Nick Alcock [Thu, 4 Jun 2020 14:38:26 +0000 (15:38 +0100)]
libctf, dump: migrate towards dumping errors rather than truncation
If we get an error emitting a single type, variable, or label, right now
we emit the error into the ctf_dprintf stream and propagate the error
all the way up the stack, causing the entire output to be silently
truncated (unless libctf debugging is on).
Instead, emit an error and keep going. (This makes sense for this use
case: if you're dumping types and a type is corrupted, you want to
know!)
Not all instances of this are fixed in this commit, only ones associated
with type formatting: more fixes will come.
libctf/
* ctf-dump.c (ctf_dump_format_type): Emit a warning.
(ctf_dump_label): Swallow errors from ctf_dump_format_type.
(ctf_dump_objts): Likewise.
(ctf_dump_var): Likewise.
(ctf_dump_type): Do not emit a duplicate message. Move to
ctf_err_warning, and swallow all errors.
Nick Alcock [Tue, 9 Jun 2020 09:45:07 +0000 (10:45 +0100)]
libctf, decl: avoid leaks of the formatted string on error
ctf_decl_sprintf builds up a formatted string in the ctf_decl_t's
cd_buf, but then on error this is hardly ever freed: we assume that
ctf_decl_fini frees it, but it leaks it instead.
Make it free it like any decent ADT should.
libctf/
* ctf-decl.c (ctf_decl_fini): Free the cd_buf.
(ctf_decl_buf): Once it escapes, don't try to free it later.
Nick Alcock [Thu, 4 Jun 2020 14:25:32 +0000 (15:25 +0100)]
libctf, types: enhance ctf_type_aname to print function arg types
Somehow this never got implemented, which makes debugging any kind of
bug that has to do with argument types fantastically confusing, because
it *looks* like the func type takes no arguments though in fact it does.
This also lets us simplify the dumper slightly (and introduces our first
uses of ctf_assert and ctf_err_warn: there will be many more).
ctf_type_aname dumps function types without including the function
pointer name itself: ctf_dump search-and-replaces it in. This seems to
give the nicest-looking results for existing users of both, even if it
is a bit fiddly.
libctf/
* ctf-types.c (ctf_type_aname): Print arg types here...
* ctf-dump.c (ctf_dump_funcs): ... not here: but do substitute
in the type name here.
Nick Alcock [Thu, 4 Jun 2020 14:07:54 +0000 (15:07 +0100)]
libctf, ld, binutils: add textual error/warning reporting for libctf
This commit adds a long-missing piece of infrastructure to libctf: the
ability to report errors and warnings using all the power of printf,
rather than being restricted to one errno value. Internally, libctf
calls ctf_err_warn() to add errors and warnings to a list: a new
iterator ctf_errwarning_next() then consumes this list one by one and
hands it to the caller, which can free it. New errors and warnings are
added until the list is consumed by the caller or the ctf_file_t is
closed, so you can dump them at intervals. The caller can of course
choose to print only those warnings it wants. (I am not sure whether we
want objdump, readelf or ld to print warnings or not: right now I'm
printing them, but maybe we only want to print errors? This entirely
depends on whether warnings are voluminous things describing e.g. the
inability to emit single types because of name clashes or something.
There are no users of this infrastructure yet, so it's hard to say.)
There is no internationalization here yet, but this at least adds a
place where internationalization can be added, to one of
ctf_errwarning_next or ctf_err_warn.
We also provide a new ctf_assert() function which uses this
infrastructure to provide non-fatal assertion failures while emitting an
assert-like string to the caller: to save space and avoid needlessly
duplicating unchanging strings, the assertion test is inlined but the
print-things-out failure case is not. All assertions in libctf will be
converted to use this machinery in future commits and propagate
assertion-failure errors up, so that the linker in particular cannot be
killed by libctf assertion failures when it could perfectly well just
print warnings and drop the CTF section.
include/
* ctf-api.h (ECTF_INTERNAL): Adjust error text.
(ctf_errwarning_next): New.
libctf/
* ctf-impl.h (ctf_assert): New.
(ctf_err_warning_t): Likewise.
(ctf_file_t) <ctf_errs_warnings>: Likewise.
(ctf_err_warn): New prototype.
(ctf_assert_fail_internal): Likewise.
* ctf-inlines.h (ctf_assert_internal): Likewise.
* ctf-open.c (ctf_file_close): Free ctf_errs_warnings.
* ctf-create.c (ctf_serialize): Copy it on serialization.
* ctf-subr.c (ctf_err_warn): New, add an error/warning.
(ctf_errwarning_next): New iterator, free and pass back
errors/warnings in succession.
* libctf.ver (ctf_errwarning_next): Add.
ld/
* ldlang.c (lang_ctf_errs_warnings): New, print CTF errors
and warnings. Assert when libctf asserts.
(lang_merge_ctf): Call it.
(land_write_ctf): Likewise.
binutils/
* objdump.c (ctf_archive_member): Print CTF errors and warnings.
* readelf.c (dump_ctf_archive_member): Likewise.
Nick Alcock [Tue, 30 Jun 2020 17:02:28 +0000 (18:02 +0100)]
libctf: pass the thunk down properly when wrapping qsort_r
When wrapping qsort_r on a system like FreeBSD on which the compar
argument comes first, we wrap the passed arg in a thunk so we can pass
down both the caller-supplied comparator function and its argument. We
should pass the *argument* down to the comparator, not the thunk, which
is basically random nonsense on the stack from the point of view of the
caller of qsort_r.
Nick Alcock [Wed, 3 Jun 2020 15:36:18 +0000 (16:36 +0100)]
libctf, next, hash: add dynhash and dynset _next iteration
This lets you iterate over dynhashes and dynsets using the _next API.
dynhashes can be iterated over in sorted order, which works by
populating an array of key/value pairs using ctf_dynhash_next itself,
then sorting it with qsort.
