Skip to contentSkip to conversation

2.6.12-rc3-mm3: Kernel BUG at "mm/slab.c":1219

PATCH RESEND In review Rafael J. Wysocki 6 participants 15 messages May 9, 2005 – May 10, 2005

lore
Share & tools
lore
#0 lore raw

PATCH RESEND 1/1 2.6.12-rc3-mm3: Kernel BUG at "mm/slab.c":1219

Hi,

I get this from 2.6.12-rc3-mm3 on a UP AMD64 box (Asus L5D), 100% of the time:

]--snip--[
ACPI: bus type pci registered
PCI: Using configuration type 1
mtrr: v2.0 (20020519)
kmem_cache_create: Early error in slab <NULL> ----------- [cut here ] --------- [please bite here ] ---------

Kernel BUG at "mm/slab.c":1219
invalid operand: 0000 [1]
CPU 0
Modules linked in:
Pid: 1, comm: swapper Not tainted 2.6.12-rc3-mm3
RIP: 0010:[<ffffffff80179eeb>] <ffffffff80179eeb>{kmem_cache_create+139}
RSP: 0000:ffff810001ca1eb8  EFLAGS: 00010292
RAX: 0000000000000034 RBX: 0000000000000000 RCX: 0000000000000000
RDX: 0000000000000000 RSI: 0000000000000dd3 RDI: ffffffff804167e0
RBP: 0000000000000005 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000010 R11: 0000000000000008 R12: 0000000000042000
R13: 0000000000000000 R14: 0000ffffffff8010 R15: 0000000000000000
FS:  0000000000000000(0000) GS:ffffffff8055a840(0000) knlGS:0000000000000000
CS:  0010 DS: 0018 ES: 0018 CR0: 000000008005003b
CR2: 0000000000004000 CR3: 0000000000101000 CR4: 00000000000006e0
Process swapper (pid: 1, threadinfo ffff810001ca0000, task ffff810001c5a7a0)
Stack: fffffffffffffff8 0000000000000000 0000000000000000 0000000000000000
       0000000000000010 0000000000000000 0000000000000005 0000000000000006
       00000000ffffffff 0000ffffffff8010
Call Trace:<ffffffff8057a11d>{init_bio+93} <ffffffff8010c0f2>{init+178}
       <ffffffff8010fc37>{child_rip+8} <ffffffff8010c040>{init+0}
       <ffffffff8010fc2f>{child_rip+0}

Greets,
Rafael

Signature · 3 lines
- Would you tell me, please, which way I ought to go from here?
- That depends a good deal on where you want to get to.
		-- Lewis Carroll "Alice's Adventures in Wonderland"
#1 lore raw
Quoted 3 lines · Rafael J. Wysocki

Hi,

I get this from 2.6.12-rc3-mm3 on a UP AMD64 box (Asus L5D), 100% of the time:

Probably a generic bug. Block layer is passing some slab flag slab doesn't like.

-Andi

Quoted 39 lines

]--snip--[
ACPI: bus type pci registered
PCI: Using configuration type 1
mtrr: v2.0 (20020519)
kmem_cache_create: Early error in slab <NULL> ----------- [cut here ] --------- [please bite here ] ---------

