Commit 16200948d8 ("ALSA: usb-audio: Fix race at stopping the stream") was
incomplete causing another more severe kernel panic, so it got reverted.
This fixes both the original problem and its fallout kernel race/crash.
The original fix is to move the endpoint member NULL clearing logic inside
wait_clear_urbs() so the irq triggering the urb completion doesn't call
retire_capture/playback_urb() after the NULL clearing and generate a panic.
However this creates a new race between snd_usb_endpoint_start()'s call
to wait_clear_urbs() and the irq urb completion handler which again calls
retire_capture/playback_urb() leading to a new NULL dereference.
We keep the EP deactivation code in snd_usb_endpoint_start() because
removing it will break the EP reference counting (see [1] [2] for info),
however we don't need the "can_sleep" mechanism anymore because a new
function was introduced (snd_usb_endpoint_sync_pending_stop()) which
synchronizes pending stops and gets called inside the pcm prepare callback.
It also makes sense to remove can_sleep because it was also removed from
deactivate_urbs() signature in [3] so we benefit from more simplification.
[1] commit 015618b90 ("ALSA: snd-usb: Fix URB cancellation at stream start")
[2] commit e9ba389c5 ("ALSA: usb-audio: Fix scheduling-while-atomic bug in PCM capture stream")
[3] commit ccc1696d5 ("ALSA: usb-audio: simplify endpoint deactivation code")
Fixes: f8114f8583 ("Revert "ALSA: usb-audio: Fix race at stopping the stream"")
Signed-off-by: Ioan-Adrian Ratiu <adi@adirat.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Some functions in mixer.c and endpoint.c receive list_head instead of
the object itself. This is not obvious and rather error-prone. Let's
pass the proper object directly instead.
The functions in midi.c still receive list_head and this can't be
changed since the object definition isn't exposed to the outside of
midi.c, so left as is.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
When a USB-audio device is disconnected while PCM is still running, we
still see some race: the disconnect callback calls
snd_usb_endpoint_free() that calls release_urbs() and then kfree()
while a PCM stream would be closed at the same time and calls
stop_endpoints() that leads to wait_clear_urbs(). That is, the EP
object might be deallocated while a PCM stream is syncing with
wait_clear_urbs() with the same EP.
Basically calling multiple wait_clear_urbs() would work fine, also
calling wait_clear_urbs() and release_urbs() would work, too, as
wait_clear_urbs() just reads some fields in ep. The problem is the
succeeding kfree() in snd_pcm_endpoint_free().
This patch moves out the EP deallocation into the later point, the
destructor callback. At this stage, all PCMs must have been already
closed, so it's safe to free the objects.
Reported-by: Alan Stern <stern@rowland.harvard.edu>
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The return value of snd_usb_endpoint_deactivate() is not used,
make the function have no return value.
Update the documentation to reflect what the function is actually
doing.
Signed-off-by: Eldad Zack <eldad@fogrefinery.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This patch changes the way URBs are allocated and their sizes are
determined for PCM playback in the snd-usb-audio driver. Currently
the driver allocates too few URBs for endpoints that don't use
implicit sync, making underruns more likely to occur. This may be a
holdover from before I/O delays could be measured accurately; in any
case, it is no longer necessary.
The patch allocates as many URBs as possible, subject to four
limitations:
The total number of URBs for the endpoint is not allowed to
exceed MAX_URBS (which the patch increases from 8 to 12).
The total number of packets per URB is not allowed to exceed
MAX_PACKS (or MAX_PACKS_HS for high-speed devices), which is
decreased from 20 to 6.
The total duration of queued data is not allowed to exceed
MAX_QUEUE, which is decreased from 24 ms to 18 ms.
The total number of ALSA frames in the output queue is not
allowed to exceed the ALSA buffer size.
The last requirement is the hardest to implement. Currently the
number of URBs needed to fill a buffer cannot be determined in
advance, because a buffer contains a fixed number of frames whereas
the number of frames in an URB varies to match shifts in the device's
clock rate. To solve this problem, the patch changes the logic for
deciding how many packets an URB should contain. Rather than using as
many as possible without exceeding an ALSA period boundary, now the
driver uses only as many packets as needed to transfer a predetermined
number of frames. As a result, unless the device's clock has an
exceedingly variable rate, the number of URBs making up each period
(and hence each buffer) will remain constant.
The overall effect of the patch is that playback works better in
low-latency settings. The user can still specify values for
frames/period and periods/buffer that exceed the capabilities of the
hardware, of course. But for values that are within those
capabilities, the performance will be improved. For example, testing
shows that a high-speed device can handle 32 frames/period and 3
periods/buffer at 48 KHz, whereas the current driver starts to get
glitchy at 64 frames/period and 2 periods/buffer.
A side effect of these changes is that the "nrpacks" module parameter
is no longer used. The patch removes it.
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
CC: Clemens Ladisch <clemens@ladisch.de>
Tested-by: Daniel Mack <zonque@gmail.com>
Tested-by: Eldad Zack <eldad@fogrefinery.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Correct spelling of snd_usb_endpoint_implict_feedback_sink in all
occurances.
