Skip to contentSkip to conversation

- make /proc/vmcore accesable on x86_64

RFC PATCH In review Jiri Olsa 2 participants 5 messages Oct 19, 2010 – Oct 31, 2010

Subsystems (2)
lore
Share & tools
lore

Jiri Olsa

author
#0 lore raw

RFC PATCH 1/1 - make /proc/vmcore accesable on x86_64

hi,

I was playing around with preserve context feature on x86_64 and was not able to get the /proc/vmcore in the kexec-ed kernel until I aplied attached patch.

I'm not at all sure this is right fix ;) but it seems that /proc/vmcore is not loaded since the machine field in the elf header is wrong.

If the x86_64 arch returns 0 (KEXEC_ARCH_DEFAULT) then following condition in load_crashdump_segments function will fill 386 as the machine for the vmcore elf header..

...

        if ((info->kexec_flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_X86_64)
        {
                elf_info.machine = EM_X86_64;
        } else {
                elf_info.machine       = EM_386;
                elf_info.lowmem_limit  = X86_MAXMEM;
                elf_info.get_note_info = get_crash_notes;
        }

...

ending with kexec-ed kernel failing check for the machine, and disabling vmcore interface...

I'm guessing that the reason is the load_crashdump_segments function being shared among 32 and 64 bits.. but again, not sure ;)

My goal is to share data within the host and kexec-ed kernel, so I'd like to use the /proc/vmcore for that.. any other ideas are welcome :)

thanks,
jirka
---

kexec/arch/i386/crashdump-x86.c +2 -0
index 06e5ae9..40f989a 100644
@@ -691,6 +691,8 @@ int load_crashdump_segments(struct kexec_info *info, char* mod_cmdline,
 	struct memory_range *mem_range, *memmap_p;
 	struct crash_elf_info elf_info;
 
+	memset(&elf_info, 0x0, sizeof(elf_info));
+
 	/* Constant parts of the elf_info */
 	elf_info.data             = ELFDATA2LSB;
 	elf_info.backup_src_start = BACKUP_SRC_START;
kexec/arch/x86_64/kexec-x86_64.c +1 -4
index 3092643..c34fd92 100644
@@ -134,10 +134,7 @@ int arch_process_options(int argc, char **argv)
 }
 
 const struct arch_map_entry arches[] = {
-	/* For compatibility with older patches
-	 * use KEXEC_ARCH_DEFAULT instead of KEXEC_ARCH_X86_64 here.
-	 */
-	{ "x86_64", KEXEC_ARCH_DEFAULT },
+	{ "x86_64", KEXEC_ARCH_X86_64 },
 	{ NULL, 0 },
 };
 

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#1 lore raw
Quoted 31 lines · Jiri Olsa

hi,

I was playing around with preserve context feature on x86_64 and was not able to get the /proc/vmcore in the kexec-ed kernel until I aplied attached patch.

I'm not at all sure this is right fix ;) but it seems that /proc/vmcore is not loaded since the machine field in the elf header is wrong.

If the x86_64 arch returns 0 (KEXEC_ARCH_DEFAULT) then following condition in load_crashdump_segments function will fill 386 as the machine for the vmcore elf header..

...

        if ((info->kexec_flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_X86_64)
        {
                elf_info.machine = EM_X86_64;
        } else {
                elf_info.machine       = EM_386;
                elf_info.lowmem_limit  = X86_MAXMEM;
                elf_info.get_note_info = get_crash_notes;
        }

...

ending with kexec-ed kernel failing check for the machine, and disabling vmcore interface...

Quoted 7 lines

I'm guessing that the reason is the load_crashdump_segments function being shared among 32 and 64 bits.. but again, not sure ;)

My goal is to share data within the host and kexec-ed kernel, so I'd like to use the /proc/vmcore for that.. any other ideas are welcome :)

This approach seems reasonable to me.
I'll push it if a I don't hear any objections
in the near future.

Quoted 25 lines

thanks,
jirka
---

kexec/arch/i386/crashdump-x86.c +2 -0
index 06e5ae9..40f989a 100644
@@ -691,6 +691,8 @@ int load_crashdump_segments(struct kexec_info *info, char* mod_cmdline,
 	struct memory_range *mem_range, *memmap_p;
 	struct crash_elf_info elf_info;
 
+	memset(&elf_info, 0x0, sizeof(elf_info));
+
 	/* Constant parts of the elf_info */
 	elf_info.data             = ELFDATA2LSB;
 	elf_info.backup_src_start = BACKUP_SRC_START;
kexec/arch/x86_64/kexec-x86_64.c +0 -0
index 3092643..c34fd92 100644
@@ -134,10 +134,7 @@ int arch_process_options(int argc, char **argv)
 }

kexec/arch/x86_64/kexec-x86_64.c int arch_process_options(int argc, char **argv) · line 140

Quoted 10 lines
kexec/arch/x86_64/kexec-x86_64.c +1 -4
 
 const struct arch_map_entry arches[] = {
-	/* For compatibility with older patches
-	 * use KEXEC_ARCH_DEFAULT instead of KEXEC_ARCH_X86_64 here.
-	 */
-	{ "x86_64", KEXEC_ARCH_DEFAULT },
+	{ "x86_64", KEXEC_ARCH_X86_64 },
 	{ NULL, 0 },
 };
 

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#2 lore raw
Quoted 36 lines · Simon Horman
Quoted 35 lines · Jiri Olsa

hi,

I was playing around with preserve context feature on x86_64 and was not able to get the /proc/vmcore in the kexec-ed kernel until I aplied attached patch.

I'm not at all sure this is right fix ;) but it seems that /proc/vmcore is not loaded since the machine field in the elf header is wrong.

