[PATCH] ArmPkg/ArmScmiDxe: Fix the calculation of RequiredArraySize in ClockDescribeRates()

[PATCH] ArmPkg/ArmScmiDxe: Fix the calculation of RequiredArraySize in ClockDescribeRates()

[Sudeep Holla]

As per the SCMI specification, section CLOCK_DESCRIBE_RATES mentions that
the value of num_rates_flags[11:0] in the response must be 3 if the return
format is the triplet. Due to the buggy firmware, this was not noticed for
long time. The firmware is now fixed resulting in ClockDescribeRates() to
fail with "Buffer Too Small" error as the RequiredArraySize gets miscalculated
as 72 instead of 24.

Fix the issue by reusing the logic for both the return format which must
work if num_rates_flags has correct value as expected from the specification.

Cc: Pierre Gondois <Pierre.Gondois@arm.com>
Cc: Girish Pathak <girish.pathak@arm.com>
Cc: Jeff Brasen <jbrasen@nvidia.com>
Reported-by: Sami Mujawar <sami.mujawar@arm.com>
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
---
 ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c | 7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)

Hi All,

Sorry for the resend, I hadn't subscribed before so resending after the
subscription.

Regards,
Sudeep

diff --git a/ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c b/ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c
index 12a7e6df5d43..1412717bc227 100644
--- a/ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c
+++ b/ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c
@@ -236,12 +236,7 @@ ClockDescribeRates (
       *TotalRates = NUM_RATES (DescribeRates->NumRatesFlags)
                     + NUM_REMAIN_RATES (DescribeRates->NumRatesFlags);
 
-      if (*Format == ScmiClockRateFormatDiscrete) {
-        RequiredArraySize = (*TotalRates) * sizeof (UINT64);
-      } else {
-        // We need to return triplet of 64 bit value for each rate
-        RequiredArraySize = (*TotalRates) * 3 * sizeof (UINT64);
-      }
+      RequiredArraySize = (*TotalRates) * sizeof (UINT64);
 
       if (RequiredArraySize > (*RateArraySize)) {
         *RateArraySize = RequiredArraySize;
-- 
2.39.1

Re: [PATCH] ArmPkg/ArmScmiDxe: Fix the calculation of RequiredArraySize in ClockDescribeRates()

[PierreGondois]

Hello Sudeep,

On 2/2/23 15:30, Sudeep Holla wrote:
> As per the SCMI specification, section CLOCK_DESCRIBE_RATES mentions that
> the value of num_rates_flags[11:0] in the response must be 3 if the return
> format is the triplet. Due to the buggy firmware, this was not noticed for
> long time. The firmware is now fixed resulting in ClockDescribeRates() to
> fail with "Buffer Too Small" error as the RequiredArraySize gets miscalculated
> as 72 instead of 24.
> 
> Fix the issue by reusing the logic for both the return format which must
> work if num_rates_flags has correct value as expected from the specification.
> 
> Cc: Pierre Gondois <Pierre.Gondois@arm.com>
> Cc: Girish Pathak <girish.pathak@arm.com>
> Cc: Jeff Brasen <jbrasen@nvidia.com>
> Reported-by: Sami Mujawar <sami.mujawar@arm.com>
> Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
> ---
>   ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c | 7 +------
>   1 file changed, 1 insertion(+), 6 deletions(-)
> 
> Hi All,
> 
> Sorry for the resend, I hadn't subscribed before so resending after the
> subscription.
> 
> Regards,
> Sudeep
> 
> diff --git a/ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c b/ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c
> index 12a7e6df5d43..1412717bc227 100644
> --- a/ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c
> +++ b/ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c
> @@ -236,12 +236,7 @@ ClockDescribeRates (
>         *TotalRates = NUM_RATES (DescribeRates->NumRatesFlags)
>                       + NUM_REMAIN_RATES (DescribeRates->NumRatesFlags);
>   
> -      if (*Format == ScmiClockRateFormatDiscrete) {
> -        RequiredArraySize = (*TotalRates) * sizeof (UINT64);
> -      } else {
> -        // We need to return triplet of 64 bit value for each rate
> -        RequiredArraySize = (*TotalRates) * 3 * sizeof (UINT64);
> -      }
> +      RequiredArraySize = (*TotalRates) * sizeof (UINT64);
>   
>         if (RequiredArraySize > (*RateArraySize)) {
>           *RateArraySize = RequiredArraySize;

I think the following also needs to be modified:
"""
for (RateNo = 0; RateNo < NUM_RATES (DescribeRates->NumRatesFlags); RateNo++) {
   // Linear clock rates from minimum to maximum in steps
   // Minimum clock rate.
   Rate                                    = &DescribeRates->Rates[RateOffset++];
   RateArray[RateIndex].ContinuousRate.Min =
	ConvertTo64Bit (Rate->Low, Rate->High);

