Talk:V4L Test Suite

From LinuxTVWiki
Jump to: navigation, search

It seems the 2 V4L2 examples (capture.c and vivi.c) don't play well with each other. I am not really sure where to post about this, so I will start here and move it to the appropriate place once I find it. so far http://bugzilla.kernel.org seems the most likely. I am dumping my notes on this page to try to keep the point of the article page from getting buried.


Bugs in Examples

capture.c and vivi.c are V4L2 examples:

capture: "Video for Linux Two API Specification: Revision 0.24 Appendix B. Video Capture Example"

vivi: "It is intended to be a working example code for current V4L2 Driver and Stub driver to facilitate the implementation of new video drivers.

They don't play well with each other. There is a memory leak, and I think vivi is telling capture to use a buffer mode that isn't valid - I call this "capabilities mismatch" but I am not really sure what the problem is.

Summary

detailed report: "vivi memory leak = kernel panic" https://bugs.launchpad.net/ubuntu/+source/linux/+bug/294951

memory leak
sudo modprobe vivi
gcc capture.c -g -o capture
valgrind ./capture --userp -d /dev/video1
==17153== malloc/free: in use at exit: 2,457,632 bytes in 5 blocks.
capabilities mismatch
./capture --userp -d /dev/video1
VIDIOC_QBUF error 22, Invalid argument

Detail

setup for both: (capture.c is the original, same problem with capture_example.c, see URL above)

wget http://www.linuxtv.org/downloads/video4linux/API/V4L2_API/v4l2spec/capture.c
gcc capture.c -g -o capture
sudo modprobe vivi
dmesg | grep /dev/video1
[ 1015.491550] vivi: V4L2 device registered as /dev/video1
memory leak
# memleak.sh
# runs capture to cause "Cannot allocate memory"
# and logs /proc/meminfo along the way

set -x

sudo modprobe vivi
dmesg|grep vivi

# make sure the above grep matchs:
DEV=/dev/video0

if [ -d proc ]; then
  # wipe out previous run
  rm -f proc/meminfo*
else
  # make a place to log proc/meminfo
  mkdir proc
fi

# 0.0 = start
let a=0
cp /proc/meminfo proc/meminfo$a.0.txt

# loop until "Cannot allocate memory"
while true; do

  ((a+=1))

  # run using valid params, therefore should not error.
  ./capture --read -d $DEV
  # unless something has sucked all the memory from somewhere.
  if [ "0" != "$?" ]; then break; fi
  # shouldn't be anything interesting, but log the state, just to be sure.
  cp /proc/meminfo proc/meminfo$a.1.txt
  
  # run with unsupported buffer mode
  ./capture --userp -d $DEV
  # which didn't release memory from who knows where, so log the state:
  cp /proc/meminfo proc/meminfo$a.2.txt

done

# x.3 = final state.  should be about the same as x.2
cp /proc/meminfo proc/meminfo$a.3.txt

resulting data: http://spreadsheets.google.com/ccc?key=pIfz0wOzPtW1-oZkXXbvcLA&hl=en

The runaways: HighFree go from 72608 to 252 VmallocUsed from 6440 to 109724 VmallocChunk from 103620 to 336


Using valgrind (a suite of tools for debugging and profiling programs)

 valgrind -v --leak-check=full --show-reachable=yes ./capture --userp -d /dev/video1
 
