| Message ID | 20190326091744.11542-1-tzimmermann@suse.de (mailing list archive) |
|---|---|
| Headers | show |
| Series | DRM driver for fbdev devices | expand |
On Tue, Mar 26, 2019 at 10:17:33AM +0100, Thomas Zimmermann wrote: > Hi, > > this RFC patch set implements fbdevdrm, a DRM driver on top of fbdev > drivers. I'd appreciate feedback on the code and the idea in general. > > The fbdev subsystem is considered legacy and will probably be removed at > some point. This would mean the loss of a signifanct number of drivers. > Some of the affected hardware is probably not in use any longer, but some > hardware is still around and provides good(-enough) framebuffers. > > OTOH, userspace programs that want to support a wide range of graphics > hardware have to implement support for both DRM and fbdev interfaces. Such > software would benefit from a single interface. > > The fbdevdrm driver provides a way of running drivers for old and new > hardware from the DRM subsystem and interfaces. > > It's not intended to add new features or drivers to fbdev. Instead fbdevdrm > is supposed to be a template for converting fbdev drivers to DRM. It contains > a number of comments (labeled 'DRM porting note') that explain the required > steps. The license is fairly liberal to allow for combination with existing > fbdev code. > > I tested the current patch set with the following drivers: atyfb, aty128fb, > matroxfb, pm2fb, s3fb, savagefb, sisfb, tdfxfb and tridentfb. I was able to > successfully start with fbcon enabled and then run weston or X. > > Thomas Zimmermann (11): > drm/fbdevdrm: Add driver skeleton > drm/fbdevdrm: Add fbdevdrm device > drm/fbdevdrm: Add memory management > drm/fbdevdrm: Add file operations > drm/fbdevdrm: Add GEM and dumb interfaces > drm/fbdevdrm: Add modesetting infrastructure > drm/fbdevdrm: Add DRM <-> fbdev pixel-format conversion > drm/fbdevdrm: Add mode conversion DRM <-> fbdev > drm/fbdevdrm: Add primary plane > drm/fbdevdrm: Add CRTC > drm/fbdevdrm: Detect and validate display modes Looks surprisingly clean, at least from a quick read. Only big thing I noticed on the implementation side is that you probably want to use the simple display helpers. At least that's a much better fit for simple display hw supported by these fbdev drivers. What I'm not sure at all on is whether this is a good idea. It's a quick way of supporting a few drivers we might need from fbdev. But I fear for long-term ecosystem health it'll be a loss: Much less motivation to port the drivers to be native kms ones, and as a result, much less opportunities to extract helpers to make driver writing for such hw easier for everyone. And the latter is really important, if you look at all the work people have done to improve the modeset helpers. Nowadays we have some examples where the kms native port of an fbdev driver turned out to be 2-3x smaller. But I also see the benefit of making the fbdev->kms transition smoother. For discussion here's a pretty wild idea that we might want to consider: - move the ttm based gem code into a helper library like the cma gem helpers. Only special piece we need to pass it is the fb_info structure, for that we can use the same hooks the cma helpers use to allow drivers to specify the right struct device to use for cma allocations. plus ofc an init function for the memory manager and all that, which drivers can put into their driver private device structure - move the modeset compat code into a helper based on top of the simple display pipe helpers. - for each fbdev driver we care about: 1. create bare-bones kms driver 2. copypaste the entire fbdev driver code into the drm driver 3. hook up the above two helper libraries, remove the register_framebuffer - I think we can still allocate the fb_info and all that for transition 4. transition the driver over. If we have the helpers a bit split up between display and buffer management side that should be a lot more gradual, and with the fbdevdrm midlayer inserted in the middle. - Improve the other helpers we have as needed - there's probably room for a simple ttm version, inspired by all these simple display chips (e.g. ast/cirrus/bochs/mga all solve similar problems like this). - Treat any such drivers as CONFIG_STAGING until they've become fully native, i.e. if no one bothers to convert them, we'll drop them again. The above is kinda similar in spirit to the transitional helpers we've used to upgrade the big drivers from legacy kms to atomic. Thoughts? Cheers, Daniel > > drivers/gpu/drm/Kconfig | 2 + > drivers/gpu/drm/Makefile | 1 + > drivers/gpu/drm/fbdevdrm/Kconfig | 13 + > drivers/gpu/drm/fbdevdrm/Makefile | 11 + > drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.c | 276 ++++++++ > drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.h | 58 ++ > drivers/gpu/drm/fbdevdrm/fbdevdrm_device.c | 96 +++ > drivers/gpu/drm/fbdevdrm/fbdevdrm_device.h | 55 ++ > drivers/gpu/drm/fbdevdrm/fbdevdrm_drv.c | 347 +++++++++ > drivers/gpu/drm/fbdevdrm/fbdevdrm_format.c | 441 ++++++++++++ > drivers/gpu/drm/fbdevdrm/fbdevdrm_format.h | 26 + > drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.c | 195 +++++ > drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.h | 53 ++ > drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.c | 746 ++++++++++++++++++++ > drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.h | 38 + > drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.c | 498 +++++++++++++ > drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.h | 27 + > drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.c | 202 ++++++ > drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.h | 35 + > 19 files changed, 3120 insertions(+) > create mode 100644 drivers/gpu/drm/fbdevdrm/Kconfig > create mode 100644 drivers/gpu/drm/fbdevdrm/Makefile > create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.c > create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.h > create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_device.c > create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_device.h > create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_drv.c > create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_format.c > create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_format.h > create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.c > create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.h > create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.c > create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.h > create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.c > create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.h > create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.c > create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.h > > -- > 2.21.0 >
Hi, first of all, thanks for the detailed feedback. Am 26.03.19 um 15:53 schrieb Daniel Vetter: > On Tue, Mar 26, 2019 at 10:17:33AM +0100, Thomas Zimmermann wrote: >> Hi, >> >> this RFC patch set implements fbdevdrm, a DRM driver on top of fbdev >> drivers. I'd appreciate feedback on the code and the idea in general. >> >> The fbdev subsystem is considered legacy and will probably be removed at >> some point. This would mean the loss of a signifanct number of drivers. >> Some of the affected hardware is probably not in use any longer, but some >> hardware is still around and provides good(-enough) framebuffers. >> >> OTOH, userspace programs that want to support a wide range of graphics >> hardware have to implement support for both DRM and fbdev interfaces. Such >> software would benefit from a single interface. >> >> The fbdevdrm driver provides a way of running drivers for old and new >> hardware from the DRM subsystem and interfaces. >> >> It's not intended to add new features or drivers to fbdev. Instead fbdevdrm >> is supposed to be a template for converting fbdev drivers to DRM. It contains >> a number of comments (labeled 'DRM porting note') that explain the required >> steps. The license is fairly liberal to allow for combination with existing >> fbdev code. >> >> I tested the current patch set with the following drivers: atyfb, aty128fb, >> matroxfb, pm2fb, s3fb, savagefb, sisfb, tdfxfb and tridentfb. I was able to >> successfully start with fbcon enabled and then run weston or X. >> >> Thomas Zimmermann (11): >> drm/fbdevdrm: Add driver skeleton >> drm/fbdevdrm: Add fbdevdrm device >> drm/fbdevdrm: Add memory management >> drm/fbdevdrm: Add file operations >> drm/fbdevdrm: Add GEM and dumb interfaces >> drm/fbdevdrm: Add modesetting infrastructure >> drm/fbdevdrm: Add DRM <-> fbdev pixel-format conversion >> drm/fbdevdrm: Add mode conversion DRM <-> fbdev >> drm/fbdevdrm: Add primary plane >> drm/fbdevdrm: Add CRTC >> drm/fbdevdrm: Detect and validate display modes > > Looks surprisingly clean, at least from a quick read. Only big thing I > noticed on the implementation side is that you probably want to use the > simple display helpers. At least that's a much better fit for simple > display hw supported by these fbdev drivers. I thought about using the simple-DRM helpers, but found that a full DRM driver would be more helpful for porting over fbdev drivers. Unless simple DRM is a hard requirement, I'd prefer to leave it this way. For those devices that only support a single pipeline, the conversion to simple DRM should then be mandatory during the porting process. > What I'm not sure at all on is whether this is a good idea. It's a quick > way of supporting a few drivers we might need from fbdev. But I fear for > long-term ecosystem health it'll be a loss: Much less motivation to port > the drivers to be native kms ones, and as a result, much less > opportunities to extract helpers to make driver writing for such hw easier > for everyone. > > And the latter is really important, if you look at all the work people > have done to improve the modeset helpers. Nowadays we have some examples > where the kms native port of an fbdev driver turned out to be 2-3x > smaller. I don't disagree with any of this. My intention is to simplify the graphics stack *without* loosing support for all the old hardware. I see fbdevdrm as help for porting drivers, or short-term solution for getting old framebuffer to work. > But I also see the benefit of making the fbdev->kms transition smoother. > For discussion here's a pretty wild idea that we might want to consider: > > - move the ttm based gem code into a helper library like the cma gem > helpers. Only special piece we need to pass it is the fb_info structure, > for that we can use the same hooks the cma helpers use to allow drivers > to specify the right struct device to use for cma allocations. plus ofc > an init function for the memory manager and all that, which drivers can > put into their driver private device structure > > - move the modeset compat code into a helper based on top of the simple > display pipe helpers. > > - for each fbdev driver we care about: > 1. create bare-bones kms driver > 2. copypaste the entire fbdev driver code into the drm driver > 3. hook up the above two helper libraries, remove the > register_framebuffer - I think we can still allocate the fb_info and all > that for transition > 4. transition the driver over. If we have the helpers a bit split up > between display and buffer management side that should be a lot more > gradual, and with the fbdevdrm midlayer inserted in the middle. I use the current approach because it does not require modifications to DRM or fbdev. Not copying the fbdev driver code has the advantage that any fix that goes into fbdev is also used by fbdevdrm. OTOH, that has some problems. At least the event-reporting hooks appear to be fragile. You mentioned that you have patches for replacing it. I'd be happy to use something else. Filtering out generic and DRM-based drivers is also not optimal. Instead of adding the porting helpers to the DRM core, I'd suggest to add them to fbdevdrm. Fbdevdrm would have to duplicate some of the fbdev functionality and provide a replacement for register_framebuffer. Porting would then mean to 1) create a new DRM driver by copying fbdevdrm, and 2) copy the fbdev driver code to the new DRM driver and rename the function calls accordingly. At this point the new driver should already work to some extend. 