| Message ID | 20171030145904.18584-1-deathsimple@vodafone.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi Christian, On 30 October 2017 at 20:29, Christian König <ckoenig.leichtzumerken@gmail.com> wrote: > From: Christian König <christian.koenig@amd.com> > > The amdgpu issue to also need signaled fences in the reservation objects > should be fixed by now. > > Optimize the list by keeping only the not signaled yet fences around. I think this warrants some more context in the commit log. Could you please do the same for the other patch as well? > > v2: temporary put the signaled fences at the end of the new container > > Signed-off-by: Christian König <christian.koenig@amd.com> > --- > drivers/dma-buf/reservation.c | 36 ++++++++++++++++++++++++++---------- > 1 file changed, 26 insertions(+), 10 deletions(-) > Best, Sumit.
Quoting Christian König (2017-10-30 14:59:03) > From: Christian König <christian.koenig@amd.com> > > The amdgpu issue to also need signaled fences in the reservation objects > should be fixed by now. > > Optimize the list by keeping only the not signaled yet fences around. > > v2: temporary put the signaled fences at the end of the new container > > Signed-off-by: Christian König <christian.koenig@amd.com> > --- > drivers/dma-buf/reservation.c | 36 ++++++++++++++++++++++++++---------- > 1 file changed, 26 insertions(+), 10 deletions(-) > > diff --git a/drivers/dma-buf/reservation.c b/drivers/dma-buf/reservation.c > index b44d9d7db347..6fc794576997 100644 > --- a/drivers/dma-buf/reservation.c > +++ b/drivers/dma-buf/reservation.c > @@ -145,8 +145,8 @@ reservation_object_add_shared_replace(struct reservation_object *obj, > struct reservation_object_list *fobj, > struct dma_fence *fence) > { > - unsigned i; > struct dma_fence *old_fence = NULL; > + unsigned i, j, k; > > dma_fence_get(fence); > > @@ -162,9 +162,7 @@ reservation_object_add_shared_replace(struct reservation_object *obj, > * references from the old struct are carried over to > * the new. > */ > - fobj->shared_count = old->shared_count; > - > - for (i = 0; i < old->shared_count; ++i) { > + for (i = 0, j = 0, k = fobj->shared_max; i < old->shared_count; ++i) { > struct dma_fence *check; > > check = rcu_dereference_protected(old->shared[i], > @@ -172,10 +170,18 @@ reservation_object_add_shared_replace(struct reservation_object *obj, > > if (!old_fence && check->context == fence->context) { > old_fence = check; > - RCU_INIT_POINTER(fobj->shared[i], fence); > - } else > - RCU_INIT_POINTER(fobj->shared[i], check); > + RCU_INIT_POINTER(fobj->shared[j++], fence); > + } else if (!dma_fence_is_signaled(check)) { > + RCU_INIT_POINTER(fobj->shared[j++], check); > + } else { > + /* > + * Temporary save the signaled fences at the end of the > + * new container > + */ > + RCU_INIT_POINTER(fobj->shared[--k], check); > + } > } > + fobj->shared_count = j; > if (!old_fence) { > RCU_INIT_POINTER(fobj->shared[fobj->shared_count], fence); > fobj->shared_count++; > @@ -192,10 +198,20 @@ reservation_object_add_shared_replace(struct reservation_object *obj, > write_seqcount_end(&obj->seq); > preempt_enable(); > > - if (old) > - kfree_rcu(old, rcu); > - > dma_fence_put(old_fence); We don't need to keep old_fence (local var) around any more in this scheme. By construction, the new shared_list is large enough to hold all the old_fences+1, so you can use a loop like j = 0; k = fobj->shared_max; RCU_INIT_POINTER(fobj->shared[j++], fence); for (i = 0; i < old->shared_count; ++i) { struct dma_fence *check; check = rcu_dereference_protected(old->shared[i], reservation_object_held(obj)); if (check->context == fence->context || dma_fence_is_signaled(check)) RCU_INIT_POINTER(fobj->shared[--k], check); else RCU_INIT_POINTER(fobj->shared[j++], check); } fobj->shared_count = j; > + > + if (!old) > + return; > + > + /* Drop the references to the signaled fences */ > + for (i = k; i < fobj->shared_max; ++i) { > + struct dma_fence *f; > + > + f = rcu_dereference_protected(fobj->shared[i], > + reservation_object_held(obj)); > + dma_fence_put(f); > + } > + kfree_rcu(old, rcu); > } Fwiw, this fixes a big problem with the lack of pruning of the reservation_object, so \o/ -Chris
diff --git a/drivers/dma-buf/reservation.c b/drivers/dma-buf/reservation.c index b44d9d7db347..6fc794576997 100644 --- a/drivers/dma-buf/reservation.c +++ b/drivers/dma-buf/reservation.c @@ -145,8 +145,8 @@ reservation_object_add_shared_replace(struct reservation_object *obj, struct reservation_object_list *fobj, struct dma_fence *fence) { - unsigned i; struct dma_fence *old_fence = NULL; + unsigned i, j, k; dma_fence_get(fence); @@ -162,9 +162,7 @@ reservation_object_add_shared_replace(struct reservation_object *obj, * references from the old struct are carried over to * the new. */ - fobj->shared_count = old->shared_count; - - for (i = 0; i < old->shared_count; ++i) { + for (i = 0, j = 0, k = fobj->shared_max; i < old->shared_count; ++i) { struct dma_fence *check; check = rcu_dereference_protected(old->shared[i], @@ -172,10 +170,18 @@ reservation_object_add_shared_replace(struct reservation_object *obj, if (!old_fence && check->context == fence->context) { old_fence = check; - RCU_INIT_POINTER(fobj->shared[i], fence); - } else - RCU_INIT_POINTER(fobj->shared[i], check); + RCU_INIT_POINTER(fobj->shared[j++], fence); + } else if (!dma_fence_is_signaled(check)) { + RCU_INIT_POINTER(fobj->shared[j++], check); + } else { + /* + * Temporary save the signaled fences at the end of the + * new container + */ + RCU_INIT_POINTER(fobj->shared[--k], check); + } } + fobj->shared_count = j; if (!old_fence) { RCU_INIT_POINTER(fobj->shared[fobj->shared_count], fence); fobj->shared_count++; @@ -192,10 +198,20 @@ reservation_object_add_shared_replace(struct reservation_object *obj, write_seqcount_end(&obj->seq); preempt_enable(); - if (old) - kfree_rcu(old, rcu); - dma_fence_put(old_fence); + + if (!old) + return; + + /* Drop the references to the signaled fences */ + for (i = k; i < fobj->shared_max; ++i) { + struct dma_fence *f; + + f = rcu_dereference_protected(fobj->shared[i], + reservation_object_held(obj)); + dma_fence_put(f); + } + kfree_rcu(old, rcu); } /**