Re: [RFC PATCH glibc 09/13] glibc: Perform rseq(2) registration at C startup and thread creation (v13)

* Mathieu Desnoyers:

Register rseq(2) TLS for each thread (including main), and unregister
for each thread (excluding main). "rseq" stands for Restartable
Sequences.

See the rseq(2) man page proposed here:
  https://lkml.org/lkml/2018/9/19/647

This patch is based on glibc-2.30. The rseq(2) system call was merged
into Linux 4.18.

This patch needs to be updated for some be/le abilist splits.
Big-endian ABI lists for arm, microblaze, sh, appear to be missing.
This is something that can be checked with build-many-glibcs.py.

   @Theglibc{}                  @code{__rseq_abi}

+Restartable Sequences system call (Linux-specific). The layout of this
+structure is defined by the Linux kernel rseq.h UAPI. Registration of each
+thread's __rseq_abi is performed by @glibcadj{} at libc initialization and

                                       @theglibc{}
           @code{__rseq_abi}

+pthread creation.

+Each supported architecture provide a RSEQ_SIG signature in sys/rseq.h. That

                                         @code{RSEQ_SIG}       @file{sys/rseq.h} 

+signature is expected to be present in the code before each Restartable
+Sequences abort handler. Failure to provide the expected signature may
+terminate the process with a Segmentation fault.

Two spaces at the end of setences, please.

The manual should use @deftypevar to create an index entry etc. for
__rseq_abi.  See argp_program_version for an example of how to do this.

I think current policy is to have documentation for at least the minimum
functionality in the manual.  I understand that it makes it a lot of
work to write patches which add system call wrappers.

Thanks,
Florian

----- On Jan 7, 2020, at 7:23 AM, Florian Weimer fweimer@redhat.com wrote:

* Mathieu Desnoyers:

quoted

Register rseq(2) TLS for each thread (including main), and unregister
for each thread (excluding main). "rseq" stands for Restartable
Sequences.

See the rseq(2) man page proposed here:
  https://lkml.org/lkml/2018/9/19/647

This patch is based on glibc-2.30. The rseq(2) system call was merged
into Linux 4.18.

This patch needs to be updated for some be/le abilist splits.
Big-endian ABI lists for arm, microblaze, sh, appear to be missing.

done.

This is something that can be checked with build-many-glibcs.py.

OK.

quoted

diff --git a/manual/threads.texi b/manual/threads.texi
index 0858ef8f92..059f781120 100644
--- a/manual/threads.texi
+++ b/manual/threads.texi

@@ -881,3 +881,20 @@ Behaves like @code{pthread_timedjoin_np} except that the

absolute time in
 @c pthread_spin_unlock
 @c pthread_testcancel
 @c pthread_yield
+
+@node Restartable Sequences
+@section Restartable Sequences
+@cindex rseq
+
+This section describes the @glibcadj{} Restartable Sequences integration.
+
+The @glibcadj{} implements a __rseq_abi TLS symbol to interact with the

  @Theglibc{}                  @code{__rseq_abi}

quoted

+Restartable Sequences system call (Linux-specific). The layout of this
+structure is defined by the Linux kernel rseq.h UAPI. Registration of each
+thread's __rseq_abi is performed by @glibcadj{} at libc initialization and

                                      @theglibc{}
          @code{__rseq_abi}

quoted

+pthread creation.

quoted

+Each supported architecture provide a RSEQ_SIG signature in sys/rseq.h. That

                                        @code{RSEQ_SIG}       @file{sys/rseq.h}

quoted

+signature is expected to be present in the code before each Restartable
+Sequences abort handler. Failure to provide the expected signature may
+terminate the process with a Segmentation fault.

Two spaces at the end of setences, please.

All above done.

The manual should use @deftypevar to create an index entry etc. for
__rseq_abi.  See argp_program_version for an example of how to do this.

OK. I've added __rseq_abi as a variable and RSEQ_SIG as a macro.

I think current policy is to have documentation for at least the minimum
functionality in the manual.  I understand that it makes it a lot of
work to write patches which add system call wrappers.

Makes sense.

Thanks!

Mathieu

Thanks,
Florian

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com