| Message ID | 20240827185616.45094-1-pstanner@redhat.com |
|---|---|
| Headers | show |
| Series | PCI: Remove pcim_iounmap_regions() | expand |
PS: This series's title should now obviously be "*Mostly* Remove pcim_iounmap_regions()". On Tue, 2024-08-27 at 20:56 +0200, Philipp Stanner wrote: > OK, so unfortunately it seems very challenging to reconcile the merge > conflict pointed up by Serge between net-next and pci-devres > regarding > "ethernet: stmicro": A patch that applies to the net-next tree does > not > apply anymore to pci-devres (and vice versa). > > So I actually think that it would be best if we just drop the > portation > of "ethernet: stmicro" for now and port it as the last user in v6.13. > > That should then be trivial. > > Changes in v4: > - Drop the "ethernet: stmicro: [...] patch since it doesn't apply > to > net-next, and making it apply to that prevents it from being > applyable to PCI ._. (Serge, me) > - Instead, deprecate pcim_iounmap_regions() and keep "ethernet: > stimicro" as the last user for now. Perform the deprecation in > the > series' first patch. Remove the Reviewed-by's givin so far to > that > patch. > - ethernet: cavium: Use PTR_ERR_OR_ZERO(). (Andy) > - vdpa: solidrun (Bugfix) Correct wrong printf string (was "psnet" > instead of > "snet"). (Christophe) > - vdpa: solidrun (Bugfix): Add missing blank line. (Andy) > - vdpa: solidrun (Portation): Use PTR_ERR_OR_ZERO(). (Andy) > - Apply Reviewed-by's from Andy and Xu Yilun. > > Changes in v3: > - fpga/dfl-pci.c: remove now surplus wrapper around > pcim_iomap_region(). (Andy) > - block: mtip32xx: remove now surplus label. (Andy) > - vdpa: solidrun: Bugfix: Include forgotten place where stack UB > occurs. (Andy, Christophe) > - Some minor wording improvements in commit messages. (Me) > > Changes in v2: > - Add a fix for the UB stack usage bug in vdap/solidrun. Separate > patch, put stable kernel on CC. (Christophe, Andy). > - Drop unnecessary pcim_release_region() in mtip32xx (Andy) > - Consequently, drop patch "PCI: Make pcim_release_region() a > public > function", since there's no user anymore. (obsoletes the squash > requested by Damien). > - vdap/solidrun: > • make 'i' an 'unsigned short' (Andy, me) > • Use 'continue' to simplify loop (Andy) > • Remove leftover blank line > - Apply given Reviewed- / acked-bys (Andy, Damien, Bartosz) > > > Important things first: > This series is based on [1] and [2] which Bjorn Helgaas has currently > queued for v6.12 in the PCI tree. > > This series shall remove pcim_iounmap_regions() in order to make way > to > remove its brother, pcim_iomap_regions(). > > @Bjorn: Feel free to squash the PCI commits. > > Regards, > P. > > [1] > https://lore.kernel.org/all/20240729093625.17561-4-pstanner@redhat.com/ > [2] > https://lore.kernel.org/all/20240807083018.8734-2-pstanner@redhat.com/ > > Philipp Stanner (7): > PCI: Deprecate pcim_iounmap_regions() > fpga/dfl-pci.c: Replace deprecated PCI functions > block: mtip32xx: Replace deprecated PCI functions > gpio: Replace deprecated PCI functions > ethernet: cavium: Replace deprecated PCI functions > vdpa: solidrun: Fix UB bug with devres > vdap: solidrun: Replace deprecated PCI functions > > drivers/block/mtip32xx/mtip32xx.c | 16 +++-- > drivers/fpga/dfl-pci.c | 16 ++--- > drivers/gpio/gpio-merrifield.c | 14 ++--- > .../net/ethernet/cavium/common/cavium_ptp.c | 6 +- > drivers/pci/devres.c | 8 ++- > drivers/vdpa/solidrun/snet_main.c | 59 ++++++++--------- > -- > include/linux/pci.h | 1 + > 7 files changed, 51 insertions(+), 69 deletions(-) >
