| Message ID | 20191025234834.28214-1-john.stultz@linaro.org (mailing list archive) |
|---|---|
| Headers | show |
| Series | Allow DMA BUF heaps to be loaded as modules | expand |
On Fri, Oct 25, 2019 at 11:48:32PM +0000, John Stultz wrote: > Now that the DMA BUF heaps core code has been queued, I wanted > to send out some of the pending changes that I've been working > on. > > For use with Android and their GKI effort, it is desired that > DMA BUF heaps are able to be loaded as modules. This is required > for migrating vendors off of ION which was also recently changed > to support modules. > > So this patch series simply provides the necessary exported > symbols and allows the system and CMA drivers to be built > as modules. > > Due to the fact that dmabuf's allocated from a heap may > be in use for quite some time, there isn't a way to safely > unload the driver once it has been loaded. Thus these > drivers do no implement module_exit() functions and will > show up in lsmod as "[permanent]" > > Feedback and thoughts on this would be greatly appreciated! Do we actually want this? I figured if we just state that vendors should set up all the right dma-buf heaps in dt, is that not enough? Exporting symbols for no real in-tree users feels fishy. -Daniel > > thanks > -john > > Cc: Laura Abbott <labbott@redhat.com> > Cc: Benjamin Gaignard <benjamin.gaignard@linaro.org> > Cc: Sumit Semwal <sumit.semwal@linaro.org> > Cc: Liam Mark <lmark@codeaurora.org> > Cc: Pratik Patel <pratikp@codeaurora.org> > Cc: Brian Starkey <Brian.Starkey@arm.com> > Cc: Andrew F. Davis <afd@ti.com> > Cc: Andrew Morton <akpm@linux-foundation.org> > Cc: Yue Hu <huyue2@yulong.com> > Cc: Mike Rapoport <rppt@linux.ibm.com> > Cc: Chenbo Feng <fengc@google.com> > Cc: Alistair Strachan <astrachan@google.com> > Cc: Sandeep Patil <sspatil@google.com> > Cc: Hridya Valsaraju <hridya@google.com> > Cc: dri-devel@lists.freedesktop.org > > John Stultz (1): > dma-buf: heaps: Allow system & cma heaps to be configured as a modules > > Sandeep Patil (1): > mm: cma: Export cma symbols for cma heap as a module > > drivers/dma-buf/dma-heap.c | 2 ++ > drivers/dma-buf/heaps/Kconfig | 4 ++-- > drivers/dma-buf/heaps/heap-helpers.c | 2 ++ > kernel/dma/contiguous.c | 1 + > mm/cma.c | 5 +++++ > 5 files changed, 12 insertions(+), 2 deletions(-) > > -- > 2.17.1 > > _______________________________________________ > dri-devel mailing list > dri-devel@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/dri-devel
On Mon, Nov 4, 2019 at 1:58 AM Daniel Vetter <daniel@ffwll.ch> wrote: > On Fri, Oct 25, 2019 at 11:48:32PM +0000, John Stultz wrote: > > Now that the DMA BUF heaps core code has been queued, I wanted > > to send out some of the pending changes that I've been working > > on. > > > > For use with Android and their GKI effort, it is desired that > > DMA BUF heaps are able to be loaded as modules. This is required > > for migrating vendors off of ION which was also recently changed > > to support modules. > > > > So this patch series simply provides the necessary exported > > symbols and allows the system and CMA drivers to be built > > as modules. > > > > Due to the fact that dmabuf's allocated from a heap may > > be in use for quite some time, there isn't a way to safely > > unload the driver once it has been loaded. Thus these > > drivers do no implement module_exit() functions and will > > show up in lsmod as "[permanent]" > > > > Feedback and thoughts on this would be greatly appreciated! > > Do we actually want this? I guess that always depends on the definition of "we" :) > I figured if we just state that vendors should set up all the right > dma-buf heaps in dt, is that not enough? So even if the heaps are configured via DT (which at the moment there is no such binding, so that's not really a valid method yet), there's still the question of if the heap is necessary/makes sense on the device. And the DT would only control the availability of the heap interface, not if the heap driver is loaded or not. On the HiKey/HiKey960 boards, we have to allocate contiguous buffers for the display framebuffer. So gralloc uses ION to allocate from the CMA heap. However on the db845c, it has no such restrictions, so the CMA heap isn't necessary. With Android's GKI effort, there needs to be one kernel that works on all the devices, and they are using modules to try to minimize the amount of memory spent on functionality that isn't universally needed. So on devices that don't need the CMA heap, they'd probably prefer not to load the CMA dmabuf heap driver, so it would be best if it could be built as a module. If we want to build the CMA heap as a module, the symbols it uses need to be exported. > Exporting symbols for no real in-tree users feels fishy. I'm submitting an in-tree user here. So I'm not sure what you mean? I suspect you're thinking there is some hidden/nefarious plan here, but really there isn't. thanks -john