Kernel BUG at "mm/slab.c":1219
invalid operand: 0000 [1]
CPU 0
Modules linked in:
Pid: 1, comm: swapper Not tainted 2.6.12-rc3-mm3
RIP: 0010:[<ffffffff80179eeb>] <ffffffff80179eeb>{kmem_cache_create+139}
RSP: 0000:ffff810001ca1eb8  EFLAGS: 00010292
RAX: 0000000000000034 RBX: 0000000000000000 RCX: 0000000000000000
RDX: 0000000000000000 RSI: 0000000000000dd3 RDI: ffffffff804167e0
RBP: 0000000000000005 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000010 R11: 0000000000000008 R12: 0000000000042000
R13: 0000000000000000 R14: 0000ffffffff8010 R15: 0000000000000000
FS:  0000000000000000(0000) GS:ffffffff8055a840(0000) knlGS:0000000000000000
CS:  0010 DS: 0018 ES: 0018 CR0: 000000008005003b
CR2: 0000000000004000 CR3: 0000000000101000 CR4: 00000000000006e0
Process swapper (pid: 1, threadinfo ffff810001ca0000, task ffff810001c5a7a0)
Stack: fffffffffffffff8 0000000000000000 0000000000000000 0000000000000000
       0000000000000010 0000000000000000 0000000000000005 0000000000000006
       00000000ffffffff 0000ffffffff8010
Call Trace:<ffffffff8057a11d>{init_bio+93} <ffffffff8010c0f2>{init+178}
       <ffffffff8010fc37>{child_rip+8} <ffffffff8010c040>{init+0}
       <ffffffff8010fc2f>{child_rip+0}

Greets,
Rafael

Signature · 3 lines
- Would you tell me, please, which way I ought to go from here?
- That depends a good deal on where you want to get to.
		-- Lewis Carroll "Alice's Adventures in Wonderland"
#2 lore raw
Quoted 7 lines · Andi Kleen
Quoted 3 lines · Rafael J. Wysocki

Hi,

I get this from 2.6.12-rc3-mm3 on a UP AMD64 box (Asus L5D), 100% of the time:

Probably a generic bug. Block layer is passing some slab flag slab doesn't like.

Some slab change, perhaps? There's nothing special about the init_bio() slab call:

bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0,
SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);

Hmm, this looks strange. That bug happens if:

        if ((!name) ||
                in_interrupt() ||
                (size < BYTES_PER_WORD) ||
                (size > (1<<MAX_OBJ_ORDER)*PAGE_SIZE) ||
                (dtor && !ctor)) {
                        printk(KERN_ERR "%s: Early error in slab %s\n",
                                        __FUNCTION__, name);
                        BUG();
                }

It must be in_interrupt() triggering, perhaps something change in the boot sequence?

Signature · 1 line
Jens Axboe
#3 lore raw

tis 2005-05-10 klockan 08:18 +0200 skrev Jens Axboe:

Quoted 28 lines · Andi Kleen
Quoted 7 lines · Rafael J. Wysocki
Quoted 3 lines

Hi,

I get this from 2.6.12-rc3-mm3 on a UP AMD64 box (Asus L5D), 100% of the time:

Probably a generic bug. Block layer is passing some slab flag slab doesn't like.

Some slab change, perhaps? There's nothing special about the init_bio() slab call:

bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0,
SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);

Hmm, this looks strange. That bug happens if:

        if ((!name) ||
                in_interrupt() ||
                (size < BYTES_PER_WORD) ||
                (size > (1<<MAX_OBJ_ORDER)*PAGE_SIZE) ||
                (dtor && !ctor)) {
                        printk(KERN_ERR "%s: Early error in slab %s\n",
                                        __FUNCTION__, name);
                        BUG();
                }
Quoted 3 lines

It must be in_interrupt() triggering, perhaps something change in the boot sequence?

The funny thing is that it seems to be the name being NULL, from the original post:
kmem_cache_create: Early error in slab <NULL>

When looking at the code I really can't see how that can be. Rafael, what setup is this compiled under?

#4 lore raw
Quoted 33 lines · Alexander Nyberg

tis 2005-05-10 klockan 08:18 +0200 skrev Jens Axboe:

Quoted 28 lines · Andi Kleen
Quoted 7 lines · Rafael J. Wysocki
Quoted 3 lines

Hi,

I get this from 2.6.12-rc3-mm3 on a UP AMD64 box (Asus L5D), 100% of the time:

Probably a generic bug. Block layer is passing some slab flag slab doesn't like.