Convenience inline functions named ctf_dyn{hash,set}_cnext are also
provided that take (-> return) const keys and values.
libctf/
* ctf-impl.h (ctf_next_hkv_t): New, kv-pairs passed to
sorting functions.
(ctf_next_t) <u.ctn_sorted_hkv>: New, sorted kv-pairs for
ctf_dynhash_next_sorted.
<cu.ctn_h>: New, pointer to the dynhash under iteration.
<cu.ctn_s>: New, pointer to the dynset under iteration.
(ctf_hash_sort_f): Sorting function passed to...
(ctf_dynhash_next_sorted): ... this new function.
(ctf_dynhash_next): New.
(ctf_dynset_next): New.
* ctf-inlines.h (ctf_dynhash_cnext_sorted): New.
(ctf_dynhash_cnext): New.
(ctf_dynset_cnext): New.
* ctf-hash.c (ctf_dynhash_next_sorted): New.
(ctf_dynhash_next): New.
(ctf_dynset_next): New.
* ctf-util.c (ctf_next_destroy): Free the u.ctn_sorted_hkv if
needed.
(ctf_next_copy): Alloc-and-copy the u.ctn_sorted_hkv if needed.
Nick Alcock [Wed, 3 Jun 2020 14:13:24 +0000 (15:13 +0100)]
libctf, next: introduce new class of easier-to-use iterators
The libctf machinery currently only provides one way to iterate over its
data structures: ctf_*_iter functions that take a callback and an arg
and repeatedly call it.
This *works*, but if you are doing a lot of iteration it is really quite
inconvenient: you have to package up your local variables into
structures over and over again and spawn lots of little functions even
if it would be clearer in a single run of code. Look at ctf-string.c
for an extreme example of how unreadable this can get, with
three-line-long functions proliferating wildly.
The deduplicator takes this to the Nth level. It iterates over a whole
bunch of things: if we'd had to use _iter-class iterators for all of
them there would be twenty additional functions in the deduplicator
alone, for no other reason than that the iterator API requires it.
Let's do something better. strtok_r gives us half the design: generators
in a number of other languages give us the other half.
The *_next API allows you to iterate over CTF-like entities in a single
function using a normal while loop. e.g. here we are iterating over all
the types in a dict:
ctf_next_t *i = NULL;
int *hidden;
ctf_id_t id;
while ((id = ctf_type_next (fp, &i, &hidden, 1)) != CTF_ERR)
{
/* do something with 'hidden' and 'id' */
}
if (ctf_errno (fp) != ECTF_NEXT_END)
/* iteration error */
Here we are walking through the members of a struct with CTF ID
'struct_type':
while ((offset = ctf_member_next (fp, struct_type, &i, &name,
&membtype)) >= 0
{
/* do something with offset, name, and membtype */
}
if (ctf_errno (fp) != ECTF_NEXT_END)
/* iteration error */
Like every other while loop, this means you have access to all the local
variables outside the loop while inside it, with no need to tiresomely
package things up in structures, move the body of the loop into a
separate function, etc, as you would with an iterator taking a callback.
ctf_*_next allocates 'i' for you on first entry (when it must be NULL),
and frees and NULLs it and returns a _next-dependent flag value when the
iteration is over: the fp errno is set to ECTF_NEXT_END when the
iteartion ends normally. If you want to exit early, call
ctf_next_destroy on the iterator. You can copy iterators using
ctf_next_copy, which copies their current iteration position so you can
remember loop positions and go back to them later (or ctf_next_destroy
them if you don't need them after all).
Each _next function returns an always-likely-to-be-useful property of
the thing being iterated over, and takes pointers to parameters for the
others: with very few exceptions all those parameters can be NULLs if
you're not interested in them, so e.g. you can iterate over only the
offsets of members of a structure this way:
If you pass an iterator in use by one iteration function to another one,
you get the new error ECTF_NEXT_WRONGFUN back; if you try to change
ctf_file_t in mid-iteration, you get ECTF_NEXT_WRONGFP back.
Internally the ctf_next_t remembers the iteration function in use,
various sizes and increments useful for almost all iterations, then
uses unions to overlap the actual entities being iterated over to keep
ctf_next_t size down.
Iterators available in the public API so far (all tested in actual use
in the deduplicator):
/* Iterate over the members of a STRUCT or UNION, returning each member's
offset and optionally name and member type in turn. On end-of-iteration,
returns -1. */
ssize_t
ctf_member_next (ctf_file_t *fp, ctf_id_t type, ctf_next_t **it,
const char **name, ctf_id_t *membtype);
/* Iterate over the members of an enum TYPE, returning each enumerand's
NAME or NULL at end of iteration or error, and optionally passing
back the enumerand's integer VALue. */
const char *
ctf_enum_next (ctf_file_t *fp, ctf_id_t type, ctf_next_t **it,
int *val);
/* Iterate over every type in the given CTF container (not including
parents), optionally including non-user-visible types, returning
each type ID and optionally the hidden flag in turn. Returns CTF_ERR
on end of iteration or error. */
ctf_id_t
ctf_type_next (ctf_file_t *fp, ctf_next_t **it, int *flag,
int want_hidden);
/* Iterate over every variable in the given CTF container, in arbitrary
order, returning the name and type of each variable in turn. The
NAME argument is not optional. Returns CTF_ERR on end of iteration
or error. */
ctf_id_t
ctf_variable_next (ctf_file_t *fp, ctf_next_t **it, const char **name);
/* Iterate over all CTF files in an archive, returning each dict in turn as a
ctf_file_t, and NULL on error or end of iteration. It is the caller's
responsibility to close it. Parent dicts may be skipped. Regardless of
whether they are skipped or not, the caller must ctf_import the parent if
need be. */
ctf_file_t *
ctf_archive_next (const ctf_archive_t *wrapper, ctf_next_t **it,
const char **name, int skip_parent, int *errp);
ctf_label_next is prototyped but not implemented yet.
