The gdbarch::max_insn_length field is used mostly to support displaced
stepping; it controls the size of the buffers allocated for the
displaced-step instruction, and is also used when first copying the
instruction, and later, when fixing up the instruction, in order to
read in and parse the instruction being stepped.
However, it has started to be used in other places in GDB, for
example, it's used in the Python disassembler API, and it is used on
amd64 as part of branch-tracing instruction classification.
The problem is that the value assigned to max_insn_length is not
always the maximum instruction length, but sometimes is a multiple of
that length, as required to support displaced stepping, see rs600,
ARM, and AArch64 for examples of this.
It seems to me that we are overloading the meaning of the
max_insn_length field, and I think that could potentially lead to
confusion.
I propose that we add a new gdbarch field,
gdbarch::displaced_step_max_buffer_length, this new field will do
exactly what it says on the tin; represent the required displaced step
buffer size. The max_insn_length field can then do exactly what it
claims to do; represent the maximum length of a single instruction.
As some architectures (e.g. i386, and amd64) only require their
displaced step buffers to be a single instruction in size, I propose
that the default for displaced_step_max_buffer_length will be the
value of max_insn_length. Architectures than need more buffer space
can then override this default as needed.
I've updated all architectures to setup the new field if appropriate,
and I've audited all calls to gdbarch_max_insn_length and switched to
gdbarch_displaced_step_max_buffer_length where appropriate.
There should be no user visible changes after this commit.
---
gdb/aarch64-linux-tdep.c | 4 +++-
gdb/arm-tdep.c | 4 +++-
gdb/displaced-stepping.c | 6 +++---
gdb/gdbarch-gen.h | 17 ++++++++++++++---
gdb/gdbarch.c | 26 ++++++++++++++++++++++++++
gdb/gdbarch_components.py | 23 ++++++++++++++++++++---
gdb/linux-tdep.c | 2 +-
gdb/rs6000-tdep.c | 6 ++++--
8 files changed, 74 insertions(+), 14 deletions(-)
On 2/28/23 11:51, Andrew Burgess via Gdb-patches wrote:
> The gdbarch::max_insn_length field is used mostly to support displaced
> stepping; it controls the size of the buffers allocated for the
> displaced-step instruction, and is also used when first copying the
> instruction, and later, when fixing up the instruction, in order to
> read in and parse the instruction being stepped.
>
> However, it has started to be used in other places in GDB, for
> example, it's used in the Python disassembler API, and it is used on
> amd64 as part of branch-tracing instruction classification.
>
> The problem is that the value assigned to max_insn_length is not
> always the maximum instruction length, but sometimes is a multiple of
> that length, as required to support displaced stepping, see rs600,
> ARM, and AArch64 for examples of this.
>
> It seems to me that we are overloading the meaning of the
> max_insn_length field, and I think that could potentially lead to
> confusion.
>
> I propose that we add a new gdbarch field,
> gdbarch::displaced_step_max_buffer_length, this new field will do
> exactly what it says on the tin; represent the required displaced step
> buffer size. The max_insn_length field can then do exactly what it
> claims to do; represent the maximum length of a single instruction.
>
> As some architectures (e.g. i386, and amd64) only require their
> displaced step buffers to be a single instruction in size, I propose
> that the default for displaced_step_max_buffer_length will be the
> value of max_insn_length. Architectures than need more buffer space
> can then override this default as needed.
>
> I've updated all architectures to setup the new field if appropriate,
> and I've audited all calls to gdbarch_max_insn_length and switched to
> gdbarch_displaced_step_max_buffer_length where appropriate.
Naming nit: the displaced step buffer length isn't variable, so I don't
think it makes sense to have "max" in the name. I think it should be
named "displaced_step_buffer_length", since it's _the_ displaced step
buffer length.
> diff --git a/gdb/gdbarch_components.py b/gdb/gdbarch_components.py
> index 1d420a513f9..a00398bb03d 100644
> --- a/gdb/gdbarch_components.py
> +++ b/gdb/gdbarch_components.py
> @@ -1752,7 +1752,10 @@ Advance PC to next instruction in order to skip a permanent breakpoint.
>
> Value(
> comment="""
> -The maximum length of an instruction on this architecture in bytes.
> +The maximum length of an instruction on this architecture in octets.
