On Mon, Sep 30, 2019 at 08:57:32AM +0000, Koenig, Christian wrote:

Am 30.09.19 um 09:22 schrieb Daniel Vetter:

...

On Sun, Sep 22, 2019 at 2:08 PM Qiang Yu yuq825@gmail.com wrote:

...

This causes kernel crash when testing lima driver.

Cc: Christian König christian.koenig@amd.com Fixes: b8c036dfc66f ("dma-buf: simplify reservation_object_get_fences_rcu a bit") Signed-off-by: Qiang Yu yuq825@gmail.com

Selftest for this would be lovely, now that the basic infrastructure is in place ...

What do you have in mind? I wouldn't even know where to start to write an unit test for this.

1. set a few fences (both excl + shared) in a dma_resv 2. get them 3. check that we got them all 4. notice that the exlusive fence isn't actually in the array (because we increment the index before storing, so the exclusive fence ended past the array). For robustness the test should check that the fences are listed in any order, not just the one the current implementation gives you.

I guess the actual crash happens when we're unlucky and overflow the allocation, which is probably more rare. But KASAN should help catch that too (run that in your CI if you don't do that yet, it's pretty impressive).

Or am I totally misunderstanding what's going wrong here? -Daniel

Christian.

...

-Daniel

...

---

drivers/dma-buf/dma-resv.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/dma-buf/dma-resv.c b/drivers/dma-buf/dma-resv.c index 42a8f3f11681..709002515550 100644 --- a/drivers/dma-buf/dma-resv.c +++ b/drivers/dma-buf/dma-resv.c @@ -471,7 +471,7 @@ int dma_resv_get_fences_rcu(struct dma_resv *obj, if (pfence_excl) *pfence_excl = fence_excl; else if (fence_excl)

•           shared[++shared_count] = fence_excl;
  

•           shared[shared_count++] = fence_excl;
  
     if (!shared_count) {
             kfree(shared);
  

-- 2.17.1

---

dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel