RE: Requested feedback on V4L2 driver design
How will the locking be handled ? I suppose opening a plane and closing a plane will just be a matter of sending one (or several) messages to the VPSS. In that case what would prevent a userspace application from sending a direct message to close a plane that a driver would have opened ? Will the firmware be able to identify if a close message comes from the same source as a previous open message, without any risk of identify spoofing ? That is a good question. Let me have one of the firmware guys answer this. I too am assuming that some other application can't issue a close on a channel if they weren't the one to open it. Puru, can you answer this? Laurent, You are correct in saying that opening of plan will be matter of sending one message but it is slightly different for closing of driver and other commands. As part of open, firmware returns specific identifier i.e. handle which has to be used for all further commands/messages related to specific plan. It means that close will also require this specific identifier i.e. handle and so it will not be possible for other application to close any driver unless specific identifier i.e. handle is also supplied along with close message. This identifier or handles helps in identifying source of open/close or any commands. Regards, Purushotam -Original Message- From: Maupin, Chase Sent: Friday, February 26, 2010 9:05 PM To: Laurent Pinchart Cc: Hans Verkuil; sakari.ai...@maxwell.research.nokia.com; mche...@infradead.org; vpss_driver_des...@list.ti.com - This list is to discuss the VPSS driver design (May contain non-TIers); linux-media@vger.kernel.org; Kamoolkar, Mugdha Subject: RE: Requested feedback on V4L2 driver design Laurent, Responses inline Puru, There is a question for you below. Can you look at it and provide an answer? Sincerely, Chase Maupin Software Applications Catalog DSP Products e-mail: chase.mau...@ti.com phone: (281) 274-3285 For support: Forums - http://community.ti.com/forums/ Wiki - http://wiki.davincidsp.com/ -Original Message- From: Laurent Pinchart [mailto:laurent.pinch...@ideasonboard.com] Sent: Thursday, February 25, 2010 6:32 PM To: Maupin, Chase Cc: Hans Verkuil; sakari.ai...@maxwell.research.nokia.com; mche...@infradead.org; vpss_driver_des...@list.ti.com - This list is to discuss the VPSS driver design (May contain non-TIers); linux- me...@vger.kernel.org; Kamoolkar, Mugdha Subject: Re: Requested feedback on V4L2 driver design Hi Chase, On Friday 12 February 2010 17:46:55 Maupin, Chase wrote: Laurent, First let me thank you for taking time to review this. You're welcome. [snip] -Original Message- From: Laurent Pinchart [mailto:laurent.pinch...@ideasonboard.com] Sent: Thursday, February 11, 2010 7:23 PM On Monday 08 February 2010 16:08:37 Maupin, Chase wrote: [snip] If you have additional questions or need more information please feel free to contact us (we have setup a mailing list at vpss_driver_des...@list.ti.com) so we can answer them. I'll answer here as the instructions provided in the wiki to subscribe to the vpss_driver_design mailing list are incorrect (http://list.ti.com/ isn't accessible, the name has no A record associated to it). I've CC'ed the list in case subscription wouldn't be required to post. The page for subscribing to the list requires a my.TI login which you can setup at https://myportal.ti.com/portal/dt?provider=TIPassLoginSingleContainerlt=m ytij5=2j3=1goto=https%3A%2F%2Fmy.ti.com%3A443%2Fcgi- bin%2Fhome.pl%3FDCMP %3DTIHeaderTracking%26HQS%3DOther%2BOT%2Bhdr_my_ti. However, your reply to the list should be fine without subscribing. Thanks for the information, but http://list.ti.com/ still can't be accessed. The host list.ti.com has no A record, an HTTP connection can't thus be established. I'm not sure what is going on here. Let me file a ticket with our IT group and see if they can fix the problem. 1. Multi-core design OMAP3 was already a dual-core system, OMAP4 (I assume all this is about the OMAP4 processors family) seems to push the concept one step further. With its heterogeneous multi-core design (ARM master CPU and slave DSPs), the OMAP architecture delivers high performances at the cost of higher development time and effort as users need to write software for completely different cores, usually using different toolchains. This is in my opinion a good (or at least acceptable) trade-off between CPU power, development time and power consumption (DSPs being more efficient at signal processing at the cost of a higher development complexity). I'm a bit puzzled, however, by how the VPSS MCU will help improving the situation compared to the OMAP3 design. The VPSS MCU will provide an API that will expose a fixed subset of the hardware capabilities. This is only a guess
RE: Requested feedback on V4L2 driver design
Laurent, Responses inline. Sincerely, Chase Maupin Software Applications Catalog DSP Products e-mail: chase.mau...@ti.com phone: (281) 274-3285 For support: Forums - http://community.ti.com/forums/ Wiki - http://wiki.davincidsp.com/ -Original Message- From: Laurent Pinchart [mailto:laurent.pinch...@ideasonboard.com] Sent: Thursday, February 25, 2010 6:35 PM To: Maupin, Chase Cc: Hans Verkuil; sakari.ai...@maxwell.research.nokia.com; mche...@infradead.org; vpss_driver_des...@list.ti.com - This list is to discuss the VPSS driver design (May contain non-TIers); linux- me...@vger.kernel.org Subject: Re: Requested feedback on V4L2 driver design Hi Chase, On Tuesday 16 February 2010 14:00:11 Maupin, Chase wrote: Laurent, To follow up with some of the comments I made before I got additional clarification about the commands supported by the proxy driver running on the VPSS MCU. The proxy will support all of the commands used by V4L2 as well as those proposed extensions to V4L2 that Hans has mentioned. Basically, the list of commands supported at initial release is not only those required today, but a full set for all the features of the VPSS. In this was as new APIs are added to V4L2 the support for those features will already be supported by the VPSS MCU proxy driver. Thank you for the clarification. A few things are still uncleared to me, as stated in my previous mail (from a few minutes ago). My main question is, if the VPSS API is full-featured and accessible from the master CPU, why do we need a proxy driver in the firmware at all ? The proxy driver is the piece of code in the firmware that is actually exposing the VPSS API to the master CPU. It is responsible for listening for requests from the master CPU and then executing those requests on the VPSS CPU. Without the proxy there is no way to tell the VPSS CPU which functions to execute. As for the license of the firmware this is still being worked. It is currently under TI proprietary license and will be distributed as binary under Technical Software Publicly Available (TSPA) which means it can be obtained by anyone. If you feel that source code is required for the firmware at launch to gain acceptance please let us know and we can start working that issue. I think it would definitely help keeping the Linux driver and the VPSS firmware in sync if the VPSS firmware source was available. The firmware source code could even be distributed along with the Linux driver. Thanks for the input. We'll keep this in mind and see what we can do. By the way, will the firmware be loaded at runtime by the driver, or will it be stored internally in the chip ? The firmware will not be stored internally on the chip and will have to be loaded at runtime. We have not settled on how the loading will be done. Currently we are thinking on loading it from u-boot similar to an FPGA firmware load but it could also be done from the kernel. -- Regards, Laurent Pinchart -- To unsubscribe from this list: send the line unsubscribe linux-media in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
RE: Requested feedback on V4L2 driver design
