linux_dsm_epyc7002/arch/x86/kernel/cpu/mtrr/generic.c

726 lines
19 KiB
C
Raw Normal View History

/* This only handles 32bit MTRR on 32bit hosts. This is strictly wrong
because MTRRs can span upto 40 bits (36bits on most modern x86) */
#include <linux/init.h>
#include <linux/slab.h>
#include <linux/mm.h>
[PATCH] i386: fix MTRR code Until not so long ago, there were system log messages pointing to inconsistent MTRR setup of the video frame buffer caused by the way vesafb and X worked. While vesafb was fixed meanwhile, I believe fixing it there only hides a shortcoming in the MTRR code itself, in that that code is not symmetric with respect to the ordering of attempts to set up two (or more) regions where one contains the other. In the current shape, it permits only setting up sub-regions of pre-exisiting ones. The patch below makes this symmetric. While working on that I noticed a few more inconsistencies in that code, namely - use of 'unsigned int' for sizes in many, but not all places (the patch is converting this to use 'unsigned long' everywhere, which specifically might be necessary for x86-64 once a processor supporting more than 44 physical address bits would become available) - the code to correct inconsistent settings during secondary processor startup tried (if necessary) to correct, among other things, the value in IA32_MTRR_DEF_TYPE, however the newly computed value would never get used (i.e. stored in the respective MSR) - the generic range validation code checked that the end of the to-be-added range would be above 1MB; the value checked should have been the start of the range - when contained regions are detected, previously this was allowed only when the old region was uncacheable; this can be symmetric (i.e. the new region can also be uncacheable) and even further as per Intel's documentation write-trough and write-back for either region is also compatible with the respective opposite in the other Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Andi Kleen <ak@suse.de>
2006-12-07 08:14:09 +07:00
#include <linux/module.h>
#include <asm/io.h>
#include <asm/mtrr.h>
#include <asm/msr.h>
#include <asm/system.h>
#include <asm/cpufeature.h>
#include <asm/processor-flags.h>
#include <asm/tlbflush.h>
#include <asm/pat.h>
#include "mtrr.h"
struct fixed_range_block {
int base_msr; /* start address of an MTRR block */
int ranges; /* number of MTRRs in this block */
};
static struct fixed_range_block fixed_range_blocks[] = {
{ MSR_MTRRfix64K_00000, 1 }, /* one 64k MTRR */
{ MTRRfix16K_80000_MSR, 2 }, /* two 16k MTRRs */
{ MTRRfix4K_C0000_MSR, 8 }, /* eight 4k MTRRs */
{}
};
static unsigned long smp_changes_mask;
static int mtrr_state_set;
u64 mtrr_tom2;
struct mtrr_state_type mtrr_state = {};
EXPORT_SYMBOL_GPL(mtrr_state);
x86: mtrr: don't modify RdDram/WrDram bits of fixed MTRRs Impact: bug fix + BIOS workaround BIOS is expected to clear the SYSCFG[MtrrFixDramModEn] on AMD CPUs after fixed MTRRs are configured. Some BIOSes do not clear SYSCFG[MtrrFixDramModEn] on BP (and on APs). This can lead to obfuscation in Linux when this bit is not cleared on BP but cleared on APs. A consequence of this is that the saved fixed-MTRR state (from BP) differs from the fixed-MTRRs of APs -- because RdDram/WrDram bits are read as zero when SYSCFG[MtrrFixDramModEn] is cleared -- and Linux tries to sync fixed-MTRR state from BP to AP. This implies that Linux sets SYSCFG[MtrrFixDramEn] and activates those bits. More important is that (some) systems change these bits in SMM when ACPI is enabled. Hence it is racy if Linux modifies RdMem/WrMem bits, too. (1) The patch modifies an old fix from Bernhard Kaindl to get suspend/resume working on some Acer Laptops. Bernhard's patch tried to sync RdMem/WrMem bits of fixed MTRR registers and that helped on those old Laptops. (Don't ask me why -- can't test it myself). But this old problem was not the motivation for the patch. (See http://lkml.org/lkml/2007/4/3/110) (2) The more important effect is to fix issues on some more current systems. On those systems Linux panics or just freezes, see http://bugzilla.kernel.org/show_bug.cgi?id=11541 (and also duplicates of this bug: http://bugzilla.kernel.org/show_bug.cgi?id=11737 http://bugzilla.kernel.org/show_bug.cgi?id=11714) The affected systems boot only using acpi=ht, acpi=off or when the kernel is built with CONFIG_MTRR=n. The acpi options prevent full enablement of ACPI. Obviously when ACPI is enabled the BIOS/SMM modfies RdMem/WrMem bits. When CONFIG_MTRR=y Linux also accesses and modifies those bits when it needs to sync fixed-MTRRs across cores (Bernhard's fix, see (1)). How do you synchronize that? You can't. As a consequence Linux shouldn't touch those bits at all (Rationale are AMD's BKDGs which recommend to clear the bit that makes RdMem/WrMem accessible). This is the purpose of this patch. And (so far) this suffices to fix (1) and (2). I suggest not to touch RdDram/WrDram bits of fixed-MTRRs and SYSCFG[MtrrFixDramEn] and to clear SYSCFG[MtrrFixDramModEn] as suggested by AMD K8, and AMD family 10h/11h BKDGs. BIOS is expected to do this anyway. This should avoid that Linux and SMM tread on each other's toes ... Signed-off-by: Andreas Herrmann <andreas.herrmann3@amd.com> Cc: trenn@suse.de Cc: Yinghai Lu <yinghai@kernel.org> LKML-Reference: <20090312163937.GH20716@alberich.amd.com> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-03-12 23:39:37 +07:00
/**
* BIOS is expected to clear MtrrFixDramModEn bit, see for example
* "BIOS and Kernel Developer's Guide for the AMD Athlon 64 and AMD
* Opteron Processors" (26094 Rev. 3.30 February 2006), section
* "13.2.1.2 SYSCFG Register": "The MtrrFixDramModEn bit should be set
* to 1 during BIOS initalization of the fixed MTRRs, then cleared to
* 0 for operation."