Some slab change, perhaps? There's nothing special about the init_bio() slab call:

bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0,
SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);

Hmm, this looks strange. That bug happens if:

        if ((!name) ||
                in_interrupt() ||
                (size < BYTES_PER_WORD) ||
                (size > (1<<MAX_OBJ_ORDER)*PAGE_SIZE) ||
                (dtor && !ctor)) {
                        printk(KERN_ERR "%s: Early error in slab %s\n",
                                        __FUNCTION__, name);
                        BUG();
                }
Quoted 3 lines

It must be in_interrupt() triggering, perhaps something change in the boot sequence?

Quoted 6 lines

The funny thing is that it seems to be the name being NULL, from the original post:
kmem_cache_create: Early error in slab <NULL>

When looking at the code I really can't see how that can be. Rafael, what setup is this compiled under?

Huh indeed, smells like bad code generation or hardware.

Signature · 1 line
Jens Axboe
#5 lore raw
In reply to Jens Axboe
Quoted 19 lines

Some slab change, perhaps? There's nothing special about the init_bio() slab call:

bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0,
SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);

Hmm, this looks strange. That bug happens if:

        if ((!name) ||
                in_interrupt() ||
                (size < BYTES_PER_WORD) ||
                (size > (1<<MAX_OBJ_ORDER)*PAGE_SIZE) ||
                (dtor && !ctor)) {
                        printk(KERN_ERR "%s: Early error in slab %s\n",
                                        __FUNCTION__, name);
                        BUG();
                }
Quoted 2 lines

It must be in_interrupt() triggering, perhaps something change in the boot sequence?

I would add a printk for all arguments on top of kmem_cache_create and decompose the big if () into smaller if (...) BUG() so that you can see which condition triggers exactly.

Then if you suspect compiler issues you can edit the Makefile and recompile the kernel with -O1

Rafael, can you try that perhaps?

-Andi

#6 lore raw
In reply to Rafael J. Wysocki
Quoted 23 lines · Rafael J. Wysocki

I get this from 2.6.12-rc3-mm3 on a UP AMD64 box (Asus L5D), 100% of the time:

]--snip--[
ACPI: bus type pci registered
PCI: Using configuration type 1
mtrr: v2.0 (20020519)
kmem_cache_create: Early error in slab <NULL> ----------- [cut here ] --------- [please bite here ] ---------

Kernel BUG at "mm/slab.c":1219
invalid operand: 0000 [1]
CPU 0
Modules linked in:
Pid: 1, comm: swapper Not tainted 2.6.12-rc3-mm3
RIP: 0010:[<ffffffff80179eeb>] <ffffffff80179eeb>{kmem_cache_create+139}
RSP: 0000:ffff810001ca1eb8  EFLAGS: 00010292
RAX: 0000000000000034 RBX: 0000000000000000 RCX: 0000000000000000
RDX: 0000000000000000 RSI: 0000000000000dd3 RDI: ffffffff804167e0
RBP: 0000000000000005 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000010 R11: 0000000000000008 R12: 0000000000042000
R13: 0000000000000000 R14: 0000ffffffff8010 R15: 0000000000000000
FS:  0000000000000000(0000) GS:ffffffff8055a840(0000) knlGS:0000000000000000
CS:  0010 DS: 0018 ES: 0018 CR0: 000000008005003b
Quoted 9 lines
CR2: 0000000000004000 CR3: 0000000000101000 CR4: 00000000000006e0
Process swapper (pid: 1, threadinfo ffff810001ca0000, task ffff810001c5a7a0)
Stack: fffffffffffffff8 0000000000000000 0000000000000000 0000000000000000
       0000000000000010 0000000000000000 0000000000000005 0000000000000006
       00000000ffffffff 0000ffffffff8010
Call Trace:<ffffffff8057a11d>{init_bio+93} <ffffffff8010c0f2>{init+178}
       <ffffffff8010fc37>{child_rip+8} <ffffffff8010c040>{init+0}
       <ffffffff8010fc2f>{child_rip+0}

Something kooky is happening.

