[PATCH] test: improve resiliency of malloc autotest

[PATCH] test: improve resiliency of malloc autotest

[Bruce Richardson]

The test case "test_multi_alloc_statistics" was brittle in that it did
some allocations and frees and then checked statistics without
considering the initial state of the malloc heaps. This meant that,
depending on what allocations/frees were done beforehand, the test can
sometimes fail.

We can improve resiliency by running the test using a new malloc heap,
which means it is unaffected by any previous allocations.

Bugzilla ID: 1579
Fixes: a40a1f8231b4 ("app: various tests update")
Cc: stable@dpdk.org

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 app/test/test_malloc.c | 28 ++++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)

diff --git a/app/test/test_malloc.c b/app/test/test_malloc.c
index 02a7d8ef20..a2f2f6f8a6 100644
--- a/app/test/test_malloc.c
+++ b/app/test/test_malloc.c
@@ -25,6 +25,7 @@
 #include <rte_malloc.h>
 #include <rte_cycles.h>
 #include <rte_random.h>
+#include <rte_eal_paging.h>
 #include <rte_string_fns.h>
 
 #define N 10000
@@ -272,6 +273,28 @@ test_multi_alloc_statistics(void)
 	size_t size = 2048;
 	int align = 1024;
 	int overhead = 0;
+	const size_t heap_size = (1 << 21);
+
+	if (rte_malloc_heap_create(__func__) != 0) {
+		printf("Failed to create test malloc heap\n");
+		return -1;
+	}
+	/* allocate some memory using malloc and add it to our test heap. */
+	void *memory = mmap(NULL, heap_size, PROT_READ | PROT_WRITE,
+			MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+	if (memory == MAP_FAILED) {
+		printf("Failed to allocate memory\n");
+		return -1;
+	}
+	if (rte_malloc_heap_memory_add(__func__, memory, heap_size, NULL, 1, heap_size) != 0) {
+		printf("Failed to add memory to heap\n");
+		return -1;
+	}
+	socket = rte_malloc_heap_get_socket(__func__);
+	if (socket < 0) {
+		printf("Failed to get socket for test malloc heap.\n");
+		return -1;
+	}
 
 	/* Dynamically calculate the overhead by allocating one cacheline and
 	 * then comparing what was allocated from the heap.
@@ -371,6 +394,11 @@ test_multi_alloc_statistics(void)
 		printf("Malloc statistics are incorrect - freed alloc\n");
 		return -1;
 	}
+
+	/* cleanup */
+	rte_malloc_heap_memory_remove(__func__, memory, heap_size);
+	rte_malloc_heap_destroy(__func__);
+	munmap(memory, heap_size);
 	return 0;
 }
 
-- 
2.43.0

[PATCH v2] test: improve resiliency of malloc autotest

[Bruce Richardson]

The test case "test_multi_alloc_statistics" was brittle in that it did
some allocations and frees and then checked statistics without
considering the initial state of the malloc heaps. This meant that,
depending on what allocations/frees were done beforehand, the test can
sometimes fail.

We can improve resiliency by running the test using a new malloc heap,
which means it is unaffected by any previous allocations.

Bugzilla ID: 1579
Fixes: a40a1f8231b4 ("app: various tests update")
Cc: stable@dpdk.org

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
v2:
* removed unnecessary extra include
* only added new code for non-windows, since using mmap for allocation.
---
 app/test/test_malloc.c | 31 +++++++++++++++++++++++++++++++
 1 file changed, 31 insertions(+)

diff --git a/app/test/test_malloc.c b/app/test/test_malloc.c
index 02a7d8ef20..62e4445ebc 100644
--- a/app/test/test_malloc.c
+++ b/app/test/test_malloc.c
@@ -272,6 +272,30 @@ test_multi_alloc_statistics(void)
 	size_t size = 2048;
 	int align = 1024;
 	int overhead = 0;
+#ifndef RTE_EXEC_ENV_WINDOWS
+	const size_t heap_size = (1 << 21);
+
+	if (rte_malloc_heap_create(__func__) != 0) {
+		printf("Failed to create test malloc heap\n");
+		return -1;
+	}
+	/* allocate some memory using malloc and add it to our test heap. */
+	void *memory = mmap(NULL, heap_size, PROT_READ | PROT_WRITE,
+			MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+	if (memory == MAP_FAILED) {
+		printf("Failed to allocate memory\n");
+		return -1;
+	}
+	if (rte_malloc_heap_memory_add(__func__, memory, heap_size, NULL, 1, heap_size) != 0) {
+		printf("Failed to add memory to heap\n");
+		return -1;
+	}
+	socket = rte_malloc_heap_get_socket(__func__);
+	if (socket < 0) {
+		printf("Failed to get socket for test malloc heap.\n");
+		return -1;
+	}
+#endif
 
