Message ID | 1500654001-20899-1-git-send-email-deathsimple@vodafone.de (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On Fri, Jul 21, 2017 at 06:20:01PM +0200, Christian König wrote: > From: Christian König <christian.koenig@amd.com> > > With hardware resets in mind it is possible that all shared fences are > signaled, but the exlusive isn't. Fix waiting for everything in this situation. How did you end up with both shared and exclusive fences on the same reservation object? At least I thought the point of exclusive was that it's exclusive (and has an implicit barrier on all previous shared fences). Same for shared fences, they need to wait for the exclusive one (and replace it). Is this fallout from the amdgpu trickery where by default you do all shared fences? I thought we've aligned semantics a while back ... -Daniel > > Signed-off-by: Christian König <christian.koenig@amd.com> > --- > drivers/dma-buf/reservation.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/drivers/dma-buf/reservation.c b/drivers/dma-buf/reservation.c > index e2eff86..ce3f9c1 100644 > --- a/drivers/dma-buf/reservation.c > +++ b/drivers/dma-buf/reservation.c > @@ -461,7 +461,7 @@ long reservation_object_wait_timeout_rcu(struct reservation_object *obj, > } > } > > - if (!shared_count) { > + if (!fence) { > struct dma_fence *fence_excl = rcu_dereference(obj->fence_excl); > > if (fence_excl && > -- > 2.7.4 > > _______________________________________________ > dri-devel mailing list > dri-devel@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/dri-devel
On 2017年07月22日 00:20, Christian König wrote: > From: Christian König <christian.koenig@amd.com> > > With hardware resets in mind it is possible that all shared fences are > signaled, but the exlusive isn't. Fix waiting for everything in this situation. > > Signed-off-by: Christian König <christian.koenig@amd.com> > --- > drivers/dma-buf/reservation.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/drivers/dma-buf/reservation.c b/drivers/dma-buf/reservation.c > index e2eff86..ce3f9c1 100644 > --- a/drivers/dma-buf/reservation.c > +++ b/drivers/dma-buf/reservation.c > @@ -461,7 +461,7 @@ long reservation_object_wait_timeout_rcu(struct reservation_object *obj, > } > } > > - if (!shared_count) { > + if (!fence) { previous code seems be a bug, the exclusive fence isn't be waited at all if shared_count != 0. With your fix, there still is a case the exclusive fence could be skipped, that when fobj->shared[shared_count-1] isn't signalled. Regards, David Zhou > struct dma_fence *fence_excl = rcu_dereference(obj->fence_excl); > > if (fence_excl &&
Am 24.07.2017 um 10:33 schrieb Daniel Vetter: > On Fri, Jul 21, 2017 at 06:20:01PM +0200, Christian König wrote: >> From: Christian König <christian.koenig@amd.com> >> >> With hardware resets in mind it is possible that all shared fences are >> signaled, but the exlusive isn't. Fix waiting for everything in this situation. > How did you end up with both shared and exclusive fences on the same > reservation object? At least I thought the point of exclusive was that > it's exclusive (and has an implicit barrier on all previous shared > fences). Same for shared fences, they need to wait for the exclusive one > (and replace it). > > Is this fallout from the amdgpu trickery where by default you do all > shared fences? I thought we've aligned semantics a while back ... No, that is perfectly normal even for other drivers. Take a look at the reservation code. The exclusive fence replaces all shared fences, but adding a shared fence doesn't replace the exclusive fence. That actually makes sense, cause when you want to add move shared fences those need to wait for the last exclusive fence as well. Now normally I would agree that when you have shared fences it is sufficient to wait for all of them cause those operations can't start before the exclusive one finishes. But with GPU reset and/or the ability to abort already submitted operations it is perfectly possible that you end up with an exclusive fence which isn't signaled and a shared fence which is signaled in the same reservation object. Christian. > -Daniel > >> Signed-off-by: Christian König <christian.koenig@amd.com> >> --- >> drivers/dma-buf/reservation.c | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/drivers/dma-buf/reservation.c b/drivers/dma-buf/reservation.c >> index e2eff86..ce3f9c1 100644 >> --- a/drivers/dma-buf/reservation.c >> +++ b/drivers/dma-buf/reservation.c >> @@ -461,7 +461,7 @@ long reservation_object_wait_timeout_rcu(struct reservation_object *obj, >> } >> } >> >> - if (!shared_count) { >> + if (!fence) { >> struct dma_fence *fence_excl = rcu_dereference(obj->fence_excl); >> >> if (fence_excl && >> -- >> 2.7.4 >> >> _______________________________________________ >> dri-devel mailing list >> dri-devel@lists.freedesktop.org >> https://lists.freedesktop.org/mailman/listinfo/dri-devel