Laurent, Responses inline Puru, There is a question for you below. Can you look at it and provide an answer? Sincerely, Chase Maupin Software Applications Catalog DSP Products e-mail: chase.mau...@ti.com phone: (281) 274-3285 For support: Forums - http://community.ti.com/forums/ Wiki - http://wiki.davincidsp.com/ -Original Message- From: Laurent Pinchart [mailto:laurent.pinch...@ideasonboard.com] Sent: Thursday, February 25, 2010 6:32 PM To: Maupin, Chase Cc: Hans Verkuil; sakari.ai...@maxwell.research.nokia.com; mche...@infradead.org; vpss_driver_des...@list.ti.com - This list is to discuss the VPSS driver design (May contain non-TIers); linux- me...@vger.kernel.org; Kamoolkar, Mugdha Subject: Re: Requested feedback on V4L2 driver design Hi Chase, On Friday 12 February 2010 17:46:55 Maupin, Chase wrote: Laurent, First let me thank you for taking time to review this. You're welcome. [snip] -Original Message- From: Laurent Pinchart [mailto:laurent.pinch...@ideasonboard.com] Sent: Thursday, February 11, 2010 7:23 PM On Monday 08 February 2010 16:08:37 Maupin, Chase wrote: [snip] If you have additional questions or need more information please feel free to contact us (we have setup a mailing list at vpss_driver_des...@list.ti.com) so we can answer them. I'll answer here as the instructions provided in the wiki to subscribe to the vpss_driver_design mailing list are incorrect (http://list.ti.com/ isn't accessible, the name has no A record associated to it). I've CC'ed the list in case subscription wouldn't be required to post. The page for subscribing to the list requires a my.TI login which you can setup at https://myportal.ti.com/portal/dt?provider=TIPassLoginSingleContainerlt=m ytij5=2j3=1goto=https%3A%2F%2Fmy.ti.com%3A443%2Fcgi- bin%2Fhome.pl%3FDCMP %3DTIHeaderTracking%26HQS%3DOther%2BOT%2Bhdr_my_ti. However, your reply to the list should be fine without subscribing. Thanks for the information, but http://list.ti.com/ still can't be accessed. The host list.ti.com has no A record, an HTTP connection can't thus be established. I'm not sure what is going on here. Let me file a ticket with our IT group and see if they can fix the problem. 1. Multi-core design OMAP3 was already a dual-core system, OMAP4 (I assume all this is about the OMAP4 processors family) seems to push the concept one step further. With its heterogeneous multi-core design (ARM master CPU and slave DSPs), the OMAP architecture delivers high performances at the cost of higher development time and effort as users need to write software for completely different cores, usually using different toolchains. This is in my opinion a good (or at least acceptable) trade-off between CPU power, development time and power consumption (DSPs being more efficient at signal processing at the cost of a higher development complexity). I'm a bit puzzled, however, by how the VPSS MCU will help improving the situation compared to the OMAP3 design. The VPSS MCU will provide an API that will expose a fixed subset of the hardware capabilities. This is only a guess, but I suppose the firmware will be fairly generic, and that TI will provide customized versions to big customers tailored for their needs and use cases. The official kernel drivers will then need to be changed, and those changes will have no chance to be accepted in the mainline kernel. This will lead to forks and fragmentation of the developers base among the big players in the embedded markets. What will be the compensation for that ? How will the VPSS MCU provide higher performances than the OMAP3 model ? The firmware on the VPSS MCU will be able to configure/control all of the functionality that the VPSS MCU has and will be the same for all customers. The only part that may change is the proxy driver of the firmware. The proxy driver is the piece that will be responsible for taking the commands from the driver and telling the firmware to execute the operation. As the proxy is the tip of the firmware iceberg, it will be all the Linux driver will care about. Whether the firmware backend is able to configure and control all of the functionality that the VPSS MCU offers is then irrelevant, as the Linux driver will have no way to access that backend directly. Is my understanding correct ? As you saw in my follow-up e-mail the proxy will support all the VPSS APIs so my statement was misleading. The initial version of the proxy will support all the standard V4L2 operations. As new operations (such as on the fly video scaling) are added to the V4L2 API the firmware may require an update to the proxy driver to handle these requests, but the underlying code will remain the same. Different proxy versions will need different version of the Linux
RE: Requested feedback on V4L2 driver design
Laurent, Some additional information from the syslink team. Yes, the current Notify code is indeed going over the kernel mailbox code. You can find this code in drivers/dsp/syslink/notify_ducatidriver/notify_ducati.c. We have already pushed some mailbox patches specific to the new mailbox register set in OMAP4/Netra to kernel. These patches are available starting in 2.6.33. On our tree, we currently have some custom patches some of which were pushed to upstream and in the process of getting rolled into the upcoming 2.6.34 releases. Sincerely, Chase Maupin Software Applications Catalog DSP Products e-mail: chase.mau...@ti.com phone: (281) 274-3285 For support: Forums - http://community.ti.com/forums/ Wiki - http://wiki.davincidsp.com/ -Original Message- From: Kamoolkar, Mugdha Sent: Wednesday, February 24, 2010 9:05 PM To: Maupin, Chase; Laurent Pinchart Cc: Hans Verkuil; sakari.ai...@maxwell.research.nokia.com; mche...@infradead.org; vpss_driver_des...@list.ti.com - This list is to discuss the VPSS driver design (May contain non-TIers); linux- me...@vger.kernel.org; Kanigeri, Hari; Shah, Bhavin; Anna, Suman Subject: RE: Requested feedback on V4L2 driver design Chase, Laurent, Sorry for the extreme delay in my response ... From the code available currently on omapzoom, our plans are to eventually have only the Notify module in kernel-space. All the other code in multicore_ipc will actually move to user-side. The Notify module gives additional functionality over the basic mailbox driver to abstract the single physical event into multiple logical events. This enables multiple clients (one of which is the DSS driver) to use the single physical interrupt for multiple different purposes in a fully modular manner. We will ensure that the kernel-side Notify module is fully integrated into the kernel in the proper way and still meets our functionality requirements, taking feedback from the community into account. We are also making several changes in the APIs for all modules to make them much easier to use. A lot of the complexity as seen by the user will vanish underneath. This is still under progress, so it's not out on omapzoom yet, but will definitely be done. As soon as this is done, we will work on moving most of the modules (except Notify) fully from kernel-user space. Once our kernel-user work has at least gone far enough ahead to allow us to make a design proposal, we will push it out for review to get your valuable feedback. I have also looped in the TI engineers who have worked on and pushed out the omapzoom SysLink code. Regards, Mugdha -Original Message- From: Maupin, Chase Sent: Friday, February 12, 2010 10:17 PM To: Laurent Pinchart Cc: Hans Verkuil; sakari.ai...@maxwell.research.nokia.com; mche...@infradead.org; vpss_driver_des...@list.ti.com - This list is to discuss the VPSS driver design (May contain non-TIers); linux- me...@vger.kernel.org; Kamoolkar, Mugdha Subject: RE: Requested feedback on V4L2 driver design Laurent, First let me thank you for taking time to review this. I have made comments below to address your concerns. Sincerely, Chase Maupin Software Applications Catalog DSP Products e-mail: chase.mau...@ti.com phone: (281) 274-3285 For support: Forums - http://community.ti.com/forums/ Wiki - http://wiki.davincidsp.com/ -Original Message- From: Laurent Pinchart [mailto:laurent.pinch...@ideasonboard.com] Sent: Thursday, February 11, 2010 7:23 PM To: Maupin, Chase Cc: Hans Verkuil; sakari.ai...@maxwell.research.nokia.com; mche...@infradead.org; vpss_driver_des...