Clearly init_bio() is not passing in a NULL `name' parameter. Maybe the backtrace is screwed due to dopey gcc autoinlining and the bad caller is really biovec_init_slabs(). Try removing the __cacheline_aligned_mostly_readonly from the declaration of bvec_slabs[].

Please tell us what gcc and binutils versions you're using.

#7 lore raw

Hi,

Quoted 34 lines · Andrew Morton
Quoted 23 lines · Rafael J. Wysocki

I get this from 2.6.12-rc3-mm3 on a UP AMD64 box (Asus L5D), 100% of the time:

]--snip--[
ACPI: bus type pci registered
PCI: Using configuration type 1
mtrr: v2.0 (20020519)
kmem_cache_create: Early error in slab <NULL> ----------- [cut here ] --------- [please bite here ] ---------

Kernel BUG at "mm/slab.c":1219
invalid operand: 0000 [1]
CPU 0
Modules linked in:
Pid: 1, comm: swapper Not tainted 2.6.12-rc3-mm3
RIP: 0010:[<ffffffff80179eeb>] <ffffffff80179eeb>{kmem_cache_create+139}
RSP: 0000:ffff810001ca1eb8  EFLAGS: 00010292
RAX: 0000000000000034 RBX: 0000000000000000 RCX: 0000000000000000
RDX: 0000000000000000 RSI: 0000000000000dd3 RDI: ffffffff804167e0
RBP: 0000000000000005 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000010 R11: 0000000000000008 R12: 0000000000042000
R13: 0000000000000000 R14: 0000ffffffff8010 R15: 0000000000000000
FS:  0000000000000000(0000) GS:ffffffff8055a840(0000) knlGS:0000000000000000
CS:  0010 DS: 0018 ES: 0018 CR0: 000000008005003b
Quoted 9 lines
CR2: 0000000000004000 CR3: 0000000000101000 CR4: 00000000000006e0
Process swapper (pid: 1, threadinfo ffff810001ca0000, task ffff810001c5a7a0)
Stack: fffffffffffffff8 0000000000000000 0000000000000000 0000000000000000
       0000000000000010 0000000000000000 0000000000000005 0000000000000006
       00000000ffffffff 0000ffffffff8010
Call Trace:<ffffffff8057a11d>{init_bio+93} <ffffffff8010c0f2>{init+178}
       <ffffffff8010fc37>{child_rip+8} <ffffffff8010c040>{init+0}
       <ffffffff8010fc2f>{child_rip+0}
Quoted 8 lines

Something kooky is happening.

Clearly init_bio() is not passing in a NULL `name' parameter. Maybe the backtrace is screwed due to dopey gcc autoinlining and the bad caller is really biovec_init_slabs(). Try removing the __cacheline_aligned_mostly_readonly from the declaration of bvec_slabs[].

Please tell us what gcc and binutils versions you're using.

rafael@albercik:~> gcc -v
]--snip--[
gcc version 3.3.5 20050117 (prerelease) (SUSE Linux)

rafael@albercik:~> ld -v
GNU ld version 2.15.94.0.2.2 20041220 (SuSE Linux)

rafael@albercik:~> as -v
GNU assembler version 2.15.94.0.2.2 (x86_64-suse-linux) using BFD version 2.15.94.0.2.2 20041220 (SuSE Linux)

It's the default setup for SUSE 9.3 64-bit.

Strangely enough, I've compiled 2.6.12-rc4 and 2.6.12-rc3-mm2 with the same gcc/binutils and they work fine.

