| Message ID | 975645970144b6c8401ab9bd1243619978a868a9.1620207353.git.mchehab+huawei@kernel.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Fix some PM runtime issues at the media subsystem | expand |
On Wed, 5 May 2021 11:41:58 +0200 Mauro Carvalho Chehab <mchehab+huawei@kernel.org> wrote: > There are several issues in the way the atmel driver handles > pm_runtime_get_sync(): > > - it doesn't check return codes; > - it doesn't properly decrement the usage_count on all places; > - it starts streaming even if pm_runtime_get_sync() fails. > - while it tries to get pm_runtime at the clock enable logic, > it doesn't check if the operation was suceeded. > > Replace all occurrences of it to use the new kAPI: > pm_runtime_resume_and_get(), which ensures that, if the > return code is not negative, the usage_count was incremented. > > With that, add additional checks when this is called, in order > to ensure that errors will be properly addressed. > > Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> Hi Mauro, I don't know media enough to know what is the right answer but in some of this series, a failure in pm_runtime_resume_and_get() leads to a bunch of buffer cleanup (patch 22 being an example) and in others return happens without doing that cleanup. It might be both are safe, or I'm missing something else, but I'm certainly not confident enough to give any tags on this one as a result of that mismatch. > --- > drivers/media/platform/atmel/atmel-isc-base.c | 30 ++++++++++++++----- > drivers/media/platform/atmel/atmel-isi.c | 19 +++++++++--- > 2 files changed, 38 insertions(+), 11 deletions(-) > > diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c > index fe3ec8d0eaee..ce8e1351fa53 100644 > --- a/drivers/media/platform/atmel/atmel-isc-base.c > +++ b/drivers/media/platform/atmel/atmel-isc-base.c > @@ -294,9 +294,13 @@ static int isc_wait_clk_stable(struct clk_hw *hw) > static int isc_clk_prepare(struct clk_hw *hw) > { > struct isc_clk *isc_clk = to_isc_clk(hw); > + int ret; > > - if (isc_clk->id == ISC_ISPCK) > - pm_runtime_get_sync(isc_clk->dev); > + if (isc_clk->id == ISC_ISPCK) { > + ret = pm_runtime_resume_and_get(isc_clk->dev); > + if (ret < 0) > + return ret; > + } > > return isc_wait_clk_stable(hw); > } > @@ -353,9 +357,13 @@ static int isc_clk_is_enabled(struct clk_hw *hw) > { > struct isc_clk *isc_clk = to_isc_clk(hw); > u32 status; > + int ret; > > - if (isc_clk->id == ISC_ISPCK) > - pm_runtime_get_sync(isc_clk->dev); > + if (isc_clk->id == ISC_ISPCK) { > + ret = pm_runtime_resume_and_get(isc_clk->dev); > + if (ret < 0) > + return 0; > + } > > regmap_read(isc_clk->regmap, ISC_CLKSR, &status); > > @@ -807,7 +815,12 @@ static int isc_start_streaming(struct vb2_queue *vq, unsigned int count) > goto err_start_stream; > } > > - pm_runtime_get_sync(isc->dev); > + ret = pm_runtime_resume_and_get(isc->dev); > + if (ret < 0) { > + v4l2_err(&isc->v4l2_dev, "RPM resume failed in subdev %d\n", > + ret); > + goto err_pm_get; > + } > > ret = isc_configure(isc); > if (unlikely(ret)) > @@ -838,7 +851,7 @@ static int isc_start_streaming(struct vb2_queue *vq, unsigned int count) > > err_configure: > pm_runtime_put_sync(isc->dev); > - > +err_pm_get: > v4l2_subdev_call(isc->current_subdev->sd, video, s_stream, 0); > > err_start_stream: > @@ -1809,6 +1822,7 @@ static void isc_awb_work(struct work_struct *w) > u32 baysel; > unsigned long flags; > u32 min, max; > + int ret; > > /* streaming is not active anymore */ > if (isc->stop) > @@ -1831,7 +1845,9 @@ static void isc_awb_work(struct work_struct *w) > ctrls->hist_id = hist_id; > baysel = isc->config.sd_format->cfa_baycfg << ISC_HIS_CFG_BAYSEL_SHIFT; > > - pm_runtime_get_sync(isc->dev); > + ret = pm_runtime_resume_and_get(isc->dev); > + if (ret < 0) > + return; > > /* > * only update if we have all the required histograms and controls > diff --git a/drivers/media/platform/atmel/atmel-isi.c