Hi Vipin!
Thanks for all of the work on this bug, it is highly appreciated. Below
are suggestions for improvements for this patch.
On 1/13/2023 1:12 PM, Vipin P R wrote:
add test case coverage to cover the ms_idx jump
This message could be expanded to be more informative. Suggested rewording:
test/fbarray: add test case for incorrect lookahead behavior
Cc: sta...@dpdk.org
Signed-off-by: Vipin P R <vip...@vmware.com>
Acked-by: Kumara Parameshwaran <kparamesh...@vmware.com>
---
Depends-on: 0001-Memory-Allocation-Fixes-ms_idx-jump-lookahead-during.patch
Depends-on: 0002-Memory-Allocation-Fixes-ms_idx-jump-lookbehind-durin.patch
This makes no difference for commit, but for future reference:
depends-on should reference link to actual patches, not a patch file name.
---
app/test/test_fbarray.c | 49 +++++++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 49 insertions(+)
diff --git a/app/test/test_fbarray.c b/app/test/test_fbarray.c
index a691bf4..275449c 100644
--- a/app/test/test_fbarray.c
+++ b/app/test/test_fbarray.c
@@ -11,6 +11,7 @@
#include <rte_debug.h>
#include <rte_errno.h>
#include <rte_fbarray.h>
+#include <rte_memory.h>
This is presumably added to get access to `struct rte_memseg`, but this
is not needed, because the bug is in the mask behavior, which does not
depend on specific data size.
#include "test.h"
@@ -402,6 +403,53 @@ static int check_used_one(void)
return 0;
}
+/* the following test case verifies that the jump in ms_idx for an fb-array is correct. */
+static int test_jump(void)
+{
+ struct rte_fbarray test_array;
+ int input[] = {1, 1070, 1, 2, 1, 2, 4, 12, 2, 2, 1, 2, 1};
I've managed to reduce this bug down to a more minimal example:
{ 63, 1, 2 }
+ int ms_idx, prev_ms_idx, delta;
+ int len;
+ ms_idx = prev_ms_idx = 0;
+
+ int ret = rte_fbarray_init(&test_array, "test", 32768, sizeof(struct
rte_memseg));
+ if (ret == 0) {
+ RTE_LOG(DEBUG, EAL, "FB array init success\n");
If the code did an early exit, an additional indentation level could've
been avoided, like so:
TEST_ASSERT(rte_fbarray_init(&test_array, "test", 256, 8) == 0,
"Failed to initialize fbarray\n");
Also, missing corresponding `rte_fbarray_destroy` call.
+ int k = 0;
Seems like the only place where this is used is in find_next_n_free, and
it never changes, so I don't think this variable is needed at all.
+ for(int i=0; i < sizeof(input)/sizeof(int); i++) {
RTE_DIM? Also, array indices are `unsigned int` rather than `int`,
compiler gives a warning.
+ if (i == 0) {
+ len = input[i];
+ } else {
+ len = input[i] + 1;
+ }
All of this could be rewritten as follows:
int len, hole;
/* if this is not the first iteration, create a hole */
hole = i != 0;
len = input[i] + hole;
+ prev_ms_idx = ms_idx;
+ ms_idx = rte_fbarray_find_next_n_free(&test_array, k, len);
Like I said above, `k` is unneeded, we can just replace it with 0.
+
+ if (i != 0) {
+ ms_idx++;
+ }
Given suggestion above, could use `if (hole)` instead, would be more
readable.
+
+ for (int j=0; j < input[i]; j++) {
Array indices are unsigned, and also could replace with
for (unsigned int j = hole; j < len; j++)
IMO would be more readable.
+ RTE_LOG(DEBUG, EAL, "ms_idx:%d\n", ms_idx);
I don't think this log is needed.
+ rte_fbarray_set_used(&test_array, ms_idx);
+ ms_idx++;
+ }
+
+ if (prev_ms_idx) {
+ /* The value of ms_idx should be monotonically increasing
+ * given the above input sequence in test_array.
+ * */
+ delta = ms_idx - prev_ms_idx;
+ if (!(delta > 0)) {
Given above suggestions, this can be replaced with `if (delta != len)`.
Also, given the `TEST_ASSERT(0)` below, I think this could just be
replaced with an assert and a message, e.g.
TEST_ASSERT(delta == len, "Incorrect fbarray index\n");
--
Thanks,
Anatoly