Commit 503d8e80 authored Feb 10, 2023 by Jim Harris

bdevperf: use rand_r() twice to get 64-bit values



rand_r() only returns up to RAND_MAX which is
INT32_MAX.  This means that on sufficiently large
bdevs, especially with smaller block sizes, bdevperf
may not be issuing I/O across the full range of the
bdev.

Found while investigating issue #2908.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/16768

 (master)

(cherry picked from commit 0d11cf93)
Change-Id: I16db684a57a96f138e709008bded4471428944b6
Signed-off-by: Krzysztof Karas <krzysztof.karas@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17693


Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>

parent b2e6e62a

examples/bdev/bdevperf/bdevperf.c

+7 −1

Original line number	Diff line number	Diff line
		@@ -1094,11 +1094,17 @@ static void
		bdevperf_submit_single(struct bdevperf_job job, struct bdevperf_task task)
		{
		uint64_t offset_in_ios;
		uint64_t rand_value;

		if (job->zipf) {
		offset_in_ios = spdk_zipf_generate(job->zipf);
		} else if (job->is_random) {
		offset_in_ios = rand_r(&job->seed) % job->size_in_ios;
		/* RAND_MAX is only INT32_MAX, so use 2 calls to rand_r to
		* get a large enough value to ensure we are issuing I/O
		* uniformly across the whole bdev.
		*/
		rand_value = (uint64_t)rand_r(&job->seed) * RAND_MAX + rand_r(&job->seed);
		offset_in_ios = rand_value % job->size_in_ios;
		} else {
		offset_in_ios = job->offset_in_ios++;
		if (job->offset_in_ios == job->size_in_ios) {

Admin message