Skip to contentSkip to conversation

gcov: fix memory allocation problem in gcov_info_dup

PATCH In review Yuan Pengfei 2 participants 4 messages Feb 19, 2014 – Feb 20, 2014 Gcov based kernel profiling

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 gcov: fix memory allocation problem in gcov_info_dup

From: Yuan Pengfei <[email protected]>

If -fprofile-values option is used, ctr->num and sci_ptr->num may be zero, resulting in zero size and cv_size, which will cause ENOMEM when opening gcov data files in debugfs.

---
 kernel/gcov/gcc_3_4.c | 2 +-
 kernel/gcov/gcc_4_7.c | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
kernel/gcov/gcc_3_4.c +1 -1
index 27bc88a..1c1f425 100644
@@ -269,7 +269,7 @@ struct gcov_info *gcov_info_dup(struct gcov_info *info)
 		dup->counts[i].num = ctr->num;
 		dup->counts[i].merge = ctr->merge;
 		dup->counts[i].values = vmalloc(size);
-		if (!dup->counts[i].values)
+		if (size && !dup->counts[i].values)
 			goto err_free;
 		memcpy(dup->counts[i].values, ctr->values, size);
 	}
kernel/gcov/gcc_4_7.c +1 -1
index 2c6e463..7465944 100644
@@ -290,7 +290,7 @@ struct gcov_info *gcov_info_dup(struct gcov_info *info)
 
 			dci_ptr->values = vmalloc(cv_size);
 
-			if (!dci_ptr->values)
+			if (cv_size && !dci_ptr->values)
 				goto err_free;
 
 			dci_ptr->num = sci_ptr->num;
Signature · 1 line
1.8.5.3���{.n�+�������+%�����ݶ��w��{.n�+����{��G�����{ay�ʇڙ�,j��f���h��������z_�(�階�ݢj"���m������G������?����&���~��iO��z��v�^�m����������?�I�
#1 lore raw
Quoted 6 lines · Yuan Pengfei

From: Yuan Pengfei <[email protected]>

If -fprofile-values option is used, ctr->num and sci_ptr->num may be zero, resulting in zero size and cv_size, which will cause ENOMEM when opening gcov data files in debugfs.

I'm unaware of any user of -fprofile-values in the kernel. Are you trying to extend gcov-kernel to also support -fprofile-values? I would expect that additional changes to the .gcda file creation logic are required to fully support this GCC option.

Regards,
Peter Oberparleiter

Signature · 2 lines
Peter Oberparleiter
Linux on System z Development - IBM Germany
#2 lore raw

Using -fprofile-values is one case that will cause the memory allocation problem. More changes are needed to fully support -fprofile-values. I submit this patch first because I think it is obvious and does no harm.

Quoted 4 lines

I'm unaware of any user of -fprofile-values in the kernel. Are you
trying to extend gcov-kernel to also support -fprofile-values? I would
expect that additional changes to the .gcda file creation logic are
required to fully support this GCC option.���{.n�+�������+%�����ݶ��w��{.n�+����{��G�����{ay�ʇڙ�,j��f���h��������z_�(�階�ݢj"���m������G������?����&���~��iO��z��v�^�m���� ������?�I�

#3 lore raw
Quoted 7 lines · Yuan Pengfei

Using -fprofile-values is one case that will cause the memory allocation problem. More changes are needed to fully support -fprofile-values. I submit this patch first because I think it is obvious and does no harm.

Quoted 4 lines

I'm unaware of any user of -fprofile-values in the kernel. Are you trying to extend gcov-kernel to also support -fprofile-values? I would expect that additional changes to the .gcda file creation logic are required to fully support this GCC option.

If there is a way to trigger the problem of gcov_info.values == NULL using the existing gcov-kernel infrastructure then I agree that this should be fixed. If not, then I would rather see the suggested change as part of a patch series that enables other profiling types.

Signature · 2 lines
Peter Oberparleiter
Linux on System z Development - IBM Germany