diff mbox series

[v4,01/10] soc: qcom: cpr3: Fix 'acc_desc' usage

Message ID 20240703091651.2820236-2-quic_varada@quicinc.com (mailing list archive)
State Handled Elsewhere, archived
Headers show
Series Enable CPR for IPQ9574 | expand

Commit Message

Varadarajan Narayanan July 3, 2024, 9:16 a.m. UTC
cpr3 code assumes that 'acc_desc' is available for SoCs
implementing CPR version 4 or less. However, IPQ9574 SoC
implements CPRv4 without ACC. This causes NULL pointer accesses
resulting in crashes. Hence, check if 'acc_desc' is populated
before using it.

Signed-off-by: Varadarajan Narayanan <quic_varada@quicinc.com>
---
v4: Undo the acc_desc validation in probe function as that could
    affect other SoC.
---
 drivers/pmdomain/qcom/cpr3.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

Comments

Dmitry Baryshkov July 3, 2024, 10:46 a.m. UTC | #1
On Wed, Jul 03, 2024 at 02:46:42PM GMT, Varadarajan Narayanan wrote:
> cpr3 code assumes that 'acc_desc' is available for SoCs
> implementing CPR version 4 or less. However, IPQ9574 SoC
> implements CPRv4 without ACC. This causes NULL pointer accesses
> resulting in crashes. Hence, check if 'acc_desc' is populated
> before using it.
> 
> Signed-off-by: Varadarajan Narayanan <quic_varada@quicinc.com>
> ---
> v4: Undo the acc_desc validation in probe function as that could
>     affect other SoC.
> ---
>  drivers/pmdomain/qcom/cpr3.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/pmdomain/qcom/cpr3.c b/drivers/pmdomain/qcom/cpr3.c
> index c7790a71e74f..6ceb7605f84d 100644
> --- a/drivers/pmdomain/qcom/cpr3.c
> +++ b/drivers/pmdomain/qcom/cpr3.c
> @@ -2399,12 +2399,12 @@ static int cpr_pd_attach_dev(struct generic_pm_domain *domain,
>  		if (ret)
>  			goto exit;
>  
> -		if (acc_desc->config)
> +		if (acc_desc && acc_desc->config)
>  			regmap_multi_reg_write(drv->tcsr, acc_desc->config,
>  					       acc_desc->num_regs_per_fuse);
>  
>  		/* Enable ACC if required */
> -		if (acc_desc->enable_mask)
> +		if (acc_desc && acc_desc->enable_mask)
>  			regmap_update_bits(drv->tcsr, acc_desc->enable_reg,
>  					   acc_desc->enable_mask,
>  					   acc_desc->enable_mask);

Should the same fix be applied to other places which access acc_desc?
For example cpr_pre_voltage() and cpr_post_voltage() which call
cpr_set_acc()?
Varadarajan Narayanan July 4, 2024, 5 a.m. UTC | #2
On Wed, Jul 03, 2024 at 01:46:54PM +0300, Dmitry Baryshkov wrote:
> On Wed, Jul 03, 2024 at 02:46:42PM GMT, Varadarajan Narayanan wrote:
> > cpr3 code assumes that 'acc_desc' is available for SoCs
> > implementing CPR version 4 or less. However, IPQ9574 SoC
> > implements CPRv4 without ACC. This causes NULL pointer accesses
> > resulting in crashes. Hence, check if 'acc_desc' is populated
> > before using it.
> >
> > Signed-off-by: Varadarajan Narayanan <quic_varada@quicinc.com>
> > ---
> > v4: Undo the acc_desc validation in probe function as that could
> >     affect other SoC.
> > ---
> >  drivers/pmdomain/qcom/cpr3.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/pmdomain/qcom/cpr3.c b/drivers/pmdomain/qcom/cpr3.c
> > index c7790a71e74f..6ceb7605f84d 100644
> > --- a/drivers/pmdomain/qcom/cpr3.c
> > +++ b/drivers/pmdomain/qcom/cpr3.c
> > @@ -2399,12 +2399,12 @@ static int cpr_pd_attach_dev(struct generic_pm_domain *domain,
> >  		if (ret)
> >  			goto exit;
> >
> > -		if (acc_desc->config)
> > +		if (acc_desc && acc_desc->config)
> >  			regmap_multi_reg_write(drv->tcsr, acc_desc->config,
> >  					       acc_desc->num_regs_per_fuse);
> >
> >  		/* Enable ACC if required */
> > -		if (acc_desc->enable_mask)
> > +		if (acc_desc && acc_desc->enable_mask)
> >  			regmap_update_bits(drv->tcsr, acc_desc->enable_reg,
> >  					   acc_desc->enable_mask,
> >  					   acc_desc->enable_mask);
>
> Should the same fix be applied to other places which access acc_desc?
> For example cpr_pre_voltage() and cpr_post_voltage() which call
> cpr_set_acc()?

