| Message ID | 1432076885-5107-1-git-send-email-sakari.ailus@iki.fi (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi Sakari, On Wed, May 20, 2015 at 02:08:05AM +0300, Sakari Ailus wrote: > The async notifier was registered before the v4l2_device was registered and > before the notifier callbacks were set. This could lead to missing the > bound() and complete() callbacks and to attempting to spin_lock() and > uninitialised spin lock. > > Also fix unregistering the async notifier in the case of an error --- the > function may not fail anymore after the notifier is registered. > > Signed-off-by: Sakari Ailus <sakari.ailus@iki.fi> I already noticed this during my Camera for N900 work and solved it the same way, so: Reviewed-By: Sebastian Reichel <sre@kernel.org> You may want to add a Fixes Tag against the patch implementing the async notifier support in omap3isp, since its quite easy to run into the callback problems (at least I did) and the missing resource freeing (due to EPROBE_AGAIN). -- Sebastian
Hi Sakari, Thank you for the patch. On Wednesday 20 May 2015 01:41:43 Sebastian Reichel wrote: > Hi Sakari, > > On Wed, May 20, 2015 at 02:08:05AM +0300, Sakari Ailus wrote: > > The async notifier was registered before the v4l2_device was registered > > and before the notifier callbacks were set. This could lead to missing the > > bound() and complete() callbacks and to attempting to spin_lock() and > > uninitialised spin lock. > > > > Also fix unregistering the async notifier in the case of an error --- the > > function may not fail anymore after the notifier is registered. > > > > Signed-off-by: Sakari Ailus <sakari.ailus@iki.fi> > > I already noticed this during my Camera for N900 work and solved it > the same way, so: > > Reviewed-By: Sebastian Reichel <sre@kernel.org> > > You may want to add a Fixes Tag against the patch implementing the > async notifier support in omap3isp, since its quite easy to run into > the callback problems (at least I did) and the missing resource > freeing (due to EPROBE_AGAIN). Applied to my tree with the Reviewed-by and Fixes tags.
On Wednesday 20 May 2015 09:57:33 Laurent Pinchart wrote: > On Wednesday 20 May 2015 01:41:43 Sebastian Reichel wrote: > > On Wed, May 20, 2015 at 02:08:05AM +0300, Sakari Ailus wrote: > > > The async notifier was registered before the v4l2_device was registered > > > and before the notifier callbacks were set. This could lead to missing > > > the bound() and complete() callbacks and to attempting to spin_lock() > > > and uninitialised spin lock. > > > > > > Also fix unregistering the async notifier in the case of an error --- > > > the function may not fail anymore after the notifier is registered. > > > > > > Signed-off-by: Sakari Ailus <sakari.ailus@iki.fi> > > > > I already noticed this during my Camera for N900 work and solved it > > the same way, so: > > > > Reviewed-By: Sebastian Reichel <sre@kernel.org> > > > > You may want to add a Fixes Tag against the patch implementing the > > async notifier support in omap3isp, since its quite easy to run into > > the callback problems (at least I did) and the missing resource > > freeing (due to EPROBE_AGAIN). > > Applied to my tree with the Reviewed-by and Fixes tags. I spoke too soon. I think you forgot to remove the v4l2_async_notifier_unregister() call from isp_register_entities(). I can fix while applying if you agree with the change.
Hi Laurent, On Wed, May 20, 2015 at 10:14:53AM +0300, Laurent Pinchart wrote: > On Wednesday 20 May 2015 09:57:33 Laurent Pinchart wrote: > > On Wednesday 20 May 2015 01:41:43 Sebastian Reichel wrote: > > > On Wed, May 20, 2015 at 02:08:05AM +0300, Sakari Ailus wrote: > > > > The async notifier was registered before the v4l2_device was registered > > > > and before the notifier callbacks were set. This could lead to missing > > > > the bound() and complete() callbacks and to attempting to spin_lock() > > > > and uninitialised spin lock. > > > > > > > > Also fix unregistering the async notifier in the case of an error --- > > > > the function may not fail anymore after the notifier is registered. > > > > > > > > Signed-off-by: Sakari Ailus <sakari.ailus@iki.fi> > > > > > > I already noticed this during my Camera for N900 work and solved it > > > the same way, so: > > > > > > Reviewed-By: Sebastian Reichel <sre@kernel.org> > > > > > > You may want to add a Fixes Tag against the patch implementing the > > > async notifier support in omap3isp, since its quite easy to run into > > > the callback problems (at least I did) and the missing resource > > > freeing (due to EPROBE_AGAIN). > > > > Applied to my tree with the Reviewed-by and Fixes tags. > > I spoke too soon. I think you forgot to remove the > v4l2_async_notifier_unregister() call from isp_register_entities(). I can fix > while applying if you agree with the change. Please do. Thanks!