@list.ti.com - This list is to discuss the VPSS driver design (May contain non-TIers); linux- me...@vger.kernel.org Subject: Re: Requested feedback on V4L2 driver design Hi Chase, On Monday 08 February 2010 16:08:37 Maupin, Chase wrote: All, Texas Instruments (TI) is working on the design for the V4L2 capture and display drivers for our next generation system-on-chip (SoC) processor and would like to solicit your feedback. Thank you very much for requesting feedback on the system design. I personally appreciate this, and I'm pretty sure that the feeling is shared by most of the Linux kernel developers. If you have additional questions or need more information please feel free to contact us (we have setup a mailing list at vpss_driver_des...@list.ti.com) so we can answer them. I'll answer here as the instructions provided in the wiki to subscribe to the vpss_driver_design mailing list are incorrect (http://list.ti.com/ isn't accessible, the name has no A record associated to it). I've CC'ed the list in case subscription wouldn't be required to post. The page for subscribing to the list requires a my.TI login which you can setup at https://myportal.ti.com/portal/dt?provider=TIPassLoginSingleContainerlt=m ytij5=2j3=1goto=https%3A%2F%2Fmy.ti.com
Re: Requested feedback on V4L2 driver design
Hi Chase, On Friday 12 February 2010 17:46:55 Maupin, Chase wrote: Laurent, First let me thank you for taking time to review this. You're welcome. [snip] -Original Message- From: Laurent Pinchart [mailto:laurent.pinch...@ideasonboard.com] Sent: Thursday, February 11, 2010 7:23 PM On Monday 08 February 2010 16:08:37 Maupin, Chase wrote: [snip] If you have additional questions or need more information please feel free to contact us (we have setup a mailing list at vpss_driver_des...@list.ti.com) so we can answer them. I'll answer here as the instructions provided in the wiki to subscribe to the vpss_driver_design mailing list are incorrect (http://list.ti.com/ isn't accessible, the name has no A record associated to it). I've CC'ed the list in case subscription wouldn't be required to post. The page for subscribing to the list requires a my.TI login which you can setup at https://myportal.ti.com/portal/dt?provider=TIPassLoginSingleContainerlt=m ytij5=2j3=1goto=https%3A%2F%2Fmy.ti.com%3A443%2Fcgi-bin%2Fhome.pl%3FDCMP %3DTIHeaderTracking%26HQS%3DOther%2BOT%2Bhdr_my_ti. However, your reply to the list should be fine without subscribing. Thanks for the information, but http://list.ti.com/ still can't be accessed. The host list.ti.com has no A record, an HTTP connection can't thus be established. 1. Multi-core design OMAP3 was already a dual-core system, OMAP4 (I assume all this is about the OMAP4 processors family) seems to push the concept one step further. With its heterogeneous multi-core design (ARM master CPU and slave DSPs), the OMAP architecture delivers high performances at the cost of higher development time and effort as users need to write software for completely different cores, usually using different toolchains. This is in my opinion a good (or at least acceptable) trade-off between CPU power, development time and power consumption (DSPs being more efficient at signal processing at the cost of a higher development complexity). I'm a bit puzzled, however, by how the VPSS MCU will help improving the situation compared to the OMAP3 design. The VPSS MCU will provide an API that will expose a fixed subset of the hardware capabilities. This is only a guess, but I suppose the firmware will be fairly generic, and that TI will provide customized versions to big customers tailored for their needs and use cases. The official kernel drivers will then need to be changed, and those changes will have no chance to be accepted in the mainline kernel. This will lead to forks and fragmentation of the developers base among the big players in the embedded markets. What will be the compensation for that ? How will the VPSS MCU provide higher performances than the OMAP3 model ? The firmware on the VPSS MCU will be able to configure/control all of the functionality that the VPSS MCU has and will be the same for all customers. The only part that may change is the proxy driver of the firmware. The proxy driver is the piece that will be responsible for taking the commands from the driver and telling the firmware to execute the operation. As the proxy is the tip of the firmware iceberg, it will be all the Linux driver will care about. Whether the firmware backend is able to configure and control all of the functionality that the VPSS MCU offers is then irrelevant, as the Linux driver will have no way to access that backend directly. Is my understanding correct ? The initial version of the proxy will support all the standard V4L2 operations. As new operations (such as on the fly video scaling) are added to the V4L2 API the firmware may require an update to the proxy driver to handle these requests, but the underlying code will remain the same. Different proxy versions will need different version of the Linux driver. That's where fragmentation of the developers base come into play, and that's the part that worries me. For customers who wish to use features of the VPSS that are not supported by the current V4L2 APIs there are OpenMax components being developed that can also talk to the VPSS and support the full set of features of the VPSS. These components allow for additional use cases such as transferring data directly from other processing blocks such as the DSP to the VPSS without ever returning to the host processor (tunneling). However, the OpenMax API does not integrate with most existing software such as applications that use V4L2 drivers for video capture and display. So the firmware backend will expose all its features through OpenMax, and the proxy will expose a V4L2-like API based on the firmware backend ? In that case why do we need a proxy at all, why can't the Linux driver access the firmware backend directly and configure it the way it sees fit ? My understanding is that both the backend (for OpenMax) and the proxy will be accessed through
Re: Requested feedback on V4L2 driver design
Hi Chase, On Tuesday 16 February 2010 14:00:11 Maupin, Chase wrote: Laurent, To follow up with some of the comments I made before I got additional clarification about the commands supported by the proxy driver running on the VPSS MCU. The proxy will support all of the commands used by V4L2 as well as those proposed extensions to V4L2 that Hans has mentioned. Basically, the list of commands supported at initial release is not only those required today, but a full set for all the features of the VPSS. In this was as new APIs are added to V4L2 the support for those features will already be supported by the VPSS MCU proxy driver. Thank you for the clarification. A few things are still uncleared to me, as stated in my previous mail (from a few minutes ago). My main question is, if the VPSS API is full-featured and accessible from the master CPU, why do we need a proxy driver in the firmware at all ? As for the license of the firmware this is still being worked. It is currently under TI proprietary license and will be distributed as binary under Technical Software Publicly Available (TSPA) which means it can be obtained by anyone. If you feel that source code is required for the firmware at launch to gain acceptance please let us know and we can start working that issue. I think it would definitely help keeping the Linux driver and the VPSS firmware in sync if the VPSS firmware source was available. The firmware source code could even be distributed along with the Linux driver. By the way, will the firmware be loaded at runtime by the driver, or will it be stored internally in the chip ? -- Regards, Laurent Pinchart -- To unsubscribe from this list: send the line unsubscribe linux-media in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
RE: Requested feedback on V4L2 driver design