If the x86_64 arch returns 0 (KEXEC_ARCH_DEFAULT) then following condition in load_crashdump_segments function will fill 386 as the machine for the vmcore elf header..

...

        if ((info->kexec_flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_X86_64)
        {
                elf_info.machine = EM_X86_64;
        } else {
                elf_info.machine       = EM_386;
                elf_info.lowmem_limit  = X86_MAXMEM;
                elf_info.get_note_info = get_crash_notes;
        }

...

ending with kexec-ed kernel failing check for the machine, and disabling vmcore interface...

I'm guessing that the reason is the load_crashdump_segments function being shared among 32 and 64 bits.. but again, not sure ;)

Quoted 7 lines
Quoted 3 lines

My goal is to share data within the host and kexec-ed kernel, so I'd like to use the /proc/vmcore for that.. any other ideas are welcome :)

This approach seems reasonable to me.
I'll push it if a I don't hear any objections
in the near future.

Hi Jiru,

could you please submit this as a formal patch with a signed-off-by line?

Quoted 23 lines
Quoted 23 lines

thanks,
jirka
---

kexec/arch/i386/crashdump-x86.c +2 -0
index 06e5ae9..40f989a 100644
@@ -691,6 +691,8 @@ int load_crashdump_segments(struct kexec_info *info, char* mod_cmdline,
 	struct memory_range *mem_range, *memmap_p;
 	struct crash_elf_info elf_info;
 
+	memset(&elf_info, 0x0, sizeof(elf_info));
+
 	/* Constant parts of the elf_info */
 	elf_info.data             = ELFDATA2LSB;
 	elf_info.backup_src_start = BACKUP_SRC_START;
kexec/arch/x86_64/kexec-x86_64.c +0 -0
index 3092643..c34fd92 100644
@@ -134,10 +134,7 @@ int arch_process_options(int argc, char **argv)
 }
 
Quoted 16 lines

kexec/arch/x86_64/kexec-x86_64.c int arch_process_options(int argc, char **argv) · line 140

Quoted 10 lines
kexec/arch/x86_64/kexec-x86_64.c +1 -4
 const struct arch_map_entry arches[] = {
-	/* For compatibility with older patches
-	 * use KEXEC_ARCH_DEFAULT instead of KEXEC_ARCH_X86_64 here.
-	 */
-	{ "x86_64", KEXEC_ARCH_DEFAULT },
+	{ "x86_64", KEXEC_ARCH_X86_64 },
 	{ NULL, 0 },
 };
 

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

author
#3 lore raw

PATCH 1/1 - make /proc/vmcore accesable on x86_64

Quoted 2 lines · Simon Horman
Quoted 1 line · Simon Horman

On Tue, Oct 19, 2010 at 05:06:30PM +0200, Jiri Olsa wrote:

SNIP

Quoted 11 lines
Quoted 7 lines
Quoted 3 lines

My goal is to share data within the host and kexec-ed kernel, so I'd like to use the /proc/vmcore for that.. any other ideas are welcome :)

This approach seems reasonable to me.
I'll push it if a I don't hear any objections
in the near future.

Hi Jiru,

could you please submit this as a formal patch with a signed-off-by line?

sure, it's attached

thanks,
jirka
---

hi,

I was playing around with preserve context feature on x86_64 and was not able to get the /proc/vmcore in the kexec-ed kernel until I aplied attached patch.

If the x86_64 arch returns 0 (KEXEC_ARCH_DEFAULT) then following condition in load_crashdump_segments function will fill 386 as the machine for the vmcore elf header..

...

        if ((info->kexec_flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_X86_64)
        {
                elf_info.machine = EM_X86_64;
        } else {
                elf_info.machine       = EM_386;
                elf_info.lowmem_limit  = X86_MAXMEM;
                elf_info.get_note_info = get_crash_notes;
        }

...

ending with kexec-ed kernel failing check for the machine, and disabling vmcore interface.

wbr,
jirka

---

kexec/arch/i386/crashdump-x86.c +2 -0
index 06e5ae9..40f989a 100644
@@ -691,6 +691,8 @@ int load_crashdump_segments(struct kexec_info *info, char* mod_cmdline,
 	struct memory_range *mem_range, *memmap_p;
 	struct crash_elf_info elf_info;
 
+	memset(&elf_info, 0x0, sizeof(elf_info));
+
 	/* Constant parts of the elf_info */
 	elf_info.data             = ELFDATA2LSB;
 	elf_info.backup_src_start = BACKUP_SRC_START;
kexec/arch/x86_64/kexec-x86_64.c +1 -4
index 3092643..c34fd92 100644
@@ -134,10 +134,7 @@ int arch_process_options(int argc, char **argv)
 }
 
 const struct arch_map_entry arches[] = {
-	/* For compatibility with older patches
-	 * use KEXEC_ARCH_DEFAULT instead of KEXEC_ARCH_X86_64 here.
-	 */
-	{ "x86_64", KEXEC_ARCH_DEFAULT },
+	{ "x86_64", KEXEC_ARCH_X86_64 },
 	{ NULL, 0 },
 };
 

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

Applied
#4 lore raw
Quoted 14 lines · Jiri Olsa
Quoted 2 lines · Simon Horman
Quoted 1 line · Simon Horman

On Tue, Oct 19, 2010 at 05:06:30PM +0200, Jiri Olsa wrote:

SNIP

Quoted 8 lines
Quoted 7 lines
Quoted 3 lines

My goal is to share data within the host and kexec-ed kernel, so I'd like to use the /proc/vmcore for that.. any other ideas are welcome :)

This approach seems reasonable to me.
I'll push it if a I don't hear any objections
in the near future.

Quoted 5 lines
Quoted 3 lines

Hi Jiru,

could you please submit this as a formal patch with a signed-off-by line?

sure, it's attached

Thanks, applied.

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec