public inbox for gdb-patches@sourceware.org
 help / color / mirror / Atom feed
From: Shahab Vahedi <shahab.vahedi@gmail.com>
To: gdb-patches@sourceware.org
Cc: Shahab Vahedi <shahab.vahedi@gmail.com>,
	Shahab Vahedi <shahab@synopsys.com>, Tom Tromey <tom@tromey.com>,
	Francois Bedard <fbedard@synopsys.com>,
	Anton Kolesov <Anton.Kolesov@synopsys.com>
Subject: [PUSHED v2 2/2] arc: Add support for signal frames for Linux targets
Date: Tue, 22 Dec 2020 12:13:40 +0100	[thread overview]
Message-ID: <20201222111340.10618-2-shahab.vahedi@gmail.com> (raw)
In-Reply-To: <20201222111340.10618-1-shahab.vahedi@gmail.com>

From: Anton Kolesov <Anton.Kolesov@synopsys.com>

Implement functions needed to unwind signal frames on ARC Linux targets.

gdb/ChangeLog

	* arc-linux-tdep.c (arc_linux_sc_reg_offsets): New static variable.
	(arc_linux_is_sigtramp): New function.
	(arc_linux_sigcontext_addr): Likewise.
	(arc_linux_init_osabi): Use them.
---
 gdb/arc-linux-tdep.c | 181 +++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 181 insertions(+)

diff --git a/gdb/arc-linux-tdep.c b/gdb/arc-linux-tdep.c
index 6e74bae8056..ad68f494b3c 100644
--- a/gdb/arc-linux-tdep.c
+++ b/gdb/arc-linux-tdep.c
@@ -33,6 +33,60 @@
 
 #define REGOFF(offset) (offset * ARC_REGISTER_SIZE)
 
+/* arc_linux_sc_reg_offsets[i] is the offset of register i in the `struct
+   sigcontext'.  Array index is an internal GDB register number, as defined in
+   arc-tdep.h:arc_regnum.
+
+   From <include/uapi/asm/sigcontext.h> and <include/uapi/asm/ptrace.h>.
+
+   The layout of this struct is tightly bound to "arc_regnum" enum
+   in arc-tdep.h.  Any change of order in there, must be reflected
+   here as well.  */
+static const int arc_linux_sc_reg_offsets[] = {
+  /* R0 - R12.  */
+  REGOFF (22), REGOFF (21), REGOFF (20), REGOFF (19),
+  REGOFF (18), REGOFF (17), REGOFF (16), REGOFF (15),
+  REGOFF (14), REGOFF (13), REGOFF (12), REGOFF (11),
+  REGOFF (10),
+
+  /* R13 - R25.  */
+  ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER,
+  ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER,
+  ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER,
+  ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER,
+  ARC_OFFSET_NO_REGISTER,
+
+  REGOFF (9),			/* R26 (GP) */
+  REGOFF (8),			/* FP */
+  REGOFF (23),			/* SP */
+  ARC_OFFSET_NO_REGISTER,	/* ILINK */
+  ARC_OFFSET_NO_REGISTER,	/* R30 */
+  REGOFF (7),			/* BLINK */
+
+  /* R32 - R59.  */
+  ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER,
+  ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER,
+  ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER,
+  ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER,
+  ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER,
+  ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER,
+  ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER,
+  ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER,
+  ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER, ARC_OFFSET_NO_REGISTER,
+  ARC_OFFSET_NO_REGISTER,
+
+  REGOFF (4),			/* LP_COUNT */
+  ARC_OFFSET_NO_REGISTER,	/* RESERVED */
+  ARC_OFFSET_NO_REGISTER,	/* LIMM */
+  ARC_OFFSET_NO_REGISTER,	/* PCL */
+
+  REGOFF (6),			/* PC  */
+  REGOFF (5),			/* STATUS32 */
+  REGOFF (2),			/* LP_START */
+  REGOFF (3),			/* LP_END */
+  REGOFF (1),			/* BTA */
+};
+
 /* arc_linux_core_reg_offsets[i] is the offset in the .reg section of GDB
    regnum i.  Array index is an internal GDB register number, as defined in
    arc-tdep.h:arc_regnum.
@@ -87,6 +141,127 @@ static const int arc_linux_core_reg_offsets[] = {
   REGOFF (6)			/* ERET */
 };
 
