Kernel panic at i686
Hi all,
I started to test linux 4.1 series with rc6. However, I was never able to boot that kernel in i686 architecture. Trying it again with VirtualBox gave me more conclusions. Using one core it simply boots up. Using more than one CPU core it crashes with:
Failed to access perfctr msr (MSR c0010007 is 0)
task: f58e0000 ti: f58e8000 task.ti: f58e800
EIP: 0060:[<c135a903>] EFLAGS: 00010206 CPU: 0 EIP is at free_cache_attributes+0x83/0xd0
EAX: 00000001 EBX: f589d46c ECX: 00000090 EDX: 360c2000
ESI: 00000000 EDI: c1724a80 EBP: f58e9ec0 ESP: f58e9ea0
DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
CR0: 8005003b CR2: 000000ac CR3: 01731000 CR4: 000006d0
In more rich detail you can find that problem on my bug-tracker for Manjaro Linux:
https://github.com/manjaro/packages-core/issues/14
I just want to know if you are aware of it. With current 4.1.3 release I still face that issue ...
kind regards
Philip Müller
--------------------------
Manjaro Project-Lead
[linux41] regression with 'x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure' on AMD i686
Hi Guenter, Sudeep,
It now came down to 'x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure'[1] as you can see here[2].
The facts are:
- You can't boot on i686 only with more than one CPU core on AMD hardware (x86_64 however works)
- Using Ubuntu config[3] it boots but creates an kernel >= 1 GB in size on Manjaro
- By reverting 0d55ba4[4] the kernel boots.
So we have to find out what is causing this issue[5].
kind regards
Philip Müller
--------------------------
Manjaro Project-Lead
[1]
https://github.com/torvalds/linux/commit/0d55ba46bfbee64fd2b492b87bfe2ec172e7b056
[2]
https://raw.githubusercontent.com/philmmanjaro/linux41/master/git-bisect.txt
[3]
https://raw.githubusercontent.com/philmmanjaro/linux41/master/config.4.1.3-040103-generic
[4]
https://github.com/manjaro/packages-core/commit/f7b77f3e84295a6313a9181d520fb48e60453b64
[5] https://github.com/manjaro/packages-core/issues/14
Quoted 30 lines · Philip Müller
Hi all,
I started to test linux 4.1 series with rc6. However, I was never able to boot that kernel in i686 architecture. Trying it again with VirtualBox gave me more conclusions. Using one core it simply boots up. Using more than one CPU core it crashes with:
Failed to access perfctr msr (MSR
c0010007is 0)task:
f58e0000ti:f58e8000task.ti: f58e800
EIP: 0060:[<c135a903>] EFLAGS: 00010206 CPU: 0 EIP is atfree_cache_attributes+0x83/0xd0
EAX: 00000001 EBX:f589d46cECX: 00000090 EDX:360c2000
ESI: 00000000 EDI:c1724a80EBP:f58e9ec0ESP:f58e9ea0
DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
CR0:8005003bCR2:000000acCR3: 01731000 CR4:000006d0In more rich detail you can find that problem on my bug-tracker for Manjaro Linux:
https://github.com/manjaro/packages-core/issues/14
I just want to know if you are aware of it. With current 4.1.3 release I still face that issue ...
kind regards
Philip Müller
--------------------------
Manjaro Project-Lead
Re: [linux41] regression with 'x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure' on AMD i686
Quoted 7 lines · Philip Müller
Quoted 7 lines
task:
f58e0000ti:f58e8000task.ti: f58e800
EIP: 0060:[<c135a903>] EFLAGS: 00010206 CPU: 0 EIP is atfree_cache_attributes+0x83/0xd0
EAX: 00000001 EBX:f589d46cECX: 00000090 EDX:360c2000
ESI: 00000000 EDI:c1724a80EBP:f58e9ec0ESP:f58e9ea0
DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
CR0:8005003bCR2:000000acCR3: 01731000 CR4:000006d0
That's a trivial NULL pointer dereference in the error/cleanup path. Patch below should fix it.
