On 7/8/25 2:57 AM, Christoph Hellwig wrote:
That still doesn't make it sensible to keep the q usage counter elevator for unlimited time. See nvme multipath for how we can keep bios around forever without elevating the usage counter which is supposed to be transient. Note that dm-multipath should in fact already be doing the right thing in bio based mode as well.
Hi Christoph, I will look into modifying the SRP tests in the blktests repository such that these use bio-based mode instead of request-based mode. However, that won't make the regressions disappear that I have reported. Many users prefer the request-based dm-multipath mode because of better support for I/O scheduling and merging. For more information, see also this cover letter from 2008: [PATCH 00/13] request-based dm-multipath (https://lwn.net/Articles/298882/). 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). * Slower booting of Linux devices that modify sysfs attributes synchronously during boot. Thanks, Bart.
