[PATCH] percpu: initialize best_upa variable

[PATCH] percpu: initialize best_upa variable

[trix]

From: Tom Rix <trix@redhat.com>

Static analysis reports this problem
percpu.c:2945:6: warning: Assigned value is garbage or undefined
        upa = best_upa;
            ^ ~~~~~~~~
best_upa may not be set, so initialize it.

Signed-off-by: Tom Rix <trix@redhat.com>
---
 mm/percpu.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/mm/percpu.c b/mm/percpu.c
index a257c3efdf18b..6578b706fae81 100644
--- a/mm/percpu.c
+++ b/mm/percpu.c
@@ -2916,6 +2916,7 @@ static struct pcpu_alloc_info * __init __flatten pcpu_build_alloc_info(
 	 * Related to atom_size, which could be much larger than the unit_size.
 	 */
 	last_allocs = INT_MAX;
+	best_upa = max_upa;
 	for (upa = max_upa; upa; upa--) {
 		int allocs = 0, wasted = 0;
 
-- 
2.26.3

Re: [PATCH] percpu: initialize best_upa variable

[Wonhyuk Yang]

On Sun, May 16, 2021 at 3:08 AM <trix@redhat.com> wrote:

> Static analysis reports this problem
> percpu.c:2945:6: warning: Assigned value is garbage or undefined
>         upa = best_upa;
>             ^ ~~~~~~~~
> best_upa may not be set, so initialize it.

Hi,

Actually, best_upa is always set in the for loop below. when upa is 1,
It will always satisfy all conditions.

> diff --git a/mm/percpu.c b/mm/percpu.c
> index a257c3efdf18b..6578b706fae81 100644
> --- a/mm/percpu.c
> +++ b/mm/percpu.c
> @@ -2916,6 +2916,7 @@ static struct pcpu_alloc_info * __init __flatten pcpu_build_alloc_info(
>          * Related to atom_size, which could be much larger than the unit_size.
>          */
>         last_allocs = INT_MAX;
> +       best_upa = max_upa;
>         for (upa = max_upa; upa; upa--) {
>                 int allocs = 0, wasted = 0;

It doesn't seem to be a problem. But, how about this?

best_upa = 1;
for (upa = max_upa; upa>1; upa--)

Re: [PATCH] percpu: initialize best_upa variable

[Dennis Zhou]

Hello,

On Sat, May 15, 2021 at 11:08:17AM -0700, trix@redhat.com wrote:
> From: Tom Rix <trix@redhat.com>
> 
> Static analysis reports this problem
> percpu.c:2945:6: warning: Assigned value is garbage or undefined
>         upa = best_upa;
>             ^ ~~~~~~~~
> best_upa may not be set, so initialize it.
> 
> Signed-off-by: Tom Rix <trix@redhat.com>
> ---
>  mm/percpu.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/mm/percpu.c b/mm/percpu.c
> index a257c3efdf18b..6578b706fae81 100644
> --- a/mm/percpu.c
> +++ b/mm/percpu.c
> @@ -2916,6 +2916,7 @@ static struct pcpu_alloc_info * __init __flatten pcpu_build_alloc_info(
>  	 * Related to atom_size, which could be much larger than the unit_size.
>  	 */
>  	last_allocs = INT_MAX;
> +	best_upa = max_upa;
>  	for (upa = max_upa; upa; upa--) {
>  		int allocs = 0, wasted = 0;
>  
> -- 
> 2.26.3
> 

I think the proper fix would be:

best_upa = 0;
for (...) { }
BUG_ON(!best_upa);
upa = best_upa;

If you're fine with this I'll make the changes and apply it to
for-5.13-fixes.

Can you also tell me what static analysis tool produced this? I'm just a
little curious because this code hasn't changed in several years so I'd
have expected some static analyzer to have caught this by now.

Thanks,
Dennis

Re: [PATCH] percpu: initialize best_upa variable

[Wonhyuk Yang]

On Mon, May 17, 2021 at 11:05 AM Dennis Zhou <dennis@kernel.org> wrote:
>
> Can you also tell me what static analysis tool produced this? I'm just a
> little curious because this code hasn't changed in several years so I'd
> have expected some static analyzer to have caught this by now.
>

It's because uninitialize_var() was gone. It was introduced commit :
3f649ab728cda8("treewide: Remove uninitialized_var() usage")

Re: [PATCH] percpu: initialize best_upa variable

[Tom Rix]