Thanks,
tglx
---
drivers/base/cacheinfo.c +4 -2
@@ -159,6 +159,9 @@ static void cache_shared_cpu_map_remove(unsigned int cpu)static void free_cache_attributes (unsigned int cpu) { +if (!per_cpu_cacheinfo (cpu)) +return ; +cache_shared_cpu_map_remove (cpu);kfree (per_cpu_cacheinfo (cpu));
@@ -514,8 +517,7 @@ static int cacheinfo_cpu_callback(struct notifier_block *nfb,break ;case CPU_DEAD :cache_remove_dev (cpu); -if (per_cpu_cacheinfo (cpu)) -free_cache_attributes (cpu); +free_cache_attributes (cpu);break ; }return notifier_from_errno (rc);
Re: [linux41] regression with 'x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure' on AMD i686
Quoted 11 lines · Thomas Gleixner
Quoted 7 lines · Philip Müller
Quoted 7 lines
task:
f58e0000ti:f58e8000task.ti: f58e800
EIP: 0060:[<c135a903>] EFLAGS: 00010206 CPU: 0 EIP is atfree_cache_attributes+0x83/0xd0
EAX: 00000001 EBX:f589d46cECX: 00000090 EDX:360c2000
ESI: 00000000 EDI:c1724a80EBP:f58e9ec0ESP:f58e9ea0
DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
CR0:8005003bCR2:000000acCR3: 01731000 CR4:000006d0That's a trivial NULL pointer dereference in the error/cleanup path. Patch below should fix it.
Well, I got a bit different, and of course totally untested possible solution:
cache_shared_cpu_map_setup() does check sib_cpu_ci->info_list before setting cpumask bits while cache_shared_cpu_map_remove() doesn't. Ballancing this out would mean:
---
drivers/base/cacheinfo.c +4 -0
@@ -148,7 +148,11 @@ static void cache_shared_cpu_map_remove(unsigned int cpu)if (sibling == cpu)/* skip itself */ continue ; + sib_cpu_ci =get_cpu_cacheinfo (sibling); +if (!sib_cpu_ci->info_list) +continue ; + sib_leaf = sib_cpu_ci->info_list + index;cpumask_clear_cpu (cpu, &sib_leaf->shared_cpu_map);cpumask_clear_cpu (sibling, &this_leaf->shared_cpu_map);
---
Now Philip can have some more fun testing :-)
Signature · 5 lines
Regards/Gruss,
Boris.
ECO tip #101: Trim your mails when you reply.
--Re: [linux41] regression with 'x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure' on AMD i686
Hi Borislav,
I can confirm your patch working. However, it might be good to use yours and Thomas' in combination to solve this properly.
kind regards
Philip
Quoted 41 lines · Borislav Petkov
Quoted 11 lines · Thomas Gleixner
Quoted 7 lines · Philip Müller
Quoted 7 lines
task:
f58e0000ti:f58e8000task.ti: f58e800
EIP: 0060:[<c135a903>] EFLAGS: 00010206 CPU: 0 EIP is atfree_cache_attributes+0x83/0xd0
EAX: 00000001 EBX:f589d46cECX: 00000090 EDX:360c2000
ESI: 00000000 EDI:c1724a80EBP:f58e9ec0ESP:f58e9ea0
DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
CR0:8005003bCR2:000000acCR3: 01731000 CR4:000006d0That's a trivial NULL pointer dereference in the error/cleanup path. Patch below should fix it.
Well, I got a bit different, and of course totally untested possible solution:
cache_shared_cpu_map_setup()does checksib_cpu_ci->info_listbefore setting cpumask bits whilecache_shared_cpu_map_remove()doesn't. Ballancing this out would mean:---
drivers/base/cacheinfo.c +4
-1@@ -148,7 +148,11 @@ static void cache_shared_cpu_map_remove(unsigned int cpu)if (sibling == cpu)/* skip itself */ continue ; + sib_cpu_ci =get_cpu_cacheinfo (sibling); +if (!sib_cpu_ci->info_list) +continue ; + sib_leaf = sib_cpu_ci->info_list + index;cpumask_clear_cpu (cpu, &sib_leaf->shared_cpu_map);cpumask_clear_cpu (sibling, &this_leaf->shared_cpu_map); ---Now Philip can have some more fun testing :-)
Re: [linux41] regression with 'x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure' on AMD i686
Quoted 2 lines · Philip Müller
I can confirm your patch working. However, it might be good to use yours and Thomas' in combination to solve this properly.
