GNU bug report logs - #63333
[PATCH] Add hurd-amd64 support

Previous Next

To add a comment to this bug, you must first unarchive it, by sending
a message to control AT debbugs.gnu.org, with unarchive 63333 in the body.
You can then email your comments to 63333 AT debbugs.gnu.org in the normal way.

Toggle the display of automated, internal messages from the tracker.

Message #5 received at submit <at> debbugs.gnu.org (full text, mbox):

From: Samuel Thibault <samuel.thibault <at> ens-lyon.org>
To: bug-diffutils <at> gnu.org, bug-grep <at> gnu.org
Cc: bug-hurd <at> gnu.org
Subject: [PATCH] Add hurd-amd64 support
Date: Sat, 6 May 2023 16:22:37 +0200

This adds SIGSEGV_FAULT_STACKPOINTER for the hurd-amd64 case

--- ./lib/sigsegv.c.original	2023-05-05 10:45:54.673751100 +0000
+++ ./lib/sigsegv.c	2023-05-05 10:48:47.903577554 +0000
@@ -351,6 +351,17 @@
    "old esp, if trapped from user".  */
 #  define SIGSEGV_FAULT_STACKPOINTER  scp->sc_uesp
 
+# elif defined __x86_64__
+
+/* scp points to a 'struct sigcontext' (defined in
+   glibc/sysdeps/mach/hurd/x86_64/bits/sigcontext.h).
+   The registers of this struct get pushed on the stack through
+   gnumach/x86_64/i386/locore.S:trapall.  */
+/* Both sc_rsp and sc_ursp appear to have the same value.
+   It appears more reliable to use sc_ursp because it is labelled as
+   "old rsp, if trapped from user".  */
+#  define SIGSEGV_FAULT_STACKPOINTER  scp->sc_ursp
+
 # endif
 
 #endif

Message #8 received at submit <at> debbugs.gnu.org (full text, mbox):

From: Jim Meyering <jim <at> meyering.net>
To: bug-diffutils <at> gnu.org, bug-grep <at> gnu.org, bug-hurd <at> gnu.org
Cc: "bug-gnulib <at> gnu.org List" <bug-gnulib <at> gnu.org>, 63333 <at> debbugs.gnu.org
Subject: Re: [bug-diffutils] bug#63333: [PATCH] Add hurd-amd64 support
Date: Mon, 8 May 2023 12:30:49 +0200

On Sat, May 6, 2023 at 8:34 PM Samuel Thibault
<samuel.thibault <at> ens-lyon.org> wrote:
> This adds SIGSEGV_FAULT_STACKPOINTER for the hurd-amd64 case
>
> --- ./lib/sigsegv.c.original    2023-05-05 10:45:54.673751100 +0000
> +++ ./lib/sigsegv.c     2023-05-05 10:48:47.903577554 +0000
> @@ -351,6 +351,17 @@
>     "old esp, if trapped from user".  */
>  #  define SIGSEGV_FAULT_STACKPOINTER  scp->sc_uesp
>
> +# elif defined __x86_64__
> +
> +/* scp points to a 'struct sigcontext' (defined in
> +   glibc/sysdeps/mach/hurd/x86_64/bits/sigcontext.h).
> +   The registers of this struct get pushed on the stack through
> +   gnumach/x86_64/i386/locore.S:trapall.  */
> +/* Both sc_rsp and sc_ursp appear to have the same value.
> +   It appears more reliable to use sc_ursp because it is labelled as
> +   "old rsp, if trapped from user".  */
> +#  define SIGSEGV_FAULT_STACKPOINTER  scp->sc_ursp
> +
>  # endif
>
>  #endif

Hi Samuel, thanks. This belongs in gnulib, so cc'ing that list.

Message #14 received at submit <at> debbugs.gnu.org (full text, mbox):

From: Bruno Haible <bruno <at> clisp.org>
To: bug-diffutils <at> gnu.org, bug-grep <at> gnu.org, bug-hurd <at> gnu.org,
 bug-gnulib <at> gnu.org
Cc: bug-gnulib <at> gnu.org, 63333 <at> debbugs.gnu.org, Jim Meyering <jim <at> meyering.net>
Subject: Re: [bug-diffutils] bug#63333: [PATCH] Add hurd-amd64 support
Date: Fri, 12 May 2023 21:38:29 +0200

Samuel Thibault <samuel.thibault <at> ens-lyon.org> wrote:
> > This adds SIGSEGV_FAULT_STACKPOINTER for the hurd-amd64 case
> >
> > --- ./lib/sigsegv.c.original    2023-05-05 10:45:54.673751100 +0000
> > +++ ./lib/sigsegv.c     2023-05-05 10:48:47.903577554 +0000
> > @@ -351,6 +351,17 @@
> >     "old esp, if trapped from user".  */
> >  #  define SIGSEGV_FAULT_STACKPOINTER  scp->sc_uesp
> >
> > +# elif defined __x86_64__
> > +
> > +/* scp points to a 'struct sigcontext' (defined in
> > +   glibc/sysdeps/mach/hurd/x86_64/bits/sigcontext.h).
> > +   The registers of this struct get pushed on the stack through
> > +   gnumach/x86_64/i386/locore.S:trapall.  */
> > +/* Both sc_rsp and sc_ursp appear to have the same value.
> > +   It appears more reliable to use sc_ursp because it is labelled as
> > +   "old rsp, if trapped from user".  */
> > +#  define SIGSEGV_FAULT_STACKPOINTER  scp->sc_ursp
> > +
> >  # endif
> >
> >  #endif

Thanks Samuel. The definition of SIGSEGV_FAULT_STACKPOINTER seems correct in
this patch, but the comments are not. I am applying this instead.

I'm calling this "tentative" support for Hurd/x86_64, because
  - I have no way to test it,
  - some of the code paths involved seem to be incorrect so far, see
    https://lists.gnu.org/archive/html/bug-hurd/2023-05/msg00207.html


2023-05-12  Bruno Haible  <bruno <at> clisp.org>

	sigsegv: Add tentative support for Hurd/x86_64.
	Reported by Samuel Thibault <samuel.thibault <at> ens-lyon.org>.
	* lib/sigsegv.c: Update from libsigsegv/src/fault-hurd-i386.h.

diff --git a/lib/sigsegv.c b/lib/sigsegv.c
index 5e943e4d5d..aadba4e060 100644
--- a/lib/sigsegv.c
+++ b/lib/sigsegv.c
@@ -365,12 +365,26 @@ int libsigsegv_version = LIBSIGSEGV_VERSION;
 # define SIGSEGV_FAULT_ADDRESS  (unsigned long) code
 # define SIGSEGV_FAULT_CONTEXT  scp
 
-# if defined __i386__
+# if defined __x86_64__
+/* 64 bit registers */
+
+/* scp points to a 'struct sigcontext' (defined in
+   glibc/sysdeps/mach/hurd/x86_64/bits/sigcontext.h).
+   The registers, at the moment the signal occurred, get pushed on the stack
+   through gnumach/x86_64/locore.S:alltraps and then copied into the struct
+   through glibc/sysdeps/mach/hurd/x86/trampoline.c.  */
+/* sc_rsp is unused (not set by gnumach/x86_64/locore.S:alltraps).  We need
+   to use sc_ursp.  */
+#  define SIGSEGV_FAULT_STACKPOINTER  scp->sc_ursp
+
+# elif defined __i386__
+/* 32 bit registers */
 
 /* scp points to a 'struct sigcontext' (defined in
    glibc/sysdeps/mach/hurd/i386/bits/sigcontext.h).
-   The registers of this struct get pushed on the stack through
-   gnumach/i386/i386/locore.S:trapall.  */
+   The registers, at the moment the signal occurred, get pushed on the stack
+   through gnumach/i386/i386/locore.S:alltraps and then copied into the struct
+   through glibc/sysdeps/mach/hurd/x86/trampoline.c.  */
 /* Both sc_esp and sc_uesp appear to have the same value.
    It appears more reliable to use sc_uesp because it is labelled as
    "old esp, if trapped from user".  */

Message #20 received at submit <at> debbugs.gnu.org (full text, mbox):

From: Bruno Haible <bruno <at> clisp.org>
To: bug-diffutils <at> gnu.org, bug-grep <at> gnu.org, bug-hurd <at> gnu.org,
 bug-gnulib <at> gnu.org
Cc: Sergey Bugaev <bugaevc <at> gmail.com>, bug-gnulib <at> gnu.org,
 63333 <at> debbugs.gnu.org, Jim Meyering <jim <at> meyering.net>
Subject: Re: [bug-diffutils] bug#63333: [PATCH] Add hurd-amd64 support
Date: Mon, 15 May 2023 19:02:13 +0200

I committed:
> 2023-05-12  Bruno Haible  <bruno <at> clisp.org>
> 
> 	sigsegv: Add tentative support for Hurd/x86_64.
> 	Reported by Samuel Thibault <samuel.thibault <at> ens-lyon.org>.
> 	* lib/sigsegv.c: Update from libsigsegv/src/fault-hurd-i386.h.

This was not sufficient. Sergey Bugaev pointed out that
  - my comments were wrong,
  - the handler's parameter list needs to include 'long code', not 'int code',
    for x86_64.

This patch fixes it, in sync with libsigsegv.