*/
static inline void k8_check_syscfg_dram_mod_en(void)
{
u32 lo, hi;
if (!((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) &&
(boot_cpu_data.x86 >= 0x0f)))
return;
rdmsr(MSR_K8_SYSCFG, lo, hi);
if (lo & K8_MTRRFIXRANGE_DRAM_MODIFY) {
printk(KERN_ERR FW_WARN "MTRR: CPU %u: SYSCFG[MtrrFixDramModEn]"
" not cleared by BIOS, clearing this bit\n",
smp_processor_id());
lo &= ~K8_MTRRFIXRANGE_DRAM_MODIFY;
mtrr_wrmsr(MSR_K8_SYSCFG, lo, hi);
}
}
/*
* Returns the effective MTRR type for the region
* Error returns:
* - 0xFE - when the range is "not entirely covered" by _any_ var range MTRR
* - 0xFF - when MTRR is not enabled
*/
u8 mtrr_type_lookup(u64 start, u64 end)
{
int i;
u64 base, mask;
u8 prev_match, curr_match;
if (!mtrr_state_set)
return 0xFF;
if (!mtrr_state.enabled)
return 0xFF;
/* Make end inclusive end, instead of exclusive */
end--;
/* Look in fixed ranges. Just return the type as per start */
if (mtrr_state.have_fixed && (start < 0x100000)) {
int idx;
if (start < 0x80000) {
idx = 0;
idx += (start >> 16);
return mtrr_state.fixed_ranges[idx];
} else if (start < 0xC0000) {
idx = 1 * 8;
idx += ((start - 0x80000) >> 14);
return mtrr_state.fixed_ranges[idx];
} else if (start < 0x1000000) {
idx = 3 * 8;
idx += ((start - 0xC0000) >> 12);
return mtrr_state.fixed_ranges[idx];
}
}
/*
* Look in variable ranges
* Look of multiple ranges matching this address and pick type
* as per MTRR precedence
*/
if (!(mtrr_state.enabled & 2)) {
return mtrr_state.def_type;
}
prev_match = 0xFF;
for (i = 0; i < num_var_ranges; ++i) {
unsigned short start_state, end_state;
if (!(mtrr_state.var_ranges[i].mask_lo & (1 << 11)))
continue;
base = (((u64)mtrr_state.var_ranges[i].base_hi) << 32) +
(mtrr_state.var_ranges[i].base_lo & PAGE_MASK);
mask = (((u64)mtrr_state.var_ranges[i].mask_hi) << 32) +
(mtrr_state.var_ranges[i].mask_lo & PAGE_MASK);
start_state = ((start & mask) == (base & mask));
end_state = ((end & mask) == (base & mask));
if (start_state != end_state)
return 0xFE;
if ((start & mask) != (base & mask)) {
continue;
}
curr_match = mtrr_state.var_ranges[i].base_lo & 0xff;
if (prev_match == 0xFF) {
prev_match = curr_match;
continue;
}
if (prev_match == MTRR_TYPE_UNCACHABLE ||
curr_match == MTRR_TYPE_UNCACHABLE) {
return MTRR_TYPE_UNCACHABLE;
}
if ((prev_match == MTRR_TYPE_WRBACK &&
curr_match == MTRR_TYPE_WRTHROUGH) ||
(prev_match == MTRR_TYPE_WRTHROUGH &&
curr_match == MTRR_TYPE_WRBACK)) {
prev_match = MTRR_TYPE_WRTHROUGH;
curr_match = MTRR_TYPE_WRTHROUGH;
}
if (prev_match != curr_match) {
return MTRR_TYPE_UNCACHABLE;
}
}
if (mtrr_tom2) {
if (start >= (1ULL<<32) && (end < mtrr_tom2))
return MTRR_TYPE_WRBACK;
}
if (prev_match != 0xFF)
return prev_match;
return mtrr_state.def_type;
}
/* Get the MSR pair relating to a var range */
static void
get_mtrr_var_range(unsigned int index, struct mtrr_var_range *vr)
{
rdmsr(MTRRphysBase_MSR(index), vr->base_lo, vr->base_hi);
rdmsr(MTRRphysMask_MSR(index), vr->mask_lo, vr->mask_hi);
}
/* fill the MSR pair relating to a var range */
void fill_mtrr_var_range(unsigned int index,
u32 base_lo, u32 base_hi, u32 mask_lo, u32 mask_hi)
{
struct mtrr_var_range *vr;
vr = mtrr_state.var_ranges;
vr[index].base_lo = base_lo;
vr[index].base_hi = base_hi;
vr[index].mask_lo = mask_lo;
vr[index].mask_hi = mask_hi;
}
static void
get_fixed_ranges(mtrr_type * frs)
{
unsigned int *p = (unsigned int *) frs;
int i;
x86: mtrr: don't modify RdDram/WrDram bits of fixed MTRRs Impact: bug fix + BIOS workaround BIOS is expected to clear the SYSCFG[MtrrFixDramModEn] on AMD CPUs after fixed MTRRs are configured. Some BIOSes do not clear SYSCFG[MtrrFixDramModEn] on BP (and on APs). This can lead to obfuscation in Linux when this bit is not cleared on BP but cleared on APs. A consequence of this is that the saved fixed-MTRR state (from BP) differs from the fixed-MTRRs of APs -- because RdDram/WrDram bits are read as zero when SYSCFG[MtrrFixDramModEn] is cleared -- and Linux tries to sync fixed-MTRR state from BP to AP. This implies that Linux sets SYSCFG[MtrrFixDramEn] and activates those bits. More important is that (some) systems change these bits in SMM when ACPI is enabled. Hence it is racy if Linux modifies RdMem/WrMem bits, too. (1) The patch modifies an old fix from Bernhard Kaindl to get suspend/resume working on some Acer Laptops. Bernhard's patch tried to sync RdMem/WrMem bits of fixed MTRR registers and that helped on those old Laptops. (Don't ask me why -- can't test it myself). But this old problem was not the motivation for the patch. (See http://lkml.org/lkml/2007/4/3/110) (2) The more important effect is to fix issues on some more current systems. On those systems Linux panics or just freezes, see http://bugzilla.kernel.org/show_bug.cgi?id=11541 (and also duplicates of this bug: http://bugzilla.kernel.org/show_bug.cgi?id=11737 http://bugzilla.kernel.org/show_bug.cgi?id=11714) The affected systems boot only using acpi=ht, acpi=off or when the kernel is built with CONFIG_MTRR=n. The acpi options prevent full enablement of ACPI. Obviously when ACPI is enabled the BIOS/SMM modfies RdMem/WrMem bits. When CONFIG_MTRR=y Linux also accesses and modifies those bits when it needs to sync fixed-MTRRs across cores (Bernhard's fix, see (1)). How do you synchronize that? You can't. As a consequence Linux shouldn't touch those bits at all (Rationale are AMD's BKDGs which recommend to clear the bit that makes RdMem/WrMem accessible). This is the purpose of this patch. And (so far) this suffices to fix (1) and (2). I suggest not to touch RdDram/WrDram bits of fixed-MTRRs and SYSCFG[MtrrFixDramEn] and to clear SYSCFG[MtrrFixDramModEn] as suggested by AMD K8, and AMD family 10h/11h BKDGs. BIOS is expected to do this anyway. This should avoid that Linux and SMM tread on each other's toes ... Signed-off-by: Andreas Herrmann <andreas.herrmann3@amd.com> Cc: trenn@suse.de Cc: Yinghai Lu <yinghai@kernel.org> LKML-Reference: <20090312163937.GH20716@alberich.amd.com> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-03-12 23:39:37 +07:00
k8_check_syscfg_dram_mod_en();
rdmsr(MSR_MTRRfix64K_00000, p[0], p[1]);
for (i = 0; i < 2; i++)
rdmsr(MTRRfix16K_80000_MSR + i, p[2 + i * 2], p[3 + i * 2]);
for (i = 0; i < 8; i++)
rdmsr(MTRRfix4K_C0000_MSR + i, p[6 + i * 2], p[7 + i * 2]);
}
void mtrr_save_fixed_ranges(void *info)
{
if (cpu_has_mtrr)
get_fixed_ranges(mtrr_state.fixed_ranges);
}
static unsigned __initdata last_fixed_start;
static unsigned __initdata last_fixed_end;
static mtrr_type __initdata last_fixed_type;
static void __init print_fixed_last(void)
{
if (!last_fixed_end)
return;
printk(KERN_DEBUG " %05X-%05X %s\n", last_fixed_start,