Please do not top-post.
We could use Thomas' too although from looking at it, detect_cache_attributes() allocates a per-CPU per_cpu_cacheinfo thing for each CPU. By the time we hit cache_shared_cpu_map_remove() in free_cache_attributes(), those per_cpu_cacheinfo(cpu) things are still allocated. We kfree them in the next step only.
But I like the moving of the check from the CPU hotplug callback to free_cache_attributes().
So I'll merge the two patches and write up a proper commit message, unless someone objects.
I'll add your Tested-by too.
Thanks.
Signature · 5 lines
Regards/Gruss,
Boris.
ECO tip #101: Trim your mails when you reply.
--Re: [linux41] regression with 'x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure' on AMD i686
Hi Borislav,
I'm fine with that decision. I tested your patch alone and the combination with Thomas' changes. Both work to solve this problem.
Do whatever suits best for this matter. Thx to you too for providing solutions so fast.
kind regards
Philip
p.s. what do you mean by top-post?
Am 26.07.2015 um 16:42 schrieb Borislav Petkov:
Quoted 22 lines · Philip Müller
Quoted 2 lines
I can confirm your patch working. However, it might be good to use yours and Thomas' in combination to solve this properly.
Please do not top-post.
We could use Thomas' too although from looking at it,
detect_cache_attributes()allocates a per-CPUper_cpu_cacheinfothing for each CPU. By the time we hitcache_shared_cpu_map_remove()infree_cache_attributes(), thoseper_cpu_cacheinfo(cpu)things are still allocated. We kfree them in the next step only.But I like the moving of the check from the CPU hotplug callback to
free_cache_attributes().So I'll merge the two patches and write up a proper commit message, unless someone objects.
I'll add your Tested-by too.
Thanks.
Re: [linux41] regression with 'x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure' on AMD i686
Quoted 5 lines · Philip Müller
Hi Borislav,
I'm fine with that decision. I tested your patch alone and the combination with Thomas' changes. Both work to solve this problem.
Quoted 8 lines
Do whatever suits best for this matter. Thx to you too for providing solutions so fast.
kind regards
Philipp.s. what do you mean by top-post?
What you just did ;-).
http://ck.wikia.com/wiki/TopPosting
Guenter
Quoted 25 lines
Am 26.07.2015 um 16:42 schrieb Borislav Petkov:
Quoted 22 lines · Philip Müller
Quoted 2 lines
I can confirm your patch working. However, it might be good to use yours and Thomas' in combination to solve this properly.
Please do not top-post.
We could use Thomas' too although from looking at it,
detect_cache_attributes()allocates a per-CPUper_cpu_cacheinfothing for each CPU. By the time we hitcache_shared_cpu_map_remove()infree_cache_attributes(), thoseper_cpu_cacheinfo(cpu)things are still allocated. We kfree them in the next step only.But I like the moving of the check from the CPU hotplug callback to
free_cache_attributes().So I'll merge the two patches and write up a proper commit message, unless someone objects.
I'll add your Tested-by too.
Thanks.
Re: [linux41] regression with 'x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure' on AMD i686
In reply to Borislav PetkovQuoted 12 lines · Borislav Petkov
Quoted 2 lines · Philip Müller
I can confirm your patch working. However, it might be good to use yours and Thomas' in combination to solve this properly.
Please do not top-post.
We could use Thomas' too although from looking at it,
detect_cache_attributes()allocates a per-CPUper_cpu_cacheinfothing for each CPU. By the time we hitcache_shared_cpu_map_remove()infree_cache_attributes(), thoseper_cpu_cacheinfo(cpu)things are still allocated. We kfree them in the next step only.
Quoted 7 lines
But I like the moving of the check from the CPU hotplug callback to
free_cache_attributes().So I'll merge the two patches and write up a proper commit message, unless someone objects.
I'll add your Tested-by too.
Did this actually happen? I don't see either fix in Linus' tree yet, the merge window is closed, and the bug happens on 4.1 and 4.2 stable kernels..
josh
Re: [linux41] regression with 'x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure' on AMD i686
Am 17.09.2015 um 01:52 schrieb Josh Boyer:
Quoted 7 lines
Did this actually happen? I don't see either fix in Linus' tree yet, the merge window is closed, and the bug happens on 4.1 and 4.2 stable kernels..
josh
Seems not yet. I don't see it neither in 4.3-rc1. Seems 4.3 will have the same issues then ...
