| Message ID | 1483569671-1462-2-git-send-email-keith.busch@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Reviewed-by: Sagi Grimberg<sagi@grimberg.me>
--
To unsubscribe from this list: send the line "unsubscribe linux-block" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed, Jan 04, 2017 at 05:41:06PM -0500, Keith Busch wrote: > This patch makes sure all online CPUs are assigned to vectors in > cases where the nodes don't have the same number of online CPUs. > The calculation for how many vectors needs to be assigned should account > for the number of CPUs for a particular node on each round of assignment > in order to ensure all online CPUs are assigned a vector when they don't > evenly divide, and calculate extras accordingly. > > Since we attempt to divide evently among the nodes, this may still result > in unused vectors if some nodes have fewer CPUs than nodes previosuly > set up, but at least every online CPU will be assigned to something. This looks fine: I think we still should switch to something like all present or possible cpus for MSI-X vector and blk-mq queue assignment, though reduce the needs for this: Reviewed-by: Christoph Hellwig <hch@lst.de> -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/kernel/irq/affinity.c b/kernel/irq/affinity.c index 4544b11..b25dce0 100644 --- a/kernel/irq/affinity.c +++ b/kernel/irq/affinity.c @@ -96,17 +96,19 @@ irq_create_affinity_masks(int nvecs, const struct irq_affinity *affd) /* Spread the vectors per node */ vecs_per_node = affv / nodes; - /* Account for rounding errors */ - extra_vecs = affv - (nodes * vecs_per_node); for_each_node_mask(n, nodemsk) { - int ncpus, v, vecs_to_assign = vecs_per_node; + int ncpus, v, vecs_to_assign; /* Get the cpus on this node which are in the mask */ cpumask_and(nmsk, cpu_online_mask, cpumask_of_node(n)); /* Calculate the number of cpus per vector */ ncpus = cpumask_weight(nmsk); + vecs_to_assign = min(vecs_per_node, ncpus); + + /* Account for rounding errors */ + extra_vecs = ncpus - vecs_to_assign; for (v = 0; curvec < last_affv && v < vecs_to_assign; curvec++, v++) { @@ -123,6 +125,7 @@ irq_create_affinity_masks(int nvecs, const struct irq_affinity *affd) if (curvec >= last_affv) break; + vecs_per_node = (affv - curvec) / --nodes; } done:
This patch makes sure all online CPUs are assigned to vectors in cases where the nodes don't have the same number of online CPUs. The calculation for how many vectors needs to be assigned should account for the number of CPUs for a particular node on each round of assignment in order to ensure all online CPUs are assigned a vector when they don't evenly divide, and calculate extras accordingly. Since we attempt to divide evently among the nodes, this may still result in unused vectors if some nodes have fewer CPUs than nodes previosuly set up, but at least every online CPU will be assigned to something. Signed-off-by: Keith Busch <keith.busch@intel.com> --- kernel/irq/affinity.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-)