On Mon, Nov 04, 2019 at 10:57:44AM -0800, John Stultz wrote: > On Mon, Nov 4, 2019 at 1:58 AM Daniel Vetter <daniel@ffwll.ch> wrote: > > On Fri, Oct 25, 2019 at 11:48:32PM +0000, John Stultz wrote: > > > Now that the DMA BUF heaps core code has been queued, I wanted > > > to send out some of the pending changes that I've been working > > > on. > > > > > > For use with Android and their GKI effort, it is desired that > > > DMA BUF heaps are able to be loaded as modules. This is required > > > for migrating vendors off of ION which was also recently changed > > > to support modules. > > > > > > So this patch series simply provides the necessary exported > > > symbols and allows the system and CMA drivers to be built > > > as modules. > > > > > > Due to the fact that dmabuf's allocated from a heap may > > > be in use for quite some time, there isn't a way to safely > > > unload the driver once it has been loaded. Thus these > > > drivers do no implement module_exit() functions and will > > > show up in lsmod as "[permanent]" > > > > > > Feedback and thoughts on this would be greatly appreciated! > > > > Do we actually want this? > > I guess that always depends on the definition of "we" :) > > > I figured if we just state that vendors should set up all the right > > dma-buf heaps in dt, is that not enough? > > So even if the heaps are configured via DT (which at the moment there > is no such binding, so that's not really a valid method yet), there's > still the question of if the heap is necessary/makes sense on the > device. And the DT would only control the availability of the heap > interface, not if the heap driver is loaded or not. Hm I thought the cma regions are configured in DT? How does that work if it's not using DT? > On the HiKey/HiKey960 boards, we have to allocate contiguous buffers > for the display framebuffer. So gralloc uses ION to allocate from the > CMA heap. However on the db845c, it has no such restrictions, so the > CMA heap isn't necessary. Why do you have a CMA region for the 2nd board if you don't need it? _That_ sounds like some serious memory waster, not a few lines of code loaded for nothing :-) > With Android's GKI effort, there needs to be one kernel that works on > all the devices, and they are using modules to try to minimize the > amount of memory spent on functionality that isn't universally needed. > So on devices that don't need the CMA heap, they'd probably prefer not > to load the CMA dmabuf heap driver, so it would be best if it could be > built as a module. If we want to build the CMA heap as a module, the > symbols it uses need to be exported. Yeah, I guess I'm disagreeing on whether dma-buf heaps are core or not. > > Exporting symbols for no real in-tree users feels fishy. > > I'm submitting an in-tree user here. So I'm not sure what you mean? I > suspect you're thinking there is some hidden/nefarious plan here, but > really there isn't. I was working under the assumption that you're only exporting the symbols for other heaps, and keep the current ones in-tree. Are there even any out-of-tree dma-buf heaps still? out-of-tree and legit different use-case I mean ofc, not just out-of-tree because inertia :-) -Daniel
On 11/5/19 4:42 AM, Daniel Vetter wrote: > On Mon, Nov 04, 2019 at 10:57:44AM -0800, John Stultz wrote: >> On Mon, Nov 4, 2019 at 1:58 AM Daniel Vetter <daniel@ffwll.ch> wrote: >>> On Fri, Oct 25, 2019 at 11:48:32PM +0000, John Stultz wrote: >>>> Now that the DMA BUF heaps core code has been queued, I wanted >>>> to send out some of the pending changes that I've been working >>>> on. >>>> >>>> For use with Android and their GKI effort, it is desired that >>>> DMA BUF heaps are able to be loaded as modules. This is required >>>> for migrating vendors off of ION which was also recently changed >>>> to support modules. >>>> >>>> So this patch series simply provides the necessary exported >>>> symbols and allows the system and CMA drivers to be built >>>> as modules. >>>> >>>> Due to the fact that dmabuf's allocated from a heap may >>>> be in use for quite some time, there isn't a way to safely >>>> unload the driver once it has been loaded. Thus these >>>> drivers do no implement module_exit() functions and will >>>> show up in lsmod as "[permanent]" >>>> >>>> Feedback and thoughts on this would be greatly appreciated! >>> >>> Do we actually want this? >> >> I guess that always depends on the definition of "we" :) >> >>> I figured if we just state that vendors should set up all the right >>> dma-buf heaps in dt, is that not enough? >> >> So even if the heaps are configured via DT (which at the moment there >> is no such binding, so that's not really a valid method yet), there's >> still the question of if the heap is necessary/makes sense on the >> device. And the DT would only control the availability of the heap >> interface, not if the heap driver is loaded or not. > > Hm I thought the cma regions are configured in DT? How does that work if > it's not using DT? > >> On the HiKey/HiKey960 boards, we have to allocate contiguous buffers >> for the display framebuffer. So gralloc uses ION to allocate from the >> CMA heap. However on the db845c, it has no such restrictions, so the >> CMA heap isn't necessary. > > Why do you have a CMA region for the 2nd board if you don't