Nick Alcock [Mon, 8 Jun 2020 08:24:01 +0000 (09:24 +0100)]
libctf: add ctf_ref
This allows you to bump the refcount on a ctf_file_t, so that you can
smuggle it out of iterators which open and close the ctf_file_t for you
around the loop body (like ctf_archive_iter).
You still can't use this to preserve a ctf_file_t for longer than the
lifetime of its containing entity (e.g. ctf_archive).
Nick Alcock [Wed, 3 Jun 2020 12:23:42 +0000 (13:23 +0100)]
libctf: add ctf_forwardable_kind
The internals of the deduplicator want to know if something is a type
that can have a forward to it fairly often, often enough that inlining
it brings a noticeable performance gain. Convert the one place in
libctf that can already benefit, even though it doesn't bring any sort
of performance gain there.
libctf/
* ctf-inlines.h (ctf_forwardable_kind): New.
* ctf-create.c (ctf_add_forward): Use it.
Nick Alcock [Tue, 2 Jun 2020 21:26:38 +0000 (22:26 +0100)]
libctf, hash: introduce the ctf_dynset
There are many places in the deduplicator which use hashtables as tiny
sets: keys with no value (and usually, but not always, no freeing
function) often with only one or a few members. For each of these, even
after the last change to not store the freeing functions, we are storing
a little malloced block for each item just to track the key/value pair,
and a little malloced block for the hash table itself just to track the
freeing function because we can't use libiberty hashtab's freeing
function because we are using that to free the little malloced per-item
block.
If we only have a key, we don't need any of that: we can ditch the
per-malloced block because we don't have a value, and we can ditch the
per-hashtab structure because we don't need to independently track the
freeing functions since libiberty hashtab is doing it for us. That
means we don't need an owner field in the (now nonexistent) item block
either.
Roughly speaking, this datatype saves about 25% in time and 20% in peak
memory usage for normal links, even fairly big ones. So this might seem
redundant, but it's really worth it.
Instead of a _lookup function, a dynset has two distinct functions:
ctf_dynset_exists, which returns true or false and an optional pointer
to the set member, and ctf_dynhash_lookup_any, which is used if all
members of the set are expected to be equivalent and we just want *any*
member and we don't care which one.
There is no iterator in this set of functions, not because we don't
iterate over dynset members -- we do, a lot -- but because the iterator
here is a member of an entirely new family of much more convenient
iteration functions, introduced in the next commit.
Nick Alcock [Tue, 2 Jun 2020 21:00:14 +0000 (22:00 +0100)]
libctf, hash: save per-item space when no key/item freeing function
The libctf dynhash hashtab abstraction supports per-hashtab arbitrary
key/item freeing functions -- but it also has a constant slot type that
holds both key and value requested by the user, so it needs to use its
own freeing function to free that -- and it has nowhere to store the
freeing functions the caller requested.
So it copies them into every hash item, bloating every slot, even though
all items in a given hash table must have the same key and value freeing
functions.
So point back to the owner using a back-pointer, but don't even spend
space in the item or the hashtab allocating those freeing functions
unless necessary: if none are needed, we can simply arrange to not pass
in ctf_dynhash_item_free as a del_f to hashtab_create_alloc, and none of
those fields will ever be accessed.
The only downside is that this makes the code sensitive to the order of
fields in the ctf_helem_t and ctf_hashtab_t: but the deduplicator
allocates so many hash tables that doing this alone cuts memory usage
during deduplication by about 10%. (libiberty hashtab itself has a lot
of per-hashtab bloat: in the future we might trim that down, or make a
trimmer version.)
libctf/
* ctf-hash.c (ctf_helem_t) <key_free>: Remove.
<value_free>: Likewise.
<owner>: New.
(ctf_dynhash_item_free): Indirect through the owner.
(ctf_dynhash_create): Only pass in ctf_dynhash_item_free and
allocate space for the key_free and value_free fields fields
if necessary.
(ctf_hashtab_insert): Likewise. Fix OOM errno value.
(ctf_dynhash_insert): Only access ctf_hashtab's key_free and
value_free if they will exist. Set the slot's owner, but only
if it exists.
(ctf_dynhash_remove): Adjust.
Nick Alcock [Tue, 2 Jun 2020 20:48:12 +0000 (21:48 +0100)]
libctf, hash: improve insertion of existing keys into dynhashes
Right now, if you insert a key/value pair into a dynhash, the old slot's
key is freed and the new one always assigned. This seemed sane to me
when I wrote it, but I got it wrong time and time again. It's much
less confusing to free the key passed in: if a key-freeing function
was passed, you are asserting that the dynhash owns the key in any
case, so if you pass in a key it is always buggy to assume it sticks
around. Freeing the old key means that you can't even safely look up a
key from out of a dynhash and hold on to it, because some other matching
key might force it to be freed at any time.
In the new model, you can always get a key out of a dynhash with
ctf_dynhash_lookup_kv and hang on to it until the kv-pair is actually
deleted from the dynhash. In the old model the pointer to the key might
be freed at any time if a matching key was inserted.
libctf/
* ctf-hash.c (ctf_hashtab_insert): Free the key passed in if
there is a key-freeing function and the key already exists.
Nick Alcock [Tue, 2 Jun 2020 20:31:45 +0000 (21:31 +0100)]
libctf: add new dynhash functions
Future commits will use these.
ctf_dynhash_elements: count elements in a dynhash
ctf_dynhash_lookup_kv: look up and return pointers to the original key
and value in a dynhash (the only way of getting
a reference to the original key)
ctf_dynhash_iter_find: iterate until an item is found, then return its
key
ctf_dynhash_cinsert: insert a const key / value into a dynhash (a thim
wrapper in a new header dedicated to inline
functions).
As with the rest of ctf_dynhash, this is not public API. No impact
on existing callers is expected.
libctf/
* ctf-inlines.h: New file.
* ctf-impl.h: Include it.
(ctf_hash_iter_find_f): New typedef.
(ctf_dynhash_elements): New.
(ctf_dynhash_lookup_kv): New.