Hans, Some follow-up from the syslink team about the driver code in the git tree. The only code to be referred on omapzoom that will actually be in the kernel is the Notify module. All the other code in multicore_ipc will actually move to user-side. The Notify module gives additional functionality over the basic mailbox driver to abstract the single physical event into multiple logical events. This enables multiple clients (one of which is the DSS driver) to use the single physical interrupt for multiple different purposes in a fully modular manner. Sincerely, Chase Maupin Software Applications Catalog DSP Products e-mail: chase.mau...@ti.com phone: (281) 274-3285 For support: Forums - http://community.ti.com/forums/ Wiki - http://wiki.davincidsp.com/ -Original Message- From: Hans Verkuil [mailto:hverk...@xs4all.nl] Sent: Tuesday, February 09, 2010 1:52 AM To: Mauro Carvalho Chehab; laurent.pinch...@ideasonboard.com Cc: Maupin, Chase; sakari.ai...@maxwell.research.nokia.com; vpss_driver_des...@list.ti.com - This list is to discuss the VPSS driver design (May contain non-TIers); linux-media@vger.kernel.org Subject: Re: Requested feedback on V4L2 driver design On Monday 08 February 2010 21:23:00 Mauro Carvalho Chehab wrote: Maupin, Chase wrote: All, Texas Instruments (TI) is working on the design for the V4L2 capture and display drivers for our next generation system-on-chip (SoC) processor and would like to solicit your feedback. Our new SoCs have been improved to allow for higher video resolutions and greater frame rates. To this end the display hardware has been moved to a separate processing block called the video processing subsystem (VPSS). The VPSS will be running a firmware image that controls the capture/display hardware and services requests from one or more host processors. Moving to a remote processor for the processing of video input and output data requires that commands to control the hardware be passed to this processing block using some form of inter-processor communication (IPC). TI would like to solicit your feedback on proposal for the V4L2 driver design to get a feel for whether or not this design would be accepted into the Linux kernel. To this end we have put together an overview of the design and usage on our wiki at http://wiki.davincidsp.com/index.php/Video_Processing_Subsystem_Driver_Des ign. We would greatly appreciate feedback from community members on the acceptability of our driver design. If you have additional questions or need more information please feel free to contact us (we have setup a mailing list at vpss_driver_des...@list.ti.com) so we can answer them. Hi Chase, I'm not sure if I got all the details on your proposal, so let me try to give my understanding. First of all, for normal usage (e.g. capturing a stream or sending an stream to an output device), the driver should work with only the standard V4L2 API. I'm assuming that the driver will provide this capability. I understand that, being a SoC hardware, there are much more that can be done than just doing the normal stream capture/output, already supported by V4L2 API. For such advanced usages, we're open to a proposal to enhance the existing API to support the needs. There are some important aspects that need to be considered when designing any Linux userspace API's: The full functionality of this device can be handled by the proposals made during last year's LPC and that are currently being implemented/prototyped for omap3. That's no coincidence, by the way :-) 1) kernel-userspace API's are forever. So, they need to be designed in a way that new technology changes won't break the old API; 2) API's are meant to be generic. So, they needed to be designed in a way that, if another hardware with similar features require an API, the planned one should fit; 3) The API's should be, as much as possible, independent of the hardware architecture. You'll see that even low-level architecture dependent stuff, like bus drivers are designed in a way that they are not bound to a particular hardware, but instead provide the same common methods to interact with the hardware to other device drivers. That's said, it would be interesting if you could give us a more deep detail on what kind of functionalities and how do you think you'll be implementing them. For me the core issue will be the communication between the main ARM and the ARM controlling the VPSS. Looking at the syslink part of the git tree it all looks way overengineered to me. In particular the multicore_ipc directory. Is all that code involved in setting up the communication path between the main and VPSS ARM? Is there some more detailed document describing how the syslink code works? What I would expect to see is standard mailbox functionality that is used in other places
Re: Requested feedback on V4L2 driver design
On Wednesday 24 February 2010 15:34:19 Maupin, Chase wrote: Hans, Some follow-up from the syslink team about the driver code in the git tree. The only code to be referred on omapzoom that will actually be in the kernel is the Notify module. All the other code in multicore_ipc will actually move to user-side. The Notify module gives additional functionality over the basic mailbox driver to abstract the single physical event into multiple logical events. This enables multiple clients (one of which is the DSS driver) to use the single physical interrupt for multiple different purposes in a fully modular manner. Hi Chase, That is good news. Will it also switch to the standard linux mailbox code? I saw a patch on the omap mailinglist recently that replaced the DSPBRIDGE mailbox code with the kernel mailbox code. I'm not sure whether this is applicable to the Notify code as well, but if it is, then that seems a sensible move. Regards, Hans Sincerely, Chase Maupin Software Applications Catalog DSP Products e-mail: chase.mau...@ti.com phone: (281) 274-3285 For support: Forums - http://community.ti.com/forums/ Wiki - http://wiki.davincidsp.com/ -Original Message- From: Hans Verkuil [mailto:hverk...@xs4all.nl] Sent: Tuesday, February 09, 2010 1:52 AM To: Mauro Carvalho Chehab; laurent.pinch...@ideasonboard.com Cc: Maupin, Chase; sakari.ai...@maxwell.research.nokia.com; vpss_driver_des...@list.ti.com - This list is to discuss the VPSS driver design (May contain non-TIers); linux-media@vger.kernel.org Subject: Re: Requested feedback on V4L2 driver design On Monday 08 February 2010 21:23:00 Mauro Carvalho Chehab wrote: Maupin, Chase wrote: All, Texas Instruments (TI) is working on the design for the V4L2 capture and display drivers for our next generation system-on-chip (SoC) processor and would like to solicit your feedback. Our new SoCs have been improved to allow for higher video resolutions and greater frame rates. To this end the display hardware has been moved to a separate processing block called the video processing subsystem (VPSS). The VPSS will be running a firmware image that controls the capture/display hardware and services requests from one or more host processors. Moving to a remote processor for the processing of video input and output data requires that commands to control the hardware be passed to this processing block using some form of inter-processor communication (IPC). TI would like to solicit your feedback on proposal for the V4L2 driver design to get a feel for whether or not this design would be accepted into the Linux kernel. To this end we have put together an overview of the design and usage on our wiki at http://wiki.davincidsp.com/index.php/Video_Processing_Subsystem_Driver_Des ign. We would greatly appreciate feedback from community members on the acceptability of our driver design. If you have additional questions or need more information please feel free to contact us (we have setup a mailing list at vpss_driver_des...@list.ti.com) so we can answer them. Hi Chase, I'm not sure if I got all the details on your proposal, so let me try to give my understanding. First of all, for normal usage (e.g. capturing a stream or sending an stream to an output device), the driver should work with only the standard V4L2 API. I'm assuming that the driver will provide this capability. I understand that, being a SoC hardware, there are much more that can be done than just doing the normal stream capture/output, already supported by V4L2 API. For such advanced usages, we're open to a proposal to enhance the existing API to support the needs. There are some important aspects that need to be considered when designing any Linux userspace API's: The full functionality of this device can be handled by the proposals made during last year's LPC and that are currently being implemented/prototyped for omap3. That's no coincidence, by the way :-) 1) kernel-userspace API's are forever. So, they need to be designed in a way that new technology changes won't break the old API; 2) API's are meant to be generic. So, they needed to be designed in a way that, if another hardware with similar features require an API, the planned one should fit; 3) The API's should be, as much as possible, independent of the hardware architecture. You'll see that even low-level architecture dependent stuff, like bus drivers are designed in a way that they are not bound to a particular hardware, but instead provide the same common methods to interact with the hardware to other device drivers. That's said, it would be interesting if you could give us a more deep detail on what kind of functionalities and how do you think you'll be implementing
