Toolchain/binutils-gdb
1
0
Fork 0
forked from Imagelibrary/binutils-gdb
Code Pull Requests Activity
Files
ce35d7163e779b1321058b22f005c70ce1524b25
binutils-gdb/gdb/arc-linux-tdep.c
Shahab Vahedi d8979e81fd gdb: Do not interrupt atomic sequences for ARC 
When stepping over thread-lock related codes (in uClibc), the inferior process
gets stuck and never manages to enter the critical section:

------8<-------
 1 size_t fwrite(const void * __restrict ptr, size_t size,
 2               size_t nmemb, register FILE * __restrict stream)
 3 {
 4     size_t retval;
 5     __STDIO_AUTO_THREADLOCK_VAR;
 6
 7 >   __STDIO_AUTO_THREADLOCK(stream);
 8
 9     retval = fwrite_unlocked(ptr, size, nmemb, stream);
10
11     __STDIO_AUTO_THREADUNLOCK(stream);
12
13     return retval;
14 }
------>8-------

Here, we are at line 7.  Using the "next" command leads no where.
However, setting a breakpoint on line 9 and issuing "continue" works.

Looking at the assembly instructions reveals that we're dealing with the
critical section entry code [1] that should never be interrupted, in this
case by the debugger's implicit breakpoints:

------8<-------
  ...
1 add_s   r0,r13,0x38
2 mov_s   r3,1
3 llock   r2,[r0]        <-.
4 brne.nt r2,0,14     --.  |
5 scond   r3,[r0]       |  |
6 bne     -10         --|--'
7 brne_s  r2,0,84     <-'
  ...
------>8-------

Lines 3 until 5 (inclusive) are supposed to be executed atomically.
Therefore, GDB should never (implicitly) insert a breakpoint on lines
4 and 5, else the program will try to acquire the lock again by jumping
back to line 3 and gets stuck in an infinite loop.

The solution is to make GDB aware of these patterns so it inserts
breakpoints after the sequence -- line 6 in this example.

[1]
https://cgit.uclibc-ng.org/cgi/cgit/uclibc-ng.git/tree/libc/sysdeps/linux/arc/bits/atomic.h#n46
------8<-------
  ({									\
	__typeof(oldval) prev;						\
									\
	__asm__ __volatile__(						\
	"1:	llock   %0, [%1]	\n"				\
	"	brne    %0, %2, 2f	\n"				\
	"	scond   %3, [%1]	\n"				\
	"	bnz     1b		\n"				\
	"2:				\n"				\
	: "=&r"(prev)							\
	: "r"(mem), "ir"(oldval),					\
	  "r"(newval) /* can't be "ir". scond can't take limm for "b" */\
	: "cc", "memory");						\
									\
	prev;								\
  })
------>8-------
"llock" (Load Locked) loads the 32-bit word pointed by the source
operand.  If the load is completed without any interruption or
exception, the physical address is remembered, in Lock Physical Address
(LPA), and the Lock Flag (LF) is set to 1.  LF is a non-architecturally
visible flag and is cleared whenever an interrupt or exception takes
place.  LF is also cleared (atomically) whenever another process writes
to the LPA.

"scond" (Store Conditional) will write to the destination address if
and only if the LF is set to 1.  When finished, with or without a write,
it atomically copies the LF value to ZF (Zero Flag).

These two instructions together provide the mechanism for entering a
critical section.  The code snippet above comes from uClibc:
-----------------------

v3 (after Tom's remarks[2]):
 handle_atomic_sequence()
  - no need to initialize the std::vector with "{}"
  - fix typo in comments: "conditial" -> "conditional"
  - add braces to the body of "if" condition because of the comment line
 arc_linux_software_single_step()
  - make the performance slightly more efficient by moving a few
    variables after the likely "return" point.

v2 (after Simon's remarks[3]):
- handle_atomic_sequence() gets a copy of an instruction instead of
  a reference.
- handle_atomic_sequence() asserts if the given instruction is an llock.