Greets,
Rafael

Signature · 3 lines
- Would you tell me, please, which way I ought to go from here?
- That depends a good deal on where you want to get to.
		-- Lewis Carroll "Alice's Adventures in Wonderland"
#8 lore raw

Re: [BUG][Resend] 2.6.12-rc3-mm3: Kernel BUG at "mm/slab.c":1219 [update]

In reply to Andrew Morton

Hi,

Quoted 4 lines · Andrew Morton
Quoted 3 lines · Rafael J. Wysocki

I get this from 2.6.12-rc3-mm3 on a UP AMD64 box (Asus L5D), 100% of the time:

Quoted 8 lines
Quoted 8 lines

]--snip--[
ACPI: bus type pci registered
PCI: Using configuration type 1
mtrr: v2.0 (20020519)
kmem_cache_create: Early error in slab <NULL> ----------- [cut here ] --------- [please bite here ] ---------

Kernel BUG at "mm/slab.c":1219

invalid operand: 0000 [1]

]--snip--[

Quoted 7 lines

Something kooky is happening.

Clearly init_bio() is not passing in a NULL `name' parameter. Maybe the backtrace is screwed due to dopey gcc autoinlining and the bad caller is really biovec_init_slabs(). Try removing the __cacheline_aligned_mostly_readonly from the declaration of bvec_slabs[].

Heh, it boots without the __cacheline_aligned_mostly_readonly (ie the BUG is only triggered if the __cacheline_aligned_mostly_readonly is present in the declaration of bvec_slabs[]). I've double-checked it. Interesting ... ;-)

Greets,
Rafael

Signature · 3 lines
- Would you tell me, please, which way I ought to go from here?
- That depends a good deal on where you want to get to.
		-- Lewis Carroll "Alice's Adventures in Wonderland"
#9 lore raw

Re: [BUG][Resend] 2.6.12-rc3-mm3: Kernel BUG at "mm/slab.c":1219 [update]

Quoted 9 lines · Rafael J. Wysocki
Quoted 5 lines

Clearly init_bio() is not passing in a NULL `name' parameter. Maybe the backtrace is screwed due to dopey gcc autoinlining and the bad caller is really biovec_init_slabs(). Try removing the __cacheline_aligned_mostly_readonly from the declaration of bvec_slabs[].

Heh, it boots without the __cacheline_aligned_mostly_readonly (ie the BUG is only triggered if the __cacheline_aligned_mostly_readonly is present in the declaration of bvec_slabs[]). I've double-checked it. Interesting ... ;-)

oops.

+#ifdef CONFIG_X86
+#define __cacheline_aligned_mostly_readonly \
+ __attribute__((__aligned__(SMP_CACHE_BYTES), \
+ __section__(".data.mostly_readonly")))
+#else
+#define __cacheline_aligned_mostly_readonly __cacheline_aligned +#endif

So on x86_64 we're putting __cacheline_aligned_mostly_readonly stuff into a section which is not implemented anywhere. I'm rather surprised that the kernel linked at all. But I'm more surprised that it mysteriously oopsed.

Oh well, I'll change that to CONFIG_X86 && !CONFIG_X86_64, thanks.

Or maybe Christoph wants to rustle up the x86_64 patch? We really should patch all architectures if we're going to do this thing.

#10 lore raw

Re: [BUG][Resend] 2.6.12-rc3-mm3: Kernel BUG at "mm/slab.c":1219 [update]

Quoted 21 lines · Andrew Morton
Quoted 9 lines · Rafael J. Wysocki
Quoted 5 lines

Clearly init_bio() is not passing in a NULL `name' parameter. Maybe the backtrace is screwed due to dopey gcc autoinlining and the bad caller is really biovec_init_slabs(). Try removing the __cacheline_aligned_mostly_readonly from the declaration of bvec_slabs[].

Heh, it boots without the __cacheline_aligned_mostly_readonly (ie the BUG is only triggered if the __cacheline_aligned_mostly_readonly is present in the declaration of bvec_slabs[]). I've double-checked it. Interesting ... ;-)

oops.

