[alsa-devel] [PATCH] Revert "ASoC: core: mark SND_SOC_BYTES_EXT as deprecated"
SND_SOC_BYTES_EXT was added by commit d98812082c87 ("ASoC: add SND_SOC_BYTES_EXT") to allow drivers to transfer arbitrary byte-aligned data to/from devices via ALSA control interface. Later, SND_SOC_BYTES_TLV was added by commit 7523a271682f ("ASoC: core: add a helper for extended byte controls using TLV") was added for the same purpose and developers expected this to obsolete the former.
However, SND_SOC_BYTES_TLV brought confusions to user land because it doesn't follow to a protocol of ALSA control interface. At least, this macro and related kernel APIs include two misunderstandings: - 'struct snd_ctl_elem_info.count' can also represent the length of TLV packet paylaod, snd_soc_bytes_info_ext() performs in this way. - 'struct snd_ctl_tlv.tlv' can include arbitrary data regardless of TLV packet structure, snd_soc_bytes_tlv_callback() performs in this way.
In a policy of kernel land development, it's quite worse to break protocols for applications. Therefore, developers are discouraged to use these kernel APIs.
In the first place, SND_SOC_BYTES_TLV was added to satisfy a request of developers who need to add control elements which transfer data larger than the size which 'struct snd_ctl_elem_value' can represent; e.g. over 512 bytes. As long as the size is less than the size, SND_SOC_BYTES_EXT is still available. Although there is the limitation of maximum data size, it's better to use this API for stable application interface till better alternative ways are implemented in future.
For these reasons, this commit reverts the previous commit which lead developers to the worse behaviour. Unfortunately, SND_SOC_BYTES_TLV was already merged in development cycle for kernel 3.17 and published, so this commit doesn't touch it for any compatibilities.
Reference: http://mailman.alsa-project.org/pipermail/alsa-devel/2013-November/069483.ht... Reference: commit d98812082c87 ("ASoC: add SND_SOC_BYTES_EXT") Reference: commit 7523a271682f ("ASoC: core: add a helper for extended byte controls using TLV") Cc: Vinod Koul vinod.koul@intel.com Acked-by: Charles Keepax ckeepax@opensource.wolfsonmicro.com Signed-off-by: Takashi Sakamoto o-takashi@sakamocchi.jp --- include/sound/soc.h | 3 --- 1 file changed, 3 deletions(-)
diff --git a/include/sound/soc.h b/include/sound/soc.h index 6144882..18f7d21 100644 --- a/include/sound/soc.h +++ b/include/sound/soc.h @@ -311,9 +311,6 @@ {.base = xbase, .num_regs = xregs, \ .mask = xmask }) }
-/* - * SND_SOC_BYTES_EXT is deprecated, please USE SND_SOC_BYTES_TLV instead - */ #define SND_SOC_BYTES_EXT(xname, xcount, xhandler_get, xhandler_put) \ { .iface = SNDRV_CTL_ELEM_IFACE_MIXER, .name = xname, \ .info = snd_soc_bytes_info_ext, \
On Fri, Sep 16, 2016 at 08:06:02AM +0900, Takashi Sakamoto wrote:
Please submit patches using subject lines reflecting the style for the subsystem. This makes it easier for people to identify relevant patches. Look at what existing commits in the area you're changing are doing and make sure your subject lines visually resemble what they're doing.
However, SND_SOC_BYTES_TLV brought confusions to user land because it doesn't follow to a protocol of ALSA control interface. At least, this macro and related kernel APIs include two misunderstandings:
- 'struct snd_ctl_elem_info.count' can also represent the length of TLV packet paylaod, snd_soc_bytes_info_ext() performs in this way.
- 'struct snd_ctl_tlv.tlv' can include arbitrary data regardless of TLV packet structure, snd_soc_bytes_tlv_callback() performs in this way.
I can't tell what problem you are trying to describe here, sorry. What are you expecting the userspace ABI to do that these controls don't, you appear to be saying that the code does actually do what's expected which obviously isn't a problem. What do you expect to happen, what is actually happening and why do you beleive this to be a problem?
In the first place, SND_SOC_BYTES_TLV was added to satisfy a request of developers who need to add control elements which transfer data larger than the size which 'struct snd_ctl_elem_value' can represent; e.g. over 512 bytes. As long as the size is less than the size, SND_SOC_BYTES_EXT is still available. Although there is the limitation of maximum data size, it's better to use this API for stable application interface till better alternative ways are implemented in future.
This doesn't make much sense to me, sorry. This is an interface for drivers. They shouldn't need to know the implementation details of different controls and we should be striving for consistency in the interface we present both at the driver level and at the userspace level. Having two randomly different interfaces is not going to help that or stop the larger controls existing, it'll just make the interface more confusing for driver authors.