3) Finally, do the refactoring and get it out of staging. If the fbdev subsystem is ever going to be removed, the remaining drivers could be moved under fbdevdrm and have their function calls adapted. If anyone cares about a driver, it'd be available for refactoring; otherwise it'd just sit there doing nothing. It's all self-contained and doesn't pollute the DRM core. > - Improve the other helpers we have as needed - there's probably room for > a simple ttm version, inspired by all these simple display chips (e.g. > ast/cirrus/bochs/mga all solve similar problems like this). Makes sense, but appears to be unrelated. Best regards Thomas > - Treat any such drivers as CONFIG_STAGING until they've become fully > native, i.e. if no one bothers to convert them, we'll drop them again. > > The above is kinda similar in spirit to the transitional helpers we've > used to upgrade the big drivers from legacy kms to atomic. > > Thoughts? > > Cheers, Daniel > >> >> drivers/gpu/drm/Kconfig | 2 + >> drivers/gpu/drm/Makefile | 1 + >> drivers/gpu/drm/fbdevdrm/Kconfig | 13 + >> drivers/gpu/drm/fbdevdrm/Makefile | 11 + >> drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.c | 276 ++++++++ >> drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.h | 58 ++ >> drivers/gpu/drm/fbdevdrm/fbdevdrm_device.c | 96 +++ >> drivers/gpu/drm/fbdevdrm/fbdevdrm_device.h | 55 ++ >> drivers/gpu/drm/fbdevdrm/fbdevdrm_drv.c | 347 +++++++++ >> drivers/gpu/drm/fbdevdrm/fbdevdrm_format.c | 441 ++++++++++++ >> drivers/gpu/drm/fbdevdrm/fbdevdrm_format.h | 26 + >> drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.c | 195 +++++ >> drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.h | 53 ++ >> drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.c | 746 ++++++++++++++++++++ >> drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.h | 38 + >> drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.c | 498 +++++++++++++ >> drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.h | 27 + >> drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.c | 202 ++++++ >> drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.h | 35 + >> 19 files changed, 3120 insertions(+) >> create mode 100644 drivers/gpu/drm/fbdevdrm/Kconfig >> create mode 100644 drivers/gpu/drm/fbdevdrm/Makefile >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.c >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.h >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_device.c >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_device.h >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_drv.c >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_format.c >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_format.h >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.c >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.h >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.c >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.h >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.c >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.h >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.c >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.h >> >> -- >> 2.21.0 >> >
On Wed, Mar 27, 2019 at 10:10 AM Thomas Zimmermann <tzimmermann@suse.de> wrote: > > Hi, > > first of all, thanks for the detailed feedback. > > Am 26.03.19 um 15:53 schrieb Daniel Vetter: > > On Tue, Mar 26, 2019 at 10:17:33AM +0100, Thomas Zimmermann wrote: > >> Hi, > >> > >> this RFC patch set implements fbdevdrm, a DRM driver on top of fbdev > >> drivers. I'd appreciate feedback on the code and the idea in general. > >> > >> The fbdev subsystem is considered legacy and will probably be removed at > >> some point. This would mean the loss of a signifanct number of drivers. > >> Some of the affected hardware is probably not in use any longer, but some > >> hardware is still around and provides good(-enough) framebuffers. > >> > >> OTOH, userspace programs that want to support a wide range of graphics > >> hardware have to implement support for both DRM and fbdev interfaces. Such > >> software would benefit from a single interface. > >> > >> The fbdevdrm driver provides a way of running drivers for old and new > >> hardware from the DRM subsystem and interfaces. > >> > >> It's not intended to add new features or drivers to fbdev. Instead fbdevdrm > >> is supposed to be a template for converting fbdev drivers to DRM. It contains > >> a number of comments (labeled 'DRM porting note') that explain the required > >> steps. The license is fairly liberal to allow for combination with existing > >> fbdev code. > >> > >> I tested the current patch set with the following drivers: atyfb, aty128fb, > >> matroxfb, pm2fb, s3fb, savagefb, sisfb, tdfxfb and tridentfb. I was able to > >> successfully start with fbcon enabled and then run weston or X. > >> > >> Thomas Zimmermann (11): > >> drm/fbdevdrm: Add driver skeleton > >> drm/fbdevdrm: Add fbdevdrm device > >> drm/fbdevdrm: Add memory management > >> drm/fbdevdrm: Add file operations > >> drm/fbdevdrm: Add GEM and dumb interfaces > >> drm/fbdevdrm: Add modesetting infrastructure > >> drm/fbdevdrm: Add DRM <-> fbdev pixel-format conversion > >> drm/fbdevdrm: Add mode conversion DRM <-> fbdev > >> drm/fbdevdrm: Add primary plane > >> drm/fbdevdrm: Add CRTC > >> drm/fbdevdrm: Detect and validate display modes > > > > Looks surprisingly clean, at least from a quick read. Only big thing I > > noticed on the implementation side is that you probably want to use the > > simple display helpers. At least that's a much better fit for simple > > display hw supported by these fbdev drivers. > > I thought about using the simple-DRM helpers, but found that a full DRM > driver would be more helpful for porting over fbdev drivers. Unless > simple DRM is a hard requirement, I'd prefer to leave it this way. > > For those devices that only support a single pipeline, the conversion to > simple DRM should then be mandatory during the porting process. fbdev only supports a single output, all multi-head extensions are driver specific ioctl hacks. Given that all you can do is switch it on or off (with a given mode), simple pipe helpers are the perfect fit for fbdev. Some drivers might support more (like multiple heads, or at least different outputs), and those we should convert over. But at least for step 1 in converting fbdev drivers over, simple pipe is the right starting point I think. > > What I'm not sure at all on is whether this is a good idea. It's a quick > > way of supporting a few drivers we might need from fbdev. But I fear for > > long-term ecosystem health it'll be a loss: Much less motivation to port > > the drivers to be native kms ones, and as a result, much less > > opportunities to extract helpers to make driver writing for such hw easier > > for everyone. > > > > And the latter is really important, if you look at all the work people > > have done to improve the modeset helpers. Nowadays we have some examples > > where the kms native port of an fbdev driver turned out to be 2-3x > > smaller. > > I don't disagree with any of this. My intention is to simplify the > graphics stack *without* loosing support for all the old hardware. I see > fbdevdrm as help for porting drivers, or short-term solution for getting > old framebuffer to work. Simplify for whom? If the goal is to make life easier for userspace I'm not sure - you already need at least legacy kms and atomic kms backends (except for the simplest stuff). And I'm not sure how many lines of code an fbdev backend really is. Definitely won't simplify things for the kernel, since this way we'll never know which fbdev drivers are dead and which ones we need to keep supporting. This isn't the first time we've just dropped a pile of old hw support on the floor. What we should aim for though is to make the transition as easy as possible for those few drivers that aren't ported yet, but still needed. > > But I also see the benefit of making the fbdev->kms transition smoother. > > For discussion here's a pretty wild idea that we might want to consider: > > > > - move the ttm based gem code into a helper library like the cma gem > > helpers. Only special piece we need to pass it is the fb_info structure, > > for that we can use the same hooks the cma helpers use to allow drivers > > to specify the right struct device to use for cma allocations. plus ofc > > an init function for the memory manager and all that, which drivers can > > put into their driver private device structure > > > > - move the modeset compat code into a helper based on top of the simple > > display pipe helpers. > > > > - for each fbdev driver we care about: > > 1. create bare-bones kms driver > > 2. copypaste the entire fbdev driver code into the drm driver > > 3. hook up the above two helper libraries, remove the > > register_framebuffer - I think we can still allocate the fb_info and all > > that for transition > > 4. transition the driver over. If we have the helpers a bit split up > > between display and buffer management side that should be a lot more > > gradual, and with the fbdevdrm midlayer inserted in the middle. > > I use the current approach because it does not require modifications to > DRM or fbdev. Not copying the fbdev driver code has the advantage that > any fix that goes into fbdev is also used by fbdevdrm. > > OTOH, that has some problems. At least the event-reporting hooks appear > to be fragile. You mentioned that you have patches for replacing it. I'd > be happy to use something else. Filtering out generic and DRM-based > drivers is also not optimal. > > Instead of adding the porting helpers to the DRM core, I'd suggest to > add them to fbdevdrm. Fbdevdrm would have to duplicate some of the fbdev > functionality and provide a replacement for register_framebuffer. > > Porting would then mean to > > 1) create a new DRM driver by copying fbdevdrm, and > 2) copy the fbdev driver code to the new DRM driver and rename the > function calls accordingly. At this point the new driver should already > work to some extend. > 3) Finally, do the refactoring and get it out of staging. Maybe the idea behind helpers isn't fully clear: https://blog.ffwll.ch/2016/12/midlayers-once-more-with-feeling.html The idea with helpers is essentially to have all the flexibility of copypasting (you can refactor as you see fit), without having to actually copypaste a lot of the code. That's why I think an fbdevdrm helper suits, at least as long as we're actively transitioning. Worked fairly well for the atomic transition at least. Wrt what you need to provide: I think all that needs to be replaced is (un)register_framebuffer - all other fbdev functions and datastructures we should be able to continue to use, even for such a frankendriver transitioning from fbdev to drm. > If the fbdev subsystem is ever going to be removed, the remaining > drivers could be moved under fbdevdrm and have their function calls > adapted. If anyone cares about a driver, it'd be available for > refactoring; otherwise it'd just sit there doing nothing. It's all > self-contained and doesn't pollute the DRM core. Maybe some confusions: - We're not