control: ucm: add ioctl to retrieve full card components

[mstrozek]

The fixed-size components field in SNDRV_CTL_IOCTL_CARD_INFO can be too small on systems with many audio devices. The kernel [1] will provide a new ioctl to read the full string while truncating the original in card_info if it grows too big. Make sure alsa-lib can read the full string if the original is truncated.

[1] link to v3

[perexg]

Thanks for your change, but the situation with new components string can be handled using new internal structure in alsa-lib like:

diff --git a/include/local.h b/include/local.h
index a6996759..97701754 100644
--- a/include/local.h
+++ b/include/local.h
@@ -101,7 +101,19 @@
 #define _snd_pcm_sw_params snd_pcm_sw_params
 #define _snd_pcm_status snd_pcm_status
 
-#define _snd_ctl_card_info snd_ctl_card_info
+/* V2 structure - increased components string from 128 bytes to 512 bytes */
+struct _snd_ctl_card_info {
+       int card;                       /* card number */
+       int pad;                        /* reserved for future (was type) */
+       unsigned char id[16];           /* ID of card (user selectable) */
+       unsigned char driver[16];       /* Driver name */
+       unsigned char name[32];         /* Short name of soundcard */
+       unsigned char longname[80];     /* name + info text about soundcard */
+       unsigned char reserved_[16];    /* reserved for future (was ID of mixer) */
+       unsigned char mixername[80];    /* visual mixer identification */
+       unsigned char components[512];  /* card components / fine identification, delimited with one space (AC97 etc..) */
+};
+
 #define _snd_ctl_elem_id snd_ctl_elem_id
 #define _snd_ctl_elem_list snd_ctl_elem_list
 #define _snd_ctl_elem_info snd_ctl_elem_info
diff --git a/src/control/control_hw.c b/src/control/control_hw.c
index 962cd796..30825a80 100644
--- a/src/control/control_hw.c
+++ b/src/control/control_hw.c
@@ -129,10 +129,14 @@ static int snd_ctl_hw_subscribe_events(snd_ctl_t *handle, int subscribe)
 static int snd_ctl_hw_card_info(snd_ctl_t *handle, snd_ctl_card_info_t *info)
 {
        snd_ctl_hw_t *hw = handle->private_data;
-       if (ioctl(hw->fd, SNDRV_CTL_IOCTL_CARD_INFO, info) < 0) {
+       struct snd_ctl_card_info kinfo;
+       if (ioctl(hw->fd, SNDRV_CTL_IOCTL_CARD_INFO, &kinfo) < 0) {
                snd_errornum(CONTROL, "SNDRV_CTL_IOCTL_CARD_INFO failed");
                return -errno;
        }
+       snd_ctl_card_info_clear(info);
+       ... copy data from kinfo to info here..
+       ... if necessary ('>' postfix in the components string) - call SNDRV_CTL_IOCTL_CARD_COMPONENTS ...
        return 0;
 }

Difference for memory allocations should be already covered using snd_ctl_card_info_sizeof(), snd_ctl_card_info_alloca() and snd_ctl_card_info_malloc() functions.

[perexg]

I don't follow the second call. If the ioctl structure defined array with 512 bytes, it cannot be beyond sizeof(comp.components). Perhaps, you may just add assert to assure that "sizeof(info.components) == sizeof(comp.components)" and keep the bogus comp.length check.

[mstrozek]

Yes, it does not make much sense if the size is fixed to 512 bytes... or do you think it might be better to make 'components' a dynamic array? and allocate it to the queried length received in the first call?

[perexg]

Yes, I think that Takashi meant to use pointer to the string array in the structure like we do in e.g. struct snd_xferi. Note: The ioctl for 32-bit code must be added to sound/core/control_compat.c (kernel) for this variant.

[mstrozek]

Done now, hopefully this makes more sense.
Kernel v3 patch with the 32-bit code will come soon

[mstrozek]

Also now added some better handling in snd_ctl_card_info_free/_clear/_copy()

[perexg]

Unfortunately, for alsa-lib, the all data must be allocated together ahead. See snd_ctl_card_info_alloca. So, please, return back fixed array to this structure.

[mstrozek]

Returned to components[512]. In this case also removed the "length query" from the ioctl, writing into info->components directly and only running a quick sanity xheck on the length, though not sure if it is necessary