Re: Per-cpu IDT patch

[Jaromír Doleček <jaromir.dolecek%gmail.com@localhost>]

I've just committed a fix for sys/arch/xen/x86/cpu.c to avoid
triggering the KASSERT(). It was just missing initialization for the
Xen variant, overlooked when the fpu_kern_enter() API was added in
November 2019.

Jaromir

Le mer. 8 juil. 2020 à 12:31, Frank Kardel <kardel%netbsd.org@localhost> a écrit :
>
> The patched GENERIC kernel booted fine directly.
>
> A patches XEN3_DOM0 kernel with Xen 4.11 hypervisor paniced. That's why
>
> I am suspecting Xen/AES changes.
>
> Frank
>
>
>
> On 07/08/20 12:10, s ymgch wrote:
> > On Mon, Jul 6, 2020 at 10:45 PM Frank Kardel <kardel%netbsd.org@localhost> wrote:
> >> Nice progress here:
> >>
> >> On bare metal 9.99.69 GENERIC boots sees all ixl, wm and nvme devices
> >> with all their interrupts initialized - great.
> >>
> >> BUT... 9.99.69-XEN3_DOM0 digs out following panic:
> >>
> >> [   2.0100519] panic: kernel diagnostic assertion "ci->ci_kfpu_spl ==
> >> -1" failed: file "/src/NetBSD/cur/src/sys/arch/x86/x86/fpu.c", line 389
> >> [   2.0100519] cpu3: Begin traceback...
> >> [   2.0100519] vpanic() at netbsd:vpanic+0x146
> >> [   2.0100519] kern_assert() at netbsd:kern_assert+0x48
> >> [   2.0100519] fpu_kern_enter() at netbsd:fpu_kern_enter+0x190
> >> [   2.0100519] cpu3: Begin traceback...
> >> [   2.0100519] vpanic() at netbsd:vpanic+0x146
> >> [   2.0100519] kern_assert() at netbsd:kern_assert+0x48
> >> [   2.0100519] fpu_kern_enter() at netbsd:fpu_kern_enter+0x190
> >> [   2.0100519] aesni_probe() at netbsd:aesni_probe+0x28
> >> [   2.0100519] aes_selftest() at netbsd:aes_selftest+0x13
> >> [   2.0100519] aes_modcmd() at netbsd:aes_modcmd+0xff
> >> [   2.0100519] module_do_builtin() at netbsd:module_do_builtin+0x173
> >> [   2.0100519] module_do_builtin() at netbsd:module_do_builtin+0x12e
> >> [   2.0100519] module_init_class() at netbsd:module_init_class+0x145
> >> [   2.0100519] main() at netbsd:main+0x472
> >> [   2.0100519] cpu3: End traceback...
> >> [   2.0100519] fatal breakpoint trap in supervisor mode
> >> [   2.0100519] trap type 1 code 0 rip 0xffffffff8023c93d cs 0xe030
> >> rflags 0x202 cr2 0 ilevel 0x8 rsp 0xffffffff81981af0
> >> [   2.0100519] curlwp 0xffffffff80e5c6c0 pid 0.0 lowest kstack
> >> 0xffffffff8197d2c0
> >> Stopped in pid 0.0 (system) at  netbsd:breakpoint+0x5:  leave
> > Did it only appear when you applied my patch?
> > It seems that the assertion happens not in IDT but fpu.
> >
> > -- yamaguchi
>