Panic on boot with r351461 (AMD ThreadRipper 2990WX)

[Konstantin Belousov]

On Sat, Aug 24, 2019 at 04:54:26PM -0600, Rebecca Cran wrote:
> On 2019-08-24 14:33, Konstantin Belousov wrote:
> > On Sat, Aug 24, 2019 at 02:22:18PM -0600, Rebecca Cran wrote:
> >> instruction pointer = 0x20: 0xffffffff811bc664
> > So what is the source line for this address ?
> 
> 
> I built a new kernel and got a new panic instruction pointer address of
> 0xffffffff8117f67c, but running it through addr2line only gave a
> function name, not a line number:
> 
> addr2line -af -e /usr/lib/debug/boot/kernel/kernel.debug 0xffffffff8117f67c
Use gdb instead.

> 
> mp_realloc_pcpu
> /usr/src/sys/amd64/amd64/mp_machdep.c:0

What was the previous bootable version of the kernel ?

Do you happen to have NUMA node without any local memory ? (Look at the
SRAT table).  If yes, try this patch.

diff --git a/sys/amd64/amd64/mp_machdep.c b/sys/amd64/amd64/mp_machdep.c
index b38c688f8b4..84ce0b779ab 100644
--- a/sys/amd64/amd64/mp_machdep.c
+++ b/sys/amd64/amd64/mp_machdep.c
@@ -402,6 +402,8 @@ mp_realloc_pcpu(int cpuid, int domain)
 		return;
 	m = vm_page_alloc_domain(NULL, 0, domain,
 	    VM_ALLOC_NORMAL | VM_ALLOC_NOOBJ);
+	if (m == NULL)
+		return;
 	na = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
 	pagecopy((void *)oa, (void *)na);
 	pmap_qenter((vm_offset_t)&__pcpu[cpuid], &m, 1);