On 15 Jul 2025, at 14:56, Mark Johnston wrote: > On Tue, Jul 15, 2025 at 10:52:35AM +0200, Kristof Provost wrote: >> On 15 Jul 2025, at 0:03, Konstantin Belousov wrote: >>> The branch main has been updated by kib: >>> >>> URL: >>> https://cgit.FreeBSD.org/src/commit/?id=4dd828c80828637452a8a4e07a64e294c82e5d8b >>> >>> commit 4dd828c80828637452a8a4e07a64e294c82e5d8b >>> Author: Konstantin Belousov <k...@freebsd.org> >>> AuthorDate: 2025-06-15 21:39:10 +0000 >>> Commit: Konstantin Belousov <k...@freebsd.org> >>> CommitDate: 2025-07-14 22:03:28 +0000 >>> >>> amd64: add kva_layout >>> >>> Reviewed by: markj >>> Tested by: pho >>> Sponsored by: The FreeBSD Foundation >>> Differential revision: https://reviews.freebsd.org/D51085 >>> --- >> .. >>> diff --git a/sys/amd64/include/vmparam.h b/sys/amd64/include/vmparam.h >>> index 0cd9bb4fa7a4..f12e5217a44b 100644 >>> --- a/sys/amd64/include/vmparam.h >>> +++ b/sys/amd64/include/vmparam.h >>> @@ -181,12 +181,9 @@ >>> * 0xffffffff80000000 KERNBASE >>> */ >>> >>> -#define VM_MIN_KERNEL_ADDRESS KV4ADDR(KPML4BASE, 0, 0, 0) >>> -#define VM_MAX_KERNEL_ADDRESS KV4ADDR(KPML4BASE + NKPML4E - 1, \ >>> - NPDPEPG-1, NPDEPG-1, NPTEPG-1) >>> - >>> -#define DMAP_MIN_ADDRESS KV4ADDR(DMPML4I, 0, 0, 0) >>> -#define DMAP_MAX_ADDRESS KV4ADDR(DMPML4I + NDMPML4E, 0, 0, 0) >>> +#define VM_MIN_KERNEL_ADDRESS_LA48 KV4ADDR(KPML4BASE, 0, 0, 0) >>> +#define VM_MIN_KERNEL_ADDRESS kva_layout.km_low >>> +#define VM_MAX_KERNEL_ADDRESS kva_layout.km_high >> >> The removal of DMAP_MIN_ADDRESS breaks the KASAN build. >> >> This seems to fix it for me, but I’m not familiar enough with this code to >> be confident to commit it: >> >> diff --git a/sys/kern/subr_asan.c b/sys/kern/subr_asan.c >> index 0edb631d1475..31204bad8c30 100644 >> --- a/sys/kern/subr_asan.c >> +++ b/sys/kern/subr_asan.c >> @@ -263,8 +263,8 @@ kasan_mark(const void *addr, size_t size, size_t >> redzsize, uint8_t code) >> if (__predict_false(!kasan_enabled)) >> return; >> >> - if ((vm_offset_t)addr >= DMAP_MIN_ADDRESS && >> - (vm_offset_t)addr < DMAP_MAX_ADDRESS) >> + if ((vm_offset_t)addr >= kva_layout.dmap_low && >> + (vm_offset_t)addr < kva_layout.dmap_high) >> return; > > That restores the old behaviour, but it would break on arm64, which > doesn't have a kva_layout. > > I pushed a different patch as commit 1f683624df878b2e4c0187137640bf2e6662c36f. > That way is also consistent with how unhandled addresses are handled in > KMSAN.
That seems to work too, thanks! — Kristof