On 5/16/21 7:05 PM, Dennis Zhou wrote:
> Hello,
>
> On Sat, May 15, 2021 at 11:08:17AM -0700, trix@redhat.com wrote:
>> From: Tom Rix <trix@redhat.com>
>>
>> Static analysis reports this problem
>> percpu.c:2945:6: warning: Assigned value is garbage or undefined
>>          upa = best_upa;
>>              ^ ~~~~~~~~
>> best_upa may not be set, so initialize it.
>>
>> Signed-off-by: Tom Rix <trix@redhat.com>
>> ---
>>   mm/percpu.c | 1 +
>>   1 file changed, 1 insertion(+)
>>
>> diff --git a/mm/percpu.c b/mm/percpu.c
>> index a257c3efdf18b..6578b706fae81 100644
>> --- a/mm/percpu.c
>> +++ b/mm/percpu.c
>> @@ -2916,6 +2916,7 @@ static struct pcpu_alloc_info * __init __flatten pcpu_build_alloc_info(
>>   	 * Related to atom_size, which could be much larger than the unit_size.
>>   	 */
>>   	last_allocs = INT_MAX;
>> +	best_upa = max_upa;
>>   	for (upa = max_upa; upa; upa--) {
>>   		int allocs = 0, wasted = 0;
>>   
>> -- 
>> 2.26.3
>>
> I think the proper fix would be:
>
> best_upa = 0;

I was looking for initializing with something that would work.

> for (...) { }
> BUG_ON(!best_upa);
WARN_ON instead?
> upa = best_upa;
>
> If you're fine with this I'll make the changes and apply it to
> for-5.13-fixes.
>
> Can you also tell me what static analysis tool produced this? I'm just a
> little curious because this code hasn't changed in several years so I'd
> have expected some static analyzer to have caught this by now.

Clang 10

Tom

>
> Thanks,
> Dennis
>

Re: [PATCH] percpu: initialize best_upa variable

[Dennis Zhou]

On Mon, May 17, 2021 at 06:17:47AM -0700, Tom Rix wrote:
> 
> On 5/16/21 7:05 PM, Dennis Zhou wrote:
> > Hello,
> > 
> > On Sat, May 15, 2021 at 11:08:17AM -0700, trix@redhat.com wrote:
> > > From: Tom Rix <trix@redhat.com>
> > > 
> > > Static analysis reports this problem
> > > percpu.c:2945:6: warning: Assigned value is garbage or undefined
> > >          upa = best_upa;
> > >              ^ ~~~~~~~~
> > > best_upa may not be set, so initialize it.
> > > 
> > > Signed-off-by: Tom Rix <trix@redhat.com>
> > > ---
> > >   mm/percpu.c | 1 +
> > >   1 file changed, 1 insertion(+)
> > > 
> > > diff --git a/mm/percpu.c b/mm/percpu.c
> > > index a257c3efdf18b..6578b706fae81 100644
> > > --- a/mm/percpu.c
> > > +++ b/mm/percpu.c
> > > @@ -2916,6 +2916,7 @@ static struct pcpu_alloc_info * __init __flatten pcpu_build_alloc_info(
> > >   	 * Related to atom_size, which could be much larger than the unit_size.
> > >   	 */
> > >   	last_allocs = INT_MAX;
> > > +	best_upa = max_upa;
> > >   	for (upa = max_upa; upa; upa--) {
> > >   		int allocs = 0, wasted = 0;
> > > -- 
> > > 2.26.3
> > > 
> > I think the proper fix would be:
> > 
> > best_upa = 0;
> 
> I was looking for initializing with something that would work.
> 

I think I prefer setting it to 0 because it forces the loop to have
succeeded vs being able to bypass it if the for loop logic was changed.

> > for (...) { }
> > BUG_ON(!best_upa);
> WARN_ON instead?

This is initialization code. So if upa == 0, it really is a problem.
Having 0 units per allocation is bogus.

> > upa = best_upa;
> > 
> > If you're fine with this I'll make the changes and apply it to
> > for-5.13-fixes.
> > 
> > Can you also tell me what static analysis tool produced this? I'm just a
> > little curious because this code hasn't changed in several years so I'd
> > have expected some static analyzer to have caught this by now.
> 
> Clang 10
> 
> Tom
> 

Thanks,
Dennis

Re: [PATCH] percpu: initialize best_upa variable

[Dennis Zhou]

