Given that the kernel has exponential backoff to cover the lack of interrupts for scrub completion status there is a reasonable likelihood that 'ndctl start-scrub' is issued while the hardware/platform scrub-state is idle, but the kernel engine poll timer has not fired.
Trigger at least one poll cycle for the kernel to re-read the scrub-state before reporting that ARS is busy. Reported-by: Krzysztof Rusocki <[email protected]> Signed-off-by: Dan Williams <[email protected]> --- ndctl/lib/libndctl.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-) diff --git a/ndctl/lib/libndctl.c b/ndctl/lib/libndctl.c index e5641feec23d..536e142cf6af 100644 --- a/ndctl/lib/libndctl.c +++ b/ndctl/lib/libndctl.c @@ -1354,8 +1354,18 @@ static int __ndctl_bus_get_scrub_state(struct ndctl_bus *bus, NDCTL_EXPORT int ndctl_bus_start_scrub(struct ndctl_bus *bus) { struct ndctl_ctx *ctx = ndctl_bus_get_ctx(bus); + int rc; + + rc = sysfs_write_attr(ctx, bus->scrub_path, "1\n"); - return sysfs_write_attr(ctx, bus->scrub_path, "1\n"); + /* + * Try at least 1 poll cycle before reporting busy in case this + * request hits the kernel's exponential backoff while the + * hardware/platform scrub state is idle. + */ + if (rc == -EBUSY && ndctl_bus_poll_scrub_completion(bus, 1, 1) == 0) + return sysfs_write_attr(ctx, bus->scrub_path, "1\n"); + return rc; } NDCTL_EXPORT int ndctl_bus_get_scrub_state(struct ndctl_bus *bus)
