diff mbox series

[v4] dmaengine: tegra210-adma: Fix runtime PM imbalance on error

Message ID 20200621054710.9915-1-dinghao.liu@zju.edu.cn
State New
Headers show
Series [v4] dmaengine: tegra210-adma: Fix runtime PM imbalance on error | expand

Commit Message

Dinghao Liu June 21, 2020, 5:47 a.m. UTC
pm_runtime_get_sync() increments the runtime PM usage counter even
when it returns an error code. Thus a pairing decrement is needed on
the error handling path to keep the counter balanced.

Signed-off-by: Dinghao Liu <dinghao.liu@zju.edu.cn>
---

v2: - Merge two patches that fix runtime PM imbalance in
      tegra_adma_probe() and tegra_adma_alloc_chan_resources()
      respectively.

v3: - Use pm_runtime_put_noidle() instead of pm_runtime_put_sync()
      in tegra_adma_alloc_chan_resources().

v4: - Use pm_runtime_put_noidle() instead of pm_runtime_put_sync()
      in tegra_adma_probe().
---
 drivers/dma/tegra210-adma.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

Comments

Jon Hunter June 23, 2020, 10:13 a.m. UTC | #1
On 21/06/2020 06:47, Dinghao Liu wrote:
> pm_runtime_get_sync() increments the runtime PM usage counter even
> when it returns an error code. Thus a pairing decrement is needed on
> the error handling path to keep the counter balanced.

So you have not mentioned here why you are using _noidle and not _put.
Furthermore, in this patch [0] you are not using _noidle to fix the same
problem in another driver. We should fix this in a consistent manner
across all drivers, otherwise it leads to more confusion.

Finally, Rafael mentions we should just use _put [0] and so I think we
should follow his recommendation.

Jon

[0] https://lkml.org/lkml/2020/5/21/601
Geert Uytterhoeven June 23, 2020, 12:08 p.m. UTC | #2
Hi Jon,

More stirring in the cesspool ;-)

On Tue, Jun 23, 2020 at 12:13 PM Jon Hunter <jonathanh@nvidia.com> wrote:
> On 21/06/2020 06:47, Dinghao Liu wrote:
> > pm_runtime_get_sync() increments the runtime PM usage counter even
> > when it returns an error code. Thus a pairing decrement is needed on
> > the error handling path to keep the counter balanced.
>
> So you have not mentioned here why you are using _noidle and not _put.
> Furthermore, in this patch [0] you are not using _noidle to fix the same
> problem in another driver. We should fix this in a consistent manner
> across all drivers, otherwise it leads to more confusion.
>
> Finally, Rafael mentions we should just use _put [0] and so I think we
> should follow his recommendation.
>
> Jon
>
> [0] https://lkml.org/lkml/2020/5/21/601

"_noidle() is the simplest one and it is sufficient."
https://lore.kernel.org/linux-i2c/CAJZ5v0i87NGcy9+kxubScdPDyByr8ypQWcGgBFn+V-wDd69BHQ@mail.gmail.com/

You never know what additional things the other put* variants
will start doing in the future...

Gr{oetje,eeting}s,

                        Geert


--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds
Jon Hunter June 23, 2020, 12:25 p.m. UTC | #3
Hi Geert,

On 23/06/2020 13:08, Geert Uytterhoeven wrote:
> Hi Jon,
> 
> More stirring in the cesspool ;-)

Ha! Indeed.

> On Tue, Jun 23, 2020 at 12:13 PM Jon Hunter <jonathanh@nvidia.com> wrote:
>> On 21/06/2020 06:47, Dinghao Liu wrote:
>>> pm_runtime_get_sync() increments the runtime PM usage counter even
>>> when it returns an error code. Thus a pairing decrement is needed on
>>> the error handling path to keep the counter balanced.
>>
>> So you have not mentioned here why you are using _noidle and not _put.
>> Furthermore, in this patch [0] you are not using _noidle to fix the same
>> problem in another driver. We should fix this in a consistent manner
>> across all drivers, otherwise it leads to more confusion.
>>
>> Finally, Rafael mentions we should just use _put [0] and so I think we
>> should follow his recommendation.
>>
>> Jon
>>
>> [0] https://lkml.org/lkml/2020/5/21/601
> 
> "_noidle() is the simplest one and it is sufficient."
> https://lore.kernel.org/linux-i2c/CAJZ5v0i87NGcy9+kxubScdPDyByr8ypQWcGgBFn+V-wDd69BHQ@mail.gmail.com/

Good to know. This detail should be spelled out in the changelog so that
it is clear why we are using _noidle and not _put. I did take a look and
it did seem to handle the usage_count OK, but I was concerned if there
could be something else in the _put path that may get missed.

Anyway, I am fine with the change, but with an updated changelog on why
_noidle is being used.

> You never know what additional things the other put* variants
> will start doing in the future...

Hopefully not, as that would be a breakage of the API itself. From what
Rafael said that all _put calls should work and if at some point in the
future they don't, then that seems like a regression.

Jon
diff mbox series

Patch

diff --git a/drivers/dma/tegra210-adma.c b/drivers/dma/tegra210-adma.c
index db58d7e4f9fe..c5fa2ef74abc 100644
--- a/drivers/dma/tegra210-adma.c
+++ b/drivers/dma/tegra210-adma.c
@@ -658,6 +658,7 @@  static int tegra_adma_alloc_chan_resources(struct dma_chan *dc)
 
 	ret = pm_runtime_get_sync(tdc2dev(tdc));
 	if (ret < 0) {
+		pm_runtime_put_noidle(tdc2dev(tdc));
 		free_irq(tdc->irq, tdc);
 		return ret;
 	}
@@ -869,8 +870,10 @@  static int tegra_adma_probe(struct platform_device *pdev)
 	pm_runtime_enable(&pdev->dev);
 
 	ret = pm_runtime_get_sync(&pdev->dev);
-	if (ret < 0)
+	if (ret < 0) {
+		pm_runtime_put_noidle(&pdev->dev);
 		goto rpm_disable;
+	}
 
 	ret = tegra_adma_init(tdma);
 	if (ret)