Am 24.07.2017 um 10:34 schrieb zhoucm1: > > > On 2017年07月22日 00:20, Christian König wrote: >> From: Christian König <christian.koenig@amd.com> >> >> With hardware resets in mind it is possible that all shared fences are >> signaled, but the exlusive isn't. Fix waiting for everything in this >> situation. >> >> Signed-off-by: Christian König <christian.koenig@amd.com> >> --- >> drivers/dma-buf/reservation.c | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/drivers/dma-buf/reservation.c >> b/drivers/dma-buf/reservation.c >> index e2eff86..ce3f9c1 100644 >> --- a/drivers/dma-buf/reservation.c >> +++ b/drivers/dma-buf/reservation.c >> @@ -461,7 +461,7 @@ long reservation_object_wait_timeout_rcu(struct >> reservation_object *obj, >> } >> } >> - if (!shared_count) { >> + if (!fence) { > previous code seems be a bug, the exclusive fence isn't be waited at > all if shared_count != 0. > > With your fix, there still is a case the exclusive fence could be > skipped, that when fobj->shared[shared_count-1] isn't signalled. Yeah, indeed that looks like it needs to be fixed as well. I'm still completely jet lagged and need to work through tons of stuff from last week. Do you have time to take care of fixing up this patch and send a v2? Thanks in advance, Christian. > > Regards, > David Zhou >> struct dma_fence *fence_excl = >> rcu_dereference(obj->fence_excl); >> if (fence_excl && >
On Mon, Jul 24, 2017 at 11:51 AM, Christian König <deathsimple@vodafone.de> wrote: > Am 24.07.2017 um 10:33 schrieb Daniel Vetter: >> >> On Fri, Jul 21, 2017 at 06:20:01PM +0200, Christian König wrote: >>> >>> From: Christian König <christian.koenig@amd.com> >>> >>> With hardware resets in mind it is possible that all shared fences are >>> signaled, but the exlusive isn't. Fix waiting for everything in this >>> situation. >> >> How did you end up with both shared and exclusive fences on the same >> reservation object? At least I thought the point of exclusive was that >> it's exclusive (and has an implicit barrier on all previous shared >> fences). Same for shared fences, they need to wait for the exclusive one >> (and replace it). >> >> Is this fallout from the amdgpu trickery where by default you do all >> shared fences? I thought we've aligned semantics a while back ... > > > No, that is perfectly normal even for other drivers. Take a look at the > reservation code. > > The exclusive fence replaces all shared fences, but adding a shared fence > doesn't replace the exclusive fence. That actually makes sense, cause when > you want to add move shared fences those need to wait for the last exclusive > fence as well. Hm right. > Now normally I would agree that when you have shared fences it is sufficient > to wait for all of them cause those operations can't start before the > exclusive one finishes. But with GPU reset and/or the ability to abort > already submitted operations it is perfectly possible that you end up with > an exclusive fence which isn't signaled and a shared fence which is signaled > in the same reservation object. How does that work? The batch(es) with the shared fence are all supposed to wait for the exclusive fence before they start, which means even if you gpu reset and restart/cancel certain things, they shouldn't be able to complete out of order. If you outright cancel a fence then you're supposed to first call dma_fence_set_error(-EIO) and then complete it. Note that atm that part might be slightly overengineered and I'm not sure about how we expose stuff to userspace, e.g. dma_fence_set_error(-EAGAIN) is (or soon, has been) used by i915 for it's internal book-keeping, which might not be the best to leak to other consumers. But completing fences (at least exported ones, where userspace or other drivers can get at them) shouldn't be possible. -Daniel
