If I run the following on rk3288-veyron-minnie (a 2GB machine) cd /sys/bus/platform/drivers/dwmmc_rockchip for i in $(seq 1 3000); do echo "========================" $i echo ff0f0000.dwmmc > unbind sleep .5 echo ff0f0000.dwmmc > bind while true; do if [ -e /dev/mmcblk2 ]; then break; fi sleep .1 done done Then I start OOMing somewhere between iteration 200 and 250. Using kmemleak, I see reports like: unreferenced object 0xe39c5580 (size 64): comm "kworker/1:0", pid 17, jiffies 4294821091 (age 96.952s) hex dump (first 32 bytes): 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ backtrace: [<ad19d10a>] __kmalloc+0x1ec/0x2dc [<a28614c3>] blk_mq_alloc_tag_set+0x27c/0x2bc [<0955ae01>] mmc_init_queue+0xa8/0x2a8 [<5102b986>] mmc_blk_alloc_req+0xf8/0x2d4 [<f1c2214f>] mmc_blk_probe+0x4a8/0x6c0 [<0dfdd9d5>] mmc_bus_probe+0x24/0x28 It's pretty clear that we're missing a call to blk_mq_free_tag_set(). Let's add it. Fixes: 81196976ed94 ("mmc: block: Add blk-mq support") Signed-off-by: Douglas Anderson <dianders@xxxxxxxxxxxx> --- drivers/mmc/core/queue.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/drivers/mmc/core/queue.c b/drivers/mmc/core/queue.c index 7c364a9c4eeb..09071e13282e 100644 --- a/drivers/mmc/core/queue.c +++ b/drivers/mmc/core/queue.c @@ -480,6 +480,8 @@ void mmc_cleanup_queue(struct mmc_queue *mq) */ flush_work(&mq->complete_work); + blk_mq_free_tag_set(&mq->tag_set); + mq->card = NULL; } -- 2.21.0.1020.gf2820cf01a-goog