This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [patch] ia64: Fix breakpoints memory shadow
- From: Jan Kratochvil <jan dot kratochvil at redhat dot com>
- To: Joel Brobecker <brobecker at adacore dot com>
- Cc: gdb-patches at sourceware dot org
- Date: Tue, 11 Nov 2008 14:17:26 +0100
- Subject: Re: [patch] ia64: Fix breakpoints memory shadow
- References: <20081028172816.GA1284@host0.dyn.jankratochvil.net> <20081029210242.GA3635@adacore.com> <20081030144841.GA26606@host0.dyn.jankratochvil.net> <20081101185410.GB15606@adacore.com>
On Sat, 01 Nov 2008 19:54:10 +0100, Joel Brobecker wrote:
> Can we maybe expand the comment explaining why we can't save the whole
> bundle in the shadow buffer?
Done, mostly copied the mail of mine there.
> > const unsigned char *
> > ia64_breakpoint_from_pc (struct gdbarch *gdbarch, CORE_ADDR *pcptr, int *lenptr)
> > {
> > - static unsigned char breakpoint[] =
> > - { 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 };
...
> I think there is a piece missing in this change. Looks like breakpoint
> is no longer initialized.
According to ISO C90 6.5.7 such a static array is (was) initialized to zeroes.
> I think that what you need to do is read the instruction bundle from memory,
> and insert the breakpoint instruction, and then copy you
> BUNDLE_LEN - 2 bytes into the breakpoint buffer.
I just kept it broken with no regressions. But OK, fixed.
> The reason why we didn't trip over this, I believe, is because this routine
> is only used when using the default insert/remove bp routines, or when using
> the remote protocol. It's also used from monitor.c.
The return value is used for the ia64 case only in bp_loc_is_permanent,
nowhere else. And for ia64 the arch interface is inappropriate but we can
keep it this way as it is probably the right interface for anything non-ia64.
> Also, the check for BUNDLE_LEN against BREAKPOINT_MAX is slightly
> over-pessimistic. Can you move it somewhere at the beginning of
> the file (this assumption is required in other parts of the code),
> and compare BREAKPOINT_MAX against the actual size of data saved
> (BUNDLE_LEN -2)?
Right, thanks, done.
Regards,
Jan
gdb/
2008-11-11 Jan Kratochvil <jan.kratochvil@redhat.com>
Fix automatic restoration of breakpoints memory for ia64.
* ia64-tdep.c: New #if check on BREAKPOINT_MAX vs. BUNDLE_LEN.
(ia64_memory_insert_breakpoint): New comment part for SHADOW_CONTENTS
content. Remova variable instr. New variable cleanup. Force
automatic breakpoints restoration. PLACED_SIZE and SHADOW_LEN are now
set larger, to BUNDLE_LEN - 2. Variable `bundle' type update.
(ia64_memory_remove_breakpoint): Rename variables bundle to bundle_mem
and instr to instr_saved. New variables bundle_saved and
instr_breakpoint. Comment new reasons why we need to disable automatic
restoration of breakpoints. Assert PLACED_SIZE and SHADOW_LEN. New
check of the original memory content.
(ia64_breakpoint_from_pc): Implement the emulation of permanent
breakpoints compatible with current bp_loc_is_permanent.
(template_encoding_table): Make it `const'.
* breakpoint.c (bp_loc_is_permanent): Support unsupported software
breakpoints. New variables `cleanup' and `retval' to protect
target_read_memory by make_show_memory_breakpoints_cleanup.
* monitor.c (monitor_insert_breakpoint): Remove unused variable `bp'.
gdb/testsuite/
2008-11-11 Jan Kratochvil <jan.kratochvil@redhat.com>
* gdb.base/breakpoint-shadow.exp, gdb.base/breakpoint-shadow.c: New.
* gdb.base/start.exp: New comment about an alternative - `runto_main'.
