Message ID | 20240909132810.1296786-1-ruanjinjie@huawei.com |
---|---|
Headers | show |
Series | spi: geni-qcom: Undo runtime PM changes at driver exit time | expand |
Hi, On Mon, Sep 9, 2024 at 6:19 AM Jinjie Ruan <ruanjinjie@huawei.com> wrote: > > Use devm_pm_runtime_enable(), devm_request_irq() and > devm_spi_register_controller() to simplify code. > > And also register a callback spi_geni_release_dma_chan() with > devm_add_action_or_reset(), to release dma channel in both error > and device detach path, which can make sure the release sequence is > consistent with the original one. > > 1. Unregister spi controller. > 2. Free the IRQ. > 3. Free DMA chans > 4. Disable runtime PM. > > So the remove function can also be removed. > > Suggested-by: Doug Anderson <dianders@chromium.org> > Signed-off-by: Jinjie Ruan <ruanjinjie@huawei.com> > --- > v4: > - Correct the "data" of devm_add_action_or_reset(). > v3: > - Land the rest of the cleanups afterwards. > --- > drivers/spi/spi-geni-qcom.c | 37 +++++++++++++------------------------ > 1 file changed, 13 insertions(+), 24 deletions(-) > > diff --git a/drivers/spi/spi-geni-qcom.c b/drivers/spi/spi-geni-qcom.c > index 6f4057330444..5cb002d7d4a6 100644 > --- a/drivers/spi/spi-geni-qcom.c > +++ b/drivers/spi/spi-geni-qcom.c > @@ -632,8 +632,10 @@ static int spi_geni_grab_gpi_chan(struct spi_geni_master *mas) > return ret; > } > > -static void spi_geni_release_dma_chan(struct spi_geni_master *mas) > +static void spi_geni_release_dma_chan(void *data) > { > + struct spi_geni_master *mas = data; > + > if (mas->rx) { > dma_release_channel(mas->rx); > mas->rx = NULL; > @@ -1132,6 +1134,12 @@ static int spi_geni_probe(struct platform_device *pdev) > if (ret) > return ret; > > + ret = devm_add_action_or_reset(dev, spi_geni_release_dma_chan, mas); > + if (ret) { > + dev_err(dev, "Unable to add action.\n"); > + return ret; > + } Use dev_err_probe() to simplify. ret = devm_add_action_or_reset(dev, spi_geni_release_dma_chan, mas); if (ret) return dev_err_probe(dev, ret, "Unable to add action.\n"); Personally I'd also rather that you do the devm_add_action_or_reset() call straight in spi_geni_grab_gpi_chan(). That makes it much more obvious what's happening. You can still use dev_err_probe() in there since it's called (indirectly) from probe. In that case you'd probably replace the "return 0;" in that function with just "return dev_err_probe(...)". > @@ -1146,33 +1154,15 @@ static int spi_geni_probe(struct platform_device *pdev) > if (mas->cur_xfer_mode == GENI_GPI_DMA) > spi->flags = SPI_CONTROLLER_MUST_TX; > > - ret = request_irq(mas->irq, geni_spi_isr, 0, dev_name(dev), spi); > + ret = devm_request_irq(dev, mas->irq, geni_spi_isr, 0, dev_name(dev), spi); > if (ret) > - goto spi_geni_release_dma; > + return ret; > > - ret = spi_register_controller(spi); > + ret = devm_spi_register_controller(dev, spi); > if (ret) > - goto spi_geni_probe_free_irq; > + return ret; > > return 0; You no longer need the "if" statement or even to assign to "ret". Just: return devm_spi_register_controller(dev, spi); Those are just nits, though. I'd be OK with: Reviewed-by: Douglas Anderson <dianders@chromium.org> ...since Mark has already landed the first two patches, your v5 would just contain this one patch. -Doug
On 2024/9/12 6:53, Doug Anderson wrote: > Hi, > > On Mon, Sep 9, 2024 at 6:19 AM Jinjie Ruan <ruanjinjie@huawei.com> wrote: >> >> Use devm_pm_runtime_enable(), devm_request_irq() and >> devm_spi_register_controller() to simplify code. >> >> And also register a callback spi_geni_release_dma_chan() with >> devm_add_action_or_reset(), to release dma channel in both error >> and device detach path, which can make sure the release sequence is >> consistent with the original one. >> >> 1. Unregister spi controller. >> 2. Free the IRQ. >> 3. Free DMA chans >> 4. Disable runtime PM. >> >> So the remove