Re: [PATCH] driver: soc: dpio: use the whole functions to protect critical zone
From: Li Yang <hidden>
Date: 2021-10-20 01:42:45
Also in:
linux-arm-kernel, lkml
On Mon, Oct 18, 2021 at 10:07 PM [off-list ref] wrote:
From: Meng Li <redacted>
In orininal code, use 2 function spin_lock() and local_irq_save() to
protect the critical zone. But when enable the kernel debug config,
there are below inconsistent lock state detected.
================================
WARNING: inconsistent lock state
5.10.63-yocto-standard #1 Not tainted
--------------------------------
inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
lock_torture_wr/226 [HC0[0]:SC1[5]:HE1:SE0] takes:
ffff002005b2dd80 (&p->access_spinlock){+.?.}-{3:3}, at: qbman_swp_enqueue_multiple_mem_back+0x44/0x270
{SOFTIRQ-ON-W} state was registered at:
lock_acquire.part.0+0xf8/0x250
lock_acquire+0x68/0x84
_raw_spin_lock+0x68/0x90
qbman_swp_enqueue_multiple_mem_back+0x44/0x270
......
cryptomgr_test+0x38/0x60
kthread+0x158/0x164
ret_from_fork+0x10/0x38
irq event stamp: 4498
hardirqs last enabled at (4498): [<ffff800010fcf980>] _raw_spin_unlock_irqrestore+0x90/0xb0
hardirqs last disabled at (4497): [<ffff800010fcffc4>] _raw_spin_lock_irqsave+0xd4/0xe0
softirqs last enabled at (4458): [<ffff8000100108c4>] __do_softirq+0x674/0x724
softirqs last disabled at (4465): [<ffff80001005b2a4>] __irq_exit_rcu+0x190/0x19c
other info that might help us debug this:
Possible unsafe locking scenario:
CPU0
----
lock(&p->access_spinlock);
<Interrupt>
lock(&p->access_spinlock);
*** DEADLOCK ***
So, in order to avoid deadlock, use the whole functinoss/functinos/functions/
spin_lock_irqsave/spin_unlock_irqrestore() to protect critical zone.
Fixes: 3b2abda7d28c ("soc: fsl: dpio: Replace QMAN array mode with ring mode enqueue")
Cc: stable@vger.kernel.org
Signed-off-by: Meng Li <redacted>Applied for fix. Thanks.
quoted hunk ↗ jump to hunk
--- drivers/soc/fsl/dpio/qbman-portal.c | 9 +++------ 1 file changed, 3 insertions(+), 6 deletions(-)diff --git a/drivers/soc/fsl/dpio/qbman-portal.c b/drivers/soc/fsl/dpio/qbman-portal.c index 845e91416b58..a56dbe4de34f 100644 --- a/drivers/soc/fsl/dpio/qbman-portal.c +++ b/drivers/soc/fsl/dpio/qbman-portal.c@@ -785,8 +785,7 @@ int qbman_swp_enqueue_multiple_mem_back(struct qbman_swp *s, int i, num_enqueued = 0; unsigned long irq_flags; - spin_lock(&s->access_spinlock); - local_irq_save(irq_flags); + spin_lock_irqsave(&s->access_spinlock, irq_flags); half_mask = (s->eqcr.pi_ci_mask>>1); full_mask = s->eqcr.pi_ci_mask;@@ -797,8 +796,7 @@ int qbman_swp_enqueue_multiple_mem_back(struct qbman_swp *s, s->eqcr.available = qm_cyc_diff(s->eqcr.pi_ring_size, eqcr_ci, s->eqcr.ci); if (!s->eqcr.available) { - local_irq_restore(irq_flags); - spin_unlock(&s->access_spinlock); + spin_unlock_irqrestore(&s->access_spinlock, irq_flags); return 0; } }@@ -837,8 +835,7 @@ int qbman_swp_enqueue_multiple_mem_back(struct qbman_swp *s, dma_wmb(); qbman_write_register(s, QBMAN_CINH_SWP_EQCR_PI, (QB_RT_BIT)|(s->eqcr.pi)|s->eqcr.pi_vb); - local_irq_restore(irq_flags); - spin_unlock(&s->access_spinlock); + spin_unlock_irqrestore(&s->access_spinlock, irq_flags); return num_enqueued; } --2.17.1