RE: Requested feedback on V4L2 driver design
Chase, Laurent, Sorry for the extreme delay in my response ... From the code available currently on omapzoom, our plans are to eventually have only the Notify module in kernel-space. All the other code in multicore_ipc will actually move to user-side. The Notify module gives additional functionality over the basic mailbox driver to abstract the single physical event into multiple logical events. This enables multiple clients (one of which is the DSS driver) to use the single physical interrupt for multiple different purposes in a fully modular manner. We will ensure that the kernel-side Notify module is fully integrated into the kernel in the proper way and still meets our functionality requirements, taking feedback from the community into account. We are also making several changes in the APIs for all modules to make them much easier to use. A lot of the complexity as seen by the user will vanish underneath. This is still under progress, so it's not out on omapzoom yet, but will definitely be done. As soon as this is done, we will work on moving most of the modules (except Notify) fully from kernel-user space. Once our kernel-user work has at least gone far enough ahead to allow us to make a design proposal, we will push it out for review to get your valuable feedback. I have also looped in the TI engineers who have worked on and pushed out the omapzoom SysLink code. Regards, Mugdha -Original Message- From: Maupin, Chase Sent: Friday, February 12, 2010 10:17 PM To: Laurent Pinchart Cc: Hans Verkuil; sakari.ai...@maxwell.research.nokia.com; mche...@infradead.org; vpss_driver_des...@list.ti.com - This list is to discuss the VPSS driver design (May contain non-TIers); linux-media@vger.kernel.org; Kamoolkar, Mugdha Subject: RE: Requested feedback on V4L2 driver design Laurent, First let me thank you for taking time to review this. I have made comments below to address your concerns. Sincerely, Chase Maupin Software Applications Catalog DSP Products e-mail: chase.mau...@ti.com phone: (281) 274-3285 For support: Forums - http://community.ti.com/forums/ Wiki - http://wiki.davincidsp.com/ -Original Message- From: Laurent Pinchart [mailto:laurent.pinch...@ideasonboard.com] Sent: Thursday, February 11, 2010 7:23 PM To: Maupin, Chase Cc: Hans Verkuil; sakari.ai...@maxwell.research.nokia.com; mche...@infradead.org; vpss_driver_des...@list.ti.com - This list is to discuss the VPSS driver design (May contain non-TIers); linux- me...@vger.kernel.org Subject: Re: Requested feedback on V4L2 driver design Hi Chase, On Monday 08 February 2010 16:08:37 Maupin, Chase wrote: All, Texas Instruments (TI) is working on the design for the V4L2 capture and display drivers for our next generation system-on-chip (SoC) processor and would like to solicit your feedback. Thank you very much for requesting feedback on the system design. I personally appreciate this, and I'm pretty sure that the feeling is shared by most of the Linux kernel developers. If you have additional questions or need more information please feel free to contact us (we have setup a mailing list at vpss_driver_des...@list.ti.com) so we can answer them. I'll answer here as the instructions provided in the wiki to subscribe to the vpss_driver_design mailing list are incorrect (http://list.ti.com/ isn't accessible, the name has no A record associated to it). I've CC'ed the list in case subscription wouldn't be required to post. The page for subscribing to the list requires a my.TI login which you can setup at https://myportal.ti.com/portal/dt?provider=TIPassLoginSingleContainerlt=mytij5=2j3=1goto=https%3A%2F%2Fmy.ti.com%3A443%2Fcgi-bin%2Fhome.pl%3FDCMP%3DTIHeaderTracking%26HQS%3DOther%2BOT%2Bhdr_my_ti. However, your reply to the list should be fine without subscribing. 1. Multi-core design OMAP3 was already a dual-core system, OMAP4 (I assume all this is about the OMAP4 processors family) seems to push the concept one step further. With its heterogeneous multi-core design (ARM master CPU and slave DSPs), the OMAP architecture delivers high performances at the cost of higher development time and effort as users need to write software for completely different cores, usually using different toolchains. This is in my opinion a good (or at least acceptable) trade-off between CPU power, development time and power consumption (DSPs being more efficient at signal processing at the cost of a higher development complexity). I'm a bit puzzled, however, by how the VPSS MCU will help improving the situation compared to the OMAP3 design. The VPSS MCU will provide an API that will expose a fixed subset of the hardware capabilities. This is only a guess, but I suppose the firmware will be fairly generic, and that TI will provide customized versions to big customers tailored for their needs and use cases. The official kernel drivers
RE: Requested feedback on V4L2 driver design
Laurent, To follow up with some of the comments I made before I got additional clarification about the commands supported by the proxy driver running on the VPSS MCU. The proxy will support all of the commands used by V4L2 as well as those proposed extensions to V4L2 that Hans has mentioned. Basically, the list of commands supported at initial release is not only those required today, but a full set for all the features of the VPSS. In this was as new APIs are added to V4L2 the support for those features will already be supported by the VPSS MCU proxy driver. As for the license of the firmware this is still being worked. It is currently under TI proprietary license and will be distributed as binary under Technical Software Publicly Available (TSPA) which means it can be obtained by anyone. If you feel that source code is required for the firmware at launch to gain acceptance please let us know and we can start working that issue. Sincerely, Chase Maupin Software Applications Catalog DSP Products e-mail: chase.mau...@ti.com phone: (281) 274-3285 For support: Forums - http://community.ti.com/forums/ Wiki - http://wiki.davincidsp.com/ -Original Message- From: Laurent Pinchart [mailto:laurent.pinch...@ideasonboard.com] Sent: Thursday, February 11, 2010 7:23 PM To: Maupin, Chase Cc: Hans Verkuil; sakari.ai...@maxwell.research.nokia.com; mche...@infradead.org; vpss_driver_des...@list.ti.com - This list is to discuss the VPSS driver design (May contain non-TIers); linux- me...@vger.kernel.org Subject: Re: Requested feedback on V4L2 driver design Hi Chase, On Monday 08 February 2010 16:08:37 Maupin, Chase wrote: All, Texas Instruments (TI) is working on the design for the V4L2 capture and display drivers for our next generation system-on-chip (SoC) processor and would like to solicit your feedback. Thank you very much for requesting feedback on the system design. I personally appreciate this, and I'm pretty sure that the feeling is shared by most of the Linux kernel developers. If you have additional questions or need more information please feel free to contact us (we have setup a mailing list at vpss_driver_des...