Profiling Support for BareMetal Target

Project Infrastructure LLVM Dev List Archives
1

Hi,

I followed two posts as my guideline

http://lists.llvm.org/pipermail/llvm-dev/2017-September/117339.html and
http://lists.llvm.org/pipermail/llvm-dev/2017-September/117156.html

I am using clang 4.0.1 and compiler-rt 4.0 release

1. I made changes in the LCF to mimic Linux Platform (for linker magic)

In data section i have added:

    __start___llvm_prf_cnts = .;
    KEEP(*(__llvm_prf_cnts))
    __stop___llvm_prf_cnts = .;
    __start___llvm_prf_data = .;
    KEEP(*(__llvm_prf_data))
    __stop___llvm_prf_data = .;
    __start___llvm_prf_names = .;
    KEEP(*(__llvm_prf_names))
    __stop___llvm_prf_names = .;
    __start___llvm_prf_vnds = .;
    KEEP(*(__llvm_prf_vnds))
    __stop___llvm_prf_vnds = .;

2. Since there are not .ctors or atexit available i called the
functions lprofSetupValueProfiler and __llvm_profile_initialize_file
from the startup before main and __llvm_profile_write_file during the
exit after main.

3. Made changes in the compiler-rt library to adapt the profile
sources for a bare metal target.

When i run the elf file of a simple return 0 main i do get a
default.profraw file and after processing the raw data using
llvm-profdata and llvm-cov all i get is
error: a.out: Failed to load coverage: No coverage data found

I could see the following :

$ llvm-profdata.exe show -detailed-summary default.profraw
Total functions: 1
Maximum function count: 1
Maximum internal block count: 0
Detailed summary:
Total number of blocks: 1
Total count: 1
0 blocks with count >= 0 account for 80 percentage of the total counts.
0 blocks with count >= 0 account for 90 percentage of the total counts.
0 blocks with count >= 0 account for 95 percentage of the total counts.
0 blocks with count >= 0 account for 99 percentage of the total counts.
0 blocks with count >= 0 account for 99.9 percentage of the total counts.
0 blocks with count >= 0 account for 99.99 percentage of the total counts.
0 blocks with count >= 0 account for 99.999 percentage of the total counts.

I am not sure what is wrong.
In the profiling library i see many entry points. I used
RegisterRuntime constructor to decide on the functions, that i
mentioned above, to be called in the startup routine . I am right in
that ?

Any idea as to what i am doing wrong ?

Regards,
Shafi

2

Hi,

Hi,

I followed two posts as my guideline

[llvm-dev] Using source-based code coverage on baremetal and
[llvm-dev] Using source-based code coverage on baremetal

I am using clang 4.0.1 and compiler-rt 4.0 release

1. I made changes in the LCF to mimic Linux Platform (for linker magic)

In data section i have added:

   __start___llvm_prf_cnts = .;
   KEEP(*(__llvm_prf_cnts))
   __stop___llvm_prf_cnts = .;
   __start___llvm_prf_data = .;
   KEEP(*(__llvm_prf_data))
   __stop___llvm_prf_data = .;
   __start___llvm_prf_names = .;
   KEEP(*(__llvm_prf_names))
   __stop___llvm_prf_names = .;
   __start___llvm_prf_vnds = .;
   KEEP(*(__llvm_prf_vnds))
   __stop___llvm_prf_vnds = .;

2. Since there are not .ctors or atexit available i called the
functions lprofSetupValueProfiler and __llvm_profile_initialize_file
from the startup before main and __llvm_profile_write_file during the
exit after main.

3. Made changes in the compiler-rt library to adapt the profile
sources for a bare metal target.

When i run the elf file of a simple return 0 main i do get a
default.profraw file and after processing the raw data using
llvm-profdata and llvm-cov all i get is
error: a.out: Failed to load coverage: No coverage data found

I could see the following :

$ llvm-profdata.exe show -detailed-summary default.profraw
Total functions: 1
Maximum function count: 1
Maximum internal block count: 0
Detailed summary:
Total number of blocks: 1
Total count: 1
0 blocks with count >= 0 account for 80 percentage of the total counts.
0 blocks with count >= 0 account for 90 percentage of the total counts.
0 blocks with count >= 0 account for 95 percentage of the total counts.
0 blocks with count >= 0 account for 99 percentage of the total counts.
0 blocks with count >= 0 account for 99.9 percentage of the total counts.
0 blocks with count >= 0 account for 99.99 percentage of the total counts.
0 blocks with count >= 0 account for 99.999 percentage of the total counts.

I am not sure what is wrong.
In the profiling library i see many entry points. I used
RegisterRuntime constructor to decide on the functions, that i
mentioned above, to be called in the startup routine . I am right in
that ?

Any idea as to what i am doing wrong ?

Perhaps there's just no __llvm_covmap section in the binary? That would explain the error. Can you retain that section with your linker script?

best,
vedant