--- gdb/breakpoint.c 3 Nov 2008 14:01:26 -0000 1.359
+++ gdb/breakpoint.c 11 Nov 2008 12:57:12 -0000
@@ -5234,19 +5234,32 @@ bp_loc_is_permanent (struct bp_location
CORE_ADDR addr;
const gdb_byte *brk;
gdb_byte *target_mem;
+ struct cleanup *cleanup;
+ int retval = 0;
gdb_assert (loc != NULL);
addr = loc->address;
brk = gdbarch_breakpoint_from_pc (current_gdbarch, &addr, &len);
+ /* Software breakpoints unsupported? */
+ if (brk == NULL)
+ return 0;
+
target_mem = alloca (len);
+ /* Enable the automatic memory restoration from breakpoints while
+ we read the memory. Otherwise we could say about our temporary
+ breakpoints they are permanent. */
+ cleanup = make_show_memory_breakpoints_cleanup (0);
+
if (target_read_memory (loc->address, target_mem, len) == 0
&& memcmp (target_mem, brk, len) == 0)
- return 1;
+ retval = 1;
- return 0;
+ do_cleanups (cleanup);
+
+ return retval;
}
--- gdb/ia64-tdep.c 11 Sep 2008 14:23:15 -0000 1.184
+++ gdb/ia64-tdep.c 11 Nov 2008 12:57:12 -0000
@@ -110,6 +110,12 @@ typedef enum instruction_type
#define BUNDLE_LEN 16
+/* See the saved memory layout comment for ia64_memory_insert_breakpoint. */
+
+#if BREAKPOINT_MAX < BUNDLE_LEN - 2
+# error "BREAKPOINT_MAX < BUNDLE_LEN - 2"
+#endif
+
static gdbarch_init_ftype ia64_gdbarch_init;
static gdbarch_register_name_ftype ia64_register_name;
@@ -442,7 +448,7 @@ replace_slotN_contents (char *bundle, lo
replace_bit_field (bundle, instr, 5+41*slotnum, 41);
}
-static enum instruction_type template_encoding_table[32][3] =
+static const enum instruction_type template_encoding_table[32][3] =
{
{ M, I, I }, /* 00 */
{ M, I, I }, /* 01 */
@@ -545,7 +551,45 @@ fetch_instruction (CORE_ADDR addr, instr
simulators. So I changed the pattern slightly to do "break.i 0x080001"
instead. But that didn't work either (I later found out that this
pattern was used by the simulator that I was using.) So I ended up
- using the pattern seen below. */
+ using the pattern seen below.
+
+ SHADOW_CONTENTS has byte-based addressing (PLACED_ADDRESS and SHADOW_LEN)
+ while we need bit-based addressing as the instructions length is 41 bits and
+ we must not modify/corrupt the adjacent slots in the same bundle.
+ Fortunately we may store larger memory incl. the adjacent bits with the
+ original memory content (not the possibly already stored breakpoints there).
+ We need to be careful in ia64_memory_remove_breakpoint to always restore
+ only the specific bits of this instruction ignoring any adjacent stored
+ bits.
+
+ We use the original addressing with the low nibble in the range <0..2> which
+ gets incorrectly interpreted by generic non-ia64 breakpoint_restore_shadows
+ as the direct byte offset of SHADOW_CONTENTS. We store whole BUNDLE_LEN
+ bytes just without these two possibly skipped bytes to not to exceed to the
+ next bundle.
+
+ If we would like to store the whole bundle to SHADOW_CONTENTS we would have
+ to store already the base address (`address & ~0x0f') into PLACED_ADDRESS.
+ In such case there is no other place where to store
+ SLOTNUM (`adress & 0x0f', value in the range <0..2>). We need to know
+ SLOTNUM in ia64_memory_remove_breakpoint.
+
+ ia64 16-byte bundle layout:
+ | 5 bits | slot 0 with 41 bits | slot 1 with 41 bits | slot 2 with 41 bits |
+
+ The current addressing used by the code below:
+ original PC placed_address placed_size required covered
+ == bp_tgt->shadow_len reqd \subset covered
+ 0xABCDE0 0xABCDE0 0xE <0x0...0x5> <0x0..0xD>
+ 0xABCDE1 0xABCDE1 0xE <0x5...0xA> <0x1..0xE>
+ 0xABCDE2 0xABCDE2 0xE <0xA...0xF> <0x2..0xF>
+
+ `objdump -d' and some other tools show a bit unjustified offsets:
+ original PC byte where starts the instruction objdump offset
+ 0xABCDE0 0xABCDE0 0xABCDE0
+ 0xABCDE1 0xABCDE5 0xABCDE6
+ 0xABCDE2 0xABCDEA 0xABCDEC
+ */
#if 0
#define IA64_BREAKPOINT 0x00002000040LL
@@ -557,34 +601,43 @@ ia64_memory_insert_breakpoint (struct gd
struct bp_target_info *bp_tgt)
{
CORE_ADDR addr = bp_tgt->placed_address;
- char bundle[BUNDLE_LEN];
+ gdb_byte bundle[BUNDLE_LEN];
int slotnum = (int) (addr & 0x0f) / SLOT_MULTIPLIER;
- long long instr;
int val;
int template;
+ struct cleanup *cleanup;
if (slotnum > 2)
error (_("Can't insert breakpoint for slot numbers greater than 2."));
addr &= ~0x0f;
+ /* Enable the automatic memory restoration from breakpoints while
+ we read our instruction bundle. Otherwise, we could store into
+ SHADOW_CONTENTS an already stored breakpoint at the same location.