(ctf_dynhash_iter_find): New.
* ctf-hash.c (ctf_dynhash_lookup_kv): New.
(ctf_traverse_find_cb_arg_t): New.
(ctf_hashtab_traverse_find): New.
(ctf_dynhash_iter_find): New.
(ctf_dynhash_elements): New.
Nick Alcock [Tue, 2 Jun 2020 20:11:25 +0000 (21:11 +0100)]
libctf: add ctf_member_count
This returns the number of members in a struct or union, or the number
of enumerations in an enum. (This was only available before now by
iterating across every member, but it can be returned much faster than
that.)
include/
* ctf-api.h (ctf_member_count): New.
libctf/
* ctf-types.c (ctf_member_count): New.
* libctf.ver: New public function.
Nick Alcock [Tue, 2 Jun 2020 20:06:18 +0000 (21:06 +0100)]
libctf: add ctf_type_name_raw
We already have a function ctf_type_aname_raw, which returns the raw
name of a type with no decoration for structures or arrays or anything
like that: just the underlying name of whatever it is that's being
ultimately pointed at.
But this can be inconvenient to use, becauswe it always allocates new
storage for the string and copies it in, so it can potentially fail.
Add ctf_type_name_raw, which just returns the string directly out of
libctf's guts: it will live until the ctf_file_t is closed (if we later
gain the ability to remove types from writable dicts, it will live as
long as the type lives).
Nick Alcock [Tue, 2 Jun 2020 20:00:35 +0000 (21:00 +0100)]
libctf: having debugging enabled is unlikely
The deduplicator can emit enormous amounts of debugging output,
so much so that a later commit will introduce a new configure flag
that configures most of it out (and configures it out by default).
It became clear that when this configure flag is on, but debugging is
not enabled via the LIBCTF_DEBUG environment variable, up to 10% of
runtime can be spent on branch mispredictions checking the _libctf_debug
variable. Mark it unlikely to be set (when it is set, performance is
likely to be the least of your concerns).
libctf/
* ctf-subr.c (ctf_dprintf): _libctf_debug is unlikely to be set.
The archive machinery mmap()s its archives when possible: so it arranges
to do appropriately-sized unmaps by recording the unmap length in the
ctfa_magic value and unmapping that.
This brilliant (horrible) trick works less well when ctf_arc_bufopen is
called with an existing buffer (which might be a readonly mapping).
ctf_arc_bufopen always returns a ctf_archive_t wrapper, so record in
there the necessity to not unmap anything when a bufopen'ed archive is
closed again.
i.e. bitfields have to be integral types, but they can be cv-qualified
integral types or typedefs of same, etc.
This is easy to fix: do a ctf_type_resolve_unsliced() at creation time
to ensure the ultimate type is integral, and ctf_type_resolve() at
lookup time so that if you somehow have e.g. a slice of a typedef of a
slice of a cv-qualified int, we pull the encoding that the topmost slice
is based on out of the subsidiary slice (and then modify it), not out of
the underlying int. (This last bit is rather academic right now, since
all slices override exactly the same properties of the underlying type,
but it's still the right thing to do.)
libctf/
* ctf-create.c (ctf_add_slice): Support slices of any kind that
resolves to an integral type.
* ctf-types.c (ctf_type_encoding): Resolve the type before
fishing its encoding out.
Nick Alcock [Tue, 2 Jun 2020 19:38:17 +0000 (20:38 +0100)]
libctf, create: fix addition of anonymous struct/union members
A Solaris-era bug causes us to check the offsets of types with no names
against the first such type when ctf_add_type()ing members to a struct
or union. Members with no names (i.e. anonymous struct/union members)
can appear as many times as you like in a struct/union, so this check
should be skipped in this case.
Nick Alcock [Tue, 2 Jun 2020 19:28:16 +0000 (20:28 +0100)]
libctf, create: member names of "" and NULL should be the same
This matters for the case of unnamed bitfields, whose names are the null
string. These are special in that they are the only members whose
"names" are allowed to be duplicated in a single struct, but we were
only handling this for the case where name == NULL. Translate "" to
NULL to help callers.
libctf/
* ctf-create.c (ctf_add_member_offset): Support names of ""
as if they were the null pointer.
Nick Alcock [Tue, 2 Jun 2020 19:26:29 +0000 (20:26 +0100)]
libctf, open: drop unnecessary historical wart around forwards
When opening, we consider a forward with a kind above the maximum
allowable set of kinds and a forward of kind CTF_K_UNKNOWN to be a
forward to a struct. Whatever CTF version it was that produced
forwards with no associated kind, it predates anything we can read:
remove this wart.
Nick Alcock [Tue, 2 Jun 2020 19:25:02 +0000 (20:25 +0100)]
libctf, types: allow ctf_type_reference of dynamic slices
One spot was missed when we rejigged ctf_update into ctf_serialize and
allowed all operations on dynamic containers: ctf_type_reference of
slices. A dynamic slice's vlen state is stored in the dtu_slice member,
so fetch it from there.
libctf/
* ctf-types.c (ctf_type_reference): Add support for dynamic slices.
Nick Alcock [Tue, 2 Jun 2020 19:23:07 +0000 (20:23 +0100)]
libctf, create: add explicit casts for variables' and slices' types
This is technically unnecessary -- the compiler is quite capable of
doing the range reduction for us -- but it does mean that all
assignments of a ctf_id_t to its final uint32_t representation now have
appropriate explicit casts.
Nick Alcock [Tue, 2 Jun 2020 19:17:51 +0000 (20:17 +0100)]
libctf, create: do not corrupt function types' arglists at insertion time
ctf_add_function assumes that function types' arglists are of type
ctf_id_t. Since they are CTF IDs, they are 32 bits wide, a uint32_t:
unfortunately ctf_id_t is a forward-compatible user-facing 64 bits wide,
and should never ever reach the CTF storage level.
