kernel/include/sound/soc.h, branch linux-5.2.y

ASoC: core: conditionally increase module refcount on component open

2019-04-08T07:15:44Z

Recently, for Intel platforms the "ignore_module_refcount" field was introduced for the component driver. In order to avoid a deadlock preventing the PCI modules from being removed even when the card was idle, the refcounts were not incremented for the device driver module during component probe. However, this change introduced a nasty side effect: the device driver module can be unloaded while a pcm stream is open. This patch proposes to change the field to be renamed as "module_get_upon_open". When this field is set, the module refcount should be incremented on pcm open amd decremented upon pcm close. This will enable modules to be removed when no PCM playback/capture happens and prevent removal when the component is actually in use. Also, align with the skylake component driver with the new name. Fixes: b450b878('ASoC: core: don't increase component module refcount unconditionally' Signed-off-by: Ranjani Sridharan Acked-by: Pierre-Louis Bossart Signed-off-by: Mark Brown

ASoC: dpcm: prevent snd_soc_dpcm use after free

2019-03-11T16:58:49Z

The dpcm get from fe_clients/be_clients may be free before use Add a spin lock at snd_soc_card level, to protect the dpcm instance. The lock may be used in atomic context, so use spin lock. Use irq spin lock version, since the lock may be used in interrupts. possible race condition between void dpcm_be_disconnect( ... list_del(&dpcm->list_be); list_del(&dpcm->list_fe); kfree(dpcm); ... and for_each_dpcm_fe() for_each_dpcm_be*() race condition example Thread 1: snd_soc_dapm_mixer_update_power() -> soc_dpcm_runtime_update() -> dpcm_be_disconnect() -> kfree(dpcm); Thread 2: dpcm_fe_dai_trigger() -> dpcm_be_dai_trigger() -> snd_soc_dpcm_can_be_free_stop() -> if (dpcm->fe == fe) Excpetion Scenario: two FE link to same BE FE1 -> BE FE2 -> Thread 1: switch of mixer between FE2 -> BE Thread 2: pcm_stop FE1 Exception: Unable to handle kernel paging request at virtual address dead0000000000e0 pc=<> [] dpcm_be_dai_trigger+0x29c/0x47c sound/soc/soc-pcm.c:3226 if (dpcm->fe == fe) lr=<> [] dpcm_fe_dai_do_trigger+0x94/0x26c Backtrace: [] notify_die+0x68/0xb8 [] die+0x118/0x2a8 [] __do_kernel_fault+0x13c/0x14c [] do_translation_fault+0x64/0xa0 [] do_mem_abort+0x4c/0xd0 [] el1_da+0x24/0x40 [] dpcm_be_dai_trigger+0x29c/0x47c [] dpcm_fe_dai_do_trigger+0x94/0x26c [] dpcm_fe_dai_trigger+0x3c/0x44 [] snd_pcm_do_stop+0x50/0x5c [] snd_pcm_action+0xb4/0x13c [] snd_pcm_drop+0xa0/0x128 [] snd_pcm_common_ioctl+0x9d8/0x30f0 [] snd_pcm_ioctl_compat+0x29c/0x2f14 [] compat_SyS_ioctl+0x128/0x244 [] el0_svc_naked+0x34/0x38 [] 0xffffffffffffffff Signed-off-by: KaiChieh Chuang Signed-off-by: Mark Brown

ASoC: core: don't increase component module refcount unconditionally

2019-02-08T18:00:20Z

The ASoC core has for the longest time increased the module reference counts, even before the transition to the component model. This is probably fine on most platforms, but it introduces a deadlock case on Intel devices with the Skylake and SOF drivers which cannot be removed due to their reference counts being modified by the core. In these 2 cases, the PCI or ACPI driver .probe creates a platform device to let the machine driver .probe register the audio card. Conversely the PCI or ACPI driver .remove will unregister the platform device which results in the card being removed by the machine driver .remove. With ascii art, this can be represented as modprobe snd_soc_skl/ soc-pci-dev/sof-acpci-dev ----------> pci/acpi probe ^ | | ---------------| | | | | V V increase register register machine refcount component platform_device ^ | | | | V component <---- register card <---- probe probe The issue is that by playing with the component's module reference counts during the card registration, it's no longer possible to remove the module which controls the component. This can be shown, e.g. with the following error: root@plb-XPS-13-9350:~# lsmod | grep snd_soc_skl snd_soc_skl 110592 1 root@plb-XPS-13-9350:~# rmmod snd_soc_skl rmmod: ERROR: Module snd_soc_skl is in use Increasing the reference count during the component probe is not useful. If the PCI/ACPI module is removed, the card will be removed anyway. To avoid breaking existing platforms and allowing Intel platforms to safely deal with module load/unload cases, this patch introduces a flag which needs to be set during the component initialization. This is a strictly opt-in capability that should only be used when the handling of the component module does not require a reference count increase to prevent removal during use. Note that this solution is not directly applicable to the legacy Atom/SST driver, which uses a different device hierarchy. There are however additional refcount issues which prevent the ACPI driver from being removed. This is a different issue which would need a different patch. Signed-off-by: Pierre-Louis Bossart Signed-off-by: Mark Brown

