = 0;
|
x86: disable the GART early, 64-bit
For K8 system: 4G RAM with memory hole remapping enabled, or more than
4G RAM installed.
when try to use kexec second kernel, and the first doesn't include
gart_shutdown. the second kernel could have different aper position than
the first kernel. and second kernel could use that hole as RAM that is
still used by GART set by the first kernel. esp. when try to kexec
2.6.24 with sparse mem enable from previous kernel (from RHEL 5 or SLES
10). the new kernel will use aper by GART (set by first kernel) for
vmemmap. and after new kernel setting one new GART. the position will be
real RAM. the _mapcount set is lost.
Bad page state in process 'swapper'
page:ffffe2000e600020 flags:0x0000000000000000 mapping:0000000000000000 mapcount:1 count:0
Trying to fix it up, but a reboot is needed
Backtrace:
Pid: 0, comm: swapper Not tainted 2.6.24-rc7-smp-gcdf71a10-dirty #13
Call Trace:
[<ffffffff8026401f>] bad_page+0x63/0x8d
[<ffffffff80264169>] __free_pages_ok+0x7c/0x2a5
[<ffffffff80ba75d1>] free_all_bootmem_core+0xd0/0x198
[<ffffffff80ba3a42>] numa_free_all_bootmem+0x3b/0x76
[<ffffffff80ba3461>] mem_init+0x3b/0x152
[<ffffffff80b959d3>] start_kernel+0x236/0x2c2
[<ffffffff80b9511a>] _sinittext+0x11a/0x121
and
[ffffe2000e600000-ffffe2000e7fffff] PMD ->ffff81001c200000 on node 0
phys addr is : 0x1c200000
RHEL 5.1 kernel -53 said:
PCI-DMA: aperture base @ 1c000000 size 65536 KB
new kernel said:
Mapping aperture over 65536 KB of RAM @ 3c000000
So could try to disable that GART if possible.
According to Ingo
> hm, i'm wondering, instead of modifying the GART, why dont we simply
> _detect_ whatever GART settings we have inherited, and propagate that
> into our e820 maps? I.e. if there's inconsistency, then punch that out
> from the memory maps and just dont use that memory.
>
> that way it would not matter whether the GART settings came from a [old
> or crashing] Linux kernel that has not called gart_iommu_shutdown(), or
> whether it's a BIOS that has set up an aperture hole inconsistent with
> the memory map it passed. (or the memory map we _think_ i tried to pass
> us)
>
> it would also be more robust to only read and do a memory map quirk
> based on that, than actively trying to change the GART so early in the
> bootup. Later on we have to re-enable the GART _anyway_ and have to
> punch a hole for it.
>
> and as a bonus, we would have shored up our defenses against crappy
> BIOSes as well.
add e820 modification for gart inconsistent setting.
gart_fix_e820=off could be used to disable e820 fix.
Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 05:33:09 -07:00
|
|
|
|
|
|
|
if (!early_pci_allowed())
|
|
|
|
return;
|
|
|
|
|
aperture_64.c: duplicated code, buggy?
Hi!
void __init early_gart_iommu_check(void)
contains
for (num = 24; num < 32; num++) {
if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00)))
continue;
loop, with very similar loop duplicated in
void __init gart_iommu_hole_init(void)
. First copy of a loop seems to be buggy, too. It uses 0 as a "nothing
set" value, which may actually bite us in last_aper_enabled case
(because it may be often zero).
(Beware, it is hard to test this patch, because this code has about
2^8 different code paths, depending on hardware and cmdline settings).
Plus, the second loop does not check for consistency of
aper_enabled. Should it?
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-05-26 12:40:47 -06:00
|
|
|
/* This is mostly duplicate of iommu_hole_init */
|
x86: disable the GART early, 64-bit
For K8 system: 4G RAM with memory hole remapping enabled, or more than
4G RAM installed.
when try to use kexec second kernel, and the first doesn't include
gart_shutdown. the second kernel could have different aper position than
the first kernel. and second kernel could use that hole as RAM that is
still used by GART set by the first kernel. esp. when try to kexec
2.6.24 with sparse mem enable from previous kernel (from RHEL 5 or SLES
10). the new kernel will use aper by GART (set by first kernel) for
vmemmap. and after new kernel setting one new GART. the position will be
real RAM. the _mapcount set is lost.
