On 15/12/2016 20:54, Chris Wilson wrote:
On Thu, Dec 15, 2016 at 11:44:13PM +0300, Dan Carpenter wrote:
Hello Tvrtko Ursulin,
The patch a19d6ff29a82: "drm/i915: Small compaction of the engine
init code" from Jun 23, 2016, leads to the following static checker
warning:
drivers/gpu/drm/i915/intel_lrc.c:1973 logical_render_ring_init()
warn: passing freed memory 'engine'
drivers/gpu/drm/i915/intel_lrc.c
1970
1971 ret = logical_ring_init(engine);
1972 if (ret) {
1973 lrc_destroy_wa_ctx_obj(engine);
The problem is that logical_ring_init() frees "engine" on the error
path so this is a use after free.
And calls lrc_destroy_wa_ctx_obj() in the process. So we can
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 067394b0a769..1c1bad8ae7b0 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -1940,12 +1940,7 @@ int logical_render_ring_init(struct intel_engine_cs *engine)
ret);
}
- ret = logical_ring_init(engine);
- if (ret) {
- lrc_destroy_wa_ctx_obj(engine);
- }
-
- return ret;
+ return logical_ring_init(engine);
}
int logical_xcs_ring_init(struct intel_engine_cs *engine)
I've marked this as TODO for later.
Interesting that it was detected only now. Even more so, the referenced
commit just moved the code around.
I think that the real issue was introduced around "drm/i915/gen8:
Re-order init pipe_control in lrc mode".
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
