futex does not wake

[vans163]

Running FFMPEG inside graphene direct or sgx FUTEX_CLOCK_REALTIME is not implemented, and it causes the FUTEX to get the wrong time (notice the -1594868808) causing infinite stalls on threads.

EDIT: The issue with not waking remains but it seems that its not related to FUTEX_CLOCK_REALTIME.

[P711262:T4:ffmpeg] trace: ---- shim_futex(0x1a0f03380, FUTEX_PRIVATE|FUTEX_CLOCK_REALTIME|FUTEX_WAIT_BITSET, 0, 0x0, 0x0, -1) ...
[P711262:T8:ffmpeg] warning: Ignoring FUTEX_CLOCK_REALTIME flag
[P711262:T1:ffmpeg] trace: ---- return from shim_read(...) = 0x8000
[P711262:T1:ffmpeg] trace: ---- shim_futex(0x1a0f03bc4, FUTEX_PRIVATE|FUTEX_WAKE, 1, 0x0, 0x1a0f03b98, -1594868808) ...
[P711262:T1:ffmpeg] trace: ---- return from shim_futex(...) = 0x1
[P711262:T10:ffmpeg] trace: ---- return from shim_futex(...) = 0x0
[P711262:T1:ffmpeg] trace: ---- shim_clock_gettime(1, 0x1a89536f0) = 0x0
[P711262:T1:ffmpeg] trace: ---- shim_futex(0x1a0f030c0, FUTEX_PRIVATE|FUTEX_WAKE, 1, 0x0, 0x1a0f03098, -1594871624) ...
[P711262:T1:ffmpeg] trace: ---- return from shim_futex(...) = 0x1
[P711262:T2:ffmpeg] trace: ---- return from shim_futex(...) = 0x0
[P711262:T1:ffmpeg] trace: ---- shim_clock_gettime(1, 0x1a89536f0) = 0x0
[P711262:T1:ffmpeg] trace: ---- shim_pselect6(1, 0x1a89523d0, 0x0, 0x0, 0x1a8952380, 0x0) ...
[P711262:T1:ffmpeg] trace: ---- return from shim_pselect6(...) = 0x1
[P711262:T1:ffmpeg] trace: ---- shim_read(0, 0x1a89523bc, 0x1) ...
[P711262:T1:ffmpeg] trace: ---- return from shim_read(...) = 0x1 

Running outside graphene notice the signed int max value for the futex.