Bad page state in process 'swapper'
page:ffffe2000e600020 flags:0x0000000000000000 mapping:0000000000000000 mapcount:1 count:0
Trying to fix it up, but a reboot is needed
Backtrace:
Pid: 0, comm: swapper Not tainted 2.6.24-rc7-smp-gcdf71a10-dirty #13
Call Trace:
[<ffffffff8026401f>] bad_page+0x63/0x8d
[<ffffffff80264169>] __free_pages_ok+0x7c/0x2a5
[<ffffffff80ba75d1>] free_all_bootmem_core+0xd0/0x198
[<ffffffff80ba3a42>] numa_free_all_bootmem+0x3b/0x76
[<ffffffff80ba3461>] mem_init+0x3b/0x152
[<ffffffff80b959d3>] start_kernel+0x236/0x2c2
[<ffffffff80b9511a>] _sinittext+0x11a/0x121
and
[ffffe2000e600000-ffffe2000e7fffff] PMD ->ffff81001c200000 on node 0
phys addr is : 0x1c200000
RHEL 5.1 kernel -53 said:
PCI-DMA: aperture base @ 1c000000 size 65536 KB
new kernel said:
Mapping aperture over 65536 KB of RAM @ 3c000000
So could try to disable that GART if possible.
According to Ingo
> hm, i'm wondering, instead of modifying the GART, why dont we simply
> _detect_ whatever GART settings we have inherited, and propagate that
> into our e820 maps? I.e. if there's inconsistency, then punch that out
> from the memory maps and just dont use that memory.
>
> that way it would not matter whether the GART settings came from a [old
> or crashing] Linux kernel that has not called gart_iommu_shutdown(), or
> whether it's a BIOS that has set up an aperture hole inconsistent with
> the memory map it passed. (or the memory map we _think_ i tried to pass
> us)
>
> it would also be more robust to only read and do a memory map quirk
> based on that, than actively trying to change the GART so early in the
> bootup. Later on we have to re-enable the GART _anyway_ and have to
> punch a hole for it.
>
> and as a bonus, we would have shored up our defenses against crappy
> BIOSes as well.
add e820 modification for gart inconsistent setting.
gart_fix_e820=off could be used to disable e820 fix.
Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 05:33:09 -07:00
|
|
|
fix = 0;
|
2008-04-19 02:31:11 -06:00
|
|
|
for (i = 0; i < ARRAY_SIZE(bus_dev_ranges); i++) {
|
|
|
|
int bus;
|
|
|
|
int dev_base, dev_limit;
|
|
|
|
|
|
|
|
bus = bus_dev_ranges[i].bus;
|
|
|
|
dev_base = bus_dev_ranges[i].dev_base;
|
|
|
|
dev_limit = bus_dev_ranges[i].dev_limit;
|
|
|
|
|
|
|
|
for (slot = dev_base; slot < dev_limit; slot++) {
|
|
|
|
if (!early_is_k8_nb(read_pci_config(bus, slot, 3, 0x00)))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
ctl = read_pci_config(bus, slot, 3, AMD64_GARTAPERTURECTL);
|
|
|
|
aper_enabled = ctl & AMD64_GARTEN;
|
|
|
|
aper_order = (ctl >> 1) & 7;
|
|
|
|
aper_size = (32 * 1024 * 1024) << aper_order;
|
|
|
|
aper_base = read_pci_config(bus, slot, 3, AMD64_GARTAPERTUREBASE) & 0x7fff;
|
|
|
|
aper_base <<= 25;
|
|
|
|
|
aperture_64.c: duplicated code, buggy?
Hi!
void __init early_gart_iommu_check(void)
contains
for (num = 24; num < 32; num++) {
if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00)))
continue;
loop, with very similar loop duplicated in
void __init gart_iommu_hole_init(void)
. First copy of a loop seems to be buggy, too. It uses 0 as a "nothing
set" value, which may actually bite us in last_aper_enabled case
(because it may be often zero).
(Beware, it is hard to test this patch, because this code has about
2^8 different code paths, depending on hardware and cmdline settings).
Plus, the second loop does not check for consistency of
aper_enabled. Should it?
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-05-26 12:40:47 -06:00
|
|
|
if (last_valid) {
|
|
|
|
if ((aper_order != last_aper_order) ||
|
|
|
|
(aper_base != last_aper_base) ||
|
|
|
|
(aper_enabled != last_aper_enabled)) {
|
|
|
|
fix = 1;
|
|
|
|
break;
|
|
|
|
}
|
2008-04-19 02:31:11 -06:00
|
|
|
}
|
aperture_64.c: duplicated code, buggy?
Hi!
void __init early_gart_iommu_check(void)
contains
for (num = 24; num < 32; num++) {
if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00)))
continue;
loop, with very similar loop duplicated in
void __init gart_iommu_hole_init(void)
. First copy of a loop seems to be buggy, too. It uses 0 as a "nothing
set" value, which may actually bite us in last_aper_enabled case
(because it may be often zero).
(Beware, it is hard to test this patch, because this code has about
2^8 different code paths, depending on hardware and cmdline settings).
Plus, the second loop does not check for consistency of
aper_enabled. Should it?
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-05-26 12:40:47 -06:00
|
|
|
|
2008-04-19 02:31:11 -06:00
|
|
|
last_aper_order = aper_order;
|
|
|
|
last_aper_base = aper_base;
|
|
|
|
last_aper_enabled = aper_enabled;
|
aperture_64.c: duplicated code, buggy?
