Skip to contentSkip to conversation

staging: vchiq_arm: Clear VLA warning

PATCH In review Tobin C. Harding 3 participants 5 messages Mar 12, 2018

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

PATCH v2 1/1 staging: vchiq_arm: Clear VLA warning

The kernel would like to have all stack VLA usage removed[1]. The array here is fixed (declared with a const variable) but it appears like a VLA to the compiler. Also, currently we are putting 768 bytes on the stack. This function is only called on the error path so performance is not critical, let's just allocate the memory instead of using the stack. This saves stack space and removes the VLA build warning.

kmalloc a buffer for dumping state instead of using the stack.

[1]: https://lkml.org/lkml/2018/3/7/621

---

v1 of this patch already merged into staging-testing branch of Greg's staging tree. This patch depends on v1 being removed, can re-do this one on top of tip of staging-testing if required.

v2:
- Use kmalloc() instead of the stack

Patch is untested.

 drivers/staging/vc04_services/interface/vchiq_arm/vchiq_arm.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/staging/vc04_services/interface/vchiq_arm/vchiq_arm.c +5 -1
index f5cefda49b22..408ea73f8da7 100644
@@ -3455,11 +3455,15 @@ vchiq_dump_service_use_state(VCHIQ_STATE_T *state)
 		int fourcc;
 		int clientid;
 		int use_count;
-	} service_data[local_max_services];
+	} *service_data;
 
 	if (!arm_state)
 		return;
 
+	service_data = kmalloc_array(local_max_services, sizeof(*service_data), GFP_KERNEL);
+	if (!service_data)
+		return;
+
 	read_lock_bh(&arm_state->susp_res_lock);
 	vc_suspend_state = arm_state->vc_suspend_state;
 	vc_resume_state  = arm_state->vc_resume_state;
Signature · 1 line
2.7.4
#1 lore raw
Quoted 7 lines · Tobin C. Harding

The kernel would like to have all stack VLA usage removed[1]. The array here is fixed (declared with a const variable) but it appears like a VLA to the compiler. Also, currently we are putting 768 bytes on the stack. This function is only called on the error path so performance is not critical, let's just allocate the memory instead of using the stack. This saves stack space and removes the VLA build warning.

Quoted 6 lines

kmalloc a buffer for dumping state instead of using the stack.

[1]: https://lkml.org/lkml/2018/3/7/621

---

Drop this please, leaks memory.

thanks,
Tobin.

#2 lore raw

Hi Tobin,

Quoted 11 lines

"Tobin C. Harding" <[email protected]> hat am 12. März 2018 um 06:46 geschrieben:

Quoted 7 lines · Tobin C. Harding

The kernel would like to have all stack VLA usage removed[1]. The array here is fixed (declared with a const variable) but it appears like a VLA to the compiler. Also, currently we are putting 768 bytes on the stack. This function is only called on the error path so performance is not critical, let's just allocate the memory instead of using the stack. This saves stack space and removes the VLA build warning.

Quoted 8 lines
Quoted 6 lines

kmalloc a buffer for dumping state instead of using the stack.

[1]: https://lkml.org/lkml/2018/3/7/621

---

Drop this please, leaks memory.

except from the leak, did you test this patch on a RPi?

Thanks
Stefan

Quoted 3 lines

thanks,
Tobin.

#3 lore raw
Quoted 15 lines · Stefan Wahren

Hi Tobin,

Quoted 13 lines

"Tobin C. Harding" <[email protected]> hat am 12. März 2018 um 06:46 geschrieben:

Quoted 9 lines · Tobin C. Harding

The kernel would like to have all stack VLA usage removed[1]. The array here is fixed (declared with a const variable) but it appears like a VLA to the compiler. Also, currently we are putting 768 bytes on the stack. This function is only called on the error path so performance is not critical, let's just allocate the memory instead of using the stack. This saves stack space and removes the VLA build warning.

kmalloc a buffer for dumping state instead of using the stack.

Quoted 8 lines
Quoted 6 lines
Quoted 4 lines

[1]: https://lkml.org/lkml/2018/3/7/621

---

Drop this please, leaks memory.

except from the leak, did you test this patch on a RPi?

No I didn't, but I can have a go at it. Will try before doing v3.

thanks,
Tobin.

#4 lore raw
In reply to Stefan Wahren
Quoted 15 lines · Stefan Wahren

Hi Tobin,

Quoted 13 lines

"Tobin C. Harding" <[email protected]> hat am 12. März 2018 um 06:46 geschrieben:

Quoted 9 lines · Tobin C. Harding

The kernel would like to have all stack VLA usage removed[1]. The array here is fixed (declared with a const variable) but it appears like a VLA to the compiler. Also, currently we are putting 768 bytes on the stack. This function is only called on the error path so performance is not critical, let's just allocate the memory instead of using the stack. This saves stack space and removes the VLA build warning.

kmalloc a buffer for dumping state instead of using the stack.

Quoted 9 lines
Quoted 6 lines
Quoted 4 lines

[1]: https://lkml.org/lkml/2018/3/7/621

---

Drop this please, leaks memory.

except from the leak, did you test this patch on a RPi?

Hm... Yeah. It looks like we're holding a mutex when we call vchiq_check_service() from vchiq_queue_message().

regards,
dan carpenter