last_fixed_end - 1, mtrr_attrib_to_str(last_fixed_type));
last_fixed_end = 0;
}
static void __init update_fixed_last(unsigned base, unsigned end,
mtrr_type type)
{
last_fixed_start = base;
last_fixed_end = end;
last_fixed_type = type;
}
static void __init print_fixed(unsigned base, unsigned step,
const mtrr_type *types)
[PATCH] i386: fix MTRR code Until not so long ago, there were system log messages pointing to inconsistent MTRR setup of the video frame buffer caused by the way vesafb and X worked. While vesafb was fixed meanwhile, I believe fixing it there only hides a shortcoming in the MTRR code itself, in that that code is not symmetric with respect to the ordering of attempts to set up two (or more) regions where one contains the other. In the current shape, it permits only setting up sub-regions of pre-exisiting ones. The patch below makes this symmetric. While working on that I noticed a few more inconsistencies in that code, namely - use of 'unsigned int' for sizes in many, but not all places (the patch is converting this to use 'unsigned long' everywhere, which specifically might be necessary for x86-64 once a processor supporting more than 44 physical address bits would become available) - the code to correct inconsistent settings during secondary processor startup tried (if necessary) to correct, among other things, the value in IA32_MTRR_DEF_TYPE, however the newly computed value would never get used (i.e. stored in the respective MSR) - the generic range validation code checked that the end of the to-be-added range would be above 1MB; the value checked should have been the start of the range - when contained regions are detected, previously this was allowed only when the old region was uncacheable; this can be symmetric (i.e. the new region can also be uncacheable) and even further as per Intel's documentation write-trough and write-back for either region is also compatible with the respective opposite in the other Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Andi Kleen <ak@suse.de>
2006-12-07 08:14:09 +07:00
{
unsigned i;
for (i = 0; i < 8; ++i, ++types, base += step) {
if (last_fixed_end == 0) {
update_fixed_last(base, base + step, *types);
continue;
}
if (last_fixed_end == base && last_fixed_type == *types) {
last_fixed_end = base + step;
continue;
}
/* new segments: gap or different type */
print_fixed_last();
update_fixed_last(base, base + step, *types);
}
[PATCH] i386: fix MTRR code Until not so long ago, there were system log messages pointing to inconsistent MTRR setup of the video frame buffer caused by the way vesafb and X worked. While vesafb was fixed meanwhile, I believe fixing it there only hides a shortcoming in the MTRR code itself, in that that code is not symmetric with respect to the ordering of attempts to set up two (or more) regions where one contains the other. In the current shape, it permits only setting up sub-regions of pre-exisiting ones. The patch below makes this symmetric. While working on that I noticed a few more inconsistencies in that code, namely - use of 'unsigned int' for sizes in many, but not all places (the patch is converting this to use 'unsigned long' everywhere, which specifically might be necessary for x86-64 once a processor supporting more than 44 physical address bits would become available) - the code to correct inconsistent settings during secondary processor startup tried (if necessary) to correct, among other things, the value in IA32_MTRR_DEF_TYPE, however the newly computed value would never get used (i.e. stored in the respective MSR) - the generic range validation code checked that the end of the to-be-added range would be above 1MB; the value checked should have been the start of the range - when contained regions are detected, previously this was allowed only when the old region was uncacheable; this can be symmetric (i.e. the new region can also be uncacheable) and even further as per Intel's documentation write-trough and write-back for either region is also compatible with the respective opposite in the other Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Andi Kleen <ak@suse.de>
2006-12-07 08:14:09 +07:00
}
static void prepare_set(void);
static void post_set(void);
static void __init print_mtrr_state(void)
{
unsigned int i;
int high_width;
printk(KERN_DEBUG "MTRR default type: %s\n",
mtrr_attrib_to_str(mtrr_state.def_type));
if (mtrr_state.have_fixed) {
printk(KERN_DEBUG "MTRR fixed ranges %sabled:\n",
mtrr_state.enabled & 1 ? "en" : "dis");
print_fixed(0x00000, 0x10000, mtrr_state.fixed_ranges + 0);
for (i = 0; i < 2; ++i)
print_fixed(0x80000 + i * 0x20000, 0x04000, mtrr_state.fixed_ranges + (i + 1) * 8);
for (i = 0; i < 8; ++i)
print_fixed(0xC0000 + i * 0x08000, 0x01000, mtrr_state.fixed_ranges + (i + 3) * 8);
/* tail */
print_fixed_last();
}
printk(KERN_DEBUG "MTRR variable ranges %sabled:\n",
mtrr_state.enabled & 2 ? "en" : "dis");
high_width = ((size_or_mask ? ffs(size_or_mask) - 1 : 32) - (32 - PAGE_SHIFT) + 3) / 4;
for (i = 0; i < num_var_ranges; ++i) {
if (mtrr_state.var_ranges[i].mask_lo & (1 << 11))
printk(KERN_DEBUG " %u base %0*X%05X000 mask %0*X%05X000 %s\n",
i,
high_width,
mtrr_state.var_ranges[i].base_hi,
mtrr_state.var_ranges[i].base_lo >> 12,
high_width,
mtrr_state.var_ranges[i].mask_hi,
mtrr_state.var_ranges[i].mask_lo >> 12,
mtrr_attrib_to_str(mtrr_state.var_ranges[i].base_lo & 0xff));
else
printk(KERN_DEBUG " %u disabled\n", i);
}
if (mtrr_tom2) {
printk(KERN_DEBUG "TOM2: %016llx aka %lldM\n",
mtrr_tom2, mtrr_tom2>>20);
}
}
/* Grab all of the MTRR state for this CPU into *state */
void __init get_mtrr_state(void)
{
unsigned int i;
struct mtrr_var_range *vrs;
unsigned lo, dummy;
unsigned long flags;
vrs = mtrr_state.var_ranges;
rdmsr(MSR_MTRRcap, lo, dummy);