Hi!
void __init early_gart_iommu_check(void)
contains
for (num = 24; num < 32; num++) {
if (!early_is_k8_nb(read_pci_config(0, num, 3, 0x00)))
continue;
loop, with very similar loop duplicated in
void __init gart_iommu_hole_init(void)
. First copy of a loop seems to be buggy, too. It uses 0 as a "nothing
set" value, which may actually bite us in last_aper_enabled case
(because it may be often zero).
(Beware, it is hard to test this patch, because this code has about
2^8 different code paths, depending on hardware and cmdline settings).
Plus, the second loop does not check for consistency of
aper_enabled. Should it?
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-05-26 12:40:47 -06:00
|
|
|
last_valid = 1;
|
x86: disable the GART early, 64-bit
For K8 system: 4G RAM with memory hole remapping enabled, or more than
4G RAM installed.
when try to use kexec second kernel, and the first doesn't include
gart_shutdown. the second kernel could have different aper position than
the first kernel. and second kernel could use that hole as RAM that is
still used by GART set by the first kernel. esp. when try to kexec
2.6.24 with sparse mem enable from previous kernel (from RHEL 5 or SLES
10). the new kernel will use aper by GART (set by first kernel) for
vmemmap. and after new kernel setting one new GART. the position will be
real RAM. the _mapcount set is lost.
Bad page state in process 'swapper'
page:ffffe2000e600020 flags:0x0000000000000000 mapping:0000000000000000 mapcount:1 count:0
Trying to fix it up, but a reboot is needed
Backtrace:
Pid: 0, comm: swapper Not tainted 2.6.24-rc7-smp-gcdf71a10-dirty #13
Call Trace:
[<ffffffff8026401f>] bad_page+0x63/0x8d
[<ffffffff80264169>] __free_pages_ok+0x7c/0x2a5
[<ffffffff80ba75d1>] free_all_bootmem_core+0xd0/0x198
[<ffffffff80ba3a42>] numa_free_all_bootmem+0x3b/0x76
[<ffffffff80ba3461>] mem_init+0x3b/0x152
[<ffffffff80b959d3>] start_kernel+0x236/0x2c2
[<ffffffff80b9511a>] _sinittext+0x11a/0x121
and
[ffffe2000e600000-ffffe2000e7fffff] PMD ->ffff81001c200000 on node 0
phys addr is : 0x1c200000
RHEL 5.1 kernel -53 said:
PCI-DMA: aperture base @ 1c000000 size 65536 KB
new kernel said:
Mapping aperture over 65536 KB of RAM @ 3c000000
So could try to disable that GART if possible.
According to Ingo
> hm, i'm wondering, instead of modifying the GART, why dont we simply
> _detect_ whatever GART settings we have inherited, and propagate that
> into our e820 maps? I.e. if there's inconsistency, then punch that out
> from the memory maps and just dont use that memory.
>
> that way it would not matter whether the GART settings came from a [old
> or crashing] Linux kernel that has not called gart_iommu_shutdown(), or
> whether it's a BIOS that has set up an aperture hole inconsistent with
> the memory map it passed. (or the memory map we _think_ i tried to pass
> us)
>
> it would also be more robust to only read and do a memory map quirk
> based on that, than actively trying to change the GART so early in the
> bootup. Later on we have to re-enable the GART _anyway_ and have to
> punch a hole for it.
>
> and as a bonus, we would have shored up our defenses against crappy
> BIOSes as well.
add e820 modification for gart inconsistent setting.
gart_fix_e820=off could be used to disable e820 fix.
Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 05:33:09 -07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!fix && !aper_enabled)
|
|
|
|
return;
|
|
|
|
|
|
|
|
if (!aper_base || !aper_size || aper_base + aper_size > 0x100000000UL)
|
|
|
|
fix = 1;
|
|
|
|
|
|
|
|
if (gart_fix_e820 && !fix && aper_enabled) {
|
2008-06-21 04:50:47 -06:00
|
|
|
if (e820_any_mapped(aper_base, aper_base + aper_size,
|
|
|
|
E820_RAM)) {
|
2008-05-20 08:27:17 -06:00
|
|
|
/* reserve it, so we can reuse it in second kernel */
|
x86: disable the GART early, 64-bit
For K8 system: 4G RAM with memory hole remapping enabled, or more than
4G RAM installed.
when try to use kexec second kernel, and the first doesn't include
gart_shutdown. the second kernel could have different aper position than
the first kernel. and second kernel could use that hole as RAM that is
still used by GART set by the first kernel. esp. when try to kexec
2.6.24 with sparse mem enable from previous kernel (from RHEL 5 or SLES
10). the new kernel will use aper by GART (set by first kernel) for
vmemmap. and after new kernel setting one new GART. the position will be
real RAM. the _mapcount set is lost.