going to remove fbdev as long as there's users left. So this scenario isn't one to optimize for. The only thing that's not going to happen is adding more features/drivers to fbdev (except through drm drivers and drm's fbdev compat code). - I don't think we need to port all drivers over to drm. Most of them are irrelevant by now, letting them quietly die feels like the best action. One reason for copypasting drivers to drm was to make sure there's someone who actually cares enough about that specific driver. > > - Improve the other helpers we have as needed - there's probably room for > > a simple ttm version, inspired by all these simple display chips (e.g. > > ast/cirrus/bochs/mga all solve similar problems like this). > > Makes sense, but appears to be unrelated. It's the main reason to have drivers in upstream - sharing code and infrastructure. At least for me the main benefit in porting a bunch more fbdev drivers over isn't the old hw support (that hw is quietly dying anyway, we just have to wait). But improving support for new hw that fills similar niches, and making it easier for everyone else. Ofc I'm not going to stop anyone who's going to do all the porting work, but we can't realistically require it (e.g. legacy kms drivers also don't support atomic, since simply not possible without rewriting the driver). Cheers, Daniel > > Best regards > Thomas > > > - Treat any such drivers as CONFIG_STAGING until they've become fully > > native, i.e. if no one bothers to convert them, we'll drop them again. > > > > The above is kinda similar in spirit to the transitional helpers we've > > used to upgrade the big drivers from legacy kms to atomic. > > > > Thoughts? > > > > Cheers, Daniel > > > >> > >> drivers/gpu/drm/Kconfig | 2 + > >> drivers/gpu/drm/Makefile | 1 + > >> drivers/gpu/drm/fbdevdrm/Kconfig | 13 + > >> drivers/gpu/drm/fbdevdrm/Makefile | 11 + > >> drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.c | 276 ++++++++ > >> drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.h | 58 ++ > >> drivers/gpu/drm/fbdevdrm/fbdevdrm_device.c | 96 +++ > >> drivers/gpu/drm/fbdevdrm/fbdevdrm_device.h | 55 ++ > >> drivers/gpu/drm/fbdevdrm/fbdevdrm_drv.c | 347 +++++++++ > >> drivers/gpu/drm/fbdevdrm/fbdevdrm_format.c | 441 ++++++++++++ > >> drivers/gpu/drm/fbdevdrm/fbdevdrm_format.h | 26 + > >> drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.c | 195 +++++ > >> drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.h | 53 ++ > >> drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.c | 746 ++++++++++++++++++++ > >> drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.h | 38 + > >> drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.c | 498 +++++++++++++ > >> drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.h | 27 + > >> drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.c | 202 ++++++ > >> drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.h | 35 + > >> 19 files changed, 3120 insertions(+) > >> create mode 100644 drivers/gpu/drm/fbdevdrm/Kconfig > >> create mode 100644 drivers/gpu/drm/fbdevdrm/Makefile > >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.c > >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.h > >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_device.c > >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_device.h > >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_drv.c > >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_format.c > >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_format.h > >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.c > >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.h > >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.c > >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.h > >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.c > >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.h > >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.c > >> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.h > >> > >> -- > >> 2.21.0 > >> > > > > -- > Thomas Zimmermann > Graphics Driver Developer > SUSE Linux GmbH, Maxfeldstrasse 5, 90409 Nuernberg, Germany > GF: Felix Imendörffer, Mary Higgins, Sri Rasiah > HRB 21284 (AG Nürnberg) >
On 2019-03-27 10:41 a.m., Daniel Vetter wrote: > On Wed, Mar 27, 2019 at 10:10 AM Thomas Zimmermann <tzimmermann@suse.de> wrote: >> Am 26.03.19 um 15:53 schrieb Daniel Vetter: >>> >>> Looks surprisingly clean, at least from a quick read. Only big thing I >>> noticed on the implementation side is that you probably want to use the >>> simple display helpers. At least that's a much better fit for simple >>> display hw supported by these fbdev drivers. >> >> I thought about using the simple-DRM helpers, but found that a full DRM >> driver would be more helpful for porting over fbdev drivers. Unless >> simple DRM is a hard requirement, I'd prefer to leave it this way. >> >> For those devices that only support a single pipeline, the conversion to >> simple DRM should then be mandatory during the porting process. > > fbdev only supports a single output, all multi-head extensions are > driver specific ioctl hacks. Given that all you can do is switch it on > or off (with a given mode), simple pipe helpers are the perfect fit > for fbdev. > > Some drivers might support more (like multiple heads, or at least > different outputs), and those we should convert over. But at least for > step 1 in converting fbdev drivers over, simple pipe is the right > starting point I think. Converting an fbdev driver to a KMS one loses some features: 1) Dynamically switching modes / colour formats via fbdev. 2) fbcon acceleration (or maybe this can be preserved?). In turn, it allows exposing additional functionality: 3) Multiple outputs via KMS. Is there any benefit in converting a driver without adding 3)? If not, is simple pipe still a good starting point?