> +This must be set for architectures that support displaced-stepping.
> +Setting this for other architectures improves error detection within
> +the Python disassembler API.
I'm not sure I understand why you would mention that this needs to be
set for architectures supporting displaced-stepping, since the point of
your change is to use displaced_step_max_buffer_length for those use
cases.
These were only minor comments, the patch LGTM in general:
Approved-By: Simon Marchi <simon.marchi@efficios.com>
Simon
@@ -2240,7 +2240,9 @@ aarch64_linux_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch)
set_gdbarch_get_syscall_number (gdbarch, aarch64_linux_get_syscall_number);
/* Displaced stepping. */
- set_gdbarch_max_insn_length (gdbarch, 4 * AARCH64_DISPLACED_MODIFIED_INSNS);
+ set_gdbarch_max_insn_length (gdbarch, 4);
+ set_gdbarch_displaced_step_max_buffer_length
+ (gdbarch, 4 * AARCH64_DISPLACED_MODIFIED_INSNS);
set_gdbarch_displaced_step_copy_insn (gdbarch,
aarch64_displaced_step_copy_insn);
set_gdbarch_displaced_step_fixup (gdbarch, aarch64_displaced_step_fixup);
@@ -10662,7 +10662,9 @@ arm_gdbarch_init (struct gdbarch_info info, struct gdbarch_list *arches)
/* Note: for displaced stepping, this includes the breakpoint, and one word
of additional scratch space. This setting isn't used for anything beside
displaced stepping at present. */
- set_gdbarch_max_insn_length (gdbarch, 4 * ARM_DISPLACED_MODIFIED_INSNS);
+ set_gdbarch_displaced_step_max_buffer_length
+ (gdbarch, 4 * ARM_DISPLACED_MODIFIED_INSNS);
+ set_gdbarch_max_insn_length (gdbarch, 4);
/* This should be low enough for everything. */
tdep->lowest_pc = 0x20;
@@ -55,7 +55,7 @@ displaced_step_buffers::prepare (thread_info *thread, CORE_ADDR &displaced_pc)
regcache *regcache = get_thread_regcache (thread);
const address_space *aspace = regcache->aspace ();
gdbarch *arch = regcache->arch ();
- ULONGEST len = gdbarch_max_insn_length (arch);
+ ULONGEST len = gdbarch_displaced_step_max_buffer_length (arch);
/* Search for an unused buffer. */
displaced_step_buffer *buffer = nullptr;
@@ -243,7 +243,7 @@ displaced_step_buffers::finish (gdbarch *arch, thread_info *thread,
below. */
thread->inf->displaced_step_state.unavailable = false;
- ULONGEST len = gdbarch_max_insn_length (arch);
+ ULONGEST len = gdbarch_displaced_step_max_buffer_length (arch);
/* Restore memory of the buffer. */
write_memory_ptid (thread->ptid, buffer->addr,
@@ -302,7 +302,7 @@ displaced_step_buffers::restore_in_ptid (ptid_t ptid)
regcache *regcache = get_thread_regcache (buffer.current_thread);
gdbarch *arch = regcache->arch ();
- ULONGEST len = gdbarch_max_insn_length (arch);
+ ULONGEST len = gdbarch_displaced_step_max_buffer_length (arch);
write_memory_ptid (ptid, buffer.addr, buffer.saved_copy.data (), len);
@@ -1018,7 +1018,10 @@ typedef void (gdbarch_skip_permanent_breakpoint_ftype) (struct regcache *regcach
extern void gdbarch_skip_permanent_breakpoint (struct gdbarch *gdbarch, struct regcache *regcache);
extern void set_gdbarch_skip_permanent_breakpoint (struct gdbarch *gdbarch, gdbarch_skip_permanent_breakpoint_ftype *skip_permanent_breakpoint);
-/* The maximum length of an instruction on this architecture in bytes. */
+/* The maximum length of an instruction on this architecture in octets.
+ This must be set for architectures that support displaced-stepping.
+ Setting this for other architectures improves error detection within
+ the Python disassembler API. */
extern bool gdbarch_max_insn_length_p (struct gdbarch *gdbarch);
@@ -1039,8 +1042,8 @@ extern void set_gdbarch_max_insn_length (struct gdbarch *gdbarch, ULONGEST max_i
see the comments in infrun.c.
