[PATCH stable-6.10 regression] Revert "soundwire: stream: fix programming slave ports for non-continous port maps"

[PATCH stable-6.10 regression] Revert "soundwire: stream: fix programming slave ports for non-continous port maps"

[Peter Ujfalusi]

The prop->src_dpn_prop and prop.sink_dpn_prop is allocated for the _number_
of ports and it is forced as 0 index based.

The original code was correct while the change to walk the bits and use
their position as index into the arrays is not correct.

For exmple we can have the prop.source_ports=0x2, which means we have one
port, but the prop.src_dpn_prop[1] is accessing outside of the allocated
memory.

This reverts commit 6fa78e9c41471fe43052cd6feba6eae1b0277ae3.

Cc: stable@vger.kernel.org # 6.10.y
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
---
Hi,

The reverted patch causes major regression on soundwire causing all audio
to fail.
Interestingly the patch is only in 6.10.8 and 6.10.9, not in mainline or linux-next.

soundwire sdw-master-0-1: Program transport params failed: -22
soundwire sdw-master-0-1: Program params failed: -22
SDW1-Playback: ASoC: error at snd_soc_link_prepare on SDW1-Playback: -22

Regards,
Peter 

 drivers/soundwire/stream.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/drivers/soundwire/stream.c b/drivers/soundwire/stream.c
index 00191b1d2260..4e9e7d2a942d 100644
--- a/drivers/soundwire/stream.c
+++ b/drivers/soundwire/stream.c
@@ -1286,18 +1286,18 @@ struct sdw_dpn_prop *sdw_get_slave_dpn_prop(struct sdw_slave *slave,
 					    unsigned int port_num)
 {
 	struct sdw_dpn_prop *dpn_prop;
-	unsigned long mask;
+	u8 num_ports;
 	int i;
 
 	if (direction == SDW_DATA_DIR_TX) {
-		mask = slave->prop.source_ports;
+		num_ports = hweight32(slave->prop.source_ports);
 		dpn_prop = slave->prop.src_dpn_prop;
 	} else {
-		mask = slave->prop.sink_ports;
+		num_ports = hweight32(slave->prop.sink_ports);
 		dpn_prop = slave->prop.sink_dpn_prop;
 	}
 
-	for_each_set_bit(i, &mask, 32) {
+	for (i = 0; i < num_ports; i++) {
 		if (dpn_prop[i].num == port_num)
 			return &dpn_prop[i];
 	}
-- 
2.46.0

Re: [PATCH stable-6.10 regression] Revert "soundwire: stream: fix programming slave ports for non-continous port maps"

[Péter Ujfalusi]

Hi,

On 10/09/2024 15:40, Peter Ujfalusi wrote:
> The prop->src_dpn_prop and prop.sink_dpn_prop is allocated for the _number_
> of ports and it is forced as 0 index based.
> 
> The original code was correct while the change to walk the bits and use
> their position as index into the arrays is not correct.
> 
> For exmple we can have the prop.source_ports=0x2, which means we have one
> port, but the prop.src_dpn_prop[1] is accessing outside of the allocated
> memory.
> 
> This reverts commit 6fa78e9c41471fe43052cd6feba6eae1b0277ae3.

I just noticed that Krzysztof already sent the revert patch but it is
not picked up for stable-6.10.y

https://lore.kernel.org/lkml/20240909164746.136629-1-krzysztof.kozlowski@linaro.org/

