Fix a long standing memory leak of stream buffers for streams that did not
have a struct video_device allocated: namely the TS and IDX streams.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Update the incoming MDL's buffers' bytesused and sync the buffers for the cpu
in one pass instead of two.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
The encoder was not honoring the MDL size sent in DE_SET_MDL mailbox commands.
This change adjusts the size of the last buffer in an MDL, as reported to the
firmware, so that the encoder will send the exact amount of bytes we specify
per MDL transfer. This eliminates tearing in YUV playback when using
non-default YUV buffer sizes.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Fix YUV capture such that the encoder will pass one frame per transfer. This
will allow the application to maintain frame alignment when a transfer from
the encoder is missed due to high system latency in service the CX23418 IRQ.
Also force YUV buffer sizes to be specified in multiples of 33.75 kB, the
smalled amount of buffer sizes need to store a complete set of HM12 4:2:0
macroblocks specifying 32 lines of the frame. A full 60Hz/525 line
screen requires 15 * 33.75 kB per frame and a full 50Hz/625 line screen
requires 18 * 33.75 kB per frame so the default buffer size is 3 * 33.75 kB,
requiring exactly 5 or 6 buffers per MDL respectively. The bytes needed per
frame and hence MDL need not be the bytes in an integer number of buffers.
However, if frame artifacts are seen with scaled screen sizes, the YUV buffer
size can be set 34 kB (33.75 kB) to get rid of the artifacts at the cost of more
copies between the kernel and userspace.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Add a Memory Descriptor List (MDL) layer to buffer handling to implement
scatter-gather I/O. Currently there is still only 1 buffer per MDL.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Rename type in anticipation of implementing a struct cx18_mdl type that
actually keeps track of a memory descriptor list.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Simplify the way outgoing work handler gets scheduled to send empty buffers
back to the firmware for use. Also reduced the memory required for scheduling
this outgoing work, by using a single, per stream work object.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
To avoid sleeps in providing buffers to user space and in handling incoming
buffers from the capture unit, converted the per stream mutex for locking
queues to 3 spin locks. There is now a spin lock per queue
to increase concurrency when moving buffers around.
Also simplified queue manipulations and buffer handling of incoming buffers
of data from the capture unit.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Renamed structure member name to be more specific to type in anticipation
of updating to the v4l2_device/v4l2_subdev framework. Too many objects named
"dev" and /v4l2_\{0,1\}dev/ would be to confusing.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Removed magic number that referred to firmware limit on the number of
MDLs the firmware can maintain for any stream at any one time.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
cx18: Avoid making firmware API calls with the queue lock held. The source
of MPEG strem corruption when not holding the queue lock was found to be that
the MPEG buffer could be retrieved by the user app before it was sync'ed for
the host cpu. Incoming buffers are now sync'ed before being put on q_full and
releasing the queue lock. We can thus avoid the sometimes lengthy call to
the firmware for CPU_DE_SET_MDL while holding the queue lock, so we can get
better performance.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
cx18: Allow more than 63 capture buffers in rotation per stream. Implement
q_busy to hold buffers the firmware has for use. q_free holds truly unused
buffers in a pool. New buffers are given to the firmware as soon as the
firmware returns one, if there are any to give to the firmware.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Eliminate q_io from stream buffer handling in anticipation of upcoming
changes in buffer handling. q_io was a holdover from ivtv and it's function
in cx18 was trivial and not necessary. We just push things back onto the
front of q_full now, instead of maintaining a 1 buffer q_io queue.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Add copyright attribution for files modified by awalls in 2008
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Restore buffers that have fallen out of the transfer rotation, and check
for coherent mailbox data when processing a stale mailbox.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Convert the per stream queue spinlocks into mutexes. All queue
manipulation happens via the work queue or system calls into the driver, and
not in an interrupt context. This reduces the amout of time the cx18 driver
keeps interrupts disabled.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
A major rewrite of interrupt handling for incoming mailbox processing, to split
the timing critical steps from the the deferrable steps as the sending XPU on
the CX23418 will time out and overwrite our incoming mailboxes rather quickly.
