Skip to contentSkip to conversation

selftests: udmabuf: fix hugepage size calculation

PATCH In review zishun yi 4 participants 5 messages Jan 18 – Jan 23 Kselftest

lore
Share & tools
lore

zishun yi

author
#0 lore raw

PATCH 1/1 selftests: udmabuf: fix hugepage size calculation

udmabuf's test 6 and 7 is intended to verify the behavior of 2MB huge pages migration, but the helper function still calculates the size using standard (4KB) pages.

---
 tools/testing/selftests/drivers/dma-buf/udmabuf.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)
tools/testing/selftests/drivers/dma-buf/udmabuf.c +6 -6
index d78aec662586..8887fb9b79df 100644
@@ -69,7 +69,7 @@ static int create_udmabuf_list(int devfd, int memfd, off64_t memfd_size)
 	for (i = 0; i < NUM_ENTRIES; i++) {
 		list->list[i].memfd  = memfd;
 		list->list[i].offset = i * (memfd_size / NUM_ENTRIES);
-		list->list[i].size   = getpagesize() * NUM_PAGES;
+		list->list[i].size   = page_size * NUM_PAGES;
 	}
 
 	list->count = NUM_ENTRIES;
@@ -115,8 +115,8 @@ static int compare_chunks(void *addr1, void *addr2, off64_t memfd_size)
 	while (i < NUM_ENTRIES) {
 		off = i * (memfd_size / NUM_ENTRIES);
 		for (j = 0; j < NUM_PAGES; j++, k++) {
-			char1 = *((char *)addr1 + off + (j * getpagesize()));
-			char2 = *((char *)addr2 + (k * getpagesize()));
+			char1 = *((char *)addr1 + off + (j * page_size));
+			char2 = *((char *)addr2 + (k * page_size));
 			if (char1 != char2) {
 				ret = -1;
 				goto err;
@@ -126,7 +126,7 @@ static int compare_chunks(void *addr1, void *addr2, off64_t memfd_size)
 	}
 err:
 	munmap(addr1, memfd_size);
-	munmap(addr2, NUM_ENTRIES * NUM_PAGES * getpagesize());
+	munmap(addr2, NUM_ENTRIES * NUM_PAGES * page_size);
 	return ret;
 }
 
@@ -240,7 +240,7 @@ int main(int argc, char *argv[])
 	addr1 = mmap_fd(memfd, size);
 	write_to_memfd(addr1, size, 'a');
 	buf = create_udmabuf_list(devfd, memfd, size);
-	addr2 = mmap_fd(buf, NUM_PAGES * NUM_ENTRIES * getpagesize());
+	addr2 = mmap_fd(buf, NUM_PAGES * NUM_ENTRIES * page_size);
 	write_to_memfd(addr1, size, 'b');
 	ret = compare_chunks(addr1, addr2, size);
 	if (ret < 0)
@@ -256,7 +256,7 @@ int main(int argc, char *argv[])
 	size = MEMFD_SIZE * page_size;
 	memfd = create_memfd_with_seals(size, true);
 	buf = create_udmabuf_list(devfd, memfd, size);
-	addr2 = mmap_fd(buf, NUM_PAGES * NUM_ENTRIES * getpagesize());
+	addr2 = mmap_fd(buf, NUM_PAGES * NUM_ENTRIES * page_size);
 	addr1 = mmap_fd(memfd, size);
 	write_to_memfd(addr1, size, 'a');
 	write_to_memfd(addr1, size, 'b');
Signature · 1 line
2.52.0
#1 lore raw
Quoted 3 lines · On Sun, 18 Jan 2026 21:51:26 +0800 zishun yi <[email protected]> wrote:

udmabuf's test 6 and 7 is intended to verify the behavior of 2MB huge pages migration, but the helper function still calculates the size using standard (4KB) pages.

lgtm. Vivek, can you please check it? Feel free to merge into another tree and I'll drop the mm.git ccopy.

