Linaro-mm-sig

linaro-mm-sig@lists.linaro.org

32 participants
5827 discussions

[PATCH v2 0/2] dma-buf: Add support for debugfs

by Sumit Semwal

The patch series adds a much-missed support for debugfs to dma-buf framework. Based on the feedback received on v1 of this patch series, support is also added to allow exporters to provide name-strings that will prove useful while debugging. Some more magic can be added for more advanced debugging, but we'll leave that for the time being. Best regards, ~Sumit. Sumit Semwal (2): dma-buf: replace dma_buf_export() with dma_buf_export_named() dma-buf: Add debugfs support Documentation/dma-buf-sharing.txt | 13 ++- drivers/base/dma-buf.c | 173 ++++++++++++++++++++++++++++++++++++- include/linux/dma-buf.h | 16 +++- 3 files changed, 193 insertions(+), 9 deletions(-) -- 1.7.10.4

12 years, 11 months

[PATCH v10 0/2] Add CMA heap for ION memory manager

by Benjamin Gaignard

The goal of those patches is to allow ION clients (drivers or userland applications) to use Contiguous Memory Allocator (CMA). To get more info about CMA: http://lists.linaro.org/pipermail/linaro-mm-sig/2012-February/001328.html patches version 10: - stop adding private field in ion_heap structure - put ion_heap into struct ion_cma_heap patches version 9: - rebased on Android kernel - make cma heap able to support ION_FLAG_CACHED flag patches version 8: - fix memory leak when release sg_table - remove virt_to_phys from ion_cma_phys patches version 7: - rebased on Android kernel - fix ion Makefile - add ion_cma_map_kernel function - remove CONFIG_CMA compilation flags from ion_heap.c patches version 6: - add private field in ion_platform_heap to pass the device linked with CMA. - rework CMA heap to use private field. - prepare CMA heap for incoming dma_common_get_sgtable function http://lists.linaro.org/pipermail/linaro-mm-sig/2012-June/002109.html - simplify ion-ux500 driver. patches version 5: - port patches on android kernel 3.4 where ION use dmabuf - add ion_cma_heap_map_dma and ion_cma_heap_unmap_dma functions patches version 4: - add ION_HEAP_TYPE_DMA heap type in ion_heap_type enum. - CMA heap is now a "native" ION heap. - add ion_heap_create_full function to keep backward compatibilty. - clean up included files in CMA heap - ux500-ion is using ion_heap_create_full instead of ion_heap_create patches version 3: - add a private field in ion_heap structure instead of expose ion_device structure to all heaps - ion_cma_heap is no more a platform driver - ion_cma_heap use ion_heap private field to store the device pointer and make the link with reserved CMA regions - provide ux500-ion driver and configuration file for snowball board to give an example of how use CMA heaps patches version 2: - fix comments done by Andy Green Benjamin Gaignard (2): gpu: ion: fix ion_platform_data definition gpu: ion: add CMA heap drivers/gpu/ion/Makefile | 1 + drivers/gpu/ion/ion_cma_heap.c | 234 ++++++++++++++++++++++++++++++++++++++++ drivers/gpu/ion/ion_heap.c | 6 ++ drivers/gpu/ion/ion_priv.h | 14 +++ include/linux/ion.h | 5 +- 5 files changed, 259 insertions(+), 1 deletion(-) create mode 100644 drivers/gpu/ion/ion_cma_heap.c -- 1.7.10