need it? > _That_ sounds like some serious memory waster, not a few lines of code > loaded for nothing :-) > >> With Android's GKI effort, there needs to be one kernel that works on >> all the devices, and they are using modules to try to minimize the >> amount of memory spent on functionality that isn't universally needed. >> So on devices that don't need the CMA heap, they'd probably prefer not >> to load the CMA dmabuf heap driver, so it would be best if it could be >> built as a module. If we want to build the CMA heap as a module, the >> symbols it uses need to be exported. > > Yeah, I guess I'm disagreeing on whether dma-buf heaps are core or not. > >>> Exporting symbols for no real in-tree users feels fishy. >> >> I'm submitting an in-tree user here. So I'm not sure what you mean? I >> suspect you're thinking there is some hidden/nefarious plan here, but >> really there isn't. > > I was working under the assumption that you're only exporting the symbols > for other heaps, and keep the current ones in-tree. Are there even any > out-of-tree dma-buf heaps still? out-of-tree and legit different use-case > I mean ofc, not just out-of-tree because inertia :-) Not sure what you mean here, hopefully all the heaps can be "in-tree" some day. https://patchwork.kernel.org/patch/10863957/ Plus some non-caching heaps and one that forces early allocation of our PAT (gart like) IP. All this stuff is going into our evil vendor tree next cycle (if not upstream by then :)), if we want some of these "specialty" heaps to go into generic kernel builds at some point they will need to be modules if the core is. Although I am still thinking Heaps should be always built in + system + CMA heaps, then the rando heaps could be modules if needed. Andrew > -Daniel >
On Tue, Nov 5, 2019 at 2:30 PM Andrew F. Davis <afd@ti.com> wrote: > > On 11/5/19 4:42 AM, Daniel Vetter wrote: > > On Mon, Nov 04, 2019 at 10:57:44AM -0800, John Stultz wrote: > >> On Mon, Nov 4, 2019 at 1:58 AM Daniel Vetter <daniel@ffwll.ch> wrote: > >>> On Fri, Oct 25, 2019 at 11:48:32PM +0000, John Stultz wrote: > >>>> Now that the DMA BUF heaps core code has been queued, I wanted > >>>> to send out some of the pending changes that I've been working > >>>> on. > >>>> > >>>> For use with Android and their GKI effort, it is desired that > >>>> DMA BUF heaps are able to be loaded as modules. This is required > >>>> for migrating vendors off of ION which was also recently changed > >>>> to support modules. > >>>> > >>>> So this patch series simply provides the necessary exported > >>>> symbols and allows the system and CMA drivers to be built > >>>> as modules. > >>>> > >>>> Due to the fact that dmabuf's allocated from a heap may > >>>> be in use for quite some time, there isn't a way to safely > >>>> unload the driver once it has been loaded. Thus these > >>>> drivers do no implement module_exit() functions and will > >>>> show up in lsmod as "[permanent]" > >>>> > >>>> Feedback and thoughts on this would be greatly appreciated! > >>> > >>> Do we actually want this? > >> > >> I guess that always depends on the definition of "we" :) > >> > >>> I figured if we just state that vendors should set up all the right > >>> dma-buf heaps in dt, is that not enough? > >> > >> So even if the heaps are configured via DT (which at the moment there > >> is no such binding, so that's not really a valid method yet), there's > >> still the question of if the heap is necessary/makes sense on the > >> device. And the DT would only control the availability of the heap > >> interface, not if the heap driver is loaded or not. > > > > Hm I thought the cma regions are configured in DT? How does that work if > > it's not using DT? > > > >> On the HiKey/HiKey960 boards, we have to allocate contiguous buffers > >> for the display framebuffer. So gralloc uses ION to allocate from the > >> CMA heap. However on the db845c, it has no such restrictions, so the > >> CMA heap isn't necessary. > > > > Why do you have a CMA region for the 2nd board if you don't need it? > > _That_ sounds like some serious memory waster, not a few lines of code > > loaded for nothing :-) > > > >> With Android's GKI effort, there needs to be one kernel that works on > >> all the devices, and they are using modules to try to minimize the > >> amount of memory spent on functionality that isn't universally needed. > >> So on devices that don't need the CMA heap, they'd probably prefer not > >> to load the CMA dmabuf heap driver, so it would be best if it could be > >> built as a module. If we want to build the CMA heap as a module, the > >> symbols it uses need to be exported. > > > > Yeah, I guess I'm disagreeing on whether dma-buf heaps are core or not. > > > >>> Exporting symbols for no real in-tree users feels fishy. > >> > >> I'm submitting an in-tree user here. So I'm not sure what you mean? I > >> suspect you're thinking there is some hidden/nefarious plan here, but > >> really there isn't. > > > > I was working under the assumption that you're only exporting the symbols > > for other heaps, and keep the current ones in-tree. Are there even any > > out-of-tree dma-buf heaps still? out-of-tree and legit different use-case > > I mean ofc, not just out-of-tree because inertia :-) > > Not sure what you mean here, hopefully all the heaps can be "in-tree" > some day. > > https://patchwork.kernel.org/patch/10863957/ No idea this is good or bad, where's the userspace for it? > Plus some non-caching heaps and one that forces early allocation of our > PAT (gart like) IP. Hm, so essentially we'd need to move _all_ drm allocators into dma-buf heaps, for all drivers? Can't just do this for TI only ... > All this stuff is going into our evil vendor tree next cycle (if not > upstream by then :)), if we want some of these "specialty" heaps to go > into generic kernel builds at some point they will need to be modules if > the core is. > > Although I am still thinking Heaps should be always built in + system + > CMA heaps, then the rando heaps could be modules if needed. Yeah that's what I'd expected to happen. Speciality heaps for when you have the relevant something unusual (and I'm not sure whether upstream does want something unusual really, the above examples from you sound a bit strange). -Daniel
On Tue, Nov 5, 2019 at 1:43 AM Daniel Vetter <daniel@ffwll.ch> wrote: > > On Mon, Nov 04, 2019 at 10:57:44AM -0800, John Stultz wrote: > > On Mon, Nov 4, 2019 at 1:58 AM Daniel Vetter <daniel@ffwll.ch> wrote: > > > On Fri, Oct 25, 2019 at 11:48:32PM +0000, John Stultz wrote: > > So even if the heaps are configured via DT (which at the moment there > > is no such binding, so that's not really a valid method yet), there's > > still the question of if the heap is necessary/makes sense on the > > device. And the DT would only control the availability of the heap > > interface, not if the heap driver is loaded or not. > > Hm I thought the cma regions are configured in DT? How does that work if > it's not using DT? So yea, CMA regions are either configured by DT or setup at build time (I think there's a cmdline option to set it up as well). But the CMA regions and the dmabuf cma heap driver are separate things. The latter uses the former. > > On the HiKey/HiKey960 boards, we have to allocate contiguous buffers > > for the display framebuffer. So gralloc uses ION to allocate from the > > CMA heap. However on the db845c, it has no such restrictions, so the > > CMA heap isn't necessary. > > Why do you have a CMA region for the 2nd board if you don't need it? > _That_ sounds like some serious memory waster, not a few lines of code > loaded for nothing :-) ??? That's not what I said above. If the db845c doesn't need CMA it won't have a CMA region. The issue at hand is that we may want to avoid loading the dmabuf CMA heap driver on a board that doesn't use CMA. > > With Android's GKI effort, there needs to be one kernel that works on > > all the devices, and they are using modules to try to minimize the > > amount of memory spent on functionality that isn't universally needed. > > So on devices that don't need the CMA heap, they'd probably prefer not > > to load the CMA dmabuf heap driver, so it would be best if it could be > > built as a module. If we want to build the CMA heap as a module, the > > symbols it uses need to be exported. > > Yeah, I guess I'm disagreeing on whether dma-buf heaps are core or not. That's fine to dispute. I'm not really in a place to assert one way or not, but the Android folks have made their ION system and CMA heaps loadable via a module, so it would seem like having the dmabuf system and CMA heaps be modular would be useful to properly replace that usage. For instance, the system heap as a module probably doesn't make much sense, as most boards that want to use the dmabuf heaps interface are likely to use that as well. CMA is more optional as not all boards will use that one, so it might make sense to avoid loading it. Sandeep: Can you chime in here as to how critical having the system and cma heaps be modules are? > > > Exporting symbols for no real in-tree users feels fishy. > > > > I'm submitting an in-tree user here. So I'm not sure what you mean? I > > suspect you're thinking there is some hidden/nefarious plan here, but > > really there isn't. > > I was working under the assumption that you're only exporting the symbols > for other heaps, and keep the current ones in-tree. No. I'm trying to allow the (hopefully-soon-to-be-in-tree) system and cma heap drivers to be loaded from a module. If other heaps need exports, they can submit their heaps upstream and argue for the exported symbols themselves. > Are there even any > out-of-tree dma-buf heaps still? out-of-tree and legit different use-case > I mean ofc, not just out-of-tree because inertia :-) So as Andrew mentioned, he has some dmabuf heaps he's working on at TI. From what I've heard the qualcomm folks like the dmabuf heaps approach, but I'm not sure if they've taken a pass at converting their vendor ION heaps to it yet. The main reason I'm only submitting system and CMA is because those are the only two I personally have a user for (HiKey/HiKey960 boards). It's my hope that once we deprecate ION in Android, vendors will migrate and we'll be able to push them to upstream their heaps as well. thanks -john
