Re: [PATCHv2] mmc: core: fix race of queue reset and card removal
From: Adrian Hunter
Date: Wed Oct 05 2022 - 09:41:58 EST
On 4/10/22 16:13, Christian Löhle wrote:
> If a recovery is active and the card is removed do not
> try to switch back partitions. Furthermore do not reference
> mq->card which might be NULLed in the meantime.
>
> This has been observed with recovery active with CQE.
> [ 1083.510578] Unable to handle kernel NULL pointer dereference at virtual address 000000000000038c
> [ 1083.511362] Mem abort info:
> [ 1083.511626] ESR = 0x96000004
> [ 1083.511912] EC = 0x25: DABT (current EL), IL = 32 bits
> [ 1083.512395] SET = 0, FnV = 0
> [ 1083.512681] EA = 0, S1PTW = 0
> [ 1083.512973] FSC = 0x04: level 0 translation fault
> [ 1083.513417] Data abort info:
> [ 1083.513686] ISV = 0, ISS = 0x00000004
> [ 1083.514039] CM = 0, WnR = 0
> [ 1083.514318] user pgtable: 4k pages, 48-bit VAs, pgdp=000000000a4c3000
> [ 1083.514899] [000000000000038c] pgd=0000000000000000, p4d=0000000000000000
> [ 1083.515854] Internal error: Oops: 96000004 [#1] SMP
> [ 1083.516295] CPU: 0 PID: 153 Comm: kworker/0:2 Tainted: G W 5.18.12-g925ff1d10c99-dirty #7
> [ 1083.517127] Hardware name: Pine64 RockPro64 v2.1 (DT)
> [ 1083.517574] Workqueue: events mmc_mq_recovery_handler
> [ 1083.518032] pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> [ 1083.518645] pc : mmc_blk_reset+0x60/0x1ac
> [ 1083.519004] lr : mmc_blk_reset+0x38/0x1ac
> [ 1083.519361] sp : ffff8000100b3cd0
> [ 1083.519654] x29: ffff8000100b3cd0 x28: 0000000000000000 x27: 0000000000000000
> [ 1083.520288] x26: ffff80000b0ba000 x25: ffff0000f6e74805 x24: ffff000004c2fdc0
> [ 1083.520922] x23: ffff000014950000 x22: ffff000004c2fc18 x21: ffff00000a33c000
> [ 1083.521556] x20: 00000000ffffff85 x19: ffff000004c2fc00 x18: ffffffffffffffff
> [ 1083.522189] x17: ffff80000cd9b200 x16: ffff80000cd9b190 x15: 0000000000000006
> [ 1083.522823] x14: 0000000000000000 x13: ffff80000b0c28f0 x12: 0000000000001707
> [ 1083.523457] x11: 00000000000007ad x10: ffff80000c6c28f0 x9 : ffff80000b0c28f0
> [ 1083.524090] x8 : 00000000fffbffff x7 : 0000000000000001 x6 : 0000000000000000
> [ 1083.524723] x5 : 0000000000000000 x4 : ffff0000f6e62d30 x3 : 0000000000000000
> [ 1083.525357] x2 : 0000000000000000 x1 : ffff00000b6e0000 x0 : 0000000000000000
> [ 1083.525990] Call trace:
> [ 1083.526209] mmc_blk_reset+0x60/0x1ac
> [ 1083.526536] mmc_blk_cqe_recovery+0x8c/0xd0
> [ 1083.526908] mmc_mq_recovery_handler+0xc4/0xd0
> [ 1083.527303] process_one_work+0x23c/0x3fc
> [ 1083.527663] worker_thread+0x74/0x420
> [ 1083.527990] kthread+0xec/0xf0
> [ 1083.528264] ret_from_fork+0x10/0x20
> [ 1083.528587] Code: d50323bf d65f03c0 f94352a0 f9404000 (b9438c01)
> [ 1083.529126] ---[ end trace 0000000000000000 ]---
>
> [ 1431.677970] Unable to handle kernel NULL pointer dereference at virtual address 0000000000000000
> [ 1431.678753] Mem abort info:
> [ 1431.679017] ESR = 0x96000004
> [ 1431.679303] EC = 0x25: DABT (current EL), IL = 32 bits
> [ 1431.679786] SET = 0, FnV = 0
> [ 1431.680072] EA = 0, S1PTW = 0
> [ 1431.680366] FSC = 0x04: level 0 translation fault
> [ 1431.680810] Data abort info:
> [ 1431.681080] ISV = 0, ISS = 0x00000004
> [ 1431.681432] CM = 0, WnR = 0
> [ 1431.681712] user pgtable: 4k pages, 48-bit VAs, pgdp=000000000bb98000
> [ 1431.682390] [0000000000000000] pgd=0000000000000000, p4d=0000000000000000
> [ 1431.683393] Internal error: Oops: 96000004 [#1] SMP
> [ 1431.683841] CPU: 0 PID: 19948 Comm: kworker/0:2 Not tainted 5.18.12-gf65532578f32-dirty #16
> [ 1431.684576] Hardware name: Pine64 RockPro64 v2.1 (DT)
> [ 1431.685024] Workqueue: events mmc_mq_recovery_handler
> [ 1431.685487] pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> [ 1431.686100] pc : mmc_put_card+0x38/0x110
> [ 1431.686453] lr : mmc_mq_recovery_handler+0x98/0xd0
> [ 1431.686879] sp : ffff800015813cf0
