[PATCH v2 0/2] fix race condition in rsnd_ssi_pointer_update

[PATCH v2 0/2] fix race condition in rsnd_ssi_pointer_update

[jiada_wang]

From: Jiada Wang <jiada_wang@mentor.com>

This patch set aims to fix the race condition in rsnd_ssi_pointer_update,
between set of byte_pos and wrap it around when new buffer starts.
 
Jiada Wang (2):
  ASoC: rsnd: ssi: fix race condition in rsnd_ssi_pointer_update
  ASoC: rsnd: ssi: remove unnesessary period_pos

Changes for v2:
 * Updated commit message

 sound/soc/sh/rcar/ssi.c |   25 +++++++++++++------------
 1 file changed, 13 insertions(+), 12 deletions(-)

-- 
1.7.9.5

[PATCH v2 1/2] ASoC: rsnd: ssi: fix race condition in rsnd_ssi_pointer_update

[jiada_wang]

From: Jiada Wang <jiada_wang@mentor.com>

Currently there is race condition between set of byte_pos and wrap
it around when new buffer starts. If .pointer is called in-between
it will result in inconsistent pointer position be returned
from .pointer callback.

This patch increments buffer pointer atomically to avoid this issue.

Signed-off-by: Jiada Wang <jiada_wang@mentor.com>
Reviewed-by: Takashi Sakamoto <takashi.sakamoto@miraclelinux.com>
---
 sound/soc/sh/rcar/ssi.c |   16 ++++++++++------
 1 file changed, 10 insertions(+), 6 deletions(-)

diff --git a/sound/soc/sh/rcar/ssi.c b/sound/soc/sh/rcar/ssi.c
index fece1e5f..cbf3bf3 100644
--- a/sound/soc/sh/rcar/ssi.c
+++ b/sound/soc/sh/rcar/ssi.c
@@ -446,25 +446,29 @@ static bool rsnd_ssi_pointer_update(struct rsnd_mod *mod,
 				    int byte)
 {
 	struct rsnd_ssi *ssi = rsnd_mod_to_ssi(mod);
+	bool ret = false;
+	int byte_pos;
 
-	ssi->byte_pos += byte;
+	byte_pos = ssi->byte_pos + byte;
 
-	if (ssi->byte_pos >= ssi->next_period_byte) {
+	if (byte_pos >= ssi->next_period_byte) {
 		struct snd_pcm_runtime *runtime = rsnd_io_to_runtime(io);
 
 		ssi->period_pos++;
 		ssi->next_period_byte += ssi->byte_per_period;
 
 		if (ssi->period_pos >= runtime->periods) {
-			ssi->byte_pos = 0;
+			byte_pos = 0;
 			ssi->period_pos = 0;
 			ssi->next_period_byte = ssi->byte_per_period;
 		}
 
-		return true;
+		ret = true;
 	}
 
-	return false;
+	WRITE_ONCE(ssi->byte_pos, byte_pos);
+
+	return ret;
 }
 
 /*
@@ -838,7 +842,7 @@ static int rsnd_ssi_pointer(struct rsnd_mod *mod,
 	struct rsnd_ssi *ssi = rsnd_mod_to_ssi(mod);
 	struct snd_pcm_runtime *runtime = rsnd_io_to_runtime(io);
 
-	*pointer = bytes_to_frames(runtime, ssi->byte_pos);
+	*pointer = bytes_to_frames(runtime, READ_ONCE(ssi->byte_pos));
 
 	return 0;
 }
-- 
1.7.9.5

[PATCH v2 2/2] ASoC: rsnd: ssi: remove unnesessary period_pos

[jiada_wang]

From: Jiada Wang <jiada_wang@mentor.com>

period_pos can always be calculated by byte_pos and
byte_per_period, there is no reason to maintain this
variable in rsnd_dai_stream.

This patch removes period_pos from rsnd_ssi and calculates
next_period_byte with consideration of actual byte_pos value.

Signed-off-by: Jiada Wang <jiada_wang@mentor.com>
---
 sound/soc/sh/rcar/ssi.c |    9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)

diff --git a/sound/soc/sh/rcar/ssi.c b/sound/soc/sh/rcar/ssi.c
index cbf3bf3..f212024 100644
--- a/sound/soc/sh/rcar/ssi.c
+++ b/sound/soc/sh/rcar/ssi.c
@@ -80,7 +80,6 @@ struct rsnd_ssi {
 	unsigned int usrcnt;
 
 	int byte_pos;
-	int period_pos;
 	int byte_per_period;
 	int next_period_byte;
 };
@@ -421,7 +420,6 @@ static void rsnd_ssi_pointer_init(struct rsnd_mod *mod,
 	struct snd_pcm_runtime *runtime = rsnd_io_to_runtime(io);
 
 	ssi->byte_pos		= 0;
-	ssi->period_pos		= 0;
 	ssi->byte_per_period	= runtime->period_size *
 				  runtime->channels *
 				  samples_to_bytes(runtime, 1);
@@ -453,13 +451,12 @@ static bool rsnd_ssi_pointer_update(struct rsnd_mod *mod,
 
 	if (byte_pos >= ssi->next_period_byte) {
 		struct snd_pcm_runtime *runtime = rsnd_io_to_runtime(io);
+		int period_pos = byte_pos / ssi->byte_per_period;
 
-		ssi->period_pos++;
-		ssi->next_period_byte += ssi->byte_per_period;
+		ssi->next_period_byte = (period_pos + 1) * ssi->byte_per_period;
 
