SYNC_BO does not constrain its size, so a request for zero bytes reaches
drm_clflush_virt_range(), which ends with an unconditional
clflushopt(end - 1). For an empty range that is the byte before the
mapping, and abo->mem.kva comes from vmap(), so the access lands in the
guard page below the vmalloc area and faults:
BUG: unable to handle page fault for address: ffffd16fbbc70fff
#PF: supervisor read access in kernel mode
Oops: Oops: 0000 [#1] SMP NOPTI
CPU: 7 UID: 1000 Comm: sync_bo_probe
RIP: 0010:drm_clflush_virt_range+0x3c/0x70
Call Trace:
amdxdna_drm_sync_bo_ioctl+0x124/0x430 [amdxdna]
drm_ioctl+0x301/0x4c0
__x64_sys_ioctl+0x115/0x2f0
do_syscall_64+0xa6/0x3d0
Any process that can open the render node can do this. Reproduced 3 of 3
times on a Strix Point NPU (1022:17f0), by calling SYNC_BO with size 0 on
an AMDXDNA_BO_SHARE object. The import arm takes the same request but
flushes the whole scatterlist, so it survives it.
Nothing needs flushing for an empty range, so answer before choosing a
path.
Fixes: e252e3f3488a ("accel/amdxdna: Revise device bo creation and free")
Cc: [email protected]
Signed-off-by: Taimuraz Kaitmazov <[email protected]>
---
Trees before amdxdna_flush_bo() existed carry the same call inline in
amdxdna_drm_sync_bo_ioctl(), with args->size passed to
drm_clflush_virt_range() unclamped, so a backport wants the guard at that
call site instead.
drivers/accel/amdxdna/amdxdna_gem.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/drivers/accel/amdxdna/amdxdna_gem.c
b/drivers/accel/amdxdna/amdxdna_gem.c
index 1c63eff0a4a8..2a16de96e6a4 100644
--- a/drivers/accel/amdxdna/amdxdna_gem.c
+++ b/drivers/accel/amdxdna/amdxdna_gem.c
@@ -1247,6 +1247,9 @@ static int amdxdna_flush_bo(struct amdxdna_gem_obj *abo,
u64 offset, u64 size)
return -EINVAL;
size = min(abo->mem.size, end) - offset;
+ if (!size)
+ return 0;
+
if (is_import_bo(abo))
drm_clflush_sg(abo->base.sgt);
else if (amdxdna_gem_vmap(abo))
--
2.55.0