@list.ti.com) so we can answer them. I'll answer here as the instructions provided in the wiki to subscribe to the vpss_driver_design mailing list are incorrect (http://list.ti.com/ isn't accessible, the name has no A record associated to it). I've CC'ed the list in case subscription wouldn't be required to post. 1. Multi-core design OMAP3 was already a dual-core system, OMAP4 (I assume all this is about the OMAP4 processors family) seems to push the concept one step further. With its heterogeneous multi-core design (ARM master CPU and slave DSPs), the OMAP architecture delivers high performances at the cost of higher development time and effort as users need to write software for completely different cores, usually using different toolchains. This is in my opinion a good (or at least acceptable) trade-off between CPU power, development time and power consumption (DSPs being more efficient at signal processing at the cost of a higher development complexity). I'm a bit puzzled, however, by how the VPSS MCU will help improving the situation compared to the OMAP3 design. The VPSS MCU will provide an API that will expose a fixed subset of the hardware capabilities. This is only a guess, but I suppose the firmware will be fairly generic, and that TI will provide customized versions to big customers tailored for their needs and use cases. The official kernel drivers will then need to be changed, and those changes will have no chance to be accepted in the mainline kernel. This will lead to forks and fragmentation of the developers base among the big players in the embedded markets. What will be the compensation for that ? How will the VPSS MCU provide higher performances than the OMAP3 model ? 2. VPSS firmware and API The wiki doesn't state under which license the VPSS MCU firmware will be released, but I suppose it won't be open sourced. The VPSS API, which seems from the information provided in the wiki to mimic the V4L2 API at least for video capture and output, will thus be controlled by TI and pretty much set into stone. This means future extensions to the V4L2 API that will provide more control over the devices to userspace applications will be stuck with access to a limited subset of the hardware capabilities, and users will not be able to use the full potential of the system. This goes in the opposite direction of what the Linux media community is trying to do today. For the past 6 months now we have been working on additions to the V4L2 subsystem to create a complete media framework, targeted at both desktop and embedded use cases. The new APIs that we are developing will let userspace applications discover
RE: Requested feedback on V4L2 driver design
Laurent, First let me thank you for taking time to review this. I have made comments below to address your concerns. Sincerely, Chase Maupin Software Applications Catalog DSP Products e-mail: chase.mau...@ti.com phone: (281) 274-3285 For support: Forums - http://community.ti.com/forums/ Wiki - http://wiki.davincidsp.com/ -Original Message- From: Laurent Pinchart [mailto:laurent.pinch...@ideasonboard.com] Sent: Thursday, February 11, 2010 7:23 PM To: Maupin, Chase Cc: Hans Verkuil; sakari.ai...@maxwell.research.nokia.com; mche...@infradead.org; vpss_driver_des...@list.ti.com - This list is to discuss the VPSS driver design (May contain non-TIers); linux- me...@vger.kernel.org Subject: Re: Requested feedback on V4L2 driver design Hi Chase, On Monday 08 February 2010 16:08:37 Maupin, Chase wrote: All, Texas Instruments (TI) is working on the design for the V4L2 capture and display drivers for our next generation system-on-chip (SoC) processor and would like to solicit your feedback. Thank you very much for requesting feedback on the system design. I personally appreciate this, and I'm pretty sure that the feeling is shared by most of the Linux kernel developers. If you have additional questions or need more information please feel free to contact us (we have setup a mailing list at vpss_driver_des...@list.ti.com) so we can answer them. I'll answer here as the instructions provided in the wiki to subscribe to the vpss_driver_design mailing list are incorrect (http://list.ti.com/ isn't accessible, the name has no A record associated to it). I've CC'ed the list in case subscription wouldn't be required to post. The page for subscribing to the list requires a my.TI login which you can setup at https://myportal.ti.com/portal/dt?provider=TIPassLoginSingleContainerlt=mytij5=2j3=1goto=https%3A%2F%2Fmy.ti.com%3A443%2Fcgi-bin%2Fhome.pl%3FDCMP%3DTIHeaderTracking%26HQS%3DOther%2BOT%2Bhdr_my_ti. However, your reply to the list should be fine without subscribing. 1. Multi-core design OMAP3 was already a dual-core system, OMAP4 (I assume all this is about the OMAP4 processors family) seems to push the concept one step further. With its heterogeneous multi-core design (ARM master CPU and slave DSPs), the OMAP architecture delivers high performances at the cost of higher development time and effort as users need to write software for completely different cores, usually using different toolchains. This is in my opinion a good (or at least acceptable) trade-off between CPU power, development time and power consumption (DSPs being more efficient at signal processing at the cost of a higher development complexity). I'm a bit puzzled, however, by how the VPSS MCU will help improving the situation compared to the OMAP3 design. The VPSS MCU will provide an API that will expose a fixed subset of the hardware capabilities. This is only a guess, but I suppose the firmware will be fairly generic, and that TI will provide customized versions to big customers tailored for their needs and use cases. The official kernel drivers will then need to be changed, and those changes will have no chance to be accepted in the mainline kernel. This will lead to forks and fragmentation of the developers base among the big players in the embedded markets. What will be the compensation for that ? How will the VPSS MCU provide higher performances than the OMAP3 model ? The firmware on the VPSS MCU will be able to configure/control all of the functionality that the VPSS MCU has and will be the same for all customers. The only part that may change is the proxy driver of the firmware. The proxy driver is the piece that will be responsible for taking the commands from the driver and telling the firmware to execute the operation. The initial version of the proxy will support all the standard V4L2 operations. As new operations (such as on the fly video scaling) are added to the V4L2 API the firmware may require an update to the proxy driver to handle these requests, but the underlying code will remain the same. For customers who wish to use features of the VPSS that are not supported by the current V4L2 APIs there are OpenMax components being developed that can also talk to the VPSS and support the full set of features of the VPSS. These components allow for additional use cases such as transferring data directly from other processing blocks such as the DSP to the VPSS without ever returning to the host processor (tunneling). However, the OpenMax API does not integrate with most existing software such as applications that use V4L2 drivers for video capture and display. What this means is that we will not be creating a bunch of one-off drivers for customers who want to use features that are not part of the V4L2 APIs. Instead those customers will be able to use the OpenMax components. The Linux V4L2
RE: Requested feedback on V4L2 driver design