> 
> Cc: stable@vger.kernel.org # 6.10.y
> Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
> ---
> Hi,
> 
> The reverted patch causes major regression on soundwire causing all audio
> to fail.
> Interestingly the patch is only in 6.10.8 and 6.10.9, not in mainline or linux-next.
> 
> soundwire sdw-master-0-1: Program transport params failed: -22
> soundwire sdw-master-0-1: Program params failed: -22
> SDW1-Playback: ASoC: error at snd_soc_link_prepare on SDW1-Playback: -22
> 
> Regards,
> Peter 
> 
>  drivers/soundwire/stream.c | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/soundwire/stream.c b/drivers/soundwire/stream.c
> index 00191b1d2260..4e9e7d2a942d 100644
> --- a/drivers/soundwire/stream.c
> +++ b/drivers/soundwire/stream.c
> @@ -1286,18 +1286,18 @@ struct sdw_dpn_prop *sdw_get_slave_dpn_prop(struct sdw_slave *slave,
>  					    unsigned int port_num)
>  {
>  	struct sdw_dpn_prop *dpn_prop;
> -	unsigned long mask;
> +	u8 num_ports;
>  	int i;
>  
>  	if (direction == SDW_DATA_DIR_TX) {
> -		mask = slave->prop.source_ports;
> +		num_ports = hweight32(slave->prop.source_ports);
>  		dpn_prop = slave->prop.src_dpn_prop;
>  	} else {
> -		mask = slave->prop.sink_ports;
> +		num_ports = hweight32(slave->prop.sink_ports);
>  		dpn_prop = slave->prop.sink_dpn_prop;
>  	}
>  
> -	for_each_set_bit(i, &mask, 32) {
> +	for (i = 0; i < num_ports; i++) {
>  		if (dpn_prop[i].num == port_num)
>  			return &dpn_prop[i];
>  	}

-- 
Péter

Re: [PATCH stable-6.10 regression] Revert "soundwire: stream: fix programming slave ports for non-continous port maps"

[Greg KH]

On Tue, Sep 10, 2024 at 04:02:29PM +0300, Péter Ujfalusi wrote:
> Hi,
> 
> On 10/09/2024 15:40, Peter Ujfalusi wrote:
> > The prop->src_dpn_prop and prop.sink_dpn_prop is allocated for the _number_
> > of ports and it is forced as 0 index based.
> > 
> > The original code was correct while the change to walk the bits and use
> > their position as index into the arrays is not correct.
> > 
> > For exmple we can have the prop.source_ports=0x2, which means we have one
> > port, but the prop.src_dpn_prop[1] is accessing outside of the allocated
> > memory.
> > 
> > This reverts commit 6fa78e9c41471fe43052cd6feba6eae1b0277ae3.
> 
> I just noticed that Krzysztof already sent the revert patch but it is
> not picked up for stable-6.10.y
> 
> https://lore.kernel.org/lkml/20240909164746.136629-1-krzysztof.kozlowski@linaro.org/

Is this in Linus's tree yet?  That's what we are waiting for.

> > Cc: stable@vger.kernel.org # 6.10.y
> > Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
> > ---
> > Hi,
> > 
> > The reverted patch causes major regression on soundwire causing all audio
> > to fail.
> > Interestingly the patch is only in 6.10.8 and 6.10.9, not in mainline or linux-next.

Really?  Commit ab8d66d132bc ("soundwire: stream: fix programming slave
ports for non-continous port maps") is in Linus's tree, why isn't it
being reverted there first?

confused,

greg k-h

Re: [PATCH stable-6.10 regression] Revert "soundwire: stream: fix programming slave ports for non-continous port maps"

[Linux regression tracking (Thorsten Leemhuis)]

On 11.09.24 14:31, Greg KH wrote:
> On Tue, Sep 10, 2024 at 04:02:29PM +0300, Péter Ujfalusi wrote:
>>> The reverted patch causes major regression on soundwire causing all audio
>>> to fail.
>>> Interestingly the patch is only in 6.10.8 and 6.10.9, not in mainline or linux-next.
> 
> Really?  Commit ab8d66d132bc ("soundwire: stream: fix programming slave
> ports for non-continous port maps") is in Linus's tree, why isn't it
> being reverted there first?