12 years, 11 months

[PATCH v9 0/3] Add CMA heap for ION memory manager

by Benjamin Gaignard

The goal of those patches is to allow ION clients (drivers or userland applications) to use Contiguous Memory Allocator (CMA). To get more info about CMA: http://lists.linaro.org/pipermail/linaro-mm-sig/2012-February/001328.html patches version 9: - rebased on Android kernel - make cma heap able to support ION_FLAG_CACHED flag patches version 8: - fix memory leak when release sg_table - remove virt_to_phys from ion_cma_phys patches version 7: - rebased on Android kernel - fix ion Makefile - add ion_cma_map_kernel function - remove CONFIG_CMA compilation flags from ion_heap.c patches version 6: - add private field in ion_platform_heap to pass the device linked with CMA. - rework CMA heap to use private field. - prepare CMA heap for incoming dma_common_get_sgtable function http://lists.linaro.org/pipermail/linaro-mm-sig/2012-June/002109.html - simplify ion-ux500 driver. patches version 5: - port patches on android kernel 3.4 where ION use dmabuf - add ion_cma_heap_map_dma and ion_cma_heap_unmap_dma functions patches version 4: - add ION_HEAP_TYPE_DMA heap type in ion_heap_type enum. - CMA heap is now a "native" ION heap. - add ion_heap_create_full function to keep backward compatibilty. - clean up included files in CMA heap - ux500-ion is using ion_heap_create_full instead of ion_heap_create patches version 3: - add a private field in ion_heap structure instead of expose ion_device structure to all heaps - ion_cma_heap is no more a platform driver - ion_cma_heap use ion_heap private field to store the device pointer and make the link with reserved CMA regions - provide ux500-ion driver and configuration file for snowball board to give an example of how use CMA heaps patches version 2: - fix comments done by Andy Green Benjamin Gaignard (3): gpu: ion: fix ion_platform_data definition gpu: ion: add private field in ion_heap and ion_platform_heap structure gpu: ion: add CMA heap drivers/gpu/ion/Makefile | 1 + drivers/gpu/ion/ion_cma_heap.c | 221 ++++++++++++++++++++++++++++++++++++++++ drivers/gpu/ion/ion_heap.c | 7 ++ drivers/gpu/ion/ion_priv.h | 16 +++ include/linux/ion.h | 5 +- 5 files changed, 249 insertions(+), 1 deletion(-) create mode 100644 drivers/gpu/ion/ion_cma_heap.c -- 1.7.10

12 years, 11 months

[PATCH 0/2] Device Tree support for CMA (Contiguous Memory Allocator)

by Marek Szyprowski

Hello, Here is my initial proposal for device tree integration for Contiguous Memory Allocator. The code is quite straightforward, however I expect that the memory bindings require some discussion. The proposed bindings allows to define contiguous memory regions of specified base address and size. Then, the defined regions can be assigned to the given device(s) by adding a property with a phanle to the defined contiguous memory region. From the device tree perspective that's all. Once the bindings are added, all the memory allocations from dma-mapping subsystem will be served from the defined contiguous memory regions. Contiguous Memory Allocator is a framework, which lets to provide a large contiguous memory buffers for (usually a multimedia) devices. The contiguous memory is reserved during early boot and then shared with kernel, which is allowed to allocate it for movable pages. Then, when device driver requests a contigouous buffer, the framework migrates movable pages out of contiguous region and gives it to the driver. When device driver frees the buffer, it is added to kernel memory pool again. For more information, please refer to commit c64be2bb1c6eb43c838b2c6d57 ("drivers: add Contiguous Memory Allocator") and d484864dd96e1830e76895 (CMA merge commit). Best regards Marek Szyprowski Samsung Poland R&D Center Patch summary: Marek Szyprowski (2): drivers: dma-contiguous: clean source code and prepare for device tree drivers: dma-contiguous: add initialization from device tree Documentation/devicetree/bindings/memory.txt | 101 ++++++++++ arch/arm/boot/dts/skeleton.dtsi | 7 +- drivers/base/dma-contiguous.c | 278 +++++++++++++++++++------- include/asm-generic/dma-contiguous.h | 4 +- include/linux/dma-contiguous.h | 32 ++- 5 files changed, 338 insertions(+), 84 deletions(-) create mode 100644 Documentation/devicetree/bindings/memory.txt -- 1.7.9.5

