d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

busdma_bounce: Make the map waiting list per-bounce-zone.

Browse Source 
When pages are freed to a bounce zone, only maps waiting for pages for
that zone can make forward progress.  If a map for a different bounce
zone is at the head of the global list, then requests that could
otherwise make forward progress will be stalled waiting on the other
bounce zone.  If bounce zones shared bounce pages then a global list
would still make sense to prevent "later" requests from starving an
earlier request but that is not a concern with per-zone bounce page
pools.

Reviewed by:	imp
Sponsored by:	Netflix
Differential Revision:	https://reviews.freebsd.org/D34966
This commit is contained in:
John Baldwin 2022-04-21 10:41:09 -07:00
parent d11f5d4762
commit 10fe9a1fb4
1 changed files with 8 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
sys/kern/subr_busdma_bounce.c
View File

@ -65,6 +65,7 @@ struct bounce_page {
struct bounce_zone {
STAILQ_ENTRY(bounce_zone) links;
STAILQ_HEAD(, bounce_page) bounce_page_list;
STAILQ_HEAD(, bus_dmamap) bounce_map_waitinglist;
int total_bpages;
int free_bpages;
int reserved_bpages;
@ -88,7 +89,6 @@ static int total_bpages;
static int busdma_zonecount;
static STAILQ_HEAD(, bounce_zone) bounce_zone_list;
static STAILQ_HEAD(, bus_dmamap) bounce_map_waitinglist;
static STAILQ_HEAD(, bus_dmamap) bounce_map_callbacklist;
static MALLOC_DEFINE(M_BOUNCE, "bounce", "busdma bounce pages");
@ -103,6 +103,7 @@ static int reserve_bounce_pages(bus_dma_tag_t dmat, bus_dmamap_t map,
static int
_bus_dmamap_reserve_pages(bus_dma_tag_t dmat, bus_dmamap_t map, int flags)
{
struct bounce_zone *bz;
/* Reserve Necessary Bounce Pages */
mtx_lock(&bounce_lock);
@ -115,7 +116,9 @@ _bus_dmamap_reserve_pages(bus_dma_tag_t dmat, bus_dmamap_t map, int flags)
} else {
if (reserve_bounce_pages(dmat, map, 1) != 0) {
/* Queue us for resources */
STAILQ_INSERT_TAIL(&bounce_map_waitinglist, map, links);
bz = dmat->bounce_zone;
STAILQ_INSERT_TAIL(&bz->bounce_map_waitinglist, map,
links);
mtx_unlock(&bounce_lock);
return (EINPROGRESS);
}
@ -131,7 +134,6 @@ init_bounce_pages(void *dummy __unused)
total_bpages = 0;
STAILQ_INIT(&bounce_zone_list);
STAILQ_INIT(&bounce_map_waitinglist);
STAILQ_INIT(&bounce_map_callbacklist);
mtx_init(&bounce_lock, "bounce pages lock", NULL, MTX_DEF);
}
@ -174,6 +176,7 @@ alloc_bounce_zone(bus_dma_tag_t dmat)
return (ENOMEM);
STAILQ_INIT(&bz->bounce_page_list);
STAILQ_INIT(&bz->bounce_map_waitinglist);
bz->free_bpages = 0;
bz->reserved_bpages = 0;
bz->active_bpages = 0;
@ -403,9 +406,9 @@ free_bounce_page(bus_dma_tag_t dmat, struct bounce_page *bpage)
STAILQ_INSERT_HEAD(&bz->bounce_page_list, bpage, links);
bz->free_bpages++;
bz->active_bpages--;
if ((map = STAILQ_FIRST(&bounce_map_waitinglist)) != NULL) {
if ((map = STAILQ_FIRST(&bz->bounce_map_waitinglist)) != NULL) {
if (reserve_bounce_pages(map->dmat, map, 1) == 0) {
STAILQ_REMOVE_HEAD(&bounce_map_waitinglist, links);
STAILQ_REMOVE_HEAD(&bz->bounce_map_waitinglist, links);
STAILQ_INSERT_TAIL(&bounce_map_callbacklist,
map, links);
bz->total_deferred++;