When inf_ptrace_xfer_partial performs a memory transfer via ptrace with
PT_READ_I, PT_WRITE_I (aka PTRACE_PEEKTEXT, PTRACE_POKETEXT), etc., then
it currently transfers at most one word. This behavior yields degraded
performance, particularly if the caller has significant preparation work
for each invocation. And indeed it has for writing, in
memory_xfer_partial in target.c, where all of the remaining data to be
transferred is copied to a temporary buffer each time, for breakpoint
shadow handling. Thus large writes have quadratic runtime and can take
hours.
Note: On GNU/Linux targets GDB usually does not use
inf_ptrace_xfer_partial for large memory transfers, but attempts a single
read/write from/to /proc/<pid>/mem instead. However, the kernel may
reject writes to /proc/<pid>/mem (such as kernels prior to 2.6.39), or
/proc may not be mounted. In both cases GDB falls back to the ptrace
mechanism.
This patch fixes the performance issue by attempting to fulfill the whole
transfer request in inf_ptrace_xfer_partial, using a loop around the
ptrace call.
gdb/ChangeLog:
PR gdb/21220
* inf-ptrace.c (inf_ptrace_xfer_partial): In "case
TARGET_OBJECT_MEMORY", extract the logic for ptrace peek/poke...
(inf_ptrace_peek_poke): ...here. New function. Now also loop
over ptrace peek/poke until end of buffer or error.
gas/ChangeLog:
2017-03-02 Kuan-Lin Chen <rufus@andestech.com>
* config/tc-riscv.c (md_apply_fix): Set fx_frag and
fx_next->fx_frag for CFA_advance_loc relocations.
Since BFD64 may be used on 32-bit address, we need to apply addr_mask
to check VMA and LMA.
* ldlang.c (lang_check_section_addresses): Use addr_mask to
check VMA and LMA.
It isn't used anywhere else than the file it's defined in.
gdb/ChangeLog:
* parse.c (length_of_subexp): Make static.
* parser-defs.h (length_of_subexp): Remove.
* readelf.c (print_gnu_build_attribute_description): Move symbol
printing code to...
(print_symbol_for_build_attribute): New function. ...here.
Add to find the best symbol to associate with an OPEN note.
Add code to cache the symbol table and string table, so that they
are not loaded every time a note is displayed.
* testsuite/binutils-all/note-2-32.s: Add a function symbol.
* testsuite/binutils-all/note-2-64.s: Likewise.
* testsuite/binutils-all/note-2-32.d: Update expected note output.
* testsuite/binutils-all/note-2-64.d: Likewise.
An optional parameter TEST has been added to get_hexadecimal_valueof in commit:
https://sourceware.org/ml/gdb-patches/2016-06/msg00469.html
This patch adds a similar optional parameter to other related methods that
retrieve expression values: get_valueof, get_integer_valueof and get_sizeof.
Thus tests that evaluate same expression multiple times can provide custom
test names, ensuring that test names will be unique.
gdb/testsuite/ChangeLog:
2017-03-14 Anton Kolesov <anton.kolesov@synopsys.com>
* lib/gdb.exp (get_valueof, get_integer_valueof, get_sizeof):
Add optional 'test' parameter.
So far linux_proc_xfer_partial refused to handle write requests. This is
still based on the assumption that the Linux kernel does not support
writes to /proc/<pid>/mem. That used to be true, but has changed with
Linux 2.6.39 released in May 2011.
This patch lifts this restriction and now exploits /proc/<pid>/mem for
writing to inferior memory as well, if possible.
gdb/ChangeLog:
* linux-nat.c (linux_proc_xfer_partial): Handle write operations
as well.