On Tue, Nov 5, 2019 at 6:41 PM John Stultz <john.stultz@linaro.org> wrote: > > On Tue, Nov 5, 2019 at 1:43 AM Daniel Vetter <daniel@ffwll.ch> wrote: > > > > On Mon, Nov 04, 2019 at 10:57:44AM -0800, John Stultz wrote: > > > On Mon, Nov 4, 2019 at 1:58 AM Daniel Vetter <daniel@ffwll.ch> wrote: > > > > On Fri, Oct 25, 2019 at 11:48:32PM +0000, John Stultz wrote: > > > So even if the heaps are configured via DT (which at the moment there > > > is no such binding, so that's not really a valid method yet), there's > > > still the question of if the heap is necessary/makes sense on the > > > device. And the DT would only control the availability of the heap > > > interface, not if the heap driver is loaded or not. > > > > Hm I thought the cma regions are configured in DT? How does that work if > > it's not using DT? > > So yea, CMA regions are either configured by DT or setup at build time > (I think there's a cmdline option to set it up as well). > > But the CMA regions and the dmabuf cma heap driver are separate > things. The latter uses the former. Huh, I assumed the plan is that whenever there's a cma region, we want to instantiate a dma-buf heap for it? Why/when would we not want to do that? > > > On the HiKey/HiKey960 boards, we have to allocate contiguous buffers > > > for the display framebuffer. So gralloc uses ION to allocate from the > > > CMA heap. However on the db845c, it has no such restrictions, so the > > > CMA heap isn't necessary. > > > > Why do you have a CMA region for the 2nd board if you don't need it? > > _That_ sounds like some serious memory waster, not a few lines of code > > loaded for nothing :-) > > ??? That's not what I said above. If the db845c doesn't need CMA it > won't have a CMA region. > > The issue at hand is that we may want to avoid loading the dmabuf CMA > heap driver on a board that doesn't use CMA. So the CMA core code is also a module, or how does that work? Not loading the cma dma-buf heap, but keeping all the other cma code around feels slightly silly. Do we have numbers that justify this silliness? > > > With Android's GKI effort, there needs to be one kernel that works on > > > all the devices, and they are using modules to try to minimize the > > > amount of memory spent on functionality that isn't universally needed. > > > So on devices that don't need the CMA heap, they'd probably prefer not > > > to load the CMA dmabuf heap driver, so it would be best if it could be > > > built as a module. If we want to build the CMA heap as a module, the > > > symbols it uses need to be exported. > > > > Yeah, I guess I'm disagreeing on whether dma-buf heaps are core or not. > > That's fine to dispute. I'm not really in a place to assert one way or > not, but the Android folks have made their ION system and CMA heaps > loadable via a module, so it would seem like having the dmabuf system > and CMA heaps be modular would be useful to properly replace that > usage. > > For instance, the system heap as a module probably doesn't make much > sense, as most boards that want to use the dmabuf heaps interface are > likely to use that as well. CMA is more optional as not all boards > will use that one, so it might make sense to avoid loading it. > > Sandeep: Can you chime in here as to how critical having the system > and cma heaps be modules are? > > > > > > Exporting symbols for no real in-tree users feels fishy. > > > > > > I'm submitting an in-tree user here. So I'm not sure what you mean? I > > > suspect you're thinking there is some hidden/nefarious plan here, but > > > really there isn't. > > > > I was working under the assumption that you're only exporting the symbols > > for other heaps, and keep the current ones in-tree. > > No. I'm trying to allow the (hopefully-soon-to-be-in-tree) system and > cma heap drivers to be loaded from a module. > If other heaps need exports, they can submit their heaps upstream and > argue for the exported symbols themselves. > > > Are there even any > > out-of-tree dma-buf heaps still? out-of-tree and legit different use-case > > I mean ofc, not just out-of-tree because inertia :-) > > So as Andrew mentioned, he has some dmabuf heaps he's working on at > TI. From what I've heard the qualcomm folks like the dmabuf heaps > approach, but I'm not sure if they've taken a pass at converting their > vendor ION heaps to it yet. > > The main reason I'm only submitting system and CMA is because those > are the only two I personally have a user for (HiKey/HiKey960 boards). > It's my hope that once we deprecate ION in Android, vendors will > migrate and we'll be able to push them to upstream their heaps as > well. I think for upstream I'd want to see those other heaps first. If they're mostly driver allocators exposed as heaps, then I think we need something different than heap modules, maybe allow dma-buf to allocate from drivers instead. But afaiui all such driver allocators we have in upstream are cma regions only right now. -Daniel