Re: [linux41] regression with 'x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure' on AMD i686
In reply to Josh BoyerQuoted 16 lines · Josh Boyer
Quoted 15 lines · Borislav Petkov
Quoted 2 lines · Philip Müller
I can confirm your patch working. However, it might be good to use yours and Thomas' in combination to solve this properly.
Please do not top-post.
We could use Thomas' too although from looking at it,
detect_cache_attributes()allocates a per-CPUper_cpu_cacheinfothing for each CPU. By the time we hitcache_shared_cpu_map_remove()infree_cache_attributes(), thoseper_cpu_cacheinfo(cpu)things are still allocated. We kfree them in the next step only.But I like the moving of the check from the CPU hotplug callback to
free_cache_attributes().
Quoted 8 lines
Quoted 4 lines
So I'll merge the two patches and write up a proper commit message, unless someone objects.
I'll add your Tested-by too.
Did this actually happen? I don't see either fix in Linus' tree yet, the merge window is closed, and the bug happens on 4.1 and 4.2 stable kernels..
Greg wanted to pick it up...
Greg, what's up?
Signature · 5 lines
Regards/Gruss,
Boris.
ECO tip #101: Trim your mails when you reply.
--Greg KH
maintainerRe: [linux41] regression with 'x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure' on AMD i686
Quoted 22 lines · Borislav Petkov
Quoted 21 lines · Josh Boyer
Quoted 19 lines · Borislav Petkov
Quoted 2 lines · Philip Müller
I can confirm your patch working. However, it might be good to use yours and Thomas' in combination to solve this properly.
Please do not top-post.
We could use Thomas' too although from looking at it,
detect_cache_attributes()allocates a per-CPUper_cpu_cacheinfothing for each CPU. By the time we hitcache_shared_cpu_map_remove()infree_cache_attributes(), thoseper_cpu_cacheinfo(cpu)things are still allocated. We kfree them in the next step only.But I like the moving of the check from the CPU hotplug callback to
free_cache_attributes().So I'll merge the two patches and write up a proper commit message, unless someone objects.
I'll add your Tested-by too.
Quoted 7 lines
Quoted 3 lines
Did this actually happen? I don't see either fix in Linus' tree yet, the merge window is closed, and the bug happens on 4.1 and 4.2 stable kernels..
Greg wanted to pick it up...
Greg, what's up?
It's in my "to-apply" queue, let me go dig it up now...
thanks for the reminder.
greg k-h
PATCH 1/1 cpu/cacheinfo: Fix teardown path
In reply to Philip MüllerFrom: Borislav Petkov <[email protected]>
Date: Mon, 27 Jul 2015 08:36:27 +0200
Subject: [PATCH] cpu/cacheinfo: Fix teardown path
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Philip Müller reported a hang when booting 32-bit 4.1 kernel on an AMD box. A fragment of the splat was enough to pinpoint the issue:
task: f58e0000 ti: f58e8000 task.ti: f58e800
EIP: 0060:[<c135a903>] EFLAGS: 00010206 CPU: 0 EIP is at free_cache_attributes+0x83/0xd0
EAX: 00000001 EBX: f589d46c ECX: 00000090 EDX: 360c2000
ESI: 00000000 EDI: c1724a80 EBP: f58e9ec0 ESP: f58e9ea0
DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
CR0: 8005003b CR2: 000000ac CR3: 01731000 CR4: 000006d0
cache_shared_cpu_map_setup() did check sibling CPUs cacheinfo descriptor while the respective teardown path cache_shared_cpu_map_remove() didn't. Fix that.
Quoted 1 line
From tglx's version: to be on the safe side, move the cacheinfo
descriptor check to free_cache_attributes(), thus cleaning up the hotplug path a little and making this even more robust.
---
Moin Thomas,
I've merged both patches and tagged it for stable. Which means, tip-urgent.
