On machines where the GART aperture is mapped over physical RAM
/proc/vmcore contains the remapped range and reading it may
cause hangs or reboots. This range needs to be excluded from /proc/vmcore.
This has originally been implemented by commit dd5f726076cc ("kexec:
support for kexec on panic using new system call").
The implementation relied on the GART code adding an iomem_resource for
this range called "GART", as implemented by commit 56dd669a138c ("[PATCH]
Insert GART region into resource map")
However, inserting the iomem_resource from the early GART code caused
resource conflicts with some AGP drivers (bko#72201), which got avoided by
reverting the patch in commit 707d4eefbdb3 ("Revert [PATCH] Insert GART
region into resource map"). This revert introduced the /proc/vmcore bug.
With the "GART" iomem_resource removed, the defunct code in crash.c has
been removed by commit f296f2634920 ("x86/kexec: Remove walk_iomem_res()
call with GART type")
The patch below stores the location of the GART region in two variables
(named gart_stolen_ram_start and gart_stolen_ram_end) and reverts/adapts
parts of f296f2634920 to exclude the region from /proc/vmcore.
Passing the information via an iomem_resource (or by reserving the range
in the e820, which would propagate into an iomem_resource) would
reintroduce bko#72201.
Signed-off-by: Jiri Bohac <jbohac at suse.cz>
Fixes: 707d4eefbdb3 ("Revert [PATCH] Insert GART region into resource map")
diff --git a/arch/x86/include/asm/gart.h b/arch/x86/include/asm/gart.h
index 1d268098ac2e..324a0a19d166 100644
--- a/arch/x86/include/asm/gart.h
+++ b/arch/x86/include/asm/gart.h
@@ -2,6 +2,7 @@
#define _ASM_X86_GART_H
#include <asm/e820/api.h>
+#include <linux/pci.h>
extern void set_up_gart_resume(u32, u32);
@@ -33,6 +34,8 @@ extern int fix_aperture;
extern int gart_iommu_aperture;
extern int gart_iommu_aperture_allowed;
extern int gart_iommu_aperture_disabled;
+extern u32 gart_stolen_ram_start;
+extern u32 gart_stolen_ram_end;
extern void early_gart_iommu_check(void);
extern int gart_iommu_init(void);
@@ -43,6 +46,8 @@ extern int gart_iommu_hole_init(void);
#define gart_iommu_aperture 0
#define gart_iommu_aperture_allowed 0
#define gart_iommu_aperture_disabled 1
+#define gart_stolen_ram_start 0
+#define gart_stolen_ram_end 0
static inline void early_gart_iommu_check(void)
{
diff --git a/arch/x86/kernel/aperture_64.c b/arch/x86/kernel/aperture_64.c
index ef2859f9fcce..bc9a9b5b88e3 100644
--- a/arch/x86/kernel/aperture_64.c
+++ b/arch/x86/kernel/aperture_64.c
@@ -49,6 +49,8 @@
int gart_iommu_aperture;
int gart_iommu_aperture_disabled __initdata;
int gart_iommu_aperture_allowed __initdata;
+u32 gart_stolen_ram_start;
+u32 gart_stolen_ram_end;
int fallback_aper_order __initdata = 1; /* 64MB */
int fallback_aper_force __initdata;
@@ -87,6 +89,9 @@ static u32 __init allocate_aperture(void)
register_nosave_region(addr >> PAGE_SHIFT,
(addr+aper_size) >> PAGE_SHIFT);
+ gart_stolen_ram_start = (u32)addr;
+ gart_stolen_ram_end = (u32)addr + aper_size - 1;
+
return (u32)addr;
}
diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
index 44404e2307bb..ce065d72656d 100644
--- a/arch/x86/kernel/crash.c
+++ b/arch/x86/kernel/crash.c
@@ -37,6 +37,7 @@
#include <asm/reboot.h>
#include <asm/virtext.h>
#include <asm/intel_pt.h>
+#include <asm/gart.h>
/* Alignment required for elf header segment */
#define ELF_CORE_HEADER_ALIGN 4096
@@ -58,7 +59,7 @@ struct crash_elf_data {
struct kimage *image;
/*
* Total number of ram ranges we have after various adjustments for
- * crash reserved region, etc.
+ * GART, crash reserved region etc.
*/
unsigned int max_nr_ranges;
@@ -217,7 +218,6 @@ static int get_nr_ram_ranges_callback(u64 start, u64 end, void *arg)
return 0;
}
-
/* Gather all the required information to prepare elf headers for ram regions */
static void fill_up_crash_elf_data(struct crash_elf_data *ced,
struct kimage *image)
@@ -231,6 +231,15 @@ static void fill_up_crash_elf_data(struct crash_elf_data *ced,
ced->max_nr_ranges = nr_ranges;
+ /*
+ * We don't create ELF headers for GART aperture as an attempt
+ * to dump this memory in second kernel leads to hang/crash.
+ * If gart aperture is mapped over RAM, one needs to exclude that region
+ * and that requires an extra phdr.
+ */
+ if (gart_stolen_ram_start)
+ ced->max_nr_ranges++;
+
/* Exclusion of crash region could split memory ranges */
ced->max_nr_ranges++;
@@ -339,6 +348,14 @@ static int elf_header_exclude_ranges(struct crash_elf_data *ced,
return ret;
}
+ /* Exclude GART region */
+ if (gart_stolen_ram_start) {
+ ret = exclude_mem_range(cmem, gart_stolen_ram_start,
+ gart_stolen_ram_end);
+ if (ret)
+ return ret;
+ }
+
return ret;
}
--
Jiri Bohac <jbohac at suse.cz>
SUSE Labs, Prague, Czechia