+/* Is THIS_FRAME a sigtramp function - the function that returns from
+   signal handler into normal execution flow? This is the case if the PC is
+   either at the start of, or in the middle of the two instructions:
+
+     mov r8, __NR_rt_sigreturn ; __NR_rt_sigreturn == 139
+     trap_s 0 ; `swi' for ARC700
+
+   On ARC uClibc Linux this function is called __default_rt_sa_restorer.
+
+   Returns TRUE if this is a sigtramp frame.  */
+
+static bool
+arc_linux_is_sigtramp (struct frame_info *this_frame)
+{
+  struct gdbarch *gdbarch = get_frame_arch (this_frame);
+  CORE_ADDR pc = get_frame_pc (this_frame);
+
+  if (arc_debug)
+    {
+      debug_printf ("arc-linux: arc_linux_is_sigtramp, pc=%s\n",
+		    paddress(gdbarch, pc));
+    }
+
+  static const gdb_byte insns_be_hs[] = {
+    0x20, 0x8a, 0x12, 0xc2,	/* mov  r8,nr_rt_sigreturn */
+    0x78, 0x1e			/* trap_s 0 */
+  };
+  static const gdb_byte insns_be_700[] = {
+    0x20, 0x8a, 0x12, 0xc2,	/* mov  r8,nr_rt_sigreturn */
+    0x22, 0x6f, 0x00, 0x3f	/* swi */
+  };
+
+  gdb_byte arc_sigtramp_insns[sizeof (insns_be_700)];
+  size_t insns_sz;
+  if (arc_mach_is_arcv2 (gdbarch))
+    {
+      insns_sz = sizeof (insns_be_hs);
+      memcpy (arc_sigtramp_insns, insns_be_hs, insns_sz);
+    }
+  else
+    {
+      insns_sz = sizeof (insns_be_700);
+      memcpy (arc_sigtramp_insns, insns_be_700, insns_sz);
+    }
+  if (gdbarch_byte_order (gdbarch) == BFD_ENDIAN_LITTLE)
+    {
+      /* On little endian targets, ARC code section is in what is called
+	 "middle endian", where half-words are in the big-endian order,
+	 only bytes inside the halfwords are in the little endian order.
+	 As a result it is very easy to convert big endian instruction to
+	 little endian, since it is needed to swap bytes in the halfwords,
+	 so there is no need to have information on whether that is a
+	 4-byte instruction or 2-byte.  */
+      gdb_assert ((insns_sz % 2) == 0);
+      for (int i = 0; i < insns_sz; i += 2)
+	std::swap (arc_sigtramp_insns[i], arc_sigtramp_insns[i+1]);
+    }
+
+  gdb_byte buf[insns_sz];
+
+  /* Read the memory at the PC.  Since we are stopped, any breakpoint must
+     have been removed.  */
+  if (!safe_frame_unwind_memory (this_frame, pc, buf, insns_sz))
+    {
+      /* Failed to unwind frame.  */
+      return FALSE;
+    }
+
+  /* Is that code the sigtramp instruction sequence?  */
+  if (memcmp (buf, arc_sigtramp_insns, insns_sz) == 0)
+    return TRUE;
+
+  /* No - look one instruction earlier in the code...  */
+  if (!safe_frame_unwind_memory (this_frame, pc - 4, buf, insns_sz))
+    {
+      /* Failed to unwind frame.  */
+      return FALSE;
+    }
+
+  return (memcmp (buf, arc_sigtramp_insns, insns_sz) == 0);
+}
+
+/* Get sigcontext structure of sigtramp frame - it contains saved
+   registers of interrupted frame.
+
+   Stack pointer points to the rt_sigframe structure, and sigcontext can
+   be found as in:
+
+   struct rt_sigframe {
+     struct siginfo info;
+     struct ucontext uc;
+     ...
+   };
+
+   struct ucontext {
+     unsigned long uc_flags;
+     struct ucontext *uc_link;
+     stack_t uc_stack;
+     struct sigcontext uc_mcontext;
+     sigset_t uc_sigmask;
+   };
+
+   sizeof (struct siginfo) == 0x80
+   offsetof (struct ucontext, uc_mcontext) == 0x14
+
+   GDB cannot include linux headers and use offsetof () because those are
+   target headers and GDB might be built for a different run host.  There
+   doesn't seem to be an established mechanism to figure out those offsets
+   via gdbserver, so the only way is to hardcode values in the GDB,
+   meaning that GDB will be broken if values will change.  That seems to
+   be a very unlikely scenario and other arches (aarch64, alpha, amd64,
+   etc) in GDB hardcode values.  */
+
+static CORE_ADDR
+arc_linux_sigcontext_addr (struct frame_info *this_frame)
+{
+  const int ucontext_offset = 0x80;
+  const int sigcontext_offset = 0x14;
+  return get_frame_sp (this_frame) + ucontext_offset + sigcontext_offset;
+}
+
 /* Implement the "cannot_fetch_register" gdbarch method.  */
 