Thanks.
drivers/base/cacheinfo.c | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-)
drivers/base/cacheinfo.c +8 -2
@@ -148,7 +148,11 @@ static void cache_shared_cpu_map_remove(unsigned int cpu)if (sibling == cpu)/* skip itself */ continue ; + sib_cpu_ci =get_cpu_cacheinfo (sibling); +if (!sib_cpu_ci->info_list) +continue ; + sib_leaf = sib_cpu_ci->info_list + index;cpumask_clear_cpu (cpu, &sib_leaf->shared_cpu_map);cpumask_clear_cpu (sibling, &this_leaf->shared_cpu_map);
@@ -159,6 +163,9 @@ static void cache_shared_cpu_map_remove(unsigned int cpu)static void free_cache_attributes (unsigned int cpu) { +if (!per_cpu_cacheinfo (cpu)) +return ; +cache_shared_cpu_map_remove (cpu);kfree (per_cpu_cacheinfo (cpu));
@@ -514,8 +521,7 @@ static int cacheinfo_cpu_callback(struct notifier_block *nfb,break ;case CPU_DEAD :cache_remove_dev (cpu); -if (per_cpu_cacheinfo (cpu)) -free_cache_attributes (cpu); +free_cache_attributes (cpu);break ; }return notifier_from_errno (rc);
Signature · 8 lines
2.5.0.rc2.28.g6003e7f
--
Regards/Gruss,
Boris.
ECO tip #101: Trim your mails when you reply.
--Re: [PATCH] cpu/cacheinfo: Fix teardown path
Quoted 26 lines · Borislav Petkov
From: Borislav Petkov <[email protected]>
Date: Mon, 27 Jul 2015 08:36:27 +0200
Subject: [PATCH] cpu/cacheinfo: Fix teardown path
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bitPhilip Müller reported a hang when booting 32-bit 4.1 kernel on an AMD box. A fragment of the splat was enough to pinpoint the issue:
task: f58e0000 ti: f58e8000 task.ti: f58e800 EIP: 0060:[<c135a903>] EFLAGS: 00010206 CPU: 0 EIP is at free_cache_attributes+0x83/0xd0 EAX: 00000001 EBX: f589d46c ECX: 00000090 EDX: 360c2000 ESI: 00000000 EDI: c1724a80 EBP: f58e9ec0 ESP: f58e9ea0 DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068 CR0: 8005003b CR2: 000000ac CR3: 01731000 CR4: 000006d0
cache_shared_cpu_map_setup()did check sibling CPUs cacheinfo descriptor while the respective teardown pathcache_shared_cpu_map_remove()didn't. Fix that.From tglx's version: to be on the safe side, move the cacheinfo
descriptor check tofree_cache_attributes(), thus cleaning up the hotplug path a little and making this even more robust.
Quoted 10 lines
Reported-by Philip Müller <[email protected]>Cc <[email protected]> # 4.1Cc Andre Przywara <[email protected]>Cc Guenter Roeck <[email protected]>Cc "H. Peter Anvin" <[email protected]>Cc Ingo Molnar <[email protected]>Cc Philip Müller <[email protected]>Cc Sudeep Holla <[email protected]>
Looks good to me. If not too late
Regards,
Sudeep
Re: [PATCH] cpu/cacheinfo: Fix teardown path
In reply to Borislav PetkovQuoted 39 lines · Borislav Petkov
From: Borislav Petkov <[email protected]>
Date: Mon, 27 Jul 2015 08:36:27 +0200
Subject: [PATCH] cpu/cacheinfo: Fix teardown path
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bitPhilip Müller reported a hang when booting 32-bit 4.1 kernel on an AMD box. A fragment of the splat was enough to pinpoint the issue:
task:
f58e0000ti:f58e8000task.ti: f58e800
EIP: 0060:[<c135a903>] EFLAGS: 00010206 CPU: 0 EIP is atfree_cache_attributes+0x83/0xd0
EAX: 00000001 EBX:f589d46cECX: 00000090 EDX:360c2000
ESI: 00000000 EDI:c1724a80EBP:f58e9ec0ESP:f58e9ea0
DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
CR0:8005003bCR2:000000acCR3: 01731000 CR4:000006d0
cache_shared_cpu_map_setup()did check sibling CPUs cacheinfo descriptor while the respective teardown pathcache_shared_cpu_map_remove()didn't. Fix that.Quoted 1 line
From tglx's version: to be on the safe side, move the cacheinfo
descriptor check to
free_cache_attributes(), thus cleaning up the hotplug path a little and making this even more robust.Reported-by Philip Müller <[email protected]>Cc <[email protected]> # 4.1Cc Andre Przywara <[email protected]>Cc Guenter Roeck <[email protected]>Cc "H. Peter Anvin" <[email protected]>Cc Ingo Molnar <[email protected]>Cc Philip Müller <[email protected]>Cc Sudeep Holla <[email protected]>Cc Thomas Gleixner <[email protected]>Signed-off-by Borislav Petkov <[email protected]>
Quoted 6 lines
---
Moin Thomas,
I've merged both patches and tagged it for stable. Which means, tip-urgent.