12 years, 11 months

lists.linaro.org maintenance completed

by IT Services

This is to let you know that the migration of lists.linaro.org has been successfully completed. As per the email I sent on Wednesday, it may take some time for the new address of the server to be seen by your computer. You can check this by trying to connect to the web site: http://lists.linaro.org/ If you are able to connect and you do not get an error, this means you are connecting to the new server and you can send email to the lists. If you experience any problems after the weekend and you find that you still cannot connect to the server, please reply to this email to let us know. Regards Philip IT Services Manager Linaro

12 years, 11 months

Planned maintenance - lists.linaro.org

by IT Services

Hello You are receiving this email because you are subscribed to one or more mailing lists provided by the lists.linaro.org server. IT Services are announcing planned maintenance for this server scheduled for *Friday 15th March 2013, starting at 2pm GMT*. The purpose of the work is to move the service to another server. There will be some disruption during this maintenance. In order to ensure that you do not accidentally try to use the service while it is being moved, the current server will be shut down at 2pm. A further email will be sent on Friday afternoon to confirm that the migration of the service is completed. However, due to the way servers are found, it may take a while before your computer is able to connect to the relocated service. After the old server has been shut down, email sent to any of the lists will be queued, but it is possible that the sending server will still trying to deliver the email to the old server rather than the new one when it is started. It is therefore *strongly* recommended that you do not send any email to an @lists.linaro.org email address until you can connect to the new service, which you will be able to test by trying to use a web browser to connect to http://lists.linaro.org after you receive the email confirming that the migration has been completed. Since the old service will be shut down, if you are able to connect, you can be sure you have connected to the new service. If by Monday you are still unable to connect to the service or you are not able to send email to an @lists.linaro.org email address, please send an email to its(a)linaro.org. Thank you. Regards Philip IT Services Manager Linaro

12 years, 11 months

[PATCH] ARM: DMA-mapping: add missing GFP_DMA flag for atomic buffer allocation

by Marek Szyprowski

Atomic pool should always be allocated from DMA zone if such zone is available in the system to avoid issues caused by limited dma mask of any of the devices used for making an atomic allocation. Reported-by: Krzysztof Halasa <khc(a)pm.waw.pl> Signed-off-by: Marek Szyprowski <m.szyprowski(a)samsung.com> --- arch/arm/mm/dma-mapping.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c index c7e3759..e9db6b4 100644 --- a/arch/arm/mm/dma-mapping.c +++ b/arch/arm/mm/dma-mapping.c @@ -342,6 +342,7 @@ static int __init atomic_pool_init(void) { struct dma_pool *pool = &atomic_pool; pgprot_t prot = pgprot_dmacoherent(pgprot_kernel); + gfp_t gfp = GFP_KERNEL | GFP_DMA; unsigned long nr_pages = pool->size >> PAGE_SHIFT; unsigned long *bitmap; struct page *page; @@ -361,8 +362,8 @@ static int __init atomic_pool_init(void) ptr = __alloc_from_contiguous(NULL, pool->size, prot, &page, atomic_pool_init); else - ptr = __alloc_remap_buffer(NULL, pool->size, GFP_KERNEL, prot, - &page, atomic_pool_init); + ptr = __alloc_remap_buffer(NULL, pool->size, gfp, prot, &page, + atomic_pool_init); if (ptr) { int i; -- 1.7.9.5