b/drivers/media/platform/atmel/atmel-isi.c > index e392b3efe363..5b1dd358f2e6 100644 > --- a/drivers/media/platform/atmel/atmel-isi.c > +++ b/drivers/media/platform/atmel/atmel-isi.c > @@ -422,7 +422,9 @@ static int start_streaming(struct vb2_queue *vq, unsigned int count) > struct frame_buffer *buf, *node; > int ret; > > - pm_runtime_get_sync(isi->dev); > + ret = pm_runtime_resume_and_get(isi->dev); > + if (ret < 0) > + return ret; This is the case I'm referring to above. > > /* Enable stream on the sub device */ > ret = v4l2_subdev_call(isi->entity.subdev, video, s_stream, 1); > @@ -782,9 +784,10 @@ static int isi_enum_frameintervals(struct file *file, void *fh, > return 0; > } > > -static void isi_camera_set_bus_param(struct atmel_isi *isi) > +static int isi_camera_set_bus_param(struct atmel_isi *isi) > { > u32 cfg1 = 0; > + int ret; > > /* set bus param for ISI */ > if (isi->pdata.hsync_act_low) > @@ -801,12 +804,16 @@ static void isi_camera_set_bus_param(struct atmel_isi *isi) > cfg1 |= ISI_CFG1_THMASK_BEATS_16; > > /* Enable PM and peripheral clock before operate isi registers */ > - pm_runtime_get_sync(isi->dev); > + ret = pm_runtime_resume_and_get(isi->dev); > + if (ret < 0) > + return ret; > > isi_writel(isi, ISI_CTRL, ISI_CTRL_DIS); > isi_writel(isi, ISI_CFG1, cfg1); > > pm_runtime_put(isi->dev); > + > + return 0; > } > > /* -----------------------------------------------------------------------*/ > @@ -1085,7 +1092,11 @@ static int isi_graph_notify_complete(struct v4l2_async_notifier *notifier) > dev_err(isi->dev, "No supported mediabus format found\n"); > return ret; > } > - isi_camera_set_bus_param(isi); > + ret = isi_camera_set_bus_param(isi); > + if (ret) { > + dev_err(isi->dev, "Can't wake up device\n"); > + return ret; > + } > > ret = isi_set_default_fmt(isi); > if (ret) {
On 5/5/21 3:08 PM, Jonathan Cameron wrote: > On Wed, 5 May 2021 11:41:58 +0200 > Mauro Carvalho Chehab <mchehab+huawei@kernel.org> wrote: > >> There are several issues in the way the atmel driver handles >> pm_runtime_get_sync(): >> >> - it doesn't check return codes; >> - it doesn't properly decrement the usage_count on all places; >> - it starts streaming even if pm_runtime_get_sync() fails. >> - while it tries to get pm_runtime at the clock enable logic, >> it doesn't check if the operation was suceeded. >> >> Replace all occurrences of it to use the new kAPI: >> pm_runtime_resume_and_get(), which ensures that, if the >> return code is not negative, the usage_count was incremented. >> >> With that, add additional checks when this is called, in order >> to ensure that errors will be properly addressed. >> >> Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> > > Hi Mauro, I don't know media enough to know what is the right answer > but in some of this series, a failure in > pm_runtime_resume_and_get() leads to a bunch of buffer cleanup > (patch 22 being an example) and in others return happens without doing > that cleanup. > > It might be both are safe, or I'm missing something else, but I'm > certainly not confident enough to give any tags on this one as a result > of that mismatch. > >> --- >> drivers/media/platform/atmel/atmel-isc-base.c | 30 ++++++++++++++----- >> drivers/media/platform/atmel/atmel-isi.c | 19 +++++++++--- >> 2 files changed, 38 insertions(+), 11 deletions(-) >> >> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c >> index fe3ec8d0eaee..ce8e1351fa53 100644 >> --- a/drivers/media/platform/atmel/atmel-isc-base.c >> +++ b/drivers/media/platform/atmel/atmel-isc-base.c >> @@ -294,9 +294,13 @@ static int isc_wait_clk_stable(struct clk_hw *hw) >> static int isc_clk_prepare(struct clk_hw *hw) >> { >> struct isc_clk *isc_clk = to_isc_clk(hw); >> + int ret; >> >> - if (isc_clk->id == ISC_ISPCK) >> - pm_runtime_get_sync(isc_clk->dev); >> + if (isc_clk->id == ISC_ISPCK) { >> + ret = pm_runtime_resume_and_get(isc_clk->dev); >> + if (ret < 0) >> + return ret; >> + } Hi Mauro, With this patch, the ISC is broken on latest media tree. It looks like pm_runtime_resume_and_get for the ISC_ISPCK clock returns -ENOACCESS and thus, the probe of the driver fails: atmel-sama5d2-isc f0008000.isc: failed to enable ispck: -13 atmel-sama5d2-isc: probe of f0008000.isc failed with error -13 Could you point out how I could fix this ? Maybe the isc_clk->dev is not properly handled/initialized in some other part of the code ? Thanks ! Eugen >> >> return isc_wait_clk_stable(hw); >> } >> @@ -353,9 +357,13 @@ static int isc_clk_is_enabled(struct clk_hw *hw) >> { >> struct isc_clk *isc_clk = to_isc_clk(hw); >> u32 status; >> + int ret; >> >> - if (isc_clk->id == ISC_ISPCK) >> - pm_runtime_get_sync(isc_clk->dev); >> + if (isc_clk->id == ISC_ISPCK) { >> + ret = pm_runtime_resume_and_get(isc_clk->dev); >> + if (ret < 0) >> + return 0; >> + } >> >> regmap_read(isc_clk->regmap, ISC_CLKSR, &status); >> >> @@ -807,7 +815,12 @@ static int isc_start_streaming(struct vb2_queue *vq, unsigned int count) >> goto err_start_stream; >> } >> >> - pm_runtime_get_sync(isc->dev); >> + ret = pm_runtime_resume_and_get(isc->dev); >> + if (ret < 0) { >> + v4l2_err(&isc->v4l2_dev, "RPM resume failed in subdev %d\n", >> + ret); >> + goto err_pm_get; >> + } >> >> ret = isc_configure(isc); >> if (unlikely(ret)) >> @@ -838,7 +851,7 @@ static int isc_start_streaming(struct vb2_queue *vq, unsigned int count) >> >> err_configure: >> pm_runtime_put_sync(isc->dev); >> - >> +err_pm_get: >> v4l2_subdev_call(isc->current_subdev->sd, video, s_stream, 0); >> >> err_start_stream: >> @@ -1809,6 +1822,7 @@ static void isc_awb_work(struct work_struct *w) >> u32 baysel; >> unsigned long flags; >> u32 min, max; >> + int ret; >> >> /* streaming is not active anymore */ >> if (isc->stop) >> @@ -1831,7 +1845,9 @@ static void isc_awb_work(struct work_struct *w) >> ctrls->hist_id = hist_id; >> baysel = isc->config.sd_format->cfa_baycfg << ISC_HIS_CFG_BAYSEL_SHIFT; >> >> - pm_runtime_get_sync(isc->dev); >> + ret = pm_runtime_resume_and_get(isc->dev); >> + if (ret < 0) >> + return; >> >> /* >> * only update if we have all the required histograms and controls >> diff --git a/drivers/media/platform/atmel/atmel-isi.c b/drivers/media/platform/atmel/atmel-isi.c >> index e392b3efe363..5b1dd358f2e6 100644 >> --- a/drivers/media/platform/atmel/atmel-isi.c >> +++ b/drivers/media/platform/atmel/atmel-isi.c >> @@ -422,7 +422,9 @@ static int start_streaming(struct vb2_queue *vq, unsigned int count) >> struct frame_buffer *buf, *node; >> int ret; >> >> - pm_runtime_get_sync(isi->dev); >> + ret = pm_runtime_resume_and_get(isi->dev); >> + if (ret < 0) >> + return ret; > This is the case I'm referring to above. > >> >> /* Enable stream on the sub device */ >> ret = v4l2_subdev_call(isi->entity.subdev, video, s_stream, 1); >> @@ -782,9 +784,10 @@ static int isi_enum_frameintervals(struct file *file, void *fh, >> return 0; >> } >> >> -static void isi_camera_set_bus_param(struct atmel_isi *isi) >> +static int isi_camera_set_bus_param(struct atmel_isi *isi) >> { >> u32 cfg1 = 0; >> + int ret; >> >> /* set bus param for ISI */ >> if (isi->pdata.hsync_act_low) >> @@ -801,12 +804,16 @@ static void isi_camera_set_bus_param(struct atmel_isi *isi) >> cfg1 |= ISI_CFG1_THMASK_BEATS_16; >> >> /* Enable PM and peripheral clock before operate isi registers */ >> - pm_runtime_get_sync(isi->dev); >> + ret = pm_runtime_resume_and_get(isi->dev); >> + if (ret < 0) >> + return ret; >> >> isi_writel(isi, ISI_CTRL, ISI_CTRL_DIS); >> isi_writel(isi, ISI_CFG1, cfg1); >> >> pm_runtime_put(isi->dev); >> + >> + return 0; >> } >> >> /* -----------------------------------------------------------------------*/ >> @@ -1085,7 +1092,11 @@ static int isi_graph_notify_complete(struct v4l2_async_notifier *notifier) >> dev_err(isi->dev, "No supported mediabus format found\n"); >> return ret; >> } >> - isi_camera_set_bus_param(isi); >> + ret = isi_camera_set_bus_param(isi); >> + if (ret) { >> + dev_err(isi->dev, "Can't wake up device\n"); >> + return ret; >> + } >> >> ret = isi_set_default_fmt(isi); >> if (ret) { >
