In the PCM read/write loop, the driver calls snd_pcm_update_hw_ptr()
at each time at the beginning of the loop. Russell King reported that
this hogs CPU significantly.
The current code assumes that the pointer callback is very fast and
cheap, also not too much fine grained. It's not true in all cases.
When the pointer advances short samples while the read/write copy has
been performed, the driver updates the hw_ptr and gets avail > 0
again. Then it tries to read/write these small chunks. This repeats
until the avail really gets to zero.
For avoiding this situation, a simple workaround is to call
snd_pcm_update_hw_ptr() only once at starting the loop, assuming that
the read/write copy is performed fast enough. If the available count
becomes short, it goes to snd_pcm_wait_avail() anyway, and this
processes right.
Tested-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Allows the constraint lists to be declared const by drivers which seems
reasonable; there's plenty of other constification we could do if we were
being complete but this was easy and quick.
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
These aren't modules, but they do make use of these macros, so
they will need export.h to get that definition. Previously,
they got it via the implicit module.h inclusion.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Add a helper function to allow drivers to disable hardware resampling
when the application has specified the SNDRV_PCM_HW_PARAMS_NORESAMPLE
flag.
Signed-off-by: Clemens Ladisch <clemens@ladisch.de>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
wait_for_avail() in pcm_lib.c has a race in it (observed in practice by an
Intel validation group).
The function is supposed to return once space in the buffer has become
available, or if some timeout happens. The entity that creates space (irq
handler of sound driver and some such) will do a wake up on a waitqueue
that this function registers for.
However there are two races in the existing code
1) If space became available between the caller noticing there was no
space and this function actually sleeping, the wakeup is missed and the
timeout condition will happen instead
2) If a wakeup happened but not sufficient space became available, the
code will loop again and wait for more space. However, if the second
wake comes in prior to hitting the schedule_timeout_interruptible(), it
will be missed, and potentially you'll wait out until the timeout
happens.
The fix consists of using more careful setting of the current state (so
that if a wakeup happens in the main loop window, the schedule_timeout()
falls through) and by checking for available space prior to going into the
schedule_timeout() loop, but after being on the waitqueue and having the
state set to interruptible.
[tiwai: the following changes have been added to Arjan's original patch:
- merged akpm's fix for waitqueue adding order into a single patch
- reduction of duplicated code of avail check
]
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: <stable@kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Formatting a PCM name is useful for module debug too.
Add snd_prefix when making function public.
[minor coding-style fixes by tiwai]
Signed-off-by: Eliot Blennerhassett <eblennerhassett@audioscience.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
When the PCM period size is set larger than 10 seconds, currently the
PCM core may abort the operation with DMA-error due to the fixed timeout
for 10 seconds. A similar problem is seen in the drain operation that
has a fixed timeout of 10 seconds, too.
This patch fixes the timeout length depending on the period size and
rate, also including the consideration of no_period_wakeup flag.
Reported-by: Raymond Yau <superquad.vortex2@gmail.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The xrun_log function was augmented with the in_interrupt parameter whereas the
empty macro definition used when xrun logging is disabled was not.
Add a third parameter to the empty macro definition so as to not cause compiler
errors when xrun logging (CONFIG_SND_PCM_XRUN_DEBUG) is disabled.
Signed-off-by: Ben Gardiner <bengardiner@nanometrics.ca>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
When debugging pcm drivers I found the "period" or "hw" prefix printed
by either XRUN_DEBUG_PERIODUPDATE or XRUN_DEBUG_PERIODUPDATE events,
respectively to be very useful is observing the interplay between
interrupt-context updates and syscall-context updates.
Similarly, when debugging overruns with XRUN_DEBUG_LOG it is useful to
see the context of the last 10 positions.
Add an in_interrupt member to hwptr_log_entry which stores the value of
the in_interrupt parameter of snd_pcm_update_hw_ptr0 when the log entry
is created. Print a "[Q]" prefix when dumping the log entries if
in_interrupt was true.
Signed-off-by: Ben Gardiner <bengardiner@nanometrics.ca>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
When period interrupts are disabled, snd_pcm_update_hw_ptr0() compares
the current time against the time estimated for the current hardware
pointer to detect xruns. The somewhat fuzzy threshold in the while loop
makes it possible that hdelta becomes negative; the comparison being
done with unsigned types then makes the loop go through the entire 263
negative range, and, depending on the value, never reach an unsigned
value that is small enough to stop the loop. Doing this with interrupts
disabled results in the machine locking up.