(quoted diff) +7 -0
+#ifdef CONFIG_X86
+#define __cacheline_aligned_mostly_readonly		\
+  __attribute__((__aligned__(SMP_CACHE_BYTES),		\
+		 __section__(".data.mostly_readonly")))
+#else
+#define __cacheline_aligned_mostly_readonly __cacheline_aligned
+#endif
Quoted 5 lines

So on x86_64 we're putting __cacheline_aligned_mostly_readonly stuff into a section which is not implemented anywhere. I'm rather surprised that the kernel linked at all. But I'm more surprised that it mysteriously oopsed.

Oh well, I'll change that to CONFIG_X86 && !CONFIG_X86_64, thanks.

Better just add the section to x86-64. Should be easy by copying the change from the i386 patch.

Ideally it would be asm-generic/vmlinux.lds and cover everybody...

-Andi

Quoted 2 lines

Or maybe Christoph wants to rustle up the x86_64 patch? We really should patch all architectures if we're going to do this thing.

#11 lore raw

Re: [BUG][Resend] 2.6.12-rc3-mm3: Kernel BUG at "mm/slab.c":1219 [update]

Quoted 2 lines · Andi Kleen

Better just add the section to x86-64. Should be easy by copying the change from the i386 patch.

Something like this?

arch/x86_64/kernel/vmlinux.lds.S +7 -0
Index: linux-2.6.11/arch/x86_64/kernel/vmlinux.lds.S
===================================================================
@@ -42,6 +42,13 @@ SECTIONS
 	CONSTRUCTORS
 	}
 
+  . = ALIGN(32);
+  .data.cacheline_aligned : { (.data.cacheline_aligned) }
+
+  /* Rarely changed data like cpu maps */
+  . = ALIGN(4096);
+  .data.mostly_readonly : { *(.data.mostly_readonly) }
+
   _edata = .;			/* End of data section */
 
   __bss_start = .;		/* BSS */
Quoted 1 line

Ideally it would be asm-generic/vmlinux.lds and cover everybody...

Hmm.. Add a definition for ALIGNED_DATA ?

#12 lore raw

Re: [BUG][Resend] 2.6.12-rc3-mm3: Kernel BUG at "mm/slab.c":1219 [update]

Quoted 12 lines · Christoph Lameter
Quoted 2 lines · Andi Kleen

Better just add the section to x86-64. Should be easy by copying the change from the i386 patch.

Something like this?

Index: linux-2.6.11/arch/x86_64/kernel/vmlinux.lds.S
===================================================================

arch/x86_64/kernel/vmlinux.lds.S SECTIONS · line 45

Quoted 4 lines
arch/x86_64/kernel/vmlinux.lds.S +1 -0
@@ -42,6 +42,13 @@ SECTIONS
 	CONSTRUCTORS
 	}
 
+  . = ALIGN(32);

This should be . = ALIGN(CONFIG_X86_L1_CACHE_BYTES)

It was wrong on i386 already btw, which needs the same.

arch/x86_64/kernel/vmlinux.lds.S SECTIONS · line 49

Quoted 4 lines
arch/x86_64/kernel/vmlinux.lds.S +4 -0
+  .data.cacheline_aligned : { (.data.cacheline_aligned) }
+
+  /* Rarely changed data like cpu maps */
+  . = ALIGN(4096);

Does it really need an 4096 byte alignment? That seems like a waste of memory. Cache line alignment should be enough.

Quoted 9 lines
arch/x86_64/kernel/vmlinux.lds.S +2 -0
+  .data.mostly_readonly : { *(.data.mostly_readonly) }
+
   _edata = .;			/* End of data section */
 
   __bss_start = .;		/* BSS */
Quoted 2 lines

Ideally it would be asm-generic/vmlinux.lds and cover everybody...

Hmm.. Add a definition for ALIGNED_DATA ?

Thinking about it again there is no portable way to do the cacheline alignment yet, so drop that suggestion please.

-Andi

Quoted 1 line
#13 lore raw

Re: [BUG][Resend] 2.6.12-rc3-mm3: Kernel BUG at "mm/slab.c":1219 [update]

Quoted 5 lines · Andi Kleen
Quoted 1 line

+ . = ALIGN(32);

This should be . = ALIGN(CONFIG_X86_L1_CACHE_BYTES)

It was wrong on i386 already btw, which needs the same.

i386 always specifies the numbers and does not use symbolic constants. We better leave that as it is.

