| Message ID | 1444743477-48819-1-git-send-email-koro.chen@mediatek.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 10/13/2015 03:37 PM, Koro Chen wrote: > The patch "ASoC: Prevent components from being bound to multiple cards" > adds check to prevent multiple cards from using the same component. > However, snd_soc_register_platform() or snd_soc_register_codec() will > also create components, and sharing the same platform by multiple cards > is then refused. This happens with a platform having multiple > independent DAIs that share the same DMA controller. > > Relax the condition by checking component->registered_as_component, > which is only true in case of snd_soc_register_component() and > will be false for components created by snd_soc_register_platform() > or snd_soc_register_codec(). Binding a component to multiple cards results in internal data structure corruption, regardless of whether it is a raw component, CODEC or platform, which is why the check was added. So the proposed change wont work. > > Signed-off-by: Koro Chen <koro.chen@mediatek.com> > --- > sound/soc/soc-core.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c > index 42575b0..eca169a 100644 > --- a/sound/soc/soc-core.c > +++ b/sound/soc/soc-core.c > @@ -1106,7 +1106,8 @@ static int soc_probe_component(struct snd_soc_card *card, > return 0; > > if (component->card) { > - if (component->card != card) { > + if (component->card != card && > + component->registered_as_component) { > dev_err(component->dev, > "Trying to bind component to card \"%s\" but is already bound to card \"%s\"\n", > card->name, component->card->name); >
On Tue, 2015-10-13 at 15:44 +0200, Lars-Peter Clausen wrote: > On 10/13/2015 03:37 PM, Koro Chen wrote: > > The patch "ASoC: Prevent components from being bound to multiple cards" > > adds check to prevent multiple cards from using the same component. > > However, snd_soc_register_platform() or snd_soc_register_codec() will > > also create components, and sharing the same platform by multiple cards > > is then refused. This happens with a platform having multiple > > independent DAIs that share the same DMA controller. > > > > Relax the condition by checking component->registered_as_component, > > which is only true in case of snd_soc_register_component() and > > will be false for components created by snd_soc_register_platform() > > or snd_soc_register_codec(). > > Binding a component to multiple cards results in internal data structure > corruption, regardless of whether it is a raw component, CODEC or platform, > which is why the check was added. So the proposed change wont work. > Thanks for your comment. Is it possible to share an example of how the data structure will be corrupted? So I can study into this further. > > > > Signed-off-by: Koro Chen <koro.chen@mediatek.com> > > --- > > sound/soc/soc-core.c | 3 ++- > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c > > index 42575b0..eca169a 100644 > > --- a/sound/soc/soc-core.c > > +++ b/sound/soc/soc-core.c > > @@ -1106,7 +1106,8 @@ static int soc_probe_component(struct snd_soc_card *card, > > return 0; > > > > if (component->card) { > > - if (component->card != card) { > > + if (component->card != card && > > + component->registered_as_component) { > > dev_err(component->dev, > > "Trying to bind component to card \"%s\" but is already bound to card \"%s\"\n", > > card->name, component->card->name); > > > > _______________________________________________ > Alsa-devel mailing list > Alsa-devel@alsa-project.org > http://mailman.alsa-project.org/mailman/listinfo/alsa-devel
On 10/13/2015 04:18 PM, Koro Chen wrote: > On Tue, 2015-10-13 at 15:44 +0200, Lars-Peter Clausen wrote: >> On 10/13/2015 03:37 PM, Koro Chen wrote: >>> The patch "ASoC: Prevent components from being bound to multiple cards" >>> adds check to prevent multiple cards from using the same component. >>> However, snd_soc_register_platform() or snd_soc_register_codec() will >>> also create components, and sharing the same platform by multiple cards >>> is then refused. This happens with a platform having multiple >>> independent DAIs that share the same DMA controller. >>> >>> Relax the condition by checking component->registered_as_component, >>> which is only true in case of snd_soc_register_component() and >>> will be false for components created by snd_soc_register_platform() >>> or snd_soc_register_codec(). >> >> Binding a component to multiple cards results in internal data structure >> corruption, regardless of whether it is a raw component, CODEC or platform, >> which is why the check was added. So the proposed change wont work. >> > Thanks for your comment. Is it possible to share an example of how the > data structure will be corrupted? So I can study into this further. Just look at soc_probe_component() and think about what happens if that runs twice for two different cards. Multiple calls to list_add() on the same list, controls are added multiple times, DAPM widgets are created multiple times, the card field will only point to the last card.