[2]
https://sourceware.org/pipermail/gdb-patches/2021-February/175805.html

[3]
https://sourceware.org/pipermail/gdb-patches/2021-January/175487.html

gdb/ChangeLog:

	PR tdep/27369
	* arc-linux-tdep.c (handle_atomic_sequence): New.
	(arc_linux_software_single_step): Call handle_atomic_sequence().
2021-02-08 13:11:44 +01:00

549 lines
17 KiB
C
Raw Blame History

/* Target dependent code for GNU/Linux ARC.
Copyright 2020-2021 Free Software Foundation, Inc.
This file is part of GDB.
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>. */
/* GDB header files. */
#include "defs.h"
#include "linux-tdep.h"
#include "objfiles.h"
#include "opcode/arc.h"
#include "osabi.h"
#include "solib-svr4.h"
/* ARC header files. */
#include "opcodes/arc-dis.h"
#include "arc-linux-tdep.h"
#include "arc-tdep.h"
#include "arch/arc.h"
#define REGOFF(offset) (offset * ARC_REGISTER_SIZE)
/* 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.
From include/uapi/asm/ptrace.h in the ARC Linux sources. */
/* 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_core_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. */
REGOFF (37), REGOFF (36), REGOFF (35), REGOFF (34),
REGOFF (33), REGOFF (32), REGOFF (31), REGOFF (30),
REGOFF (29), REGOFF (28), REGOFF (27), REGOFF (26),
REGOFF (25),
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 (39), /* PC */
REGOFF (5), /* STATUS32 */
REGOFF (2), /* LP_START */
REGOFF (3), /* LP_END */
REGOFF (1), /* BTA */
REGOFF (6) /* ERET */
};
/* Implement the "cannot_fetch_register" gdbarch method. */
static int
arc_linux_cannot_fetch_register (struct gdbarch *gdbarch, int regnum)
{
/* Assume that register is readable if it is unknown. */
switch (regnum)
{
case ARC_ILINK_REGNUM:
case ARC_RESERVED_REGNUM:
case ARC_LIMM_REGNUM:
return true;
case ARC_R30_REGNUM:
case ARC_R58_REGNUM:
case ARC_R59_REGNUM:
return !arc_mach_is_arcv2 (gdbarch);
}
return (regnum > ARC_BLINK_REGNUM) && (regnum < ARC_LP_COUNT_REGNUM);
}
/* Implement the "cannot_store_register" gdbarch method. */
static int
arc_linux_cannot_store_register (struct gdbarch *gdbarch, int regnum)
{
/* Assume that register is writable if it is unknown. */
switch (regnum)
{
case ARC_ILINK_REGNUM:
case ARC_RESERVED_REGNUM:
case ARC_LIMM_REGNUM:
case ARC_PCL_REGNUM:
return true;
case ARC_R30_REGNUM:
case ARC_R58_REGNUM:
case ARC_R59_REGNUM:
return !arc_mach_is_arcv2 (gdbarch);
}
return (regnum > ARC_BLINK_REGNUM) && (regnum < ARC_LP_COUNT_REGNUM);
}
/* For ARC Linux, breakpoints use the 16-bit TRAP_S 1 instruction, which
is 0x3e78 (little endian) or 0x783e (big endian). */
static const gdb_byte arc_linux_trap_s_be[] = { 0x78, 0x3e };
static const gdb_byte arc_linux_trap_s_le[] = { 0x3e, 0x78 };
static const int trap_size = 2; /* Number of bytes to insert "trap". */
/* Implement the "breakpoint_kind_from_pc" gdbarch method. */
static int
arc_linux_breakpoint_kind_from_pc (struct gdbarch *gdbarch, CORE_ADDR *pcptr)
{
return trap_size;
}
/* Implement the "sw_breakpoint_from_kind" gdbarch method. */
static const gdb_byte *
arc_linux_sw_breakpoint_from_kind (struct gdbarch *gdbarch,
int kind, int *size)
{
*size = kind;
return ((gdbarch_byte_order (gdbarch) == BFD_ENDIAN_BIG)
? arc_linux_trap_s_be
: arc_linux_trap_s_le);
}
/* Check for an atomic sequence of instructions beginning with an
LLOCK instruction and ending with a SCOND instruction.
These patterns are hand coded in libc's (glibc and uclibc). Take
a look at [1] for instance:
main+14: llock r2,[r0]
main+18: brne.nt r2,0,main+30
main+22: scond r3,[r0]
main+26: bne main+14
main+30: mov_s r0,0
If such a sequence is found, attempt to step over it.
