Re: Linux 5.4.252 FPU initialization warnings in stable kernels 5.4/5.10

From: Salvatore Bonaccorso
Date: Wed Aug 16 2023 - 01:53:24 EST

Hi,

On Tue, Aug 15, 2023 at 11:16:52PM +0200, Greg KH wrote:
> On Tue, Aug 15, 2023 at 08:15:39PM +0000, Shaoying Xu wrote:
> > Hi Thomas/Greg
> >
> > We are seeing “get of unsupported state” warnings during FPU initialization in the v5.4.252 and v5.10.189
> > kernel booted on AWS EC2 instances with Intel processors based on Nitro system. These warnings are observed
> > in EC2 c5.18xlarge instance:
> >
> > [ 1.204495] ------------[ cut here ]------------
> > [ 1.204495] get of unsupported state
> > [ 1.204495] WARNING: CPU: 0 PID: 0 at arch/x86/kernel/fpu/xstate.c:879 get_xsave_addr+0x81/0x90
> > [ 1.204495] Modules linked in:
> > [ 1.204495] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.4.252 #10
> > [ 1.204495] Hardware name: Amazon EC2 c5.18xlarge/, BIOS 1.0 10/16/2017
> > [ 1.204495] RIP: 0010:get_xsave_addr+0x81/0x90
> > [ 1.204495] Code: 5b c3 48 83 c4 08 31 c0 5b c3 80 3d 7c f0 78 01 00 75 c1 48 c7 c7 34 be 03 b2 89 4c 24 04 c6 05 68 f0 78 01 01 e8 ef 41 05 00 <0f> 0b 48 63 4c 24 04 eb a1 31 c0 c3 0f 1f 00 0f 1f 44 00 00 41 54
> > [ 1.204495] RSP: 0000:ffffffffb2603ed0 EFLAGS: 00010282
> > [ 1.204495] RAX: 0000000000000000 RBX: ffffffffb27ebe80 RCX: 0000000047cb2486
> > [ 1.204495] RDX: 0000000000000018 RSI: ffffffffb39e99a0 RDI: ffffffffb39e756c
> > [ 1.204495] RBP: ffffffffb27ebd40 R08: 7520666f20746567 R09: 74726f707075736e
> > [ 1.204495] R10: 00000000000962fc R11: 6574617473206465 R12: ffffffffb2d89b60
> > [ 1.204495] R13: 0000000000000246 R14: 0000000000000000 R15: 0000000000000000
> > [ 1.204495] FS: 0000000000000000(0000) GS:ffff96d031400000(0000) knlGS:0000000000000000
> > [ 1.204495] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > [ 1.204495] CR2: ffff96e277fff000 CR3: 000000103060a001 CR4: 00000000007200b0
> > [ 1.204495] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> > [ 1.204495] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
> > [ 1.204495] Call Trace:
> > [ 1.204495] ? __warn+0x85/0xd0
> > [ 1.204495] ? get_xsave_addr+0x81/0x90
> > [ 1.204495] ? report_bug+0xb6/0x130
> > [ 1.204495] ? get_xsave_addr+0x81/0x90
> > [ 1.204495] ? fixup_bug.part.12+0x18/0x30
> > [ 1.204495] ? do_error_trap+0x95/0xb0
> > [ 1.204495] ? do_invalid_op+0x36/0x40
> > [ 1.204495] ? get_xsave_addr+0x81/0x90
> > [ 1.204495] ? invalid_op+0x1e/0x30
> > [ 1.204495] ? get_xsave_addr+0x81/0x90
> > [ 1.204495] identify_cpu+0x422/0x510
> > [ 1.204495] identify_boot_cpu+0xc/0x94
> > [ 1.204495] arch_cpu_finalize_init+0x5/0x47
> > [ 1.204495] start_kernel+0x468/0x511
> > [ 1.204495] secondary_startup_64+0xa4/0xb0
> > [ 1.204495] ---[ end trace dffac81ff531fcf2 ]---
> >
> > The issue can be easily reproduced on both virtualized and bare metal instances but interesting thing is
> > it can’t be found in other latest stable kernels v4.14, v4.19, v5.15 and newer. We tried to bisect between v5.4.251 and v5.4.252 and
> > were able to find below commit to be the culprit. Also, reverting it in v5.4.252 and v5.10.189 resolved above warnings completely.
> >
> > x86/fpu: Move FPU initialization into arch_cpu_finalize_init()
> > commit b81fac906a8f9e682e513ddd95697ec7a20878d4 upstream