On Wed, Mar 27, 2019 at 10:55 AM Michel Dänzer <michel@daenzer.net> wrote: > > On 2019-03-27 10:41 a.m., Daniel Vetter wrote: > > On Wed, Mar 27, 2019 at 10:10 AM Thomas Zimmermann <tzimmermann@suse.de> wrote: > >> Am 26.03.19 um 15:53 schrieb Daniel Vetter: > >>> > >>> Looks surprisingly clean, at least from a quick read. Only big thing I > >>> noticed on the implementation side is that you probably want to use the > >>> simple display helpers. At least that's a much better fit for simple > >>> display hw supported by these fbdev drivers. > >> > >> I thought about using the simple-DRM helpers, but found that a full DRM > >> driver would be more helpful for porting over fbdev drivers. Unless > >> simple DRM is a hard requirement, I'd prefer to leave it this way. > >> > >> For those devices that only support a single pipeline, the conversion to > >> simple DRM should then be mandatory during the porting process. > > > > fbdev only supports a single output, all multi-head extensions are > > driver specific ioctl hacks. Given that all you can do is switch it on > > or off (with a given mode), simple pipe helpers are the perfect fit > > for fbdev. > > > > Some drivers might support more (like multiple heads, or at least > > different outputs), and those we should convert over. But at least for > > step 1 in converting fbdev drivers over, simple pipe is the right > > starting point I think. > > Converting an fbdev driver to a KMS one loses some features: > > 1) Dynamically switching modes / colour formats via fbdev. Possible to add that to drm_fb_helper.c, it's just after 10+ years still no one cares, so it didn't happen. We have added a bunch of fbdev emulation features for stuff people wanted, so patches are welcome. And will be merged, if they exist. Ofc will be rather limited since fbdev only has one output. But should be possible to at least faithfully map for the simple case of only 1 connected output. Ofc we still can't reallocate the fbdev bo, because fbdev really doesn't like when the aperture moves around. Fixing that is possible, if you fix the fbdev vs. fbcon locking, which is probably a few man-years of engineering (I looked at it recently for reasons, it's totally broken and full of ABBA issues, solved by simply not taking the other lock). > 2) fbcon acceleration (or maybe this can be preserved?). gma500 has fbcon accel. i915 had patches, but wasn't worth it. It's possible, but demand for this seems approximately nil. We do support fbdev page flipping, through the overallocate parameter. > In turn, it allows exposing additional functionality: > > 3) Multiple outputs via KMS. 4) multi-master and multi-client (at least if we properly manage the vram with ttm for discrete, socs should use the cma helpers). I think that's the big one, and why compositors might want to have kms even for dumb display hw. 5) on socs: prime buffer sharing, that seems to be the main reason we gained a few fbdev conversion in kms recently > Is there any benefit in converting a driver without adding 3)? If not, > is simple pipe still a good starting point? simple pipe still allows multiple outputs if they're entirely separate. Only thing it limits is: 1) only 1 fixed plane per crtc 2) fixed crtc->output routing (but with the option to plug in drm_bridge transcoders into that fixed pipeline for code sharing). I still think simple_pipe is a good starting point for most fbdev drivers left. All the fancy ones that support multiple outputs and output switching do have at least a basic kms driver already. Since atomic helpers are fairly modular (and simple pipe builds on top of atomic helpers) it should be easy to extend from simple pipe to full atomic once necessary. -Daniel
Hi Am 27.03.19 um 10:41 schrieb Daniel Vetter: >> I use the current approach because it does not require modifications to >> DRM or fbdev. Not copying the fbdev driver code has the advantage that >> any fix that goes into fbdev is also used by fbdevdrm. >> >> OTOH, that has some problems. At least the event-reporting hooks appear >> to be fragile. You mentioned that you have patches for replacing it. I'd >> be happy to use something else. Filtering out generic and DRM-based >> drivers is also not optimal. >> >> Instead of adding the porting helpers to the DRM core, I'd suggest to >> add them to fbdevdrm. Fbdevdrm would have to duplicate some of the fbdev >> functionality and provide a replacement for register_framebuffer. >> >> Porting would then mean to >> >> 1) create a new DRM driver by copying fbdevdrm, and >> 2) copy the fbdev driver code to the new DRM driver and rename the >> function calls accordingly. At this point the new driver should already >> work to some extend. >> 3) Finally, do the refactoring and get it out of staging. > > Maybe the idea behind helpers isn't fully clear: > > https://blog.ffwll.ch/2016/12/midlayers-once-more-with-feeling.html > > The idea with helpers is essentially to have all the flexibility of > copypasting (you can refactor as you see fit), without having to > actually copypaste a lot of the code. That's why I think an fbdevdrm > helper suits, at least as long as we're actively transitioning. Worked > fairly well for the atomic transition at least. That's not really a problem from my side, but maybe we misunderstood each other. Modeling fbdevdrm after CMA helpers or simple drm would put it next to central DRM core and helper modules. That seems like an odd place for transitional helper functions that are supposed to be replaced by the actual DRM helpers. I don't want some kind of abstraction layer. I'm proposing a design like tinydrm, where a separate module offers