On Tue, Nov 5, 2019 at 11:19 AM Daniel Vetter <daniel@ffwll.ch> wrote: > On Tue, Nov 5, 2019 at 6:41 PM John Stultz <john.stultz@linaro.org> wrote: > > On Tue, Nov 5, 2019 at 1:43 AM Daniel Vetter <daniel@ffwll.ch> wrote: > > > > > > On Mon, Nov 04, 2019 at 10:57:44AM -0800, John Stultz wrote: > > > > On Mon, Nov 4, 2019 at 1:58 AM Daniel Vetter <daniel@ffwll.ch> wrote: > > > > > On Fri, Oct 25, 2019 at 11:48:32PM +0000, John Stultz wrote: > > > > So even if the heaps are configured via DT (which at the moment there > > > > is no such binding, so that's not really a valid method yet), there's > > > > still the question of if the heap is necessary/makes sense on the > > > > device. And the DT would only control the availability of the heap > > > > interface, not if the heap driver is loaded or not. > > > > > > Hm I thought the cma regions are configured in DT? How does that work if > > > it's not using DT? > > > > So yea, CMA regions are either configured by DT or setup at build time > > (I think there's a cmdline option to set it up as well). > > > > But the CMA regions and the dmabuf cma heap driver are separate > > things. The latter uses the former. > > Huh, I assumed the plan is that whenever there's a cma region, we want > to instantiate a dma-buf heap for it? Why/when would we not want to do > that? Not quite. Andrew noted that we may not want to expose all CMA regions via dmabuf heaps, so right now we only expose the default region. I have follow on patches that I sent out earlier (which requires a to-be-finalized DT binding) which allows us to specify which other CMA regions to expose. > > > > On the HiKey/HiKey960 boards, we have to allocate contiguous buffers > > > > for the display framebuffer. So gralloc uses ION to allocate from the > > > > CMA heap. However on the db845c, it has no such restrictions, so the > > > > CMA heap isn't necessary. > > > > > > Why do you have a CMA region for the 2nd board if you don't need it? > > > _That_ sounds like some serious memory waster, not a few lines of code > > > loaded for nothing :-) > > > > ??? That's not what I said above. If the db845c doesn't need CMA it > > won't have a CMA region. > > > > The issue at hand is that we may want to avoid loading the dmabuf CMA > > heap driver on a board that doesn't use CMA. > > So the CMA core code is also a module, or how does that work? Not No. CMA core isn't available as a module. > loading the cma dma-buf heap, but keeping all the other cma code > around feels slightly silly. Do we have numbers that justify this > silliness? I agree that is maybe not the most critical item on the list, but its one of many that do add up over time. Again, I'll defer to Sandeep or other folks on the Google side to help with the importance here. Mostly I'm trying to ensure there is functional parity to ION so we don't give folks any reason they could object to deprecating it. > > The main reason I'm only submitting system and CMA is because those > > are the only two I personally have a user for (HiKey/HiKey960 boards). > > It's my hope that once we deprecate ION in Android, vendors will > > migrate and we'll be able to push them to upstream their heaps as > > well. > > I think for upstream I'd want to see those other heaps first. If > they're mostly driver allocators exposed as heaps, then I think we > need something different than heap modules, maybe allow dma-buf to > allocate from drivers instead. But afaiui all such driver allocators > we have in upstream are cma regions only right now. I'm sorry, I'm not sure I understand what you mean here (I'm not sure what action I should be taking). Could you clarify this point? thanks -john
On Tue, Nov 5, 2019 at 8:48 PM John Stultz <john.stultz@linaro.org> wrote: > > On Tue, Nov 5, 2019 at 11:19 AM Daniel Vetter <daniel@ffwll.ch> wrote: > > On Tue, Nov 5, 2019 at 6:41 PM John Stultz <john.stultz@linaro.org> wrote: > > > On Tue, Nov 5, 2019 at 1:43 AM Daniel Vetter <daniel@ffwll.ch> wrote: > > > > > > > > On Mon, Nov 04, 2019 at 10:57:44AM -0800, John Stultz wrote: > > > > > On Mon, Nov 4, 2019 at 1:58 AM Daniel Vetter <daniel@ffwll.ch> wrote: > > > > > > On Fri, Oct 25, 2019 at 11:48:32PM +0000, John Stultz wrote: > > > > > So even if the heaps are configured via DT (which at the moment there > > > > > is no such binding, so that's not really a valid method yet), there's > > > > > still the question of if the heap is necessary/makes sense on the > > > > > device. And the DT would only control the availability of the heap > > > > > interface, not if the heap driver is loaded or not. > > > > > > > > Hm I thought the cma regions are configured in DT? How does that work if > > > > it's not using DT? > > > > > > So yea, CMA regions are either configured by DT or setup at build time > > > (I think there's a cmdline option to set it up as well). > > > > > > But the CMA regions and the dmabuf cma heap driver are separate > > > things. The latter uses the former. > > > > Huh, I assumed the plan is that whenever there's a cma region, we want > > to instantiate a dma-buf heap for it? Why/when would we not want to do > > that? > > Not quite. Andrew noted that we may not want to expose