Il 08/05/24 11:51, Jason-JH.Lin ha scritto:
When we run kernel with lockdebug option, we will get the BUG below:
[ 106.692124] BUG: sleeping function called from invalid context at drivers/base/power/runtime.c:1164
[ 106.692190] in_atomic(): 1, irqs_disabled(): 128, non_block: 0, pid: 3616, name: kworker/u17:3
[ 106.692226] preempt_count: 1, expected: 0
[ 106.692254] RCU nest depth: 0, expected: 0
[ 106.692282] INFO: lockdep is turned off.
[ 106.692306] irq event stamp: 0
[ 106.692331] hardirqs last enabled at (0): [<0000000000000000>] 0x0
[ 106.692376] hardirqs last disabled at (0): [<ffffffee15d37fa0>] copy_process+0xc90/0x2ac0
[ 106.692429] softirqs last enabled at (0): [<ffffffee15d37fc4>] copy_process+0xcb4/0x2ac0
[ 106.692473] softirqs last disabled at (0): [<0000000000000000>] 0x0
[ 106.692513] CPU: 1 PID: 3616 Comm: kworker/u17:3 Not tainted 6.1.87-lockdep-14133-g26e933aca785 #1 6839942e1cf34914b0a366137843dd2366f52aa9
[ 106.692556] Hardware name: Google Ciri sku0/unprovisioned board (DT)
[ 106.692586] Workqueue: imgsys_runner imgsys_runner_func
[ 106.692638] Call trace:
[ 106.692662] dump_backtrace+0x100/0x120
[ 106.692702] show_stack+0x20/0x2c
[ 106.692737] dump_stack_lvl+0x84/0xb4
[ 106.692775] dump_stack+0x18/0x48
[ 106.692809] __might_resched+0x354/0x4c0
[ 106.692847] __might_sleep+0x98/0xe4
[ 106.692883] __pm_runtime_resume+0x70/0x124
[ 106.692921] cmdq_mbox_send_data+0xe4/0xb1c
[ 106.692964] msg_submit+0x194/0x2dc
[ 106.693003] mbox_send_message+0x190/0x330
[ 106.693043] imgsys_cmdq_sendtask+0x1618/0x2224
[ 106.693082] imgsys_runner_func+0xac/0x11c
[ 106.693118] process_one_work+0x638/0xf84
[ 106.693158] worker_thread+0x808/0xcd0
[ 106.693196] kthread+0x24c/0x324
[ 106.693231] ret_from_fork+0x10/0x20
We found that there is a spin_lock_irqsave protection in msg_submit()
of mailbox.c and it is in the atomic context.
So when cmdq driver calls pm_runtime_get_sync() in cmdq_mbox_send_data(),
it will get this BUG report.
1. Change pm_runtime_get_sync() to pm_runtime_get() to avoid using sleep
in atomic context.
2. Move clk_bulk_enable() outside cmdq_runtime_resume() to ensure GCE
clocks are enabled before configuring GCE register.
3. Add used_count to avoid cmdq_runtime_suspend() being called before
calling cmdq_runtime_resume().
Fixes: 8afe816b0c99 ("mailbox: mtk-cmdq-mailbox: Implement Runtime PM with autosuspend")
Signed-off-by: Jason-JH.Lin <jason-jh.lin@xxxxxxxxxxxx>
---
drivers/mailbox/mtk-cmdq-mailbox.c | 24 +++++++++++++++---------
1 file changed, 15 insertions(+), 9 deletions(-)
diff --git a/drivers/mailbox/mtk-cmdq-mailbox.c b/drivers/mailbox/mtk-cmdq-mailbox.c
index 033aff11f87c..b50f42e69aab 100644
--- a/drivers/mailbox/mtk-cmdq-mailbox.c
+++ b/drivers/mailbox/mtk-cmdq-mailbox.c
@@ -82,6 +82,7 @@ struct cmdq {
const struct gce_plat *pdata;
struct cmdq_thread *thread;
struct clk_bulk_data clocks[CMDQ_GCE_NUM_MAX];
+ atomic_t used_count;
bool suspended;
};
@@ -317,14 +318,21 @@ static int cmdq_runtime_resume(struct device *dev)
{
struct cmdq *cmdq = dev_get_drvdata(dev);
- return clk_bulk_enable(cmdq->pdata->gce_num, cmdq->clocks);
+ atomic_inc(&cmdq->used_count);
+ return 0;
}
static int cmdq_runtime_suspend(struct device *dev)
{
struct cmdq *cmdq = dev_get_drvdata(dev);
+ if (atomic_read(&cmdq->used_count) == 0) {
+ dev_warn(dev, "%s when used_count is 0!", __func__);
+ return -EINVAL;
+ }
+
clk_bulk_disable(cmdq->pdata->gce_num, cmdq->clocks);
+ atomic_dec(&cmdq->used_count);
return 0;
}
@@ -392,9 +400,8 @@ static int cmdq_mbox_send_data(struct mbox_chan *chan, void *data)
/* Client should not flush new tasks if suspended. */
WARN_ON(cmdq->suspended);
- ret = pm_runtime_get_sync(cmdq->mbox.dev);
- if (ret < 0)
- return ret;
+ WARN_ON(pm_runtime_get(cmdq->mbox.dev) < 0);
That's a bit sketchy, and I'm afraid that this will break in some other ways.