-Original Message- From: Hans Verkuil [mailto:hverk...@xs4all.nl] Sent: Tuesday, February 09, 2010 1:52 AM To: Mauro Carvalho Chehab; laurent.pinch...@ideasonboard.com Cc: Maupin, Chase; sakari.ai...@maxwell.research.nokia.com; vpss_driver_des...@list.ti.com - This list is to discuss the VPSS driver design (May contain non-TIers); linux-media@vger.kernel.org Subject: Re: Requested feedback on V4L2 driver design On Monday 08 February 2010 21:23:00 Mauro Carvalho Chehab wrote: Maupin, Chase wrote: All, Texas Instruments (TI) is working on the design for the V4L2 capture and display drivers for our next generation system-on-chip (SoC) processor and would like to solicit your feedback. Our new SoCs have been improved to allow for higher video resolutions and greater frame rates. To this end the display hardware has been moved to a separate processing block called the video processing subsystem (VPSS). The VPSS will be running a firmware image that controls the capture/display hardware and services requests from one or more host processors. Moving to a remote processor for the processing of video input and output data requires that commands to control the hardware be passed to this processing block using some form of inter-processor communication (IPC). TI would like to solicit your feedback on proposal for the V4L2 driver design to get a feel for whether or not this design would be accepted into the Linux kernel. To this end we have put together an overview of the design and usage on our wiki at http://wiki.davincidsp.com/index.php/Video_Processing_Subsystem_Driver_Des ign. We would greatly appreciate feedback from community members on the acceptability of our driver design. If you have additional questions or need more information please feel free to contact us (we have setup a mailing list at vpss_driver_des...@list.ti.com) so we can answer them. Hi Chase, I'm not sure if I got all the details on your proposal, so let me try to give my understanding. First of all, for normal usage (e.g. capturing a stream or sending an stream to an output device), the driver should work with only the standard V4L2 API. I'm assuming that the driver will provide this capability. I understand that, being a SoC hardware, there are much more that can be done than just doing the normal stream capture/output, already supported by V4L2 API. For such advanced usages, we're open to a proposal to enhance the existing API to support the needs. There are some important aspects that need to be considered when designing any Linux userspace API's: The full functionality of this device can be handled by the proposals made during last year's LPC and that are currently being implemented/prototyped for omap3. That's no coincidence, by the way :-) Our initial goal is to enable the current V4L2 APIs and functionality that exist today and then to continue working to add new features that use new functionality as it becomes available. Of course we will be working with the V4L2 community on these features. 1) kernel-userspace API's are forever. So, they need to be designed in a way that new technology changes won't break the old API; 2) API's are meant to be generic. So, they needed to be designed in a way that, if another hardware with similar features require an API, the planned one should fit; 3) The API's should be, as much as possible, independent of the hardware architecture. You'll see that even low-level architecture dependent stuff, like bus drivers are designed in a way that they are not bound to a particular hardware, but instead provide the same common methods to interact with the hardware to other device drivers. That's said, it would be interesting if you could give us a more deep detail on what kind of functionalities and how do you think you'll be implementing them. For me the core issue will be the communication between the main ARM and the ARM controlling the VPSS. Looking at the syslink part of the git tree it all looks way overengineered to me. In particular the multicore_ipc directory. Is all that code involved in setting up the communication path between the main and VPSS ARM? Is there some more detailed document describing how the syslink code works? Agreed. The most important aspect that we are hoping to get feedback on is using the syslink notify IPC for communication between the cores. In this use case much of the functionality of syslink is un-needed. The focus for this driver will be on using simple IPC where events are registered to send data between the host MCU and the VPSS. I'll get someone from the syslink team to comment here with more details but the basic idea is that the host MCU and the VPSS will have an event configured for handling requests. When the host MCU wants to sent a request to the VPSS it will create
RE: Requested feedback on V4L2 driver design
-Original Message- From: Hans Verkuil [mailto:hverk...@xs4all.nl] Sent: Tuesday, February 09, 2010 1:52 AM To: Mauro Carvalho Chehab; laurent.pinch...@ideasonboard.com Cc: Maupin, Chase; sakari.ai...@maxwell.research.nokia.com; vpss_driver_des...@list.ti.com - This list is to discuss the VPSS driver design (May contain non-TIers); linux-media@vger.kernel.org Subject: Re: Requested feedback on V4L2 driver design On Monday 08 February 2010 21:23:00 Mauro Carvalho Chehab wrote: Maupin, Chase wrote: All, Texas Instruments (TI) is working on the design for the V4L2 capture and display drivers for our next generation system-on-chip (SoC) processor and would like to solicit your feedback. Our new SoCs have been improved to allow for higher video resolutions and greater frame rates. To this end the display hardware has been moved to a separate processing block called the video processing subsystem (VPSS). The VPSS will be running a firmware image that controls the capture/display hardware and services requests from one or more host processors. Moving to a remote processor for the processing of video input and output data requires that commands to control the hardware be passed to this processing block using some form of inter-processor communication (IPC). TI would like to solicit your feedback on proposal for the V4L2 driver design to get a feel for whether or not this design would be accepted into the Linux kernel. To this end we have put together an overview of the design and usage on our wiki at http://wiki.davincidsp.com/index.php/Video_Processing_Subsystem_Driver_Des ign. We would greatly appreciate feedback from community members on the acceptability of our driver design. If you have additional questions or need more information please feel free to contact us (we have setup a mailing list at vpss_driver_des...@list.ti.com) so we can answer them. Hi Chase, I'm not sure if I got all the details on your proposal, so let me try to give my understanding. First of all, for normal usage (e.g. capturing a stream or sending an stream to an output device), the driver should work with only the standard V4L2 API. I'm assuming that the driver will provide this capability. I understand that, being a SoC hardware, there are much more that can be done than just doing the normal stream capture/output, already supported by V4L2 API. For such advanced usages, we're open to a proposal to enhance the existing API to support the needs. There are some important aspects that need to be considered when designing any Linux userspace API's: The full functionality of this device can be handled by the proposals made during last year's LPC and that are currently being implemented/prototyped for omap3. That's no coincidence, by the way :-) 1) kernel-userspace API's are forever. So, they need to be designed in a way that new technology changes won't break the old API; 2) API's are meant to be generic. So, they needed to be designed in a way that, if another hardware with similar features require an API, the planned one should fit; 3) The API's should be, as much as possible, independent of the hardware architecture. You'll see that even low-level architecture dependent stuff, like bus drivers are designed in a way that they are not bound to a particular hardware, but instead provide the same common methods to interact with the hardware to other device drivers. That's said, it would be interesting if you could give us a more deep detail on what kind of functionalities and how do you think you'll be implementing them. For me the core issue will be the communication between the main ARM and the ARM controlling the VPSS. Looking at the syslink part of the git tree it all looks way overengineered to me. In particular the multicore_ipc directory. Is all that code involved in setting up the communication path between the main and VPSS ARM? Is there some more detailed document describing how the syslink code works? I uploaded a preliminary version of the syslink User's Guide to the wiki page in the syslink section. You can find the pdf at http://wiki.davincidsp.com/images/3/30/Sprugo6a.pdf. Section 3.7 covers the Notify functionality and usage. Please note that for these video drivers much of the communication settings would be static (i.e. the even number). This configuration would be done by the driver and would not require the user to set it up. What I would expect to see is standard mailbox functionality that is used in other places as well. I gather that at the bottom there actually seems to be a mailbox involved with syslink, but there also seems to be a lot of layers on top of that. Regards, Hans -- Hans Verkuil - video4linux developer - sponsored by TANDBERG -- To unsubscribe from this list: send the line unsubscribe linux