   Rate = &DescribeRates->Rates[RateOffset++];
   // Maximum clock rate.
   RateArray[RateIndex].ContinuousRate.Max =
	ConvertTo64Bit (Rate->Low, Rate->High);

   Rate = &DescribeRates->Rates[RateOffset++];
   // Step.
   RateArray[RateIndex++].ContinuousRate.Step =
	ConvertTo64Bit (Rate->Low, Rate->High);
}
"""

If the clock format is continuous, then per spec:
   NUM_RATES (DescribeRates->NumRatesFlags) == 3
However 3 rates are read per loop here. So I think the
loop needs to be removed aswell.

Without this additional change, I get:

"""
Synchronous Exception at 0xA2D0E6EBD20F8604

Synchronous Exception at 0xA2D0E6EBD20F8604
PC 0xA2D0E6EBD20F8604

Recursive exception occurred while dumping the CPU state
"""

With the above loop removed:
Reviewed-by: Pierre Gondois <pierre.gondois@arm.com>
Tested-by: Pierre Gondois <pierre.gondois@arm.com>

Regards,
Pierre

Re: [PATCH] ArmPkg/ArmScmiDxe: Fix the calculation of RequiredArraySize in ClockDescribeRates()

[Sudeep Holla]

On Fri, Feb 03, 2023 at 12:10:38PM +0100, Pierre Gondois wrote:
> Hello Sudeep,
> 
> On 2/2/23 15:30, Sudeep Holla wrote:
> > As per the SCMI specification, section CLOCK_DESCRIBE_RATES mentions that
> > the value of num_rates_flags[11:0] in the response must be 3 if the return
> > format is the triplet. Due to the buggy firmware, this was not noticed for
> > long time. The firmware is now fixed resulting in ClockDescribeRates() to
> > fail with "Buffer Too Small" error as the RequiredArraySize gets miscalculated
> > as 72 instead of 24.
> > 
> > Fix the issue by reusing the logic for both the return format which must
> > work if num_rates_flags has correct value as expected from the specification.
> > 
> > Cc: Pierre Gondois <Pierre.Gondois@arm.com>
> > Cc: Girish Pathak <girish.pathak@arm.com>
> > Cc: Jeff Brasen <jbrasen@nvidia.com>
> > Reported-by: Sami Mujawar <sami.mujawar@arm.com>
> > Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
> > ---
> >   ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c | 7 +------
> >   1 file changed, 1 insertion(+), 6 deletions(-)
> > 
> > Hi All,
> > 
> > Sorry for the resend, I hadn't subscribed before so resending after the
> > subscription.
> > 
> > Regards,
> > Sudeep
> > 
> > diff --git a/ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c b/ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c
> > index 12a7e6df5d43..1412717bc227 100644
> > --- a/ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c
> > +++ b/ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c
> > @@ -236,12 +236,7 @@ ClockDescribeRates (
> >         *TotalRates = NUM_RATES (DescribeRates->NumRatesFlags)
> >                       + NUM_REMAIN_RATES (DescribeRates->NumRatesFlags);
> > -      if (*Format == ScmiClockRateFormatDiscrete) {
> > -        RequiredArraySize = (*TotalRates) * sizeof (UINT64);
> > -      } else {
> > -        // We need to return triplet of 64 bit value for each rate
> > -        RequiredArraySize = (*TotalRates) * 3 * sizeof (UINT64);
> > -      }
> > +      RequiredArraySize = (*TotalRates) * sizeof (UINT64);
> >         if (RequiredArraySize > (*RateArraySize)) {
> >           *RateArraySize = RequiredArraySize;
> 
> I think the following also needs to be modified:
> """
> for (RateNo = 0; RateNo < NUM_RATES (DescribeRates->NumRatesFlags); RateNo++) {
>   // Linear clock rates from minimum to maximum in steps
>   // Minimum clock rate.
>   Rate                                    = &DescribeRates->Rates[RateOffset++];
>   RateArray[RateIndex].ContinuousRate.Min =
> 	ConvertTo64Bit (Rate->Low, Rate->High);
> 
>   Rate = &DescribeRates->Rates[RateOffset++];
>   // Maximum clock rate.
>   RateArray[RateIndex].ContinuousRate.Max =
> 	ConvertTo64Bit (Rate->Low, Rate->High);
> 
>   Rate = &DescribeRates->Rates[RateOffset++];
>   // Step.
>   RateArray[RateIndex++].ContinuousRate.Step =
> 	ConvertTo64Bit (Rate->Low, Rate->High);
> }
> """
>

Right, good catch. For some reason, I don't see any exceptions in release
build :(. I will try the debug build and check. Thanks for the review.
> 
> With the above loop removed:
> Reviewed-by: Pierre Gondois <pierre.gondois@arm.com>
> Tested-by: Pierre Gondois <pierre.gondois@arm.com>
>

Thanks!