Commit c8b23b3f89 ("Add constructor and destructor to
demangle_parse_info") a while ago broke the "test-cp-name-parser"
build:
$ make test-cp-name-parser
[...]
src/gdb/cp-name-parser.y: In function ‘int main(int, char**)’:
src/gdb/cp-name-parser.y:2190:9: error: cannot convert ‘std::unique_ptr<demangle_parse_info>’ to ‘demangle_parse_info*’ in assignment
result = cp_demangled_name_to_comp (str2, &errmsg);
^
src/gdb/cp-name-parser.y:2199:38: error: ‘cp_demangled_name_parse_free’ was not declared in this scope
cp_demangled_name_parse_free (result);
^
src/gdb/cp-name-parser.y:2211:14: error: cannot convert ‘std::unique_ptr<demangle_parse_info>’ to ‘demangle_parse_info*’ in assignment
result = cp_demangled_name_to_comp (argv[arg], &errmsg);
^
src/gdb/cp-name-parser.y:2219:43: error: ‘cp_demangled_name_parse_free’ was not declared in this scope
cp_demangled_name_parse_free (result);
^
Makefile:2107: recipe for target 'test-cp-name-parser.o' failed
make: *** [test-cp-name-parser.o] Error 1
This commit restores it.
gdb/ChangeLog:
2017-03-14 Pedro Alves <palves@redhat.com>
* cp-name-parser.y (cp_demangled_name_to_comp): Update comment.
(main): Use std::unique_ptr. Remove calls to
cp_demangled_name_parse_free.
gdb/ChangeLog:
* amd64-linux-nat.c (amd64_linux_fetch_inferior_registers,
amd64_linux_fetch_inferior_registers): Use regcache->ptid
instead of inferior_ptid.
We are currently assuming that regcache->ptid is equal to inferior_ptid
when we call target_fetch/store_registers. These asserts just validate
that assumption. Also, since the following patches will change target
code to use regcache->ptid instead of inferior_ptid, asserting that they
are the same should ensure that our changes don't have any unintended
consequences.
gdb/ChangeLog:
* target.c (target_fetch_registers, target_store_registers): Add
assert.
This patch introduces the regcache_get_ptid function, which can be used
to retrieve the ptid a regcache is connected to. It is used in
subsequent patches.
gdb/ChangeLog:
* regcache.h (regcache_get_ptid): New function.
* regcache.c (regcache_get_ptid): New function.
I noticed that backslash_in_multi_line_command_test in
gdb.base/commands.exp failed on our RHEL6 servers. I traced it to the
old version of DejaGnu (1.4.4). I have found that instead of receiving
the expected:
"print \\\nargc\n"
gdb received:
"print argc\n"
thus breaking the test and its purpose. Versionof DejaGnu < 1.5 mess
up sending "\\\n", it somehow gets replaced with a space. I found that
the following commit in DejaGnu fixed the issue:
http://git.savannah.gnu.org/cgit/dejagnu.git/commit/lib/remote.exp?id=3f39294f5cd6802858838d3bcc0ccce847ae17f2
Even though the commit is almost 10 years old, the following release of
DejaGnu was only in 2013, which is why we still have systems with the
old code.
If the DejaGnu version is < 1.5, we just skip the test.
gdb/testsuite/ChangeLog:
* gdb.base/commands.exp (backslash_in_multi_line_command_test):
Skip for versions of DejaGnu < 1.5.
The next patch will require checking the DejaGnu version. There is
already a test that does this,
gdb.threads/attach-many-short-lived-threads.exp. This patch introduces
a new procedure, dejagnu_version, and makes that test use it.
The version number is "right-padded" with zeroes, to make sure that we
always return a triplet (major, minor, patch).
The procedure does not consider the DejaGnu versions from git. For
example, if you used DejaGnu from its current master branch, the version
would be "1.6.1-git", meaning that 1.6.1 will be the next release. I
figured we'll cross that bridge when (and if) we get there.
gdb/testsuite/ChangeLog:
* lib/gdb.exp (dejagnu_version): New proc.
* gdb.threads/attach-many-short-lived-threads.exp (bad_dejagnu):
Use dejagnu_version.
While integrating the d_printing recursion guard change into gdb I
noticed we forgot to initialize the demangle_component d_printing
field in cplus_demangle_fill_{name,extended_operator,ctor,dtor}.