function can also be removed. >> >> Suggested-by: Doug Anderson <dianders@chromium.org> >> Signed-off-by: Jinjie Ruan <ruanjinjie@huawei.com> >> --- >> v4: >> - Correct the "data" of devm_add_action_or_reset(). >> v3: >> - Land the rest of the cleanups afterwards. >> --- >> drivers/spi/spi-geni-qcom.c | 37 +++++++++++++------------------------ >> 1 file changed, 13 insertions(+), 24 deletions(-) >> >> diff --git a/drivers/spi/spi-geni-qcom.c b/drivers/spi/spi-geni-qcom.c >> index 6f4057330444..5cb002d7d4a6 100644 >> --- a/drivers/spi/spi-geni-qcom.c >> +++ b/drivers/spi/spi-geni-qcom.c >> @@ -632,8 +632,10 @@ static int spi_geni_grab_gpi_chan(struct spi_geni_master *mas) >> return ret; >> } >> >> -static void spi_geni_release_dma_chan(struct spi_geni_master *mas) >> +static void spi_geni_release_dma_chan(void *data) >> { >> + struct spi_geni_master *mas = data; >> + >> if (mas->rx) { >> dma_release_channel(mas->rx); >> mas->rx = NULL; >> @@ -1132,6 +1134,12 @@ static int spi_geni_probe(struct platform_device *pdev) >> if (ret) >> return ret; >> >> + ret = devm_add_action_or_reset(dev, spi_geni_release_dma_chan, mas); >> + if (ret) { >> + dev_err(dev, "Unable to add action.\n"); >> + return ret; >> + } > > Use dev_err_probe() to simplify. > > ret = devm_add_action_or_reset(dev, spi_geni_release_dma_chan, mas); > if (ret) > return dev_err_probe(dev, ret, "Unable to add action.\n"); It seems that if it only return -ENOMEM or 0, using dev_err_probe() has not not much value for many community maintainers. > > > Personally I'd also rather that you do the devm_add_action_or_reset() > call straight in spi_geni_grab_gpi_chan(). That makes it much more Yes, it will be more clear. > obvious what's happening. You can still use dev_err_probe() in there > since it's called (indirectly) from probe. In that case you'd probably > replace the "return 0;" in that function with just "return > dev_err_probe(...)". > > >> @@ -1146,33 +1154,15 @@ static int spi_geni_probe(struct platform_device *pdev) >> if (mas->cur_xfer_mode == GENI_GPI_DMA) >> spi->flags = SPI_CONTROLLER_MUST_TX; >> >> - ret = request_irq(mas->irq, geni_spi_isr, 0, dev_name(dev), spi); >> + ret = devm_request_irq(dev, mas->irq, geni_spi_isr, 0, dev_name(dev), spi); >> if (ret) >> - goto spi_geni_release_dma; >> + return ret; >> >> - ret = spi_register_controller(spi); >> + ret = devm_spi_register_controller(dev, spi); >> if (ret) >> - goto spi_geni_probe_free_irq; >> + return ret; >> >> return 0; > > You no longer need the "if" statement or even to assign to "ret". Just: > > return devm_spi_register_controller(dev, spi); Right! > > > Those are just nits, though. I'd be OK with: > > Reviewed-by: Douglas Anderson <dianders@chromium.org> > > ...since Mark has already landed the first two patches, your v5 would > just contain this one patch. > > -Doug
Hi, On Wed, Sep 11, 2024 at 8:53 PM Jinjie Ruan <ruanjinjie@huawei.com> wrote: > > >> @@ -1132,6 +1134,12 @@ static int spi_geni_probe(struct platform_device *pdev) > >> if (ret) > >> return ret; > >> > >> + ret = devm_add_action_or_reset(dev, spi_geni_release_dma_chan, mas); > >> + if (ret) { > >> + dev_err(dev, "Unable to add action.\n"); > >> + return ret; > >> + } > > > > Use dev_err_probe() to simplify. > > > > ret = devm_add_action_or_reset(dev, spi_geni_release_dma_chan, mas); > > if (ret) > > return dev_err_probe(dev, ret, "Unable to add action.\n"); > > It seems that if it only return -ENOMEM or 0, using dev_err_probe() has > not not much value for many community maintainers. While I won't insist, it still has some value to use dev_err_probe() as I talked about in commit 7065f92255bb ("driver core: Clarify that dev_err_probe() is OK even w/out -EPROBE_DEFER"). -Doug