A breakpoint is placed at the end of the sequence.
This function expects the INSN to be a "llock(d)" instruction.
[1]
https://cgit.uclibc-ng.org/cgi/cgit/uclibc-ng.git/tree/libc/ \
sysdeps/linux/arc/bits/atomic.h#n46
*/
static std::vector<CORE_ADDR>
handle_atomic_sequence (arc_instruction insn, disassemble_info &di)
{
const int atomic_seq_len = 24; /* Instruction sequence length. */
std::vector<CORE_ADDR> next_pcs;
/* Sanity check. */
gdb_assert (insn.insn_class == LLOCK);
/* Data size we are dealing with: LLOCK vs. LLOCKD */
arc_ldst_data_size llock_data_size_mode = insn.data_size_mode;
/* Indicator if any conditional branch is found in the sequence. */
bool found_bc = false;
/* Becomes true if "LLOCK(D) .. SCOND(D)" sequence is found. */
bool is_pattern_valid = false;
for (int insn_count = 0; insn_count < atomic_seq_len; ++insn_count)
{
arc_insn_decode (arc_insn_get_linear_next_pc (insn),
&di, arc_delayed_print_insn, &insn);
if (insn.insn_class == BRCC)
{
/* If more than one conditional branch is found, this is not
the pattern we are interested in. */
if (found_bc)
break;
found_bc = true;
continue;
}
/* This is almost a happy ending. */
if (insn.insn_class == SCOND)
{
/* SCOND should match the LLOCK's data size. */
if (insn.data_size_mode == llock_data_size_mode)
is_pattern_valid = true;
break;
}
}
if (is_pattern_valid)
{
/* Get next instruction after scond(d). There is no limm. */
next_pcs.push_back (insn.address + insn.length);
}
return next_pcs;
}
/* Implement the "software_single_step" gdbarch method. */
static std::vector<CORE_ADDR>
arc_linux_software_single_step (struct regcache *regcache)
{
struct gdbarch *gdbarch = regcache->arch ();
struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
struct disassemble_info di = arc_disassemble_info (gdbarch);
/* Read current instruction. */
struct arc_instruction curr_insn;
arc_insn_decode (regcache_read_pc (regcache), &di, arc_delayed_print_insn,
&curr_insn);
if (curr_insn.insn_class == LLOCK)
return handle_atomic_sequence (curr_insn, di);
CORE_ADDR next_pc = arc_insn_get_linear_next_pc (curr_insn);
std::vector<CORE_ADDR> next_pcs;
/* For instructions with delay slots, the fall thru is not the
instruction immediately after the current instruction, but the one
after that. */
if (curr_insn.has_delay_slot)
{
struct arc_instruction next_insn;
arc_insn_decode (next_pc, &di, arc_delayed_print_insn, &next_insn);
next_pcs.push_back (arc_insn_get_linear_next_pc (next_insn));
}
else
next_pcs.push_back (next_pc);
ULONGEST status32;
regcache_cooked_read_unsigned (regcache, gdbarch_ps_regnum (gdbarch),
&status32);
if (curr_insn.is_control_flow)
{
CORE_ADDR branch_pc = arc_insn_get_branch_target (curr_insn);
if (branch_pc != next_pc)
next_pcs.push_back (branch_pc);
}
/* Is current instruction the last in a loop body? */
else if (tdep->has_hw_loops)
{
/* If STATUS32.L is 1, then ZD-loops are disabled. */
if ((status32 & ARC_STATUS32_L_MASK) == 0)
{
ULONGEST lp_end, lp_start, lp_count;
regcache_cooked_read_unsigned (regcache, ARC_LP_START_REGNUM,
&lp_start);
regcache_cooked_read_unsigned (regcache, ARC_LP_END_REGNUM, &lp_end);
regcache_cooked_read_unsigned (regcache, ARC_LP_COUNT_REGNUM,
&lp_count);
if (arc_debug)
{
debug_printf ("arc-linux: lp_start = %s, lp_end = %s, "
"lp_count = %s, next_pc = %s\n",
paddress (gdbarch, lp_start),
paddress (gdbarch, lp_end),
pulongest (lp_count),
paddress (gdbarch, next_pc));
}
if (next_pc == lp_end && lp_count > 1)
{
/* The instruction is in effect a jump back to the start of
the loop. */
next_pcs.push_back (lp_start);
}
}
}
/* Is this a delay slot? Then next PC is in BTA register. */