Setup a pool of work "order forms" for the irq handler to send jobs to the new
work handler routine which uses the kernel default work queue to do the
deferrable work. Started optimizing some of the cx18-io calls as they are now
the low hanging fruit for recoving microseconds back from the timeline.
Future optimizations will get rid of mmio read retries, mmio stats logging, and
combine smaller functions in the irq path into the larger ones to save ~2 us
each.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
cx18: Move DVB buffer transfer handling from irq handler to work_queue thread.
In order to properly lock the epu2cpu mailbox for driver to CX23418 commands,
the DVB/TS buffer handling needs to be moved from the IRQ handler and IRQ
context to a work queue. This work_queue implmentation is strikingly similar
to the ivtv implementation - for better or worse.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
cx18: Throttle mmio to/from the CX23418 so boards work in older systems. The
CX23418 couldn't reliably handle mmio at the rate at which the cx18 driver was
attempting to access the chip. The PCI bridge arrangements and settings on
modern motherboards still allowed the CX23418 to work OK, but it didn't work
well on many older motherboards: mysterious I2C errors, firmware loading errors,
etc. This patch adds a throttle to *all* mmio access to the CX23418. It
defaults to a delay of 31 ns, but is adjustable by the mmio_ndelay module parm.
My HVR-1600 and Raptor PAL/SECAM card now function for analog capture on a
motherboard with an Intel 82810E Northbridge and 82801AA Southbridge.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
cx18: Create cx18_ specific wrappers for all pci mmio accessesors. This is a
first step in instrumenting all CX23418 PCI bus IO, to debug problems with
accessing the CX23418's PCI memory mapped IO.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
cx18: Convert cx18_queue buffers member to atomic_t. This allows safe
concurrent access to check if a queue has data without having to acquire the
queue spinlock.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
cx18: Remove redundant struct cx18_queue length member. It can be trivially
computed from queue->buffers * stream->buf_size, if ever really needed.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
cx18: Simplify queue flush logic to prevent oops in cx18_flush_queues().
If accounting of a queue is in error, logic borrowed from ivtv will cause
an oops when flushing the queues for a stream. This change greatly
simplifies the queue flush logic, and sets the queue back to sane
defaults on a flush.
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Fixed a lot of sparse warnings: mostly warnings about shadowed variables
and signed/unsigned mismatches.
Signed-off-by: Hans Verkuil <hverkuil@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
cx18: Add missing lock for when the irq handler manipulates the queues. This
was a potential source of stream queue corruption. Also changed the name of
cx18_queue_find_buf() to cx18_queue_get_buf_irq().
Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
This patch contains the following possible cleanups:
- cx18-i2c.c should #include "cx18-i2c.h" for getting the prototypes of
it's global functions
- make the following needlessly global functions static:
- cx18-fileops.c:cx18_claim_stream()
- cx18-fileops.c:cx18_release_stream()
- cx18-queue.c:cx18_queue_move()
- remove the following unused functions:
- cx18-driver.c:cx18_waitq()
- cx18-queue.c:cx18_buf_copy_from_user()
Signed-off-by: Adrian Bunk <bunk@kernel.org>
Reviewed-by: Hans Verkuil <hverkuil@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
- don't show kernel backtrace when the allocation of the buffers fails: the
normal ivtv/cx18 messages are clear enough and the backtrace scares users.
- fix cleanup after the buffer allocation fails (caused kernel panic).
Signed-off-by: Hans Verkuil <hverkuil@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Many thanks to Steve Toth from Hauppauge and Nattu Dakshinamurthy from
Conexant for their support. I am in particular thankful to Hauppauge
since without their help this driver would not exist. It should also
be noted that Steve did the work to get the DVB part up and running.
Thank you!
Signed-off-by: Hans Verkuil <hverkuil@xs4all.nl>
Signed-off-by: Steven Toth <stoth@hauppauge.com>
Signed-off-by: Michael Krufky <mkrufky@linuxtv.org>
Signed-off-by: G. Andrew Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Andy Walls