12 years, 12 months

[PATCH] gpu: ion: Copy name string in client and heap create functions

by Johan Mossberg

ion_client_create and ion_heap_create assumes the caller will keep the name string valid forever. In my opinion this is not the expected behavior, it also makes it difficult to have dynamically created names. Signed-off-by: Johan Mossberg <johan.mossberg(a)stericsson.com> --- drivers/gpu/ion/ion.c | 11 +++++++---- drivers/gpu/ion/ion_heap.c | 3 ++- drivers/gpu/ion/ion_priv.h | 4 +++- 3 files changed, 12 insertions(+), 6 deletions(-) diff --git a/drivers/gpu/ion/ion.c b/drivers/gpu/ion/ion.c index 8fd61b3..b0ad01b 100644 --- a/drivers/gpu/ion/ion.c +++ b/drivers/gpu/ion/ion.c @@ -57,6 +57,8 @@ struct ion_device { struct dentry *debug_root; }; +#define CLIENT_NAME_LEN 16 + /** * struct ion_client - a process/hw block local address space * @node: node in the tree of all clients @@ -75,7 +77,7 @@ struct ion_client { struct ion_device *dev; struct rb_root handles; struct mutex lock; - const char *name; + char name[CLIENT_NAME_LEN]; struct task_struct *task; pid_t pid; struct dentry *debug_root; @@ -598,14 +600,14 @@ static int ion_debug_client_show(struct seq_file *s, void *unused) names[id] = handle->buffer->heap->name; sizes[id] += handle->buffer->size; } - mutex_unlock(&client->lock); - seq_printf(s, "%16.16s: %16.16s\n", "heap_name", "size_in_bytes"); for (i = 0; i < ION_NUM_HEAP_IDS; i++) { if (!names[i]) continue; seq_printf(s, "%16.16s: %16u\n", names[i], sizes[i]); } + mutex_unlock(&client->lock); + return 0; } @@ -655,7 +657,8 @@ struct ion_client *ion_client_create(struct ion_device *dev, client->dev = dev; client->handles = RB_ROOT; mutex_init(&client->lock); - client->name = name; + strncpy(client->name, name, CLIENT_NAME_LEN); + client->name[CLIENT_NAME_LEN - 1] = '\0'; client->task = task; client->pid = pid; diff --git a/drivers/gpu/ion/ion_heap.c b/drivers/gpu/ion/ion_heap.c index 225ef94..96dd70b 100644 --- a/drivers/gpu/ion/ion_heap.c +++ b/drivers/gpu/ion/ion_heap.c @@ -160,7 +160,8 @@ struct ion_heap *ion_heap_create(struct ion_platform_heap *heap_data) return ERR_PTR(-EINVAL); } - heap->name = heap_data->name; + strncpy(heap->name, heap_data->name, HEAP_NAME_LEN); + heap->name[HEAP_NAME_LEN - 1] = '\0'; heap->id = heap_data->id; return heap; } diff --git a/drivers/gpu/ion/ion_priv.h b/drivers/gpu/ion/ion_priv.h index c116921..3296ba4 100644 --- a/drivers/gpu/ion/ion_priv.h +++ b/drivers/gpu/ion/ion_priv.h @@ -107,6 +107,8 @@ struct ion_heap_ops { struct vm_area_struct *vma); }; +#define HEAP_NAME_LEN 16 + /** * struct ion_heap - represents a heap in the system * @node: rb node to put the heap on the device's tree of heaps @@ -131,7 +133,7 @@ struct ion_heap { enum ion_heap_type type; struct ion_heap_ops *ops; unsigned int id; - const char *name; + char name[HEAP_NAME_LEN]; int (*debug_show)(struct ion_heap *heap, struct seq_file *, void *); }; -- 1.7.4.3