Hi Mark,
On Sep 16 2016 19:52, Mark Brown wrote:
On Fri, Sep 16, 2016 at 08:06:02AM +0900, Takashi Sakamoto wrote:
Please submit patches using subject lines reflecting the style for the subsystem. This makes it easier for people to identify relevant patches. Look at what existing commits in the area you're changing are doing and make sure your subject lines visually resemble what they're doing.
OK. But the different fashion in ALSA SoC part from ALSA itself often puzzles me.
However, SND_SOC_BYTES_TLV brought confusions to user land because it doesn't follow to a protocol of ALSA control interface. At least, this macro and related kernel APIs include two misunderstandings:
- 'struct snd_ctl_elem_info.count' can also represent the length of TLV packet paylaod, snd_soc_bytes_info_ext() performs in this way.
- 'struct snd_ctl_tlv.tlv' can include arbitrary data regardless of TLV packet structure, snd_soc_bytes_tlv_callback() performs in this way.
I can't tell what problem you are trying to describe here, sorry. What are you expecting the userspace ABI to do that these controls don't, you appear to be saying that the code does actually do what's expected which obviously isn't a problem.
These two APIs don't perform what to be expected in a point of application interface. These two items are not what to be expected in a point of application interface, but what they do wrong.
What do you expect to happen, what is actually happening and why do you beleive this to be a problem?
This patchset comes from a conclusion of this thread: [alsa-devel] [RFC][PATCH 0/4] ALSA: control: return payload length of TLV operation http://mailman.alsa-project.org/pipermail/alsa-devel/2016-August/112242.html http://mailman.alsa-project.org/pipermail/alsa-devel/2016-September/112479.h... (they were cross month end.)
If you need a minute, please read Iwai-san's post: http://mailman.alsa-project.org/pipermail/alsa-devel/2016-September/112534.h...
In the first place, SND_SOC_BYTES_TLV was added to satisfy a request of developers who need to add control elements which transfer data larger than the size which 'struct snd_ctl_elem_value' can represent; e.g. over 512 bytes. As long as the size is less than the size, SND_SOC_BYTES_EXT is still available. Although there is the limitation of maximum data size, it's better to use this API for stable application interface till better alternative ways are implemented in future.
This doesn't make much sense to me, sorry. This is an interface for drivers. They shouldn't need to know the implementation details of different controls and we should be striving for consistency in the interface we present both at the driver level and at the userspace level.
I have no objections to this indication. I agree with it, as a driver developer.
Having two randomly different interfaces is not going to help that or stop the larger controls existing, it'll just make the interface more confusing for driver authors.
On the other hand, when some driver developers actually follow to the message which I'd like to purge, then abuses of the interface are increased. Near future, it will bring confusions and disorder to ALSA applications. And the cause is hard to be identified from user land, because few developers think drivers lose protocol of application interface in fact.
Of cource, there's a need to work more to solve this issue. This is quite a small part of it.
Regards
Takashi Sakamoto
On Fri, Sep 16, 2016 at 10:21:12PM +0900, Takashi Sakamoto wrote:
On Sep 16 2016 19:52, Mark Brown wrote:
Please submit patches using subject lines reflecting the style for the subsystem. This makes it easier for people to identify relevant patches. Look at what existing commits in the area you're changing are doing and make sure your subject lines visually resemble what they're doing.
OK. But the different fashion in ALSA SoC part from ALSA itself often puzzles me.
ALSA patches don't usually start "Revert" either...
However, SND_SOC_BYTES_TLV brought confusions to user land because it doesn't follow to a protocol of ALSA control interface. At least, this macro and related kernel APIs include two misunderstandings:
- 'struct snd_ctl_elem_info.count' can also represent the length of TLV packet paylaod, snd_soc_bytes_info_ext() performs in this way.
- 'struct snd_ctl_tlv.tlv' can include arbitrary data regardless of TLV packet structure, snd_soc_bytes_tlv_callback() performs in this way.
I can't tell what problem you are trying to describe here, sorry. What are you expecting the userspace ABI to do that these controls don't, you appear to be saying that the code does actually do what's expected which obviously isn't a problem.
These two APIs don't perform what to be expected in a point of application interface. These two items are not what to be expected in a point of application interface, but what they do wrong.
In what concrete way do they not do what is expected?
What do you expect to happen, what is actually happening and why do you beleive this to be a problem?
This patchset comes from a conclusion of this thread:
We need someone reading the changelog to be able to identify what the change is doing, especially for something like this which is visible outside the kernel.
Having two randomly different interfaces is not going to help that or stop the larger controls existing, it'll just make the interface more confusing for driver authors.
On the other hand, when some driver developers actually follow to the message which I'd like to purge, then abuses of the interface are increased. Near future, it will bring confusions and disorder to ALSA applications. And the cause is hard to be identified from user land, because few developers think drivers lose protocol of application interface in fact.
Of cource, there's a need to work more to solve this issue. This is quite a small part of it.
This isn't really going to help that too much unfortunately given that one of the users of the new interface is the Intel drivers which are presumably going to be one of the most common things applications encounter.
participants (2)
-
Mark Brown -
Takashi Sakamoto