On 2017年07月24日 19:57, Daniel Vetter wrote: > On Mon, Jul 24, 2017 at 11:51 AM, Christian König > <deathsimple@vodafone.de> wrote: >> Am 24.07.2017 um 10:33 schrieb Daniel Vetter: >>> On Fri, Jul 21, 2017 at 06:20:01PM +0200, Christian König wrote: >>>> From: Christian König <christian.koenig@amd.com> >>>> >>>> With hardware resets in mind it is possible that all shared fences are >>>> signaled, but the exlusive isn't. Fix waiting for everything in this >>>> situation. >>> How did you end up with both shared and exclusive fences on the same >>> reservation object? At least I thought the point of exclusive was that >>> it's exclusive (and has an implicit barrier on all previous shared >>> fences). Same for shared fences, they need to wait for the exclusive one >>> (and replace it). >>> >>> Is this fallout from the amdgpu trickery where by default you do all >>> shared fences? I thought we've aligned semantics a while back ... >> >> No, that is perfectly normal even for other drivers. Take a look at the >> reservation code. >> >> The exclusive fence replaces all shared fences, but adding a shared fence >> doesn't replace the exclusive fence. That actually makes sense, cause when >> you want to add move shared fences those need to wait for the last exclusive >> fence as well. > Hm right. > >> Now normally I would agree that when you have shared fences it is sufficient >> to wait for all of them cause those operations can't start before the >> exclusive one finishes. But with GPU reset and/or the ability to abort >> already submitted operations it is perfectly possible that you end up with >> an exclusive fence which isn't signaled and a shared fence which is signaled >> in the same reservation object. > How does that work? The batch(es) with the shared fence are all > supposed to wait for the exclusive fence before they start, which > means even if you gpu reset and restart/cancel certain things, they > shouldn't be able to complete out of order. Hi Daniel, Do you mean exclusive fence must be signalled before any shared fence? Where could I find this restriction? Thanks, David Zhou > > If you outright cancel a fence then you're supposed to first call > dma_fence_set_error(-EIO) and then complete it. Note that atm that > part might be slightly overengineered and I'm not sure about how we > expose stuff to userspace, e.g. dma_fence_set_error(-EAGAIN) is (or > soon, has been) used by i915 for it's internal book-keeping, which > might not be the best to leak to other consumers. But completing > fences (at least exported ones, where userspace or other drivers can > get at them) shouldn't be possible. > -Daniel
On Tue, Jul 25, 2017 at 02:16:55PM +0800, zhoucm1 wrote: > > > On 2017年07月24日 19:57, Daniel Vetter wrote: > > On Mon, Jul 24, 2017 at 11:51 AM, Christian König > > <deathsimple@vodafone.de> wrote: > > > Am 24.07.2017 um 10:33 schrieb Daniel Vetter: > > > > On Fri, Jul 21, 2017 at 06:20:01PM +0200, Christian König wrote: > > > > > From: Christian König <christian.koenig@amd.com> > > > > > > > > > > With hardware resets in mind it is possible that all shared fences are > > > > > signaled, but the exlusive isn't. Fix waiting for everything in this > > > > > situation. > > > > How did you end up with both shared and exclusive fences on the same > > > > reservation object? At least I thought the point of exclusive was that > > > > it's exclusive (and has an implicit barrier on all previous shared > > > > fences). Same for shared fences, they need to wait for the exclusive one > > > > (and replace it). > > > > > > > > Is this fallout from the amdgpu trickery where by default you do all > > > > shared fences? I thought we've aligned semantics a while back ... > > > > > > No, that is perfectly normal even for other drivers. Take a look at the > > > reservation code. > > > > > > The exclusive fence replaces all shared fences, but adding a shared fence > > > doesn't replace the exclusive fence. That actually makes sense, cause when > > > you want to add move shared fences those need to wait for the last exclusive > > > fence as well. > > Hm right. > > > > > Now normally I would agree that when you have shared fences it is sufficient > > > to wait for all of them cause those operations can't start before the > > > exclusive one finishes. But with GPU reset and/or the ability to abort > > > already submitted operations it is perfectly possible that you end up with > > > an exclusive fence which isn't signaled and a shared fence which is signaled > > > in the same reservation object. > > How does that work? The batch(es) with the shared fence are all > > supposed to wait for the exclusive fence before they start, which > > means even if you