13 years

[RFC][PATCH 00/30] staging: Android sync driver

by John Stultz

As proposed yesterday, here's the Android sync driver patches for staging. I've preserved the commit history, but moved all the changes over to be against the staging directory (instead of drivers/base). The goal of submitting this driver to staging is to try to get more collaberation, as there are some similar efforts going on in the community with dmabuf-fences. My email from yesterday with more details for how I hope this goes is here: http://comments.gmane.org/gmane.linux.kernel/1448420 Erik also provided a nice background on the patch set in his reply yesterday, which I'll quote here: "In Honeycomb where we introduced the Hardware Composer HAL. This is a userspace layer that allows composition acceleration on a per platform basis. Different SoC vendors have implemented this using overlays, 2d blitters, a combinations of both, or other clever/disgusting means. Along with the HWC we consolidated a lot of our camera and media pipeline to allow their input to be fed into the GPU or display(overlay.) In order to exploit parallelism the the graphics pipeline, this introduced lots of implicit synchronization dependancies. After a couple years of working with many different SoC vendors, we found that it was really difficult to communicate our system's expectations of the implicit contract and it was difficult for the SoC vendors to properly implement the implicit contract in each of their IP blocks (display, gpu, camera, video codecs). It was also incredibly difficult to debug when problems/deadlocks arose. In an effort to clean up the situation we decided to create set of simple synchronization primitives and have our compositor (SurfaceFlinger) manage the synchronization contract explicitly. We designed these primitives so that they can be passed across processes (much like ion/dma_buf handles), can be backed by hardware synchronization primitives, and can be combined with other sync dependancies in a heterogeneous manner. We also added enough debugging information to make pinpointing a synchronization deadlock bug easier. There are also OpenGL extensions added (which I believe have been ratified by Khronos) to convert a "native" sync object to a gl fence object and vise versa. So far shipped this system on two products (the Nexus 10 and 4) with two different SoCs (Samsung Exynos5250 and Qualcomm MSM8064.) These two projects were much easier to work out the kinks in the graphics/compositing pipelines. In addition we were able to use the telemetry and tracing features to track down the causes of dropped frames aka "jank." As for the implementation, I started with having the main driver op primitive be a wait() op. I quickly noticed that most of the tricky race condition prone code was ending up in the drivers wait() op. It also made handling asynchronous waits of more than one type of sync_pt difficult to manage. In the end I opted for something roughly like poll() where all the heavy lifting is done at the high level and the drivers only need to implement a simple check function." Anyway, let me know what you think of the patches, and hopefully this is something that could be considered for staging for 3.10 thanks -john Cc: Maarten Lankhorst <maarten.lankhorst(a)canonical.com> Cc: Erik Gilling <konkers(a)android.com> Cc: Daniel Vetter <daniel.vetter(a)ffwll.ch> Cc: Rob Clark <robclark(a)gmail.com> Cc: Sumit Semwal <sumit.semwal(a)linaro.org> Cc: Greg KH <gregkh(a)linuxfoundation.org> Cc: dri-devel(a)lists.freedesktop.org Cc: linaro-mm-sig(a)lists.linaro.org Cc: Android Kernel Team <kernel-team(a)android.com> Erik Gilling (26): staging: sync: Add synchronization framework staging: sw_sync: Add cpu based sync driver staging: sync: Add timestamps to sync_pts staging: sync: Add debugfs support staging: sw_sync: Add debug support staging: sync: Add ioctl to get fence data staging: sw_sync: Add fill_driver_data support staging: sync: Add poll support staging: sync: Allow async waits to be canceled staging: sync: Export sync API symbols staging: sw_sync: Export sw_sync API staging: sync: Reorder sync_fence_release staging: sync: Optimize fence merges staging: sync: Add internal refcounting to fences staging: sync: Add reference counting to timelines staging: sync: Change wait timeout to mirror poll semantics staging: sync: Dump sync state to console on timeout staging: sync: Improve timeout dump messages staging: sync: Dump sync state on fence errors staging: sync: Protect unlocked access to fence status staging: sync: Update new fence status with sync_fence_signal_pt staging: sync: Use proper barriers when waiting indefinitely staging: sync: Refactor sync debug printing staging: sw_sync: Convert to use new value_str debug ops staging: sync: Add tracepoint support staging: sync: Don't log wait timeouts when timeout = 0 Jamie Gennis (1): staging: sync: Fix timeout = 0 wait behavior Rebecca Schultz Zavin (2): staging: sync: Fix error paths staging: sw_sync: Fix error paths Ørjan Eide (1): staging: sync: Fix race condition between merge and signal drivers/staging/android/Kconfig | 27 + drivers/staging/android/Makefile | 2 + drivers/staging/android/sw_sync.c | 263 +++++++++ drivers/staging/android/sw_sync.h | 58 ++ drivers/staging/android/sync.c | 1016 ++++++++++++++++++++++++++++++++++ drivers/staging/android/sync.h | 426 ++++++++++++++ drivers/staging/android/trace/sync.h | 82 +++ 7 files changed, 1874 insertions(+) create mode 100644 drivers/staging/android/sw_sync.c create mode 100644 drivers/staging/android/sw_sync.h create mode 100644 drivers/staging/android/sync.c create mode 100644 drivers/staging/android/sync.h create mode 100644 drivers/staging/android/trace/sync.h -- 1.7.10.4