With this patch alone, if acc_desc is NULL, cpr_probe() will fail
at the start itself because of this check

	if (!data->acc_desc && desc->cpr_type < CTRL_TYPE_CPRH)
		return -EINVAL;

After applying this patch series, cpr_probe will cross the above
check to accomodate IPQ9574. However, the check below will ensure
drv->tcsr is not initialized.

	if (desc->cpr_type < CTRL_TYPE_CPRH &&
	    !of_device_is_compatible(dev->of_node, "qcom,ipq9574-cpr4"))

cpr_pre_voltage() and cpr_post_voltage() call cpr_set_acc() only
if drv->tcsr is not NULL. Hence acc_desc need not be checked.

Will add the check to cpr_pre_voltage() and cpr_post_voltage() if
you feel it will make it more robust regardless of the changes to
cpr_probe in future. Please let me know.

Thanks
Varada
Dmitry Baryshkov July 5, 2024, 3:16 p.m. UTC | #3
On Thu, Jul 04, 2024 at 10:30:50AM GMT, Varadarajan Narayanan wrote:
> On Wed, Jul 03, 2024 at 01:46:54PM +0300, Dmitry Baryshkov wrote:
> > On Wed, Jul 03, 2024 at 02:46:42PM GMT, Varadarajan Narayanan wrote:
> > > cpr3 code assumes that 'acc_desc' is available for SoCs
> > > implementing CPR version 4 or less. However, IPQ9574 SoC
> > > implements CPRv4 without ACC. This causes NULL pointer accesses
> > > resulting in crashes. Hence, check if 'acc_desc' is populated
> > > before using it.
> > >
> > > Signed-off-by: Varadarajan Narayanan <quic_varada@quicinc.com>
> > > ---
> > > v4: Undo the acc_desc validation in probe function as that could
> > >     affect other SoC.
> > > ---
> > >  drivers/pmdomain/qcom/cpr3.c | 4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/pmdomain/qcom/cpr3.c b/drivers/pmdomain/qcom/cpr3.c
> > > index c7790a71e74f..6ceb7605f84d 100644
> > > --- a/drivers/pmdomain/qcom/cpr3.c
> > > +++ b/drivers/pmdomain/qcom/cpr3.c
> > > @@ -2399,12 +2399,12 @@ static int cpr_pd_attach_dev(struct generic_pm_domain *domain,
> > >  		if (ret)
> > >  			goto exit;
> > >
> > > -		if (acc_desc->config)
> > > +		if (acc_desc && acc_desc->config)
> > >  			regmap_multi_reg_write(drv->tcsr, acc_desc->config,
> > >  					       acc_desc->num_regs_per_fuse);
> > >
> > >  		/* Enable ACC if required */
> > > -		if (acc_desc->enable_mask)
> > > +		if (acc_desc && acc_desc->enable_mask)
> > >  			regmap_update_bits(drv->tcsr, acc_desc->enable_reg,
> > >  					   acc_desc->enable_mask,
> > >  					   acc_desc->enable_mask);
> >
> > Should the same fix be applied to other places which access acc_desc?
> > For example cpr_pre_voltage() and cpr_post_voltage() which call
> > cpr_set_acc()?
> 
> With this patch alone, if acc_desc is NULL, cpr_probe() will fail
> at the start itself because of this check
> 
> 	if (!data->acc_desc && desc->cpr_type < CTRL_TYPE_CPRH)
> 		return -EINVAL;
> 
> After applying this patch series, cpr_probe will cross the above
> check to accomodate IPQ9574. However, the check below will ensure
> drv->tcsr is not initialized.
> 
> 	if (desc->cpr_type < CTRL_TYPE_CPRH &&
> 	    !of_device_is_compatible(dev->of_node, "qcom,ipq9574-cpr4"))
> 
> cpr_pre_voltage() and cpr_post_voltage() call cpr_set_acc() only
> if drv->tcsr is not NULL. Hence acc_desc need not be checked.
> 
> Will add the check to cpr_pre_voltage() and cpr_post_voltage() if
> you feel it will make it more robust regardless of the changes to
> cpr_probe in future. Please let me know.