All the CTF code other than ctf_add_function correctly assumes that
function arglists outside dynamic containers are 32 bits wide, so the
serialization machinery ends up cutting off half the arglist, corrupting
all args but the first (a good sign is a bunch of args of ID 0, the
unimplemented type, popping up).
Fix this by copying the arglist into place item by item, casting it
properly, at the same time as we validate the arg types. Fix the type
of the dtu_argv in the dynamic container and drop the now-unnecessary
cast in the serializer.
libctf/
* ctf-impl.h (ctf_dtdef_t) <dtu_argv>: Fix type.
* ctf-create.c (ctf_add_function): Check for unimplemented type
and populate at the same time. Populate one-by-one, not via
memcpy.
(ctf_serialize): Remove unnecessary cast.
* ctf-types.c (ctf_func_type_info): Likewise.
(ctf_func_type_args): Likewise. Fix comment typo.
Nick Alcock [Tue, 2 Jun 2020 19:04:24 +0000 (20:04 +0100)]
libctf, create: support addition of references to the unimplemented type
The deduplicating linker adds types from the linker inputs to the output
via the same API everyone else does, so it's important that we can emit
everything that the compiler wants us to. Unfortunately, the compiler
may represent the unimplemented type (used for compiler constructs that
CTF cannot currently encode) as type zero or as a type of kind
CTF_K_UNKNOWN, and we don't allow the addition of types that cite the
former.
Adding this support adds a tiny bit of extra complexity: additions of
structure members immediately following a member of the unimplemented
type must be via ctf_add_member_offset or ctf_add_member_encoded, since
we have no idea how big members of the unimplemented type are.
(Attempts to do otherwise return -ECTF_NONREPRESENTABLE, like other
attempts to do forbidden things with the unimplemented type.)
Even slices of the unimplemented type are permitted: this is the only
case in which you can slice a type that terminates in a non-integral
type, on the grounds that it was likely integral in the source code,
it's just that we can't represent that sort of integral type properly
yet.
libctf/
* ctf-create.c (ctf_add_reftype): Support refs to type zero.
(ctf_add_array): Support array contents of type zero.
(ctf_add_function): Support arguments and return types of
type zero.
(ctf_add_typedef): Support typedefs to type zero.
(ctf_add_member_offset): Support members of type zero,
unless added at unspecified (naturally-aligned) offset.
Nick Alcock [Tue, 2 Jun 2020 18:07:08 +0000 (19:07 +0100)]
libctf: restructure error handling to reduce relocations
Jose Marchesi noted that the traditional-Unix error array in ctf-error.c
introduces one reloc per error to initialize the array: 58 so far. We
can reduce this to zero using an array of carefully-sized individual
members which is used to construct a string table, that is then
referenced by the lookup functions: but doing this automatically is a
pain.
Bruno Haible wrote suitable code years ago: I got permission to reuse it
(Bruno says "... which I hereby put in the public domain"); I modified
it a tiny bit (similarly to what Ulrich Drepper did in the dsohowto
text, but I redid it from scratch), commented it up a bit, and shifted
the error table into that form, migrating it into the new file
ctf-error.h.
This has the advantage that it spotted both typos in the text of the
errors in the comments in ctf-api.h and typos in the error defines in
the comments in ctf-error.c, and places where the two were simply not
in sync. All are now fixed.
One new constant exists in ctf-api.h: CTF_NERR, since the old method of
working out the number of errors in ctf-error.c was no longer usable,
and it seems that the number of CTF errors is something users might
reasonably want as well. It should be pretty easy to keep up to date as
new errors are introduced.
There should be no user visible changes after this commit.
gdb/ChangeLog:
* python/py-registers.c (gdbpy_register_object_data_init): Remove
redundant local variable.
(gdbpy_get_register_descriptor): Extract descriptor vector as a
reference, not pointer, update code accordingly.
Nick Clifton [Wed, 22 Jul 2020 15:07:48 +0000 (16:07 +0100)]
Fix problems in CTF handling code exposed by the Coverity static analysis tool.
readelf * readelf.c (parse_args): Silence potential warnings about a
memory resource leak when allocating space for ctf option values.
(dump_section_as_ctf): Fix typo checking dump_ctf_strtab_name
variable.
libctf * ctf-archive.c (ctf_arc_write): Avoid calling close twice on the
same file descriptor.
Simon Marchi [Wed, 22 Jul 2020 13:56:08 +0000 (15:56 +0200)]
gdb/jit: skip jit symbol lookup if already detected the symbols don't exist
To detect whether an objfile is a JITer, we lookup JIT interface
symbols in the objfile. If an objfile does not have these symbols, we
conclude that it is not a JITer. An objfile that does not have the
symbols will never have them. Therefore, once we do a lookup and find
out that the objfile does not have JIT symbols, just set a flag so
that we can skip symbol lookup for that objfile the next time we reset
JIT breakpoints.
gdb/ChangeLog:
2020-07-22 Simon Marchi <simon.marchi@polymtl.ca>
Tankut Baris Aktemur <tankut.baris.aktemur@intel.com>
* objfiles.h (struct objfile) <skip_jit_symbol_lookup>: New field.
* jit.c (jit_breakpoint_re_set_internal): Use the
`skip_jit_symbol_lookup` field.
Simon Marchi [Wed, 22 Jul 2020 13:56:08 +0000 (15:56 +0200)]
gdb/jit: apply minor cleanup and modernization
gdb/ChangeLog:
2020-07-22 Simon Marchi <simon.marchi@polymtl.ca>
Tankut Baris Aktemur <tankut.baris.aktemur@intel.com>
* jit.c (jit_read_descriptor): Define the descriptor address once,
use twice.
(jit_breakpoint_deleted): Move the declaration of the loop variable
`iter` into the loop header.
(jit_breakpoint_re_set_internal): Move the declaration of the local
variable `objf_data` to the first point of definition.
(jit_event_handler): Move the declaration of local variables
`code_entry`, `entry_addr`, and `objf` to their first point of use.
Rename `objf` to `jited`.