+ In practice it is already being prevented by the DUPLICATE field and
+ update_global_location_list. */
+ cleanup = make_show_memory_breakpoints_cleanup (0);
val = target_read_memory (addr, bundle, BUNDLE_LEN);
/* Check for L type instruction in 2nd slot, if present then
bump up the slot number to the 3rd slot */
template = extract_bit_field (bundle, 0, 5);
- if (slotnum == 1 && template_encoding_table[template][1] == L)
- {
- slotnum = 2;
- }
+ if (slotnum == 1 && template_encoding_table[template][slotnum] == L)
+ slotnum = 2;
+
+ /* Slot number 2 may skip at most 2 bytes at the beginning. */
+ bp_tgt->placed_size = bp_tgt->shadow_len = BUNDLE_LEN - 2;
+
+ /* Store the whole bundle, except for the initial skipped bytes by the slot
+ number interpreted as bytes offset in PLACED_ADDRESS. */
+ memcpy (bp_tgt->shadow_contents, bundle + slotnum, bp_tgt->shadow_len);
- instr = slotN_contents (bundle, slotnum);
- memcpy (bp_tgt->shadow_contents, &instr, sizeof (instr));
- bp_tgt->placed_size = bp_tgt->shadow_len = sizeof (instr);
replace_slotN_contents (bundle, IA64_BREAKPOINT, slotnum);
if (val == 0)
- target_write_memory (addr, bundle, BUNDLE_LEN);
+ target_write_memory (addr + slotnum, bundle + slotnum, bp_tgt->shadow_len);
+ do_cleanups (cleanup);
return val;
}
@@ -593,9 +646,9 @@ ia64_memory_remove_breakpoint (struct gd
struct bp_target_info *bp_tgt)
{
CORE_ADDR addr = bp_tgt->placed_address;
- char bundle[BUNDLE_LEN];
+ gdb_byte bundle_mem[BUNDLE_LEN], bundle_saved[BUNDLE_LEN];
int slotnum = (addr & 0x0f) / SLOT_MULTIPLIER;
- long long instr;
+ long long instr_breakpoint, instr_saved;
int val;
int template;
struct cleanup *cleanup;
@@ -604,40 +657,93 @@ ia64_memory_remove_breakpoint (struct gd
/* Disable the automatic memory restoration from breakpoints while
we read our instruction bundle. Otherwise, the general restoration
- mechanism kicks in and ends up corrupting our bundle, because it
- is not aware of the concept of instruction bundles. */
+ mechanism kicks in and we would possibly remove parts of the adjacent
+ placed breakpoints. It is due to our SHADOW_CONTENTS overlapping the real
+ breakpoint instruction bits region. */
cleanup = make_show_memory_breakpoints_cleanup (1);
- val = target_read_memory (addr, bundle, BUNDLE_LEN);
+ val = target_read_memory (addr, bundle_mem, BUNDLE_LEN);
/* Check for L type instruction in 2nd slot, if present then
bump up the slot number to the 3rd slot */
- template = extract_bit_field (bundle, 0, 5);
- if (slotnum == 1 && template_encoding_table[template][1] == L)
- {
- slotnum = 2;
- }
-
- memcpy (&instr, bp_tgt->shadow_contents, sizeof instr);
- replace_slotN_contents (bundle, instr, slotnum);
+ template = extract_bit_field (bundle_mem, 0, 5);
+ if (slotnum == 1 && template_encoding_table[template][slotnum] == L)
+ slotnum = 2;
+
+ gdb_assert (bp_tgt->placed_size == BUNDLE_LEN - 2);
