From: "Steven Rostedt (Google)" <rostedt@xxxxxxxxxxx>
On PowerPC 64 which has 64K pages, it screws up the accounting of some
calculations used for tests. For instance, 1% of the ring buffer may not
be more than a page. So testing 1% and then subtracting the number of
events per page is going to lead to a negative number. This will obviously
fail.
Take into account that the subbuffer may be very large, and to make a
minimum percent to use in case a subbuffer size is greater than 1%.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=219358
Reported-by: Adrien Nader <adrien@xxxxxxxx>
Signed-off-by: Steven Rostedt (Google) <rostedt@xxxxxxxxxxx>
---
utest/tracefs-utest.c | 11 +++++++++++
1 file changed, 11 insertions(+)
diff --git a/utest/tracefs-utest.c b/utest/tracefs-utest.c
index 742f4546bef0..b5095a18bb16 100644
--- a/utest/tracefs-utest.c
+++ b/utest/tracefs-utest.c
@@ -1340,6 +1340,17 @@ static void test_cpu_read_buf_percent(struct test_cpu_data *data, int percent)
/* For percent == 0, just test for any data */
if (percent) {
+ int min_percent;
+
+ /*
+ * For architectures like PowerPC with 64K PAGE_SIZE and thus
+ * large sub buffers, where we will not have over 100 sub buffers
+ * percent must at least cover more than 1 sub buffer.
+ */
+ min_percent = (100 + (data->nr_subbufs - 1)) / data->nr_subbufs;
+ if (percent < min_percent)
+ percent = min_percent;
+
expect = data->nr_subbufs * data->events_per_buf * percent / 100;
/* Add just under the percent */
--
2.45.2
|