To prevent this, ensure that the loop condition uses signed types for
both operands so that the comparison is correctly done.
Many thanks to Kelly Anderson for debugging this.
Reported-by: Nix <nix@esperi.org.uk>
Reported-by: "Christopher K." <c.krooss@googlemail.com>
Reported-and-tested-by: Kelly Anderson <kelly@silka.with-linux.com>
Signed-off-by: Kelly Anderson <kelly@silka.with-linux.com>
[cl: remove unneeded casts; use a temp variable]
Signed-off-by: Clemens Ladisch <clemens@ladisch.de>
Cc: 2.6.38 <stable@kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The Coverity checker spotted that we do not always remember to call
va_end() on 'args' in failure paths in snd_pcm_hw_rule_add().
Here's a patch to fix that up (compile tested only) - it also removes
some annoying trailing whitespace that caught my eye while I was in the
area..
Signed-off-by: Jesper Juhl <jj@chaosbits.net>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Add a lightweight condition on top of the xrun checking so that we can
avoid the division when the application is calling the update function
often enough.
Suggested-by: Jaroslav Kysela <perex@perex.cz>
Signed-off-by: Clemens Ladisch <clemens@ladisch.de>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
When period wakeups are disabled, successive calls to the pointer update
function do not have a maximum allowed distance, so xruns cannot be
detected with the pointer value only.
To detect xruns, compare the actually elapsed time with the time that
should have theoretically elapsed since the last update. When the
hardware pointer has wrapped around due to an xrun, the actually elapsed
time will be too big by about hw_ptr_buffer_jiffies.
Signed-off-by: Clemens Ladisch <clemens@ladisch.de>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This patch allows to disable period interrupts which are
not needed when the application relies on a system timer
to wake-up and refill the ring buffer. The behavior of
the driver is left unchanged, and interrupts are only
disabled if the application requests this configuration.
The behavior in case of underruns is slightly different,
instead of being detected during the period interrupts the
underruns are detected when the application calls
snd_pcm_update_avail, which in turns forces a refresh of the
hw pointer and shows the buffer is empty.
More specifically this patch makes a lot of sense when
PulseAudio relies on timer-based scheduling to access audio
devices such as HDAudio or Intel SST. Disabling interrupts
removes two unwanted wake-ups due to period elapsed events
in low-power playback modes. It also simplifies PulseAudio
voice modules used for speech calls.
To quote Lennart "This patch looks very interesting and
desirable. This is something have long been waiting for."
Support for this in hardware drivers is optional.
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@intel.com>
Signed-off-by: Clemens Ladisch <clemens@ladisch.de>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The current code in pcm_lib.c do all checks using only the position
in the ring buffer. Unfortunately, where the interrupts gets delayed or
merged into one, we need another timing source to check when the
buffer size boundary overlaps to avoid the wrong updating of the
ring buffer pointers.
This code uses jiffies to check the right time window without any
performance impact.
Signed-off-by: Jaroslav Kysela <perex@perex.cz>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
In situation when appl_ptr is far greater then hw_ptr, the hw_avail value
can be greater than buffer_size. Check for this.
Signed-off-by: Jaroslav Kysela <perex@perex.cz>
When using poll() to wait for the next period -- or avail_min samples --
one gets a consistent delay for each system call that is usually just a
little short of the selected period time. However, When using
snd_pcm_read/write(), one gets a jittery delay that alternates between
less than a millisecond and approximately two period times. This is
caused by snd_pcm_lib_{read,write}1() transferring any available samples
to the user's buffer and adjusting the application pointer prior to
sleeping to the end of the current period. When the next period
interrupt occurs, there is then less than avail_min samples remaining to
be transferred in the period, so we end up sleeping until a second
period occurs.
This is solved by using runtime->twake as the number of samples needed
for a wakeup in addition to selecting the proper wait queue to wake in
snd_pcm_update_state(). This requires twake to be non-zero when used
by snd_pcm_lib_{read,write}1() even if avail_min is zero.
