schedule event doesn't work with time-filter

[honggyukim]

Please see below. The time-filter is set to 10ms, but there are events which are under the time threshold.

$ uftrace record -t 10ms ./clang hello.c

$ uftrace replay
# DURATION     TID     FUNCTION
            [178116] | _GLOBAL__sub_I_ARMFrameLowering.cpp() {
   5.945 ms [178116] |   /* linux:schedule */
            [178116] |   llvm::cl::Option::addArgument() {
 271.693 us [178116] |     /* linux:schedule */
            [178116] |     llvm::ManagedStaticBase::RegisterManagedStatic() {
   1.760 ms [178116] |       /* linux:schedule */
   9.018 ms [178116] |       /* linux:schedule */
 197.215 us [178116] |       /* linux:schedule */
  11.100 ms [178116] |     } /* llvm::ManagedStaticBase::RegisterManagedStatic */
 191.312 us [178116] |     /* linux:schedule */
  11.631 ms [178116] |   } /* llvm::cl::Option::addArgument */
  17.620 ms [178116] | } /* _GLOBAL__sub_I_ARMFrameLowering.cpp */
            [178116] | _GLOBAL__sub_I_ARMAsmParser.cpp() {
            [178116] |   __static_initialization_and_destruction_0() {
  19.550 ms [178116] |     /* linux:schedule */
   2.124 ms [178116] |     /* linux:schedule */
  21.767 ms [178116] |   } /* __static_initialization_and_destruction_0 */
  21.768 ms [178116] | } /* _GLOBAL__sub_I_ARMAsmParser.cpp */
            [178116] | _GLOBAL__sub_I_MachineScheduler.cpp() {
            [178116] |   __static_initialization_and_destruction_0() {
   2.450 ms [178116] |     /* linux:schedule */
   7.352 ms [178116] |     /* linux:schedule */
 344.297 us [178116] |     /* linux:schedule */
  10.342 ms [178116] |   } /* __static_initialization_and_destruction_0 */
  10.343 ms [178116] | } /* _GLOBAL__sub_I_MachineScheduler.cpp */
            [178116] | _GLOBAL__sub_I_DwarfDebug.cpp() {
            [178116] |   __static_initialization_and_destruction_0() {
  19.927 ms [178116] |     /* linux:schedule */
  20.027 ms [178116] |   } /* __static_initialization_and_destruction_0 */
  20.027 ms [178116] | } /* _GLOBAL__sub_I_DwarfDebug.cpp */
            [178116] | _GLOBAL__sub_I_ModuleSummaryIndex.cpp() {
  15.123 ms [178116] |   /* linux:schedule */
  15.185 ms [178116] | } /* _GLOBAL__sub_I_ModuleSummaryIndex.cpp */
            [178116] | main() {
   8.016 ms [178116] |   /* linux:schedule */
   5.461 ms [178116] |   /* linux:schedule */
   8.761 ms [178116] |   /* linux:schedule */
            [178116] |   LLVMInitializeARMTarget() {
   9.735 ms [178116] |     /* linux:schedule */
 170.202 us [178116] |     /* linux:schedule */
 339.266 us [178116] |     /* linux:schedule */
 338.426 us [178116] |     /* linux:schedule */
 314.810 us [178116] |     /* linux:schedule */
  11.775 ms [178116] |   } /* LLVMInitializeARMTarget */
 220.885 us [178116] |   /* linux:schedule */
            [178116] |   LLVMInitializeX86Target() {
        ...

[namhyung]

The scheduling and task events are recorded asynchronously using perf event and it doesn't aware of the time filter setting. You need to set it for replay again.. sorry.

[honggyukim]

We may force behind to use time-filter setting of recorded data for replay and other analysis commands.

[namhyung]

That'd be a possible way..

[honggyukim]

Should we add one more special info bit to uftrace.data/info for time-filter setting?

[honggyukim]

As mentioned in #699 (comment), this problem makes the replay and other outputs kind of ugly in some cases. So I always have to say the users to use time-filter for all the other analysis commands.