   520.702 ( 0.002 ms): ffmpeg/1611 futex(uaddr: 0x55d35291fa08, op: WAKE|PRIVATE_FLAG, val: 1)           = 0
   520.706 (         ): ffmpeg/1611 futex(uaddr: 0x55d35291f9a0, op: WAIT_BITSET|PRIVATE_FLAG|CLOCK_REALTIME, val3: MATCH_ANY) ...
   523.278 ( 0.005 ms): ffmpeg/1612 futex(uaddr: 0x55d35291fb68, op: WAKE|PRIVATE_FLAG, val: 1)           = 0
   518.458 ( 6.058 ms): ffmpeg/1610  ... [continued]: futex())                                            = 0
   523.285 (         ): ffmpeg/1612 futex(uaddr: 0x55d35291fb00, op: WAIT_BITSET|PRIVATE_FLAG|CLOCK_REALTIME, val3: MATCH_ANY) ...
   526.510 ( 0.002 ms): ffmpeg/1610 futex(uaddr: 0x55d35291f8a8, op: WAKE|PRIVATE_FLAG, val: 1)           = 0
   526.513 (         ): ffmpeg/1610 futex(uaddr: 0x55d35291f844, op: WAIT_BITSET|PRIVATE_FLAG|CLOCK_REALTIME, val3: MATCH_ANY) ...
   524.484 ( 0.007 ms): ffmpeg/1609 futex(uaddr: 0x55d35291f840, op: WAKE|PRIVATE_FLAG, val: 1)           = 1
   526.393 ( 0.007 ms): ffmpeg/1609 futex(uaddr: 0x55d35291f9a0, op: WAKE|PRIVATE_FLAG, val: 1)           = 1
   520.706 ( 5.696 ms): ffmpeg/1611  ... [continued]: futex())                                            = 0
   526.404 (         ): ffmpeg/1611 futex(uaddr: 0x55d35291fa08, op: WAIT|PRIVATE_FLAG, val: 2)        ...
   526.407 ( 0.002 ms): ffmpeg/1609 futex(uaddr: 0x55d35291fa08, op: WAKE|PRIVATE_FLAG, val: 1)           = 1
   526.404 ( 0.006 ms): ffmpeg/1611  ... [continued]: futex())                                            = 0
   528.647 ( 0.006 ms): ffmpeg/1611 futex(uaddr: 0x55d35291fa08, op: WAKE|PRIVATE_FLAG, val: 1)           = 0
   528.654 (         ): ffmpeg/1611 futex(uaddr: 0x55d35291f9a4, op: WAIT_BITSET|PRIVATE_FLAG|CLOCK_REALTIME, val3: MATCH_ANY) ...
   529.458 ( 0.006 ms): ffmpeg/1609 write(fd: 2</dev/pts/0>, buf: 0x7fff57160904, count: 81)              = 81
   529.559 ( 0.004 ms): ffmpeg/1609 futex(uaddr: 0x55d35291fb00, op: WAKE|PRIVATE_FLAG, val: 1)           = 1
   523.285 ( 6.279 ms): ffmpeg/1612  ... [continued]: futex())                                            = 0
   529.566 (         ): ffmpeg/1612 futex(uaddr: 0x55d35291fb68, op: WAIT|PRIVATE_FLAG, val: 2)        ...
   529.569 ( 0.002 ms): ffmpeg/1609 futex(uaddr: 0x55d35291fb68, op: WAKE|PRIVATE_FLAG, val: 1)           = 1
   529.566 ( 0.005 ms): ffmpeg/1612  ... [continued]: futex())                                            = 0
   526.513 ( 6.165 ms): ffmpeg/1610  ... [continued]: futex())                                            = 0
   532.681 (         ): ffmpeg/1610 futex(uaddr: 0x55d35291f8a8, op: WAIT|PRIVATE_FLAG, val: 2)        ...
   532.665 ( 0.010 ms): ffmpeg/1609 futex(uaddr: 0x55d35291f844, op: WAKE|PRIVATE_FLAG, val: 1)           = 1
   532.681 ( 0.540 ms): ffmpeg/1610  ... [continued]: futex())                                            = 0
   533.215 ( 0.004 ms): ffmpeg/1609 futex(uaddr: 0x55d35291f8a8, op: WAKE|PRIVATE_FLAG, val: 1)           = 1
   535.520 ( 0.009 ms): ffmpeg/1609 futex(uaddr: 0x55d35291f9a4, op: WAKE|PRIVATE_FLAG, val: 1)           = 1
   528.654 ( 6.935 ms): ffmpeg/1611  ... [continued]: futex())                                            = 0
   535.595 (         ): ffmpeg/1611 futex(uaddr: 0x55d35291fa08, op: WAIT|PRIVATE_FLAG, val: 2)        ...
   535.634 (         ): ffmpeg/1610 futex(uaddr: 0x55d35291fb34, op: WAIT_BITSET|PRIVATE_FLAG|CLOCK_REALTIME, val3: MATCH_ANY) ...
   535.717 ( 0.003 ms): ffmpeg/1612 futex(uaddr: 0x55d35291fb34, op: WAKE|PRIVATE_FLAG, val: 2147483647)  = 1

[dimakuv]

Looks like a question to @boryspoplawski

[boryspoplawski]

Running FFMPEG inside graphene direct or sgx FUTEX_CLOCK_REALTIME is not implemented, and it causes the FUTEX to get the wrong time (notice the -1594868808) causing infinite stalls on threads.

In your trace -1594868808 is the last argument to futex, which is unused (ignored) in FUTEX_WAKE.
Graphene implements only one clock, so ignoring FUTEX_CLOCK_REALTIME should be safe (as application cannot get different times for different clocks).

[vans163]

Running FFMPEG inside graphene direct or sgx FUTEX_CLOCK_REALTIME is not implemented, and it causes the FUTEX to get the wrong time (notice the -1594868808) causing infinite stalls on threads.

In your trace -1594868808 is the last argument to futex, which is unused (ignored) in FUTEX_WAKE.
Graphene implements only one clock, so ignoring FUTEX_CLOCK_REALTIME should be safe (as application cannot get different times for different clocks).

Okay thanks, it seems the issue stems from something else then (not related to FUTEX_CLOCK_REALTIME).

[boryspoplawski]

Unfortunately there seems to be nothing wrong in the logs you've provided, so I won't be able to help more without more information.

[mkow]

I believe there's no problem with futex syscall here, and FFMPEG troubles were caused by #2553, so I'm closing this issue.
@vans163 Feel free to reopen if you think that there are still some problems with futexes.