Em Thu, 10 Jun 2021 09:04:07 +0000 <Eugen.Hristev@microchip.com> escreveu: > >> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c > >> index fe3ec8d0eaee..ce8e1351fa53 100644 > >> --- a/drivers/media/platform/atmel/atmel-isc-base.c > >> +++ b/drivers/media/platform/atmel/atmel-isc-base.c > >> @@ -294,9 +294,13 @@ static int isc_wait_clk_stable(struct clk_hw *hw) > >> static int isc_clk_prepare(struct clk_hw *hw) > >> { > >> struct isc_clk *isc_clk = to_isc_clk(hw); > >> + int ret; > >> > >> - if (isc_clk->id == ISC_ISPCK) > >> - pm_runtime_get_sync(isc_clk->dev); > >> + if (isc_clk->id == ISC_ISPCK) { > >> + ret = pm_runtime_resume_and_get(isc_clk->dev); > >> + if (ret < 0) > >> + return ret; > >> + } > > Hi Mauro, > > With this patch, the ISC is broken on latest media tree. It looks like > pm_runtime_resume_and_get for the ISC_ISPCK clock returns -ENOACCESS and > thus, the probe of the driver fails: > > atmel-sama5d2-isc f0008000.isc: failed to enable ispck: -13 > atmel-sama5d2-isc: probe of f0008000.isc failed with error -13 > > > Could you point out how I could fix this ? Maybe the isc_clk->dev is not > properly handled/initialized in some other part of the code ? Looking at RPM implementation: static int rpm_resume(struct device *dev, int rpmflags) { ... if (dev->power.runtime_error) retval = -EINVAL; else if (dev->power.disable_depth == 1 && dev->power.is_suspended && dev->power.runtime_status == RPM_ACTIVE) retval = 1; else if (dev->power.disable_depth > 0) retval = -EACCES; ... It sounds that RPM is disabled for this clock. Did you call pm_runtime_enable() before calling isc_clk_prepare()? Thanks, Mauro
On 6/10/21 12:38 PM, Mauro Carvalho Chehab wrote: > EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe > > Em Thu, 10 Jun 2021 09:04:07 +0000 > <Eugen.Hristev@microchip.com> escreveu: > >>>> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c >>>> index fe3ec8d0eaee..ce8e1351fa53 100644 >>>> --- a/drivers/media/platform/atmel/atmel-isc-base.c >>>> +++ b/drivers/media/platform/atmel/atmel-isc-base.c >>>> @@ -294,9 +294,13 @@ static int isc_wait_clk_stable(struct clk_hw *hw) >>>> static int isc_clk_prepare(struct clk_hw *hw) >>>> { >>>> struct isc_clk *isc_clk = to_isc_clk(hw); >>>> + int ret; >>>> >>>> - if (isc_clk->id == ISC_ISPCK) >>>> - pm_runtime_get_sync(isc_clk->dev); >>>> + if (isc_clk->id == ISC_ISPCK) { >>>> + ret = pm_runtime_resume_and_get(isc_clk->dev); >>>> + if (ret < 0) >>>> + return ret; >>>> + } >> >> Hi Mauro, >> >> With this patch, the ISC is broken on latest media tree. It looks like >> pm_runtime_resume_and_get for the ISC_ISPCK clock returns -ENOACCESS and >> thus, the probe of the driver fails: >> >> atmel-sama5d2-isc f0008000.isc: failed to enable ispck: -13 >> atmel-sama5d2-isc: probe of f0008000.isc failed with error -13 >> >> >> Could you point out how I could fix this ? Maybe the isc_clk->dev is not >> properly handled/initialized in some other part of the code ? > > Looking at RPM implementation: > > static int rpm_resume(struct device *dev, int rpmflags) > { > ... > if (dev->power.runtime_error) > retval = -EINVAL; > else if (dev->power.disable_depth == 1 && dev->power.is_suspended > && dev->power.runtime_status == RPM_ACTIVE) > retval = 1; > else if (dev->power.disable_depth > 0) > retval = -EACCES; > ... > > It sounds that RPM is disabled for this clock. Did you call > pm_runtime_enable() before calling isc_clk_prepare()? > > Thanks, > Mauro > I tried to call pm_runtime_enable for the clk at clk_init time, but doing that makes the runtime for the ISC fail like this: atmel-sama5d2-isc f0008000.isc: Unbalanced pm_runtime_enable!