 static int
@@ -429,6 +604,12 @@ arc_linux_init_osabi (struct gdbarch_info info, struct gdbarch *gdbarch)
   if (arc_debug)
     debug_printf ("arc-linux: GNU/Linux OS/ABI initialization.\n");
 
+  /* Fill in target-dependent info in ARC-private structure.  */
+  tdep->is_sigtramp = arc_linux_is_sigtramp;
+  tdep->sigcontext_addr = arc_linux_sigcontext_addr;
+  tdep->sc_reg_offset = arc_linux_sc_reg_offsets;
+  tdep->sc_num_regs = ARRAY_SIZE (arc_linux_sc_reg_offsets);
+
   /* If we are using Linux, we have in uClibc
      (libc/sysdeps/linux/arc/bits/setjmp.h):
 
-- 
2.29.2


      reply	other threads:[~2020-12-22 11:13 UTC|newest]

Thread overview: 12+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-10-22 16:51 [PATCH 0/2] arc: Add signal handling support Shahab Vahedi
2020-10-22 16:51 ` [PATCH 1/2] arc: Add support for signal handlers Shahab Vahedi
2020-11-12 17:47   ` Tom Tromey
2020-10-22 16:51 ` [PATCH 2/2] arc: Add support for signal frames for Linux targets Shahab Vahedi
2020-11-12 17:51   ` Tom Tromey
2020-11-10 19:12 ` [PING^1][PATCH 0/2] arc: Add signal handling support Shahab Vahedi
2020-12-07 16:36 ` [PATCH v2 " Shahab Vahedi
2020-12-07 16:36   ` [PATCH v2 1/2] arc: Add support for signal handlers Shahab Vahedi
2020-12-07 16:36   ` [PATCH v2 2/2] arc: Add support for signal frames for Linux targets Shahab Vahedi
2020-12-10 19:07   ` [PATCH v2 0/2] arc: Add signal handling support Tom Tromey
2020-12-22 11:13 ` [PUSHED v2 1/2] arc: Add support for signal handlers Shahab Vahedi
2020-12-22 11:13   ` Shahab Vahedi [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20201222111340.10618-2-shahab.vahedi@gmail.com \
    --to=shahab.vahedi@gmail.com \
    --cc=Anton.Kolesov@synopsys.com \
    --cc=fbedard@synopsys.com \
    --cc=gdb-patches@sourceware.org \
    --cc=shahab@synopsys.com \
    --cc=tom@tromey.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).