rr replay error at signal handler. mcontext->__reserved[i] data doesn't fully copyed after syscall

[IamHuskar]

arch : aarch64
os:ububtu 22.04

mcontext->__reserved[i] data doesn't fully copyed after syscall

// pls remove .txt suffix .
aarch64_mcontext_poc.pls_remove_sufixx.txt
aarch64_mcontext_poc.cpp.txt

g++ -g -o aarch64_mcontext_poc aarch64_mcontext_poc.cpp

excepted result:
Accessing address 0... pc skip 4 bytes at signall handler
arm ok value 92000045 (SIGSEGV).
after skip

rr record aarch64_mcontext_poc
rr replay

[FATAL src/ReplaySession.cc:582:cont_syscall_boundary()]
(task 983741 (rec:983353) at time 138)
-> Assertion `false' failed to hold. Replay got unrecorded signal {signo:SIGSEGV,errno:SUCCESS,code:SEGV_MAPERR,addr:0xffffd66a4dd3}

set breakpoint at GetArmESR
MContext->__reserved[0] is ok . but the data in MContext->__reserved[1] when running rr record is different from the data when running in gdb

static inline uint64_t GetArmESR(void* ucontext) {
  auto MContext = GetMContext(ucontext);

  size_t i = 0;
  /*
   *  maybe not full  MContext->__reserved[i] copyed after return from syscall
   * rr replay 
   * after i+=HostState->Size
   *  *HostState is not valid data.
   * 
   */
  auto HostState = (struct HostCTXHeader*)(&MContext->__reserved[i]);
  do {
    if (HostState->Magic == ESR1_MAGIC) {
      auto ESR = (struct HostESRState*)(HostState);
      //gdb got here
      return ESR->ESR;
    }
    i += HostState->Size;
    HostState = (struct HostCTXHeader*)(&MContext->__reserved[i]);
    
  } while (HostState->Size != 0);
  
  // rr replay got here
  return 0;
}


static inline uint32_t GetProtectFlags(void* ucontext) {
  uint64_t ESR = GetArmESR(ucontext);
  if (!((ESR & ESR1_EC) == ESR1_EC_DataAbort || (ESR & ESR1_EC) == ESR1_EC_InstAbort)) {
    printf("arm error "); // rr replay got here
    ::exit(-100);
  }
  else{
	  printf("arm ok ");//gdb ot here
  }
  return ESR;
}

[IamHuskar]

@rocallahan

I can reproduce this bug, but I really don't know how to fix it. rr is very important for my debugging work. I need help
I stopped the analysis because this bug prevented me from debugging further.

[rocallahan]

This is pretty simple, the captured sigframe size is simply too low.

[IamHuskar]

This is pretty simple, the captured sigframe size is simply too low.

thank you very much. I was trying to fix this .but I am not very familiar with rr source code.
rr is a very useful tool for me ,BigBigBigBig thanks to you . @rocallahan you are my sunshine ,my only sunshine

[alehander92]

I seriously think this is the best rr testimonial i've read lately