On Tue, May 19, 2015 at 6:08 PM, Sakari Ailus <sakari.ailus@iki.fi> wrote: > > @@ -2557,18 +2553,27 @@ static int isp_probe(struct platform_device *pdev) > if (ret < 0) > goto error_iommu; > > - isp->notifier.bound = isp_subdev_notifier_bound; > - isp->notifier.complete = isp_subdev_notifier_complete; > - > ret = isp_register_entities(isp); > if (ret < 0) > goto error_modules; > > + if (IS_ENABLED(CONFIG_OF) && pdev->dev.of_node) { So I have a question (for Laurent, I guess) that's unrelated to this patch. Why is the "IS_ENABLED(CONFIG_OF)" important? If CONFIG_OF is disabled, isn't pdev->dev.of_node always NULL anyway?
Hi Timur, On Wednesday 15 July 2015 16:04:01 Timur Tabi wrote: > On Tue, May 19, 2015 at 6:08 PM, Sakari Ailus <sakari.ailus@iki.fi> wrote: > > @@ -2557,18 +2553,27 @@ static int isp_probe(struct platform_device *pdev) > > if (ret < 0) > > goto error_iommu; > > > > - isp->notifier.bound = isp_subdev_notifier_bound; > > - isp->notifier.complete = isp_subdev_notifier_complete; > > - > > ret = isp_register_entities(isp); > > if (ret < 0) > > goto error_modules; > > > > + if (IS_ENABLED(CONFIG_OF) && pdev->dev.of_node) { > > So I have a question (for Laurent, I guess) that's unrelated to this patch. > > Why is the "IS_ENABLED(CONFIG_OF)" important? If CONFIG_OF is > disabled, isn't pdev->dev.of_node always NULL anyway? IS_ENABLED(CONFIG_OF) can be evaluated at compile time, so it will allow the compiler to remove the code block completely when OF support is disabled. pdev->dev.of_node is a runtime-only check which would allow the same optimization.
Laurent Pinchart wrote: > IS_ENABLED(CONFIG_OF) can be evaluated at compile time, so it will allow the > compiler to remove the code block completely when OF support is disabled. > pdev->dev.of_node is a runtime-only check which would allow the same > optimization. Ok, thanks. I was thinking that there was more to it than just optimization, but I guess not.
diff --git a/drivers/media/platform/omap3isp/isp.c b/drivers/media/platform/omap3isp/isp.c index 18d0a87..0fa95cc 100644 --- a/drivers/media/platform/omap3isp/isp.c +++ b/drivers/media/platform/omap3isp/isp.c @@ -2423,10 +2423,6 @@ static int isp_probe(struct platform_device *pdev) ret = isp_of_parse_nodes(&pdev->dev, &isp->notifier); if (ret < 0) return ret; - ret = v4l2_async_notifier_register(&isp->v4l2_dev, - &isp->notifier); - if (ret) - return ret; } else { isp->pdata = pdev->dev.platform_data; isp->syscon = syscon_regmap_lookup_by_pdevname("syscon.0"); @@ -2557,18 +2553,27 @@ static int isp_probe(struct platform_device *pdev) if (ret < 0) goto error_iommu; - isp->notifier.bound = isp_subdev_notifier_bound; - isp->notifier.complete = isp_subdev_notifier_complete; - ret = isp_register_entities(isp); if (ret < 0) goto error_modules; + if (IS_ENABLED(CONFIG_OF) && pdev->dev.of_node) { + isp->notifier.bound = isp_subdev_notifier_bound; + isp->notifier.complete = isp_subdev_notifier_complete; + + ret = v4l2_async_notifier_register(&isp->v4l2_dev, + &isp->notifier); + if (ret) + goto error_register_entities; + } + isp_core_init(isp, 1); omap3isp_put(isp); return 0; +error_register_entities: + isp_unregister_entities(isp); error_modules: isp_cleanup_modules(isp); error_iommu:
The async notifier was registered before the v4l2_device was registered and before the notifier callbacks were set. This could lead to missing the bound() and complete() callbacks and to attempting to spin_lock() and uninitialised spin lock. Also fix unregistering the async notifier in the case of an error --- the function may not fail anymore after the notifier is registered. Signed-off-by: Sakari Ailus <sakari.ailus@iki.fi> --- drivers/media/platform/omap3isp/isp.c | 19 ++++++++++++------- 1 file changed, 12 insertions(+), 7 deletions(-)