Documentation/driver-api/pci/pci.rst | 3 + drivers/gpu/drm/vboxvideo/vbox_main.c | 24 +- drivers/pci/devres.c | 996 ++++++++++++++++++++++---- drivers/pci/iomap.c | 18 + drivers/pci/pci.c | 124 +++- drivers/pci/pci.h | 24 - include/linux/pci.h | 17 + 7 files changed, 989 insertions(+), 217 deletions(-)
¡Hola! PCI's devres API suffers several weaknesses: 1. There are functions prefixed with pcim_. Those are always managed counterparts to never-managed functions prefixed with pci_ – or so one would like to think. There are some apparently unmanaged functions (all region-request / release functions, and pci_intx()) which suddenly become managed once the user has initialized the device with pcim_enable_device() instead of pci_enable_device(). This "sometimes yes, sometimes no" nature of those functions is confusing and therefore bug-provoking. In fact, it has already caused a bug in DRM. The last patch in this series fixes that bug. 2. iomappings: Instead of giving each mapping its own callback, the existing API uses a statically allocated struct tracking one mapping per bar. This is not extensible. Especially, you can't create _ranged_ managed mappings that way, which many drivers want. 3. Managed request functions only exist as "plural versions" with a bit-mask as a parameter. That's quite over-engineered considering that each user only ever mapps one, maybe two bars. This series: - add a set of new "singular" devres functions that use devres the way its intended, with one callback per resource. - deprecates the existing iomap-table mechanism. - deprecates the hybrid nature of pci_ functions. - preserves backwards compatibility so that drivers using the existing API won't notice any changes. - adds documentation, especially some warning users about the complicated nature of PCI's devres. Note that this series is based on my "unify pci_iounmap"-series from a few weeks ago. [1] I tested this on a x86 VM with a simple pci test-device with two regions. Operates and reserves resources as intended on my system. Kasan and kmemleak didn't find any problems. I believe this series cleans the API up as much as possible without having to port all existing drivers to the new API. Especially, I think that this implementation is easy to extend if the need for new managed functions arises :) Greetings, P. [1] https://lore.kernel.org/lkml/20240111085540.7740-1-pstanner@redhat.com/ Philipp Stanner (10): pci: add new set of devres functions pci: deprecate iomap-table functions pci: warn users about complicated devres nature pci: devres: make devres region requests consistent pci: move enabled status bit to pci_dev struct pci: move pinned status bit to pci_dev struct pci: devres: give mwi its own callback pci: devres: give pci(m)_intx its own callback pci: devres: remove legacy pcim_release() drm/vboxvideo: fix mapping leaks Documentation/driver-api/pci/pci.rst | 3 + drivers/gpu/drm/vboxvideo/vbox_main.c | 24 +- drivers/pci/devres.c | 996 ++++++++++++++++++++++---- drivers/pci/iomap.c | 18 + drivers/pci/pci.c | 124 +++- drivers/pci/pci.h | 24 - include/linux/pci.h | 17 + 7 files changed, 989 insertions(+), 217 deletions(-) -- 2.43.0
Mon, Jan 15, 2024 at 03:46:11PM +0100, Philipp Stanner kirjoitti: > ¡Hola! i? Vim user? :-) > PCI's devres API suffers several weaknesses: > > 1. There are functions prefixed with pcim_. Those are always managed > counterparts to never-managed functions prefixed with pci_ – or so one > would like to think. There are some apparently unmanaged functions > (all region-request / release functions, and pci_intx()) which > suddenly become managed once the user has initialized the device with > pcim_enable_device() instead of pci_enable_device(). This "sometimes > yes, sometimes no" nature of those functions is confusing and > therefore bug-provoking. In fact, it has already caused a bug in DRM. > The last patch in this series fixes that bug. > 2. iomappings: Instead of giving each mapping its own callback, the > existing API uses a statically allocated struct tracking one mapping > per bar. This is not extensible. Especially, you can't create > _ranged_ managed mappings that way, which many drivers want. > 3. Managed request functions only exist as "plural versions" with a > bit-mask as a parameter. That's quite over-engineered considering > that each user only ever mapps one, maybe two bars. > > This series: > - add a set of new "singular" devres functions that use devres the way > its intended, with one callback per resource. > - deprecates the existing iomap-table mechanism. > - deprecates the hybrid nature of pci_ functions. > - preserves backwards compatibility so that drivers using the existing > API won't notice any changes. > - adds documentation, especially some warning users about the > complicated nature of PCI's devres. Instead of adding pcim_intx(), please provide proper one for pci_alloc_irq_vectors(). Ideally it would be nice to deprecate old IRQ management functions in PCI core and delete them in the future. -- With Best Regards, Andy Shevchenko
On Tue, 2024-01-16 at 23:17 +0200, andy.shevchenko@gmail.com wrote: > Mon, Jan 15, 2024 at 03:46:11PM +0100, Philipp Stanner kirjoitti: > > ¡Hola! > > i? Vim user? :-) The Dark Side of the Force is the path to many abilities, that some consider to be... unnatural https://www.neo-layout.org/ > > > PCI's devres API suffers several weaknesses: > > > > 1. There are functions prefixed with pcim_. Those are always > > managed > > counterparts to never-managed functions prefixed with pci_ – or > > so one > > would like to think. There are some apparently unmanaged > > functions > > (all region-request / release functions, and pci_intx()) which > > suddenly become managed once the user has initialized the device > > with > > pcim_enable_device() instead of pci_enable_device(). This > > "sometimes > > yes, sometimes no" nature of those functions is confusing and > > therefore bug-provoking. In fact, it has already caused a bug in > > DRM. > > The last patch in this series fixes that bug. > > 2. iomappings: Instead of giving each mapping its own callback, the > > existing API uses a statically allocated struct tracking one > > mapping > > per bar. This is not extensible. Especially, you can't create > > _ranged_ managed mappings that way, which many drivers want. > > 3. Managed request functions only exist as "plural versions" with a > > bit-mask as a parameter. That's quite over-engineered > > considering > > that each user only ever mapps one, maybe two bars. > > > > This series: > > - add a set of new "singular" devres functions that use devres the > > way > > its intended, with one callback per resource. > > - deprecates the existing iomap-table mechanism. > > - deprecates the hybrid nature of pci_ functions. > > - preserves backwards compatibility so that drivers using the > > existing > > API won't notice any changes. > > - adds documentation, especially some warning users about the > > complicated nature of PCI's devres. > > Instead of adding pcim_intx(), please provide proper one for > pci_alloc_irq_vectors(). Ideally it would be nice to deprecate > old IRQ management functions in PCI core and delete them in the > future. > In order to deprecate the intermingling with half-managed hyprid devres in pci.c, you need to have pci_intx() be backwards compatible. Unless you can remove it at once. And the least broken way to do that I thought would be pcim_intx(), because that's consistent with how I make pci_request_region() & Co. call into their managed counterparts. There are 25 users of pci_intx(). We'd have to look how many of them call pcim_enable_device() and how easy they would be to port to... pci_alloc_irq_vectors() you say? I haven't used that before. Would have to look into it and see how we could do that. P.