On 8/27/24 12:56 PM, Philipp Stanner wrote: > pcim_iomap_regions() and pcim_iomap_table() have been deprecated by the > PCI subsystem in commit e354bb84a4c1 ("PCI: Deprecate > pcim_iomap_table(), pcim_iomap_regions_request_all()"). > > In mtip32xx, these functions can easily be replaced by their respective > successors, pcim_request_region() and pcim_iomap(). Moreover, the > driver's calls to pcim_iounmap_regions() in probe()'s error path and in > remove() are not necessary. Cleanup can be performed by PCI devres > automatically. > > Replace pcim_iomap_regions() and pcim_iomap_table(). > > Remove the calls to pcim_iounmap_regions(). Looks fine to me - since it depends on other trees, feel free to take it through those: Reviewed-by: Jens Axboe <axboe@kernel.dk>
On Tue, 2024-08-27 at 20:56 +0200, Philipp Stanner wrote: > pcim_iomap_regions() and pcim_iomap_table() have been deprecated by > the PCI subsystem in commit e354bb84a4c1 ("PCI: Deprecate > pcim_iomap_table(), pcim_iomap_regions_request_all()"). > > Furthermore, the driver contains an unneeded call to > pcim_iounmap_regions() in its probe() function's error unwind path. > > Replace the deprecated PCI functions with pcim_iomap_region(). > > Remove the unnecessary call to pcim_iounmap_regions(). > > Signed-off-by: Philipp Stanner <pstanner@redhat.com> > --- > drivers/net/ethernet/cavium/common/cavium_ptp.c | 6 ++---- > 1 file changed, 2 insertions(+), 4 deletions(-) > > diff --git a/drivers/net/ethernet/cavium/common/cavium_ptp.c > b/drivers/net/ethernet/cavium/common/cavium_ptp.c > index 9fd717b9cf69..914ccc8eaf5e 100644 > --- a/drivers/net/ethernet/cavium/common/cavium_ptp.c > +++ b/drivers/net/ethernet/cavium/common/cavium_ptp.c > @@ -239,12 +239,11 @@ static int cavium_ptp_probe(struct pci_dev > *pdev, > if (err) > goto error_free; > > - err = pcim_iomap_regions(pdev, 1 << PCI_PTP_BAR_NO, > pci_name(pdev)); > + clock->reg_base = pcim_iomap_region(pdev, PCI_PTP_BAR_NO, > pci_name(pdev)); > + err = PTR_ERR_OR_ZERO(clock->reg_base); > if (err) > goto error_free; > > - clock->reg_base = pcim_iomap_table(pdev)[PCI_PTP_BAR_NO]; > - > spin_lock_init(&clock->spin_lock); > > cc = &clock->cycle_counter; > @@ -292,7 +291,6 @@ static int cavium_ptp_probe(struct pci_dev *pdev, > clock_cfg = readq(clock->reg_base + PTP_CLOCK_CFG); > clock_cfg &= ~PTP_CLOCK_CFG_PTP_EN; > writeq(clock_cfg, clock->reg_base + PTP_CLOCK_CFG); > - pcim_iounmap_regions(pdev, 1 << PCI_PTP_BAR_NO); I think I removed that by accident – thinking about it, we should not remove it since the driver later returns 0 from its probe(). So we should not keep blocking the region. Has to be addressed in v5. P. > > error_free: > devm_kfree(dev, clock);
On Tue, 2024-08-27 at 13:05 -0600, Jens Axboe wrote: > On 8/27/24 12:56 PM, Philipp Stanner wrote: > > pcim_iomap_regions() and pcim_iomap_table() have been deprecated by > > the > > PCI subsystem in commit e354bb84a4c1 ("PCI: Deprecate > > pcim_iomap_table(), pcim_iomap_regions_request_all()"). > > > > In mtip32xx, these functions can easily be replaced by their > > respective > > successors, pcim_request_region() and pcim_iomap(). Moreover, the > > driver's calls to pcim_iounmap_regions() in probe()'s error path > > and in > > remove() are not necessary. Cleanup can be performed by PCI devres > > automatically. > > > > Replace pcim_iomap_regions() and pcim_iomap_table(). > > > > Remove the calls to pcim_iounmap_regions(). > > Looks fine to me - since it depends on other trees, feel free to take > it > through those: > > Reviewed-by: Jens Axboe <axboe@kernel.dk> Thank you for the review. I have to provide a v5 because of an issue in another patch. While I'm at it, I'd modify this patch here so that the comment above pcim_request_region() is descriptive of the actual events: - /* Map BAR5 to memory. */ + /* Request BAR5. */ I'd keep your Reviewed-by if that's OK. It's the only change I'd do. Regards, P.
