So far the setup of BDL table is performed at the prepare stage, where
all PCM parameters have been already set up. When something wrong
happens at it, we return -EINVAL; it's supposed to be a rare case
since the involved memory allocation is a small chunk of kmalloc for
the table.
However, when we receive too many small non-contiguous pages in highly
fragmented memories, it may overflow the max table size, resulting in
the same -EINVAL error from the prepare, too. A bad scenario is that
user-space cannot know what went wrong (as it's an error from the
prepare stage) and -EINVAL, hence it may retry with the same
parameters, failing again repeatedly.
In this patch, we try to set up the BDL table at hw_params right after
the buffer allocation, and return -ENOMEM if it overflows.
This allows user-space knowing that it should reduce the buffer size
request accordingly and may retry with more fitting parameters.
Link: https://lore.kernel.org/r/20240221100607.6565-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
#include <sound/core.h>
#include <sound/initval.h>
+#include <sound/pcm_params.h>
#include "hda_controller.h"
#include "hda_local.h"
struct azx_pcm *apcm = snd_pcm_substream_chip(substream);
struct azx *chip = apcm->chip;
struct azx_dev *azx_dev = get_azx_dev(substream);
+ struct hdac_stream *hdas = azx_stream(azx_dev);
int ret = 0;
trace_azx_pcm_hw_params(chip, azx_dev);
goto unlock;
}
- azx_dev->core.bufsize = 0;
- azx_dev->core.period_bytes = 0;
- azx_dev->core.format_val = 0;
+ /* Set up BDLEs here, return -ENOMEM if too many BDLEs are required */
+ hdas->bufsize = params_buffer_bytes(hw_params);
+ hdas->period_bytes = params_period_bytes(hw_params);
+ hdas->format_val = 0;
+ hdas->no_period_wakeup =
+ (hw_params->info & SNDRV_PCM_INFO_NO_PERIOD_WAKEUP) &&
+ (hw_params->flags & SNDRV_PCM_HW_PARAMS_NO_PERIOD_WAKEUP);
+ if (snd_hdac_stream_setup_periods(hdas) < 0)
+ ret = -ENOMEM;
unlock:
dsp_unlock(azx_dev);