As is done in cplus_demangle_fill_{component,builtin_type,operator}.
It happened to work because in gcc all demangle_components were
allocated through d_make_empty. But gdb has its own allocation
mechanism (as might other users).
libiberty/ChangeLog:
* cp-demangle.c (cplus_demangle_fill_name): Initialize
demangle_component d_printing.
(cplus_demangle_fill_extended_operator): Likewise.
(cplus_demangle_fill_ctor): Likewise.
(cplus_demangle_fill_dtor): Likewise.
gdb/ChangeLog:
* cp-name-parser.y (make_empty): Initialize d_printing to zero.
If the source file is more recent than the object file, line number
information in the object may no longer match the source. So print a
warning message.
* objdump.c (update_source_path): Add abfd param. Add struct
stat vars. Pass to try_print_file_open. Warn if source is more
recent than object.
(try_print_file_open, slurp_file): Add struct stat param to
return fstat.
(show_line): Call update_source_path with bfd.
ld/
* ldlang.c (lang_check_section_addresses): Check for address space
overflow.
* testsuite/ld-checks/checks.exp (overflow_check): New procedure
* testsuite/ld-checks/over.s: New test source.
* testsuite/ld-checks/over.d: New test.
* testsuite/ld-checks/over2.s: New test source.
* testsuite/ld-checks/over2.d: New test.
First, need to match against just the CPU name, not the whole triplet.
Otherwise, the test picks up "*le-*" pattern from x86_64-apple-darwin
triplet.
Second, it should be testing for $target, not $host. Host may be
little endian by default, and the sysroot directory layout shouldn't
depend on whether it is built on LE or BE machine.
* emulparams/elf32ppccommon.sh (LIBPATH_SUFFIX): Set from target
cpu, not host.
Relative paths shouldn't have the sysroot prefix added. The patch
also makes some attempt at supporting DOS paths, and tidies code using
the new add_sysroot.
* emultempl/elf32.em (gld${EMULATION_NAME}_add_sysroot): Rewrite.
Only prefix absolute paths with sysroot. Handle DOS paths.
(gld${EMULATION_NAME}_check_ld_elf_hints): Constify variable.
(gld${EMULATION_NAME}_check_ld_so_conf): Likewise.
(gld${EMULATION_NAME}_after_open): Short-circuit NULL path
searches. Rename variable. Simplify get_runpath search.
This gcc option isn't well supported, so use the actual linker option
we want to test.
* testsuite/ld-elf/shared.exp: Use -Wl,-export-dynamic rather
than -rdynamic.
These targets use the generic ELF support, so don't handle orphans
well. The patch also updates the orphan doco to reflect this fact,
and deletes some ELF details that don't really add anything.
* ld.texinfo (Orphan Sections): Mention that not all targets
handle orphans well. Delete ELF details.
* testsuite/ld-elf/orphan-9.d: Don't run for i860 and i960.
* testsuite/ld-elf/orphan-10.d: Likewise.
For a long time now, c++/8218 has noted that GDB is printing argument types
for destructors:
(gdb) ptype A
type = class A {
public:
~A(int);
}
This happens because cp_type_print_method_args doesn't ignore artificial
arguments. [It ignores the first `this' pointer because it simply skips
the first argument for any non-static function.]
This patch fixes this:
(gdb) ptype A
type = class A {
public:
~A();
}
I've adjusted gdb.cp/templates.exp to account for this and added a new
passing regexp.
gdb/ChangeLog
PR c++/8218
* c-typeprint.c (cp_type_print_method_args): Skip artificial arguments.
gdb/testsuite/ChangeLog
PR c++/8128
* gdb.cp/templates.exp (test_ptype_of_templates): Remove argument
type from destructor regexps.
Add a branch which actually passes the test.
Adjust "ptype t5i" test names.
