On 7/21/22 14:15, Jason Gunthorpe wrote:
> On Fri, Jul 15, 2022 at 01:24:15PM -0500, Bob Pearson wrote:
>> @@ -173,12 +172,11 @@ int rxe_mr_init_user(struct rxe_pd *pd, u64 start, u64 length, u64 iova,
>> void *vaddr;
>> int err;
>>
>> - umem = ib_umem_get(pd->ibpd.device, start, length, access);
>> + mr->umem = umem = ib_umem_get(&rxe->ib_dev, start, length, access);
>> if (IS_ERR(umem)) {
>
> So this puts an err_ptr into mr->umem and then later the cleanup:
>
> if (mr->umem)
> ib_umem_release(mr->umem);
>
> Will not like it very much.
>
> I'm OK with the idea of initializing structures that are cleaned up by
> their single master cleanup, but you have to do it in a way that
> doesn't put corrupted data into the object..
>
> Jason
I fixed this patch but it collides with the map set revert Li
sent in.
Bob