helpers for fbdev drivers that are about to be converted. The difference might not be that significant, but just a question of were the code is located. >>> - Improve the other helpers we have as needed - there's probably room for >>> a simple ttm version, inspired by all these simple display chips (e.g. >>> ast/cirrus/bochs/mga all solve similar problems like this). >> >> Makes sense, but appears to be unrelated. > > It's the main reason to have drivers in upstream - sharing code and > infrastructure. I meant that creating 'simple TTM' is unrelated to fbdevdrm. It would make sense in any case. Best regards Thomas At least for me the main benefit in porting a bunch > more fbdev drivers over isn't the old hw support (that hw is quietly > dying anyway, we just have to wait). But improving support for new hw > that fills similar niches, and making it easier for everyone else. Ofc > I'm not going to stop anyone who's going to do all the porting work, > but we can't realistically require it (e.g. legacy kms drivers also > don't support atomic, since simply not possible without rewriting the > driver). > > Cheers, Daniel >> >> Best regards >> Thomas >> >>> - Treat any such drivers as CONFIG_STAGING until they've become fully >>> native, i.e. if no one bothers to convert them, we'll drop them again. >>> >>> The above is kinda similar in spirit to the transitional helpers we've >>> used to upgrade the big drivers from legacy kms to atomic. >>> >>> Thoughts? >>> >>> Cheers, Daniel >>> >>>> >>>> drivers/gpu/drm/Kconfig | 2 + >>>> drivers/gpu/drm/Makefile | 1 + >>>> drivers/gpu/drm/fbdevdrm/Kconfig | 13 + >>>> drivers/gpu/drm/fbdevdrm/Makefile | 11 + >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.c | 276 ++++++++ >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.h | 58 ++ >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_device.c | 96 +++ >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_device.h | 55 ++ >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_drv.c | 347 +++++++++ >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_format.c | 441 ++++++++++++ >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_format.h | 26 + >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.c | 195 +++++ >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.h | 53 ++ >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.c | 746 ++++++++++++++++++++ >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.h | 38 + >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.c | 498 +++++++++++++ >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.h | 27 + >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.c | 202 ++++++ >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.h | 35 + >>>> 19 files changed, 3120 insertions(+) >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/Kconfig >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/Makefile >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.c >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.h >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_device.c >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_device.h >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_drv.c >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_format.c >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_format.h >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.c >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.h >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.c >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.h >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.c >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.h >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.c >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.h >>>> >>>> -- >>>> 2.21.0 >>>> >>> >> >> -- >> Thomas Zimmermann >> Graphics Driver Developer >> SUSE Linux GmbH, Maxfeldstrasse 5, 90409 Nuernberg, Germany >> GF: Felix Imendörffer, Mary Higgins, Sri Rasiah >> HRB 21284 (AG Nürnberg) >> > >
On Wed, Mar 27, 2019 at 3:46 PM Thomas Zimmermann <tzimmermann@suse.de> wrote: > > Hi > > Am 27.03.19 um 10:41 schrieb Daniel Vetter: > >> I use the current approach because it does not require modifications to > >> DRM or fbdev. Not copying the fbdev driver code has the advantage that > >> any fix that goes into fbdev is also used by fbdevdrm. > >> > >> OTOH, that has some problems. At least the event-reporting hooks appear > >> to be fragile. You mentioned that you have patches for replacing it. I'd > >> be happy to use something else. Filtering out generic and DRM-based > >> drivers is also not optimal. > >> > >> Instead of adding the porting helpers to the DRM core, I'd suggest to > >> add them to fbdevdrm. Fbdevdrm would have to duplicate some of the fbdev > >> functionality and provide a replacement for register_framebuffer. > >> > >> Porting would then mean to > >> > >> 1) create a new DRM driver by copying fbdevdrm, and > >> 2) copy the fbdev driver code to the new DRM driver and rename the > >> function calls accordingly. At this point the new driver should already > >> work to some extend. > >> 3) Finally, do the refactoring and get it out of staging. > > > > Maybe the idea behind helpers isn't fully clear: > > > > https://blog.ffwll.ch/2016/12/midlayers-once-more-with-feeling.html > > > > The idea with helpers is essentially to have all the flexibility of > > copypasting (you can refactor as you see fit), without having to > > actually copypaste a lot of the code. That's why I think an fbdevdrm > > helper suits, at least as long as we're actively transitioning. Worked > > fairly