Philip Müller
Acked-byRe: [PATCH] cpu/cacheinfo: Fix teardown path
Am 27.07.2015 um 13:10 schrieb Thomas Gleixner:
Quoted 9 lines
Quoted 6 lines
---
Moin Thomas,
I've merged both patches and tagged it for stable. Which means, tip-urgent.
Reviewed-by Thomas Gleixner <[email protected]>
Hi Borislav,
I also reviewed your new code and also tested it.
tip-bot for Borislav Petkov
merge bot[tip:x86/urgent] x86/cpu/cacheinfo: Fix teardown path
In reply to Borislav Petkovx86/cpu/cacheinfo: Fix teardown path
Commit-ID: 680ac028240f8747f31c03986fbcf18b2b521e93 Gitweb: http://git.kernel.org/tip/680ac028240f8747f31c03986fbcf18b2b521e93 Author: Borislav Petkov <[email protected]> AuthorDate: Mon, 27 Jul 2015 09:58:05 +0200 Committer: Ingo Molnar <[email protected]> CommitDate: Wed, 5 Aug 2015 10:08:17 +0200
x86/cpu/cacheinfo: Fix teardown path
Philip Müller reported a hang when booting 32-bit 4.1 kernel on an AMD box. A fragment of the splat was enough to pinpoint the issue:
task: f58e0000 ti: f58e8000 task.ti: f58e800
EIP: 0060:[<c135a903>] EFLAGS: 00010206 CPU: 0 EIP is at free_cache_attributes+0x83/0xd0
EAX: 00000001 EBX: f589d46c ECX: 00000090 EDX: 360c2000
ESI: 00000000 EDI: c1724a80 EBP: f58e9ec0 ESP: f58e9ea0
DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
CR0: 8005003b CR2: 000000ac CR3: 01731000 CR4: 000006d0
cache_shared_cpu_map_setup() did check sibling CPUs cacheinfo descriptor while the respective teardown path cache_shared_cpu_map_remove() didn't. Fix that.
Quoted 1 line
From tglx's version: to be on the safe side, move the cacheinfo
descriptor check to free_cache_attributes(), thus cleaning up the hotplug path a little and making this even more robust.
--- drivers/base/cacheinfo.c | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-)
drivers/base/cacheinfo.c +8 -2
@@ -148,7 +148,11 @@ static void cache_shared_cpu_map_remove(unsigned int cpu)if (sibling == cpu)/* skip itself */ continue ; + sib_cpu_ci =get_cpu_cacheinfo (sibling); +if (!sib_cpu_ci->info_list) +continue ; + sib_leaf = sib_cpu_ci->info_list + index;cpumask_clear_cpu (cpu, &sib_leaf->shared_cpu_map);cpumask_clear_cpu (sibling, &this_leaf->shared_cpu_map);
@@ -159,6 +163,9 @@ static void cache_shared_cpu_map_remove(unsigned int cpu)static void free_cache_attributes (unsigned int cpu) { +if (!per_cpu_cacheinfo (cpu)) +return ; +cache_shared_cpu_map_remove (cpu);kfree (per_cpu_cacheinfo (cpu));
@@ -514,8 +521,7 @@ static int cacheinfo_cpu_callback(struct notifier_block *nfb,break ;case CPU_DEAD :cache_remove_dev (cpu); -if (per_cpu_cacheinfo (cpu)) -free_cache_attributes (cpu); +free_cache_attributes (cpu);break ; }return notifier_from_errno (rc);
Re: [PATCH] cpu/cacheinfo: Fix teardown path
In reply to Borislav PetkovQuoted 2 lines · Borislav Petkov
Philip Müller reported a hang when booting 32-bit 4.1 kernel on an AMD box. A fragment of the splat was enough to pinpoint the issue:
Bah, this goes to Greg and not to tip. Anyway, here's a version with updated tags.