Bad page state in process 'swapper'
page:ffffe2000e600020 flags:0x0000000000000000 mapping:0000000000000000 mapcount:1 count:0
Trying to fix it up, but a reboot is needed
Backtrace:
Pid: 0, comm: swapper Not tainted 2.6.24-rc7-smp-gcdf71a10-dirty #13
Call Trace:
[<ffffffff8026401f>] bad_page+0x63/0x8d
[<ffffffff80264169>] __free_pages_ok+0x7c/0x2a5
[<ffffffff80ba75d1>] free_all_bootmem_core+0xd0/0x198
[<ffffffff80ba3a42>] numa_free_all_bootmem+0x3b/0x76
[<ffffffff80ba3461>] mem_init+0x3b/0x152
[<ffffffff80b959d3>] start_kernel+0x236/0x2c2
[<ffffffff80b9511a>] _sinittext+0x11a/0x121
and
[ffffe2000e600000-ffffe2000e7fffff] PMD ->ffff81001c200000 on node 0
phys addr is : 0x1c200000
RHEL 5.1 kernel -53 said:
PCI-DMA: aperture base @ 1c000000 size 65536 KB
new kernel said:
Mapping aperture over 65536 KB of RAM @ 3c000000
So could try to disable that GART if possible.
According to Ingo
> hm, i'm wondering, instead of modifying the GART, why dont we simply
> _detect_ whatever GART settings we have inherited, and propagate that
> into our e820 maps? I.e. if there's inconsistency, then punch that out
> from the memory maps and just dont use that memory.
>
> that way it would not matter whether the GART settings came from a [old
> or crashing] Linux kernel that has not called gart_iommu_shutdown(), or
> whether it's a BIOS that has set up an aperture hole inconsistent with
> the memory map it passed. (or the memory map we _think_ i tried to pass
> us)
>
> it would also be more robust to only read and do a memory map quirk
> based on that, than actively trying to change the GART so early in the
> bootup. Later on we have to re-enable the GART _anyway_ and have to
> punch a hole for it.
>
> and as a bonus, we would have shored up our defenses against crappy
> BIOSes as well.
add e820 modification for gart inconsistent setting.
gart_fix_e820=off could be used to disable e820 fix.
Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 05:33:09 -07:00
|
|
|
printk(KERN_INFO "update e820 for GART\n");
|
2008-06-15 19:58:51 -06:00
|
|
|
e820_add_region(aper_base, aper_size, E820_RESERVED);
|
x86: disable the GART early, 64-bit
For K8 system: 4G RAM with memory hole remapping enabled, or more than
4G RAM installed.
when try to use kexec second kernel, and the first doesn't include
gart_shutdown. the second kernel could have different aper position than
the first kernel. and second kernel could use that hole as RAM that is
still used by GART set by the first kernel. esp. when try to kexec
2.6.24 with sparse mem enable from previous kernel (from RHEL 5 or SLES
10). the new kernel will use aper by GART (set by first kernel) for
vmemmap. and after new kernel setting one new GART. the position will be
real RAM. the _mapcount set is lost.
Bad page state in process 'swapper'
page:ffffe2000e600020 flags:0x0000000000000000 mapping:0000000000000000 mapcount:1 count:0
Trying to fix it up, but a reboot is needed
Backtrace:
Pid: 0, comm: swapper Not tainted 2.6.24-rc7-smp-gcdf71a10-dirty #13
Call Trace:
[<ffffffff8026401f>] bad_page+0x63/0x8d
[<ffffffff80264169>] __free_pages_ok+0x7c/0x2a5
[<ffffffff80ba75d1>] free_all_bootmem_core+0xd0/0x198
[<ffffffff80ba3a42>] numa_free_all_bootmem+0x3b/0x76
[<ffffffff80ba3461>] mem_init+0x3b/0x152
[<ffffffff80b959d3>] start_kernel+0x236/0x2c2
[<ffffffff80b9511a>] _sinittext+0x11a/0x121
and
[ffffe2000e600000-ffffe2000e7fffff] PMD ->ffff81001c200000 on node 0
phys addr is : 0x1c200000
RHEL 5.1 kernel -53 said:
PCI-DMA: aperture base @ 1c000000 size 65536 KB
new kernel said:
Mapping aperture over 65536 KB of RAM @ 3c000000
So could try to disable that GART if possible.
According to Ingo
> hm, i'm wondering, instead of modifying the GART, why dont we simply
> _detect_ whatever GART settings we have inherited, and propagate that
> into our e820 maps? I.e. if there's inconsistency, then punch that out
> from the memory maps and just dont use that memory.
