2017-12-16 11:51:47

by Jia-Ju Bai

Subject: [PATCH V2 1/2] bdisp: Fix a possible sleep-in-atomic bug in bdisp_hw_reset

The driver may sleep under a spinlock.
The function call path is:
bdisp_device_run (acquire the spinlock)
bdisp_hw_reset
msleep --> may sleep

To fix it, readl_poll_timeout_atomic is used to replace msleep.

This bug is found by my static analysis tool(DSAC) and
checked by my code review.

Signed-off-by: Jia-Ju Bai <[email protected]>
---
drivers/media/platform/sti/bdisp/bdisp-hw.c | 16 ++++++++--------
1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/drivers/media/platform/sti/bdisp/bdisp-hw.c b/drivers/media/platform/sti/bdisp/bdisp-hw.c
index b7892f3..e94a371 100644
--- a/drivers/media/platform/sti/bdisp/bdisp-hw.c
+++ b/drivers/media/platform/sti/bdisp/bdisp-hw.c
@@ -5,6 +5,7 @@
*/

#include <linux/delay.h>
+#include <linux/iopoll.h>

#include "bdisp.h"
#include "bdisp-filter.h"
@@ -366,7 +367,7 @@ struct bdisp_filter_addr {
*/
int bdisp_hw_reset(struct bdisp_dev *bdisp)
{
- unsigned int i;
+ u32 tmp;

dev_dbg(bdisp->dev, "%s\n", __func__);

@@ -379,15 +380,14 @@ int bdisp_hw_reset(struct bdisp_dev *bdisp)
writel(0, bdisp->regs + BLT_CTL);

/* Wait for reset done */
- for (i = 0; i < POLL_RST_MAX; i++) {
- if (readl(bdisp->regs + BLT_STA1) & BLT_STA1_IDLE)
- break;
- msleep(POLL_RST_DELAY_MS);
- }
- if (i == POLL_RST_MAX)
+ if (readl_poll_timeout_atomic(bdisp->regs + BLT_STA1, tmp,
+ (tmp & BLT_STA1_IDLE), POLL_RST_DELAY_MS,
+ POLL_RST_DELAY_MS * POLL_RST_MAX)) {
dev_err(bdisp->dev, "Reset timeout\n");
+ return -EAGAIN;
+ }

- return (i == POLL_RST_MAX) ? -EAGAIN : 0;
+ return 0;
}

/**
--
1.7.9.5

2017-12-19 10:43:30

by Fabien DESSENNE

Subject: Re: [PATCH V2 1/2] bdisp: Fix a possible sleep-in-atomic bug in bdisp_hw_reset

Hi,


On 16/12/17 12:54, Jia-Ju Bai wrote:
> The driver may sleep under a spinlock.
> The function call path is:
> bdisp_device_run (acquire the spinlock)
> bdisp_hw_reset
> msleep --> may sleep
>
> To fix it, readl_poll_timeout_atomic is used to replace msleep.
>
> This bug is found by my static analysis tool(DSAC) and
> checked by my code review.
>
> Signed-off-by: Jia-Ju Bai <[email protected]>
> ---
> drivers/media/platform/sti/bdisp/bdisp-hw.c | 16 ++++++++--------
> 1 file changed, 8 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/media/platform/sti/bdisp/bdisp-hw.c b/drivers/media/platform/sti/bdisp/bdisp-hw.c
> index b7892f3..e94a371 100644
> --- a/drivers/media/platform/sti/bdisp/bdisp-hw.c
> +++ b/drivers/media/platform/sti/bdisp/bdisp-hw.c
> @@ -5,6 +5,7 @@
> */
>
> #include <linux/delay.h>

This delay.h include is no more needed, remove it.