GDB's JIT handler stores an objfile (and data associated with it) per
program space to keep track of JIT breakpoint information. This assumes
that there is at most one JITer objfile in the program space. However,
there may be multiple. If so, only the first JITer's hook breakpoints
would be realized and the JIT events from the other JITers would be
missed.
This patch removes that assumption, allowing an arbitrary number of
objfiles within a program space to be JITers.
- The "unique" program_space -> JITer objfile pointer in
jit_program_space_data is removed. In fact, jit_program_space_data
becomes empty, so it is removed entirely.
- jit_breakpoint_deleted is modified, it now has to assume that any
objfile in a program space is a potential JITer. It now iterates on
all objfiles, checking if they are indeed JITers, and if they are,
whether the deleted breakpoint belongs to them.
- jit_breakpoint_re_set_internal also has to assume that any objfile in
a program space is a potential JITer. It creates (or updates) one
jiter_objfile_data structure for each JITer it finds.
- Same for jit_inferior_init. It now iterates all objfiles to read the
initial JIT object list.
gdb/ChangeLog:
2020-07-22 Tankut Baris Aktemur <tankut.baris.aktemur@intel.com>
Simon Marchi <simon.marchi@polymtl.ca>
* jit.c (struct jit_program_space_data): Remove.
(jit_program_space_key): Remove.
(jiter_objfile_data::~jiter_objfile_data): Remove program space
stuff.
(get_jit_program_space_data): Remove.
(jit_breakpoint_deleted): Iterate on all of the program space's
objfiles.
(jit_inferior_init): Likewise.
(jit_breakpoint_re_set_internal): Likewise. Also change return
type to void.
(jit_breakpoint_re_set): Pass current_program_space to
jit_breakpoint_re_set_internal.
gdb/testsuite/ChangeLog:
2020-07-22 Tankut Baris Aktemur <tankut.baris.aktemur@intel.com>
* gdb.base/jit-reader-simple.exp: Add a scenario for a binary that
loads two JITers.
Simon Marchi [Wed, 22 Jul 2020 13:56:07 +0000 (15:56 +0200)]
gdb/jit: move cached_code_address and jit_breakpoint to jiter_objfile_data
This is in preparation for allowing more than one JITer objfile per
program space. Once we do that, each JITer objfile will have its own
JIT breakpoint (on the __jit_debug_register_code function it provides).
The cached_code_address field is just the runtime / relocated address of
that symbol.
Since they are going to become JITer-objfile-specific and not
program-space-specific, move these fields from jit_program_space_data to
jiter_objfile_data.
gdb/ChangeLog:
2020-07-22 Simon Marchi <simon.marchi@polymtl.ca>
* jit.h (struct jiter_objfile_data) <cached_code_address,
jit_breakpoint>: Move to here from ...
* jit.c (jit_program_space_data): ... here.
(jiter_objfile_data::~jiter_objfile_data): Update.
(jit_breakpoint_deleted): Update.
(jit_breakpoint_re_set_internal): Update.
Simon Marchi [Wed, 22 Jul 2020 13:56:07 +0000 (15:56 +0200)]
gdb/jit: apply some simplifications and assertions
Following patch "gdb/jit: split jit_objfile_data in two", there are some
simplifications we can make. The invariants described there mean that
we can assume / assert some things instead of checking them using
conditionals.
If an instance of jiter_objfile_data exists for a given objfile, it's
because the required JIT interface symbols were found. Therefore, in
~jiter_objfile_data, the `register_code` field can't be NULL. It was
previously used to differentiate a jit_objfile_data object used for a
JITer vs a JITed. We can remove that check.
If an instance of jiter_objfile_data exists for a given objfile, it's
because it's the sole JITer objfile in the scope of its program space
(jit_program_space_data::objfile points to it). At the moment,
jit_breakpoint_re_set_internal won't create a second instance of
jiter_objfile_data for a given program space. Therefore, it's not
necessary to check for `ps_data != NULL` in ~jiter_objfile_data: we know
a jit_program_space_data for that program space exists. We also don't
need to check for `ps_data->objfile == this->objfile`, because we know
the objfile is the sole JITer in this program space. Replace these two
conditions with assertions.
A pre-condition for calling the jit_read_descriptor function (which is
respected in the two call sites) is that the objfile `jiter` _is_ a
JITer - it already has a jiter_objfile_data attached to it. When a
jiter_objfile_data exists, its `descriptor` field is necessarily set:
had the descriptor symbol not been found, jit_breakpoint_re_set_internal
would not have created the jiter_objfile_data. Remove the check and
early return in jit_read_descriptor. Access objfile's `jiter_data` field
directly instead of calling `get_jiter_objfile_data` (which creates the
jiter_objfile_data if it doesn't exist yet) and assert that the result
is not nullptr.
Finally, `jit_event_handler` is always passed a JITer objfile. So, add
an assertion to ensure that.
gdb/ChangeLog:
2020-07-22 Simon Marchi <simon.marchi@polymtl.ca>
* jit.c (jiter_objfile_data::~jiter_objfile_data): Remove some
checks.
(jit_read_descriptor): Remove NULL check.
(jit_event_handler): Add an assertion.
Simon Marchi [Wed, 22 Jul 2020 13:56:07 +0000 (15:56 +0200)]
gdb/jit: split jit_objfile_data in two
The jit_objfile_data is currently used to hold information about both
objfiles that are the result of JIT compilation (JITed) and objfiles
that can produce JITed objfiles (JITers). I think that this double use
of the type is confusing, and that things would be more obvious if we
had one type for each role.
This patch splits it into:
- jited_objfile_data: for data about an objfile that is the result of a
JIT compilation
- jiter_objfile_data: for data about an objfile which produces JITed
objfiles
There are now two JIT-related fields in an objfile, one for each kind.