 ==5166== Memcheck, a memory error detector.
 ==5166== Copyright (C) 2002-2007, and GNU GPL'd, by Julian Seward et al.
 ==5166== Using LibVEX rev 1854, a library for dynamic binary translation.
 ==5166== Copyright (C) 2004-2007, and GNU GPL'd, by OpenWorks LLP.
 ==5166== Using valgrind-3.3.1-Debian, a dynamic binary instrumentation framework.
 ==5166== Copyright (C) 2000-2007, and GNU GPL'd, by Julian Seward et al.
 ==5166==
 --5166-- Command line
 --5166--    ./capture
 --5166--    --userp
 --5166--    -d
 --5166--    /dev/video1
 --5166-- Startup, with flags:
 --5166--    --suppressions=/usr/lib/valgrind/debian-libc6-dbg.supp
 --5166--    -v
 --5166--    --leak-check=full
 --5166--    --show-reachable=yes
 --5166-- Contents of /proc/version:
 --5166--   Linux version 2.6.27-7-generic (buildd@palmer) (gcc version 4.3.2
 (Ubuntu 4.3.2-1ubuntu11) ) #1 SMP Thu Oct 30 04:18:38 UTC 2008
 --5166-- Arch and hwcaps: X86, x86-sse1-sse2
 --5166-- Page sizes: currently 4096, max supported 4096
 --5166-- Valgrind library directory: /usr/lib/valgrind
 --5166-- Reading syms from /lib/ld-2.8.90.so (0x4000000)
 --5166-- Reading debug info from /lib/ld-2.8.90.so...
 --5166-- ... CRC mismatch (computed 371b8ee6 wanted cc0a418a)
 --5166-- Reading debug info from /usr/lib/debug/lib/ld-2.8.90.so...
 --5166-- Reading syms from /home/juser/vga2usb/v4l.org/examples/capture (0x8048000)
 --5166-- Reading syms from /usr/lib/valgrind/x86-linux/memcheck (0x38000000)
 --5166--    object doesn't have a dynamic symbol table
 --5166-- Reading suppressions file: /usr/lib/valgrind/debian-libc6-dbg.supp
 --5166-- Reading suppressions file: /usr/lib/valgrind/default.supp
 --5166-- REDIR: 0x40155d0 (index) redirected to 0x3802cf63
 (vgPlain_x86_linux_REDIR_FOR_index)
 --5166-- Reading syms from /usr/lib/valgrind/x86-linux/vgpreload_core.so (0x401F000)
 --5166-- Reading syms from /usr/lib/valgrind/x86-linux/vgpreload_memcheck.so
 (0x4022000)
 ==5166== WARNING: new redirection conflicts with existing -- ignoring it
 --5166--     new: 0x040155d0 (index               ) R-> 0x040261a0 index
 --5166-- REDIR: 0x40157c0 (strlen) redirected to 0x4026450 (strlen)
 --5166-- Reading syms from /usr/lib/debug/libc-2.8.90.so (0x402A000)
 --5166-- REDIR: 0x409d2f0 (rindex) redirected to 0x4026080 (rindex)
 --5166-- REDIR: 0x409cf00 (strlen) redirected to 0x4026430 (strlen)
 --5166-- REDIR: 0x409d110 (strncmp) redirected to 0x40266a0 (strncmp)
 --5166-- REDIR: 0x409c7e0 (index) redirected to 0x4026170 (index)
 --5166-- REDIR: 0x409df70 (memset) redirected to 0x4027340 (memset)
 --5166-- REDIR: 0x4099c50 (calloc) redirected to 0x4023d20 (calloc)
 --5166-- REDIR: 0x409a190 (memalign) redirected to 0x4023b80 (memalign)
 --5166-- REDIR: 0x409e490 (memcpy) redirected to 0x40268a0 (memcpy)
 --5166-- REDIR: 0x409e180 (stpcpy) redirected to 0x40270d0 (stpcpy)
 --5166-- REDIR: 0x409dfd0 (mempcpy) redirected to 0x4027470 (mempcpy)
 --5166-- REDIR: 0x409f010 (strchrnul) redirected to 0x4027410 (strchrnul)
 