Quoted 5 lines

Review

Quoted 2 lines
(quoted diff) +2 -0
+  /* Rarely changed data like cpu maps */
+  . = ALIGN(4096);

Does it really need an 4096 byte alignment? That seems like a waste of memory. Cache line alignment should be enough.

Ok.

arch/i386/kernel/vmlinux.lds.S +1 -1
Index: linux-2.6.11/arch/i386/kernel/vmlinux.lds.S
===================================================================
@@ -58,7 +58,7 @@ SECTIONS
   }
 
   /* rarely changed data like cpu maps */
-  . = ALIGN(4096);
+  . = ALIGN(32);
   .data.mostly_readonly : AT(ADDR(.data.mostly_readonly) - LOAD_OFFSET) {
 	*(.data.mostly_readonly)
   }
arch/x86_64/kernel/vmlinux.lds.S +7 -0
Index: linux-2.6.11/arch/x86_64/kernel/vmlinux.lds.S
===================================================================
@@ -42,6 +42,13 @@ SECTIONS
 	CONSTRUCTORS
 	}
 
+  . = ALIGN(CONFIG_X86_L1_CACHE_BYTES);
+  .data.cacheline_aligned : { (.data.cacheline_aligned) }
+
+  /* Rarely changed data like cpu maps */
+  . = ALIGN(CONFIG_X86_L1_CACHE_BYTES);
+  .data.mostly_readonly : { *(.data.mostly_readonly) }
+
   _edata = .;			/* End of data section */
 
   __bss_start = .;		/* BSS */
#14 lore raw

Re: [BUG][Resend] 2.6.12-rc3-mm3: Kernel BUG at "mm/slab.c":1219 [update]

Quoted 10 lines · Christoph Lameter
Quoted 5 lines · Andi Kleen
Quoted 1 line

+ . = ALIGN(32);

This should be . = ALIGN(CONFIG_X86_L1_CACHE_BYTES)

It was wrong on i386 already btw, which needs the same.

i386 always specifies the numbers and does not use symbolic constants. We better leave that as it is.

That's broken then and needs to be fixed.

Quoted 8 lines
Quoted 6 lines

Review

Quoted 3 lines
(quoted diff) +2 -0
+  /* Rarely changed data like cpu maps */
+  . = ALIGN(4096);

Does it really need an 4096 byte alignment? That seems like a waste of memory. Cache line alignment should be enough.

Ok.

Looks good.

-Andi

Quoted 32 lines

Index: linux-2.6.11/arch/i386/kernel/vmlinux.lds.S
===================================================================

arch/i386/kernel/vmlinux.lds.S +1 -1
@@ -58,7 +58,7 @@ SECTIONS
   }
 
   /* rarely changed data like cpu maps */
-  . = ALIGN(4096);
+  . = ALIGN(32);
   .data.mostly_readonly : AT(ADDR(.data.mostly_readonly) - LOAD_OFFSET) {
 	*(.data.mostly_readonly)
   }

Index: linux-2.6.11/arch/x86_64/kernel/vmlinux.lds.S
===================================================================

arch/x86_64/kernel/vmlinux.lds.S +7 -0
@@ -42,6 +42,13 @@ SECTIONS
 	CONSTRUCTORS
 	}
 
+  . = ALIGN(CONFIG_X86_L1_CACHE_BYTES);
+  .data.cacheline_aligned : { (.data.cacheline_aligned) }
+
+  /* Rarely changed data like cpu maps */
+  . = ALIGN(CONFIG_X86_L1_CACHE_BYTES);
+  .data.mostly_readonly : { *(.data.mostly_readonly) }
+
   _edata = .;			/* End of data section */
 
   __bss_start = .;		/* BSS */