With this change, the following invariants hold:
- an objfile has a non-null `jiter_data` field iff it defines the required
symbols of the JIT interface
- an objfile has a non-null `jited_data` field iff it is the product of
JIT compilation (has been produced by some JITer)
gdb/ChangeLog:
2020-07-22 Simon Marchi <simon.marchi@polymtl.ca>
* jit.h (struct jit_objfile_data): Split into...
(struct jiter_objfile_data): ... this ...
(struct jited_objfile_data): ... and this.
* objfiles.h (struct objfile) <jit_data>: Remove.
<jiter_data, jited_data>: New fields.
* jit.c (jit_objfile_data::~jit_objfile_data): Rename to ...
(jiter_objfile_data::~jiter_objfile_data): ... this.
(get_jit_objfile_data): Rename to ...
(get_jiter_objfile_data): ... this.
(add_objfile_entry): Update.
(jit_read_descriptor): Use get_jiter_objfile_data.
(jit_find_objf_with_entry_addr): Use objfile's jited_data field.
(jit_breakpoint_re_set_internal): Use get_jiter_objfile_data.
(jit_inferior_exit_hook): Use objfile's jited_data field.
Simon Marchi [Wed, 22 Jul 2020 13:56:06 +0000 (15:56 +0200)]
gdb/jit: link to jit_objfile_data directly from the objfile struct
Remove the use of objfile_data to associate a jit_objfile_data with an
objfile. Instead, directly link to a jit_objfile_data from an objfile
struct. The goal is to eliminate unnecessary abstraction.
The free_objfile_data function naturally becomes the destructor of
jit_objfile_data. However, free_objfile_data accesses the objfile to
which the data is attached, which the destructor of jit_objfile_data
doesn't have access to. To work around this, add a backlink to the
owning objfile in jit_objfile_data. This is however temporary, it goes
away in a subsequent patch.
gdb/ChangeLog:
2020-07-22 Simon Marchi <simon.marchi@polymtl.ca>
* jit.h: Forward-declare `struct minimal_symbol`.
(struct jit_objfile_data): Migrate to here from jit.c; also add a
constructor, destructor, and an objfile* field.
* jit.c (jit_objfile_data): Remove.
(struct jit_objfile_data): Migrate from here to jit.h.
(jit_objfile_data::~jit_objfile_data): New destructor
implementation with code moved from free_objfile_data.
(free_objfile_data): Delete.
(get_jit_objfile_data): Update to use the jit_data field of objfile.
(jit_find_objf_with_entry_addr): Ditto.
(jit_inferior_exit_hook): Ditto.
(_initialize_jit): Remove the call to
register_objfile_data_with_cleanup.
* objfiles.h (struct objfile) <jit_data>: New field.
gdb/jit: pass the jiter objfile as an argument to jit_event_handler
This is a refactoring that adds a new parameter to the `jit_event_handler`
function: the JITer objfile. The goal is to distinguish which JITer
triggered the JIT event, in case there are multiple JITers -- a capability
that is added in a subsequent patch.
gdb/ChangeLog:
2020-07-22 Tankut Baris Aktemur <tankut.baris.aktemur@intel.com>
* jit.h: Forward-declare `struct objfile`.
(jit_event_handler): Add a second parameter, the JITer objfile.
* jit.c (jit_read_descriptor): Change the signature to take the
JITer objfile as an argument instead of the jit_program_space_data.
(jit_inferior_init): Update the call to jit_read_descriptor.
(jit_event_handler): Use the new JITer objfile argument when calling
jit_read_descriptor.
* breakpoint.c (handle_jit_event): Update the call to
jit_event_handler to pass the JITer objfile.
MIPS/GAS/testsuite: Fix JALR relocation tests for IRIX targets
With IRIX targets the JALR hint relocation is not produced for the o32
ABI, where it is considered a GNU extension. Consequently several tests
fail as the output produced by GAS fails to match patterns expecting the
relocation to be present where appropriate, even though output produced
is indeed correct.
As the absence of the relocation is expected, fix the tests by providing
respective alternative dump patterns with any JALR relocations removed,
removing numerous failures with `*-*-irix*' targets:
Strictly speaking no MIPSr6 or microMIPS target is supported by IRIX,
but GAS supports such configurations on the basis of uniformity, so
provide the relevant patterns too rather than excluding the combinations
from testing.
gas/
* testsuite/gas/mips/jal-svr4pic-irix.d: New file.
* testsuite/gas/mips/mips1@jal-svr4pic-irix.d: New file.
* testsuite/gas/mips/mipsr6@jal-svr4pic-irix.d: New file.
* testsuite/gas/mips/micromips@jal-svr4pic-irix.d: New file.
* testsuite/gas/mips/r3000@jal-svr4pic-irix.d: New file.
* testsuite/gas/mips/jal-svr4pic-local-irix.d: New file.
* testsuite/gas/mips/mips1@jal-svr4pic-local-irix.d: New file.
* testsuite/gas/mips/micromips@jal-svr4pic-local-irix.d: New
file.
* testsuite/gas/mips/r3000@jal-svr4pic-local-irix.d: New file.
* testsuite/gas/mips/jal-svr4pic-noreorder-irix.d: New file.
* testsuite/gas/mips/mips1@jal-svr4pic-noreorder-irix.d: New
file.
* testsuite/gas/mips/mipsr6@jal-svr4pic-noreorder-irix.d: New
file.
* testsuite/gas/mips/micromips@jal-svr4pic-noreorder-irix.d: New
file.
* testsuite/gas/mips/r3000@jal-svr4pic-noreorder-irix.d: New
file.
* testsuite/gas/mips/jal-xgot-irix.d: New file.
* testsuite/gas/mips/jalr2-irix.d: New file.
* testsuite/gas/mips/micromips-branch-relax-insn32-pic-irix.d:
New file.
* testsuite/gas/mips/micromips-branch-relax-pic-irix.d: New
file.
* testsuite/gas/mips/mips-abi32-pic2-irix.d: New file.
* testsuite/gas/mips/jal-svr4pic-local.d: Don't exclude
`*-*-irix*' targets. Add source file designator.