[PATCH] i386: fix MTRR code Until not so long ago, there were system log messages pointing to inconsistent MTRR setup of the video frame buffer caused by the way vesafb and X worked. While vesafb was fixed meanwhile, I believe fixing it there only hides a shortcoming in the MTRR code itself, in that that code is not symmetric with respect to the ordering of attempts to set up two (or more) regions where one contains the other. In the current shape, it permits only setting up sub-regions of pre-exisiting ones. The patch below makes this symmetric. While working on that I noticed a few more inconsistencies in that code, namely - use of 'unsigned int' for sizes in many, but not all places (the patch is converting this to use 'unsigned long' everywhere, which specifically might be necessary for x86-64 once a processor supporting more than 44 physical address bits would become available) - the code to correct inconsistent settings during secondary processor startup tried (if necessary) to correct, among other things, the value in IA32_MTRR_DEF_TYPE, however the newly computed value would never get used (i.e. stored in the respective MSR) - the generic range validation code checked that the end of the to-be-added range would be above 1MB; the value checked should have been the start of the range - when contained regions are detected, previously this was allowed only when the old region was uncacheable; this can be symmetric (i.e. the new region can also be uncacheable) and even further as per Intel's documentation write-trough and write-back for either region is also compatible with the respective opposite in the other Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Andi Kleen <ak@suse.de>
2006-12-07 08:14:09 +07:00
mtrr_state.have_fixed = (lo >> 8) & 1;
for (i = 0; i < num_var_ranges; i++)
get_mtrr_var_range(i, &vrs[i]);
[PATCH] i386: fix MTRR code Until not so long ago, there were system log messages pointing to inconsistent MTRR setup of the video frame buffer caused by the way vesafb and X worked. While vesafb was fixed meanwhile, I believe fixing it there only hides a shortcoming in the MTRR code itself, in that that code is not symmetric with respect to the ordering of attempts to set up two (or more) regions where one contains the other. In the current shape, it permits only setting up sub-regions of pre-exisiting ones. The patch below makes this symmetric. While working on that I noticed a few more inconsistencies in that code, namely - use of 'unsigned int' for sizes in many, but not all places (the patch is converting this to use 'unsigned long' everywhere, which specifically might be necessary for x86-64 once a processor supporting more than 44 physical address bits would become available) - the code to correct inconsistent settings during secondary processor startup tried (if necessary) to correct, among other things, the value in IA32_MTRR_DEF_TYPE, however the newly computed value would never get used (i.e. stored in the respective MSR) - the generic range validation code checked that the end of the to-be-added range would be above 1MB; the value checked should have been the start of the range - when contained regions are detected, previously this was allowed only when the old region was uncacheable; this can be symmetric (i.e. the new region can also be uncacheable) and even further as per Intel's documentation write-trough and write-back for either region is also compatible with the respective opposite in the other Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Andi Kleen <ak@suse.de>
2006-12-07 08:14:09 +07:00
if (mtrr_state.have_fixed)
get_fixed_ranges(mtrr_state.fixed_ranges);
rdmsr(MTRRdefType_MSR, lo, dummy);
mtrr_state.def_type = (lo & 0xff);
mtrr_state.enabled = (lo & 0xc00) >> 10;
[PATCH] i386: fix MTRR code Until not so long ago, there were system log messages pointing to inconsistent MTRR setup of the video frame buffer caused by the way vesafb and X worked. While vesafb was fixed meanwhile, I believe fixing it there only hides a shortcoming in the MTRR code itself, in that that code is not symmetric with respect to the ordering of attempts to set up two (or more) regions where one contains the other. In the current shape, it permits only setting up sub-regions of pre-exisiting ones. The patch below makes this symmetric. While working on that I noticed a few more inconsistencies in that code, namely - use of 'unsigned int' for sizes in many, but not all places (the patch is converting this to use 'unsigned long' everywhere, which specifically might be necessary for x86-64 once a processor supporting more than 44 physical address bits would become available) - the code to correct inconsistent settings during secondary processor startup tried (if necessary) to correct, among other things, the value in IA32_MTRR_DEF_TYPE, however the newly computed value would never get used (i.e. stored in the respective MSR) - the generic range validation code checked that the end of the to-be-added range would be above 1MB; the value checked should have been the start of the range - when contained regions are detected, previously this was allowed only when the old region was uncacheable; this can be symmetric (i.e. the new region can also be uncacheable) and even further as per Intel's documentation write-trough and write-back for either region is also compatible with the respective opposite in the other Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Andi Kleen <ak@suse.de>
2006-12-07 08:14:09 +07:00
if (amd_special_default_mtrr()) {
unsigned low, high;
/* TOP_MEM2 */
rdmsr(MSR_K8_TOP_MEM2, low, high);
mtrr_tom2 = high;
mtrr_tom2 <<= 32;
mtrr_tom2 |= low;
mtrr_tom2 &= 0xffffff800000ULL;
}
print_mtrr_state();
mtrr_state_set = 1;
/* PAT setup for BP. We need to go through sync steps here */
local_irq_save(flags);
prepare_set();
pat_init();
post_set();
local_irq_restore(flags);
}
/* Some BIOS's are fucked and don't set all MTRRs the same! */
void __init mtrr_state_warn(void)
{
unsigned long mask = smp_changes_mask;
if (!mask)
return;
if (mask & MTRR_CHANGE_MASK_FIXED)
printk(KERN_WARNING "mtrr: your CPUs had inconsistent fixed MTRR settings\n");
if (mask & MTRR_CHANGE_MASK_VARIABLE)
printk(KERN_WARNING "mtrr: your CPUs had inconsistent variable MTRR settings\n");
if (mask & MTRR_CHANGE_MASK_DEFTYPE)
printk(KERN_WARNING "mtrr: your CPUs had inconsistent MTRRdefType settings\n");
printk(KERN_INFO "mtrr: probably your BIOS does not setup all CPUs.\n");
printk(KERN_INFO "mtrr: corrected configuration.\n");
}
/* Doesn't attempt to pass an error out to MTRR users
because it's quite complicated in some cases and probably not
worth it because the best error handling is to ignore it. */
void mtrr_wrmsr(unsigned msr, unsigned a, unsigned b)
{
if (wrmsr_safe(msr, a, b) < 0)
printk(KERN_ERR
"MTRR: CPU %u: Writing MSR %x to %x:%x failed\n",
smp_processor_id(), msr, a, b);
}
/**
* set_fixed_range - checks & updates a fixed-range MTRR if it differs from the value it should have
* @msr: MSR address of the MTTR which should be checked and updated
* @changed: pointer which indicates whether the MTRR needed to be changed
* @msrwords: pointer to the MSR values which the MSR should have
*/
static void set_fixed_range(int msr, bool *changed, unsigned int *msrwords)
{
unsigned lo, hi;
rdmsr(msr, lo, hi);
if (lo != msrwords[0] || hi != msrwords[1]) {
mtrr_wrmsr(msr, msrwords[0], msrwords[1]);
*changed = true;
}
}
/**
* generic_get_free_region - Get a free MTRR.
