Eric, or anyone with a powerpc64 with a lot of ram, does this diff
prevent panics in uvm_fault or crashes on "trap type 300"?
I did reproduce a "trap type 300" with 8g ram once on 2024-11-05, but
failed to reproduce it again. I have now found some code in
powerpc64's pte_insert that might put a page table entry in the wrong
bucket, but only if the primary and secondary buckets are both full.
With only 8g ram, my powerpc64 rarely fills these buckets before it
starts swapping. I'm trying to fix a problem that almost never
happens to me.
--gkoehler
Index: arch/powerpc64/powerpc64/pmap.c
===================================================================
RCS file: /cvs/src/sys/arch/powerpc64/powerpc64/pmap.c,v
diff -u -p -r1.62 pmap.c
--- arch/powerpc64/powerpc64/pmap.c 4 Jun 2024 17:31:59 -0000 1.62
+++ arch/powerpc64/powerpc64/pmap.c 17 Nov 2024 03:20:57 -0000
@@ -816,8 +816,8 @@ pte_insert(struct pte_desc *pted)
pted->pted_va &= ~(PTED_VA_HID_M|PTED_VA_PTEGIDX_M);
pted->pted_va |= off & (PTED_VA_PTEGIDX_M|PTED_VA_HID_M);
- idx ^= (PTED_HID(pted) ? pmap_ptab_mask : 0);
- pte = pmap_ptable + (idx * 8);
+ pte = pmap_ptable;
+ pte += (idx ^ (PTED_HID(pted) ? pmap_ptab_mask : 0)) * 8;
pte += PTED_PTEGIDX(pted); /* increment by index into pteg */
if ((pte->pte_hi & PTE_WIRED) == 0)
