Since each slab is a collection of independent mini-slabs, we can
fire up a bunch of threads to sort the mini-slabs in parallel.
This speeds up the sorting phase of the rmapbt rebuilding if we
have a large number of mini slabs.
Signed-off-by: Darrick J. Wong <darrick.wong@xxxxxxxxxx>
---
repair/slab.c | 46 ++++++++++++++++++++++++++++++++++++++++++++--
1 file changed, 44 insertions(+), 2 deletions(-)
diff --git a/repair/slab.c b/repair/slab.c
index 97c13d3..8609270 100644
--- a/repair/slab.c
+++ b/repair/slab.c
@@ -201,6 +201,27 @@ slab_add(
return 0;
}
+#include "threads.h"
+
+struct qsort_slab {
+ struct xfs_slab *slab;
+ struct xfs_slab_hdr *hdr;
+ int (*compare_fn)(const void *, const void *);
+};
+
+static void
+qsort_slab_helper(
+ struct work_queue *wq,
+ xfs_agnumber_t agno,
+ void *arg)
+{
+ struct qsort_slab *qs = arg;
+
+ qsort(slab_ptr(qs->slab, qs->hdr, 0), qs->hdr->sh_inuse,
+ qs->slab->s_item_sz, qs->compare_fn);
+ free(qs);
+}
+
/*
* Sort the items in the slab. Do not run this method if there are any
* cursors holding on to the slab.
@@ -210,14 +231,35 @@ qsort_slab(
struct xfs_slab *slab,
int (*compare_fn)(const void *, const void *))
{
+ struct work_queue wq;
struct xfs_slab_hdr *hdr;
+ struct qsort_slab *qs;
+
+ /*
+ * If we don't have that many slabs, we're probably better
+ * off skipping all the thread overhead.
+ */
+ if (slab->s_nr_slabs <= 4) {
+ hdr = slab->s_first;
+ while (hdr) {
+ qsort(slab_ptr(slab, hdr, 0), hdr->sh_inuse,
+ slab->s_item_sz, compare_fn);
+ hdr = hdr->sh_next;
+ }
+ return;
+ }
+ create_work_queue(&wq, NULL, libxfs_nproc());
hdr = slab->s_first;
while (hdr) {
- qsort(slab_ptr(slab, hdr, 0), hdr->sh_inuse, slab->s_item_sz,
- compare_fn);
+ qs = malloc(sizeof(struct qsort_slab));
+ qs->slab = slab;
+ qs->hdr = hdr;
+ qs->compare_fn = compare_fn;
+ queue_work(&wq, qsort_slab_helper, 0, qs);
hdr = hdr->sh_next;
}
+ destroy_work_queue(&wq);
}
/*
--
To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