* @base: The starting (base) address of the region.
* @size: The size (in bytes) of the region.
* @replace_reg: mtrr index to be replaced; set to invalid value if none.
*
* Returns: The index of the region on success, else negative on error.
*/
[PATCH] i386: fix MTRR code Until not so long ago, there were system log messages pointing to inconsistent MTRR setup of the video frame buffer caused by the way vesafb and X worked. While vesafb was fixed meanwhile, I believe fixing it there only hides a shortcoming in the MTRR code itself, in that that code is not symmetric with respect to the ordering of attempts to set up two (or more) regions where one contains the other. In the current shape, it permits only setting up sub-regions of pre-exisiting ones. The patch below makes this symmetric. While working on that I noticed a few more inconsistencies in that code, namely - use of 'unsigned int' for sizes in many, but not all places (the patch is converting this to use 'unsigned long' everywhere, which specifically might be necessary for x86-64 once a processor supporting more than 44 physical address bits would become available) - the code to correct inconsistent settings during secondary processor startup tried (if necessary) to correct, among other things, the value in IA32_MTRR_DEF_TYPE, however the newly computed value would never get used (i.e. stored in the respective MSR) - the generic range validation code checked that the end of the to-be-added range would be above 1MB; the value checked should have been the start of the range - when contained regions are detected, previously this was allowed only when the old region was uncacheable; this can be symmetric (i.e. the new region can also be uncacheable) and even further as per Intel's documentation write-trough and write-back for either region is also compatible with the respective opposite in the other Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Andi Kleen <ak@suse.de>
2006-12-07 08:14:09 +07:00
int generic_get_free_region(unsigned long base, unsigned long size, int replace_reg)
{
int i, max;
mtrr_type ltype;
[PATCH] i386: fix MTRR code Until not so long ago, there were system log messages pointing to inconsistent MTRR setup of the video frame buffer caused by the way vesafb and X worked. While vesafb was fixed meanwhile, I believe fixing it there only hides a shortcoming in the MTRR code itself, in that that code is not symmetric with respect to the ordering of attempts to set up two (or more) regions where one contains the other. In the current shape, it permits only setting up sub-regions of pre-exisiting ones. The patch below makes this symmetric. While working on that I noticed a few more inconsistencies in that code, namely - use of 'unsigned int' for sizes in many, but not all places (the patch is converting this to use 'unsigned long' everywhere, which specifically might be necessary for x86-64 once a processor supporting more than 44 physical address bits would become available) - the code to correct inconsistent settings during secondary processor startup tried (if necessary) to correct, among other things, the value in IA32_MTRR_DEF_TYPE, however the newly computed value would never get used (i.e. stored in the respective MSR) - the generic range validation code checked that the end of the to-be-added range would be above 1MB; the value checked should have been the start of the range - when contained regions are detected, previously this was allowed only when the old region was uncacheable; this can be symmetric (i.e. the new region can also be uncacheable) and even further as per Intel's documentation write-trough and write-back for either region is also compatible with the respective opposite in the other Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Andi Kleen <ak@suse.de>
2006-12-07 08:14:09 +07:00
unsigned long lbase, lsize;
max = num_var_ranges;
[PATCH] i386: fix MTRR code Until not so long ago, there were system log messages pointing to inconsistent MTRR setup of the video frame buffer caused by the way vesafb and X worked. While vesafb was fixed meanwhile, I believe fixing it there only hides a shortcoming in the MTRR code itself, in that that code is not symmetric with respect to the ordering of attempts to set up two (or more) regions where one contains the other. In the current shape, it permits only setting up sub-regions of pre-exisiting ones. The patch below makes this symmetric. While working on that I noticed a few more inconsistencies in that code, namely - use of 'unsigned int' for sizes in many, but not all places (the patch is converting this to use 'unsigned long' everywhere, which specifically might be necessary for x86-64 once a processor supporting more than 44 physical address bits would become available) - the code to correct inconsistent settings during secondary processor startup tried (if necessary) to correct, among other things, the value in IA32_MTRR_DEF_TYPE, however the newly computed value would never get used (i.e. stored in the respective MSR) - the generic range validation code checked that the end of the to-be-added range would be above 1MB; the value checked should have been the start of the range - when contained regions are detected, previously this was allowed only when the old region was uncacheable; this can be symmetric (i.e. the new region can also be uncacheable) and even further as per Intel's documentation write-trough and write-back for either region is also compatible with the respective opposite in the other Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Andi Kleen <ak@suse.de>
2006-12-07 08:14:09 +07:00
if (replace_reg >= 0 && replace_reg < max)
return replace_reg;
for (i = 0; i < max; ++i) {
mtrr_if->get(i, &lbase, &lsize, &ltype);
if (lsize == 0)
return i;
}
return -ENOSPC;
}
static void generic_get_mtrr(unsigned int reg, unsigned long *base,
[PATCH] i386: fix MTRR code Until not so long ago, there were system log messages pointing to inconsistent MTRR setup of the video frame buffer caused by the way vesafb and X worked. While vesafb was fixed meanwhile, I believe fixing it there only hides a shortcoming in the MTRR code itself, in that that code is not symmetric with respect to the ordering of attempts to set up two (or more) regions where one contains the other. In the current shape, it permits only setting up sub-regions of pre-exisiting ones. The patch below makes this symmetric. While working on that I noticed a few more inconsistencies in that code, namely - use of 'unsigned int' for sizes in many, but not all places (the patch is converting this to use 'unsigned long' everywhere, which specifically might be necessary for x86-64 once a processor supporting more than 44 physical address bits would become available) - the code to correct inconsistent settings during secondary processor startup tried (if necessary) to correct, among other things, the value in IA32_MTRR_DEF_TYPE, however the newly computed value would never get used (i.e. stored in the respective MSR) - the generic range validation code checked that the end of the to-be-added range would be above 1MB; the value checked should have been the start of the range - when contained regions are detected, previously this was allowed only when the old region was uncacheable; this can be symmetric (i.e. the new region can also be uncacheable) and even further as per Intel's documentation write-trough and write-back for either region is also compatible with the respective opposite in the other Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Andi Kleen <ak@suse.de>
2006-12-07 08:14:09 +07:00
unsigned long *size, mtrr_type *type)
{
unsigned int mask_lo, mask_hi, base_lo, base_hi;
unsigned int tmp, hi;
int cpu;
/*
* get_mtrr doesn't need to update mtrr_state, also it could be called
* from any cpu, so try to print it out directly.
