Quoting Tvrtko Ursulin (2019-01-15 10:27:02) > > On 14/01/2019 21:17, Chris Wilson wrote: > > + if (err) > > + goto err_unlock; > > + > > + err = __i915_gem_userptr_get_pages_schedule(obj); > > + if (err == -EAGAIN) > > __i915_gem_userptr_set_active(obj, true); > > > > - if (IS_ERR(pages)) > > - release_pages(pvec, pinned); > > - kvfree(pvec); > > +err_unlock: > > + up_read(&mm->mmap_sem); > > May be safer to drop the lock earlier, immediately after probe. I don't > see holding it while queuing the worker and doing internal book-keeping > is useful and might just create more lock chain dependencies. Hmm, I thought we need to cover up to set-active (probe + queue + insert into rbtree) as I thought the mmu-invalidate was under the mmap_sem wlock. -Chris _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