 	/* Dynamically calculate the overhead by allocating one cacheline and
 	 * then comparing what was allocated from the heap.
@@ -371,6 +395,13 @@ test_multi_alloc_statistics(void)
 		printf("Malloc statistics are incorrect - freed alloc\n");
 		return -1;
 	}
+
+#ifndef RTE_EXEC_ENV_WINDOWS
+	/* cleanup */
+	rte_malloc_heap_memory_remove(__func__, memory, heap_size);
+	rte_malloc_heap_destroy(__func__);
+	munmap(memory, heap_size);
+#endif
 	return 0;
 }
 
-- 
2.43.0

Re: [PATCH v2] test: improve resiliency of malloc autotest

[Dmitry Kozlyuk]

2025-01-17 13:52 (UTC+0000), Bruce Richardson:
> The test case "test_multi_alloc_statistics" was brittle in that it did
> some allocations and frees and then checked statistics without
> considering the initial state of the malloc heaps. This meant that,
> depending on what allocations/frees were done beforehand, the test can
> sometimes fail.
> 
> We can improve resiliency by running the test using a new malloc heap,
> which means it is unaffected by any previous allocations.
> 
> Bugzilla ID: 1579
> Fixes: a40a1f8231b4 ("app: various tests update")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
> ---
> v2:
> * removed unnecessary extra include
> * only added new code for non-windows, since using mmap for allocation.

Why is it necessary to use `mmap()` and not portable `malloc()`?
Even the comment in the patch says "malloc" :)

Re: [PATCH v2] test: improve resiliency of malloc autotest

[Bruce Richardson]

On Fri, Jan 17, 2025 at 05:20:41PM +0300, Dmitry Kozlyuk wrote:
> 2025-01-17 13:52 (UTC+0000), Bruce Richardson:
> > The test case "test_multi_alloc_statistics" was brittle in that it did
> > some allocations and frees and then checked statistics without
> > considering the initial state of the malloc heaps. This meant that,
> > depending on what allocations/frees were done beforehand, the test can
> > sometimes fail.
> > 
> > We can improve resiliency by running the test using a new malloc heap,
> > which means it is unaffected by any previous allocations.
> > 
> > Bugzilla ID: 1579
> > Fixes: a40a1f8231b4 ("app: various tests update")
> > Cc: stable@dpdk.org
> > 
> > Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
> > ---
> > v2:
> > * removed unnecessary extra include
> > * only added new code for non-windows, since using mmap for allocation.
> 
> Why is it necessary to use `mmap()` and not portable `malloc()`?
> Even the comment in the patch says "malloc" :)

I did originally use malloc, but malloc didn't give us aligned memory so
the call to add the memory to the heap was subsequently failing.

However, I see that the unit tests in the CI are failing on some
architectures, probably because of alignment again, because of using a
single 2MB block of memory. I was going to do a v3 where I queried the
pagesize and used N*pgsize as the parameter to "add" rather than saying
it's a 1x2MB block. Instead, though, I'll rework the code
to use malloc and then manually align instead.

/Bruce

[PATCH v3] test: improve resiliency of malloc autotest

[Bruce Richardson]

The test case "test_multi_alloc_statistics" was brittle in that it did
some allocations and frees and then checked statistics without
considering the initial state of the malloc heaps. This meant that,
depending on what allocations/frees were done beforehand, the test can
sometimes fail.

We can improve resiliency by running the test using a new malloc heap,
which means it is unaffected by any previous allocations.