*/
cpu = get_cpu();
rdmsr(MTRRphysMask_MSR(reg), mask_lo, mask_hi);
if ((mask_lo & 0x800) == 0) {
/* Invalid (i.e. free) range */
*base = 0;
*size = 0;
*type = 0;
goto out_put_cpu;
}
rdmsr(MTRRphysBase_MSR(reg), base_lo, base_hi);
/* Work out the shifted address mask: */
tmp = mask_hi << (32 - PAGE_SHIFT) | mask_lo >> PAGE_SHIFT;
mask_lo = size_or_mask | tmp;
/* Expand tmp with high bits to all 1s: */
hi = fls(tmp);
if (hi > 0) {
tmp |= ~((1<<(hi - 1)) - 1);
if (tmp != mask_lo) {
WARN_ONCE(1, KERN_INFO "mtrr: your BIOS has set up an incorrect mask, fixing it up.\n");
mask_lo = tmp;
}
}
/*
* This works correctly if size is a power of two, i.e. a
* contiguous range:
*/
*size = -mask_lo;
*base = base_hi << (32 - PAGE_SHIFT) | base_lo >> PAGE_SHIFT;
*type = base_lo & 0xff;
out_put_cpu:
put_cpu();
}
/**
* set_fixed_ranges - checks & updates the fixed-range MTRRs if they differ from the saved set
* @frs: pointer to fixed-range MTRR values, saved by get_fixed_ranges()
*/
static int set_fixed_ranges(mtrr_type * frs)
{
unsigned long long *saved = (unsigned long long *) frs;
bool changed = false;
int block=-1, range;
x86: mtrr: don't modify RdDram/WrDram bits of fixed MTRRs Impact: bug fix + BIOS workaround BIOS is expected to clear the SYSCFG[MtrrFixDramModEn] on AMD CPUs after fixed MTRRs are configured. Some BIOSes do not clear SYSCFG[MtrrFixDramModEn] on BP (and on APs). This can lead to obfuscation in Linux when this bit is not cleared on BP but cleared on APs. A consequence of this is that the saved fixed-MTRR state (from BP) differs from the fixed-MTRRs of APs -- because RdDram/WrDram bits are read as zero when SYSCFG[MtrrFixDramModEn] is cleared -- and Linux tries to sync fixed-MTRR state from BP to AP. This implies that Linux sets SYSCFG[MtrrFixDramEn] and activates those bits. More important is that (some) systems change these bits in SMM when ACPI is enabled. Hence it is racy if Linux modifies RdMem/WrMem bits, too. (1) The patch modifies an old fix from Bernhard Kaindl to get suspend/resume working on some Acer Laptops. Bernhard's patch tried to sync RdMem/WrMem bits of fixed MTRR registers and that helped on those old Laptops. (Don't ask me why -- can't test it myself). But this old problem was not the motivation for the patch. (See http://lkml.org/lkml/2007/4/3/110) (2) The more important effect is to fix issues on some more current systems. On those systems Linux panics or just freezes, see http://bugzilla.kernel.org/show_bug.cgi?id=11541 (and also duplicates of this bug: http://bugzilla.kernel.org/show_bug.cgi?id=11737 http://bugzilla.kernel.org/show_bug.cgi?id=11714) The affected systems boot only using acpi=ht, acpi=off or when the kernel is built with CONFIG_MTRR=n. The acpi options prevent full enablement of ACPI. Obviously when ACPI is enabled the BIOS/SMM modfies RdMem/WrMem bits. When CONFIG_MTRR=y Linux also accesses and modifies those bits when it needs to sync fixed-MTRRs across cores (Bernhard's fix, see (1)). How do you synchronize that? You can't. As a consequence Linux shouldn't touch those bits at all (Rationale are AMD's BKDGs which recommend to clear the bit that makes RdMem/WrMem accessible). This is the purpose of this patch. And (so far) this suffices to fix (1) and (2). I suggest not to touch RdDram/WrDram bits of fixed-MTRRs and SYSCFG[MtrrFixDramEn] and to clear SYSCFG[MtrrFixDramModEn] as suggested by AMD K8, and AMD family 10h/11h BKDGs. BIOS is expected to do this anyway. This should avoid that Linux and SMM tread on each other's toes ... Signed-off-by: Andreas Herrmann <andreas.herrmann3@amd.com> Cc: trenn@suse.de Cc: Yinghai Lu <yinghai@kernel.org> LKML-Reference: <20090312163937.GH20716@alberich.amd.com> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-03-12 23:39:37 +07:00
k8_check_syscfg_dram_mod_en();
while (fixed_range_blocks[++block].ranges)
for (range=0; range < fixed_range_blocks[block].ranges; range++)
set_fixed_range(fixed_range_blocks[block].base_msr + range,
&changed, (unsigned int *) saved++);
return changed;
}
/* Set the MSR pair relating to a var range. Returns TRUE if
changes are made */
static bool set_mtrr_var_ranges(unsigned int index, struct mtrr_var_range *vr)
{
unsigned int lo, hi;
bool changed = false;
rdmsr(MTRRphysBase_MSR(index), lo, hi);
if ((vr->base_lo & 0xfffff0ffUL) != (lo & 0xfffff0ffUL)
|| (vr->base_hi & (size_and_mask >> (32 - PAGE_SHIFT))) !=
(hi & (size_and_mask >> (32 - PAGE_SHIFT)))) {
mtrr_wrmsr(MTRRphysBase_MSR(index), vr->base_lo, vr->base_hi);
changed = true;
}
rdmsr(MTRRphysMask_MSR(index), lo, hi);
if ((vr->mask_lo & 0xfffff800UL) != (lo & 0xfffff800UL)
|| (vr->mask_hi & (size_and_mask >> (32 - PAGE_SHIFT))) !=
(hi & (size_and_mask >> (32 - PAGE_SHIFT)))) {
mtrr_wrmsr(MTRRphysMask_MSR(index), vr->mask_lo, vr->mask_hi);
changed = true;
}
return changed;
}
[PATCH] i386: fix MTRR code Until not so long ago, there were system log messages pointing to inconsistent MTRR setup of the video frame buffer caused by the way vesafb and X worked. While vesafb was fixed meanwhile, I believe fixing it there only hides a shortcoming in the MTRR code itself, in that that code is not symmetric with respect to the ordering of attempts to set up two (or more) regions where one contains the other. In the current shape, it permits only setting up sub-regions of pre-exisiting ones. The patch below makes this symmetric. While working on that I noticed a few more inconsistencies in that code, namely - use of 'unsigned int' for sizes in many, but not all places (the patch is converting this to use 'unsigned long' everywhere, which specifically might be necessary for x86-64 once a processor supporting more than 44 physical address bits would become available) - the code to correct inconsistent settings during secondary processor startup tried (if necessary) to correct, among other things, the value in IA32_MTRR_DEF_TYPE, however the newly computed value would never get used (i.e. stored in the respective MSR) - the generic range validation code checked that the end of the to-be-added range would be above 1MB; the value checked should have been the start of the range - when contained regions are detected, previously this was allowed only when the old region was uncacheable; this can be symmetric (i.e. the new region can also be uncacheable) and even further as per Intel's documentation write-trough and write-back for either region is also compatible with the respective opposite in the other Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Andi Kleen <ak@suse.de>
2006-12-07 08:14:09 +07:00
static u32 deftype_lo, deftype_hi;
/**
* set_mtrr_state - Set the MTRR state for this CPU.
