lkml.org 
[lkml]   [2021]   [Jul]   [22]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
SubjectRe: [PATCH] uart: mediatek: fix memory corruption issue
From
Date
On Thu, 2021-07-22 at 12:12 +0200, Greg KH wrote:
> On Thu, Jul 22, 2021 at 06:03:32PM +0800, zhiyong tao wrote:
> > On Wed, 2021-07-21 at 12:46 +0200, Greg KH wrote:
> > > On Sat, Jul 10, 2021 at 05:01:03PM +0800, Zhiyong Tao wrote:
> > > > This patch is used to fix memory corruption issue when rx power off.
> > > > 1. add spin lock in mtk8250_dma_rx_complete function in APDMA mode.
> > >
> > > What does a lock protect from? Please be explicit and detailed.
> >
> > ==> Hi Gregkh,
> >
> > when uart is used as a communication port with external device(GPS).
> > when external device(GPS) power off, the power of rx pin is also from
> > 1.8v to 0v. Even if there is not any data in rx. But uart rx pin can
> > capture the data "0".
>
> That sounds like a broken hardware design.
>
> > If uart don't receive any data in specified cycle, uart will generates
> > BI(Break interrupt) interrupt.
> > If external device(GPS) power off, we found that BI interrupt appeared
> > continuously and very frequently.
> > When uart interrupt type is BI, uart IRQ handler(8250 framwork
> > API:serial8250_handle_irq) will push data to tty buffer.
> > The code path:
> > https://elixir.bootlin.com/linux/latest/source/drivers/tty/serial/8250/8250_port.c#L1917
> >
> > mtk8250_dma_rx_complete is a task of mtk_uart_apdma_rx_handler.
> > mtk8250_dma_rx_complete priority is lower than uart irq
> > handler(serial8250_handle_irq).
> > if we are in process of mtk8250_dma_rx_complete, uart appear BI
> > interrupt:1)serial8250_handle_irq will priority execution.2)it may cause
> > write tty buffer conflict in mtk8250_dma_rx_complete.
> > So the spin lock protect the rx receive data process is not break.
>
> Then put something like this in the changelog text, as it is, it is not
> descriptive at all.

==>
Thanks for your suggestion.
we will add it in changelog text([PATCH 0/1] Mediatek uart patch) in v1.

>
> > > > 2. add processing mechanism which count value is 0
> > >
> > > What does this do? And why is it needed?
> >
> > ==> when count value is 0, we don't need push data to tty buffer.
> > so we add it.
>
> But that does not actually do anything different from what it does
> today. And it has nothing to do with the lock, so this should be 2
> different patches, right?
==>
yes, it has nothing to do with the lock. we will separate it in v1.

>
> > > > Signed-off-by: Zhiyong Tao <zhiyong.tao@mediatek.com>
> > >
> > > What commit does this fix? Does this need to go to stable kernel trees?
> > > If so, how far back?
> > >
> > > > ---
> > > > drivers/tty/serial/8250/8250_mtk.c | 15 +++++++++++----
> > > > 1 file changed, 11 insertions(+), 4 deletions(-)
> > > >
> > > > diff --git a/drivers/tty/serial/8250/8250_mtk.c b/drivers/tty/serial/8250/8250_mtk.c
> > > > index f7d3023f860f..09f7d2166315 100644
> > > > --- a/drivers/tty/serial/8250/8250_mtk.c
> > > > +++ b/drivers/tty/serial/8250/8250_mtk.c
> > > > @@ -91,12 +91,15 @@ static void mtk8250_dma_rx_complete(void *param)
> > > > struct mtk8250_data *data = up->port.private_data;
> > > > struct tty_port *tty_port = &up->port.state->port;
> > > > struct dma_tx_state state;
> > > > - int copied, total, cnt;
> > > > + unsigned int copied, total, cnt;
> > > > unsigned char *ptr;
> > > > + unsigned long flags;
> > > >
> > > > if (data->rx_status == DMA_RX_SHUTDOWN)
> > > > return;
> > > >
> > > > + spin_lock_irqsave(&up->port.lock, flags);
> > > > +
> > > > dmaengine_tx_status(dma->rxchan, dma->rx_cookie, &state);
> > > > total = dma->rx_size - state.residue;
> > > > cnt = total;
> > > > @@ -104,9 +107,11 @@ static void mtk8250_dma_rx_complete(void *param)
> > > > if ((data->rx_pos + cnt) > dma->rx_size)
> > > > cnt = dma->rx_size - data->rx_pos;
> > > >
> > > > - ptr = (unsigned char *)(data->rx_pos + dma->rx_buf);
> > > > - copied = tty_insert_flip_string(tty_port, ptr, cnt);
> > > > - data->rx_pos += cnt;
> > > > + if (cnt != 0) {
> > >
> > > Why does cnt matter here? If cnt is 0, the code above should not do
> > > anything at all, right?
> >
> > ==> yes, if the counter value is 0, we don't need push data to the tty
> > buffer.
>
> But this does not change the logic as if cnt is 0, nothing gets pushed
> with the current code either, right?
==>
yes, it is.

>
> > > Or if it does, should we change tty_insert_flip_string() to always check
> > > for cnt != 0 before it does the first loop? Hm, it looks like it will
> > > abort if cnt is 0, so what is this change really doing? Why do you need
> > > it? What is it "fixing"?
> > >
> > ==> It is not fix anything, we just think if count value is 0, we don't
> > need do anything.
>
> Then make it a separate patch, independant from the lock patch, and we
> can discuss it there. Do NOT have patches do multiple things.
==>
I will make it a separate patch in v1.

Thanks.
>
> thanks,
>
> greg k-h

\
 
 \ /
  Last update: 2021-07-22 19:34    [W:0.067 / U:0.824 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site