2023-05-15  Bruno Haible  <bruno <at> clisp.org>

	sigsegv: Add tentative support for Hurd/x86_64.
	Based on explanations by Sergey Bugaev <bugaevc <at> gmail.com>.
	* lib/sigsegv.c: Update from libsigsegv/src/fault-hurd-i386-old.h.

diff --git a/lib/sigsegv.c b/lib/sigsegv.c
index aadba4e060..8263d9b7bd 100644
--- a/lib/sigsegv.c
+++ b/lib/sigsegv.c
@@ -361,7 +361,7 @@ int libsigsegv_version = LIBSIGSEGV_VERSION;
 
 #if defined __GNU__ /* Hurd */
 
-# define SIGSEGV_FAULT_HANDLER_ARGLIST  int sig, int code, struct sigcontext *scp
+# define SIGSEGV_FAULT_HANDLER_ARGLIST  int sig, long code, struct sigcontext *scp
 # define SIGSEGV_FAULT_ADDRESS  (unsigned long) code
 # define SIGSEGV_FAULT_CONTEXT  scp
 
@@ -370,11 +370,29 @@ int libsigsegv_version = LIBSIGSEGV_VERSION;
 
 /* scp points to a 'struct sigcontext' (defined in
    glibc/sysdeps/mach/hurd/x86_64/bits/sigcontext.h).
-   The registers, at the moment the signal occurred, get pushed on the stack
-   through gnumach/x86_64/locore.S:alltraps and then copied into the struct
-   through glibc/sysdeps/mach/hurd/x86/trampoline.c.  */
-/* sc_rsp is unused (not set by gnumach/x86_64/locore.S:alltraps).  We need
-   to use sc_ursp.  */
+   The registers, at the moment the signal occurred, get pushed on the kernel
+   stack through gnumach/x86_64/locore.S:alltraps. They are denoted by a
+   'struct i386_saved_state' (defined in gnumach/i386/i386/thread.h).
+   Upon invocation of the Mach interface function thread_get_state
+   <https://www.gnu.org/software/hurd/gnumach-doc/Thread-Execution.html>
+   (= __thread_get_state in glibc), defined in gnumach/kern/thread.c,
+   the function thread_getstatus, defined in gnumach/i386/i386/pcb.c, copies the
+   register values in a different arrangement into a 'struct i386_thread_state',
+   defined in gnumach/i386/include/mach/i386/thread_status.h. (Different
+   arrangement: trapno, err get dropped; different order of r8...r15; also rsp
+   gets set to 0.)
+   This 'struct i386_thread_state' is actually the 'basic' part of a
+   'struct machine_thread_all_state', defined in
+   glibc/sysdeps/mach/x86/thread_state.h.
+   From there, the function _hurd_setup_sighandler, defined in
+   glibc/sysdeps/mach/hurd/x86/trampoline.c,
+   1. sets rsp to the same value as ursp,
+   2. copies the 'struct i386_thread_state' into the appropriate part of a
+      'struct sigcontext', defined in
+      glibc/sysdeps/mach/hurd/x86_64/bits/sigcontext.h.  */
+/* Both sc_rsp and sc_ursp have the same value.
+   It appears more reliable to use sc_ursp because sc_rsp is marked as
+   "not used".  */
 #  define SIGSEGV_FAULT_STACKPOINTER  scp->sc_ursp
 
 # elif defined __i386__
@@ -382,12 +400,28 @@ int libsigsegv_version = LIBSIGSEGV_VERSION;
 
 /* scp points to a 'struct sigcontext' (defined in
    glibc/sysdeps/mach/hurd/i386/bits/sigcontext.h).
-   The registers, at the moment the signal occurred, get pushed on the stack
-   through gnumach/i386/i386/locore.S:alltraps and then copied into the struct
-   through glibc/sysdeps/mach/hurd/x86/trampoline.c.  */
-/* Both sc_esp and sc_uesp appear to have the same value.
-   It appears more reliable to use sc_uesp because it is labelled as
-   "old esp, if trapped from user".  */
+   The registers, at the moment the signal occurred, get pushed on the kernel
+   stack through gnumach/i386/i386/locore.S:alltraps. They are denoted by a
+   'struct i386_saved_state' (defined in gnumach/i386/i386/thread.h).
+   Upon invocation of the Mach interface function thread_get_state
+   <https://www.gnu.org/software/hurd/gnumach-doc/Thread-Execution.html>
+   (= __thread_get_state in glibc), defined in gnumach/kern/thread.c,
+   the function thread_getstatus, defined in gnumach/i386/i386/pcb.c, copies the
+   register values in a different arrangement into a 'struct i386_thread_state',
+   defined in gnumach/i386/include/mach/i386/thread_status.h. (Different
+   arrangement: trapno, err get dropped; also esp gets set to 0.)
+   This 'struct i386_thread_state' is actually the 'basic' part of a
+   'struct machine_thread_all_state', defined in
+   glibc/sysdeps/mach/x86/thread_state.h.
+   From there, the function _hurd_setup_sighandler, defined in
+   glibc/sysdeps/mach/hurd/x86/trampoline.c,
+   1. sets esp to the same value as uesp,
+   2. copies the 'struct i386_thread_state' into the appropriate part of a
+      'struct sigcontext', defined in
+      glibc/sysdeps/mach/hurd/i386/bits/sigcontext.h.  */
+/* Both sc_esp and sc_uesp have the same value.
+   It appears more reliable to use sc_uesp because sc_esp is marked as
+   "not used".  */
 #  define SIGSEGV_FAULT_STACKPOINTER  scp->sc_uesp
 
 # endif

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.