ASoC: add helper to change platform name for all dailinks

2019-01-28T18:05:38Z

To reuse the same machine drivers with Atom/SST, Skylake and SOF, we need to change the default platform_name (or platforms->name in the "modern" representation). So far, this override was done with an automatic override, which was broken by a set of changes for DT platforms related to deferred probe handling. This automatic override is actually not really needed, the machine driver can already receive the platform name as a platform_data parameter. This is used e.g. for HDaudio support where we have different PCI aliases used for different platforms. We can reuse the same mechanism and modify the machine drivers to override the dailinks prior to registrating the card. This will require additional work for SOF, but with this helper it'll be just two lines of additional code per machine driver which is reused, not the end of the world. This helper can be simplified when all drivers have transitioned to the "modern" representation of dailinks. Signed-off-by: Pierre-Louis Bossart Signed-off-by: Mark Brown

ASoC: soc-core: add .num_platform for dai_link

2019-01-21T18:12:19Z

Current snd_soc_dai_link is starting to use snd_soc_dai_link_component (= modern) style for Platform, but it is still assuming single Platform so far. We will need to have multi Platform support in the not far future. Currently only simple card is using it as sound card driver, and other drivers are converted to it from legacy style by snd_soc_init_platform(). To avoid future problem of multi Platform support, let's add num_platforms before it is too late. In the same time, to make it same naming mothed, "platform" should be "platforms". This patch fixup it too. Signed-off-by: Kuninori Morimoto Signed-off-by: Mark Brown

ASoC: soc.h: add explanation of legacy/modern style of dai_link

2019-01-18T19:14:43Z

Current ALSA SoC is assuming 1 CPU 1 Platform (= DMA) style system. Because of this background, it is directly using xxx_name / xxx_of_node / xxx_dai_name on dai_link. Let's call it as legacy style here. More complex style system like multi CPU multi Platform (= DMA) will coming. To supporting it, we can use snd_soc_dai_link_component on dai_link. Let's call it as modern style here. But current ALSA SoC can't support it so far. Thus, we need to have multi CPU / multi Codec / multi Platform style in the future on ALSA SoC. Currently we already have multi Codec support. Platform is starting to use modern style on dai_link, but still style only. Multi Platform is not yet implemented. And we still don't have multi CPU support on ALSA SoC, and not have modern style either. Currently, if driver is using legacy style Codec/Platform, it will be converted to modern style on soc-core. This means, we are using glue code for legacy vs modern style so far on ALSA SoC. We can fully switch to modern style on all drivers if ALSA SoC supported modern style for CPU, and then, legacy style code will be removed from ALSA SoC. Untile then, we need to keep both legacy/modern style and its glue code. This patch adds such future plan and background on soc.h Signed-off-by: Kuninori Morimoto Signed-off-by: Mark Brown

ASoC: soc-core: fix init platform memory handling

2019-01-14T22:48:16Z

snd_soc_init_platform initializes pointers to snd_soc_dai_link which is statically allocated and it does this by devm_kzalloc. In the event of an EPROBE_DEFER the memory will be freed and the pointers are left dangling. snd_soc_init_platform sees the dangling pointers and assumes they are pointing to initialized memory and does not reallocate them on the second probe attempt which results in a use after free bug since devm has freed the memory from the first probe attempt. Since the intention for snd_soc_dai_link->platform is that it can be set statically by the machine driver we need to respect the pointer in the event we did not set it but still catch dangling pointers. The solution is to add a flag to track whether the pointer was dynamically allocated or not. Signed-off-by: Curtis Malainey Signed-off-by: Mark Brown

Merge branch 'for-4.20' of https://git.kernel.org/pub/scm/linux/kernel/git/broonie/sound into asoc-4.21 intel dep

2018-12-13T14:36:46Z

ASoC: soc.h: makes snd_soc_of_parse_audio_prefix() inline

2018-11-26T12:23:22Z

commit 3b7103562c03c ("ASoC: soc-core: add snd_soc_of_parse_node_prefix()") maked snd_soc_of_parse_audio_prefix() as #define. But it'd be better to make this a static inline rather than a #define. It helps with error messages and type safety. This patch makes it inline. Signed-off-by: Kuninori Morimoto Signed-off-by: Mark Brown

ASoC: soc-core: add snd_soc_of_parse_node_prefix()

2018-11-23T14:01:47Z

Current ASoC has snd_soc_of_parse_audio_prefix() to get codec_conf settings from DT which is used to avoid DAI naming conflict when CPU/Codec matching. Currently, it is parsing from "top node", but, we want to parse from "each sub node" if sound card had multi cpus/codecs. This patch adds new snd_soc_of_parse_node_prefix() to allow parsing settings from selected node. It is keeping existing snd_soc_of_parse_audio_prefix() by using macro. Signed-off-by: Kuninori Morimoto Signed-off-by: Mark Brown