all CMA regions > via dmabuf heaps, so right now we only expose the default region. I > have follow on patches that I sent out earlier (which requires a > to-be-finalized DT binding) which allows us to specify which other CMA > regions to expose. Why do we not want to expose them all? I figured if there's a cma heap, then a device you have needs it, and if that's the case you might want to allocate for that device from the heap? Maybe link to the discussion? > > > > > On the HiKey/HiKey960 boards, we have to allocate contiguous buffers > > > > > for the display framebuffer. So gralloc uses ION to allocate from the > > > > > CMA heap. However on the db845c, it has no such restrictions, so the > > > > > CMA heap isn't necessary. > > > > > > > > Why do you have a CMA region for the 2nd board if you don't need it? > > > > _That_ sounds like some serious memory waster, not a few lines of code > > > > loaded for nothing :-) > > > > > > ??? That's not what I said above. If the db845c doesn't need CMA it > > > won't have a CMA region. > > > > > > The issue at hand is that we may want to avoid loading the dmabuf CMA > > > heap driver on a board that doesn't use CMA. > > > > So the CMA core code is also a module, or how does that work? Not > > No. CMA core isn't available as a module. > > > loading the cma dma-buf heap, but keeping all the other cma code > > around feels slightly silly. Do we have numbers that justify this > > silliness? > > I agree that is maybe not the most critical item on the list, but its > one of many that do add up over time. > > Again, I'll defer to Sandeep or other folks on the Google side to help > with the importance here. Mostly I'm trying to ensure there is > functional parity to ION so we don't give folks any reason they could > object to deprecating it. > > > > The main reason I'm only submitting system and CMA is because those > > > are the only two I personally have a user for (HiKey/HiKey960 boards). > > > It's my hope that once we deprecate ION in Android, vendors will > > > migrate and we'll be able to push them to upstream their heaps as > > > well. > > > > I think for upstream I'd want to see those other heaps first. If > > they're mostly driver allocators exposed as heaps, then I think we > > need something different than heap modules, maybe allow dma-buf to > > allocate from drivers instead. But afaiui all such driver allocators > > we have in upstream are cma regions only right now. > > I'm sorry, I'm not sure I understand what you mean here (I'm not sure > what action I should be taking). Could you clarify this point? I'm not sold on the use-case for this, but maybe if I see the actual use-cases I might be swayed. A very basic/minimal "register a new dma-buf heap" function should be all that's really needed for android, so maybe we can start with that? -Daniel
On Tue, Nov 05, 2019 at 09:41:44AM -0800, John Stultz wrote: > On Tue, Nov 5, 2019 at 1:43 AM Daniel Vetter <daniel@ffwll.ch> wrote: > > > > On Mon, Nov 04, 2019 at 10:57:44AM -0800, John Stultz wrote: > > > On Mon, Nov 4, 2019 at 1:58 AM Daniel Vetter <daniel@ffwll.ch> wrote: > > > > On Fri, Oct 25, 2019 at 11:48:32PM +0000, John Stultz wrote: > > > So even if the heaps are configured via DT (which at the moment there > > > is no such binding, so that's not really a valid method yet), there's > > > still the question of if the heap is necessary/makes sense on the > > > device. And the DT would only control the availability of the heap > > > interface, not if the heap driver is loaded or not. > > > > Hm I thought the cma regions are configured in DT? How does that work if > > it's not using DT? > > So yea, CMA regions are either configured by DT or setup at build time > (I think there's a cmdline option to set it up as well). > > But the CMA regions and the dmabuf cma heap driver are separate > things. The latter uses the former. > > > > On the HiKey/HiKey960 boards, we have to allocate contiguous buffers > > > for the display framebuffer. So gralloc uses ION to allocate from the > > > CMA heap. However on the db845c, it has no such restrictions, so the > > > CMA heap isn't necessary. > > > > Why do you have a CMA region for the 2nd board if you don't need it? > > _That_ sounds like some serious memory waster, not a few lines of code > > loaded for nothing :-) > > ??? That's not what I said above. If the db845c doesn't need CMA it > won't have a CMA region. > > The issue at hand is that we may want to avoid loading the dmabuf CMA > heap driver on a board that doesn't use CMA. > > > > > With Android's GKI effort, there needs to be one kernel that works on > > > all the devices, and they are using modules to try to minimize the > > > amount of memory spent on functionality that isn't universally needed. > > > So on devices that don't need the CMA heap, they'd probably prefer not > > > to load the CMA dmabuf heap driver, so it would be best if it could be > > > built as a module. If we want to build the CMA heap as a module, the > > > symbols it uses need to be exported. > > > > Yeah, I guess I'm disagreeing on whether dma-buf heaps are core or