[DeviceSanitizer] Implement symbolizer for more readable information

[zhaomaosu]

SYCLOS Part: intel/llvm#14513

[yingcong-wu]

Could you provide a comparison of the output text so that we can see the result difference?

[zhaomaosu]

Take the case https://github.com/intel/llvm/blob/sycl/sycl/test-e2e/AddressSanitizer/use-after-free/use-after-free.cpp as a example.

Before this change, the output just like:

====ERROR: DeviceSanitizer: use-after-free on address 0x118d0f0
READ of size 1 at kernel <typeinfo name for main::{lambda(sycl::_V1::handler&)#1}::operator()(sycl::_V1::handler&) const::MyKernel> LID(0, 0, 0) GID(0, 0, 0)
  #0 main::'lambda'(sycl::_V1::handler&)::operator()(sycl::_V1::handler&) const::'lambda'(sycl::_V1::nd_item<1>)::operator()(sycl::_V1::nd_item<1>) const .*use-after-free.cpp:16

0x118d0f0 is located inside of Device USM region [0x118d0f0, 0x118d4f0)
allocated here:
  #0 ./use-after-free() [0x4061fe]
  #1 ./use-after-free() [0x404975]
  #2 ./use-after-free() [0x4036e9]
  #3 /lib64/libc.so.6(+0x44e50) [0x7fe60a1a2e50]
  #4 /lib64/libc.so.6(__libc_start_main+0x7c) [0x7fe60a1a2efc]
  #5 ./use-after-free() [0x403595]

released here:
  #0 ./use-after-free() [0x403750]
  #1 /lib64/libc.so.6(+0x44e50) [0x7fe60a1a2e50]
  #2 /lib64/libc.so.6(__libc_start_main+0x7c) [0x7fe60a1a2efc]
  #3 ./use-after-free() [0x403595]

After this change, the output just like:

====ERROR: DeviceSanitizer: use-after-free on address 0xff7c2c
READ of size 1 at kernel <typeinfo name for main::{lambda(sycl::_V1::handler&)#1}::operator()(sycl::_V1::handler&) const::MyKernel> LID(0, 0, 0) GID(300, 0, 0)
  #0 main::'lambda'(sycl::_V1::handler&)::operator()(sycl::_V1::handler&) const::'lambda'(sycl::_V1::nd_item<1>)::operator()(sycl::_V1::nd_item<1>) const .*use-after-free.cpp:16

0xff7c2c is located inside of Device USM region [0xff7b00, 0xff7f00)
allocated here:
  #0 in char* sycl::_V1::malloc_device<char>(unsigned long, sycl::_V1::device const&, sycl::_V1::context const&, sycl::_V1::property_list const&, sycl::_V1::detail::code_location const&) .*syclos/build/bin/../include/sycl/usm.hpp:172:3
  #1 in char* sycl::_V1::malloc_device<char>(unsigned long, sycl::_V1::queue const&, sycl::_V1::property_list const&, sycl::_V1::detail::code_location const&) .*syclos/build/bin/../include/sycl/usm.hpp:180:10
  #2 in main .*syclos/sycl/test-e2e/AddressSanitizer/use-after-free/use-after-free.cpp:10:17
  #3 in __stop___libc_freeres_ptrs (/lib64/libc.so.6+0x7f811dea7e50)
  #4 in __stop___libc_freeres_ptrs (/lib64/libc.so.6+0x7f811dea7efc)
  #5 in _start (./use-after-free+0x403595)

released here:
  #0 in main .*syclos/sycl/test-e2e/AddressSanitizer/use-after-free/use-after-free.cpp:11:3
  #1 in __stop___libc_freeres_ptrs (/lib64/libc.so.6+0x7f811dea7e50)
  #2 in __stop___libc_freeres_ptrs (/lib64/libc.so.6+0x7f811dea7efc)
  #3 in _start (./use-after-free+0x403595)

[AllanZyne]

I know I was a bit hasty in implementing this part at first, in fact, the symbolization should be invoked as late as possible.
Can you save as few as possible information when doing backtrace, and symbolize them when really need it?
Thank you!

[AllanZyne]

I think we can get handle of the module, not file path

[yingcong-wu]

I know I was a bit hasty in implementing this part at first, in fact, the symbolization should be invoked as late as possible. Can you save as few as possible information when doing backtrace, and symbolize them when really need it? Thank you!

I agree, the symbolization takes time, so we should only do it when necessary (in error reporting stage I think).

[zhaomaosu]

nice suggestions. updated.

[yingcong-wu]

The new error reporting message looks good and more useful, nice improvement!

[yingcong-wu]

lgtm

[yingcong-wu]

Please provide the link to the PR of intel/llvm.