Re: Requested feedback on V4L2 driver design
Maupin, Chase wrote: All, Texas Instruments (TI) is working on the design for the V4L2 capture and display drivers for our next generation system-on-chip (SoC) processor and would like to solicit your feedback. Our new SoCs have been improved to allow for higher video resolutions and greater frame rates. To this end the display hardware has been moved to a separate processing block called the video processing subsystem (VPSS). The VPSS will be running a firmware image that controls the capture/display hardware and services requests from one or more host processors. Moving to a remote processor for the processing of video input and output data requires that commands to control the hardware be passed to this processing block using some form of inter-processor communication (IPC). TI would like to solicit your feedback on proposal for the V4L2 driver design to get a feel for whether or not this design would be accepted into the Linux kernel. To this end we have put together an overview of the design and usage on our wiki at http://wiki.davincidsp.com/index.php/Video_Processing_Subsystem_Driver_Design. We would greatly appreciate feedback from community members on the acceptability of our driver design. If you have additional questions or need more information please feel free to contact us (we have setup a mailing list at vpss_driver_des...@list.ti.com) so we can answer them. Hi Chase, I'm not sure if I got all the details on your proposal, so let me try to give my understanding. First of all, for normal usage (e.g. capturing a stream or sending an stream to an output device), the driver should work with only the standard V4L2 API. I'm assuming that the driver will provide this capability. I understand that, being a SoC hardware, there are much more that can be done than just doing the normal stream capture/output, already supported by V4L2 API. For such advanced usages, we're open to a proposal to enhance the existing API to support the needs. There are some important aspects that need to be considered when designing any Linux userspace API's: 1) kernel-userspace API's are forever. So, they need to be designed in a way that new technology changes won't break the old API; 2) API's are meant to be generic. So, they needed to be designed in a way that, if another hardware with similar features require an API, the planned one should fit; 3) The API's should be, as much as possible, independent of the hardware architecture. You'll see that even low-level architecture dependent stuff, like bus drivers are designed in a way that they are not bound to a particular hardware, but instead provide the same common methods to interact with the hardware to other device drivers. That's said, it would be interesting if you could give us a more deep detail on what kind of functionalities and how do you think you'll be implementing them. -- Cheers, Mauro -- To unsubscribe from this list: send the line unsubscribe linux-media in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: Requested feedback on V4L2 driver design
On Monday 08 February 2010 21:23:00 Mauro Carvalho Chehab wrote: Maupin, Chase wrote: All, Texas Instruments (TI) is working on the design for the V4L2 capture and display drivers for our next generation system-on-chip (SoC) processor and would like to solicit your feedback. Our new SoCs have been improved to allow for higher video resolutions and greater frame rates. To this end the display hardware has been moved to a separate processing block called the video processing subsystem (VPSS). The VPSS will be running a firmware image that controls the capture/display hardware and services requests from one or more host processors. Moving to a remote processor for the processing of video input and output data requires that commands to control the hardware be passed to this processing block using some form of inter-processor communication (IPC). TI would like to solicit your feedback on proposal for the V4L2 driver design to get a feel for whether or not this design would be accepted into the Linux kernel. To this end we have put together an overview of the design and usage on our wiki at http://wiki.davincidsp.com/index.php/Video_Processing_Subsystem_Driver_Design. We would greatly appreciate feedback from community members on the acceptability of our driver design. If you have additional questions or need more information please feel free to contact us (we have setup a mailing list at vpss_driver_des...@list.ti.com) so we can answer them. Hi Chase, I'm not sure if I got all the details on your proposal, so let me try to give my understanding. First of all, for normal usage (e.g. capturing a stream or sending an stream to an output device), the driver should work with only the standard V4L2 API. I'm assuming that the driver will provide this capability. I understand that, being a SoC hardware, there are much more that can be done than just doing the normal stream capture/output, already supported by V4L2 API. For such advanced usages, we're open to a proposal to enhance the existing API to support the needs. There are some important aspects that need to be considered when designing any Linux userspace API's: The full functionality of this device can be handled by the proposals made during last year's LPC and that are currently being implemented/prototyped for omap3. That's no coincidence, by the way :-) 1) kernel-userspace API's are forever. So, they need to be designed in a way that new technology changes won't break the old API; 2) API's are meant to be generic. So, they needed to be designed in a way that, if another hardware with similar features require an API, the planned one should fit; 3) The API's should be, as much as possible, independent of the hardware architecture. You'll see that even low-level architecture dependent stuff, like bus drivers are designed in a way that they are not bound to a particular hardware, but instead provide the same common methods to interact with the hardware to other device drivers. That's said, it would be interesting if you could give us a more deep detail on what kind of functionalities and how do you think you'll be implementing them. For me the core issue will be the communication between the main ARM and the ARM controlling the VPSS. Looking at the syslink part of the git tree it all looks way overengineered to me. In particular the multicore_ipc directory. Is all that code involved in setting up the communication path between the main and VPSS ARM? Is there some more detailed document describing how the syslink code works? What I would expect to see is standard mailbox functionality that is used in other places as well. I gather that at the bottom there actually seems to be a mailbox involved with syslink, but there also seems to be a lot of layers on top of that. Regards, Hans -- Hans Verkuil - video4linux developer - sponsored by TANDBERG -- To unsubscribe from this list: send the line unsubscribe linux-media in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
