On Thu, Nov 24, 2016 at 09:11:41AM -0800, Andy Lutomirski wrote: > I gett this when booting a 32-bit 4.9-rc6-ish on Skylake: > > [ 0.564506] ------------[ cut here ]------------ > [ 0.564994] WARNING: CPU: 0 PID: 1 at > ./arch/x86/include/asm/fpu/internal.h:368 fpu__restore+0x203/0x210 > [ 0.565737] Modules linked in: > [ 0.566040] CPU: 0 PID: 1 Comm: sh Not tainted 4.9.0-rc6+ #488 > [ 0.566502] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), > BIOS 1.9.3-1.fc25 04/01/2014 > [ 0.567174] c78a9e5c c135a6d0 00000000 c1ac4b4c c78a9e8c c10aeb42 > c1ad53f0 00000000 > [ 0.567896] 00000001 c1ac4b4c 00000170 c107e753 00000170 c78a06c0 > 00000000 c78a0700 > [ 0.568583] c78a9ea0 c10aec05 00000009 00000000 00000000 c78a9eb8 > c107e753 c78a0700 > [ 0.569245] Call Trace: > [ 0.569440] [<c135a6d0>] dump_stack+0x58/0x78 > [ 0.569783] [<c10aeb42>] __warn+0xe2/0x100 > [ 0.570109] [<c107e753>] ? fpu__restore+0x203/0x210 > [ 0.570519] [<c10aec05>] warn_slowpath_null+0x25/0x30 > [ 0.570943] [<c107e753>] fpu__restore+0x203/0x210 > [ 0.571312] [<c107ff5c>] __fpu__restore_sig+0x1fc/0x580 > [ 0.571719] [<c108050a>] fpu__restore_sig+0x2a/0x50 > [ 0.572103] [<c107413d>] restore_sigcontext.isra.10+0xbd/0xd0 > [ 0.572546] [<c1074a11>] sys_sigreturn+0x81/0x90 > [ 0.572908] [<c1001837>] do_int80_syscall_32+0x57/0xc0 > [ 0.573306] [<c190eb06>] entry_INT80_32+0x2a/0x2a > [ 0.573677] ---[ end trace 88038c46b2a9d23a ]--- > > Telling KVM to disable XSAVES makes the warning go away. > > I seem to be the only person testing 32-bit kernels on CPUs this new :-/
Well skylake added XRSTORS, which is used in place of XRSTOR if supported by the CPU, but XRSTORS requires CPL=0, which XRSTOR did not as far as I can tell. Older CPUs don't have XRSTORS so this would not be an issue there. I don't know, but would not be surprised if running under kvm means the guest kernel is not running with CPL=0 and hence the XRSTORS feature ought not to be exposed as supported by the CPU to the guest kernel. Just a guess. Does this happen with a 64 bit kvm guest too? Does it happen if the 32 bit kernel is booted on bare hardware? My guess if I am thinking the right thing is that the answers are yes and no respectively. Looks like this was hit in jvm a couple of years ago: https://www.spinics.net/lists/kvm/msg110434.html No idea what the resolution was if any. -- Len Sorensen