On Tue, Jul 08, 2025 at 09:11:41AM -0700, Bart Van Assche wrote: > I will look into modifying the SRP tests in the blktests repository such > that these use bio-based mode instead of request-based mode. Note that this just fixes the test case. The fact that request based dm-multipath keeps active requests and thus an elevated q_usage_counter around still exists then, with effects both to sysfs and other users of queue freezing. > So the question remains what to do about these two regressions: > * The deadlock triggered by modifying a sysfs attribute of a > dm-multipath device configured with "queue_if_no_path" and no paths > (temporarily). That's not a deadlock by the classic definition, but yes, it is hang that should be addressed. > * Slower booting of Linux devices that modify sysfs attributes > synchronously during boot. So which attributes are regularly modified? Note that for read-ahead it should be safe to drop the freeze as unlike the others it is not used for splitting I/O to the limits accepted by the hardware. So we could probably drop the freeze IFF the patch documents that it is safe. But it still won't fix the root cause.
