Skip to contentSkip to conversation

percpu: initialize best_upa variable

PATCH In review [email protected] 3 participants 8 messages May 15, 2021 – May 27, 2021

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 percpu: initialize best_upa variable

From: Tom Rix <[email protected]>

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.

---
 mm/percpu.c | 1 +
 1 file changed, 1 insertion(+)
mm/percpu.c +1 -0
index a257c3efdf18b..6578b706fae81 100644
@@ -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;
 
Signature · 1 line
2.26.3
#1 lore raw
Quoted 5 lines · On Sun, May 16, 2021 at 3:08 AM <[email protected]> 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.

mm/percpu.c static struct pcpu_alloc_info * __init __flatten pcpu_build_alloc_info( · line 2921

Quoted 11 lines
mm/percpu.c +1 -0
index a257c3efdf18b..6578b706fae81 100644
@@ -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--)

maintainer
#2 lore raw
In reply to [email protected]

Hello,

Quoted 28 lines · On Sat, May 15, 2021 at 11:08:17AM -0700, [email protected] wrote:

From: Tom Rix <[email protected]>

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.

---
 mm/percpu.c | 1 +
 1 file changed, 1 insertion(+)
mm/percpu.c +1 -0
index a257c3efdf18b..6578b706fae81 100644
@@ -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;
 
Signature · 1 line
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

#3 lore raw
Quoted 5 lines · Dennis Zhou

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")

author
#4 lore raw
In reply to Dennis Zhou
Quoted 28 lines · Dennis Zhou

Hello,

Quoted 25 lines · On Sat, May 15, 2021 at 11:08:17AM -0700, [email protected] wrote:

From: Tom Rix <[email protected]>

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.

---
  mm/percpu.c | 1 +
  1 file changed, 1 insertion(+)
mm/percpu.c +1 -0
index a257c3efdf18b..6578b706fae81 100644
@@ -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;
 
Quoted 6 lines
Quoted 3 lines
Signature · 1 line
2.26.3

I think the proper fix would be:

best_upa = 0;

I was looking for initializing with something that would work.

Quoted 2 lines

for (...) { }
BUG_ON(!best_upa);

WARN_ON instead?

Quoted 8 lines

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

Quoted 4 lines

Thanks,
Dennis

maintainer
#5 lore raw
Quoted 34 lines · Tom Rix
Quoted 32 lines · Dennis Zhou

Hello,

Quoted 27 lines · On Sat, May 15, 2021 at 11:08:17AM -0700, [email protected] wrote:

From: Tom Rix <[email protected]>

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.

---
  mm/percpu.c | 1 +
  1 file changed, 1 insertion(+)
mm/percpu.c +1 -0
index a257c3efdf18b..6578b706fae81 100644
@@ -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;
Signature · 1 line
2.26.3

I think the proper fix would be:

Quoted 4 lines
Quoted 1 line

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.

Quoted 3 lines
Quoted 2 lines

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.

Quoted 5 lines
Quoted 5 lines

upa = best_upa;

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

Quoted 8 lines
Quoted 3 lines

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

maintainer
#6 lore raw

Hello,

Quoted 59 lines · Dennis Zhou
Quoted 34 lines · Tom Rix
Quoted 32 lines · Dennis Zhou

Hello,

Quoted 27 lines · On Sat, May 15, 2021 at 11:08:17AM -0700, [email protected] wrote:

From: Tom Rix <[email protected]>

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.

---
  mm/percpu.c | 1 +
  1 file changed, 1 insertion(+)
mm/percpu.c +1 -0
index a257c3efdf18b..6578b706fae81 100644
@@ -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;
Signature · 1 line
2.26.3

I think the proper fix would be:

Quoted 4 lines
Quoted 1 line

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.

Quoted 3 lines
Quoted 2 lines

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.

Quoted 9 lines
Quoted 8 lines

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.

Quoted 7 lines
Quoted 4 lines

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

author
#7 lore raw
Quoted 59 lines · Dennis Zhou

Hello,

Quoted 56 lines · Dennis Zhou
Quoted 33 lines · Tom Rix
Quoted 32 lines · Dennis Zhou

Hello,

Quoted 27 lines · On Sat, May 15, 2021 at 11:08:17AM -0700, [email protected] wrote:

From: Tom Rix <[email protected]>

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.

---
   mm/percpu.c | 1 +
   1 file changed, 1 insertion(+)
mm/percpu.c +1 -0
index a257c3efdf18b..6578b706fae81 100644
@@ -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;
Signature · 1 line
2.26.3

I think the proper fix would be:

Quoted 3 lines
Quoted 1 line

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.

Quoted 3 lines
Quoted 2 lines

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.

Quoted 10 lines
Quoted 8 lines

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

Quoted 7 lines
Quoted 4 lines
Quoted 2 lines

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

Quoted 6 lines

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

Thanks,
Dennis