Em Thu, 10 Jun 2021 12:00:42 +0000 <Eugen.Hristev@microchip.com> escreveu: > On 6/10/21 12:38 PM, Mauro Carvalho Chehab wrote: > > EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe > > > > Em Thu, 10 Jun 2021 09:04:07 +0000 > > <Eugen.Hristev@microchip.com> escreveu: > > > >>>> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c > >>>> index fe3ec8d0eaee..ce8e1351fa53 100644 > >>>> --- a/drivers/media/platform/atmel/atmel-isc-base.c > >>>> +++ b/drivers/media/platform/atmel/atmel-isc-base.c > >>>> @@ -294,9 +294,13 @@ static int isc_wait_clk_stable(struct clk_hw *hw) > >>>> static int isc_clk_prepare(struct clk_hw *hw) > >>>> { > >>>> struct isc_clk *isc_clk = to_isc_clk(hw); > >>>> + int ret; > >>>> > >>>> - if (isc_clk->id == ISC_ISPCK) > >>>> - pm_runtime_get_sync(isc_clk->dev); > >>>> + if (isc_clk->id == ISC_ISPCK) { > >>>> + ret = pm_runtime_resume_and_get(isc_clk->dev); > >>>> + if (ret < 0) > >>>> + return ret; > >>>> + } > >> > >> Hi Mauro, > >> > >> With this patch, the ISC is broken on latest media tree. It looks like > >> pm_runtime_resume_and_get for the ISC_ISPCK clock returns -ENOACCESS and > >> thus, the probe of the driver fails: > >> > >> atmel-sama5d2-isc f0008000.isc: failed to enable ispck: -13 > >> atmel-sama5d2-isc: probe of f0008000.isc failed with error -13 > >> > >> > >> Could you point out how I could fix this ? Maybe the isc_clk->dev is not > >> properly handled/initialized in some other part of the code ? > > > > Looking at RPM implementation: > > > > static int rpm_resume(struct device *dev, int rpmflags) > > { > > ... > > if (dev->power.runtime_error) > > retval = -EINVAL; > > else if (dev->power.disable_depth == 1 && dev->power.is_suspended > > && dev->power.runtime_status == RPM_ACTIVE) > > retval = 1; > > else if (dev->power.disable_depth > 0) > > retval = -EACCES; > > ... > > > > It sounds that RPM is disabled for this clock. Did you call > > pm_runtime_enable() before calling isc_clk_prepare()? > > > > Thanks, > > Mauro > > > > I tried to call pm_runtime_enable for the clk at clk_init time, but > doing that makes the runtime for the ISC fail like this: > > atmel-sama5d2-isc f0008000.isc: Unbalanced pm_runtime_enable! Making RPM balanced is complex ;-) Yet, clearly there's something weird at the PM code. I mean, ignoring a -ENOACCESS error like the original code sounds wrong, as it basically means that pm_runtime_get_sync() were doing nothing (except by incrementing a refcount). Some drivers call clk_prepare()/clk_unprepare() at their .prepare/.unprepare ops. Those functions internally call RPM get logic recursively, to ensure that the RPM state of the parents are also at the right state. Thanks, Mauro
Hi Eugen, Em Thu, 10 Jun 2021 12:00:42 +0000 <Eugen.Hristev@microchip.com> escreveu: > On 6/10/21 12:38 PM, Mauro Carvalho Chehab wrote: > > EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe > > > > Em Thu, 10 Jun 2021 09:04:07 +0000 > > <Eugen.Hristev@microchip.com> escreveu: > > > >>>> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c > >>>> index fe3ec8d0eaee..ce8e1351fa53 100644 > >>>> --- a/drivers/media/platform/atmel/atmel-isc-base.c > >>>> +++ b/drivers/media/platform/atmel/atmel-isc-base.c > >>>> @@ -294,9 +294,13 @@ static int isc_wait_clk_stable(struct clk_hw *hw) > >>>> static int isc_clk_prepare(struct clk_hw *hw) > >>>> { > >>>> struct isc_clk *isc_clk = to_isc_clk(hw); > >>>> + int ret; > >>>> > >>>> - if (isc_clk->id == ISC_ISPCK) > >>>> - pm_runtime_get_sync(isc_clk->dev); > >>>> + if (isc_clk->id == ISC_ISPCK) { > >>>> + ret = pm_runtime_resume_and_get(isc_clk->dev); > >>>> + if (ret < 0) > >>>> + return ret; > >>>> + } > >> > >> Hi Mauro, > >> > >> With this patch, the ISC is broken on latest media tree. It looks like > >> pm_runtime_resume_and_get for the ISC_ISPCK clock returns -ENOACCESS and > >> thus, the probe of the driver fails: > >> > >> atmel-sama5d2-isc f0008000.isc: failed to enable ispck: -13 > >> atmel-sama5d2-isc: probe of f0008000.isc failed with