>
> that way it would not matter whether the GART settings came from a [old
> or crashing] Linux kernel that has not called gart_iommu_shutdown(), or
> whether it's a BIOS that has set up an aperture hole inconsistent with
> the memory map it passed. (or the memory map we _think_ i tried to pass
> us)
>
> it would also be more robust to only read and do a memory map quirk
> based on that, than actively trying to change the GART so early in the
> bootup. Later on we have to re-enable the GART _anyway_ and have to
> punch a hole for it.
>
> and as a bonus, we would have shored up our defenses against crappy
> BIOSes as well.
add e820 modification for gart inconsistent setting.
gart_fix_e820=off could be used to disable e820 fix.
Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 05:33:09 -07:00
|
|
|
update_e820();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-05-26 13:17:30 -06:00
|
|
|
if (!fix)
|
|
|
|
return;
|
|
|
|
|
x86: disable the GART early, 64-bit
For K8 system: 4G RAM with memory hole remapping enabled, or more than
4G RAM installed.
when try to use kexec second kernel, and the first doesn't include
gart_shutdown. the second kernel could have different aper position than
the first kernel. and second kernel could use that hole as RAM that is
still used by GART set by the first kernel. esp. when try to kexec
2.6.24 with sparse mem enable from previous kernel (from RHEL 5 or SLES
10). the new kernel will use aper by GART (set by first kernel) for
vmemmap. and after new kernel setting one new GART. the position will be
real RAM. the _mapcount set is lost.
Bad page state in process 'swapper'
page:ffffe2000e600020 flags:0x0000000000000000 mapping:0000000000000000 mapcount:1 count:0
Trying to fix it up, but a reboot is needed
Backtrace:
Pid: 0, comm: swapper Not tainted 2.6.24-rc7-smp-gcdf71a10-dirty #13
Call Trace:
[<ffffffff8026401f>] bad_page+0x63/0x8d
[<ffffffff80264169>] __free_pages_ok+0x7c/0x2a5
[<ffffffff80ba75d1>] free_all_bootmem_core+0xd0/0x198
[<ffffffff80ba3a42>] numa_free_all_bootmem+0x3b/0x76
[<ffffffff80ba3461>] mem_init+0x3b/0x152
[<ffffffff80b959d3>] start_kernel+0x236/0x2c2
[<ffffffff80b9511a>] _sinittext+0x11a/0x121
and
[ffffe2000e600000-ffffe2000e7fffff] PMD ->ffff81001c200000 on node 0
phys addr is : 0x1c200000
RHEL 5.1 kernel -53 said:
PCI-DMA: aperture base @ 1c000000 size 65536 KB
new kernel said:
Mapping aperture over 65536 KB of RAM @ 3c000000
So could try to disable that GART if possible.
According to Ingo
> hm, i'm wondering, instead of modifying the GART, why dont we simply
> _detect_ whatever GART settings we have inherited, and propagate that
> into our e820 maps? I.e. if there's inconsistency, then punch that out
> from the memory maps and just dont use that memory.
>
> that way it would not matter whether the GART settings came from a [old
> or crashing] Linux kernel that has not called gart_iommu_shutdown(), or
> whether it's a BIOS that has set up an aperture hole inconsistent with
> the memory map it passed. (or the memory map we _think_ i tried to pass
> us)
>
> it would also be more robust to only read and do a memory map quirk
> based on that, than actively trying to change the GART so early in the
> bootup. Later on we have to re-enable the GART _anyway_ and have to
> punch a hole for it.
>
> and as a bonus, we would have shored up our defenses against crappy
> BIOSes as well.
add e820 modification for gart inconsistent setting.
gart_fix_e820=off could be used to disable e820 fix.
Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 05:33:09 -07:00
|
|
|
/* different nodes have different setting, disable them all at first*/
|
2008-04-19 02:31:11 -06:00
|
|
|
for (i = 0; i < ARRAY_SIZE(bus_dev_ranges); i++) {
|
|
|
|
int bus;
|
|
|
|
int dev_base, dev_limit;
|
|
|
|
|
|
|
|
bus = bus_dev_ranges[i].bus;
|
|
|
|
dev_base = bus_dev_ranges[i].dev_base;
|
|
|
|
dev_limit = bus_dev_ranges[i].dev_limit;
|
x86: disable the GART early, 64-bit
For K8 system: 4G RAM with memory hole remapping enabled, or more than
4G RAM installed.
when try to use kexec second kernel, and the first doesn't include
gart_shutdown. the second kernel could have different aper position than
the first kernel. and second kernel could use that hole as RAM that is
still used by GART set by the first kernel. esp. when try to kexec
2.6.24 with sparse mem enable from previous kernel (from RHEL 5 or SLES
10). the new kernel will use aper by GART (set by first kernel) for
vmemmap. and after new kernel setting one new GART. the position will be
real RAM. the _mapcount set is lost.
