2021-06-11 09:05:08

by Takashi Iwai

Subject: [PATCH] mmc: sdhci: Clear unused bounce buffer at DMA mmap error path

When DMA-mapping of the bounce buffer fails, the driver tries to fall
back, but it leaves the allocated host->bounce_buffer although its
size is zero. Later on, the driver checks the use of bounce buffer
with host->bounce_buffer pointer, and it tries to use the buffer
incorrectly, resulting in Oops.

This patch clears the release the unused buffer and clears the
bounce_buffer pointer for addressing the problem.

Signed-off-by: Takashi Iwai <[email protected]>
---
drivers/mmc/host/sdhci.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/mmc/host/sdhci.c b/drivers/mmc/host/sdhci.c
index bf238ade1602..5f467b98ca88 100644
--- a/drivers/mmc/host/sdhci.c
+++ b/drivers/mmc/host/sdhci.c
@@ -4072,9 +4072,13 @@ static void sdhci_allocate_bounce_buffer(struct sdhci_host *host)
bounce_size,
DMA_BIDIRECTIONAL);
ret = dma_mapping_error(mmc_dev(mmc), host->bounce_addr);
- if (ret)
+ if (ret) {
+ devm_kfree(mmc->parent, host->bounce_buffer);
+ host->bounce_buffer = NULL;
/* Again fall back to max_segs == 1 */
return;
+ }
+
host->bounce_buffer_size = bounce_size;

/* Lie about this since we're bouncing */
--
2.26.2

2021-06-11 09:59:39

by Adrian Hunter

Subject: Re: [PATCH] mmc: sdhci: Clear unused bounce buffer at DMA mmap error path

On 11/06/21 12:03 pm, Takashi Iwai wrote:
> When DMA-mapping of the bounce buffer fails, the driver tries to fall
> back, but it leaves the allocated host->bounce_buffer although its
> size is zero. Later on, the driver checks the use of bounce buffer
> with host->bounce_buffer pointer, and it tries to use the buffer
> incorrectly, resulting in Oops.
>
> This patch clears the release the unused buffer and clears the
> bounce_buffer pointer for addressing the problem.
>
> Signed-off-by: Takashi Iwai <[email protected]>

One minor comment below, otherwise:

Acked-by: Adrian Hunter <[email protected]>

> ---
> drivers/mmc/host/sdhci.c | 6 +++++-
> 1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/mmc/host/sdhci.c b/drivers/mmc/host/sdhci.c
> index bf238ade1602..5f467b98ca88 100644
> --- a/drivers/mmc/host/sdhci.c
> +++ b/drivers/mmc/host/sdhci.c
> @@ -4072,9 +4072,13 @@ static void sdhci_allocate_bounce_buffer(struct sdhci_host *host)
> bounce_size,
> DMA_BIDIRECTIONAL);
> ret = dma_mapping_error(mmc_dev(mmc), host->bounce_addr);
> - if (ret)
> + if (ret) {
> + devm_kfree(mmc->parent, host->bounce_buffer);

Everywhere else in this function mmc_dev(mmc) is used, so maybe
use it here too instead of mmc->parent.

> + host->bounce_buffer = NULL;
> /* Again fall back to max_segs == 1 */
> return;
> + }
> +
> host->bounce_buffer_size = bounce_size;
>
> /* Lie about this since we're bouncing */
>

2021-06-11 10:08:21

by Takashi Iwai

Subject: Re: [PATCH] mmc: sdhci: Clear unused bounce buffer at DMA mmap error path

On Fri, 11 Jun 2021 11:56:16 +0200,
Adrian Hunter wrote:
>
> On 11/06/21 12:03 pm, Takashi Iwai wrote:
> > When DMA-mapping of the bounce buffer fails, the driver tries to fall
> > back, but it leaves the allocated host->bounce_buffer although its
> > size is zero. Later on, the driver checks the use of bounce buffer
> > with host->bounce_buffer pointer, and it tries to use the buffer
> > incorrectly, resulting in Oops.
> >
> > This patch clears the release the unused buffer and clears the
> > bounce_buffer pointer for addressing the problem.
> >
> > Signed-off-by: Takashi Iwai <[email protected]>
>
> One minor comment below, otherwise:
>
> Acked-by: Adrian Hunter <[email protected]>
>
> > ---
> > drivers/mmc/host/sdhci.c | 6 +++++-
> > 1 file changed, 5 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/mmc/host/sdhci.c b/drivers/mmc/host/sdhci.c
> > index bf238ade1602..5f467b98ca88 100644
> > --- a/drivers/mmc/host/sdhci.c
> > +++ b/drivers/mmc/host/sdhci.c
> > @@ -4072,9 +4072,13 @@ static void sdhci_allocate_bounce_buffer(struct sdhci_host *host)
> > bounce_size,
> > DMA_BIDIRECTIONAL);
> > ret = dma_mapping_error(mmc_dev(mmc), host->bounce_addr);
> > - if (ret)
> > + if (ret) {
> > + devm_kfree(mmc->parent, host->bounce_buffer);
>
> Everywhere else in this function mmc_dev(mmc) is used, so maybe
> use it here too instead of mmc->parent.

Doh, right, I wrote a patch based on the older kernel and refreshed
afterwards, but I forgot correcting this place. Will resubmit the
fixed version.


thanks,

Takashi