Signed-off-by: Eldad Zack <eldad@fogrefinery.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
As we are stopping the endpoints asynchronously now, it's better to
trigger the stop of both data and sync endpoints and wait for pending
stopping operations, instead of the sequential trigger-and-wait
procedure.
So the wait argument in snd_usb_endpoint_stop() is dropped, and it's
expected that the caller synchronizes explicitly by calling
snd_usb_endpoint_sync_pending_stop(). (Actually there is only one
place calling this, so it was safe to change.)
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Reduce the redundant arguments for snd_usb_endpoint_start() and
snd_usb_endpoint_stop(). Also replaced from int to bool.
No functional changes by this commit.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
There are bug reports of a crash with USB-audio devices when PCM
prepare is performed immediately after the stream is stopped via
trigger callback. It turned out that the problem is that we don't
wait until all URBs are killed.
This patch adds a new function to synchronize the pending stop
operation on an endpoint, and calls in the prepare callback for
avoiding the crash above.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=49181
Reported-and-tested-by: Artem S. Tashkinov <t.artem@lycos.com>
Cc: <stable@vger.kernel.org> [v3.6]
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Change the interface to configure an endpoint so that it doesn't require
a hw_params struct. This will allow it to be called from prepare
instead of hw_params, configuring it after system resume.
Signed-off-by: Dylan Reid <dgreid@chromium.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
In order to support devices with implicit feedback streaming models,
packet sizes are now stored with each individual urb, and the PCM
handling code which fills the buffers purely relies on the size fields
now.
However, calling snd_usb_audio_next_packet_size() for all possible
packets in an URB at once, prior to letting the PCM code do its job
does in fact not lead to the same behaviour than what the old code did:
The PCM code will break its loop once a period boundary is reached,
consequently using up less packets that it really could.
As snd_usb_audio_next_packet_size() implements a feedback mechanism to
the endpoints phase accumulator, the number of calls to that function
matters, and when called too often, the data rate runs out of bounds.
Fix this by making the next_packet function public, and call it from the
PCM code as before if the packet data sizes are not defined.
Signed-off-by: Daniel Mack <zonque@gmail.com>
Cc: stable@kernel.org [v3.5+]
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Commit e9ba389c5 ("ALSA: usb-audio: Fix scheduling-while-atomic bug in
PCM capture stream") fixed a scheduling-while-atomic bug that happened
when snd_usb_endpoint_start was called from the trigger callback, which
is an atmic context. However, the patch breaks the idea of the endpoints
reference counting, which is the reason why the driver has been
refactored lately.
Revert that commit and let snd_usb_endpoint_start() take care of the URB
cancellation again. As this function is called from both atomic and
non-atomic context, add a flag to denote whether the function may sleep.
Signed-off-by: Daniel Mack <zonque@gmail.com>
Cc: stable@kernel.org [3.5+]
Signed-off-by: Takashi Iwai <tiwai@suse.de>
With the previous commit that added the new streaming model, all
endpoint and streaming related code is now in endpoint.c, and pcm.c
only acts as a wrapper for handling the packet's payload.
Signed-off-by: Daniel Mack <zonque@gmail.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This patch adds a new generic streaming logic for audio over USB.
It defines a model (snd_usb_endpoint) that handles everything that
is related to an USB endpoint and its streaming. There are functions to
activate and deactivate an endpoint (which call usb_set_interface()),
and to start and stop its URBs. It also has function pointers to be
called when data was received or is about to be sent, and pointer to
a sync slave (another snd_usb_endpoint) that is informed when data has
been received.
A snd_usb_endpoint knows about its state and implements a refcounting,
so only the first user will actually start the URBs and only the last
one to stop it will tear them down again.
With this sort of abstraction, the actual streaming is decoupled from
the pcm handling, which makes the "implicit feedback" mechanisms easy to
implement.
In order to split changes properly, this patch only adds the new
implementation but leaves the old one around, so the the driver doesn't
change its behaviour. The switch to actually use the new code is
submitted separately.
Signed-off-by: Daniel Mack <zonque@gmail.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
No code altered at this point, simply preparing for upcoming
refactorizations.
Signed-off-by: Daniel Mack <zonque@gmail.com>
Acked-by: Clemens Ladisch <clemens@ladisch.de>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Move code from endpoint.c into a new file called stream.c and rename
functions so that their names actually reflect what they're doing.
This way, endpoint.c will be available to functions that hold all the
endpoint logic.
Signed-off-by: Daniel Mack <zonque@gmail.com>
Acked-by: Clemens Ladisch <clemens@ladisch.de>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Clean up the usb audio driver by factoring out a lot of functions to
separate files. Code for procfs, quirks, urbs, format parsers etc all
got a new home now.
Moved almost all special quirk handling to quirks.c and introduced new
generic functions to handle them, so the exceptions do not pollute the
whole driver.
Renamed usbaudio.c to card.c because this is what it actually does now.
Renamed usbmidi.c to midi.c for namespace clarity.
Removed more things from usbaudio.h.
The non-standard drivers were adopted accordingly.
Signed-off-by: Daniel Mack <daniel@caiaq.de>
Cc: Clemens Ladisch <clemens@ladisch.de>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