The TO area is only guaranteed to have space for
- gdbarch_max_insn_length (arch) bytes, so this function must not
- write more bytes than that to that area.
+ gdbarch_displaced_step_max_buffer_length (arch) octets, so this
+ function must not write more octets than that to this area.
If you do not provide this function, GDB assumes that the
architecture does not support displaced stepping.
@@ -1122,6 +1125,14 @@ typedef void (gdbarch_displaced_step_restore_all_in_ptid_ftype) (inferior *paren
extern void gdbarch_displaced_step_restore_all_in_ptid (struct gdbarch *gdbarch, inferior *parent_inf, ptid_t child_ptid);
extern void set_gdbarch_displaced_step_restore_all_in_ptid (struct gdbarch *gdbarch, gdbarch_displaced_step_restore_all_in_ptid_ftype *displaced_step_restore_all_in_ptid);
+/* The maximum length in octets required for a displaced-step instruction
+ buffer. By default this will be the same as gdbarch::max_insn_length,
+ but should be overridden for architectures that might expand a
+ displaced-step instruction to multiple replacement instructions. */
+
+extern ULONGEST gdbarch_displaced_step_max_buffer_length (struct gdbarch *gdbarch);
+extern void set_gdbarch_displaced_step_max_buffer_length (struct gdbarch *gdbarch, ULONGEST displaced_step_max_buffer_length);
+
/* Relocate an instruction to execute at a different address. OLDLOC
is the address in the inferior memory where the instruction to
relocate is currently at. On input, TO points to the destination
@@ -192,6 +192,7 @@ struct gdbarch
gdbarch_displaced_step_finish_ftype *displaced_step_finish = NULL;
gdbarch_displaced_step_copy_insn_closure_by_addr_ftype *displaced_step_copy_insn_closure_by_addr = nullptr;
gdbarch_displaced_step_restore_all_in_ptid_ftype *displaced_step_restore_all_in_ptid = nullptr;
+ ULONGEST displaced_step_max_buffer_length = 0;
gdbarch_relocate_instruction_ftype *relocate_instruction = NULL;
gdbarch_overlay_update_ftype *overlay_update = nullptr;
gdbarch_core_read_description_ftype *core_read_description = nullptr;
@@ -453,6 +454,10 @@ verify_gdbarch (struct gdbarch *gdbarch)
log.puts ("\n\tdisplaced_step_finish");
/* Skip verify of displaced_step_copy_insn_closure_by_addr, has predicate. */
/* Skip verify of displaced_step_restore_all_in_ptid, invalid_p == 0 */
+ if (gdbarch->displaced_step_max_buffer_length == 0)
+ gdbarch->displaced_step_max_buffer_length = gdbarch->max_insn_length;
+ if (gdbarch->displaced_step_max_buffer_length < gdbarch->max_insn_length)
+ log.puts ("\n\tdisplaced_step_max_buffer_length");
/* Skip verify of relocate_instruction, has predicate. */
/* Skip verify of overlay_update, has predicate. */
/* Skip verify of core_read_description, has predicate. */
@@ -1111,6 +1116,9 @@ gdbarch_dump (struct gdbarch *gdbarch, struct ui_file *file)
gdb_printf (file,
"gdbarch_dump: displaced_step_restore_all_in_ptid = <%s>\n",
host_address_to_string (gdbarch->displaced_step_restore_all_in_ptid));
+ gdb_printf (file,
+ "gdbarch_dump: displaced_step_max_buffer_length = %s\n",
+ plongest (gdbarch->displaced_step_max_buffer_length));
gdb_printf (file,
"gdbarch_dump: gdbarch_relocate_instruction_p() = %d\n",
gdbarch_relocate_instruction_p (gdbarch));
@@ -4153,6 +4161,24 @@ set_gdbarch_displaced_step_restore_all_in_ptid (struct gdbarch *gdbarch,
gdbarch->displaced_step_restore_all_in_ptid = displaced_step_restore_all_in_ptid;
}
+ULONGEST
+gdbarch_displaced_step_max_buffer_length (struct gdbarch *gdbarch)
+{
+ gdb_assert (gdbarch != NULL);
+ /* Check variable is valid. */
+ gdb_assert (!(gdbarch->displaced_step_max_buffer_length < gdbarch->max_insn_length));
+ if (gdbarch_debug >= 2)
+ gdb_printf (gdb_stdlog, "gdbarch_displaced_step_max_buffer_length called\n");
+ return gdbarch->displaced_step_max_buffer_length;
+}
+
+void
+set_gdbarch_displaced_step_max_buffer_length (struct gdbarch *gdbarch,
+ ULONGEST displaced_step_max_buffer_length)
+{
+ gdbarch->displaced_step_max_buffer_length = displaced_step_max_buffer_length;
+}
+
bool
gdbarch_relocate_instruction_p (struct gdbarch *gdbarch)
{
@@ -1752,7 +1752,10 @@ Advance PC to next instruction in order to skip a permanent breakpoint.