-		if (ssi->period_pos >= runtime->periods) {
+		if (period_pos >= runtime->periods) {
 			byte_pos = 0;
-			ssi->period_pos = 0;
 			ssi->next_period_byte = ssi->byte_per_period;
 		}
 
-- 
1.7.9.5

Re: [PATCH v2 0/2] fix race condition in rsnd_ssi_pointer_update

[Kuninori Morimoto]

Hi

> This patch set aims to fix the race condition in rsnd_ssi_pointer_update,
> between set of byte_pos and wrap it around when new buffer starts.
>  
> Jiada Wang (2):
>   ASoC: rsnd: ssi: fix race condition in rsnd_ssi_pointer_update
>   ASoC: rsnd: ssi: remove unnesessary period_pos
> 
> Changes for v2:
>  * Updated commit message
> 
>  sound/soc/sh/rcar/ssi.c |   25 +++++++++++++------------
>  1 file changed, 13 insertions(+), 12 deletions(-)

for all patches

Acked-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>

Re: [PATCH v2 2/2] ASoC: rsnd: ssi: remove unnesessary period_pos

[Kuninori Morimoto]

Hi Jiada

Thank you for your feedback

> >> Further more, if the passed 'byte' amount to
> >> rsnd_ssi_pointer_update() is more than byte_per_period.
> >> the calculation of next_period_byte isn't correct.
> > Is it really happen ??
> >
> > Basically, I have no objection about this patch,
> > but this explanation is very strange for me...
> No, I didn't see the issue,
> but the implementation of rsnd_ssi_pointer_update(), behaves like
> it knows all caller will always pass 'byte' no larger than byte_per_period,
> without any check internally.
> 
> I am ok to remove this explanation from commit message,
> what do you think?

This function is used from PIO mode only now, and "byte" is sizeof(u32)
(Its size was "byte_per_period" when DMA mode used).
This "Further more" case never happen.
Removing from commit message is better for reader, IMO.

Best regards
---
Kuninori Morimoto

Re: [PATCH v2 2/2] ASoC: rsnd: ssi: remove unnesessary period_pos

[Jiada Wang]

Hi Morimoto-san

On 12/07/2017 01:58 AM, Kuninori Morimoto wrote:
> Hi Jiada
>
>> Further more, if the passed 'byte' amount to
>> rsnd_ssi_pointer_update() is more than byte_per_period.
>> the calculation of next_period_byte isn't correct.
> Is it really happen ??
>
> Basically, I have no objection about this patch,
> but this explanation is very strange for me...
No, I didn't see the issue,
but the implementation of rsnd_ssi_pointer_update(), behaves like
it knows all caller will always pass 'byte' no larger than byte_per_period,
without any check internally.

I am ok to remove this explanation from commit message,
what do you think?

Thanks,
Jiada
> Best regards
> ---
> Kuninori Morimoto

Re: [PATCH v2 2/2] ASoC: rsnd: ssi: remove unnesessary period_pos

[Kuninori Morimoto]

Hi Jiada

> Further more, if the passed 'byte' amount to
> rsnd_ssi_pointer_update() is more than byte_per_period.
> the calculation of next_period_byte isn't correct.

Is it really happen ??

Basically, I have no objection about this patch,
but this explanation is very strange for me...

Best regards
---
Kuninori Morimoto

[PATCH v2 2/2] ASoC: rsnd: ssi: remove unnesessary period_pos

[jiada_wang]

From: Jiada Wang <jiada_wang@mentor.com>

period_pos can always be calculated by byte_pos and
byte_per_period, there is no reason to maintain this
variable in rsnd_dai_stream.
Further more, if the passed 'byte' amount to
rsnd_ssi_pointer_update() is more than byte_per_period.
the calculation of next_period_byte isn't correct.

This patch removes period_pos from rsnd_ssi and calculates
next_period_byte with consideration of actual byte_pos value.

Signed-off-by: Jiada Wang <jiada_wang@mentor.com>
---
 sound/soc/sh/rcar/ssi.c |    9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)

diff --git a/sound/soc/sh/rcar/ssi.c b/sound/soc/sh/rcar/ssi.c
index cbf3bf3..f212024 100644
--- a/sound/soc/sh/rcar/ssi.c
+++ b/sound/soc/sh/rcar/ssi.c
@@ -80,7 +80,6 @@ struct rsnd_ssi {
 	unsigned int usrcnt;
 
 	int byte_pos;
-	int period_pos;
 	int byte_per_period;
 	int next_period_byte;
 };
@@ -421,7 +420,6 @@ static void rsnd_ssi_pointer_init(struct rsnd_mod *mod,
 	struct snd_pcm_runtime *runtime = rsnd_io_to_runtime(io);
 
 	ssi->byte_pos		= 0;
-	ssi->period_pos		= 0;
 	ssi->byte_per_period	= runtime->period_size *
 				  runtime->channels *
 				  samples_to_bytes(runtime, 1);
@@ -453,13 +451,12 @@ static bool rsnd_ssi_pointer_update(struct rsnd_mod *mod,
 
 	if (byte_pos >= ssi->next_period_byte) {
 		struct snd_pcm_runtime *runtime = rsnd_io_to_runtime(io);
+		int period_pos = byte_pos / ssi->byte_per_period;
 
-		ssi->period_pos++;
-		ssi->next_period_byte += ssi->byte_per_period;
+		ssi->next_period_byte = (period_pos + 1) * ssi->byte_per_period;
 
-		if (ssi->period_pos >= runtime->periods) {
+		if (period_pos >= runtime->periods) {
 			byte_pos = 0;
-			ssi->period_pos = 0;
 			ssi->next_period_byte = ssi->byte_per_period;
 		}
 
-- 
1.7.9.5