On 2024/9/12 21:38, Doug Anderson wrote: > Hi, > > On Wed, Sep 11, 2024 at 8:53 PM Jinjie Ruan <ruanjinjie@huawei.com> wrote: >> >>>> @@ -1132,6 +1134,12 @@ static int spi_geni_probe(struct platform_device *pdev) >>>> if (ret) >>>> return ret; >>>> >>>> + ret = devm_add_action_or_reset(dev, spi_geni_release_dma_chan, mas); >>>> + if (ret) { >>>> + dev_err(dev, "Unable to add action.\n"); >>>> + return ret; >>>> + } >>> >>> Use dev_err_probe() to simplify. >>> >>> ret = devm_add_action_or_reset(dev, spi_geni_release_dma_chan, mas); >>> if (ret) >>> return dev_err_probe(dev, ret, "Unable to add action.\n"); >> >> It seems that if it only return -ENOMEM or 0, using dev_err_probe() has >> not not much value for many community maintainers. > > While I won't insist, it still has some value to use dev_err_probe() > as I talked about in commit 7065f92255bb ("driver core: Clarify that > dev_err_probe() is OK even w/out -EPROBE_DEFER") The main difference is that when use dev_err_probe(),there will print anything on -ENOMEM now. > > -Doug
Hi, On Thu, Sep 12, 2024 at 11:44 PM Jinjie Ruan <ruanjinjie@huawei.com> wrote: > > On 2024/9/12 21:38, Doug Anderson wrote: > > Hi, > > > > On Wed, Sep 11, 2024 at 8:53 PM Jinjie Ruan <ruanjinjie@huawei.com> wrote: > >> > >>>> @@ -1132,6 +1134,12 @@ static int spi_geni_probe(struct platform_device *pdev) > >>>> if (ret) > >>>> return ret; > >>>> > >>>> + ret = devm_add_action_or_reset(dev, spi_geni_release_dma_chan, mas); > >>>> + if (ret) { > >>>> + dev_err(dev, "Unable to add action.\n"); > >>>> + return ret; > >>>> + } > >>> > >>> Use dev_err_probe() to simplify. > >>> > >>> ret = devm_add_action_or_reset(dev, spi_geni_release_dma_chan, mas); > >>> if (ret) > >>> return dev_err_probe(dev, ret, "Unable to add action.\n"); > >> > >> It seems that if it only return -ENOMEM or 0, using dev_err_probe() has > >> not not much value for many community maintainers. > > > > While I won't insist, it still has some value to use dev_err_probe() > > as I talked about in commit 7065f92255bb ("driver core: Clarify that > > dev_err_probe() is OK even w/out -EPROBE_DEFER") > The main difference is that when use dev_err_probe(),there will print > anything on -ENOMEM now. Oh, I see. You're saying that we should just get rid of the print altogether because the only error case is -ENOMEM and the kernel already splats there? Yeah, that sounds right to me. That doesn't match what you did in v5, though... -Doug
On 2024/9/14 0:27, Doug Anderson wrote: > Hi, > > On Thu, Sep 12, 2024 at 11:44 PM Jinjie Ruan <ruanjinjie@huawei.com> wrote: >> >> On 2024/9/12 21:38, Doug Anderson wrote: >>> Hi, >>> >>> On Wed, Sep 11, 2024 at 8:53 PM Jinjie Ruan <ruanjinjie@huawei.com> wrote: >>>> >>>>>> @@ -1132,6 +1134,12 @@ static int spi_geni_probe(struct platform_device *pdev) >>>>>> if (ret) >>>>>> return ret; >>>>>> >>>>>> + ret = devm_add_action_or_reset(dev, spi_geni_release_dma_chan, mas); >>>>>> + if (ret) { >>>>>> + dev_err(dev, "Unable to add action.\n"); >>>>>> + return ret; >>>>>> + } >>>>> >>>>> Use dev_err_probe() to simplify. >>>>> >>>>> ret = devm_add_action_or_reset(dev, spi_geni_release_dma_chan, mas); >>>>> if (ret) >>>>> return dev_err_probe(dev, ret, "Unable to add action.\n"); >>>> >>>> It seems that if it only return -ENOMEM or 0, using dev_err_probe() has >>>> not not much value for many community maintainers. >>> >>> While I won't insist, it still has some value to use dev_err_probe() >>> as I talked about in commit 7065f92255bb ("driver core: Clarify that >>> dev_err_probe() is OK even w/out -EPROBE_DEFER") >> The main difference is that when use dev_err_probe(),there will print >> anything on -ENOMEM now. > > Oh, I see. You're saying that we should just get rid of the print > altogether because the only error case is -ENOMEM and the kernel > already splats there? Yeah, that sounds right to me. That doesn't > match what you did in v5, though... I think the following 2 soultion is both fine: 1、return ret directly. 2、dev_err() and return. > > -Doug