re: kern/57737: netbsd-10 panics on current Epyc CPU

[matthew green <mrg%eterna23.net@localhost>]

> 	netbsd-10 panics early on current multi-core Ryzen cpus.
>
> 	See the boot log for an Epyc 9554P cpu on a Gigabyte R263-Z70
> 	board at
>
> 	<ftp://oak.causeuse.org/pub/NetBSD/netbsd-10-GA_R263-Z70_epyc9554p.bootlog.gz>
>
> 	and the related discussion on current-users, where Martin
> 	suggested
>
> 	"That sounds like an fpu xsave size issue Taylor looked at
> 	recently (but it is not fixed)."

there are multiple issues with this system, ouch.

no CPUs attach in this dmesg.  cpu0 remains half-attached.  this
is some problem with the MADT parser i guess (i don't know this
very well.)

[   1.0000040] bogus MADT X2APIC entry (id = 0x0)
[   1.0000040] bogus MADT X2APIC entry (id = 0x2)
...
[   1.0000040] bogus MADT X2APIC entry (id = 0x7e)
...
[   1.0000040] bogus MADT X2APIC entry (id = 0x5e)
[   1.0000040] bogus MADT X2APIC entry (id = 0x1)
...
[   1.0000040] bogus MADT X2APIC entry (id = 0x7f)
...
[   1.0000040] bogus MADT X2APIC entry (id = 0x5f)
 
ie, 128 cpu threads fail to attach (which matches the specs for
epyc 9554p - 64c/128t.)  some devices still attach things to
cpu0 for affinity, even though it's in UP mode:

[   1.0525126] nvme0: for io queue 1 interrupting at msix0 vec 1 affinity to cpu0
... plus nvme1/2/3.

some of the dmesg items seem to have 'nul' chars in them:

[   1.0000040] ACPI: XSDT 0x00000000A4E13728 000^@0DC (v01 GBT   BTUACPI 03042021 AMI  01000013)

[   1.0525126] AMD 19h/1xh RCEC (Root Complex^@ Event Collectosystem) at pci0 dev 0 function 3 not configured

and then the final crash as reported in this PR:

[   1.0525126] fatal privileged instruction fault in supervisor mode
[   1.0525126] trap type 0 code 0 rip 0xffffffff8023c24e cs 0x8 rflags 0x10256 cr2 0 ilevel 0x6 rsp 0xffffffff81d3bab8
[   1.0525126] curlwp 0xffffffff8188ac00 pid 0.0 lowest kstack 0xffffffff81d362c0
kernel: privileged instruction fault trap, code=0
Stopped in pid 0.0 (system) at  netbsd:xrstor+0xa:      fxsavel
xrstor() at netbsd:xrstor+0xa
aes_selftest() at netbsd:aes_selftest+0x26
aes_modcmd() at netbsd:aes_modcmd+0xe9
module_do_builtin() at netbsd:module_do_builtin+0x142
module_do_builtin() at netbsd:module_do_builtin+0xfa
module_init_class() at netbsd:module_init_class+0x142


.mrg.