> >
> > We used to speculate the fix might be similar to commit 3f8968f1f0ad (“x86/xen: Fix secondary processors' FPU initialization”) but
> > since only kernel 5.4/5.10 are impacted, we’re not quite sure how this commit affects them in practice. Could you please take a look and share your insights?
> >
> > Also put stack traces from v5.10.189:
> >
> > [ 1.210910] ------------[ cut here ]------------
> > [ 1.210910] get of unsupported state
> > [ 1.210910] WARNING: CPU: 0 PID: 0 at arch/x86/kernel/fpu/xstate.c:974 get_xsave_addr+0x89/0xa0
> > [ 1.210910] Modules linked in:
> > [ 1.210910] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.10.189 #4
> > [ 1.210910] Hardware name: Amazon EC2 c5.18xlarge/, BIOS 1.0 10/16/2017
> > [ 1.210910] RIP: 0010:get_xsave_addr+0x89/0xa0
> > [ 1.210910] Code: c4 08 31 c0 5b e9 17 a4 bc 00 80 3d e7 75 eb 01 00 75 b9 48 c7 c7 b7 f4 09 ab 89 4c 24 04 c6 05 d3 75 eb 01 01 e8 17 98 05 00 <0f> 0b 48 63 4c 24 04 eb 99 31 c0 e9 e7 a3 bc 00 0f 1f 80 00 00 00
> > [ 1.210910] RSP: 0000:ffffffffab603ec8 EFLAGS: 00010286
> > [ 1.210910] RAX: 0000000000000000 RBX: ffffffffabf25bc0 RCX: 00000000fffeffff
> > [ 1.210910] RDX: ffffffffab603cd0 RSI: 00000000fffeffff RDI: ffffffffad1a3dec
> > [ 1.210910] RBP: ffffffffabf25a60 R08: 0000000000000000 R09: 0000000000000001
> > [ 1.210910] R10: 0000000000000000 R11: ffffffffab603cc8 R12: ffffffffac539b40
> > [ 1.210910] R13: 0000000000000246 R14: 0000000000000000 R15: 0000000000000000
> > [ 1.210910] FS: 0000000000000000(0000) GS:ffff9150f1600000(0000) knlGS:0000000000000000
> > [ 1.210910] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > [ 1.210910] CR2: ffff915702801000 CR3: 0000001780610001 CR4: 00000000007300b0
> > [ 1.210910] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> > [ 1.210910] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
> > [ 1.210910] Call Trace:
> > [ 1.210910] ? __warn+0x7d/0xe0
> > [ 1.210910] ? get_xsave_addr+0x89/0xa0
> > [ 1.210910] ? report_bug+0xbb/0x140
> > [ 1.210910] ? handle_bug+0x3f/0x70
> > [ 1.210910] ? exc_invalid_op+0x13/0x60
> > [ 1.210910] ? asm_exc_invalid_op+0x12/0x20
> > [ 1.210910] ? get_xsave_addr+0x89/0xa0
> > [ 1.210910] ? get_xsave_addr+0x89/0xa0
> > [ 1.210910] identify_cpu+0x42a/0x550
> > [ 1.210910] identify_boot_cpu+0xc/0x94
> > [ 1.210910] arch_cpu_finalize_init+0x5/0x47
> > [ 1.210910] start_kernel+0x4bc/0x56b
> > [ 1.210910] secondary_startup_64_no_verify+0xb0/0xbb
> > [ 1.210910] ---[ end trace 14850c6f8ee0875d ]---
>
> I think this is fixed with commit b3607269ff57 ("x86/pkeys: Revert
> a5eff7259790 ("x86/pkeys: Add PKRU value to init_fpstate")"), which is
> queued up for the next 5.4 and 5.10.y releases to happen "soon". Can
> you test the released -rc1 versions of this to verify it is resolved or
> not?

While not the reporter, I can confirm that the commit resolved the
issue.

The issue was reported some days back in Debian as
https://bugs.debian.org/1044518

I verified that with 5.10.190-rc1 the above warning is gone:
https://bugs.debian.org/1044518#34

(as it was fixed already with 5.10.190-rc1 pending, I did not
botherred to as well report it upstream after it was clear it is not a
Debian specific issue anymore).

Regards,
Salvatore