On 8/28/24 1:10 AM, Philipp Stanner wrote: > On Tue, 2024-08-27 at 13:05 -0600, Jens Axboe wrote: >> On 8/27/24 12:56 PM, Philipp Stanner wrote: >>> pcim_iomap_regions() and pcim_iomap_table() have been deprecated by >>> the >>> PCI subsystem in commit e354bb84a4c1 ("PCI: Deprecate >>> pcim_iomap_table(), pcim_iomap_regions_request_all()"). >>> >>> In mtip32xx, these functions can easily be replaced by their >>> respective >>> successors, pcim_request_region() and pcim_iomap(). Moreover, the >>> driver's calls to pcim_iounmap_regions() in probe()'s error path >>> and in >>> remove() are not necessary. Cleanup can be performed by PCI devres >>> automatically. >>> >>> Replace pcim_iomap_regions() and pcim_iomap_table(). >>> >>> Remove the calls to pcim_iounmap_regions(). >> >> Looks fine to me - since it depends on other trees, feel free to take >> it >> through those: >> >> Reviewed-by: Jens Axboe <axboe@kernel.dk> > > Thank you for the review. > > I have to provide a v5 because of an issue in another patch. While I'm > at it, I'd modify this patch here so that the comment above > pcim_request_region() is descriptive of the actual events: > > - /* Map BAR5 to memory. */ > + /* Request BAR5. */ > > > I'd keep your Reviewed-by if that's OK. It's the only change I'd do. That's fine.
OK, so unfortunately it seems very challenging to reconcile the merge conflict pointed up by Serge between net-next and pci-devres regarding "ethernet: stmicro": A patch that applies to the net-next tree does not apply anymore to pci-devres (and vice versa). So I actually think that it would be best if we just drop the portation of "ethernet: stmicro" for now and port it as the last user in v6.13. That should then be trivial. Changes in v4: - Drop the "ethernet: stmicro: [...] patch since it doesn't apply to net-next, and making it apply to that prevents it from being applyable to PCI ._. (Serge, me) - Instead, deprecate pcim_iounmap_regions() and keep "ethernet: stimicro" as the last user for now. Perform the deprecation in the series' first patch. Remove the Reviewed-by's givin so far to that patch. - ethernet: cavium: Use PTR_ERR_OR_ZERO(). (Andy) - vdpa: solidrun (Bugfix) Correct wrong printf string (was "psnet" instead of "snet"). (Christophe) - vdpa: solidrun (Bugfix): Add missing blank line. (Andy) - vdpa: solidrun (Portation): Use PTR_ERR_OR_ZERO(). (Andy) - Apply Reviewed-by's from Andy and Xu Yilun. Changes in v3: - fpga/dfl-pci.c: remove now surplus wrapper around pcim_iomap_region(). (Andy) - block: mtip32xx: remove now surplus label. (Andy) - vdpa: solidrun: Bugfix: Include forgotten place where stack UB occurs. (Andy, Christophe) - Some minor wording improvements in commit messages. (Me) Changes in v2: - Add a fix for the UB stack usage bug in vdap/solidrun. Separate patch, put stable kernel on CC. (Christophe, Andy). - Drop unnecessary pcim_release_region() in mtip32xx (Andy) - Consequently, drop patch "PCI: Make pcim_release_region() a public function", since there's no user anymore. (obsoletes the squash requested by Damien). - vdap/solidrun: • make 'i' an 'unsigned short' (Andy, me) • Use 'continue' to simplify loop (Andy) • Remove leftover blank line - Apply given Reviewed- / acked-bys (Andy, Damien, Bartosz) Important things first: This series is based on [1] and [2] which Bjorn Helgaas has currently queued for v6.12 in the PCI tree. This series shall remove pcim_iounmap_regions() in order to make way to remove its brother, pcim_iomap_regions(). @Bjorn: Feel free to squash the PCI commits. Regards, P. [1] https://lore.kernel.org/all/20240729093625.17561-4-pstanner@redhat.com/ [2] https://lore.kernel.org/all/20240807083018.8734-2-pstanner@redhat.com/ Philipp Stanner (7): PCI: Deprecate pcim_iounmap_regions() fpga/dfl-pci.c: Replace deprecated PCI functions block: mtip32xx: Replace deprecated PCI functions gpio: Replace deprecated PCI functions ethernet: cavium: Replace deprecated PCI functions vdpa: solidrun: Fix UB bug with devres vdap: solidrun: Replace deprecated PCI functions drivers/block/mtip32xx/mtip32xx.c | 16 +++-- drivers/fpga/dfl-pci.c | 16 ++--- drivers/gpio/gpio-merrifield.c | 14 ++--- .../net/ethernet/cavium/common/cavium_ptp.c | 6 +- drivers/pci/devres.c | 8 ++- drivers/vdpa/solidrun/snet_main.c | 59 ++++++++----------- include/linux/pci.h | 1 + 7 files changed, 51 insertions(+), 69 deletions(-)