[PATCH 1/1] ufs: core: fix &hwq->cq_lock deadlock issue

[Alice Chao]

[name:lockdep&]WARNING: inconsistent lock state
[name:lockdep&]--------------------------------
[name:lockdep&]inconsistent {IN-HARDIRQ-W} -> {HARDIRQ-ON-W} usage.
[name:lockdep&]kworker/u16:4/260 [HC0[0]:SC0[0]:HE1:SE1] takes:
  ffffff8028444600 (&hwq->cq_lock){?.-.}-{2:2}, at:
ufshcd_mcq_poll_cqe_lock+0x30/0xe0
[name:lockdep&]{IN-HARDIRQ-W} state was registered at:
  lock_acquire+0x17c/0x33c
  _raw_spin_lock+0x5c/0x7c
  ufshcd_mcq_poll_cqe_lock+0x30/0xe0
  ufs_mtk_mcq_intr+0x60/0x1bc [ufs_mediatek_mod]
  __handle_irq_event_percpu+0x140/0x3ec
  handle_irq_event+0x50/0xd8
  handle_fasteoi_irq+0x148/0x2b0
  generic_handle_domain_irq+0x4c/0x6c
  gic_handle_irq+0x58/0x134
  call_on_irq_stack+0x40/0x74
  do_interrupt_handler+0x84/0xe4
  el1_interrupt+0x3c/0x78
<snip>

Possible unsafe locking scenario:
       CPU0
       ----
  lock(&hwq->cq_lock);
  <Interrupt>
    lock(&hwq->cq_lock);
  *** DEADLOCK ***
2 locks held by kworker/u16:4/260:

[name:lockdep&]
 stack backtrace:
CPU: 7 PID: 260 Comm: kworker/u16:4 Tainted: G S      W  OE
6.1.17-mainline-android14-2-g277223301adb #1
Workqueue: ufs_eh_wq_0 ufshcd_err_handler

 Call trace:
  dump_backtrace+0x10c/0x160
  show_stack+0x20/0x30
  dump_stack_lvl+0x98/0xd8
  dump_stack+0x20/0x60
  print_usage_bug+0x584/0x76c
  mark_lock_irq+0x488/0x510
  mark_lock+0x1ec/0x25c
  __lock_acquire+0x4d8/0xffc
  lock_acquire+0x17c/0x33c
  _raw_spin_lock+0x5c/0x7c
  ufshcd_mcq_poll_cqe_lock+0x30/0xe0
  ufshcd_poll+0x68/0x1b0
  ufshcd_transfer_req_compl+0x9c/0xc8
  ufshcd_err_handler+0x3bc/0xea0
  process_one_work+0x2f4/0x7e8
  worker_thread+0x234/0x450
  kthread+0x110/0x134
  ret_from_fork+0x10/0x20

ufs_mtk_mcq_intr() could refer to
https://lore.kernel.org/all/20230328103423.10970-3-powen.kao@xxxxxxxxxxxx/

When ufshcd_err_handler() is executed, CQ event interrupt can enter
waiting for the same lock. It could happened in upstream code path
ufshcd_handle_mcq_cq_events() and also in ufs_mtk_mcq_intr(). This
warning message will be generated when &hwq->cq_lock is used in IRQ
context with IRQ enabled. Use ufshcd_mcq_poll_cqe_lock() with
spin_lock_irqsave instead of spin_lock to resolve the deadlock issue.

Signed-off-by: Alice Chao <alice.chao@xxxxxxxxxxxx>
---
 drivers/ufs/core/ufs-mcq.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/ufs/core/ufs-mcq.c b/drivers/ufs/core/ufs-mcq.c
index 31df052fbc41..202ff71e1b58 100644
--- a/drivers/ufs/core/ufs-mcq.c
+++ b/drivers/ufs/core/ufs-mcq.c
@@ -299,11 +299,11 @@ EXPORT_SYMBOL_GPL(ufshcd_mcq_poll_cqe_nolock);
 unsigned long ufshcd_mcq_poll_cqe_lock(struct ufs_hba *hba,
 				       struct ufs_hw_queue *hwq)
 {
-	unsigned long completed_reqs;
+	unsigned long completed_reqs, flags;
 
-	spin_lock(&hwq->cq_lock);
+	spin_lock_irqsave(&hwq->cq_lock, flags);
 	completed_reqs = ufshcd_mcq_poll_cqe_nolock(hba, hwq);
-	spin_unlock(&hwq->cq_lock);
+	spin_unlock_irqrestore(&hwq->cq_lock, flags);
 
 	return completed_reqs;
 }
-- 
2.18.0