error -13 What's the current status of this issue? If the bug still happens, we need a fix for it. We might revert this patch, but this would just be masking some other hidden bug. Regards, Mauro > >> > >> > >> Could you point out how I could fix this ? Maybe the isc_clk->dev is not > >> properly handled/initialized in some other part of the code ? > > > > Looking at RPM implementation: > > > > static int rpm_resume(struct device *dev, int rpmflags) > > { > > ... > > if (dev->power.runtime_error) > > retval = -EINVAL; > > else if (dev->power.disable_depth == 1 && dev->power.is_suspended > > && dev->power.runtime_status == RPM_ACTIVE) > > retval = 1; > > else if (dev->power.disable_depth > 0) > > retval = -EACCES; > > ... > > > > It sounds that RPM is disabled for this clock. Did you call > > pm_runtime_enable() before calling isc_clk_prepare()? > > > > Thanks, > > Mauro > > > > I tried to call pm_runtime_enable for the clk at clk_init time, but > doing that makes the runtime for the ISC fail like this: > > atmel-sama5d2-isc f0008000.isc: Unbalanced pm_runtime_enable! Thanks, Mauro
On 9/6/21 11:03 AM, Mauro Carvalho Chehab wrote: > Hi Eugen, > > Em Thu, 10 Jun 2021 12:00:42 +0000 > <Eugen.Hristev@microchip.com> escreveu: > >> On 6/10/21 12:38 PM, Mauro Carvalho Chehab wrote: >>> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe >>> >>> Em Thu, 10 Jun 2021 09:04:07 +0000 >>> <Eugen.Hristev@microchip.com> escreveu: >>> >>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c >>>>>> index fe3ec8d0eaee..ce8e1351fa53 100644 >>>>>> --- a/drivers/media/platform/atmel/atmel-isc-base.c >>>>>> +++ b/drivers/media/platform/atmel/atmel-isc-base.c >>>>>> @@ -294,9 +294,13 @@ static int isc_wait_clk_stable(struct clk_hw *hw) >>>>>> static int isc_clk_prepare(struct clk_hw *hw) >>>>>> { >>>>>> struct isc_clk *isc_clk = to_isc_clk(hw); >>>>>> + int ret; >>>>>> >>>>>> - if (isc_clk->id == ISC_ISPCK) >>>>>> - pm_runtime_get_sync(isc_clk->dev); >>>>>> + if (isc_clk->id == ISC_ISPCK) { >>>>>> + ret = pm_runtime_resume_and_get(isc_clk->dev); >>>>>> + if (ret < 0) >>>>>> + return ret; >>>>>> + } >>>> >>>> Hi Mauro, >>>> >>>> With this patch, the ISC is broken on latest media tree. It looks like >>>> pm_runtime_resume_and_get for the ISC_ISPCK clock returns -ENOACCESS and >>>> thus, the probe of the driver fails: >>>> >>>> atmel-sama5d2-isc f0008000.isc: failed to enable ispck: -13 >>>> atmel-sama5d2-isc: probe of f0008000.isc failed with error -13 > > What's the current status of this issue? Hi Mauro, Currently, as far as I know, the ISC is broken in 5.14 and current Linux master. I plan to investigate this issue this week (or the next...), together with some other things. I want to tryout the PM part of the driver to see where is the problem. When I come up with a fix or patch, I will send it on the mailing list. If you have any ideas that I can try meanwhile, feel free to contact me to send a patch that I can test. Eugen > > If the bug still happens, we need a fix for it. > > We might revert this patch, but this would just be masking some other > hidden bug. > > Regards, > Mauro > > > >>>> >>>> >>>> Could you point out how I could fix this ? Maybe the isc_clk->dev is not >>>> properly handled/initialized in some other part of the code ? >>> >>> Looking at RPM implementation: >>> >>> static int rpm_resume(struct device *dev, int rpmflags) >>> { >>> ... >>> if (dev->power.runtime_error) >>> retval = -EINVAL; >>> else if (dev->power.disable_depth == 1 && dev->power.is_suspended >>> && dev->power.runtime_status == RPM_ACTIVE) >>> retval = 1; >>> else if (dev->power.disable_depth > 0) >>> retval = -EACCES; >>> ... >>> >>> It sounds that RPM is disabled for this clock. Did you call >>> pm_runtime_enable() before calling isc_clk_prepare()? >>> >>> Thanks, >>> Mauro >>> >> >> I tried to call pm_runtime_enable for the clk at clk_init time, but >> doing that makes the runtime for the ISC fail like this: >> >> atmel-sama5d2-isc f0008000.isc: Unbalanced pm_runtime_enable! > > > > Thanks, > Mauro >