On Tue, 2015-10-13 at 16:42 +0200, Lars-Peter Clausen wrote: > On 10/13/2015 04:18 PM, Koro Chen wrote: > > On Tue, 2015-10-13 at 15:44 +0200, Lars-Peter Clausen wrote: > >> On 10/13/2015 03:37 PM, Koro Chen wrote: > >>> The patch "ASoC: Prevent components from being bound to multiple cards" > >>> adds check to prevent multiple cards from using the same component. > >>> However, snd_soc_register_platform() or snd_soc_register_codec() will > >>> also create components, and sharing the same platform by multiple cards > >>> is then refused. This happens with a platform having multiple > >>> independent DAIs that share the same DMA controller. > >>> > >>> Relax the condition by checking component->registered_as_component, > >>> which is only true in case of snd_soc_register_component() and > >>> will be false for components created by snd_soc_register_platform() > >>> or snd_soc_register_codec(). > >> > >> Binding a component to multiple cards results in internal data structure > >> corruption, regardless of whether it is a raw component, CODEC or platform, > >> which is why the check was added. So the proposed change wont work. > >> > > Thanks for your comment. Is it possible to share an example of how the > > data structure will be corrupted? So I can study into this further. > > Just look at soc_probe_component() and think about what happens if that runs > twice for two different cards. Multiple calls to list_add() on the same > list, controls are added multiple times, DAPM widgets are created multiple > times, the card field will only point to the last card. > When multiple binding happens, soc_probe_component() just returns zero without doing anything after this patch (actually it also returned zero before the patch "ASoC: Prevent components from being bound to multiple cards"). So the component still binds to the first card. For this case I think it should be fine? > > _______________________________________________ > Linux-mediatek mailing list > Linux-mediatek@lists.infradead.org > http://lists.infradead.org/mailman/listinfo/linux-mediatek
On 10/14/2015 03:19 AM, Koro Chen wrote: > On Tue, 2015-10-13 at 16:42 +0200, Lars-Peter Clausen wrote: >> On 10/13/2015 04:18 PM, Koro Chen wrote: >>> On Tue, 2015-10-13 at 15:44 +0200, Lars-Peter Clausen wrote: >>>> On 10/13/2015 03:37 PM, Koro Chen wrote: >>>>> The patch "ASoC: Prevent components from being bound to multiple cards" >>>>> adds check to prevent multiple cards from using the same component. >>>>> However, snd_soc_register_platform() or snd_soc_register_codec() will >>>>> also create components, and sharing the same platform by multiple cards >>>>> is then refused. This happens with a platform having multiple >>>>> independent DAIs that share the same DMA controller. >>>>> >>>>> Relax the condition by checking component->registered_as_component, >>>>> which is only true in case of snd_soc_register_component() and >>>>> will be false for components created by snd_soc_register_platform() >>>>> or snd_soc_register_codec(). >>>> >>>> Binding a component to multiple cards results in internal data structure >>>> corruption, regardless of whether it is a raw component, CODEC or platform, >>>> which is why the check was added. So the proposed change wont work. >>>> >>> Thanks for your comment. Is it possible to share an example of how the >>> data structure will be corrupted? So I can study into this further. >> >> Just look at soc_probe_component() and think about what happens if that runs >> twice for two different cards. Multiple calls to list_add() on the same >> list, controls are added multiple times, DAPM widgets are created multiple >> times, the card field will only point to the last card. >> > When multiple binding happens, soc_probe_component() just returns zero > without doing anything after this patch (actually it also returned zero > before the patch "ASoC: Prevent components from being bound to multiple > cards"). So the component still binds to the first card. For this case I > think it should be fine? Same answer as before more or less. There is a lot of card specific initialization happening in soc_probe_component(). The card field is initialized to the card the component is bound to, the DAPM widgets are created for the component in the cards DAPM graph, DAI link widgets are created and so on. If you bind the component to multiple cards the initialization will only be done for one card, the first one it was bound to, that doesn't make too much sense. If you look at the snd_soc_component data structure it should be obvious that it does not make sense to associate it with more than a single card. It was never intended that it is possible to bind a component to multiple cards. That it was possible was a bug that was overlooked and some people tried to do it which caused apparently random crashes later on, caused by the data structure corruption. This is why we added the check to catch this kind of mistake early and to avoid the crashes. - Lars
On Wed, Oct 14, 2015 at 11:00:01AM +0200, Lars-Peter Clausen wrote: > It was never intended that it is possible to bind a component to multiple > cards. That it was possible was a bug that was overlooked and some people > tried to do it which caused apparently random crashes later on, caused by > the data structure corruption. This is why we added the check to catch this > kind of mistake early and to avoid the crashes. This is true, but I do think it's something that we should have some story on supporting for some of this hardware that has a bunch of channels in one IP block that can't really interact with each other. It's going to make it a lot easier for people to think about the hardware and how to describe it.