 VIDIOC_QBUF error 22, Invalid argument
 
 --5166-- REDIR: 0x4097730 (free) redirected to 0x4024a90 (free)
 ==5166==
 ==5166== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 11 from 1)
 --5166--
 --5166-- supp:     11 dl-hack3-cond-1
 ==5166== malloc/free: in use at exit: 2,457,632 bytes in 5 blocks.
 ==5166== malloc/free: 5 allocs, 0 frees, 2,457,632 bytes allocated.
 ==5166==
 ==5166== searching for pointers to 5 not-freed blocks.
 ==5166== checked 51,772 bytes.
 ==5166==
 ==5166== 32 bytes in 1 blocks are still reachable in loss record 1 of 2
 ==5166==    at 0x4023DE2: calloc (vg_replace_malloc.c:397)
 ==5166==    by 0x80494AF: init_userp (in
 /home/juser/vga2usb/v4l.org/examples/capture)
 ==5166==    by 0x80498B0: init_device (in
 /home/juser/vga2usb/v4l.org/examples/capture)
 ==5166==    by 0x8049B63: main (in /home/juser/vga2usb/v4l.org/examples/capture)
 ==5166==
 ==5166==
 ==5166== 2,457,600 bytes in 4 blocks are still reachable in loss record 2 of 2
 ==5166==    at 0x4023C4A: memalign (vg_replace_malloc.c:460)
 ==5166==    by 0x8049536: init_userp (in
 /home/juser/vga2usb/v4l.org/examples/capture)
 ==5166==    by 0x80498B0: init_device (in
 /home/juser/vga2usb/v4l.org/examples/capture)
 ==5166==    by 0x8049B63: main (in /home/juser/vga2usb/v4l.org/examples/capture)
 ==5166==
 ==5166== LEAK SUMMARY:
 ==5166==    definitely lost: 0 bytes in 0 blocks.
 ==5166==      possibly lost: 0 bytes in 0 blocks.
 ==5166==    still reachable: 2,457,632 bytes in 5 blocks.
 ==5166==         suppressed: 0 bytes in 0 blocks.
 --5166--  memcheck: sanity checks: 0 cheap, 1 expensive
 --5166--  memcheck: auxmaps: 0 auxmap entries (0k, 0M) in use
 --5166--  memcheck: auxmaps_L1: 0 searches, 0 cmps, ratio 0:10
 --5166--  memcheck: auxmaps_L2: 0 searches, 0 nodes
 --5166--  memcheck: SMs: n_issued      = 10 (160k, 0M)
 --5166--  memcheck: SMs: n_deissued    = 0 (0k, 0M)
 --5166--  memcheck: SMs: max_noaccess  = 65535 (1048560k, 1023M)
 --5166--  memcheck: SMs: max_undefined = 34 (544k, 0M)
 --5166--  memcheck: SMs: max_defined   = 20 (320k, 0M)
 --5166--  memcheck: SMs: max_non_DSM   = 10 (160k, 0M)
 --5166--  memcheck: max sec V bit nodes:    0 (0k, 0M)
 --5166--  memcheck: set_sec_vbits8 calls: 0 (new: 0, updates: 0)
 --5166--  memcheck: max shadow mem size:   464k, 0M
 --5166-- translate:            fast SP updates identified: 1,933 ( 89.4%)
 --5166-- translate:   generic_known SP updates identified: 128 (  5.9%)
 --5166-- translate: generic_unknown SP updates identified: 100 (  4.6%)
 --5166--     tt/tc: 4,312 tt lookups requiring 4,351 probes
 --5166--     tt/tc: 4,312 fast-cache updates, 3 flushes
 --5166--  transtab: new        2,035 (43,121 -> 617,673; ratio 143:10) [0 scs]
 --5166--  transtab: dumped     0 (0 -> ??)
 --5166--  transtab: discarded  8 (200 -> ??)
 --5166-- scheduler: 25,512 jumps (bb entries).
 --5166-- scheduler: 0/2,355 major/minor sched events.
 --5166--    sanity: 1 cheap, 1 expensive checks.
 --5166--    exectx: 769 lists, 8 contexts (avg 0 per list)
 --5166--    exectx: 16 searches, 8 full compares (500 per 1000)
 --5166--    exectx: 4 cmp2, 26 cmp4, 0 cmpAll
 --5166--  errormgr: 8 supplist searches, 271 comparisons during search
 --5166--  errormgr: 11 errlist searches, 26 comparisons during search

from v4l list Andy Walls:

It looks like init_userp() allocated a bunch of "buffers", which has to happen for a program to use user pointer mode of v4l2. The function errno_exit() doesn't bother to clean up when the VIDIOC_QBUF ioctl() call fails. free() is only called by uninit_device(). Since the alternate flow of the program through errno_exit() to termination doesn't call free() on "buffers", you should have a process heap memory leak on error exit.

program didn't close it's file descriptors with the driver on errno_exit()

capabilities mismatch

Looking at the source, capture.c should have reported "does not support user pointer" by:

411 init_userp                      (unsigned int           buffer_size)
421     req.count               = 4;
       req.type                = V4L2_BUF_TYPE_VIDEO_CAPTURE;
       req.memory              = V4L2_MEMORY_USERPTR;
427        if (-1 == xioctl (fd, VIDIOC_REQBUFS, &req)) {
               if (EINVAL == errno) {
                       fprintf (stderr, "%s does not support "
                                "user pointer i/o\n", dev_name);
                       exit (EXIT_FAILURE);


But instead it errors here:

250 start_capturing                 (void)
260        case IO_METHOD_USERPTR:
                       if (-1 == xioctl (fd, VIDIOC_QBUF, &buf))
                               errno_exit ("VIDIOC_QBUF");

Canonical source of capture and vivi