not. > > That's fine to dispute. I'm not really in a place to assert one way or > not, but the Android folks have made their ION system and CMA heaps > loadable via a module, so it would seem like having the dmabuf system > and CMA heaps be modular would be useful to properly replace that > usage. > > For instance, the system heap as a module probably doesn't make much > sense, as most boards that want to use the dmabuf heaps interface are > likely to use that as well. CMA is more optional as not all boards > will use that one, so it might make sense to avoid loading it. > > Sandeep: Can you chime in here as to how critical having the system > and cma heaps be modules are? With ion, we are making sure there are *standard* heaps that Android should be able to rely on to exist in all kernels [1]. That list is based on what default heaps ion had out-of-tree. As of today, even from those that ion had, Android vendor independent code only relies on 'system heap' and 'cma/dma heaps' so, can safely ignore the carveout and other ion heaps. system heap is really the one that is realistically 'hardware independent', so that can be in kernel. The cma heaps and their existence is optional, so it will be nice to be able to load them as modules. <snip> - ssp 1. https://android.googlesource.com/platform/system/core/+/refs/heads/master/libion/kernel-headers/linux/ion_4.19.h#28
On Tue, Nov 05, 2019 at 11:47:53AM -0800, John Stultz wrote: > On Tue, Nov 5, 2019 at 11:19 AM Daniel Vetter <daniel@ffwll.ch> wrote: > > On Tue, Nov 5, 2019 at 6:41 PM John Stultz <john.stultz@linaro.org> wrote: > > > On Tue, Nov 5, 2019 at 1:43 AM Daniel Vetter <daniel@ffwll.ch> wrote: > > > > > > > > On Mon, Nov 04, 2019 at 10:57:44AM -0800, John Stultz wrote: > > > > > On Mon, Nov 4, 2019 at 1:58 AM Daniel Vetter <daniel@ffwll.ch> wrote: > > > > > > On Fri, Oct 25, 2019 at 11:48:32PM +0000, John Stultz wrote: > > > > > So even if the heaps are configured via DT (which at the moment there > > > > > is no such binding, so that's not really a valid method yet), there's > > > > > still the question of if the heap is necessary/makes sense on the > > > > > device. And the DT would only control the availability of the heap > > > > > interface, not if the heap driver is loaded or not. > > > > > > > > Hm I thought the cma regions are configured in DT? How does that work if > > > > it's not using DT? > > > > > > So yea, CMA regions are either configured by DT or setup at build time > > > (I think there's a cmdline option to set it up as well). > > > > > > But the CMA regions and the dmabuf cma heap driver are separate > > > things. The latter uses the former. > > > > Huh, I assumed the plan is that whenever there's a cma region, we want > > to instantiate a dma-buf heap for it? Why/when would we not want to do > > that? > > Not quite. Andrew noted that we may not want to expose all CMA regions > via dmabuf heaps, so right now we only expose the default region. I > have follow on patches that I sent out earlier (which requires a > to-be-finalized DT binding) which allows us to specify which other CMA > regions to expose. > > > > > > On the HiKey/HiKey960 boards, we have to allocate contiguous buffers > > > > > for the display framebuffer. So gralloc uses ION to allocate from the > > > > > CMA heap. However on the db845c, it has no such restrictions, so the > > > > > CMA heap isn't necessary. > > > > > > > > Why do you have a CMA region for the 2nd board if you don't need it? > > > > _That_ sounds like some serious memory waster, not a few lines of code > > > > loaded for nothing :-) > > > > > > ??? That's not what I said above. If the db845c doesn't need CMA it > > > won't have a CMA region. > > > > > > The issue at hand is that we may want to avoid loading the dmabuf CMA > > > heap driver on a board that doesn't use CMA. > > > > So the CMA core code is also a module, or how does that work? Not > > No. CMA core isn't available as a module. > > > loading the cma dma-buf heap, but keeping all the other cma code > > around feels slightly silly. Do we have numbers that justify this > > silliness? > > I agree that is maybe not the most critical item on the list, but its > one of many that do add up over time. > > Again, I'll defer to Sandeep or other folks on the Google side to help > with the importance here. Mostly I'm trying to ensure there is > functional parity to ION so we don't give folks any reason they could > object to deprecating it. Parity with ION will definitely be nice. For now, however, even if we achieve that parity with UAPI and think about the cma-heap-as-module bit later, I guess that's ok. The real problem is the need for these heaps to be a module in the first place. I'd much rather have an upstream user to show the need for cache maintenance operations that have been talked about so many times, so we can make them happen for dma-buf-heaps in upstream. None of this has to be a module if that happens :(. The reason for the "modularization" for ion heaps is also the CMOs for Android use cases. Unfortunately we haven't had any luck with proving the need for. John, CMIIW. - ssp