+ gdb_assert (bp_tgt->placed_size == bp_tgt->shadow_len);
+
+ instr_breakpoint = slotN_contents (bundle_mem, slotnum);
+ if (instr_breakpoint != IA64_BREAKPOINT)
+ warning (_("Breakpoint removal cannot find the placed breakpoint at %s"),
+ paddr_nz (bp_tgt->placed_address));
+
+ /* Extract the original saved instruction from SLOTNUM normalizing its
+ bit-shift for INSTR_SAVED. */
+ memcpy (bundle_saved, bundle_mem, BUNDLE_LEN);
+ memcpy (bundle_saved + slotnum, bp_tgt->shadow_contents, bp_tgt->shadow_len);
+ instr_saved = slotN_contents (bundle_saved, slotnum);
+
+ /* In BUNDLE_MEM be careful to modify only the bits belonging to SLOTNUM and
+ never any other possibly also stored in SHADOW_CONTENTS. */
+ replace_slotN_contents (bundle_mem, instr_saved, slotnum);
if (val == 0)
- target_write_memory (addr, bundle, BUNDLE_LEN);
+ target_write_memory (addr, bundle_mem, BUNDLE_LEN);
do_cleanups (cleanup);
return val;
}
-/* We don't really want to use this, but remote.c needs to call it in order
- to figure out if Z-packets are supported or not. Oh, well. */
-const unsigned char *
+/* Our caller bp_loc_is_permanent uses plain memcmp expecting byte ranges of
+ the instructions. We provide the extended range as described for
+ ia64_memory_insert_breakpoint simulating a placed breakpoint there - memcmp
+ will return true iff there was already a breakpoint (and so we returned the
+ original memory). We just take care of preserving the `break' instruction
+ 21-bit (or 62-bit) parameter as the memcmp match would fail otherwise. */
+
+static const gdb_byte *
ia64_breakpoint_from_pc (struct gdbarch *gdbarch, CORE_ADDR *pcptr, int *lenptr)
{
- static unsigned char breakpoint[] =
- { 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 };
- *lenptr = sizeof (breakpoint);
-#if 0
- *pcptr &= ~0x0f;
-#endif
- return breakpoint;
+ CORE_ADDR addr = *pcptr;
+ static gdb_byte bundle[BUNDLE_LEN];
+ int slotnum = (int) (*pcptr & 0x0f) / SLOT_MULTIPLIER;
+ long long instr_fetched;
+ int val;
+ int template;
+ struct cleanup *cleanup;
+
+ if (slotnum > 2)
+ error (_("Can't insert breakpoint for slot numbers greater than 2."));
+
+ addr &= ~0x0f;
+
+ /* Enable the automatic memory restoration from breakpoints while
+ we read our instruction bundle to match bp_loc_is_permanent. */
+ cleanup = make_show_memory_breakpoints_cleanup (0);
+ val = target_read_memory (addr, bundle, BUNDLE_LEN);
+ do_cleanups (cleanup);
+
+ /* It may be currently unreachable breakpoint which is never permanent. */
+ if (val != 0)
+ return NULL;
+
+ /* Check for L type instruction in 2nd slot, if present then
+ bump up the slot number to the 3rd slot */
+ template = extract_bit_field (bundle, 0, 5);
+ if (slotnum == 1 && template_encoding_table[template][slotnum] == L)
+ slotnum = 2;
+
+ /* We need to keep the possible `break' instruction parameter value intact.
+ Opcode with cleared all the bits (except the parameter value) is `break'.