We could - again - simply change the msg_submit() function in mailbox.c, so that
it takes into account that a driver may need PM done.
A low effort example (which may be good enough or not) is:
static void msg_submit(struct mbox_chan *chan)
{
unsigned count, idx;
unsigned long flags;
void *data;
int err;
if (chan->mbox->ops->pm_off) {
err = chan->mbox->ops->pm_on();
if (err)
return err;
}
spin_lock_irqsave(&chan->lock, flags);
if (!chan->msg_count || chan->active_req) {
err = -EBUSY;
goto exit;
}
......
exit:
spin_unlock_irqrestore(&chan->lock, flags);
if (!err && (chan->txdone_method & TXDONE_BY_POLL)) {
/* kick start the timer immediately to avoid delays */
spin_lock_irqsave(&chan->mbox->poll_hrt_lock, flags);
hrtimer_start(&chan->mbox->poll_hrt, 0, HRTIMER_MODE_REL);
spin_unlock_irqrestore(&chan->mbox->poll_hrt_lock, flags);
}
/* I guess setting it OFF can't fail anyway, so this would be a void function */
if (chan->mbox->ops->pm_off)
chan->mbox->ops->pm_off();
}
Then we can wire up the two functions in the MediaTek driver - where our mailbox
only simply needs pm_runtime_get_sync() and pm_runtime_put().
The reason why I'm suggesting a callback is that this would catch the case of
mailboxes that need "more complicated" flows for ON/OFF.
Jassi, do you like the idea?
If you do, I can eventually go for a real commit with better names than pm_off/on
if I can find any - so that then Jason can go on with his mtk-cmdq fix on top of
that.
Cheers,
Angelo
+ WARN_ON(clk_bulk_enable(cmdq->pdata->gce_num, cmdq->clocks));
task = kzalloc(sizeof(*task), GFP_ATOMIC);
if (!task) {
@@ -465,7 +472,8 @@ static void cmdq_mbox_shutdown(struct mbox_chan *chan)
struct cmdq_task *task, *tmp;
unsigned long flags;
- WARN_ON(pm_runtime_get_sync(cmdq->mbox.dev) < 0);
+ WARN_ON(pm_runtime_get(cmdq->mbox.dev) < 0);
+ WARN_ON(clk_bulk_enable(cmdq->pdata->gce_num, cmdq->clocks));
spin_lock_irqsave(&thread->chan->lock, flags);
if (list_empty(&thread->task_busy_list))
@@ -507,11 +515,9 @@ static int cmdq_mbox_flush(struct mbox_chan *chan, unsigned long timeout)
struct cmdq_task *task, *tmp;
unsigned long flags;
u32 enable;
- int ret;
- ret = pm_runtime_get_sync(cmdq->mbox.dev);
- if (ret < 0)
- return ret;
+ WARN_ON(pm_runtime_get(cmdq->mbox.dev) < 0);
+ WARN_ON(clk_bulk_enable(cmdq->pdata->gce_num, cmdq->clocks));
spin_lock_irqsave(&thread->chan->lock, flags);
if (list_empty(&thread->task_busy_list))