13 years

Summary of CDF BoF @ELC 2013

by Laurent Pinchart

Hi everybody, Here's a summary of the CDF BoF that took place at the ELC 2013. I'd like to start by thanking all the participants who provided valuable feedback (and those who didn't, but who now know a bit more about CDF and will, I have no doubt about that, contribute in the future :-)). Thank you also to Linus Walleij and Jesse Barker for taking notes during the meeting while I was presenting. And obviously, thank you to Jesse Barker for organizing the BoF. I've tried to be as accurate as possible in this summary, but I might have made mistakes. If you have attended the meeting, please point out any issue, inconsistency, or just points I might have forgotten. ---- As not all attendees were familiar with CDF I started by briefly introducing the problems that prompted me to start working on CDF. CDF started as GPF, the Generic Panel Framework. While working on DT support for a display controller driver I realized that panel control code was located in board file. Moving the code somewhere in drivers/ was thus a prerequisite, but it turned out that no framework existed in the kernel to support that tasks. Several major display controller drivers (TI DSS and Samsung Exynos to name a few) had a platform-specific panel driver framework, but the resulting panel drivers wouldn't be reusable across different display controllers. A need for a new framework became pretty evident to me. After drafting an initial proposal and discussing it with several people online and offline (in Helsinki with Tomi Valkeinen from TI, in Copenhagen at Linaro Connect with Marcus Lorentzon from ST-Ericsson, and in Brussels during a BoF at the FOSDEM) the need to support encoders in addition to panels quickly arose, and GPF turned into CDF. I then pursued with an overview of the latest CDF code and its key concepts. While I was expecting this to be a short overview followed by more in-depth discussions, it turned out to support our discussions for the whole 2 hours meeting. The latest available version at the time of the BoF (posted to the linaro-mm- sig mailing list in reply to the BoF's annoucement) was the "non-quite-v3" version. It incorporated feedback received on v2 but hadn't been properly tested yet. The basic CDF building block is called a display entity, modeled as an instance of struct display_entity. They have sink ports through which they receive video data and/or source ports through which they transmit video data. Entities are chained via their ports to create a display pipeline. >From the outside world entities are interfaced through two sets of abstract operations they must provide: - Control operations are called from "upper layers" (usually to implement userspace requests) to get and set entity parameters (such as the physical size, video modes, operation states, bus parameters, ...). Those operations are implemented at the entity level. Google asked how partial updates were handled, I answered they're not handled yet (this is a key concept behind the CDF RFCs: while I try to make sure all devices can be supported, I first concentrate on hardware features required for the devices I work on). Linus Walleij mentioned he thought that partial updates were becoming out of fashion, but larger display sizes might keep them useful in the future. - Video operations control video streams. They're implemented by entities on their source ports, and are called in the upstream (from a video pipeline point of view) direction. A panel will call video operations of the entity it gets its video stream from (this could be an HDMI transmitter, the display controller directly, ...) to control the video stream it receives. Video operations are split in a set of common operations and sets of display bus specific operations (for DPI, DBI, DSI, ...). Some discussion around ops that might be needed in some cases but not others indicate that the ops structures are not quite finished for all bus types (and/or that some ops might be considered for "promotion" to common). In particular the current DSI implementation is copied from a proposal posted by Tomasz Figa from Samsung. As I have no DSI hardware to test it on I have kept it as-is. Jesse Barker pointed out that to make this fly we willl need to get CDF into a number of implementations, in particular the Samsung Exynos SoCs (needing DSI). Several efforts are ongoing: - Marcus Lorentzon (ST Ericsson, Linaro) is working on porting ST Ericsson code to CDF, and in particular on the DSI interface. - Tomasz Figa (Samsung) has worked on porting the Exynos display controller driver to CDF and provided a DSI implementation. - Tomi Valkeinen (TI) is working on porting the TI DSS driver to