Re: [PATCH v2 06/10] perf trace: Pretty print struct data

[Arnaldo Carvalho de Melo <acme@kernel.org>]

On Fri, Aug 23, 2024 at 09:41:41AM -0300, Arnaldo Carvalho de Melo wrote:
> One other thing I think that the skel patch should come before these,
> so that at this point in the series I could _test_ struct printing.

So I did that, need more polishing I have now:

⬢[acme@toolbox perf-tools-next]$ git log --oneline -10
78bb4b917b942f7d (HEAD -> perf-tools-next) perf trace: Pretty print struct data
027ae076fa7bc068 perf trace: Pass the richer 'struct syscall_arg' pointer to trace__btf_scnprintf()
4f2ac4e8a8d9488a perf trace: Add trace__bpf_sys_enter_beauty_map() to prepare for fetching data in BPF
545008dcff9e06f6 perf trace: Collect augmented data using BPF
0b5a34a3cf98843a perf trace: Fix perf trace -p <PID>
23da4ec3640538fa perf evlist: Introduce method to find if there is a bpf-output event
00dc514612fe98cf perf python: Disable -Wno-cast-function-type-mismatch if present on clang
b81162302001f411 perf python: Allow checking for the existence of warning options in clang
1cfd01eb602d73b9 perf annotate-data: Copy back variable types after move
895891dad7353d60 perf annotate-data: Update stack slot for the store
⬢[acme@toolbox perf-tools-next]$

And the comment on the struct pretty printer, added to the patch where
it is introduced.

Committer testing:

After moving the changes to the BPF skel to _before_ this patch, we're
able to test this patch at this point in the series:

  root@number:~# perf trace -e connect ssh localhost
       0.000 ( 0.008 ms): ssh/762249 connect(fd: 3, uservaddr: {2,}, addrlen: 16)                          = 0
       0.014 ( 0.005 ms): ssh/762249 connect(fd: 3, uservaddr: {10,}, addrlen: 28)                         = 0
       0.030 ( 0.032 ms): ssh/762249 connect(fd: 3, uservaddr: {10,}, addrlen: 28)                         = 0
  root@localhost's password:     63.031 ( 0.035 ms): ssh/762249 connect(fd: 4, uservaddr: {1,{['/','v','a','r','/','r','u','n','/','.','h','e','i','m',],},}, addrlen: 110) = 0
      64.037 ( 0.024 ms): ssh/762249 connect(fd: 4, uservaddr: {1,{['/','v','a','r','/','r','u','n','/','.','h','e','i','m',],},}, addrlen: 110) = 0
 
  root@number:~# strace -e connect ssh localhost
  connect(3, {sa_family=AF_INET, sin_port=htons(22), sin_addr=inet_addr("127.0.0.1")}, 16) = 0
  connect(3, {sa_family=AF_INET6, sin6_port=htons(22), sin6_flowinfo=htonl(0), inet_pton(AF_INET6, "::1", &sin6_addr), sin6_scope_id=0}, 28) = 0
  connect(3, {sa_family=AF_INET6, sin6_port=htons(22), sin6_flowinfo=htonl(0), inet_pton(AF_INET6, "::1", &sin6_addr), sin6_scope_id=0}, 28) = 0
  connect(4, {sa_family=AF_UNIX, sun_path="/var/run/.heim_org.h5l.kcm-socket"}, 110) = 0
  connect(4, {sa_family=AF_UNIX, sun_path="/var/run/.heim_org.h5l.kcm-socket"}, 110) = 0
  root@localhost's password:

Which while getting the struct contents produces an end result that 
conveys less info than the specilized connect BPF program we have, so we
need to only use this generic BTF dumper approach when we _don't_ have
an specialized one, at least at this point.

In the future we really should get the BTF dumper to use the more
specialized pretty printers that knows about how to pretty print network
specific addresses based on the network family, etc.