[v1] RAS/AMD/ATL: Fix unintended sign extension issue from coverity

[PATCH] RAS/AMD/ATL: Fix unintended sign extension issue from coverity

Posted by Karan Sanghavi 1 year, 3 months ago

Explicit cast pc to u32 to avoid sign extension while left shift

Issue reported by coverity with CID: 1593397

Signed-off-by: Karan Sanghavi <karansanghvi98@gmail.com>
---
Coverity  Link: 
https://scan7.scan.coverity.com/#/project-view/51975/11354?selectedIssue=1593397
---
 drivers/ras/amd/atl/umc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/ras/amd/atl/umc.c b/drivers/ras/amd/atl/umc.c
index dc8aa12f63c8..916c867faaf8 100644
--- a/drivers/ras/amd/atl/umc.c
+++ b/drivers/ras/amd/atl/umc.c
@@ -293,7 +293,7 @@ static unsigned long convert_dram_to_norm_addr_mi300(unsigned long addr)
 	}
 
 	/* PC bit */
-	addr |= pc << bit_shifts.pc;
+	addr |= (u32)pc << bit_shifts.pc;
 
 	/* SID bits */
 	for (i = 0; i < NUM_SID_BITS; i++) {

---
base-commit: 81983758430957d9a5cb3333fe324fd70cf63e7e
change-id: 20241104-coverity1593397signextension-78c9b2c21d51

Best regards,
-- 
Karan Sanghavi <karansanghvi98@gmail.com>

Re: [PATCH] RAS/AMD/ATL: Fix unintended sign extension issue from coverity

Posted by Shuah Khan 1 year, 3 months ago

On 11/4/24 11:34, Karan Sanghavi wrote:
> Explicit cast pc to u32 to avoid sign extension while left shift
> 
> Issue reported by coverity with CID: 1593397
> 
> Signed-off-by: Karan Sanghavi <karansanghvi98@gmail.com>
> ---
> Coverity  Link:
> https://scan7.scan.coverity.com/#/project-view/51975/11354?selectedIssue=1593397

Please include the coverity message instead of this link so
reviewers without coverity accounts can see the report.

> ---
>   drivers/ras/amd/atl/umc.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/ras/amd/atl/umc.c b/drivers/ras/amd/atl/umc.c
> index dc8aa12f63c8..916c867faaf8 100644
> --- a/drivers/ras/amd/atl/umc.c
> +++ b/drivers/ras/amd/atl/umc.c
> @@ -293,7 +293,7 @@ static unsigned long convert_dram_to_norm_addr_mi300(unsigned long addr)
>   	}
>   
>   	/* PC bit */
> -	addr |= pc << bit_shifts.pc;
> +	addr |= (u32)pc << bit_shifts.pc;

How did you determine this is the right fix and how did
test this change?

>   
>   	/* SID bits */
>   	for (i = 0; i < NUM_SID_BITS; i++) {
> 
> ---
> base-commit: 81983758430957d9a5cb3333fe324fd70cf63e7e
> change-id: 20241104-coverity1593397signextension-78c9b2c21d51
> 
> Best regards,

thanks,
-- Shuah

Re: [PATCH] RAS/AMD/ATL: Fix unintended sign extension issue from coverity

Posted by Karan Sanghavi 1 year, 3 months ago

On Mon, Nov 04, 2024 at 02:51:56PM -0700, Shuah Khan wrote:
> On 11/4/24 11:34, Karan Sanghavi wrote:
> > Explicit cast pc to u32 to avoid sign extension while left shift
> > 
> > Issue reported by coverity with CID: 1593397
> > 
> > Signed-off-by: Karan Sanghavi <karansanghvi98@gmail.com>
> > ---
> > Coverity  Link:
> > https://scan7.scan.coverity.com/#/project-view/51975/11354?selectedIssue=1593397
> 
> Please include the coverity message instead of this link so
> reviewers without coverity accounts can see the report.
>
sure will keep it in mind. 
> > ---
> >   drivers/ras/amd/atl/umc.c | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/drivers/ras/amd/atl/umc.c b/drivers/ras/amd/atl/umc.c
> > index dc8aa12f63c8..916c867faaf8 100644
> > --- a/drivers/ras/amd/atl/umc.c
> > +++ b/drivers/ras/amd/atl/umc.c
> > @@ -293,7 +293,7 @@ static unsigned long convert_dram_to_norm_addr_mi300(unsigned long addr)
> >   	}
> >   	/* PC bit */
> > -	addr |= pc << bit_shifts.pc;
> > +	addr |= (u32)pc << bit_shifts.pc;
> 
> How did you determine this is the right fix and how did
> test this change?
>
#define ADDR_SEL_2_CHAN GENMASK(15, 12)

bit_shifts.pc = 5 + FIELD_GET(ADDR_SEL_2_CHAN, temp);