CDF (or rather his own version of CDF as a first step, to avoid depending on an ever- moving target right now) independently from Linaro. - Alison Chaiken (Mentor Embedded Software) mentioned that Pengutronix is working on panels support for the Freescale i.MX family. - Linaro can probably also help extending the test coverage to various platforms from its member companies. - Finally, I'm working on CDF support for two display controllers found in Renesas SoCs. One of them support DBI and DPI, the other supports DPI only. However, I can't easily test DBI support, as I don't have access to the necessary hardware. I explained at that point that there is currently no clear agreement on a bus and operations model. The initial CDF proposal created a Linux busses for DBI and DSI (similar to I2C and SPI busses), with access to the control bus implemented through those Linux busses, and access to the video bus implemented through video operations on display entities. Tomi Valkeinen then advocated for getting rid of the DBI and DSI Linux busses and implementing access to both control and video through the display entity operations, while Marcus Lorentzon wanted to implement all those operations at the Linux bus level instead. The best way to arbitrate this will probably to work on several implementations and find out which one works better. SONY Mobile currently supports DSI auto-probing, with plug-n-play detection of DSI panels. The panel ID is first retrieved, and the correct panel driver is then loaded. We will likely need to support a similar model. Another option would be to write a single panel-dcs driver to support all DSI panels that conform with the DSI and DCS standards (although we will very likely need panel-specific quirks in that case). The two options could also coexist. We then moved to how display entities should be handled by KMS drivers and mapped to KMS objects. The KMS model hardcodes the following fixed pipeline CRTC -> encoder -> connector The CRTC is controlled by the display controller driver, and panels can be mapped to KMS connector objects. What goes in-between is more of a gray area, as hardware pipeline can have several encoders chained together. I've presented one possible control flow that could solution the problem by grouping multiple objects into an abstract entity. The right-most entity would be a standalone entity, and every encoder but the left-most one in the chain would hide the entities connected at their output. This results in a "russian dolls" model, where encoders forward control operations to the entities they embed, and forward video operations to the entity at their sink side. This can quickly become very complex, especially when locking and reference counting are added to the model. Furthermore, this solution could only handle linear pipelines, which will likely become a severe limitation in the future, especially on embedded devices (for instance splitting a video stream between two panels at the encoder level is a common use case, or driving a two-inputs panel from two CRTCs). Google asked whether this model tries to address both panels and VGA(/HDMI/...) outputs. From what I've seen so far the only limits come from the hardware engineers (often^H^H^H^H^Hsometimes troubled) minds, all kinds of data streams may appear in practice. As most systems will have one CRTS, one encoder and one panel (or connector), we should probably try to keep the model simple to start with with 1:1 mappings between the KMS CRTC/encoder/connector model and the CDF model. If we try to solve every possible problem right now the complexity will explode and we won't be able to handle it. Getting a simple solution upstream now and refactoring it later (there is no userspace API involved, so no backward compatibility issue) might be the right answer. I have no strong feeling about it, but I certainly want something I can get upstream in a reasonable time frame. Keith Packard bluntly (and totally rightfully) whether CDF is not just duplicating part of the KMS API, and whether we shouldn't instead extend the in-kernel KMS model to handle multiple encoders. One reason that drove the creation of CDF outside of KMS was to support sharing a single driver between multiple subsystems. For instance an HDMI encoder could be connected to the output of a display controller handled by a KMS driver on one board, and to the output of a video processor handled by a V4L2 driver on another board. A panel could also be connected to a display controller handled by a KMS driver on one board, and to a display controller handled by an FBDEV driver on another board. Having a single driver for those encoders or panels is one of the goals of CDF. After publishing the