Hi, here is an updated version of the patchset, differing from the previous version only in that it reinstates the missing extra check pointed out in [2]. For your convenience, the content of the cover letter for the previous version follows.
While doing some tests with the null_blk device driver, we bumped into two problems: first, unjustified and in some cases high throughput losses; second, actual hangs. These problems seem to be the consequence of the combination of three causes, and this patchset introduces a fix for each of these causes. In particular, changes address: . an apparent flaw in the logic with which delayed completions are implemented: this flaw causes, with unlucky but non-pathological workloads, actual request-completion delays to become arbitrarily larger than the configured delay; . the missing restart of the device queue on the completion of a request in single-queue non-delayed mode; . the overflow of the request-delay parameter, when extremely high values are used (e.g., to spot bugs). To avoid possible confusion, we stress that these fixes *do not* have anything to do with the problems highlighted in [1] (tests of the multiqueue xen-blkfront and xen-blkback modules with null_blk). You can find more details in the patch descriptions. Thanks, Paolo and Arianna [1] https://lkml.org/lkml/2015/8/19/181 [2] https://lkml.org/lkml/2015/11/2/433 Arianna Avanzini (2): null_blk: guarantee device restart in all irq modes null_blk: change type of completion_nsec to unsigned long Paolo Valente (1): null_blk: set a separate timer for each command drivers/block/null_blk.c | 94 +++++++++++++++++------------------------------- 1 file changed, 33 insertions(+), 61 deletions(-) -- 1.9.1 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