Bad page state in process 'swapper'
page:ffffe2000e600020 flags:0x0000000000000000 mapping:0000000000000000 mapcount:1 count:0
Trying to fix it up, but a reboot is needed
Backtrace:
Pid: 0, comm: swapper Not tainted 2.6.24-rc7-smp-gcdf71a10-dirty #13
Call Trace:
[<ffffffff8026401f>] bad_page+0x63/0x8d
[<ffffffff80264169>] __free_pages_ok+0x7c/0x2a5
[<ffffffff80ba75d1>] free_all_bootmem_core+0xd0/0x198
[<ffffffff80ba3a42>] numa_free_all_bootmem+0x3b/0x76
[<ffffffff80ba3461>] mem_init+0x3b/0x152
[<ffffffff80b959d3>] start_kernel+0x236/0x2c2
[<ffffffff80b9511a>] _sinittext+0x11a/0x121
and
[ffffe2000e600000-ffffe2000e7fffff] PMD ->ffff81001c200000 on node 0
phys addr is : 0x1c200000
RHEL 5.1 kernel -53 said:
PCI-DMA: aperture base @ 1c000000 size 65536 KB
new kernel said:
Mapping aperture over 65536 KB of RAM @ 3c000000
So could try to disable that GART if possible.
According to Ingo
> hm, i'm wondering, instead of modifying the GART, why dont we simply
> _detect_ whatever GART settings we have inherited, and propagate that
> into our e820 maps? I.e. if there's inconsistency, then punch that out
> from the memory maps and just dont use that memory.
>
> that way it would not matter whether the GART settings came from a [old
> or crashing] Linux kernel that has not called gart_iommu_shutdown(), or
> whether it's a BIOS that has set up an aperture hole inconsistent with
> the memory map it passed. (or the memory map we _think_ i tried to pass
> us)
>
> it would also be more robust to only read and do a memory map quirk
> based on that, than actively trying to change the GART so early in the
> bootup. Later on we have to re-enable the GART _anyway_ and have to
> punch a hole for it.
>
> and as a bonus, we would have shored up our defenses against crappy
> BIOSes as well.
add e820 modification for gart inconsistent setting.
gart_fix_e820=off could be used to disable e820 fix.
Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 05:33:09 -07:00
|
|
|
|
2008-04-19 02:31:11 -06:00
|
|
|
for (slot = dev_base; slot < dev_limit; slot++) {
|
|
|
|
if (!early_is_k8_nb(read_pci_config(bus, slot, 3, 0x00)))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
ctl = read_pci_config(bus, slot, 3, AMD64_GARTAPERTURECTL);
|
|
|
|
ctl &= ~AMD64_GARTEN;
|
|
|
|
write_pci_config(bus, slot, 3, AMD64_GARTAPERTURECTL, ctl);
|
|
|
|
}
|
x86: disable the GART early, 64-bit
For K8 system: 4G RAM with memory hole remapping enabled, or more than
4G RAM installed.
when try to use kexec second kernel, and the first doesn't include
gart_shutdown. the second kernel could have different aper position than
the first kernel. and second kernel could use that hole as RAM that is
still used by GART set by the first kernel. esp. when try to kexec
2.6.24 with sparse mem enable from previous kernel (from RHEL 5 or SLES
10). the new kernel will use aper by GART (set by first kernel) for
vmemmap. and after new kernel setting one new GART. the position will be
real RAM. the _mapcount set is lost.
Bad page state in process 'swapper'
page:ffffe2000e600020 flags:0x0000000000000000 mapping:0000000000000000 mapcount:1 count:0
Trying to fix it up, but a reboot is needed
Backtrace:
Pid: 0, comm: swapper Not tainted 2.6.24-rc7-smp-gcdf71a10-dirty #13
Call Trace:
[<ffffffff8026401f>] bad_page+0x63/0x8d
[<ffffffff80264169>] __free_pages_ok+0x7c/0x2a5
[<ffffffff80ba75d1>] free_all_bootmem_core+0xd0/0x198
[<ffffffff80ba3a42>] numa_free_all_bootmem+0x3b/0x76
[<ffffffff80ba3461>] mem_init+0x3b/0x152
[<ffffffff80b959d3>] start_kernel+0x236/0x2c2
[<ffffffff80b9511a>] _sinittext+0x11a/0x121
and
[ffffe2000e600000-ffffe2000e7fffff] PMD ->ffff81001c200000 on node 0
phys addr is : 0x1c200000
RHEL 5.1 kernel -53 said:
PCI-DMA: aperture base @ 1c000000 size 65536 KB
new kernel said:
Mapping aperture over 65536 KB of RAM @ 3c000000
So could try to disable that GART if possible.
According to Ingo
> hm, i'm wondering, instead of modifying the GART, why dont we simply
> _detect_ whatever GART settings we have inherited, and propagate that
> into our e820 maps? I.e. if there's inconsistency, then punch that out
> from the memory maps and just dont use that memory.