+ For L+X slot pair we are at the X slot (3) so we should not touch the
+ L slot - the upper 41 bits of the parameter. */
+ instr_fetched = slotN_contents (bundle, slotnum);
+ instr_fetched &= 0x1003ffffc0;
+ replace_slotN_contents (bundle, instr_fetched, slotnum);
+
+ *lenptr = BUNDLE_LEN - 2;
+
+ /* SLOTNUM is possibly already locally modified - use caller's *PCPTR. */
+ return bundle + (*pcptr & 0x0f);
}
static CORE_ADDR
--- gdb/monitor.c 9 Nov 2008 11:27:17 -0000 1.82
+++ gdb/monitor.c 11 Nov 2008 12:57:12 -0000
@@ -2026,7 +2026,6 @@ monitor_insert_breakpoint (struct bp_tar
{
CORE_ADDR addr = bp_tgt->placed_address;
int i;
- const unsigned char *bp;
int bplen;
monitor_debug ("MON inst bkpt %s\n", paddr (addr));
@@ -2037,7 +2036,7 @@ monitor_insert_breakpoint (struct bp_tar
addr = gdbarch_addr_bits_remove (current_gdbarch, addr);
/* Determine appropriate breakpoint size for this address. */
- bp = gdbarch_breakpoint_from_pc (current_gdbarch, &addr, &bplen);
+ gdbarch_breakpoint_from_pc (current_gdbarch, &addr, &bplen);
bp_tgt->placed_address = addr;
bp_tgt->placed_size = bplen;
--- /dev/null 1 Jan 1970 00:00:00 -0000
+++ gdb/testsuite/gdb.base/breakpoint-shadow.c 11 Nov 2008 12:57:13 -0000
@@ -0,0 +1,27 @@
+/* This testcase is part of GDB, the GNU debugger.
+
+ Copyright 2008 Free Software Foundation, Inc.
+
+ 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/>. */
+
+int
+main (void)
+{
+ volatile int i;
+
+ i = 1; /* break-first */
+ i = 2; /* break-second */
+
+ return 0;
+}
--- /dev/null 1 Jan 1970 00:00:00 -0000
+++ gdb/testsuite/gdb.base/breakpoint-shadow.exp 11 Nov 2008 12:57:13 -0000
@@ -0,0 +1,63 @@
+# Copyright 2008 Free Software Foundation, Inc.
+
+# 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/>.
+
+set testfile breakpoint-shadow
+set srcfile ${testfile}.c
+set binfile ${objdir}/${subdir}/${testfile}
+if { [gdb_compile "${srcdir}/${subdir}/${srcfile}" "${binfile}" executable {debug}] != "" } {
+ untested "Couldn't compile test program"
+ return -1
+}
+
+gdb_exit
+gdb_start
+gdb_reinitialize_dir $srcdir/$subdir
+gdb_load ${binfile}
+
+# We need to start the inferior to place the breakpoints in the memory at all.
+if ![runto_main] {
+ untested start
+ return -1
+}
+
+# The default "auto" mode removes all the breakpoints when we stop (and not
+# running the nonstop mode). We would not be able to test the shadow.
+gdb_test "set breakpoint always-inserted on"
+gdb_test "show breakpoint always-inserted" "Always inserted breakpoint mode is on."
+
+set match "\nDump of assembler code for function main:\r\n(.*)End of assembler dump.\r\n$gdb_prompt $"
+
+set test "disassembly without breakpoints"
+gdb_test_multiple "disass main" $test {
+ -re $match {
+ set orig $expect_out(1,string)
+ pass $test
+ }
+}
+
+gdb_test "b [gdb_get_line_number "break-first"]" "Breakpoint \[0-9\] at .*" "First breakpoint placed"
+gdb_test "b [gdb_get_line_number "break-second"]" "Breakpoint \[0-9\] at .*" "Second breakpoint placed"
+
+set test "disassembly with breakpoints"
+gdb_test_multiple "disass main" $test {
+ -re $match {
+ set got $expect_out(1,string)
+ if [string equal -nocase $orig $got] {
+ pass $test
+ } else {
+ fail $test
+ }
+ }
+}
--- gdb/testsuite/gdb.base/start.exp 1 Jan 2008 22:53:19 -0000 1.5
+++ gdb/testsuite/gdb.base/start.exp 11 Nov 2008 12:57:13 -0000
@@ -35,6 +35,9 @@ gdb_start
gdb_reinitialize_dir $srcdir/$subdir
gdb_load ${binfile}
+# This is a testcase specifically for the `start' GDB command. For regular
+# stop-in-main goal in the testcases please use `runto_main' instead.
+
# For C programs, "start" should stop in main().
if { [gdb_start_cmd] < 0 } {
untested start
@@ -44,4 +47,3 @@ if { [gdb_start_cmd] < 0 } {
gdb_test "" \
"main \\(\\) at .*start.c.*" \
"start"
-