slub crashes with recent -git
Subsystems (3)
PATCH 1/1 slub crashes with recent -git
Linus, Christoph,
recent slub commits in -git cause this bootup crash:
Freeing unused kernel memory: 324k freed Write protecting the kernel read-only data: 1294k
------------[ cut here ]------------
kernel BUG at mm/slub.c:2401!
invalid opcode: 0000 [#1]
PREEMPT SMP
Modules linked in:
CPU: 0
EIP: 0060:[<c017dac3>] Not tainted VLI
EFLAGS: 00010046 (2.6.22 #1)
EIP is at ksize+0x13/0x42
eax: 00000000 ebx: 00000000 ecx: 00000020 edx: 00000000
esi: f76a4000 edi: 00000004 ebp: f7b11e74 esp: f7b11e74
ds: 007b es: 007b fs: 00d8 gs: 0033 ss: 0068
Process udevd (pid: 824, ti=f7b11000 task=f7ca5000 task.ti=f7b11000)
Stack: f7b11e94 c016c28b f768cb00 00000020 f7ca5000 00000004 f76a4000 fffffff4
f7b11eb4 c03cf158 00000002 f76a4000 00000020 f768cb00 f76a4000 f7b11ed8
f7b11ed0 c03cfbc6 f768cb00 f768cb00 c046bf80 f768cb00 0000000c f7b11f6c
Call Trace:
[<c0105e3e>] show_trace_log_lvl+0x19/0x2e
[<c0105ef0>] show_stack_log_lvl+0x9d/0xa5
[<c010628f>] show_registers+0x1f5/0x334
[<c01064e6>] die+0x118/0x1fc
[<c0426e7f>] do_trap+0x8e/0xa8
[<c0106ac3>] do_invalid_op+0x88/0x92
[<c0426a92>] error_code+0x72/0x78
[<c016c28b>] krealloc+0x27/0x6d
[<c03cf158>] netlink_realloc_groups+0x61/0xd9
[<c03cfbc6>] netlink_bind+0x4f/0x121
[<c03afe8d>] sys_bind+0x67/0x86
[<c03b11e3>] sys_socketcall+0x8f/0x244
[<c0104ef2>] sysenter_past_esp+0x6b/0xb5
=======================
Code: 40 02 00 75 03 8b 52 0c 8b 02 5d 84 c0 b8 00 00 00 00 0f 49 d0 89 d0 c3 55 31 d2 83 f8 10 89 e5 74 34 e8 bc ff ff ff 85 c0 75 04 <0f> 0b eb fe 8b 40 10 85 c0 75 04 0f 0b eb fe 8b 10 f6 c6 0c 74 i had to apply the patch below to make the kernel boot again.
mm/slub.c +1 -1
@@ -2394,7 +2394,7 @@ size_t ksize(const void *object)struct page *page;struct kmem_cache *s; -if (object ==ZERO_SIZE_PTR ) +if (object ==ZERO_SIZE_PTR || !object)return 0 ; page =get_object_page (object);
Quoted 2 lines · Ingo Molnar
i had to apply the patch below to make the kernel boot again.
A better patch should be the appended. Does that work for you too?
Linus
--- mm/slub.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
mm/slub.c +1 -1
@@ -2394,7 +2394,7 @@ size_t ksize(const void *object)struct page *page;struct kmem_cache *s; -if (object ==ZERO_SIZE_PTR ) +if (ZERO_OR_NULL_PTR (object))return 0 ; page =get_object_page (object);
Ingo Molnar
Acked-byQuoted 5 lines · Linus Torvalds
Quoted 2 lines · Ingo Molnar
i had to apply the patch below to make the kernel boot again.
A better patch should be the appended. Does that work for you too?
yeah, works fine!
Ingo
Quoted 2 lines · Linus Torvalds
A better patch should be the appended. Does that work for you too?
Btw, I already committed this as obvious.
I did the same for the SLAB __do_kmalloc() thing. Let's hope that that was the extent of the damage.
Linus
Quoted 8 lines · Linus Torvalds
Quoted 2 lines · Linus Torvalds
A better patch should be the appended. Does that work for you too?
Btw, I already committed this as obvious.
I did the same for the SLAB
__do_kmalloc()thing. Let's hope that that was the extent of the damage.
My 32bit crashes seem to be gone with latest git head, both for slab and for slub. Even the serial console seems to work again. Thanks
-Andi
Quoted 4 lines · On Thu, 19 Jul 2007 13:25:07 -0700 (PDT) Linus Torvalds…
Quoted 1 line · Linus Torvalds
Quoted 8 lines
Quoted 1 line
A better patch should be the appended. Does that work for you too?