>
> that way it would not matter whether the GART settings came from a [old
> or crashing] Linux kernel that has not called gart_iommu_shutdown(), or
> whether it's a BIOS that has set up an aperture hole inconsistent with
> the memory map it passed. (or the memory map we _think_ i tried to pass
> us)
>
> it would also be more robust to only read and do a memory map quirk
> based on that, than actively trying to change the GART so early in the
> bootup. Later on we have to re-enable the GART _anyway_ and have to
> punch a hole for it.
>
> and as a bonus, we would have shored up our defenses against crappy
> BIOSes as well.
add e820 modification for gart inconsistent setting.
gart_fix_e820=off could be used to disable e820 fix.
Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 05:33:09 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
}
|
|
|
|
|
2008-04-13 19:42:31 -06:00
|
|
|
static int __initdata printed_gart_size_msg;
|
|
|
|
|
2007-10-24 04:49:50 -06:00
|
|
|
void __init gart_iommu_hole_init(void)
|
2008-01-30 05:30:09 -07:00
|
|
|
{
|
2008-04-13 19:42:31 -06:00
|
|
|
u32 agp_aper_base = 0, agp_aper_order = 0;
|
2005-11-05 09:25:53 -07:00
|
|
|
u32 aper_size, aper_alloc = 0, aper_order = 0, last_aper_order = 0;
|
2005-04-16 16:20:36 -06:00
|
|
|
u64 aper_base, last_aper_base = 0;
|
2008-04-19 02:31:11 -06:00
|
|
|
int fix, slot, valid_agp = 0;
|
|
|
|
int i, node;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2007-10-24 04:49:50 -06:00
|
|
|
if (gart_iommu_aperture_disabled || !fix_aperture ||
|
|
|
|
!early_pci_allowed())
|
2005-04-16 16:20:36 -06:00
|
|
|
return;
|
|
|
|
|
2007-07-21 09:11:36 -06:00
|
|
|
printk(KERN_INFO "Checking aperture...\n");
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2008-04-13 19:42:31 -06:00
|
|
|
if (!fallback_aper_force)
|
|
|
|
agp_aper_base = search_agp_bridge(&agp_aper_order, &valid_agp);
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
fix = 0;
|
2008-01-30 05:33:18 -07:00
|
|
|
node = 0;
|
2008-04-19 02:31:11 -06:00
|
|
|
for (i = 0; i < ARRAY_SIZE(bus_dev_ranges); i++) {
|
|
|
|
int bus;
|
|
|
|
int dev_base, dev_limit;
|
|
|
|
|
|
|
|
bus = bus_dev_ranges[i].bus;
|
|
|
|
dev_base = bus_dev_ranges[i].dev_base;
|
|
|
|
dev_limit = bus_dev_ranges[i].dev_limit;
|
|
|
|
|
|
|
|
for (slot = dev_base; slot < dev_limit; slot++) {
|
|
|
|
if (!early_is_k8_nb(read_pci_config(bus, slot, 3, 0x00)))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
iommu_detected = 1;
|
|
|
|
gart_iommu_aperture = 1;
|
|
|
|
|
|
|
|
aper_order = (read_pci_config(bus, slot, 3, AMD64_GARTAPERTURECTL) >> 1) & 7;
|
|
|
|
aper_size = (32 * 1024 * 1024) << aper_order;
|
|
|
|
aper_base = read_pci_config(bus, slot, 3, AMD64_GARTAPERTUREBASE) & 0x7fff;
|
|
|
|
aper_base <<= 25;
|
|
|
|
|
|
|
|
printk(KERN_INFO "Node %d: aperture @ %Lx size %u MB\n",
|
|
|
|
node, aper_base, aper_size >> 20);
|
|
|
|
node++;
|
|
|
|
|
|
|
|
if (!aperture_valid(aper_base, aper_size, 64<<20)) {
|
|
|
|
if (valid_agp && agp_aper_base &&
|
|
|
|
agp_aper_base == aper_base &&
|
|
|
|
agp_aper_order == aper_order) {
|
|
|
|
/* the same between two setting from NB and agp */
|
2008-06-24 23:14:09 -06:00
|
|
|
if (!no_iommu &&
|
|
|
|
max_pfn > MAX_DMA32_PFN &&
|
|
|
|
!printed_gart_size_msg) {
|
2008-04-19 02:31:11 -06:00
|
|
|
printk(KERN_ERR "you are using iommu with agp, but GART size is less than 64M\n");
|
|
|
|
printk(KERN_ERR "please increase GART size in your BIOS setup\n");
|
|
|
|
printk(KERN_ERR "if BIOS doesn't have that option, contact your HW vendor!\n");
|
|
|
|