*
* NOTE: The CPU must already be in a safe state for MTRR changes.
* RETURNS: 0 if no changes made, else a mask indicating what was changed.
*/
[PATCH] i386: fix MTRR code Until not so long ago, there were system log messages pointing to inconsistent MTRR setup of the video frame buffer caused by the way vesafb and X worked. While vesafb was fixed meanwhile, I believe fixing it there only hides a shortcoming in the MTRR code itself, in that that code is not symmetric with respect to the ordering of attempts to set up two (or more) regions where one contains the other. In the current shape, it permits only setting up sub-regions of pre-exisiting ones. The patch below makes this symmetric. While working on that I noticed a few more inconsistencies in that code, namely - use of 'unsigned int' for sizes in many, but not all places (the patch is converting this to use 'unsigned long' everywhere, which specifically might be necessary for x86-64 once a processor supporting more than 44 physical address bits would become available) - the code to correct inconsistent settings during secondary processor startup tried (if necessary) to correct, among other things, the value in IA32_MTRR_DEF_TYPE, however the newly computed value would never get used (i.e. stored in the respective MSR) - the generic range validation code checked that the end of the to-be-added range would be above 1MB; the value checked should have been the start of the range - when contained regions are detected, previously this was allowed only when the old region was uncacheable; this can be symmetric (i.e. the new region can also be uncacheable) and even further as per Intel's documentation write-trough and write-back for either region is also compatible with the respective opposite in the other Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Andi Kleen <ak@suse.de>
2006-12-07 08:14:09 +07:00
static unsigned long set_mtrr_state(void)
{
unsigned int i;
unsigned long change_mask = 0;
for (i = 0; i < num_var_ranges; i++)
if (set_mtrr_var_ranges(i, &mtrr_state.var_ranges[i]))
change_mask |= MTRR_CHANGE_MASK_VARIABLE;
[PATCH] i386: fix MTRR code Until not so long ago, there were system log messages pointing to inconsistent MTRR setup of the video frame buffer caused by the way vesafb and X worked. While vesafb was fixed meanwhile, I believe fixing it there only hides a shortcoming in the MTRR code itself, in that that code is not symmetric with respect to the ordering of attempts to set up two (or more) regions where one contains the other. In the current shape, it permits only setting up sub-regions of pre-exisiting ones. The patch below makes this symmetric. While working on that I noticed a few more inconsistencies in that code, namely - use of 'unsigned int' for sizes in many, but not all places (the patch is converting this to use 'unsigned long' everywhere, which specifically might be necessary for x86-64 once a processor supporting more than 44 physical address bits would become available) - the code to correct inconsistent settings during secondary processor startup tried (if necessary) to correct, among other things, the value in IA32_MTRR_DEF_TYPE, however the newly computed value would never get used (i.e. stored in the respective MSR) - the generic range validation code checked that the end of the to-be-added range would be above 1MB; the value checked should have been the start of the range - when contained regions are detected, previously this was allowed only when the old region was uncacheable; this can be symmetric (i.e. the new region can also be uncacheable) and even further as per Intel's documentation write-trough and write-back for either region is also compatible with the respective opposite in the other Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Andi Kleen <ak@suse.de>
2006-12-07 08:14:09 +07:00
if (mtrr_state.have_fixed && set_fixed_ranges(mtrr_state.fixed_ranges))
change_mask |= MTRR_CHANGE_MASK_FIXED;
/* Set_mtrr_restore restores the old value of MTRRdefType,
so to set it we fiddle with the saved value */
if ((deftype_lo & 0xff) != mtrr_state.def_type
|| ((deftype_lo & 0xc00) >> 10) != mtrr_state.enabled) {
[PATCH] i386: fix MTRR code Until not so long ago, there were system log messages pointing to inconsistent MTRR setup of the video frame buffer caused by the way vesafb and X worked. While vesafb was fixed meanwhile, I believe fixing it there only hides a shortcoming in the MTRR code itself, in that that code is not symmetric with respect to the ordering of attempts to set up two (or more) regions where one contains the other. In the current shape, it permits only setting up sub-regions of pre-exisiting ones. The patch below makes this symmetric. While working on that I noticed a few more inconsistencies in that code, namely - use of 'unsigned int' for sizes in many, but not all places (the patch is converting this to use 'unsigned long' everywhere, which specifically might be necessary for x86-64 once a processor supporting more than 44 physical address bits would become available) - the code to correct inconsistent settings during secondary processor startup tried (if necessary) to correct, among other things, the value in IA32_MTRR_DEF_TYPE, however the newly computed value would never get used (i.e. stored in the respective MSR) - the generic range validation code checked that the end of the to-be-added range would be above 1MB; the value checked should have been the start of the range - when contained regions are detected, previously this was allowed only when the old region was uncacheable; this can be symmetric (i.e. the new region can also be uncacheable) and even further as per Intel's documentation write-trough and write-back for either region is also compatible with the respective opposite in the other Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Andi Kleen <ak@suse.de>
2006-12-07 08:14:09 +07:00
deftype_lo = (deftype_lo & ~0xcff) | mtrr_state.def_type | (mtrr_state.enabled << 10);
change_mask |= MTRR_CHANGE_MASK_DEFTYPE;
}
return change_mask;
}
static unsigned long cr4 = 0;
static DEFINE_SPINLOCK(set_atomicity_lock);
/*
* Since we are disabling the cache don't allow any interrupts - they
* would run extremely slow and would only increase the pain. The caller must
* ensure that local interrupts are disabled and are reenabled after post_set()
* has been called.