On 9/6/21 11:13 AM, Eugen Hristev - M18282 wrote: > On 9/6/21 11:03 AM, Mauro Carvalho Chehab wrote: >> Hi Eugen, >> >> Em Thu, 10 Jun 2021 12:00:42 +0000 >> <Eugen.Hristev@microchip.com> escreveu: >> >>> On 6/10/21 12:38 PM, Mauro Carvalho Chehab wrote: >>>> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe >>>> >>>> Em Thu, 10 Jun 2021 09:04:07 +0000 >>>> <Eugen.Hristev@microchip.com> escreveu: >>>> >>>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c >>>>>>> index fe3ec8d0eaee..ce8e1351fa53 100644 >>>>>>> --- a/drivers/media/platform/atmel/atmel-isc-base.c >>>>>>> +++ b/drivers/media/platform/atmel/atmel-isc-base.c >>>>>>> @@ -294,9 +294,13 @@ static int isc_wait_clk_stable(struct clk_hw *hw) >>>>>>> static int isc_clk_prepare(struct clk_hw *hw) >>>>>>> { >>>>>>> struct isc_clk *isc_clk = to_isc_clk(hw); >>>>>>> + int ret; >>>>>>> >>>>>>> - if (isc_clk->id == ISC_ISPCK) >>>>>>> - pm_runtime_get_sync(isc_clk->dev); >>>>>>> + if (isc_clk->id == ISC_ISPCK) { >>>>>>> + ret = pm_runtime_resume_and_get(isc_clk->dev); >>>>>>> + if (ret < 0) >>>>>>> + return ret; >>>>>>> + } >>>>> >>>>> Hi Mauro, >>>>> >>>>> With this patch, the ISC is broken on latest media tree. It looks like >>>>> pm_runtime_resume_and_get for the ISC_ISPCK clock returns -ENOACCESS and >>>>> thus, the probe of the driver fails: >>>>> >>>>> atmel-sama5d2-isc f0008000.isc: failed to enable ispck: -13 >>>>> atmel-sama5d2-isc: probe of f0008000.isc failed with error -13 >> >> What's the current status of this issue? > > Hi Mauro, > > Currently, as far as I know, the ISC is broken in 5.14 and current Linux > master. I plan to investigate this issue this week (or the next...), > together with some other things. I want to tryout the PM part of the > driver to see where is the problem. > When I come up with a fix or patch, I will send it on the mailing list. > If you have any ideas that I can try meanwhile, feel free to contact me > to send a patch that I can test. > > Hi Mauro, Regarding this issue, I sent a patch on the ML that should solve it. https://patchwork.linuxtv.org/project/linux-media/patch/20210913102254.108638-1-eugen.hristev@microchip.com/ Eugen > Eugen > > >> >> If the bug still happens, we need a fix for it. >> >> We might revert this patch, but this would just be masking some other >> hidden bug. >> >> Regards, >> Mauro >> >> >> >>>>> >>>>> >>>>> Could you point out how I could fix this ? Maybe the isc_clk->dev is not >>>>> properly handled/initialized in some other part of the code ? >>>> >>>> Looking at RPM implementation: >>>> >>>> static int rpm_resume(struct device *dev, int rpmflags) >>>> { >>>> ... >>>> if (dev->power.runtime_error) >>>> retval = -EINVAL; >>>> else if (dev->power.disable_depth == 1 && dev->power.is_suspended >>>> && dev->power.runtime_status == RPM_ACTIVE) >>>> retval = 1; >>>> else if (dev->power.disable_depth > 0) >>>> retval = -EACCES; >>>> ... >>>> >>>> It sounds that RPM is disabled for this clock. Did you call >>>> pm_runtime_enable() before calling isc_clk_prepare()? >>>> >>>> Thanks, >>>> Mauro >>>> >>> >>> I tried to call pm_runtime_enable for the clk at clk_init time, but >>> doing that makes the runtime for the ISC fail like this: >>> >>> atmel-sama5d2-isc f0008000.isc: Unbalanced pm_runtime_enable! >> >> >> >> Thanks, >> Mauro >> >
diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c index fe3ec8d0eaee..ce8e1351fa53 100644 --- a/drivers/media/platform/atmel/atmel-isc-base.c +++ b/drivers/media/platform/atmel/atmel-isc-base.c @@ -294,9 +294,13 @@ static int isc_wait_clk_stable(struct clk_hw *hw) static int isc_clk_prepare(struct clk_hw *hw) { struct isc_clk *isc_clk = to_isc_clk(hw); + int ret; - if (isc_clk->id == ISC_ISPCK) - pm_runtime_get_sync(isc_clk->dev); + if (isc_clk->id == ISC_ISPCK) { + ret = pm_runtime_resume_and_get(isc_clk->dev); + if (ret < 0) + return ret; + } return isc_wait_clk_stable(hw); } @@ -353,9 +357,13 @@ static int isc_clk_is_enabled(struct clk_hw *hw) { struct isc_clk *isc_clk = to_isc_clk(hw); u32 status; + int ret; - if (isc_clk->id == ISC_ISPCK) - pm_runtime_get_sync(isc_clk->dev); + if (isc_clk->id == ISC_ISPCK) { + ret = pm_runtime_resume_and_get(isc_clk->dev); + if (ret < 0) + return 0; + } regmap_read(isc_clk->regmap, ISC_CLKSR, &status); @@ -807,7 +815,12 @@ static int isc_start_streaming(struct vb2_queue *vq, unsigned int count) goto err_start_stream; } - pm_runtime_get_sync(isc->dev); + ret = pm_runtime_resume_and_get(isc->dev); + if (ret < 0) { + v4l2_err(&isc->v4l2_dev, "RPM resume failed in subdev %d\n", + ret); + goto err_pm_get; + } ret = isc_configure(isc); if (unlikely(ret)) @@ -838,7 +851,7 @@ static int isc_start_streaming(struct vb2_queue *vq, unsigned int count) err_configure: pm_runtime_put_sync(isc->dev); - +err_pm_get: v4l2_subdev_call(isc->current_subdev->sd, video, s_stream, 0); err_start_stream: @@ -1809,6 +1822,7 @@ static void isc_awb_work(struct work_struct *w) u32 baysel; unsigned long flags; u32 min, max; + int ret; /* streaming is not active anymore */ if (isc->stop) @@ -1831,7 +1845,9 @@ static void isc_awb_work(struct work_struct *w) ctrls->hist_id = hist_id; baysel = isc->config.sd_format->cfa_baycfg << ISC_HIS_CFG_BAYSEL_SHIFT; - pm_runtime_get_sync(isc->dev); + ret = pm_runtime_resume_and_get(isc->dev); + if (ret < 0) + return; /* * only update if we have all the required histograms and controls diff --git a/drivers/media/platform/atmel/atmel-isi.c b/drivers/media/platform/atmel/atmel-isi.c index e392b3efe363..5b1dd358f2e6 100644 --- a/drivers/media/platform/atmel/atmel-isi.c +++ b/drivers/media/platform/atmel/atmel-isi.c @@ -422,7 +422,9 @@ static int start_streaming(struct vb2_queue *vq, unsigned int count) struct frame_buffer *buf, *node; int ret; - pm_runtime_get_sync(isi->dev); + ret = pm_runtime_resume_and_get(isi->dev); + if (ret < 0) + return ret; /* Enable stream on the sub device */ ret = v4l2_subdev_call(isi->entity.subdev, video, s_stream, 1); @@ -782,9 +784,10 @@ static int isi_enum_frameintervals(struct file *file, void *fh, return 0; } -static void isi_camera_set_bus_param(struct atmel_isi *isi) +static int isi_camera_set_bus_param(struct atmel_isi *isi) { u32 cfg1 = 0; + int ret; /* set bus param for ISI */ if (isi->pdata.hsync_act_low) @@ -801,12 +804,16 @@ static void isi_camera_set_bus_param(struct atmel_isi *isi) cfg1 |= ISI_CFG1_THMASK_BEATS_16; /* Enable PM and peripheral clock before operate isi registers */ - pm_runtime_get_sync(isi->dev); + ret = pm_runtime_resume_and_get(isi->dev); + if (ret < 0) + return ret; isi_writel(isi, ISI_CTRL, ISI_CTRL_DIS); isi_writel(isi, ISI_CFG1, cfg1); pm_runtime_put(isi->dev); + + return 0; } /* -----------------------------------------------------------------------*/ @@ -1085,7 +1092,11 @@ static int isi_graph_notify_complete(struct v4l2_async_notifier *notifier) dev_err(isi->dev, "No supported mediabus format found\n"); return ret; } - isi_camera_set_bus_param(isi); + ret = isi_camera_set_bus_param(isi); + if (ret) { + dev_err(isi->dev, "Can't wake up device\n"); + return ret; + } ret = isi_set_default_fmt(isi); if (ret) {
There are several issues in the way the atmel driver handles pm_runtime_get_sync(): - it doesn't check return codes; - it doesn't properly decrement the usage_count on all places; - it starts streaming even if pm_runtime_get_sync() fails. - while it tries to get pm_runtime at the clock enable logic, it doesn't check if the operation was suceeded. Replace all occurrences of it to use the new kAPI: pm_runtime_resume_and_get(), which ensures that, if the return code is not negative, the usage_count was incremented. With that, add additional checks when this is called, in order to ensure that errors will be properly addressed. Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> --- drivers/media/platform/atmel/atmel-isc-base.c | 30 ++++++++++++++----- drivers/media/platform/atmel/atmel-isi.c | 19 +++++++++--- 2 files changed, 38 insertions(+), 11 deletions(-)