printed_gart_size_msg = 1;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
fix = 1;
|
|
|
|
goto out;
|
2008-04-13 19:42:31 -06:00
|
|
|
}
|
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2008-04-19 02:31:11 -06:00
|
|
|
if ((last_aper_order && aper_order != last_aper_order) ||
|
|
|
|
(last_aper_base && aper_base != last_aper_base)) {
|
|
|
|
fix = 1;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
last_aper_order = aper_order;
|
|
|
|
last_aper_base = aper_base;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
2008-01-30 05:30:09 -07:00
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2008-04-19 02:31:11 -06:00
|
|
|
out:
|
2006-09-26 02:52:40 -06:00
|
|
|
if (!fix && !fallback_aper_force) {
|
|
|
|
if (last_aper_base) {
|
|
|
|
unsigned long n = (32 * 1024 * 1024) << last_aper_order;
|
2008-01-30 05:30:09 -07:00
|
|
|
|
2006-09-26 02:52:40 -06:00
|
|
|
insert_aperture_resource((u32)last_aper_base, n);
|
|
|
|
}
|
2008-01-30 05:30:09 -07:00
|
|
|
return;
|
2006-09-26 02:52:40 -06:00
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2008-04-13 19:42:31 -06:00
|
|
|
if (!fallback_aper_force) {
|
|
|
|
aper_alloc = agp_aper_base;
|
|
|
|
aper_order = agp_aper_order;
|
|
|
|
}
|
2008-01-30 05:30:09 -07:00
|
|
|
|
|
|
|
if (aper_alloc) {
|
2005-04-16 16:20:36 -06:00
|
|
|
/* Got the aperture from the AGP bridge */
|
2005-09-12 10:49:24 -06:00
|
|
|
} else if (swiotlb && !valid_agp) {
|
|
|
|
/* Do nothing */
|
2008-06-24 23:14:09 -06:00
|
|
|
} else if ((!no_iommu && max_pfn > MAX_DMA32_PFN) ||
|
2005-04-16 16:20:36 -06:00
|
|
|
force_iommu ||
|
|
|
|
valid_agp ||
|
2008-01-30 05:30:09 -07:00
|
|
|
fallback_aper_force) {
|
2008-01-30 05:30:10 -07:00
|
|
|
printk(KERN_ERR
|
|
|
|
"Your BIOS doesn't leave a aperture memory hole\n");
|
|
|
|
printk(KERN_ERR
|
|
|
|
"Please enable the IOMMU option in the BIOS setup\n");
|
|
|
|
printk(KERN_ERR
|
|
|
|
"This costs you %d MB of RAM\n",
|
|
|
|
32 << fallback_aper_order);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
aper_order = fallback_aper_order;
|
|
|
|
aper_alloc = allocate_aperture();
|
2008-01-30 05:30:09 -07:00
|
|
|
if (!aper_alloc) {
|
|
|
|
/*
|
|
|
|
* Could disable AGP and IOMMU here, but it's
|
|
|
|
* probably not worth it. But the later users
|
|
|
|
* cannot deal with bad apertures and turning
|
|
|
|
* on the aperture over memory causes very
|
|
|
|
* strange problems, so it's better to panic
|
|
|
|
* early.
|
|
|
|
*/
|
2005-04-16 16:20:36 -06:00
|
|
|
panic("Not enough memory for aperture");
|
|
|
|
}
|
2008-01-30 05:30:09 -07:00
|
|
|
} else {
|
|
|
|
return;
|
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
/* Fix up the north bridges */
|
2008-04-19 02:31:11 -06:00
|
|
|
for (i = 0; i < ARRAY_SIZE(bus_dev_ranges); i++) {
|
|
|
|
int bus;
|
|
|
|
int dev_base, dev_limit;
|
|
|
|
|
|
|
|
bus = bus_dev_ranges[i].bus;
|
|
|
|
dev_base = bus_dev_ranges[i].dev_base;
|
|
|
|
dev_limit = bus_dev_ranges[i].dev_limit;
|
|
|
|
for (slot = dev_base; slot < dev_limit; slot++) {
|
|
|
|
if (!early_is_k8_nb(read_pci_config(bus, slot, 3, 0x00)))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
/* Don't enable translation yet. That is done later.
|
|
|
|
Assume this BIOS didn't initialise the GART so
|
|
|
|
just overwrite all previous bits */
|
|
|
|
write_pci_config(bus, slot, 3, AMD64_GARTAPERTURECTL, aper_order << 1);
|
|
|
|
write_pci_config(bus, slot, 3, AMD64_GARTAPERTUREBASE, aper_alloc >> 25);
|
|
|
|
}
|
2008-01-30 05:30:09 -07:00
|
|
|
}
|
2008-06-09 16:10:48 -06:00
|
|
|
|
|
|
|
set_up_gart_resume(aper_order, aper_alloc);
|
2008-01-30 05:30:09 -07:00
|
|
|
}
|