-- 
Regards,
Sudeep

Re: [edk2-devel] [PATCH] ArmPkg/ArmScmiDxe: Fix the calculation of RequiredArraySize in ClockDescribeRates()

[Ard Biesheuvel]

On Fri, 3 Feb 2023 at 12:20, Sudeep Holla <sudeep.holla@arm.com> wrote:
>
> On Fri, Feb 03, 2023 at 12:10:38PM +0100, Pierre Gondois wrote:
> > Hello Sudeep,
> >
> > On 2/2/23 15:30, Sudeep Holla wrote:
> > > As per the SCMI specification, section CLOCK_DESCRIBE_RATES mentions that
> > > the value of num_rates_flags[11:0] in the response must be 3 if the return
> > > format is the triplet. Due to the buggy firmware, this was not noticed for
> > > long time. The firmware is now fixed resulting in ClockDescribeRates() to
> > > fail with "Buffer Too Small" error as the RequiredArraySize gets miscalculated
> > > as 72 instead of 24.
> > >
> > > Fix the issue by reusing the logic for both the return format which must
> > > work if num_rates_flags has correct value as expected from the specification.
> > >
> > > Cc: Pierre Gondois <Pierre.Gondois@arm.com>
> > > Cc: Girish Pathak <girish.pathak@arm.com>
> > > Cc: Jeff Brasen <jbrasen@nvidia.com>
> > > Reported-by: Sami Mujawar <sami.mujawar@arm.com>
> > > Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
> > > ---
> > >   ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c | 7 +------
> > >   1 file changed, 1 insertion(+), 6 deletions(-)
> > >
> > > Hi All,
> > >
> > > Sorry for the resend, I hadn't subscribed before so resending after the
> > > subscription.
> > >
> > > Regards,
> > > Sudeep
> > >
> > > diff --git a/ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c b/ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c
> > > index 12a7e6df5d43..1412717bc227 100644
> > > --- a/ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c
> > > +++ b/ArmPkg/Drivers/ArmScmiDxe/ScmiClockProtocol.c
> > > @@ -236,12 +236,7 @@ ClockDescribeRates (
> > >         *TotalRates = NUM_RATES (DescribeRates->NumRatesFlags)
> > >                       + NUM_REMAIN_RATES (DescribeRates->NumRatesFlags);
> > > -      if (*Format == ScmiClockRateFormatDiscrete) {
> > > -        RequiredArraySize = (*TotalRates) * sizeof (UINT64);
> > > -      } else {
> > > -        // We need to return triplet of 64 bit value for each rate
> > > -        RequiredArraySize = (*TotalRates) * 3 * sizeof (UINT64);
> > > -      }
> > > +      RequiredArraySize = (*TotalRates) * sizeof (UINT64);
> > >         if (RequiredArraySize > (*RateArraySize)) {
> > >           *RateArraySize = RequiredArraySize;
> >
> > I think the following also needs to be modified:
> > """
> > for (RateNo = 0; RateNo < NUM_RATES (DescribeRates->NumRatesFlags); RateNo++) {
> >   // Linear clock rates from minimum to maximum in steps
> >   // Minimum clock rate.
> >   Rate                                    = &DescribeRates->Rates[RateOffset++];
> >   RateArray[RateIndex].ContinuousRate.Min =
> >       ConvertTo64Bit (Rate->Low, Rate->High);
> >
> >   Rate = &DescribeRates->Rates[RateOffset++];
> >   // Maximum clock rate.
> >   RateArray[RateIndex].ContinuousRate.Max =
> >       ConvertTo64Bit (Rate->Low, Rate->High);
> >
> >   Rate = &DescribeRates->Rates[RateOffset++];
> >   // Step.
> >   RateArray[RateIndex++].ContinuousRate.Step =
> >       ConvertTo64Bit (Rate->Low, Rate->High);
> > }
> > """
> >
>
> Right, good catch. For some reason, I don't see any exceptions in release
> build :(. I will try the debug build and check. Thanks for the review.
> >
> > With the above loop removed:
> > Reviewed-by: Pierre Gondois <pierre.gondois@arm.com>
> > Tested-by: Pierre Gondois <pierre.gondois@arm.com>
> >
>
> Thanks!
>

Excellent. Could someone post the final version please?