*/
static void prepare_set(void) __acquires(set_atomicity_lock)
{
unsigned long cr0;
/* Note that this is not ideal, since the cache is only flushed/disabled
for this CPU while the MTRRs are changed, but changing this requires
more invasive changes to the way the kernel boots */
spin_lock(&set_atomicity_lock);
/* Enter the no-fill (CD=1, NW=0) cache mode and flush caches. */
cr0 = read_cr0() | X86_CR0_CD;
write_cr0(cr0);
wbinvd();
/* Save value of CR4 and clear Page Global Enable (bit 7) */
if ( cpu_has_pge ) {
cr4 = read_cr4();
write_cr4(cr4 & ~X86_CR4_PGE);
}
/* Flush all TLBs via a mov %cr3, %reg; mov %reg, %cr3 */
__flush_tlb();
/* Save MTRR state */
rdmsr(MTRRdefType_MSR, deftype_lo, deftype_hi);
/* Disable MTRRs, and set the default type to uncached */
[PATCH] i386: fix MTRR code Until not so long ago, there were system log messages pointing to inconsistent MTRR setup of the video frame buffer caused by the way vesafb and X worked. While vesafb was fixed meanwhile, I believe fixing it there only hides a shortcoming in the MTRR code itself, in that that code is not symmetric with respect to the ordering of attempts to set up two (or more) regions where one contains the other. In the current shape, it permits only setting up sub-regions of pre-exisiting ones. The patch below makes this symmetric. While working on that I noticed a few more inconsistencies in that code, namely - use of 'unsigned int' for sizes in many, but not all places (the patch is converting this to use 'unsigned long' everywhere, which specifically might be necessary for x86-64 once a processor supporting more than 44 physical address bits would become available) - the code to correct inconsistent settings during secondary processor startup tried (if necessary) to correct, among other things, the value in IA32_MTRR_DEF_TYPE, however the newly computed value would never get used (i.e. stored in the respective MSR) - the generic range validation code checked that the end of the to-be-added range would be above 1MB; the value checked should have been the start of the range - when contained regions are detected, previously this was allowed only when the old region was uncacheable; this can be symmetric (i.e. the new region can also be uncacheable) and even further as per Intel's documentation write-trough and write-back for either region is also compatible with the respective opposite in the other Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Andi Kleen <ak@suse.de>
2006-12-07 08:14:09 +07:00
mtrr_wrmsr(MTRRdefType_MSR, deftype_lo & ~0xcff, deftype_hi);
}
static void post_set(void) __releases(set_atomicity_lock)
{
/* Flush TLBs (no need to flush caches - they are disabled) */
__flush_tlb();
/* Intel (P6) standard MTRRs */
mtrr_wrmsr(MTRRdefType_MSR, deftype_lo, deftype_hi);
/* Enable caches */
write_cr0(read_cr0() & 0xbfffffff);
/* Restore value of CR4 */
if ( cpu_has_pge )
write_cr4(cr4);
spin_unlock(&set_atomicity_lock);
}
static void generic_set_all(void)
{
unsigned long mask, count;
unsigned long flags;
local_irq_save(flags);
prepare_set();
/* Actually set the state */
[PATCH] i386: fix MTRR code Until not so long ago, there were system log messages pointing to inconsistent MTRR setup of the video frame buffer caused by the way vesafb and X worked. While vesafb was fixed meanwhile, I believe fixing it there only hides a shortcoming in the MTRR code itself, in that that code is not symmetric with respect to the ordering of attempts to set up two (or more) regions where one contains the other. In the current shape, it permits only setting up sub-regions of pre-exisiting ones. The patch below makes this symmetric. While working on that I noticed a few more inconsistencies in that code, namely - use of 'unsigned int' for sizes in many, but not all places (the patch is converting this to use 'unsigned long' everywhere, which specifically might be necessary for x86-64 once a processor supporting more than 44 physical address bits would become available) - the code to correct inconsistent settings during secondary processor startup tried (if necessary) to correct, among other things, the value in IA32_MTRR_DEF_TYPE, however the newly computed value would never get used (i.e. stored in the respective MSR) - the generic range validation code checked that the end of the to-be-added range would be above 1MB; the value checked should have been the start of the range - when contained regions are detected, previously this was allowed only when the old region was uncacheable; this can be symmetric (i.e. the new region can also be uncacheable) and even further as per Intel's documentation write-trough and write-back for either region is also compatible with the respective opposite in the other Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Andi Kleen <ak@suse.de>
2006-12-07 08:14:09 +07:00
mask = set_mtrr_state();
/* also set PAT */
pat_init();
post_set();
local_irq_restore(flags);
/* Use the atomic bitops to update the global mask */
for (count = 0; count < sizeof mask * 8; ++count) {
if (mask & 0x01)
set_bit(count, &smp_changes_mask);
mask >>= 1;
}
}
static void generic_set_mtrr(unsigned int reg, unsigned long base,
unsigned long size, mtrr_type type)
/* [SUMMARY] Set variable MTRR register on the local CPU.
<reg> The register to set.
<base> The base address of the region.
<size> The size of the region. If this is 0 the region is disabled.
<type> The type of the region.
[RETURNS] Nothing.
*/
{
unsigned long flags;
struct mtrr_var_range *vr;
vr = &mtrr_state.var_ranges[reg];
local_irq_save(flags);
prepare_set();
if (size == 0) {
/* The invalid bit is kept in the mask, so we simply clear the
relevant mask register to disable a range. */
mtrr_wrmsr(MTRRphysMask_MSR(reg), 0, 0);
memset(vr, 0, sizeof(struct mtrr_var_range));
} else {
vr->base_lo = base << PAGE_SHIFT | type;
vr->base_hi = (base & size_and_mask) >> (32 - PAGE_SHIFT);
vr->mask_lo = -size << PAGE_SHIFT | 0x800;
vr->mask_hi = (-size & size_and_mask) >> (32 - PAGE_SHIFT);
mtrr_wrmsr(MTRRphysBase_MSR(reg), vr->base_lo, vr->base_hi);
mtrr_wrmsr(MTRRphysMask_MSR(reg), vr->mask_lo, vr->mask_hi);
}
post_set();
local_irq_restore(flags);
}
int generic_validate_add_page(unsigned long base, unsigned long size, unsigned int type)
{
unsigned long lbase, last;
/* For Intel PPro stepping <= 7, must be 4 MiB aligned
and not touch 0x70000000->0x7003FFFF */
if (is_cpu(INTEL) && boot_cpu_data.x86 == 6 &&
boot_cpu_data.x86_model == 1 &&
boot_cpu_data.x86_mask <= 7) {
if (base & ((1 << (22 - PAGE_SHIFT)) - 1)) {
printk(KERN_WARNING "mtrr: base(0x%lx000) is not 4 MiB aligned\n", base);
return -EINVAL;
}
if (!(base + size < 0x70000 || base > 0x7003F) &&
(type == MTRR_TYPE_WRCOMB
|| type == MTRR_TYPE_WRBACK)) {
printk(KERN_WARNING "mtrr: writable mtrr between 0x70000000 and 0x7003FFFF may hang the CPU.\n");
return -EINVAL;
}
}
/* Check upper bits of base and last are equal and lower bits are 0
for base and 1 for last */
last = base + size - 1;
for (lbase = base; !(lbase & 1) && (last & 1);
lbase = lbase >> 1, last = last >> 1) ;
if (lbase != last) {
printk(KERN_WARNING "mtrr: base(0x%lx000) is not aligned on a size(0x%lx000) boundary\n",
base, size);
return -EINVAL;
}
return 0;
}
static int generic_have_wrcomb(void)
{
unsigned long config, dummy;
rdmsr(MSR_MTRRcap, config, dummy);
return (config & (1 << 10));
}
int positive_have_wrcomb(void)
{
return 1;
}
/* generic structure...
*/
struct mtrr_ops generic_mtrr_ops = {
.use_intel_if = 1,
.set_all = generic_set_all,
.get = generic_get_mtrr,
.get_free_region = generic_get_free_region,
.set = generic_set_mtrr,
.validate_add_page = generic_validate_add_page,
.have_wrcomb = generic_have_wrcomb,
};