gpu reset and restart/cancel certain things, they > > shouldn't be able to complete out of order. > Hi Daniel, > > Do you mean exclusive fence must be signalled before any shared fence? Where > could I find this restriction? Yes, Christian also described it above. Could be that we should have better kerneldoc to document this ... -Daniel > > Thanks, > David Zhou > > > > If you outright cancel a fence then you're supposed to first call > > dma_fence_set_error(-EIO) and then complete it. Note that atm that > > part might be slightly overengineered and I'm not sure about how we > > expose stuff to userspace, e.g. dma_fence_set_error(-EAGAIN) is (or > > soon, has been) used by i915 for it's internal book-keeping, which > > might not be the best to leak to other consumers. But completing > > fences (at least exported ones, where userspace or other drivers can > > get at them) shouldn't be possible. > > -Daniel >
On 2017年07月25日 14:50, Daniel Vetter wrote: > On Tue, Jul 25, 2017 at 02:16:55PM +0800, zhoucm1 wrote: >> >> On 2017年07月24日 19:57, Daniel Vetter wrote: >>> On Mon, Jul 24, 2017 at 11:51 AM, Christian König >>> <deathsimple@vodafone.de> wrote: >>>> Am 24.07.2017 um 10:33 schrieb Daniel Vetter: >>>>> On Fri, Jul 21, 2017 at 06:20:01PM +0200, Christian König wrote: >>>>>> From: Christian König <christian.koenig@amd.com> >>>>>> >>>>>> With hardware resets in mind it is possible that all shared fences are >>>>>> signaled, but the exlusive isn't. Fix waiting for everything in this >>>>>> situation. >>>>> How did you end up with both shared and exclusive fences on the same >>>>> reservation object? At least I thought the point of exclusive was that >>>>> it's exclusive (and has an implicit barrier on all previous shared >>>>> fences). Same for shared fences, they need to wait for the exclusive one >>>>> (and replace it). >>>>> >>>>> Is this fallout from the amdgpu trickery where by default you do all >>>>> shared fences? I thought we've aligned semantics a while back ... >>>> No, that is perfectly normal even for other drivers. Take a look at the >>>> reservation code. >>>> >>>> The exclusive fence replaces all shared fences, but adding a shared fence >>>> doesn't replace the exclusive fence. That actually makes sense, cause when >>>> you want to add move shared fences those need to wait for the last exclusive >>>> fence as well. >>> Hm right. >>> >>>> Now normally I would agree that when you have shared fences it is sufficient >>>> to wait for all of them cause those operations can't start before the >>>> exclusive one finishes. But with GPU reset and/or the ability to abort >>>> already submitted operations it is perfectly possible that you end up with >>>> an exclusive fence which isn't signaled and a shared fence which is signaled >>>> in the same reservation object. >>> How does that work? The batch(es) with the shared fence are all >>> supposed to wait for the exclusive fence before they start, which >>> means even if you gpu reset and restart/cancel certain things, they >>> shouldn't be able to complete out of order. >> Hi Daniel, >> >> Do you mean exclusive fence must be signalled before any shared fence? Where >> could I find this restriction? > Yes, Christian also described it above. Could be that we should have > better kerneldoc to document this ... Is that a known assumption? if that way, it doesn't matter even that we always wait exclusive fence, right? Just one more line checking. Thanks, David Zhou > -Daniel > >> Thanks, >> David Zhou >>> If you outright cancel a fence then you're supposed to first call >>> dma_fence_set_error(-EIO) and then complete it. Note that atm that >>> part might be slightly overengineered and I'm not sure about how we >>> expose stuff to userspace, e.g. dma_fence_set_error(-EAGAIN) is (or >>> soon, has been) used by i915 for it's internal book-keeping, which >>> might not be the best to leak to other consumers. But completing >>> fences (at least exported ones, where userspace or other drivers can >>> get at them) shouldn't be possible. >>> -Daniel