Greg, please queue for 4.2 as it fixes a hang.
Thanks.
---
From: Borislav Petkov <[email protected]>
Date: Mon, 27 Jul 2015 08:36:27 +0200
Subject: [PATCH] cpu/cacheinfo: Fix teardown path
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Philip Müller reported a hang when booting 32-bit 4.1 kernel on an AMD box. A fragment of the splat was enough to pinpoint the issue:
task: f58e0000 ti: f58e8000 task.ti: f58e800
EIP: 0060:[<c135a903>] EFLAGS: 00010206 CPU: 0 EIP is at free_cache_attributes+0x83/0xd0
EAX: 00000001 EBX: f589d46c ECX: 00000090 EDX: 360c2000
ESI: 00000000 EDI: c1724a80 EBP: f58e9ec0 ESP: f58e9ea0
DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
CR0: 8005003b CR2: 000000ac CR3: 01731000 CR4: 000006d0
cache_shared_cpu_map_setup() did check sibling CPUs cacheinfo descriptor while the respective teardown path cache_shared_cpu_map_remove() didn't. Fix that.
Quoted 1 line
From tglx's version: to be on the safe side, move the cacheinfo
descriptor check to free_cache_attributes(), thus cleaning up the hotplug path a little and making this even more robust.
--- drivers/base/cacheinfo.c | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-)
drivers/base/cacheinfo.c +8 -2
@@ -148,7 +148,11 @@ static void cache_shared_cpu_map_remove(unsigned int cpu)if (sibling == cpu)/* skip itself */ continue ; + sib_cpu_ci =get_cpu_cacheinfo (sibling); +if (!sib_cpu_ci->info_list) +continue ; + sib_leaf = sib_cpu_ci->info_list + index;cpumask_clear_cpu (cpu, &sib_leaf->shared_cpu_map);cpumask_clear_cpu (sibling, &this_leaf->shared_cpu_map);
@@ -159,6 +163,9 @@ static void cache_shared_cpu_map_remove(unsigned int cpu)static void free_cache_attributes (unsigned int cpu) { +if (!per_cpu_cacheinfo (cpu)) +return ; +cache_shared_cpu_map_remove (cpu);kfree (per_cpu_cacheinfo (cpu));
@@ -514,8 +521,7 @@ static int cacheinfo_cpu_callback(struct notifier_block *nfb,break ;case CPU_DEAD :cache_remove_dev (cpu); -if (per_cpu_cacheinfo (cpu)) -free_cache_attributes (cpu); +free_cache_attributes (cpu);break ; }return notifier_from_errno (rc);
Signature · 8 lines
2.5.0.rc2.28.g6003e7f
--
Regards/Gruss,
Boris.
ECO tip #101: Trim your mails when you reply.
--Greg KH
maintainerRe: [PATCH] cpu/cacheinfo: Fix teardown path
Quoted 8 lines · Borislav Petkov
Quoted 2 lines · Borislav Petkov
Philip Müller reported a hang when booting 32-bit 4.1 kernel on an AMD box. A fragment of the splat was enough to pinpoint the issue:
Bah, this goes to Greg and not to tip. Anyway, here's a version with updated tags.
Greg, please queue for 4.2 as it fixes a hang.
What commit caused this issue?