Signed-off-by: Dave Dillow <dave@thedillows.org>
Signed-off-by: Jaroslav Kysela <perex@perex.cz>
In the cleanup of the hw_ptr update functions in 2.6.33, the calculation
of the delta value was changed to use the modulo operator to protect
against a negative difference due to the pointer wrapping around at the
boundary.
However, the ptr variables are unsigned, so a negative difference would
result in the two complement's value which has no relation to the actual
difference relative to the boundary; the result is typically some value
near LONG_MAX-boundary. Furthermore, even if the modulo operation would
be done with signed types, the result of a negative dividend could be
negative.
The invalid delta value is then caught by the following checks, but this
means that the pointer update is ignored.
To fix this, use a range check as in the other pointer calculations.
Signed-off-by: Clemens Ladisch <clemens@ladisch.de>
Cc: <stable@kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Commit 7910b4a1db in 2.6.34 changed the
runtime->boundary calculation to make this value a multiple of both the
buffer_size and the period_size, because the latter is assumed by the
runtime->hw_ptr_interrupt calculation.
However, due to the lack of a ioctl that could read the software
parameters before they are set, the kernel requires that alsa-lib
calculates the boundary value, too. The changed algorithm leads to
a different boundary value used by alsa-lib, which makes, e.g., mplayer
fail to play a 44.1 kHz file because the silence_size parameter is now
invalid; bug report:
<https://bugtrack.alsa-project.org/alsa-bug/view.php?id=5015>.
This patch reverts the change to the boundary calculation, and instead
fixes the hw_ptr_interrupt calculation to be period-aligned regardless
of the boundary value.
Signed-off-by: Clemens Ladisch <clemens@ladisch.de>
Cc: <stable@kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The commit 4d96eb255c broke the interrupt
time xrun functionality (stream stop etc.) if the CONFIG_SND_PCM_XRUN_DEBUG
is not set. This is because the xrun() is null defined without it.
Fix this by letting the function xrun() to be always defined as it was
before.
Signed-off-by: Jarkko Nikula <jhnikula@gmail.com>
Cc: Jaroslav Kysela <perex@perex.cz>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Clemens Ladisch noted for hw_ptr_removal in "cleanup & merge hw_ptr
update functions" commit:
"It is possible for the status/delay ioctls to be called when the sound
card's pointer register alreay shows a position at the beginning of the
new period, but immediately before the interrupt is actually executed.
(This happens regularly on a SMP machine with mplayer.) When that
happens, the code thinks that the position must be at least one period
ahead of the current position and drops an entire buffer of data."
Return back the hw_ptr_interrupt variable. The last interrupt pointer
is always computed from the latest hw_ptr instead of tracking it
separately (in this case all hw_ptr checks and modifications might
influence also hw_ptr_interrupt and it is difficult to keep it
consistent).
Signed-off-by: Jaroslav Kysela <perex@perex.cz>
This change fixes the "ALSA: pcm_lib - optimize wake_up() calls for PCM I/O"
commit. New sleeping queue is introduced to separate user space and kernel
space wake_ups. runtime->nowake is renamed to twake (transfer wake).
Signed-off-by: Jaroslav Kysela <perex@perex.cz>
When runtime->periods == 1 or when pointer crosses end of ring buffer,
the delta might be greater than buffer_size.
Signed-off-by: Jaroslav Kysela <perex@perex.cz>
As noted by pl bossart <bossart.nospam@gmail.com>, the PCM I/O routines
(snd_pcm_lib_write1, snd_pcm_lib_read1) should block wake_up() calls
until all samples are not processed.
Signed-off-by: Jaroslav Kysela <perex@perex.cz>
Do general cleanup in snd_pcm_update_hw_ptr*() routines and merge them.
The main change is hw_ptr_interrupt variable removal to simplify code
logic. This variable can be computed directly from hw_ptr.
Ensure that updated hw_ptr is not lower than previous one (it was possible
with old code in some obscure situations when interrupt was delayed or
the lowlevel driver returns wrong ring buffer position value).
Signed-off-by: Jaroslav Kysela <perex@perex.cz>
In some debug cases, it might be usefull to see previous ring buffer
positions to determine position problems from the lowlevel drivers.
Signed-off-by: Jaroslav Kysela <perex@perex.cz>
Refine the rate selection by choosing the rate
closer to the requested one in case of selecting
single frequency. Previously, the higher rate was
always selected.
