On Tue, 20 Sep 2022, Sherry Sun wrote:
>
>
> > -----Original Message-----
> > From: Ilpo Järvinen <ilpo.jarvinen@xxxxxxxxxxxxxxx>
> > Sent: 2022年9月20日 20:26
> > To: Sherry Sun <sherry.sun@xxxxxxx>
> > Cc: gregkh@xxxxxxxxxxxxxxxxxxx; jirislaby@xxxxxxxxxx;
> > bhuvanchandra.dv@xxxxxxxxxxx; linux-serial@xxxxxxxxxxxxxxx; linux-
> > kernel@xxxxxxxxxxxxxxx; dl-linux-imx <linux-imx@xxxxxxx>
> > Subject: Re: [PATCH] tty: serial: fsl_lpuart: disable dma rx/tx use flags in
> > lpuart_dma_shutdown
> >
> > On Tue, 20 Sep 2022, Sherry Sun wrote:
> >
> > > lpuart_dma_shutdown tears down lpuart dma, but lpuart_flush_buffer can
> > > still occur which in turn tries to access dma apis if
> > > lpuart_dma_tx_use flag is true. At this point since dma is torn down,
> > > these dma apis can abort. Set lpuart_dma_tx_use and the corresponding
> > > rx flag lpuart_dma_rx_use to false in lpuart_dma_shutdown so that dmas
> > > are not accessed after they are relinquished.
> > >
> > > Otherwise, when try to kill btattach, kernel may panic. This patch may
> > > fix this issue.
> > > root@imx8ulpevk:~# btattach -B /dev/ttyLP2 -S 115200
> > > ^C[ 90.182296] Internal error: synchronous external abort: 96000210 [#1]
> > PREEMPT SMP
> > > [ 90.189806] Modules linked in: moal(O) mlan(O)
> > > [ 90.194258] CPU: 0 PID: 503 Comm: btattach Tainted: G O
> > 5.15.32-06136-g34eecdf2f9e4 #37
> > > [ 90.203554] Hardware name: NXP i.MX8ULP 9X9 EVK (DT)
> > > [ 90.208513] pstate: 600000c5 (nZCv daIF -PAN -UAO -TCO -DIT -SSBS
> > BTYPE=--)
> > > [ 90.215470] pc : fsl_edma3_disable_request+0x8/0x60
> > > [ 90.220358] lr : fsl_edma3_terminate_all+0x34/0x20c
> > > [ 90.225237] sp : ffff800013f0bac0
> > > [ 90.228548] x29: ffff800013f0bac0 x28: 0000000000000001 x27:
> > ffff000008404800
> > > [ 90.235681] x26: ffff000008404960 x25: ffff000008404a08 x24:
> > ffff000008404a00
> > > [ 90.242813] x23: ffff000008404a60 x22: 0000000000000002 x21:
> > 0000000000000000
> > > [ 90.249946] x20: ffff800013f0baf8 x19: ffff00000559c800 x18:
> > 0000000000000000
> > > [ 90.257078] x17: 0000000000000000 x16: 0000000000000000 x15:
> > 0000000000000000
> > > [ 90.264211] x14: 0000000000000003 x13: 0000000000000000 x12:
> > 0000000000000040
> > > [ 90.271344] x11: ffff00000600c248 x10: ffff800013f0bb10 x9 :
> > ffff000057bcb090
> > > [ 90.278477] x8 : fffffc0000241a08 x7 : ffff00000534ee00 x6 :
> > ffff000008404804
> > > [ 90.285609] x5 : 0000000000000000 x4 : 0000000000000000 x3 :
> > ffff0000055b3480
> > > [ 90.292742] x2 : ffff8000135c0000 x1 : ffff00000534ee00 x0 :
> > ffff00000559c800
> > > [ 90.299876] Call trace:
> > > [ 90.302321] fsl_edma3_disable_request+0x8/0x60
> > > [ 90.306851] lpuart_flush_buffer+0x40/0x160
> > > [ 90.311037] uart_flush_buffer+0x88/0x120
> > > [ 90.315050] tty_driver_flush_buffer+0x20/0x30
> > > [ 90.319496] hci_uart_flush+0x44/0x90
> > > [ 90.323162] +0x34/0x12c
> > > [ 90.327253] tty_ldisc_close+0x38/0x70
> > > [ 90.331005] tty_ldisc_release+0xa8/0x190
> > > [ 90.335018] tty_release_struct+0x24/0x8c
> > > [ 90.339022] tty_release+0x3ec/0x4c0
> > > [ 90.342593] __fput+0x70/0x234
> > > [ 90.345652] ____fput+0x14/0x20
> > > [ 90.348790] task_work_run+0x84/0x17c
> > > [ 90.352455] do_exit+0x310/0x96c
> > > [ 90.355688] do_group_exit+0x3c/0xa0
> > > [ 90.359259] __arm64_sys_exit_group+0x1c/0x20
> > > [ 90.363609] invoke_syscall+0x48/0x114
> > > [ 90.367362] el0_svc_common.constprop.0+0xd4/0xfc
> > > [ 90.372068] do_el0_svc+0x2c/0x94
> > > [ 90.375379] el0_svc+0x28/0x80
> > > [ 90.378438] el0t_64_sync_handler+0xa8/0x130
> > > [ 90.382711] el0t_64_sync+0x1a0/0x1a4
> > > [ 90.386376] Code: 17ffffda d503201f d503233f f9409802 (b9400041)
> > > [ 90.392467] ---[ end trace 2f60524b4a43f1f6 ]---
> > > [ 90.397073] note: btattach[503] exited with preempt_count 1
> > > [ 90.402636] Fixing recursive fault but reboot is needed!