And it's a bit late for 4.2, as you say 4.1 is also affected, I'll wait for 4.3-rc1 to give this a chance to get some testing.
thanks,
greg k-h
Re: [PATCH] cpu/cacheinfo: Fix teardown path
Hi Greg,
I bi-sected it to following commit:
0d55ba46bfbee64fd2b492b87bfe2ec172e7b056 is the first bad commit commit 0d55ba46bfbee64fd2b492b87bfe2ec172e7b056
Author: Sudeep Holla <[email protected]> Date: Wed Mar 4 12:00:16 2015 +0000
x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure
You can follow it on my github repo in rich detail:
https://github.com/philmmanjaro/linux41/blob/master/git-bisect.txt
kind regards
Philip Müller
Am 08.08.2015 um 17:41 schrieb Greg KH:
Quoted 19 lines · Borislav Petkov
Quoted 8 lines · Borislav Petkov
Quoted 2 lines
Philip Müller reported a hang when booting 32-bit 4.1 kernel on an AMD box. A fragment of the splat was enough to pinpoint the issue:
Bah, this goes to Greg and not to tip. Anyway, here's a version with updated tags.
Greg, please queue for 4.2 as it fixes a hang.
What commit caused this issue?
And it's a bit late for 4.2, as you say 4.1 is also affected, I'll wait for 4.3-rc1 to give this a chance to get some testing.
thanks,
greg k-h
Re: [PATCH] cpu/cacheinfo: Fix teardown path
Quoted 6 lines · Philip Müller
Hi Greg,
I bi-sected it to following commit:
0d55ba46bfbee64fd2b492b87bfe2ec172e7b056is the first bad commit commit0d55ba46bfbee64fd2b492b87bfe2ec172e7b056
Quoted 8 lines
Author: Sudeep Holla <[email protected]> Date: Wed Mar 4 12:00:16 2015 +0000x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure
You can follow it on my github repo in rich detail:
https://github.com/philmmanjaro/linux41/blob/master/git-bisect.txt
Philip,
what is with you and top-posting? How hard is it not to do it?!
Please stop with the top-posting already.
Signature · 5 lines
Regards/Gruss,
Boris.
ECO tip #101: Trim your mails when you reply.
--Re: [PATCH] cpu/cacheinfo: Fix teardown path
In reply to Greg KHQuoted 1 line · Greg KH
What commit caused this issue?
Apparently
0d55ba46bfbe ("x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure")
Looks like moving x86 to the generic cacheinfo stuff uncovered this shortcoming there...
Quoted 2 lines
And it's a bit late for 4.2, as you say 4.1 is also affected, I'll wait for 4.3-rc1 to give this a chance to get some testing.
Right, I guess that's fine too as it'll trickle to stable eventually...
Thanks.
Signature · 5 lines
Regards/Gruss,
Boris.
ECO tip #101: Trim your mails when you reply.
--Re: [PATCH] cpu/cacheinfo: Fix teardown path
Quoted 10 lines · Borislav Petkov
Quoted 1 line · Greg KH
What commit caused this issue?
Apparently
0d55ba46bfbe("x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure")Looks like moving x86 to the generic cacheinfo stuff uncovered this shortcoming there...
Quoted 7 lines
Quoted 2 lines
And it's a bit late for 4.2, as you say 4.1 is also affected, I'll wait for 4.3-rc1 to give this a chance to get some testing.
Right, I guess that's fine too as it'll trickle to stable eventually...
Thanks.
Just a note from my end. Seems this patch didn't made it into 4.3-rc1. Any reason why?
drivers/base/cacheinfo.c +8 -2
@@ -148,7 +148,11 @@ static void cache_shared_cpu_map_remove(unsigned int cpu)if (sibling == cpu)/* skip itself */ continue ; + sib_cpu_ci =get_cpu_cacheinfo (sibling); +if (!sib_cpu_ci->info_list) +continue ; + sib_leaf = sib_cpu_ci->info_list + index;cpumask_clear_cpu (cpu, &sib_leaf->shared_cpu_map);cpumask_clear_cpu (sibling, &this_leaf->shared_cpu_map);
@@ -159,6 +163,9 @@ static void cache_shared_cpu_map_remove(unsigned int cpu)static void free_cache_attributes (unsigned int cpu) { +if (!per_cpu_cacheinfo (cpu)) +return ; +cache_shared_cpu_map_remove (cpu);kfree (per_cpu_cacheinfo (cpu));
@@ -514,8 +521,7 @@ static int cacheinfo_cpu_callback(struct notifier_block *nfb,break ;case CPU_DEAD :cache_remove_dev (cpu); -if (per_cpu_cacheinfo (cpu)) -free_cache_attributes (cpu); +free_cache_attributes (cpu);break ; }return notifier_from_errno (rc);