On Tue, Jul 25, 2017 at 02:55:14PM +0800, zhoucm1 wrote: > > > On 2017年07月25日 14:50, Daniel Vetter wrote: > > On Tue, Jul 25, 2017 at 02:16:55PM +0800, zhoucm1 wrote: > > > > > > On 2017年07月24日 19:57, Daniel Vetter wrote: > > > > On Mon, Jul 24, 2017 at 11:51 AM, Christian König > > > > <deathsimple@vodafone.de> wrote: > > > > > Am 24.07.2017 um 10:33 schrieb Daniel Vetter: > > > > > > On Fri, Jul 21, 2017 at 06:20:01PM +0200, Christian König wrote: > > > > > > > From: Christian König <christian.koenig@amd.com> > > > > > > > > > > > > > > With hardware resets in mind it is possible that all shared fences are > > > > > > > signaled, but the exlusive isn't. Fix waiting for everything in this > > > > > > > situation. > > > > > > How did you end up with both shared and exclusive fences on the same > > > > > > reservation object? At least I thought the point of exclusive was that > > > > > > it's exclusive (and has an implicit barrier on all previous shared > > > > > > fences). Same for shared fences, they need to wait for the exclusive one > > > > > > (and replace it). > > > > > > > > > > > > Is this fallout from the amdgpu trickery where by default you do all > > > > > > shared fences? I thought we've aligned semantics a while back ... > > > > > No, that is perfectly normal even for other drivers. Take a look at the > > > > > reservation code. > > > > > > > > > > The exclusive fence replaces all shared fences, but adding a shared fence > > > > > doesn't replace the exclusive fence. That actually makes sense, cause when > > > > > you want to add move shared fences those need to wait for the last exclusive > > > > > fence as well. > > > > Hm right. > > > > > > > > > Now normally I would agree that when you have shared fences it is sufficient > > > > > to wait for all of them cause those operations can't start before the > > > > > exclusive one finishes. But with GPU reset and/or the ability to abort > > > > > already submitted operations it is perfectly possible that you end up with > > > > > an exclusive fence which isn't signaled and a shared fence which is signaled > > > > > in the same reservation object. > > > > How does that work? The batch(es) with the shared fence are all > > > > supposed to wait for the exclusive fence before they start, which > > > > means even if you gpu reset and restart/cancel certain things, they > > > > shouldn't be able to complete out of order. > > > Hi Daniel, > > > > > > Do you mean exclusive fence must be signalled before any shared fence? Where > > > could I find this restriction? > > Yes, Christian also described it above. Could be that we should have > > better kerneldoc to document this ... > Is that a known assumption? if that way, it doesn't matter even that we > always wait exclusive fence, right? Just one more line checking. The problem is that amdgpu breaks that assumption over gpu reset, and that might have implications _everywhere_, not just in this code here. Are you sure this case won't pull the i915 driver over the table when sharing dma-bufs with it? Did you audit the code (plus all the other drivers too ofc). -Daniel
On 2017年07月25日 15:02, Daniel Vetter wrote: > On Tue, Jul 25, 2017 at 02:55:14PM +0800, zhoucm1 wrote: >> >> On 2017年07月25日 14:50, Daniel Vetter wrote: >>> On Tue, Jul 25, 2017 at 02:16:55PM +0800, zhoucm1 wrote: >>>> On 2017年07月24日 19:57, Daniel Vetter wrote: >>>>> On Mon, Jul 24, 2017 at 11:51 AM, Christian König >>>>> <deathsimple@vodafone.de> wrote: >>>>>> Am 24.07.2017 um 10:33 schrieb Daniel Vetter: >>>>>>> On Fri, Jul 21, 2017 at 06:20:01PM +0200, Christian König wrote: >>>>>>>> From: Christian König <christian.koenig@amd.com> >>>>>>>> >>>>>>>> With hardware resets in mind it is possible that all shared fences are >>>>>>>> signaled, but the exlusive isn't. Fix waiting for everything in this >>>>>>>> situation. >>>>>>> How did you end up with both shared and exclusive fences on the same >>>>>>> reservation object? At least I thought the point of exclusive was that >>>>>>> it's exclusive (and has an implicit barrier on all previous shared >>>>>>> fences). Same for shared fences, they need to wait for the exclusive one >>>>>>> (and replace it). >>>>>>> >>>>>>> Is this fallout from the amdgpu trickery where by default you do all >>>>>>> shared fences? I thought we've aligned semantics a while back ... >>>>>> No, that is perfectly normal even for other drivers. Take a look at the >>>>>> reservation code. >>>>>> >>>>>> The exclusive fence replaces all shared fences, but adding a shared fence >>>>>> doesn't replace the exclusive fence. That actually makes sense, cause when >>>>>> you want to add move shared fences those need to wait for the last exclusive >>>>>> fence as well. >>>>> Hm right. >>>>> >>>>>> Now normally I would agree that when you have shared fences it is sufficient >>>>>> to wait for all of them cause those operations can't start before