> > >
> > > Fixes: 6250cc30c4c4 ("tty: serial: fsl_lpuart: Use scatter/gather DMA
> > > for Tx")
> > > Signed-off-by: Thara Gopinath <tgopinath@xxxxxxxxxxxxx>
> > > Signed-off-by: Sherry Sun <sherry.sun@xxxxxxx>
> > > ---
> > > drivers/tty/serial/fsl_lpuart.c | 2 ++
> > > 1 file changed, 2 insertions(+)
> > >
> > > diff --git a/drivers/tty/serial/fsl_lpuart.c
> > > b/drivers/tty/serial/fsl_lpuart.c index f21915015d67..064bd1f33c21
> > > 100644
> > > --- a/drivers/tty/serial/fsl_lpuart.c
> > > +++ b/drivers/tty/serial/fsl_lpuart.c
> > > @@ -1771,6 +1771,7 @@ static void lpuart_dma_shutdown(struct
> > lpuart_port *sport)
> > > if (sport->lpuart_dma_rx_use) {
> > > del_timer_sync(&sport->lpuart_timer);
> > > lpuart_dma_rx_free(&sport->port);
> > > + sport->lpuart_dma_rx_use = false;
> > > }
> > >
> > > if (sport->lpuart_dma_tx_use) {
> > > @@ -1779,6 +1780,7 @@ static void lpuart_dma_shutdown(struct
> > lpuart_port *sport)
> > > sport->dma_tx_in_progress = false;
> > > dmaengine_terminate_all(sport->dma_tx_chan);
> > > }
> > > + sport->lpuart_dma_tx_use = false;
> > > }
> > >
> > > if (sport->dma_tx_chan)
> >
> > Isn't it still racy because lpuart_dma_shutdown() is called outside of port's
> > lock?
>
> Hi Ilpo, do you think add the following changes should work for this?
>
> diff --git a/drivers/tty/serial/fsl_lpuart.c b/drivers/tty/serial/fsl_lpuart.c
> index 064bd1f33c21..b3c53935fbc8 100644
> --- a/drivers/tty/serial/fsl_lpuart.c
> +++ b/drivers/tty/serial/fsl_lpuart.c
> @@ -1768,10 +1768,14 @@ static int lpuart32_startup(struct uart_port *port)
>
> static void lpuart_dma_shutdown(struct lpuart_port *sport)
> {
> + unsigned long flags;
> +
> if (sport->lpuart_dma_rx_use) {
> del_timer_sync(&sport->lpuart_timer);
> lpuart_dma_rx_free(&sport->port);
> + spin_lock_irqsave(&sport->port.lock, flags);
> sport->lpuart_dma_rx_use = false;
> + spin_unlock_irqrestore(&sport->port.lock, flags);
> }
>
> if (sport->lpuart_dma_tx_use) {
> @@ -1780,7 +1784,9 @@ static void lpuart_dma_shutdown(struct lpuart_port *sport)
> sport->dma_tx_in_progress = false;
> dmaengine_terminate_all(sport->dma_tx_chan);
> }
> + spin_lock_irqsave(&sport->port.lock, flags);
> sport->lpuart_dma_tx_use = false;
> + spin_unlock_irqrestore(&sport->port.lock, flags);
> }
After reading a bit more, I think the original patch is likely fine since
shutdown doesn't seem to be racing with flush_buffer.
--
i.