Given a linker script fragment like this:
SECTIONS {
. = 0x1000;
.text : AT(0x100) { *(.text) }
.data : AT(0x200) { *(.data) }
.rodata : AT(0x300) { *(.rodata) }
}
and an input file containing sections, '.text', '.data.1', and
'.rodata', then we'd expect the linker to place '.text' and '.rodata' in
the obvious way, and the '.data.1' orphan section would be located after
the '.data' section (assuming similar section properties).
Further, I believe that the expectation would be that the LMA for the
orphan '.data.1' section would start from 0x200 (as there is no '.data'
content).
However, right now, the LMA for '.data.1' would be 0x101, following on
from the '.text' section, this is because the change in LMA for the
'.data' section is not noticed by the linker, if there's no content in
the '.data' section.
What can be even more confusing to a user (though the cause is obvious
once you understand what's going on) is that adding some content to
'.data' will cause the orphan '.data.1' to switch to an LMA based off of
0x200.
This commit changes the behaviour so that an empty section that is in
the default lma region, and sets its lma, will adjust the lma of the
default region, this change will then be reflected in following sections
within the default lma memory region.
There's a new test to cover this issue that passes on a range of
targets, however, some targets generate additional sections, or have
stricter memory region size requirements that make it harder to come
up with a generic pass pattern, that still tests the required
features. For now I've set the test to ignore these targets.
ld/ChangeLog:
* ldlang.c (lang_size_sections_1): Shortcut loop only after
tracking changes to the default regions LMA.
* testsuite/ld-elf/orphan-9.ld: Extend header comment.
* testsuite/ld-elf/orphan-10.d: New file.
* testsuite/ld-elf/orphan-10.s: New file.
* NEWS: Mention change in behaviour.
When picking an lma_region for an orphan section we currently create a
new lang_output_section_statement_type and then populate this with the
orphan section.
The problem is that the lang_output_section_statement_type has a prev
pointer that links back to the previous output section. For non-orphan
output sections, that are created in linker script order, the prev
pointer will point to the output section that appears previous in linker
script order, as you'd probably expect.
The problem is that orphan sections are placed after processing the
linker script, and so, in the case of an output section created for an
orphan input section, the prev pointer actually points to the last
output section created.
This causes some unexpected behaviour when the orphan section is not
placed after the last non-orphan section that was created.
For example, consider this linker script:
MEMORY {
TEXT : ORIGIN = 0x200, LENGTH = 0x10
RODATA : ORIGIN = 0x400, LENGTH = 0x10
}
SECTIONS {
.text : {*(.text) } AT>TEXT
.data : AT(0x300) { *(.data) }
.rodata : { *(.rodata) } AT>RODATA
}
If we are processing an orphan section '.data.1' and decide to place
this after '.data', then the output section created will have a prev
pointer that references the '.rodata' output section. The result of
this is that '.data.1' will actually be assigned to the RODATA lma
region, which is probably not the expected behaviour.
The reason why '.data.1' is placed into the lma region of the '.rodata'
section is that lma region propagation is done at the time we create the
output section, based on the previous output section pointer, which is
really just a last-output-section-created pointer at that point in time,
though the prev point is fixed up later to reflect the true order of the
output sections.
The solution I propose in this commit is to move the propagation of lma
regions into a separate pass of the linker, rather than performing this
as part of the enter/exit of output sections during linker script
parsing.
During this later phase we have all of the output sections to hand, and
the prev/next points have been fixed up by this point to reflect the
actual placement ordering.
There's a new test to cover this issue that passes on a range of
targets, however, some targets generate additional sections, or have
stricter memory region size requirements that make it harder to come
up with a generic pass pattern, that still tests the required
features. For now I've set the test to ignore these targets.
ld/ChangeLog:
* ldlang.c (lang_leave_output_section_statement): Move lma_region
logic to...
(lang_propagate_lma_regions): ...this new function.
(lang_process): Call new function.
* testsuite/ld-elf/orphan-9.d: New file.
* testsuite/ld-elf/orphan-9.ld: New file.
* testsuite/ld-elf/orphan-9.s: New file.
* NEWS: Mention change in behaviour.
