Re: [PATCH] tty: serial: fixed uart irq maybe cause irq storm

[Andy Shevchenko]

On Mon, Apr 01, 2024 at 06:19:06PM +0800, fengchunguo@xxxxxxx wrote:
> From: "Gary Feng" <chunguo.feng@xxxxxxxxxxxxx>
> 
> if not disable uart irq before disable clk,  uart irq maybe triggered after
> disabled clk immediately, then maybe cause irq storm.
> 
> Reproduced the below call trace, see i2c not be connected, but irq storm
> was triggered.

> i2c_designware 30b70000.i2c: controller timed out
> CPU: 2 PID: 2932 Comm: gnss@1.0-servic 
> Tainted: G O 5.14.61-00094-geaa0149416cc-dirty #8
> Hardware name: Semidrive kunlun x9 REF Board (DT)
> Call trace:
> [<ffff00000808a3cc>] dump_backtrace+0x0/0x3c0
> [<ffff00000808a7a0>] show_stack+0x14/0x1c
> [<ffff000008cef43c>] dump_stack+0xc4/0xfc
> [<ffff00000814eb80>] __report_bad_irq+0x50/0xe0
> [<ffff00000814eaec>] note_interrupt+0x248/0x28c
> [<ffff00000814c0e8>] handle_irq_event+0x78/0xa4
> [<ffff00000814fcb8>] handle_fasteoi_irq+0xe4/0x1b4
> [<ffff00000814b060>] __handle_domain_irq+0x7c/0xbc
> [<ffff00000808176c>] gic_handle_irq+0x4c/0xb8
> [<ffff000008083230>] el1_irq+0xb0/0x124
> [<ffff000008d09f5c>] _raw_spin_unlock_irqrestore+0x10/0x44
> [<ffff00000865b784>] dw8250_set_termios+0x48/0xf4
> [<ffff0000086545a4>] serial8250_set_termios+0x14/0x28
> [<ffff00000864c4f4>] uart_change_speed+0x38/0x10c
> [<ffff00000864e458>] uart_set_termios+0xd0/0x17c
> [<ffff000008630ad4>] tty_set_termios+0x160/0x1e4
> [<ffff00000863165c>] set_termios+0x32c/0x3bc
> [<ffff000008631248>] tty_mode_ioctl+0x6f0/0x7d8
> [<ffff000008631a6c>] n_tty_ioctl_helper+0x10c/0x1e8
> [<ffff00000862d2c4>] n_tty_ioctl+0x120/0x194
> [<ffff00000862a724>] tty_ioctl+0x658/0xa34
> [<ffff0000082a8f40>] do_vfs_ioctl+0x554/0x810
> [<ffff0000082a9368>] SyS_ioctl+0x88/0x94
> Exception stack(0xffff00000ccf3ec0 to 0xffff00000ccf4000

Please read
https://www.kernel.org/doc/html/latest/process/submitting-patches.html#backtraces-in-commit-messages
and update the commit message accordingly.

..

> +		/*Need disable uart irq before disabled clk, because uart irq maybe triggered after
> +		 * disabled clk immediately, then cause irq storm.
> +		 */


/*
 * Use canonical multi-line comment
 * style and make sure the lines are not
 * too long.
 */

..

FWIW, the patch that compares this with current clock rate most likely will be
reverted, meaning your fix might be not needed anymore, or has to be rebased.

-- 
With Best Regards,
Andy Shevchenko