Bugzilla ID: 1579
Fixes: a40a1f8231b4 ("app: various tests update")
Cc: stable@dpdk.org

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
v3:
* switched allocation from mmap to malloc allowing it work on windows
* use explicit alignment of the malloc return value to ensure memory
  added to heap is page-aligned.

v2:
* removed unnecessary extra include
* only added new code for non-windows, since using mmap for allocation.
---
 app/test/test_malloc.c | 35 +++++++++++++++++++++++++++++++++++
 1 file changed, 35 insertions(+)

diff --git a/app/test/test_malloc.c b/app/test/test_malloc.c
index 02a7d8ef20..9e73c0da09 100644
--- a/app/test/test_malloc.c
+++ b/app/test/test_malloc.c
@@ -25,6 +25,7 @@
 #include <rte_malloc.h>
 #include <rte_cycles.h>
 #include <rte_random.h>
+#include <rte_eal_paging.h>
 #include <rte_string_fns.h>
 
 #define N 10000
@@ -272,6 +273,34 @@ test_multi_alloc_statistics(void)
 	size_t size = 2048;
 	int align = 1024;
 	int overhead = 0;
+	const size_t pgsz = rte_mem_page_size();
+	const size_t heap_size = (1 << 21);
+
+	if (pgsz < heap_size) {
+		printf("Page size is smaller than heap size\n");
+		return TEST_SKIPPED;
+	}
+
+	if (rte_malloc_heap_create(__func__) != 0) {
+		printf("Failed to create test malloc heap\n");
+		return -1;
+	}
+	/* Allocate some memory using malloc and add it to our test heap. */
+	void *unaligned_memory = malloc(heap_size + pgsz);
+	if (unaligned_memory == NULL) {
+		printf("Failed to allocate memory\n");
+		return -1;
+	}
+	void *memory = RTE_PTR_ALIGN(unaligned_memory, pgsz);
+	if (rte_malloc_heap_memory_add(__func__, memory, heap_size, NULL, 1, heap_size) != 0) {
+		printf("Failed to add memory to heap\n");
+		return -1;
+	}
+	socket = rte_malloc_heap_get_socket(__func__);
+	if (socket < 0) {
+		printf("Failed to get socket for test malloc heap.\n");
+		return -1;
+	}
 
 	/* Dynamically calculate the overhead by allocating one cacheline and
 	 * then comparing what was allocated from the heap.
@@ -371,6 +400,12 @@ test_multi_alloc_statistics(void)
 		printf("Malloc statistics are incorrect - freed alloc\n");
 		return -1;
 	}
+
+	/* cleanup */
+	rte_malloc_heap_memory_remove(__func__, memory, heap_size);
+	rte_malloc_heap_destroy(__func__);
+	free(unaligned_memory);
+
 	return 0;
 }
 
-- 
2.43.0

Re: [PATCH v2] test: improve resiliency of malloc autotest

[Stephen Hemminger]

On Fri, 17 Jan 2025 14:26:01 +0000
Bruce Richardson <bruce.richardson@intel.com> wrote:

> On Fri, Jan 17, 2025 at 05:20:41PM +0300, Dmitry Kozlyuk wrote:
> > 2025-01-17 13:52 (UTC+0000), Bruce Richardson:  
> > > The test case "test_multi_alloc_statistics" was brittle in that it did
> > > some allocations and frees and then checked statistics without
> > > considering the initial state of the malloc heaps. This meant that,
> > > depending on what allocations/frees were done beforehand, the test can
> > > sometimes fail.
> > > 
> > > We can improve resiliency by running the test using a new malloc heap,
> > > which means it is unaffected by any previous allocations.
> > > 
> > > Bugzilla ID: 1579
> > > Fixes: a40a1f8231b4 ("app: various tests update")
> > > Cc: stable@dpdk.org
> > > 
> > > Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
> > > ---
> > > v2:
> > > * removed unnecessary extra include
> > > * only added new code for non-windows, since using mmap for allocation.  
> > 
> > Why is it necessary to use `mmap()` and not portable `malloc()`?
> > Even the comment in the patch says "malloc" :)  
> 
> I did originally use malloc, but malloc didn't give us aligned memory so
> the call to add the memory to the heap was subsequently failing.

Use posix_memalign() or aligned_alloc() for that?