the >>>>>> exclusive one finishes. But with GPU reset and/or the ability to abort >>>>>> already submitted operations it is perfectly possible that you end up with >>>>>> an exclusive fence which isn't signaled and a shared fence which is signaled >>>>>> in the same reservation object. >>>>> How does that work? The batch(es) with the shared fence are all >>>>> supposed to wait for the exclusive fence before they start, which >>>>> means even if you gpu reset and restart/cancel certain things, they >>>>> shouldn't be able to complete out of order. >>>> Hi Daniel, >>>> >>>> Do you mean exclusive fence must be signalled before any shared fence? Where >>>> could I find this restriction? >>> Yes, Christian also described it above. Could be that we should have >>> better kerneldoc to document this ... >> Is that a known assumption? if that way, it doesn't matter even that we >> always wait exclusive fence, right? Just one more line checking. > The problem is that amdgpu breaks that assumption over gpu reset, and that > might have implications _everywhere_, not just in this code here. Are you > sure this case won't pull the i915 driver over the table when sharing > dma-bufs with it? Ah, I finally got your mean. So as you mentioned before, we at least should have better describe for this assumption, the best place is comments in reservation.c, every newer would know it when reading code first time. Thanks, David Zhou > Did you audit the code (plus all the other drivers too > ofc). > -Daniel
Am 24.07.2017 um 13:57 schrieb Daniel Vetter: > On Mon, Jul 24, 2017 at 11:51 AM, Christian König > <deathsimple@vodafone.de> wrote: >> Am 24.07.2017 um 10:33 schrieb Daniel Vetter: >>> On Fri, Jul 21, 2017 at 06:20:01PM +0200, Christian König wrote: >>>> From: Christian König <christian.koenig@amd.com> >>>> >>>> With hardware resets in mind it is possible that all shared fences are >>>> signaled, but the exlusive isn't. Fix waiting for everything in this >>>> situation. >>> How did you end up with both shared and exclusive fences on the same >>> reservation object? At least I thought the point of exclusive was that >>> it's exclusive (and has an implicit barrier on all previous shared >>> fences). Same for shared fences, they need to wait for the exclusive one >>> (and replace it). >>> >>> Is this fallout from the amdgpu trickery where by default you do all >>> shared fences? I thought we've aligned semantics a while back ... >> >> No, that is perfectly normal even for other drivers. Take a look at the >> reservation code. >> >> The exclusive fence replaces all shared fences, but adding a shared fence >> doesn't replace the exclusive fence. That actually makes sense, cause when >> you want to add move shared fences those need to wait for the last exclusive >> fence as well. > Hm right. > >> Now normally I would agree that when you have shared fences it is sufficient >> to wait for all of them cause those operations can't start before the >> exclusive one finishes. But with GPU reset and/or the ability to abort >> already submitted operations it is perfectly possible that you end up with >> an exclusive fence which isn't signaled and a shared fence which is signaled >> in the same reservation object. > How does that work? The batch(es) with the shared fence are all > supposed to wait for the exclusive fence before they start, which > means even if you gpu reset and restart/cancel certain things, they > shouldn't be able to complete out of order. Assume the following: 1. The exclusive fence is some move operation by the kernel which executes on a DMA engine. 2. The shared fence is a 3D operation submitted by userspace which executes on the 3D engine. Now we found the 3D engine to be hung and needs a reset, all currently submitted jobs are aborted, marked with an error code and their fences put into the signaled state. Since we only reset the 3D engine, the move operation (fortunately) isn't affected by this. I think this applies to all drivers and isn't something amdgpu specific. Regards, Christian. > > If you outright cancel a fence then you're supposed to first call > dma_fence_set_error(-EIO) and then complete it. Note that atm that > part might be slightly overengineered and I'm not sure about how we > expose stuff to userspace, e.g. dma_fence_set_error(-EAGAIN) is (or > soon, has been) used by i915 for it's internal book-keeping, which > might not be the best to leak to other consumers. But completing > fences (at least exported ones, where userspace or other drivers can > get at them) shouldn't be possible. > -Daniel