> +#include <linux/iopoll.h>
>
> #include "bdisp.h"
> #include "bdisp-filter.h"
> @@ -366,7 +367,7 @@ struct bdisp_filter_addr {
> */
> int bdisp_hw_reset(struct bdisp_dev *bdisp)
> {
> - unsigned int i;
> + u32 tmp;
>
> dev_dbg(bdisp->dev, "%s\n", __func__);
>
> @@ -379,15 +380,14 @@ int bdisp_hw_reset(struct bdisp_dev *bdisp)
> writel(0, bdisp->regs + BLT_CTL);
>
> /* Wait for reset done */
> - for (i = 0; i < POLL_RST_MAX; i++) {
> - if (readl(bdisp->regs + BLT_STA1) & BLT_STA1_IDLE)
> - break;
> - msleep(POLL_RST_DELAY_MS);
> - }
> - if (i == POLL_RST_MAX)

As recommended by Mauro, please add this comment:
Despite the large timeout, most of the time the reset happens without
needing any delays

> + if (readl_poll_timeout_atomic(bdisp->regs + BLT_STA1, tmp,
> + (tmp & BLT_STA1_IDLE), POLL_RST_DELAY_MS,
> + POLL_RST_DELAY_MS * POLL_RST_MAX)) {

read_poll_timeout expects US timings, not MS.

> dev_err(bdisp->dev, "Reset timeout\n");
> + return -EAGAIN;
> + }
>
> - return (i == POLL_RST_MAX) ? -EAGAIN : 0;
> + return 0;
> }
>
> /**

2017-12-19 13:57:40

by Jia-Ju Bai

Subject: Re: [PATCH V2 1/2] bdisp: Fix a possible sleep-in-atomic bug in bdisp_hw_reset

On 2017/12/19 18:43, Fabien DESSENNE wrote:
> Hi,
>
>
> On 16/12/17 12:54, Jia-Ju Bai wrote:
>> The driver may sleep under a spinlock.
>> The function call path is:
>> bdisp_device_run (acquire the spinlock)
>> bdisp_hw_reset
>> msleep --> may sleep
>>
>> To fix it, readl_poll_timeout_atomic is used to replace msleep.
>>
>> This bug is found by my static analysis tool(DSAC) and
>> checked by my code review.
>>
>> Signed-off-by: Jia-Ju Bai <[email protected]>
>> ---
>> drivers/media/platform/sti/bdisp/bdisp-hw.c | 16 ++++++++--------
>> 1 file changed, 8 insertions(+), 8 deletions(-)
>>
>> diff --git a/drivers/media/platform/sti/bdisp/bdisp-hw.c b/drivers/media/platform/sti/bdisp/bdisp-hw.c
>> index b7892f3..e94a371 100644
>> --- a/drivers/media/platform/sti/bdisp/bdisp-hw.c
>> +++ b/drivers/media/platform/sti/bdisp/bdisp-hw.c
>> @@ -5,6 +5,7 @@
>> */
>>
>> #include <linux/delay.h>
> This delay.h include is no more needed, remove it.
>
>> +#include <linux/iopoll.h>
>>
>> #include "bdisp.h"
>> #include "bdisp-filter.h"
>> @@ -366,7 +367,7 @@ struct bdisp_filter_addr {
>> */
>> int bdisp_hw_reset(struct bdisp_dev *bdisp)
>> {
>> - unsigned int i;
>> + u32 tmp;
>>
>> dev_dbg(bdisp->dev, "%s\n", __func__);
>>
>> @@ -379,15 +380,14 @@ int bdisp_hw_reset(struct bdisp_dev *bdisp)
>> writel(0, bdisp->regs + BLT_CTL);
>>
>> /* Wait for reset done */
>> - for (i = 0; i < POLL_RST_MAX; i++) {
>> - if (readl(bdisp->regs + BLT_STA1) & BLT_STA1_IDLE)
>> - break;
>> - msleep(POLL_RST_DELAY_MS);
>> - }
>> - if (i == POLL_RST_MAX)
> As recommended by Mauro, please add this comment:
> Despite the large timeout, most of the time the reset happens without
> needing any delays
>
>> + if (readl_poll_timeout_atomic(bdisp->regs + BLT_STA1, tmp,
>> + (tmp & BLT_STA1_IDLE), POLL_RST_DELAY_MS,
>> + POLL_RST_DELAY_MS * POLL_RST_MAX)) {
> read_poll_timeout expects US timings, not MS.
>
>> dev_err(bdisp->dev, "Reset timeout\n");
>> + return -EAGAIN;
>> + }
>>
>> - return (i == POLL_RST_MAX) ? -EAGAIN : 0;
>> + return 0;
>> }
>>
>> /**

Hi,

Okay, I have submitted a new patch according to your advice :)


Thanks,
Jia-Ju Bai