On Thu, 2015-10-15 at 13:10 +0100, Mark Brown wrote: > On Wed, Oct 14, 2015 at 11:00:01AM +0200, Lars-Peter Clausen wrote: > > > It was never intended that it is possible to bind a component to multiple > > cards. That it was possible was a bug that was overlooked and some people > > tried to do it which caused apparently random crashes later on, caused by > > the data structure corruption. This is why we added the check to catch this > > kind of mistake early and to avoid the crashes. > > This is true, but I do think it's something that we should have some > story on supporting for some of this hardware that has a bunch of > channels in one IP block that can't really interact with each other. > It's going to make it a lot easier for people to think about the > hardware and how to describe it. Yes, and if in some cases we must use multiple cards, it seems the only option left is to separate an ASoC platform driver into multiple drivers. It does not make sense to me since there is only one HW block.
On 10/15/2015 02:10 PM, Mark Brown wrote: > On Wed, Oct 14, 2015 at 11:00:01AM +0200, Lars-Peter Clausen wrote: > >> It was never intended that it is possible to bind a component to multiple >> cards. That it was possible was a bug that was overlooked and some people >> tried to do it which caused apparently random crashes later on, caused by >> the data structure corruption. This is why we added the check to catch this >> kind of mistake early and to avoid the crashes. > > This is true, but I do think it's something that we should have some > story on supporting for some of this hardware that has a bunch of > channels in one IP block that can't really interact with each other. > It's going to make it a lot easier for people to think about the > hardware and how to describe it. I'm not saying we shouldn't support it, just that we can't support it with the current code. And adding support for it will require a fair bit of restructuring. If a hardware block as multiple independent channels the best approach in my opinion is to register multiple components (Which we can't do at the moment, because there can only be one component per device). From a framework point of view there is no difference between a single device with multiple independent channels and multiple independent devices with one channel each. Both have the same logical topology. - Lars
On Thu, 2015-10-15 at 15:26 +0200, Lars-Peter Clausen wrote: > On 10/15/2015 02:10 PM, Mark Brown wrote: > > On Wed, Oct 14, 2015 at 11:00:01AM +0200, Lars-Peter Clausen wrote: > > > >> It was never intended that it is possible to bind a component to multiple > >> cards. That it was possible was a bug that was overlooked and some people > >> tried to do it which caused apparently random crashes later on, caused by > >> the data structure corruption. This is why we added the check to catch this > >> kind of mistake early and to avoid the crashes. > > > > This is true, but I do think it's something that we should have some > > story on supporting for some of this hardware that has a bunch of > > channels in one IP block that can't really interact with each other. > > It's going to make it a lot easier for people to think about the > > hardware and how to describe it. > > I'm not saying we shouldn't support it, just that we can't support it with > the current code. And adding support for it will require a fair bit of > restructuring. > > If a hardware block as multiple independent channels the best approach in my > opinion is to register multiple components (Which we can't do at the moment, > because there can only be one component per device). From a framework point Yes... I have tried to register 2 platforms in my ASoC platform driver but alsa considered they are the same platform since they are from the same device. > of view there is no difference between a single device with multiple > independent channels and multiple independent devices with one channel each. > Both have the same logical topology. > Yes, but in my case there is only one HW, one set of registers, and one set of clocks, it should be a single device node in the device tree. > - Lars > >
diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c index 42575b0..eca169a 100644 --- a/sound/soc/soc-core.c +++ b/sound/soc/soc-core.c @@ -1106,7 +1106,8 @@ static int soc_probe_component(struct snd_soc_card *card, return 0; if (component->card) { - if (component->card != card) { + if (component->card != card && + component->registered_as_component) { dev_err(component->dev, "Trying to bind component to card \"%s\" but is already bound to card \"%s\"\n", card->name, component->card->name);
The patch "ASoC: Prevent components from being bound to multiple cards" adds check to prevent multiple cards from using the same component. However, snd_soc_register_platform() or snd_soc_register_codec() will also create components, and sharing the same platform by multiple cards is then refused. This happens with a platform having multiple independent DAIs that share the same DMA controller. Relax the condition by checking component->registered_as_component, which is only true in case of snd_soc_register_component() and will be false for components created by snd_soc_register_platform() or snd_soc_register_codec(). Signed-off-by: Koro Chen <koro.chen@mediatek.com> --- sound/soc/soc-core.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)