first CDF RFC I realized there was a global consensus in the kernel display community to deprecate FBDEV at some point. Sharing panel drivers between KMS and FBDEV then became a "nice to have, but not important" feature. As V4L2 doesn't handle panels (and shouldn't be extended to do so) only encoders drivers would need to be shared, between KMS and V4L2. It's important to note here that we don't need to share a given encoder between two subsystems at runtime. On a given board the encoder will need to be controlled by KMS or V4L2, but never both at the same time. In the CDF context driver sharing refers to the ability to control a given driver from either the KMS or V4L2 subsystem. The discussion then moved to why V4L2 drivers for devices connected to an encoder couldn't be moved to KMS. All display devices should be handled by KMS, but we still have use cases where V4L2 need to handle video outputs. For instance a system with the following pipeline HDMI con. -> HDMI RX -> Processing Engine -> HDMI TX -> HDMI con. doesn't involve memory buffers in the processing pipeline. This can't be handled by KMS, as KMS cannot reporesent a video pipeline without memory in- between the receiving side and the display side. Hans Verkuil also mentioned that for certain applications one prefers to center the API around frames, and that V4L2 is ideal for instance for video conferencing/telephony. Keith Packard thought we should just extend KMS to handle the V4L2 use cases. V4L2 would then (somehow) plug its infrastructure into KMS. This topic has already been discussed in the past, and I agree that extending the KMS model to support "live sources" for CRTCs will be needed in the near future. This could be the basis of other KMS enhancements to support more complex pipelines. Making KMS and V4L2 cooperate is also desirable on the display side to write the output of the CRTC back to memory. KMS has no write-back feature in the API, V4L2 could come to the rescue there. With this kind of extension it might be possible to handle the display part of memory-less pipelines in KMS, although that might be quite a challenge. There was no clear consensus on whether this was desirable. Furthermore, only two HDMI encoders currently need to be shared (both are only supported out-of-tree at the moment). As we don't expect more than a handful of such use cases in the near future, it might not be worth the hasle to create a complete infrastructure to handle a use case that might disappear if we later move all the display-side drivers to KMS. Another solution mentioned by Hans Verkuil would be to create helper functions to translate V4L2 calls to KMS calls (to be clear, this only covers in-kernel calls to encoders). There was no clear consensus on this topic. We then moved on to the hot-plug (and hot-unplug) issues following a question from Google. Hot-plug is currently not supported. We would need to add hot- plugging notifiers and possibly a couple of other operations. However, the video common operations structure has bind/unbind operations, that can serve as a basis. The hard part in hot-plugging support is actually hot-unplugging, as we need to ensure that devices don't disappear all of a sudden while still in use. This was a design goal of CDF from the start, and any issue there will need to be resolved. Panels shouldn't be handled differently than HDMI connectors, CDF will provide a common hot-plugging model. Keith Packard then explained that DRM and KMS will likely be split in the future. The main link between the DRM and KMS APIs is GEM objects. With the recent addition of dmabuf to the Linux kernel the DRM and KMS APIs could be split and use dmabuf to share buffers. DRM and KMS would then be exposed on two separate device nodes. It would be a good idea to revisit the whole KMS/V4L2 unification discussion when DRM and KMS will be split. We briefly touched the subject of namespaces, and whether CDF should use the KMS namespace (drm_*). There is some resistance on the V4L2 side on having CDF structures be KMS objects. It was then time to wrap up the meeting, and I asked the audience one final question: should we shoehorn complex pipelines into the KMS three-stages model, or should we extend the KMS model? That was unfortunately answered by silence, showing that more thinking is needed. A couple more minutes of offline discussions briefly touched the topics of GPU driver reverse engineering and whether we could, after the KMS/DRM split, set a kernel-side standard for embedded GPU drivers. As interesting as this topic is, CDF will not solve that problem :-) -- Regards, Laurent Pinchart

13 years

Jump to page:

2026

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

Linaro-mm-sig