> [ 1431.687173] x29: ffff800015813cf0 x28: 0000000000000000 x27: 0000000000000000
> [ 1431.687807] x26: ffff80000b0ba000 x25: ffff0000f6e74805 x24: ffff000013bd65c0
> [ 1431.688441] x23: ffff000013b96120 x22: ffff000013bd6418 x21: 0000000000000000
> [ 1431.689075] x20: ffff800008ed1c70 x19: ffff8000091767d8 x18: ffffffffffffffff
> [ 1431.689709] x17: 31335b1b6d375b1b x16: 6d305b1b47554245 x15: 0000000000000006
> [ 1431.690343] x14: 0000000000000000 x13: 0000000000000000 x12: 0000000000000000
> [ 1431.690976] x11: ffff000013bd6570 x10: 0000000000000001 x9 : ffff80000ea69228
> [ 1431.691611] x8 : ffff80000df892c8 x7 : 0000000000000000 x6 : 0000000000000001
> [ 1431.692245] x5 : 0000000000000001 x4 : 0000000000000002 x3 : ffff80000e6feac8
> [ 1431.692879] x2 : 000000000000036e x1 : ffff800008ed1c70 x0 : 0000000000000000
> [ 1431.693513] Call trace:
> [ 1431.693732] mmc_put_card+0x38/0x110
> [ 1431.694055] mmc_mq_recovery_handler+0x98/0xd0
> [ 1431.694452] process_one_work+0x23c/0x3fc
> [ 1431.694812] worker_thread+0x74/0x420
> [ 1431.695139] kthread+0xec/0xf0
> [ 1431.695414] ret_from_fork+0x10/0x20
> [ 1431.695738] Code: f9001bf7 aa0103f6 aa0003f5 aa1403e1 (f9400017)
> [ 1431.696278] ---[ end trace 0000000000000000 ]---
>
> Signed-off-by: Christian Loehle <cloehle@xxxxxxxxxxxxxx>
Thanks for finding these issues. A couple of comments below.
> ---
> drivers/mmc/core/block.c | 4 ++--
> drivers/mmc/core/queue.c | 5 +++--
> 2 files changed, 5 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/mmc/core/block.c b/drivers/mmc/core/block.c
> index ce89611a136e..0cd3a7065629 100644
> --- a/drivers/mmc/core/block.c
> +++ b/drivers/mmc/core/block.c
> @@ -997,8 +997,8 @@ static int mmc_blk_reset(struct mmc_blk_data *md, struct mmc_host *host,
>
> md->reset_done |= type;
> err = mmc_hw_reset(host->card);
> - /* Ensure we switch back to the correct partition */
> - if (err) {
> + /* Ensure we switch back to the correct partition on successful reset */
> + if (!err) {
This isn't quite right. Originally, this was err != -EOPNOTSUPP
so "always" unless the reset was not attempted at all. When the -EOPNOTSUPP
return value went away, this should have become unconditional.
Also this change should be a separate patch, and have a fixes tag
i.e.
Fixes: fefdd3c91e0a ("mmc: core: Drop superfluous validations in mmc_hw|sw_reset()")
> struct mmc_blk_data *main_md =
> dev_get_drvdata(&host->card->dev);
> int part_err;
> diff --git a/drivers/mmc/core/queue.c b/drivers/mmc/core/queue.c
> index fefaa901b50f..6931fa082ea7 100644
> --- a/drivers/mmc/core/queue.c
> +++ b/drivers/mmc/core/queue.c
> @@ -137,9 +137,10 @@ static void mmc_mq_recovery_handler(struct work_struct *work)
> struct mmc_queue *mq = container_of(work, struct mmc_queue,
> recovery_work);
> struct request_queue *q = mq->queue;
> + struct mmc_card *card = mq->card;
> struct mmc_host *host = mq->card->host;
>
> - mmc_get_card(mq->card, &mq->ctx);
> + mmc_get_card(card, &mq->ctx);
>
> mq->in_recovery = true;
>
> @@ -157,7 +158,7 @@ static void mmc_mq_recovery_handler(struct work_struct *work)
> if (host->hsq_enabled)
> host->cqe_ops->cqe_recovery_finish(host);
>
> - mmc_put_card(mq->card, &mq->ctx);
> + mmc_put_card(card, &mq->ctx);
>
> blk_mq_run_hw_queues(q, true);
> }
Please try this instead:
diff --git a/drivers/mmc/core/queue.c b/drivers/mmc/core/queue.c
index 6931fa082ea7..d8d9115c51f6 100644
--- a/drivers/mmc/core/queue.c
+++ b/drivers/mmc/core/queue.c
@@ -494,6 +494,13 @@ void mmc_cleanup_queue(struct mmc_queue *mq)
if (blk_queue_quiesced(q))
blk_mq_unquiesce_queue(q);
+ /*
+ * If the recovery completes the last (and only remaining) request in
+ * the queue, and the card has been removed, we could end up here with
+ * the recovery not quite finished yet, so flush it.
+ */
+ flush_work(&mq->recovery_work);
+
blk_mq_free_tag_set(&mq->tag_set);
/*