if ((status32 & ARC_STATUS32_DE_MASK) != 0)
{
ULONGEST bta;
regcache_cooked_read_unsigned (regcache, ARC_BTA_REGNUM, &bta);
next_pcs.push_back (bta);
}
return next_pcs;
}
/* Implement the "skip_solib_resolver" gdbarch method.
See glibc_skip_solib_resolver for details. */
static CORE_ADDR
arc_linux_skip_solib_resolver (struct gdbarch *gdbarch, CORE_ADDR pc)
{
/* For uClibc 0.9.26+.
An unresolved PLT entry points to "__dl_linux_resolve", which calls
"_dl_linux_resolver" to do the resolving and then eventually jumps to
the function.
So we look for the symbol `_dl_linux_resolver', and if we are there,
gdb sets a breakpoint at the return address, and continues. */
struct bound_minimal_symbol resolver
= lookup_minimal_symbol ("_dl_linux_resolver", NULL, NULL);
if (arc_debug)
{
if (resolver.minsym != nullptr)
{
CORE_ADDR res_addr = BMSYMBOL_VALUE_ADDRESS (resolver);
debug_printf ("arc-linux: skip_solib_resolver (): "
"pc = %s, resolver at %s\n",
print_core_address (gdbarch, pc),
print_core_address (gdbarch, res_addr));
}
else
{
debug_printf ("arc-linux: skip_solib_resolver (): "
"pc = %s, no resolver found\n",
print_core_address (gdbarch, pc));
}
}
if (resolver.minsym != nullptr && BMSYMBOL_VALUE_ADDRESS (resolver) == pc)
{
/* Find the return address. */
return frame_unwind_caller_pc (get_current_frame ());
}
else
{
/* No breakpoint required. */
return 0;
}
}
void
arc_linux_supply_gregset (const struct regset *regset,
struct regcache *regcache,
int regnum, const void *gregs, size_t size)
{
gdb_static_assert (ARC_LAST_REGNUM
< ARRAY_SIZE (arc_linux_core_reg_offsets));
const bfd_byte *buf = (const bfd_byte *) gregs;
for (int reg = 0; reg <= ARC_LAST_REGNUM; reg++)
if (arc_linux_core_reg_offsets[reg] != ARC_OFFSET_NO_REGISTER)
regcache->raw_supply (reg, buf + arc_linux_core_reg_offsets[reg]);
}
void
arc_linux_supply_v2_regset (const struct regset *regset,
struct regcache *regcache, int regnum,
const void *v2_regs, size_t size)
{
const bfd_byte *buf = (const bfd_byte *) v2_regs;
/* user_regs_arcv2 is defined in linux arch/arc/include/uapi/asm/ptrace.h. */
regcache->raw_supply (ARC_R30_REGNUM, buf);
regcache->raw_supply (ARC_R58_REGNUM, buf + REGOFF (1));
regcache->raw_supply (ARC_R59_REGNUM, buf + REGOFF (2));
}
/* Populate BUF with register REGNUM from the REGCACHE. */
static void
collect_register (const struct regcache *regcache, struct gdbarch *gdbarch,
int regnum, gdb_byte *buf)
{
int offset;
/* Skip non-existing registers. */
if (arc_linux_core_reg_offsets[regnum] == ARC_OFFSET_NO_REGISTER)
return;
/* The address where the execution has stopped is in pseudo-register
STOP_PC. However, when kernel code is returning from the exception,
it uses the value from ERET register. Since, TRAP_S (the breakpoint
instruction) commits, the ERET points to the next instruction. In
other words: ERET != STOP_PC. To jump back from the kernel code to
the correct address, ERET must be overwritten by GDB's STOP_PC. Else,
the program will continue at the address after the current instruction.
*/
if (regnum == gdbarch_pc_regnum (gdbarch))
offset = arc_linux_core_reg_offsets[ARC_ERET_REGNUM];
else
offset = arc_linux_core_reg_offsets[regnum];
regcache->raw_collect (regnum, buf + offset);
}
void
arc_linux_collect_gregset (const struct regset *regset,
const struct regcache *regcache,
int regnum, void *gregs, size_t size)
{
gdb_static_assert (ARC_LAST_REGNUM
< ARRAY_SIZE (arc_linux_core_reg_offsets));
gdb_byte *buf = (gdb_byte *) gregs;
struct gdbarch *gdbarch = regcache->arch ();
/* regnum == -1 means writing all the registers. */
if (regnum == -1)
for (int reg = 0; reg <= ARC_LAST_REGNUM; reg++)