Hello,

On Mon, May 17, 2021 at 02:39:21PM +0000, Dennis Zhou wrote:
> On Mon, May 17, 2021 at 06:17:47AM -0700, Tom Rix wrote:
> > 
> > On 5/16/21 7:05 PM, Dennis Zhou wrote:
> > > Hello,
> > > 
> > > On Sat, May 15, 2021 at 11:08:17AM -0700, trix@redhat.com wrote:
> > > > From: Tom Rix <trix@redhat.com>
> > > > 
> > > > Static analysis reports this problem
> > > > percpu.c:2945:6: warning: Assigned value is garbage or undefined
> > > >          upa = best_upa;
> > > >              ^ ~~~~~~~~
> > > > best_upa may not be set, so initialize it.
> > > > 
> > > > Signed-off-by: Tom Rix <trix@redhat.com>
> > > > ---
> > > >   mm/percpu.c | 1 +
> > > >   1 file changed, 1 insertion(+)
> > > > 
> > > > diff --git a/mm/percpu.c b/mm/percpu.c
> > > > index a257c3efdf18b..6578b706fae81 100644
> > > > --- a/mm/percpu.c
> > > > +++ b/mm/percpu.c
> > > > @@ -2916,6 +2916,7 @@ static struct pcpu_alloc_info * __init __flatten pcpu_build_alloc_info(
> > > >   	 * Related to atom_size, which could be much larger than the unit_size.
> > > >   	 */
> > > >   	last_allocs = INT_MAX;
> > > > +	best_upa = max_upa;
> > > >   	for (upa = max_upa; upa; upa--) {
> > > >   		int allocs = 0, wasted = 0;
> > > > -- 
> > > > 2.26.3
> > > > 
> > > I think the proper fix would be:
> > > 
> > > best_upa = 0;
> > 
> > I was looking for initializing with something that would work.
> > 
> 
> I think I prefer setting it to 0 because it forces the loop to have
> succeeded vs being able to bypass it if the for loop logic was changed.
> 
> > > for (...) { }
> > > BUG_ON(!best_upa);
> > WARN_ON instead?
> 
> This is initialization code. So if upa == 0, it really is a problem.
> Having 0 units per allocation is bogus.
> 
> > > upa = best_upa;
> > > 
> > > If you're fine with this I'll make the changes and apply it to
> > > for-5.13-fixes.
> > > 
> > > Can you also tell me what static analysis tool produced this? I'm just a
> > > little curious because this code hasn't changed in several years so I'd
> > > have expected some static analyzer to have caught this by now.
> > 
> > Clang 10
> > 
> > Tom
> > 
> 
> Thanks,
> Dennis

Following up here. Are you find with me making the changes and
attributing it to you? Otherwise I can just spin another patch real
quick.

At this point I've already sent my PR for-5.13-fixes. So I'll queue some
fix for-5.14.
 
Thanks,
Dennis

Re: [PATCH] percpu: initialize best_upa variable

[Tom Rix]

On 5/27/21 1:24 PM, Dennis Zhou wrote:
> Hello,
>
> On Mon, May 17, 2021 at 02:39:21PM +0000, Dennis Zhou wrote:
>> On Mon, May 17, 2021 at 06:17:47AM -0700, Tom Rix wrote:
>>> On 5/16/21 7:05 PM, Dennis Zhou wrote:
>>>> Hello,
>>>>
>>>> On Sat, May 15, 2021 at 11:08:17AM -0700, trix@redhat.com wrote:
>>>>> From: Tom Rix <trix@redhat.com>
>>>>>
>>>>> Static analysis reports this problem
>>>>> percpu.c:2945:6: warning: Assigned value is garbage or undefined
>>>>>           upa = best_upa;
>>>>>               ^ ~~~~~~~~
>>>>> best_upa may not be set, so initialize it.
>>>>>
>>>>> Signed-off-by: Tom Rix <trix@redhat.com>
>>>>> ---
>>>>>    mm/percpu.c | 1 +
>>>>>    1 file changed, 1 insertion(+)
>>>>>
>>>>> diff --git a/mm/percpu.c b/mm/percpu.c
>>>>> index a257c3efdf18b..6578b706fae81 100644
>>>>> --- a/mm/percpu.c
>>>>> +++ b/mm/percpu.c
>>>>> @@ -2916,6 +2916,7 @@ static struct pcpu_alloc_info * __init __flatten pcpu_build_alloc_info(
>>>>>    	 * Related to atom_size, which could be much larger than the unit_size.
>>>>>    	 */
>>>>>    	last_allocs = INT_MAX;
>>>>> +	best_upa = max_upa;
>>>>>    	for (upa = max_upa; upa; upa--) {
>>>>>    		int allocs = 0, wasted = 0;
>>>>> -- 
>>>>> 2.26.3
>>>>>
>>>> I think the proper fix would be:
>>>>
>>>> best_upa = 0;
>>> I was looking for initializing with something that would work.
>>>
>> I think I prefer setting it to 0 because it forces the loop to have
>> succeeded vs being able to bypass it if the for loop logic was changed.
>>
>>>> for (...) { }
>>>> BUG_ON(!best_upa);
>>> WARN_ON instead?
>> This is initialization code. So if upa == 0, it really is a problem.
>> Having 0 units per allocation is bogus.
>>
>>>> upa = best_upa;
>>>>
>>>> If you're fine with this I'll make the changes and apply it to
>>>> for-5.13-fixes.
>>>>
>>>> Can you also tell me what static analysis tool produced this? I'm just a
>>>> little curious because this code hasn't changed in several years so I'd
>>>> have expected some static analyzer to have caught this by now.
>>> Clang 10
>>>
>>> Tom
>>>
>> Thanks,
>> Dennis
> Following up here. Are you find with me making the changes and
> attributing it to you? Otherwise I can just spin another patch real
> quick.

I am fine with you respinning, no need to attribute the change to me.

If you would like a review, include me on the cc.

Thanks!

Tom

> At this point I've already sent my PR for-5.13-fixes. So I'll queue some
> fix for-5.14.
>   
> Thanks,
> Dennis
>