Quoted 51 lines
tools/testing/selftests/drivers/dma-buf/udmabuf.c +6 -6
@@ -69,7 +69,7 @@ static int create_udmabuf_list(int devfd, int memfd, off64_t memfd_size)
 	for (i = 0; i < NUM_ENTRIES; i++) {
 		list->list[i].memfd  = memfd;
 		list->list[i].offset = i * (memfd_size / NUM_ENTRIES);
-		list->list[i].size   = getpagesize() * NUM_PAGES;
+		list->list[i].size   = page_size * NUM_PAGES;
 	}
 
 	list->count = NUM_ENTRIES;
@@ -115,8 +115,8 @@ static int compare_chunks(void *addr1, void *addr2, off64_t memfd_size)
 	while (i < NUM_ENTRIES) {
 		off = i * (memfd_size / NUM_ENTRIES);
 		for (j = 0; j < NUM_PAGES; j++, k++) {
-			char1 = *((char *)addr1 + off + (j * getpagesize()));
-			char2 = *((char *)addr2 + (k * getpagesize()));
+			char1 = *((char *)addr1 + off + (j * page_size));
+			char2 = *((char *)addr2 + (k * page_size));
 			if (char1 != char2) {
 				ret = -1;
 				goto err;
@@ -126,7 +126,7 @@ static int compare_chunks(void *addr1, void *addr2, off64_t memfd_size)
 	}
 err:
 	munmap(addr1, memfd_size);
-	munmap(addr2, NUM_ENTRIES * NUM_PAGES * getpagesize());
+	munmap(addr2, NUM_ENTRIES * NUM_PAGES * page_size);
 	return ret;
 }
 
@@ -240,7 +240,7 @@ int main(int argc, char *argv[])
 	addr1 = mmap_fd(memfd, size);
 	write_to_memfd(addr1, size, 'a');
 	buf = create_udmabuf_list(devfd, memfd, size);
-	addr2 = mmap_fd(buf, NUM_PAGES * NUM_ENTRIES * getpagesize());
+	addr2 = mmap_fd(buf, NUM_PAGES * NUM_ENTRIES * page_size);
 	write_to_memfd(addr1, size, 'b');
 	ret = compare_chunks(addr1, addr2, size);
 	if (ret < 0)
@@ -256,7 +256,7 @@ int main(int argc, char *argv[])
 	size = MEMFD_SIZE * page_size;
 	memfd = create_memfd_with_seals(size, true);
 	buf = create_udmabuf_list(devfd, memfd, size);
-	addr2 = mmap_fd(buf, NUM_PAGES * NUM_ENTRIES * getpagesize());
+	addr2 = mmap_fd(buf, NUM_PAGES * NUM_ENTRIES * page_size);
 	addr1 = mmap_fd(memfd, size);
 	write_to_memfd(addr1, size, 'a');
 	write_to_memfd(addr1, size, 'b');
Signature · 1 line
2.52.0
#2 lore raw

Hi Andrew, Zishun Yi,

Quoted 9 lines

Subject: Re: [PATCH v1] selftests: udmabuf: fix hugepage size calculation

Quoted 1 line · On Sun, 18 Jan 2026 21:51:26 +0800 zishun yi <[email protected]> wrote:

udmabuf's test 6 and 7 is intended to verify the behavior of 2MB huge

pages migration,

Quoted 1 line

but the helper function still calculates the size using standard (4KB)

pages.

This behavior is intentional given that a udmabuf is often created for memfd ranges that do not coincide with huge page (size) boundaries. In other words, udmabuf is not always populated with full huge pages. Instead, the ranges mostly align with regular 4K sized page boundaries (which means subpages of huge page), which is exactly what the tests verify.

Quoted 4 lines
Quoted 1 line

lgtm. Vivek, can you please check it? Feel free to merge into another tree and I'll drop the mm.git ccopy.

I think you can go ahead and drop it as this patch is not fixing a bug but is modifying the tests' behavior to create the udmabuf with huge page sized chunks which, although not wrong, but is not the intended objective for the given tests.

Thanks,
Vivek

Quoted 57 lines

tools/testing/selftests/drivers/dma-buf/udmabuf.c static int create_udmabuf_list(int devfd, int

Quoted 4 lines
tools/testing/selftests/drivers/dma-buf/udmabuf.c +0 -0
@@ -69,7 +69,7 @@ static int create_udmabuf_list(int devfd, int

memfd, off64_t memfd_size)

tools/testing/selftests/drivers/dma-buf/udmabuf.c static int compare_chunks(void *addr1, void

