General protection fault in raid5, load_balance

Bug 45351 NeilBrown 2 participants 2 messages Nov 29, 2012 2 messages not in the archive MD RAID

lore

Share & tools

lore

A message that is not in the archive (sent off-list or not yet received): [email protected]/

A message that is not in the archive (sent off-list or not yet received): [email protected]

Jim Kukunas

2012-11-29 21:54 UTC #3 lore raw

Quoted 15 lines · Neil Brown

Quoted 12 lines · On Wed, 28 Nov 2012 10:33:06 +0000 (UTC) [email protected]…
https://bugzilla.kernel.org/show_bug.cgi?id=45351
Cyril B. <[email protected]> changed:
What |Removed |Added
----------------------------------------------------------------------------
     Kernel Version|3.5.0                       |3.5.0, 3.6.8

Quoted 6 lines

Quoted 3 lines
--- Comment #1 from Cyril B. <[email protected]> 2012-11-28 10:33:05 --- I've just tested 3.6.8, I still get the same bug/trace.
Hi Jim,
could you look at this bug please?

Hi Neil,

Thank you for bringing this to my attention.

Quoted 26 lines

https://bugzilla.kernel.org/show_bug.cgi?id=45351

It seems to be crashing in xor_avx_4:

[48595.135046] general protection fault: 0000 [#1] SMP
[48595.135093] CPU 0
[48595.135098] Modules linked in: nf_conntrack_ipv6 nf_defrag_ipv6
nf_conntrack_ipv4 nf_defrag_ipv4 xt_state nf_conntrack xt_multiport coretemp
hwmon i2c_i801 shpchp pci_hotplug ehci_hcd usbcore usb_common netconsole e1000e
[last unloaded: scsi_wait_scan]
[48595.135211]
[48595.135224] Pid: 2429, comm: md4_raid5 Not tainted 3.5.0 #2                 
/DH67BL
[48595.135263] RIP: 0010:[<ffffffff813512d8>]  [<ffffffff813512d8>] xor_avx_4+0x48/0x350
[48595.135303] RSP: 0018:ffff880213a259d0  EFLAGS: 00010282
[48595.135323] RAX: 000000008005003b RBX: 0000000000000008 RCX: ffff8802130b5000
[48595.135346] RDX: ffff880212c9f000 RSI: ffff880212c9e000 RDI: 0000000000001000
[48595.135368] RBP: ffff880213a25ac0 R08: ffff8802130b4000 R09: ffff880212c9e000
[48595.135391] R10: ffff880212c9e000 R11: 0000000000000000 R12: 000000008005003b
[48595.135413] R13: 0000000000000003 R14: ffff880213a25cd0 R15: 0000000000001000
[48595.135436] FS:  0000000000000000(0000) GS:ffff88021fa00000(0000) knlGS:0000000000000000
[48595.135471] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[48595.135492] CR2: 000000000235f570 CR3: 0000000001c0b000 CR4: 00000000000407f0
[48595.135514] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[48595.135537] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400

Quoted 3 lines

....

[48595.136063] Code: b5 30 ff ff ff 48 89 95 28 ff ff ff 48 89 8d 20 ff ff ff 4c 89 85 18 ff ff ff e8 c4 04 ce ff 66 90 49 89 c4 0f 06 66 66 90 66 90 <c5> fc 29 85 50 ff ff ff c5 fc 29 8d 70 ff ff ff c5 fc 29 55 90

The code dump above is quiet revealing. The relevant instructions are:

	clts
	vmovaps	%ymm0,	-0xb0(%rbp)
	vmovaps %ymm1, 	-0x90(%rbp)
	vmovaps %ymm2,	-0x70(%rbp)

These instructions save the floating point state before we begin the actual xor work. Looking at the register dump, -0xb0(%rbp) is not properly aligned to 32 bytes, hence the #GP.

The question is whether the #GP still occurs after 841e3604d35aa70d399146abdc526d8c89a2c2f5.

Before that commit, we manually saved and restored the floating point state to the stack with the YMMS_{SAVE,RESTORE} macros. After that commit, we use the kernel_fpu_{begin,end} routines. In the former case, it would seem GCC is ignoring our request to align the stack variable to 32-bytes and 841e3604d35aa70d399146abdc526d8c89a2c2f5 should resolve the problem. In the later case, we will need to investigate further.

Thanks.

Signature · 2 lines

Jim Kukunas
Intel Open Source Technology Center