Having !acc_desc check instead of the of_device_is_compatible would
solve the issue.
Varadarajan Narayanan July 9, 2024, 8:29 a.m. UTC | #4
On Fri, Jul 05, 2024 at 06:16:51PM +0300, Dmitry Baryshkov wrote:
> On Thu, Jul 04, 2024 at 10:30:50AM GMT, Varadarajan Narayanan wrote:
> > On Wed, Jul 03, 2024 at 01:46:54PM +0300, Dmitry Baryshkov wrote:
> > > On Wed, Jul 03, 2024 at 02:46:42PM GMT, Varadarajan Narayanan wrote:
> > > > cpr3 code assumes that 'acc_desc' is available for SoCs
> > > > implementing CPR version 4 or less. However, IPQ9574 SoC
> > > > implements CPRv4 without ACC. This causes NULL pointer accesses
> > > > resulting in crashes. Hence, check if 'acc_desc' is populated
> > > > before using it.
> > > >
> > > > Signed-off-by: Varadarajan Narayanan <quic_varada@quicinc.com>
> > > > ---
> > > > v4: Undo the acc_desc validation in probe function as that could
> > > >     affect other SoC.
> > > > ---
> > > >  drivers/pmdomain/qcom/cpr3.c | 4 ++--
> > > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > >
> > > > diff --git a/drivers/pmdomain/qcom/cpr3.c b/drivers/pmdomain/qcom/cpr3.c
> > > > index c7790a71e74f..6ceb7605f84d 100644
> > > > --- a/drivers/pmdomain/qcom/cpr3.c
> > > > +++ b/drivers/pmdomain/qcom/cpr3.c
> > > > @@ -2399,12 +2399,12 @@ static int cpr_pd_attach_dev(struct generic_pm_domain *domain,
> > > >  		if (ret)
> > > >  			goto exit;
> > > >
> > > > -		if (acc_desc->config)
> > > > +		if (acc_desc && acc_desc->config)
> > > >  			regmap_multi_reg_write(drv->tcsr, acc_desc->config,
> > > >  					       acc_desc->num_regs_per_fuse);
> > > >
> > > >  		/* Enable ACC if required */
> > > > -		if (acc_desc->enable_mask)
> > > > +		if (acc_desc && acc_desc->enable_mask)
> > > >  			regmap_update_bits(drv->tcsr, acc_desc->enable_reg,
> > > >  					   acc_desc->enable_mask,
> > > >  					   acc_desc->enable_mask);
> > >
> > > Should the same fix be applied to other places which access acc_desc?
> > > For example cpr_pre_voltage() and cpr_post_voltage() which call
> > > cpr_set_acc()?
> >
> > With this patch alone, if acc_desc is NULL, cpr_probe() will fail
> > at the start itself because of this check
> >
> > 	if (!data->acc_desc && desc->cpr_type < CTRL_TYPE_CPRH)
> > 		return -EINVAL;
> >
> > After applying this patch series, cpr_probe will cross the above
> > check to accomodate IPQ9574. However, the check below will ensure
> > drv->tcsr is not initialized.
> >
> > 	if (desc->cpr_type < CTRL_TYPE_CPRH &&
> > 	    !of_device_is_compatible(dev->of_node, "qcom,ipq9574-cpr4"))
> >
> > cpr_pre_voltage() and cpr_post_voltage() call cpr_set_acc() only
> > if drv->tcsr is not NULL. Hence acc_desc need not be checked.
> >
> > Will add the check to cpr_pre_voltage() and cpr_post_voltage() if
> > you feel it will make it more robust regardless of the changes to
> > cpr_probe in future. Please let me know.
>
> Having !acc_desc check instead of the of_device_is_compatible would
> solve the issue.

Ok. Will post next version with that.

Thanks
Varada
diff mbox series

Patch

diff --git a/drivers/pmdomain/qcom/cpr3.c b/drivers/pmdomain/qcom/cpr3.c
index c7790a71e74f..6ceb7605f84d 100644
--- a/drivers/pmdomain/qcom/cpr3.c
+++ b/drivers/pmdomain/qcom/cpr3.c
@@ -2399,12 +2399,12 @@  static int cpr_pd_attach_dev(struct generic_pm_domain *domain,
 		if (ret)
 			goto exit;
 
-		if (acc_desc->config)
+		if (acc_desc && acc_desc->config)
 			regmap_multi_reg_write(drv->tcsr, acc_desc->config,
 					       acc_desc->num_regs_per_fuse);
 
 		/* Enable ACC if required */
-		if (acc_desc->enable_mask)
+		if (acc_desc && acc_desc->enable_mask)
 			regmap_update_bits(drv->tcsr, acc_desc->enable_reg,
 					   acc_desc->enable_mask,
 					   acc_desc->enable_mask);