Quoted 9 lines
tools/testing/selftests/drivers/dma-buf/udmabuf.c +1 -1
 	for (i = 0; i < NUM_ENTRIES; i++) {
 		list->list[i].memfd  = memfd;
 		list->list[i].offset = i * (memfd_size / NUM_ENTRIES);
-		list->list[i].size   = getpagesize() * NUM_PAGES;
+		list->list[i].size   = page_size * NUM_PAGES;
 	}
 
 	list->count = NUM_ENTRIES;
@@ -115,8 +115,8 @@ static int compare_chunks(void *addr1, void

*addr2, off64_t memfd_size)

tools/testing/selftests/drivers/dma-buf/udmabuf.c static int compare_chunks(void *addr1, void

Quoted 11 lines
tools/testing/selftests/drivers/dma-buf/udmabuf.c +2 -2
 	while (i < NUM_ENTRIES) {
 		off = i * (memfd_size / NUM_ENTRIES);
 		for (j = 0; j < NUM_PAGES; j++, k++) {
-			char1 = *((char *)addr1 + off + (j * getpagesize()));
-			char2 = *((char *)addr2 + (k * getpagesize()));
+			char1 = *((char *)addr1 + off + (j * page_size));
+			char2 = *((char *)addr2 + (k * page_size));
 			if (char1 != char2) {
 				ret = -1;
 				goto err;
@@ -126,7 +126,7 @@ static int compare_chunks(void *addr1, void

*addr2, off64_t memfd_size)

tools/testing/selftests/drivers/dma-buf/udmabuf.c int main(int argc, char *argv[]) · line 243

Quoted 13 lines
tools/testing/selftests/drivers/dma-buf/udmabuf.c +1 -2
 	}
 err:
 	munmap(addr1, memfd_size);
-	munmap(addr2, NUM_ENTRIES * NUM_PAGES * getpagesize());
+	munmap(addr2, NUM_ENTRIES * NUM_PAGES * page_size);
 	return ret;
 }
 
@@ -240,7 +240,7 @@ int main(int argc, char *argv[])
 	addr1 = mmap_fd(memfd, size);
 	write_to_memfd(addr1, size, 'a');
 	buf = create_udmabuf_list(devfd, memfd, size);
-	addr2 = mmap_fd(buf, NUM_PAGES * NUM_ENTRIES *

getpagesize());

tools/testing/selftests/drivers/dma-buf/udmabuf.c int main(int argc, char *argv[]) · line 259

Quoted 9 lines
tools/testing/selftests/drivers/dma-buf/udmabuf.c +1 -1
+	addr2 = mmap_fd(buf, NUM_PAGES * NUM_ENTRIES * page_size);
 	write_to_memfd(addr1, size, 'b');
 	ret = compare_chunks(addr1, addr2, size);
 	if (ret < 0)
@@ -256,7 +256,7 @@ int main(int argc, char *argv[])
 	size = MEMFD_SIZE * page_size;
 	memfd = create_memfd_with_seals(size, true);
 	buf = create_udmabuf_list(devfd, memfd, size);
-	addr2 = mmap_fd(buf, NUM_PAGES * NUM_ENTRIES *

getpagesize());

Quoted 6 lines
tools/testing/selftests/drivers/dma-buf/udmabuf.c +1 -1
+	addr2 = mmap_fd(buf, NUM_PAGES * NUM_ENTRIES * page_size);
 	addr1 = mmap_fd(memfd, size);
 	write_to_memfd(addr1, size, 'a');
 	write_to_memfd(addr1, size, 'b');
--

2.52.0

#3 lore raw
Quoted 8 lines · Vivek
Quoted 4 lines
Quoted 1 line

lgtm. Vivek, can you please check it? Feel free to merge into another tree and I'll drop the mm.git ccopy.

I think you can go ahead and drop it as this patch is not fixing a bug but is modifying the tests' behavior to create the udmabuf with huge page sized chunks which, although not wrong, but is not the intended objective for the given tests.

Great, thanks, I've removed this patch from mm.git.

#4 lore raw
In reply to Kasireddy, Vivek
Quoted 5 lines · Vivek

This behavior is intentional given that a udmabuf is often created for memfd ranges that do not coincide with huge page (size) boundaries. In other words, udmabuf is not always populated with full huge pages. Instead, the ranges mostly align with regular 4K sized page boundaries (which means subpages of huge page), which is exactly what the tests verify.

Thanks for the clarification.
I agree with dropping this patch. Thanks for your review.

Best regards, Zishun