FWIW, the revert should land in mainline tomorrow afaics:
https://lore.kernel.org/all/ZuFcBcJztAgicjNt@vaman/

BTW, in case anyone cares: I think this is another report about the
problem, this time with 6.6.y:
https://bugzilla.kernel.org/show_bug.cgi?id=219256

Ciao, Thorsten

Re: [PATCH stable-6.10 regression] Revert "soundwire: stream: fix programming slave ports for non-continous port maps"

[Vinod Koul]

On 11-09-24, 14:31, Greg KH wrote:
> On Tue, Sep 10, 2024 at 04:02:29PM +0300, Péter Ujfalusi wrote:
> > Hi,
> > 
> > On 10/09/2024 15:40, Peter Ujfalusi wrote:
> > > The prop->src_dpn_prop and prop.sink_dpn_prop is allocated for the _number_
> > > of ports and it is forced as 0 index based.
> > > 
> > > The original code was correct while the change to walk the bits and use
> > > their position as index into the arrays is not correct.
> > > 
> > > For exmple we can have the prop.source_ports=0x2, which means we have one
> > > port, but the prop.src_dpn_prop[1] is accessing outside of the allocated
> > > memory.
> > > 
> > > This reverts commit 6fa78e9c41471fe43052cd6feba6eae1b0277ae3.
> > 
> > I just noticed that Krzysztof already sent the revert patch but it is
> > not picked up for stable-6.10.y
> > 
> > https://lore.kernel.org/lkml/20240909164746.136629-1-krzysztof.kozlowski@linaro.org/
> 
> Is this in Linus's tree yet?  That's what we are waiting for.

Yes I was waiting for that as well, the pull request has been sent to
Linus, this should be in his tree, hopefully tomorow..

> 
> > > Cc: stable@vger.kernel.org # 6.10.y
> > > Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
> > > ---
> > > Hi,
> > > 
> > > The reverted patch causes major regression on soundwire causing all audio
> > > to fail.
> > > Interestingly the patch is only in 6.10.8 and 6.10.9, not in mainline or linux-next.
> 
> Really?  Commit ab8d66d132bc ("soundwire: stream: fix programming slave
> ports for non-continous port maps") is in Linus's tree, why isn't it
> being reverted there first?

I guess Peter jumped the gun, I was planning to ask you once this is
picked up Linus

-- 
~Vinod

Re: [PATCH stable-6.10 regression] Revert "soundwire: stream: fix programming slave ports for non-continous port maps"

[Vinod Koul]

On 13-09-24, 17:21, Vinod Koul wrote:
> On 11-09-24, 14:31, Greg KH wrote:
> > On Tue, Sep 10, 2024 at 04:02:29PM +0300, Péter Ujfalusi wrote:
> > > Hi,
> > > 
> > > On 10/09/2024 15:40, Peter Ujfalusi wrote:
> > > > The prop->src_dpn_prop and prop.sink_dpn_prop is allocated for the _number_
> > > > of ports and it is forced as 0 index based.
> > > > 
> > > > The original code was correct while the change to walk the bits and use
> > > > their position as index into the arrays is not correct.
> > > > 
> > > > For exmple we can have the prop.source_ports=0x2, which means we have one
> > > > port, but the prop.src_dpn_prop[1] is accessing outside of the allocated
> > > > memory.
> > > > 
> > > > This reverts commit 6fa78e9c41471fe43052cd6feba6eae1b0277ae3.
> > > 
> > > I just noticed that Krzysztof already sent the revert patch but it is
> > > not picked up for stable-6.10.y
> > > 
> > > https://lore.kernel.org/lkml/20240909164746.136629-1-krzysztof.kozlowski@linaro.org/
> > 
> > Is this in Linus's tree yet?  That's what we are waiting for.
> 
> Yes I was waiting for that as well, the pull request has been sent to
> Linus, this should be in his tree, hopefully tomorow..

It is in Linus's tree now. Greg would you like to drop commit
6fa78e9c41471fe43052cd6feba6eae1b0277ae3 or carry it and the
revert...?

