Skip to contentSkip to conversation

mmap18: fix when PTHREAD_STACK_MIN < PAGE_SIZE

PATCH New Alexey Kodanev 2 participants 3 messages Jul 12, 2021 – Jul 13, 2021

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mmap18: fix when PTHREAD_STACK_MIN < PAGE_SIZE

PTHREAD_STACK_MIN can be less than the page size, to be more precise 2048 on musl. This value is used in the test as an offset for a new stack address with mmap() + MAP_FIXED flag. Though it might not be aligned to the page size.

This breaks the test with musl:

tst_test.c:1311: TINFO: Timeout per run is 0h 05m 00s mmap18.c:98: TBROK: mmap(0x7fe67e2ee800,2048,3,306,-1,0) failed: EINVAL (22) mmap18.c:169: TFAIL: Child: exited with 2
[...]

The fix is to align mapped_size arg to the page size.

---
 testcases/kernel/syscalls/mmap/mmap18.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
testcases/kernel/syscalls/mmap/mmap18.c +4 -3
index dc2926454..b37b29890 100644
@@ -200,11 +200,12 @@ static void grow_stack_fail(size_t stack_size, size_t mapped_size)
 
 static void run_test(void)
 {
-	size_t stack_size = 8 * PTHREAD_STACK_MIN;
+	size_t pthread_stack = LTP_ALIGN(PTHREAD_STACK_MIN, getpagesize());
+	size_t stack_size = 8 * pthread_stack;
 
-	grow_stack_success(stack_size, PTHREAD_STACK_MIN);
+	grow_stack_success(stack_size, pthread_stack);
 	grow_stack_success(stack_size, stack_size/2);
-	grow_stack_fail(stack_size, PTHREAD_STACK_MIN);
+	grow_stack_fail(stack_size, pthread_stack);
 	grow_stack_fail(stack_size, stack_size/2);
 }
 
Signature · 1 line
2.25.1
#1 lore raw

PATCH 1/1 mmap18: fix when PTHREAD_STACK_MIN < PAGE_SIZE

Hi!

Quoted 24 lines

PTHREAD_STACK_MIN can be less than the page size, to be more precise 2048 on musl. This value is used in the test as an offset for a new stack address with mmap() + MAP_FIXED flag. Though it might not be aligned to the page size.

This breaks the test with musl:

tst_test.c:1311: TINFO: Timeout per run is 0h 05m 00s mmap18.c:98: TBROK: mmap(0x7fe67e2ee800,2048,3,306,-1,0) failed: EINVAL (22) mmap18.c:169: TFAIL: Child: exited with 2
[...]

The fix is to align mapped_size arg to the page size.

---
 testcases/kernel/syscalls/mmap/mmap18.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
testcases/kernel/syscalls/mmap/mmap18.c +0 -0
index dc2926454..b37b29890 100644

testcases/kernel/syscalls/mmap/mmap18.c static void grow_stack_fail(size_t stack_size, size_t mapped_size) · line 204

Quoted 6 lines
testcases/kernel/syscalls/mmap/mmap18.c +2 -1
@@ -200,11 +200,12 @@ static void grow_stack_fail(size_t stack_size, size_t mapped_size)
 
 static void run_test(void)
 {
-	size_t stack_size = 8 * PTHREAD_STACK_MIN;
+	size_t pthread_stack = LTP_ALIGN(PTHREAD_STACK_MIN, getpagesize());
+	size_t stack_size = 8 * pthread_stack;

Looking at the code we have to align both stack size and mapped size so this is correct. I guess that we can save a bit by aligning each of them independently since 8 * PTHREAD_STACK_MIN would in most cases be aligned allready.

	size_t pthread_stack = LTP_ALIGN(PTHREAD_STACK_MIN, getpagesize());
	size_t stack_size = LTP_ALIGN(8 * PTHREAD_STACK_MIN, getpagesize());

Either way this looks good so:

Quoted 14 lines
testcases/kernel/syscalls/mmap/mmap18.c +2 -2
-	grow_stack_success(stack_size, PTHREAD_STACK_MIN);
+	grow_stack_success(stack_size, pthread_stack);
 	grow_stack_success(stack_size, stack_size/2);
-	grow_stack_fail(stack_size, PTHREAD_STACK_MIN);
+	grow_stack_fail(stack_size, pthread_stack);
 	grow_stack_fail(stack_size, stack_size/2);
 }
 
Signature · 5 lines
2.25.1


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
Signature · 2 lines
Cyril Hrubis
[email protected]
#2 lore raw

PATCH 1/1 mmap18: fix when PTHREAD_STACK_MIN < PAGE_SIZE

Hi Cyril,

Quoted 37 lines · Cyril Hrubis

Hi!

Quoted 24 lines

PTHREAD_STACK_MIN can be less than the page size, to be more precise 2048 on musl. This value is used in the test as an offset for a new stack address with mmap() + MAP_FIXED flag. Though it might not be aligned to the page size.

This breaks the test with musl:

tst_test.c:1311: TINFO: Timeout per run is 0h 05m 00s mmap18.c:98: TBROK: mmap(0x7fe67e2ee800,2048,3,306,-1,0) failed: EINVAL (22) mmap18.c:169: TFAIL: Child: exited with 2
[...]

The fix is to align mapped_size arg to the page size.

---
 testcases/kernel/syscalls/mmap/mmap18.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
testcases/kernel/syscalls/mmap/mmap18.c +0 -0
index dc2926454..b37b29890 100644

testcases/kernel/syscalls/mmap/mmap18.c static void grow_stack_fail(size_t stack_size, size_t mapped_size) · line 204

Quoted 6 lines
testcases/kernel/syscalls/mmap/mmap18.c +2 -1
@@ -200,11 +200,12 @@ static void grow_stack_fail(size_t stack_size, size_t mapped_size)
 
 static void run_test(void)
 {
-	size_t stack_size = 8 * PTHREAD_STACK_MIN;
+	size_t pthread_stack = LTP_ALIGN(PTHREAD_STACK_MIN, getpagesize());
+	size_t stack_size = 8 * pthread_stack;

Looking at the code we have to align both stack size and mapped size so this is correct. I guess that we can save a bit by aligning each of them independently since 8 * PTHREAD_STACK_MIN would in most cases be aligned allready.

Quoted 6 lines
	size_t pthread_stack = LTP_ALIGN(PTHREAD_STACK_MIN, getpagesize());
	size_t stack_size = LTP_ALIGN(8 * PTHREAD_STACK_MIN, getpagesize());

Either way this looks good so:

Thanks for review!

Applied the original patch, it keeps the original proportion between stack_size and mapped_size for both libc, also stack_size divided by two after that for the mapped_size arg (though it doesn't really matter with 8 * 2048):

Quoted 8 lines
Quoted 8 lines
testcases/kernel/syscalls/mmap/mmap18.c +2 -2
-	grow_stack_success(stack_size, PTHREAD_STACK_MIN);
+	grow_stack_success(stack_size, pthread_stack);
 	grow_stack_success(stack_size, stack_size/2);
-	grow_stack_fail(stack_size, PTHREAD_STACK_MIN);
+	grow_stack_fail(stack_size, pthread_stack);
 	grow_stack_fail(stack_size, stack_size/2);
 }