After reviewing the code, I found that bit_shifts.pc can reach a maximum value of 20. 
Left-shifting a u16 pc by this amount results in an implicit promotion to an int64_t, 
which can cause sign extension and lead to unintended negative values.

To avoid this, casting to a larger data type (such as u64) woulbe be most 
appropriate solution here. 

Also,using u64 would be more appropriate rather than u32. 

Should I send a new patch with u64?  

> >   	/* SID bits */
> >   	for (i = 0; i < NUM_SID_BITS; i++) {
> > 
> > ---
> > base-commit: 81983758430957d9a5cb3333fe324fd70cf63e7e
> > change-id: 20241104-coverity1593397signextension-78c9b2c21d51
> > 
> > Best regards,
> 
> thanks,
> -- Shuah

Thank you,
Karan.

Re: [PATCH] RAS/AMD/ATL: Fix unintended sign extension issue from coverity

Posted by Yazen Ghannam 1 year, 3 months ago

On Tue, Nov 05, 2024 at 04:20:27PM +0000, Karan Sanghavi wrote:
> On Mon, Nov 04, 2024 at 02:51:56PM -0700, Shuah Khan wrote:
> > On 11/4/24 11:34, Karan Sanghavi wrote:
> > > Explicit cast pc to u32 to avoid sign extension while left shift
> > > 
> > > Issue reported by coverity with CID: 1593397
> > > 
> > > Signed-off-by: Karan Sanghavi <karansanghvi98@gmail.com>
> > > ---
> > > Coverity  Link:
> > > https://scan7.scan.coverity.com/#/project-view/51975/11354?selectedIssue=1593397
> > 
> > Please include the coverity message instead of this link so
> > reviewers without coverity accounts can see the report.
> >
> sure will keep it in mind. 

Please do share this as it'll help provide context.

> > > ---
> > >   drivers/ras/amd/atl/umc.c | 2 +-
> > >   1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > diff --git a/drivers/ras/amd/atl/umc.c b/drivers/ras/amd/atl/umc.c
> > > index dc8aa12f63c8..916c867faaf8 100644
> > > --- a/drivers/ras/amd/atl/umc.c
> > > +++ b/drivers/ras/amd/atl/umc.c
> > > @@ -293,7 +293,7 @@ static unsigned long convert_dram_to_norm_addr_mi300(unsigned long addr)
> > >   	}
> > >   	/* PC bit */
> > > -	addr |= pc << bit_shifts.pc;
> > > +	addr |= (u32)pc << bit_shifts.pc;
> > 
> > How did you determine this is the right fix and how did
> > test this change?
> >
> #define ADDR_SEL_2_CHAN GENMASK(15, 12)
> 
> bit_shifts.pc = 5 + FIELD_GET(ADDR_SEL_2_CHAN, temp);
> 
> After reviewing the code, I found that bit_shifts.pc can reach a maximum value of 20. 
> Left-shifting a u16 pc by this amount results in an implicit promotion to an int64_t, 
> which can cause sign extension and lead to unintended negative values.
> 

The 'pc' variable holds a single bit in practice.

	#define MI300_UMC_MCA_PC        BIT(25)
	pc   = FIELD_GET(MI300_UMC_MCA_PC,   addr);

> To avoid this, casting to a larger data type (such as u64) woulbe be most 
> appropriate solution here. 
> 
> Also,using u64 would be more appropriate rather than u32. 
> 
> Should I send a new patch with u64?  
>

Another option is to follow the style in the rest of the function and
use the 'temp' variable.


	        /* PC bit */
		-       addr |= pc << bit_shifts.pc;
		+       temp  = pc;
		+       addr |= temp << bit_shifts.pc;


Or we can change the variable declarations to all be 'unsigned long' to
match input/return value.

Thanks,
Yazen