well for the atomic transition at least. > > That's not really a problem from my side, but maybe we misunderstood > each other. > > Modeling fbdevdrm after CMA helpers or simple drm would put it next to > central DRM core and helper modules. That seems like an odd place for > transitional helper functions that are supposed to be replaced by the > actual DRM helpers. That's exactly what we've done with the transitional atomic helpers too. I think that's perfectly fine, as long as we put a solid explanation next to the code what it is all about and how to use it. > I don't want some kind of abstraction layer. I'm proposing a design like > tinydrm, where a separate module offers helpers for fbdev drivers that > are about to be converted. We merged tinydrm because reworking it all into proper helpers took a bit longer than what's reasonable to do out-of-tree. But it's disappearing (the very last patch series is pending merge I think now). It worked out well in the end, but I don't think it's a good model to emulate. There's even a patch to rename the directory to make it clear they're full drm drivers, just small one-file ones. Having an entire driver sit in between the actual driver and drm core feels a lot like a midlayer, and we try very hard to avoid those. -Daniel > The difference might not be that significant, but just a question of > were the code is located. > > > >>> - Improve the other helpers we have as needed - there's probably room for > >>> a simple ttm version, inspired by all these simple display chips (e.g. > >>> ast/cirrus/bochs/mga all solve similar problems like this). > >> > >> Makes sense, but appears to be unrelated. > > > > It's the main reason to have drivers in upstream - sharing code and > > infrastructure. > > I meant that creating 'simple TTM' is unrelated to fbdevdrm. It would > make sense in any case. > > Best regards > Thomas > > At least for me the main benefit in porting a bunch > > more fbdev drivers over isn't the old hw support (that hw is quietly > > dying anyway, we just have to wait). But improving support for new hw > > that fills similar niches, and making it easier for everyone else. Ofc > > I'm not going to stop anyone who's going to do all the porting work, > > but we can't realistically require it (e.g. legacy kms drivers also > > don't support atomic, since simply not possible without rewriting the > > driver). > > > > Cheers, Daniel > >> > >> Best regards > >> Thomas > >> > >>> - Treat any such drivers as CONFIG_STAGING until they've become fully > >>> native, i.e. if no one bothers to convert them, we'll drop them again. > >>> > >>> The above is kinda similar in spirit to the transitional helpers we've > >>> used to upgrade the big drivers from legacy kms to atomic. > >>> > >>> Thoughts? > >>> > >>> Cheers, Daniel > >>> > >>>> > >>>> drivers/gpu/drm/Kconfig | 2 + > >>>> drivers/gpu/drm/Makefile | 1 + > >>>> drivers/gpu/drm/fbdevdrm/Kconfig | 13 + > >>>> drivers/gpu/drm/fbdevdrm/Makefile | 11 + > >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.c | 276 ++++++++ > >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.h | 58 ++ > >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_device.c | 96 +++ > >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_device.h | 55 ++ > >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_drv.c | 347 +++++++++ > >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_format.c | 441 ++++++++++++ > >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_format.h | 26 + > >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.c | 195 +++++ > >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.h | 53 ++ > >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.c | 746 ++++++++++++++++++++ > >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.h | 38 + > >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.c | 498 +++++++++++++ > >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.h | 27 + > >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.c | 202 ++++++ > >>>> drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.h | 35 + > >>>> 19 files changed, 3120 insertions(+) > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/Kconfig > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/Makefile > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.c > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_bo.h > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_device.c > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_device.h > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_drv.c > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_format.c > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_format.h > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.c > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modes.h > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.c > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_modeset.h > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.c > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_primary.h > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.c > >>>> create mode 100644 drivers/gpu/drm/fbdevdrm/fbdevdrm_ttm.h > >>>> > >>>> -- > >>>> 2.21.0 > >>>> > >>> > >> > >> -- > >> Thomas Zimmermann > >> Graphics Driver Developer > >> SUSE Linux GmbH, Maxfeldstrasse 5, 90409 Nuernberg, Germany > >> GF: Felix Imendörffer, Mary Higgins, Sri Rasiah > >> HRB 21284 (AG Nürnberg) > >> > > > > > > -- > Thomas Zimmermann > Graphics Driver Developer > SUSE Linux GmbH, Maxfeldstrasse 5, 90409 Nuernberg, Germany > GF: Felix Imendörffer, Mary Higgins, Sri Rasiah > HRB 21284 (AG Nürnberg) >