Btw, I already committed this as obvious.
I did the same for the SLAB
__do_kmalloc()thing. Let's hope that that was the extent of the damage.Linus
Hmmmm.. The issue is really in krealloc which can be called with a NULL parameter (a special case). However, krealloc should not call ksize with NULL.
The merged patch above makes ksize(NULL) return 0. So we are returning zero size for an object that we have not allocated. Better fail if someone tries that.
The __do_kmalloc issue looks like a hunk that was somehow dropped.
IMHO: The right fix for the ksize issue would be the following patch:
mm/util.c +5 -1
@@ -88,7 +88,11 @@ void *krealloc(const void *p, size_t newreturn ZERO_SIZE_PTR ; } - ks =ksize (p); +if (p) + ks =ksize (p); +else + ks =0 ; +if (ks >= new_size)return (void *)p;
Quoted 6 lines · Ingo Molnar
Linus, Christoph,
recent slub commits in -git cause this bootup crash:
Freeing unused kernel memory: 324k freed Write protecting the kernel read-only data: 1294k
Just curious, are the crashes even possible in 2.6.22.1? (I see the same patchable code snippet in the source). Just wondering if I should also apply this to 2.6.21.1-rt4...
-- Fernando
Quoted 55 lines
------------[ cut here ]------------ kernel BUG at mm/slub.c:2401! invalid opcode: 0000 [#1] PREEMPT SMP Modules linked in: CPU: 0 EIP: 0060:[<c017dac3>] Not tainted VLI EFLAGS: 00010046 (2.6.22 #1) EIP is at ksize+0x13/0x42 eax: 00000000 ebx: 00000000 ecx: 00000020 edx: 00000000 esi: f76a4000 edi: 00000004 ebp: f7b11e74 esp: f7b11e74 ds: 007b es: 007b fs: 00d8 gs: 0033 ss: 0068 Process udevd (pid: 824, ti=f7b11000 task=f7ca5000 task.ti=f7b11000) Stack: f7b11e94 c016c28b f768cb00 00000020 f7ca5000 00000004 f76a4000 fffffff4 f7b11eb4 c03cf158 00000002 f76a4000 00000020 f768cb00 f76a4000 f7b11ed8 f7b11ed0 c03cfbc6 f768cb00 f768cb00 c046bf80 f768cb00 0000000c f7b11f6c Call Trace: [<c0105e3e>] show_trace_log_lvl+0x19/0x2e [<c0105ef0>] show_stack_log_lvl+0x9d/0xa5 [<c010628f>] show_registers+0x1f5/0x334 [<c01064e6>] die+0x118/0x1fc [<c0426e7f>] do_trap+0x8e/0xa8 [<c0106ac3>] do_invalid_op+0x88/0x92 [<c0426a92>] error_code+0x72/0x78 [<c016c28b>] krealloc+0x27/0x6d [<c03cf158>] netlink_realloc_groups+0x61/0xd9 [<c03cfbc6>] netlink_bind+0x4f/0x121 [<c03afe8d>] sys_bind+0x67/0x86 [<c03b11e3>] sys_socketcall+0x8f/0x244 [<c0104ef2>] sysenter_past_esp+0x6b/0xb5 ======================= Code: 40 02 00 75 03 8b 52 0c 8b 02 5d 84 c0 b8 00 00 00 00 0f 49 d0 89 d0 c3 55 31 d2 83 f8 10 89 e5 74 34 e8 bc ff ff ff 85 c0 75 04 <0f> 0b eb fe 8b 40 10 85 c0 75 04 0f 0b eb fe 8b 10 f6 c6 0c 74i had to apply the patch below to make the kernel boot again.
Signed-off-by Ingo Molnar <[email protected]>Index:
linux/mm/slub.c
===================================================================mm/slub.c +1
-2@@ -2394,7 +2394,7 @@ size_t ksize(const void *object)struct page *page;struct kmem_cache *s; -if (object ==ZERO_SIZE_PTR ) +if (object ==ZERO_SIZE_PTR || !object)return 0 ; page =get_object_page (object); -To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Quoted 11 lines · Fernando Lopez-Lezcano
Quoted 6 lines · Ingo Molnar
Linus, Christoph,
recent slub commits in -git cause this bootup crash:
Freeing unused kernel memory: 324k freed Write protecting the kernel read-only data: 1294k
Just curious, are the crashes even possible in 2.6.22.1? (I see the same patchable code snippet in the source). Just wondering if I should also apply this to 2.6.21.1-rt4...
no, i think they are unique to .23, due to the recent stricter behavior introduced for zero-sized allocs.
Ingo