Also, fix problem with the best_diff unsigned int
value wrapping (turning negative).
Signed-off-by: Krzysztof Helt <krzysztof.h1@wp.pl>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The direction of rounding is incorrect in the snd_interval_ratnum()
It was detected with following parameters (sb8 driver playing
8kHz stereo file):
- num is always 1000000
- requested frequency rate is from 7999 to 7999 (single frequency)
The first loop calculates div_down(num, freq->min) which is 125.
Thus, a frequency range's minimum value is 1000000 / 125 = 8000 Hz.
The second loop calculates div_up(num, freq->max) which is 126
The frequency range's maximum value is 1000000 / 126 = 7936 Hz.
The range maximum is lower than the range minimum so the function
fails due to empty result range.
Signed-off-by: Krzysztof Helt <krzysztof.h1@wp.pl>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
snd_interval_list() expected a sorted list but did not document this, so
there are drivers that give it an unsorted list. To fix this, change
the algorithm to work with any list.
This fixes the "Slave PCM not usable" error with USB devices that have
multiple alternate settings with sample rates in decreasing order, such
as the Philips Askey VC010 WebCam.
http://bugzilla.kernel.org/show_bug.cgi?id=14028
Reported-and-tested-by: Andrzej <adkadk@gmail.com>
Signed-off-by: Clemens Ladisch <clemens@ladisch.de>
Cc: <stable@kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The current PCM core has the following problems regarding PCM draining
in non-blocking mode:
- the current f_flags isn't checked in snd_pcm_drain(), thus changing
the mode dynamically via snd_pcm_nonblock() after open doesn't work.
- calling drain in non-blocking mode just return -EAGAIN error, but
doesn't provide any way to sync with draining.
This patch fixes these issues.
- check file->f_flags in snd_pcm_drain() properly
- when O_NONBLOCK is set, PCM core sets the stream(s) to DRAIN state
but quits ioctl immediately without waiting the whole drain; the
caller can sync the drain manually via poll()
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The fix 79452f0a28 introduced another
bug due to the missing offset for the overlapped hwptr.
When the hwptr goes back to zero, the delta value has to be corrected
with the buffer size. Otherwise this causes looping sounds.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Added the logging functionality to xrun_debug to record the hwptr
updates via snd_pcm_update_hw_ptr() and snd_pcm_update_hwptr_interrupt(),
corresponding to 16 and 8, respectively.
For example,
# echo 9 > /proc/asound/card0/pcm0p/xrun_debug
will record the position and other parameters at each period interrupt
together with the normal XRUN debugging.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
VMware tends to report PCM positions and period updates at utterly
wrong timing. This screws up the recent PCM core code that tries
to correct the position based on the irq timing.
Now, when a backward irq position is detected, skip the update
instead of rebasing. (This is almost the old behavior before
2.6.30.)
Signed-off-by: Takashi Iwai <tiwai@suse.de>
* topic/pcm-jiffies-check:
ALSA: pcm - A helper function to compose PCM stream name for debug prints
ALSA: pcm - Fix update of runtime->hw_ptr_interrupt
ALSA: pcm - Fix a typo in hw_ptr update check
ALSA: PCM midlevel: lower jiffies check margin using runtime->delay value
ALSA: PCM midlevel: Do not update hw_ptr_jiffies when hw_ptr is not changed
ALSA: PCM midlevel: introduce mask for xrun_debug() macro
ALSA: PCM midlevel: improve fifo_size handling
The commit 13f040f9e5 made another
regression, the missing update of runtime->hw_ptr_interrupt.
Since this field is only checked in snd_pcmupdate__hw_ptr_interrupt(),
not in snd_pcm_update_hw_ptr(), it must be updated before the hw_ptr
change check.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Fix a typo in the commit 13f040f9e5
ALSA: PCM midlevel: Do not update hw_ptr_jiffies when hw_ptr is not changed
which causes obvious problems with PA.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Takashi Iwai
Mark Brown
Paul Gortmaker
Clemens Ladisch
Arjan van de Ven
Eliot Blennerhassett
Ben Gardiner
Kelly Anderson
Jesper Juhl
Joe Perches
Jaroslav Kysela
David Dillow
Jarkko Nikula
Krzysztof Helt
Linus Torvalds