Skip to contentSkip to conversation

mmc: core: fix race of queue reset and card removal

PATCH In review Christian Löhle 2 participants 4 messages Oct 4, 2022 – Oct 6, 2022 MMC, SD and SDIO

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 mmc: core: fix race of queue reset and card removal

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 ]---
---
 drivers/mmc/core/block.c | 4 ++--
 drivers/mmc/core/queue.c | 5 +++--
 2 files changed, 5 insertions(+), 4 deletions(-)
drivers/mmc/core/block.c +2 -2
index ce89611a136e..0cd3a7065629 100644
@@ -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) {
 		struct mmc_blk_data *main_md =
 			dev_get_drvdata(&host->card->dev);
 		int part_err;
drivers/mmc/core/queue.c +3 -2
index fefaa901b50f..6931fa082ea7 100644
@@ -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);
 }
Signature · 5 lines
2.37.3

Hyperstone GmbH | Reichenaustr. 39a  | 78467 Konstanz
Managing Director: Dr. Jan Peter Berns.
Commercial register of local courts: Freiburg HRB381782
#1 lore raw

I guess treat both versions more like RFCs.
V1 seems a bit nicer, v2 avoids the dangling mq->card, which I don't see an issue with but am easily convinced otherwise. Both fix the issue for me.

-----Original Message-----
From: Christian Löhle <[email protected]>
Sent: Dienstag, 4. Oktober 2022 15:14
To: [email protected]; Adrian Hunter <[email protected]>; Linux MMC List <[email protected]>; [email protected]; Christian Löhle <[email protected]>
Cc: Avri Altman <[email protected]>
Subject: [PATCHv2] mmc: core: fix race of queue reset and card removal

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 ]---
---
 drivers/mmc/core/block.c | 4 ++--
 drivers/mmc/core/queue.c | 5 +++--
 2 files changed, 5 insertions(+), 4 deletions(-)
drivers/mmc/core/block.c +2 -2
@@ -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) {
 		struct mmc_blk_data *main_md =
 			dev_get_drvdata(&host->card->dev);
 		int part_err;
drivers/mmc/core/queue.c +3 -2
@@ -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);
 }

--
2.37.3

Hyperstone GmbH | Reichenaustr. 39a | 78467 Konstanz Managing Director: Dr. Jan Peter Berns. Commercial register of local courts: Freiburg HRB381782

#2 lore raw
In reply to Christian Löhle
Quoted 86 lines · Christian Löhle

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 ]---
Quoted 1 line

Thanks for finding these issues. A couple of comments below.

Quoted 10 lines
---
 drivers/mmc/core/block.c | 4 ++--
 drivers/mmc/core/queue.c | 5 +++--
 2 files changed, 5 insertions(+), 4 deletions(-)
drivers/mmc/core/block.c +0 -0
index ce89611a136e..0cd3a7065629 100644

drivers/mmc/core/block.c static int mmc_blk_reset(struct mmc_blk_data *md, struct mmc_host *host, · line 1001

Quoted 7 lines
drivers/mmc/core/block.c +2 -2
@@ -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()")
Quoted 20 lines
drivers/mmc/core/block.c +0 -0
 		struct mmc_blk_data *main_md =
 			dev_get_drvdata(&host->card->dev);
 		int part_err;
drivers/mmc/core/queue.c +2 -1
index fefaa901b50f..6931fa082ea7 100644
@@ -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;
 

drivers/mmc/core/queue.c static void mmc_mq_recovery_handler(struct work_struct *work) · line 164

Quoted 8 lines
drivers/mmc/core/queue.c +1 -1
@@ -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:

drivers/mmc/core/queue.c +7 -0
index 6931fa082ea7..d8d9115c51f6 100644
@@ -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);
 
 	/*
#3 lore raw

Thanks Adrian for the comments and hints, implemented and submitted. They also fix both issues.

-----Original Message-----
From: Adrian Hunter <[email protected]>
Sent: Mittwoch, 5. Oktober 2022 15:42
To: Christian Löhle <[email protected]>; [email protected]; Linux MMC List <[email protected]>; [email protected]
Cc: Avri Altman <[email protected]>
Subject: Re: [PATCHv2] mmc: core: fix race of queue reset and card removal

Quoted 80 lines · Christian Löhle

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 
Quoted 4 lines

aa0003f5 aa1403e1 (f9400017) [ 1431.696278] ---[ end trace 0000000000000000 ]---

Thanks for finding these issues. A couple of comments below.

Quoted 10 lines
---
 drivers/mmc/core/block.c | 4 ++--
 drivers/mmc/core/queue.c | 5 +++--
 2 files changed, 5 insertions(+), 4 deletions(-)
drivers/mmc/core/block.c index renamed from drivers/mmc/core/block.c +0 -0

ce89611a136e..0cd3a7065629 100644

drivers/mmc/core/block.c static int mmc_blk_reset(struct mmc_blk_data *md, · line 1002

Quoted 8 lines
drivers/mmc/core/block.c +2 -2
@@ -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()")
Quoted 20 lines
drivers/mmc/core/block.c +0 -0
 		struct mmc_blk_data *main_md =
 			dev_get_drvdata(&host->card->dev);
drivers/mmc/core/block.c +0 -0
 		int part_err;
drivers/mmc/core/queue.c index renamed from drivers/mmc/core/queue.c +0 -0

fefaa901b50f..6931fa082ea7 100644

drivers/mmc/core/queue.c +2 -1
@@ -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;
 

drivers/mmc/core/queue.c static void mmc_mq_recovery_handler(struct work_struct *work) · line 164

Quoted 8 lines
drivers/mmc/core/queue.c +1 -1
@@ -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:

drivers/mmc/core/queue.c +7 -0
@@ -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);
 
 	/*

Hyperstone GmbH | Reichenaustr. 39a | 78467 Konstanz Managing Director: Dr. Jan Peter Berns. Commercial register of local courts: Freiburg HRB381782