On Tue, Jul 25, 2017 at 11:11:35AM +0200, Christian König wrote: > Am 24.07.2017 um 13:57 schrieb Daniel Vetter: > > On Mon, Jul 24, 2017 at 11:51 AM, Christian König > > <deathsimple@vodafone.de> wrote: > > > Am 24.07.2017 um 10:33 schrieb Daniel Vetter: > > > > On Fri, Jul 21, 2017 at 06:20:01PM +0200, Christian König wrote: > > > > > From: Christian König <christian.koenig@amd.com> > > > > > > > > > > With hardware resets in mind it is possible that all shared fences are > > > > > signaled, but the exlusive isn't. Fix waiting for everything in this > > > > > situation. > > > > How did you end up with both shared and exclusive fences on the same > > > > reservation object? At least I thought the point of exclusive was that > > > > it's exclusive (and has an implicit barrier on all previous shared > > > > fences). Same for shared fences, they need to wait for the exclusive one > > > > (and replace it). > > > > > > > > Is this fallout from the amdgpu trickery where by default you do all > > > > shared fences? I thought we've aligned semantics a while back ... > > > > > > No, that is perfectly normal even for other drivers. Take a look at the > > > reservation code. > > > > > > The exclusive fence replaces all shared fences, but adding a shared fence > > > doesn't replace the exclusive fence. That actually makes sense, cause when > > > you want to add move shared fences those need to wait for the last exclusive > > > fence as well. > > Hm right. > > > > > Now normally I would agree that when you have shared fences it is sufficient > > > to wait for all of them cause those operations can't start before the > > > exclusive one finishes. But with GPU reset and/or the ability to abort > > > already submitted operations it is perfectly possible that you end up with > > > an exclusive fence which isn't signaled and a shared fence which is signaled > > > in the same reservation object. > > How does that work? The batch(es) with the shared fence are all > > supposed to wait for the exclusive fence before they start, which > > means even if you gpu reset and restart/cancel certain things, they > > shouldn't be able to complete out of order. > > Assume the following: > 1. The exclusive fence is some move operation by the kernel which executes > on a DMA engine. > 2. The shared fence is a 3D operation submitted by userspace which executes > on the 3D engine. > > Now we found the 3D engine to be hung and needs a reset, all currently > submitted jobs are aborted, marked with an error code and their fences put > into the signaled state. > > Since we only reset the 3D engine, the move operation (fortunately) isn't > affected by this. > > I think this applies to all drivers and isn't something amdgpu specific. Not i915 because: - At first we only had system wide gpu reset that killed everything, which means all requests will be completed, not just on a single engine. - Now we have per-engine reset, but combined with replaying them (the offending one gets a no-op batch to avoid re-hanging), to make sure the depency tree doesn't fall apart. Now I see that doing this isn't all that simple, and either way we still have the case where one driver resets but not the other (in multi-gpu), but I'm not exactly sure how to best handle this. What exactly is the downside of not dropping this assumption, i.e. why do you want this patch? What blows up? -Daniel > > Regards, > Christian. > > > > > If you outright cancel a fence then you're supposed to first call > > dma_fence_set_error(-EIO) and then complete it. Note that atm that > > part might be slightly overengineered and I'm not sure about how we > > expose stuff to userspace, e.g. dma_fence_set_error(-EAGAIN) is (or > > soon, has been) used by i915 for it's internal book-keeping, which > > might not be the best to leak to other consumers. But completing > > fences (at least exported ones, where userspace or other drivers can > > get at them) shouldn't be possible. > > -Daniel > >
diff --git a/drivers/dma-buf/reservation.c b/drivers/dma-buf/reservation.c index e2eff86..ce3f9c1 100644 --- a/drivers/dma-buf/reservation.c +++ b/drivers/dma-buf/reservation.c @@ -461,7 +461,7 @@ long reservation_object_wait_timeout_rcu(struct reservation_object *obj, } } - if (!shared_count) { + if (!fence) { struct dma_fence *fence_excl = rcu_dereference(obj->fence_excl); if (fence_excl &&