On Wed, 2024-01-17 at 10:59 +0100, Philipp Stanner wrote: > On Tue, 2024-01-16 at 23:17 +0200, andy.shevchenko@gmail.com wrote: > > Mon, Jan 15, 2024 at 03:46:11PM +0100, Philipp Stanner kirjoitti: [...] > > > > > > > PCI's devres API suffers several weaknesses: > > > > > > 1. There are functions prefixed with pcim_. Those are always > > > managed > > > counterparts to never-managed functions prefixed with pci_ – > > > or > > > so one > > > would like to think. There are some apparently unmanaged > > > functions > > > (all region-request / release functions, and pci_intx()) which > > > suddenly become managed once the user has initialized the > > > device > > > with > > > pcim_enable_device() instead of pci_enable_device(). This > > > "sometimes > > > yes, sometimes no" nature of those functions is confusing and > > > therefore bug-provoking. In fact, it has already caused a bug > > > in > > > DRM. > > > The last patch in this series fixes that bug. > > > 2. iomappings: Instead of giving each mapping its own callback, > > > the > > > existing API uses a statically allocated struct tracking one > > > mapping > > > per bar. This is not extensible. Especially, you can't create > > > _ranged_ managed mappings that way, which many drivers want. > > > 3. Managed request functions only exist as "plural versions" with > > > a > > > bit-mask as a parameter. That's quite over-engineered > > > considering > > > that each user only ever mapps one, maybe two bars. > > > > > > This series: > > > - add a set of new "singular" devres functions that use devres > > > the > > > way > > > its intended, with one callback per resource. > > > - deprecates the existing iomap-table mechanism. > > > - deprecates the hybrid nature of pci_ functions. > > > - preserves backwards compatibility so that drivers using the > > > existing > > > API won't notice any changes. > > > - adds documentation, especially some warning users about the > > > complicated nature of PCI's devres. > > > > Instead of adding pcim_intx(), please provide proper one for > > pci_alloc_irq_vectors(). Ideally it would be nice to deprecate > > old IRQ management functions in PCI core and delete them in the > > future. > > > > In order to deprecate the intermingling with half-managed hyprid > devres > in pci.c, you need to have pci_intx() be backwards compatible. Unless > you can remove it at once. > And the least broken way to do that I thought would be pcim_intx(), > because that's consistent with how I make pci_request_region() & Co. > call into their managed counterparts. > > There are 25 users of pci_intx(). > We'd have to look how many of them call pcim_enable_device() and how > easy they would be to port to... pci_alloc_irq_vectors() you say? I > haven't used that before. Would have to look into it and see how we > could do that. Alright, so I thought about that a bit. So pci_intx() is the old way to do it and you would like to deprecate it for good. Understood, makes sense This series, however, is about deprecating PCI's broken devres implementation – not about deprecating outdated PCI features in general. So I wouldn't like to touch anything here unless cleaning up devres demands it. Now the question would be: how can we solve this? My suggestion would be: Let's implement pcim_intx(), but only make it visible through drivers/pci/pci.h. So we won't make it usable for other drivers, don't EXPORT_SYMBOL() it and basically only have it as a tool to move the devres-part clearly and cleanly from pci.c to devres.c Further deprecating old PCI stuff could then be done in a separate series. ACK? P. > > > P.
On Mon, Jan 15, 2024 at 03:46:11PM +0100, Philipp Stanner wrote:
> ...
> pci: add new set of devres functions
> pci: deprecate iomap-table functions
> pci: warn users about complicated devres nature
> pci: devres: make devres region requests consistent
> pci: move enabled status bit to pci_dev struct
> pci: move pinned status bit to pci_dev struct
> pci: devres: give mwi its own callback
> pci: devres: give pci(m)_intx its own callback
> pci: devres: remove legacy pcim_release()
> drm/vboxvideo: fix mapping leaks
If/when you update these, take a look at the drivers/pci/ subject line
history and capitalize these to match.
We haven't really used the "devres" prefix in drivers/pci.
The de facto convention is:
- "PCI/AER:" for major features defined by the PCIe base spec (AER,
DPC, VGA, ASPM, PM, etc).
- "PCI: iproc:" for controller drivers (iproc, dwc, qcom, mvebu,
etc).
- "PCI:" for everything else
© 2016 - 2025 Red Hat, Inc.