What is the usual process for you to handle reverts?
-- 
~Vinod

Re: [PATCH stable-6.10 regression] Revert "soundwire: stream: fix programming slave ports for non-continous port maps"

[Greg KH]

On Sun, Sep 15, 2024 at 12:45:25AM +0530, Vinod Koul wrote:
> On 13-09-24, 17:21, Vinod Koul wrote:
> > On 11-09-24, 14:31, Greg KH wrote:
> > > On Tue, Sep 10, 2024 at 04:02:29PM +0300, Péter Ujfalusi wrote:
> > > > Hi,
> > > > 
> > > > On 10/09/2024 15:40, Peter Ujfalusi wrote:
> > > > > The prop->src_dpn_prop and prop.sink_dpn_prop is allocated for the _number_
> > > > > of ports and it is forced as 0 index based.
> > > > > 
> > > > > The original code was correct while the change to walk the bits and use
> > > > > their position as index into the arrays is not correct.
> > > > > 
> > > > > For exmple we can have the prop.source_ports=0x2, which means we have one
> > > > > port, but the prop.src_dpn_prop[1] is accessing outside of the allocated
> > > > > memory.
> > > > > 
> > > > > This reverts commit 6fa78e9c41471fe43052cd6feba6eae1b0277ae3.
> > > > 
> > > > I just noticed that Krzysztof already sent the revert patch but it is
> > > > not picked up for stable-6.10.y
> > > > 
> > > > https://lore.kernel.org/lkml/20240909164746.136629-1-krzysztof.kozlowski@linaro.org/
> > > 
> > > Is this in Linus's tree yet?  That's what we are waiting for.
> > 
> > Yes I was waiting for that as well, the pull request has been sent to
> > Linus, this should be in his tree, hopefully tomorow..
> 
> It is in Linus's tree now. Greg would you like to drop commit
> 6fa78e9c41471fe43052cd6feba6eae1b0277ae3 or carry it and the
> revert...?

I can not "drop" a commit that is already in a realease for obvious
reasons :(

> What is the usual process for you to handle reverts?

We just take them like normal.  What is the git id of the revert in
Linus's tree?

thanks,

greg k-h

Re: [PATCH stable-6.10 regression] Revert "soundwire: stream: fix programming slave ports for non-continous port maps"

[Vinod Koul]

On 15-09-24, 15:23, Greg KH wrote:
> On Sun, Sep 15, 2024 at 12:45:25AM +0530, Vinod Koul wrote:
> > > > 
> > > > Is this in Linus's tree yet?  That's what we are waiting for.
> > > 
> > > Yes I was waiting for that as well, the pull request has been sent to
> > > Linus, this should be in his tree, hopefully tomorow..
> > 
> > It is in Linus's tree now. Greg would you like to drop commit
> > 6fa78e9c41471fe43052cd6feba6eae1b0277ae3 or carry it and the
> > revert...?
> 
> I can not "drop" a commit that is already in a realease for obvious
> reasons :(
> 
> > What is the usual process for you to handle reverts?
> 
> We just take them like normal.  What is the git id of the revert in
> Linus's tree?

Sure, makes sense. It has been auto-picked so this is fine.

fwiw: 
ab8d66d132bc: soundwire: stream: fix programming slave ports for non-continous port maps

Thanks
-- 
~Vinod

Re: [PATCH stable-6.10 regression] Revert "soundwire: stream: fix programming slave ports for non-continous port maps"

[Vinod Koul]

On 11-09-24, 16:31, Linux regression tracking (Thorsten Leemhuis) wrote:
> On 11.09.24 14:31, Greg KH wrote:
> > On Tue, Sep 10, 2024 at 04:02:29PM +0300, Péter Ujfalusi wrote:
> >>> The reverted patch causes major regression on soundwire causing all audio
> >>> to fail.
> >>> Interestingly the patch is only in 6.10.8 and 6.10.9, not in mainline or linux-next.
> > 
> > Really?  Commit ab8d66d132bc ("soundwire: stream: fix programming slave
> > ports for non-continous port maps") is in Linus's tree, why isn't it
> > being reverted there first?
> 
> FWIW, the revert should land in mainline tomorrow afaics:
> https://lore.kernel.org/all/ZuFcBcJztAgicjNt@vaman/
> 
> BTW, in case anyone cares: I think this is another report about the
> problem, this time with 6.6.y:
> https://bugzilla.kernel.org/show_bug.cgi?id=219256

Revert has been applied to 6.6 and other stable kernel so this should be
fixed now

-- 
~Vinod