collect_register (regcache, gdbarch, reg, buf);
else if (regnum <= ARC_LAST_REGNUM)
collect_register (regcache, gdbarch, regnum, buf);
else
gdb_assert_not_reached ("Invalid regnum in arc_linux_collect_gregset.");
}
void
arc_linux_collect_v2_regset (const struct regset *regset,
const struct regcache *regcache, int regnum,
void *v2_regs, size_t size)
{
bfd_byte *buf = (bfd_byte *) v2_regs;
regcache->raw_collect (ARC_R30_REGNUM, buf);
regcache->raw_collect (ARC_R58_REGNUM, buf + REGOFF (1));
regcache->raw_collect (ARC_R59_REGNUM, buf + REGOFF (2));
}
/* Linux regset definitions. */
static const struct regset arc_linux_gregset = {
arc_linux_core_reg_offsets,
arc_linux_supply_gregset,
arc_linux_collect_gregset,
};
static const struct regset arc_linux_v2_regset = {
NULL,
arc_linux_supply_v2_regset,
arc_linux_collect_v2_regset,
};
/* Implement the `iterate_over_regset_sections` gdbarch method. */
static void
arc_linux_iterate_over_regset_sections (struct gdbarch *gdbarch,
iterate_over_regset_sections_cb *cb,
void *cb_data,
const struct regcache *regcache)
{
/* There are 40 registers in Linux user_regs_struct, although some of
them are now just a mere paddings, kept to maintain binary
compatibility with older tools. */
const int sizeof_gregset = 40 * ARC_REGISTER_SIZE;
cb (".reg", sizeof_gregset, sizeof_gregset, &arc_linux_gregset, NULL,
cb_data);
cb (".reg-arc-v2", ARC_LINUX_SIZEOF_V2_REGSET, ARC_LINUX_SIZEOF_V2_REGSET,
&arc_linux_v2_regset, NULL, cb_data);
}
/* Implement the `core_read_description` gdbarch method. */
static const struct target_desc *
arc_linux_core_read_description (struct gdbarch *gdbarch,
struct target_ops *target,
bfd *abfd)
{
arc_arch_features features
= arc_arch_features_create (abfd,
gdbarch_bfd_arch_info (gdbarch)->mach);
return arc_lookup_target_description (features);
}
/* Initialization specific to Linux environment. */
static void
arc_linux_init_osabi (struct gdbarch_info info, struct gdbarch *gdbarch)
{
struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
if (arc_debug)
debug_printf ("arc-linux: GNU/Linux OS/ABI initialization.\n");
/* If we are using Linux, we have in uClibc
(libc/sysdeps/linux/arc/bits/setjmp.h):
typedef int __jmp_buf[13+1+1+1]; //r13-r25, fp, sp, blink
Where "blink" is a stored PC of a caller function.
*/
tdep->jb_pc = 15;
linux_init_abi (info, gdbarch);
/* Set up target dependent GDB architecture entries. */
set_gdbarch_cannot_fetch_register (gdbarch, arc_linux_cannot_fetch_register);
set_gdbarch_cannot_store_register (gdbarch, arc_linux_cannot_store_register);
set_gdbarch_breakpoint_kind_from_pc (gdbarch,
arc_linux_breakpoint_kind_from_pc);
set_gdbarch_sw_breakpoint_from_kind (gdbarch,
arc_linux_sw_breakpoint_from_kind);
set_gdbarch_fetch_tls_load_module_address (gdbarch,
svr4_fetch_objfile_link_map);
set_gdbarch_software_single_step (gdbarch, arc_linux_software_single_step);
set_gdbarch_skip_trampoline_code (gdbarch, find_solib_trampoline_target);
set_gdbarch_skip_solib_resolver (gdbarch, arc_linux_skip_solib_resolver);
set_gdbarch_iterate_over_regset_sections
(gdbarch, arc_linux_iterate_over_regset_sections);
set_gdbarch_core_read_description (gdbarch, arc_linux_core_read_description);
/* GNU/Linux uses SVR4-style shared libraries, with 32-bit ints, longs
and pointers (ILP32). */
set_solib_svr4_fetch_link_map_offsets (gdbarch,
svr4_ilp32_fetch_link_map_offsets);
}
/* Suppress warning from -Wmissing-prototypes. */
extern initialize_file_ftype _initialize_arc_linux_tdep;
void
_initialize_arc_linux_tdep ()
{
gdbarch_register_osabi (bfd_arch_arc, 0, GDB_OSABI_LINUX,
arc_linux_init_osabi);
}
View Git Blame Copy Permalink