Value(
comment="""
-The maximum length of an instruction on this architecture in bytes.
+The maximum length of an instruction on this architecture in octets.
+This must be set for architectures that support displaced-stepping.
+Setting this for other architectures improves error detection within
+the Python disassembler API.
""",
type="ULONGEST",
name="max_insn_length",
@@ -1777,8 +1780,8 @@ For a general explanation of displaced stepping and how GDB uses it,
see the comments in infrun.c.
The TO area is only guaranteed to have space for
-gdbarch_max_insn_length (arch) bytes, so this function must not
-write more bytes than that to that area.
+gdbarch_displaced_step_max_buffer_length (arch) octets, so this
+function must not write more octets than that to this area.
If you do not provide this function, GDB assumes that the
architecture does not support displaced stepping.
@@ -1890,6 +1893,20 @@ contents of all displaced step buffers in the child's address space.
invalid=False,
)
+Value(
+ comment="""
+The maximum length in octets required for a displaced-step instruction
+buffer. By default this will be the same as gdbarch::max_insn_length,
+but should be overridden for architectures that might expand a
+displaced-step instruction to multiple replacement instructions.
+""",
+ type="ULONGEST",
+ name="displaced_step_max_buffer_length",
+ predefault="0",
+ postdefault="gdbarch->max_insn_length",
+ invalid="gdbarch->displaced_step_max_buffer_length < gdbarch->max_insn_length",
+)
+
Method(
comment="""
Relocate an instruction to execute at a different address. OLDLOC
@@ -2603,7 +2603,7 @@ linux_displaced_step_prepare (gdbarch *arch, thread_info *thread,
at DISP_STEP_BUF_ADDR. They are all of size BUF_LEN. */
CORE_ADDR disp_step_buf_addr
= linux_displaced_step_location (thread->inf->gdbarch);
- int buf_len = gdbarch_max_insn_length (arch);
+ int buf_len = gdbarch_displaced_step_max_buffer_length (arch);
linux_gdbarch_data *gdbarch_data = get_linux_gdbarch_data (arch);
gdb_assert (gdbarch_data->num_disp_step_buffers > 0);
@@ -889,7 +889,8 @@ ppc_displaced_step_copy_insn (struct gdbarch *gdbarch,
CORE_ADDR from, CORE_ADDR to,
struct regcache *regs)
{
- size_t len = gdbarch_max_insn_length (gdbarch);
+ size_t len = gdbarch_displaced_step_max_buffer_length (gdbarch);
+ gdb_assert (len > PPC_INSN_SIZE);
std::unique_ptr<ppc_displaced_step_copy_insn_closure> closure
(new ppc_displaced_step_copy_insn_closure (len));
gdb_byte *buf = closure->buf.data ();
@@ -8363,8 +8364,9 @@ rs6000_gdbarch_init (struct gdbarch_info info, struct gdbarch_list *arches)
set_gdbarch_displaced_step_finish (gdbarch, ppc_displaced_step_finish);
set_gdbarch_displaced_step_restore_all_in_ptid
(gdbarch, ppc_displaced_step_restore_all_in_ptid);
+ set_gdbarch_displaced_step_max_buffer_length (gdbarch, 2 * PPC_INSN_SIZE);
- set_gdbarch_max_insn_length (gdbarch, 2 * PPC_INSN_SIZE);
+ set_gdbarch_max_insn_length (gdbarch, PPC_INSN_SIZE);
/* Hook in ABI-specific overrides, if they have been registered. */
info.target_desc = tdesc;