* testsuite/gas/mips/mips1@jal-svr4pic-local.d: Don't exclude
`*-*-irix*' targets.
* testsuite/gas/mips/r3000@jal-svr4pic-local.d: Likewise.
* testsuite/gas/mips/micromips@jal-svr4pic-local.d: Likewise.
* testsuite/gas/mips/jalr2.d: Add name designator.
* testsuite/gas/mips/mips.exp: Use respective IRIX variants for
tests involving the JALR relocation throughout.
Pedro Alves [Wed, 22 Jul 2020 11:32:53 +0000 (12:32 +0100)]
gdbserver: handle running threads in qXfer:threads:read
On some systems, the gdb.multi/multi-target.exp testcase occasionally
fails like so:
Running src/gdb/testsuite/gdb.multi/multi-target.exp ...
FAIL: gdb.multi/multi-target.exp: info-inferiors: multi_process=on: inferior 1: info connections
FAIL: gdb.multi/multi-target.exp: info-inferiors: multi_process=on: inferior 1: info inferiors
FAIL: gdb.multi/multi-target.exp: info-inferiors: multi_process=on: inferior 2: info connections
FAIL: gdb.multi/multi-target.exp: info-inferiors: multi_process=on: inferior 2: info inferiors
FAIL: gdb.multi/multi-target.exp: info-inferiors: multi_process=on: inferior 3: inferior 3
... many more cascading fails.
The problem starts when the testcase runs an inferior against GDBserver:
(gdb) run
Starting program: build/gdb/testsuite/outputs/gdb.multi/multi-target/multi-target
Reading /lib64/ld-linux-x86-64.so.2 from remote target...
warning: File transfers from remote targets can be slow. Use "set sysroot" to access files locally instead.
Reading /lib64/ld-linux-x86-64.so.2 from remote target...
Reading /lib64/ld-2.31.so from remote target...
Reading /lib64/.debug/ld-2.31.so from remote target...
Reading /usr/lib/debug//lib64/ld-2.31.so from remote target...
Reading /usr/lib/debug/lib64//ld-2.31.so from remote target...
Reading target:/usr/lib/debug/lib64//ld-2.31.so from remote target...
Reading /lib/x86_64-linux-gnu/libpthread.so.0 from remote target...
Reading /lib/x86_64-linux-gnu/libc.so.6 from remote target...
Reading /lib/x86_64-linux-gnu/libc-2.31.so from remote target...
Reading /lib/x86_64-linux-gnu/.debug/libc-2.31.so from remote target...
Reading /usr/lib/debug//lib/x86_64-linux-gnu/libc-2.31.so from remote target...
Reading /usr/lib/debug//lib/x86_64-linux-gnu/libc-2.31.so from remote target...
Remote connection closed
...
Note the "Remote connection closed" message. That means GDBserver
exited abruptly.
I traced it down to the fact that GDB fetches the thread list from
GDBserver while the main thread of the process is still running. On
my main system where I wrote the testcase, I have not observed the
failure because it is slow enough that the thread stops before
GDBserver fetches the thread list in the problem scenario which I'll
describe below.
With some --remote-debug logging from GDBserver side, we see the last
packets before the connection closes:
Note the vCont;c , which sets the program running, and then a
qXfer:threads:read packet at the end.
The problem happens when the thread list refresh (qXfer:threads:read)
is sent just while the main thread is running and it still hasn't
initialized its libpthread id internally. In that state, the main
thread's lwp will remain with the thread_known flag clear. See in
find_one_thread:
/* If the new thread ID is zero, a final thread ID will be available
later. Do not enable thread debugging yet. */
if (ti.ti_tid == 0)
return 0;
Now, back in server.cc, to handle the qXfer:threads:read, we reach
handle_qxfer_threads -> handle_qxfer_threads_proper, and the latter
then calls handle_qxfer_threads_worker for each known thread. In
handle_qxfer_threads_worker, we call target_thread_handle. This ends
up in thread_db_thread_handle, here:
if (!lwp->thread_known && !find_one_thread (thread->id))
return false;
Since the thread ID isn't known yet, we call find_one_thread. This
calls into libthread_db.so, which accesses memory. Because the
current thread is running, that fails and we throw an error, here:
/* Get information about this thread. */
err = thread_db->td_ta_map_lwp2thr_p (thread_db->thread_agent, lwpid, &th);
if (err != TD_OK)
error ("Cannot get thread handle for LWP %d: %s",
lwpid, thread_db_err_str (err));
The current design is that whenever GDB-facing packets/requests need
to accesses memory, server.cc is supposed to prepare the target for
the access. See gdb_read_memory / gdb_write_memory. This preparation
means pausing threads if in non-stop mode (someday we could lift this
requirement, but we will still need to pause to access registers or do
other related ptrace accesses like PTRACE_GET_THREAD_AREA). Note that
the multi-target.exp testcase forces "maint set target-non-stop on".
So the fix here is to prepare the target to access memory when
handling qXfer:threads:read too.
gdbserver/ChangeLog:
* inferiors.cc (switch_to_process): New, moved here from
thread-db.cc, and made extern.
* inferiors.h (switch_to_process): Declare.
* server.cc: Include "gdbsupport/scoped_restore.h".
(handle_qxfer_threads_proper): Now returns bool. Prepare to
access memory around target calls.
(handle_qxfer_threads): Handle errors.
* thread-db.cc (switch_to_process): Moved to inferiors.cc.
PR ld/26262
PR ld/26267
* ldlang.c (lang_process): Set lto_all_symbols_read after all
LTO IR symbols have been read.
* plugin.c (plugin_notice): Override the IR definition only if
all LTO IR symbols have been read or the new definition is
non-weak and the the IR definition is weak
* testsuite/ld-plugin/lto.exp: Run PR ld/26262 and ld/26267
tests.
* testsuite/ld-plugin/pr26262a.c: New file.
* testsuite/ld-plugin/pr26262b.c: Likewise.
* testsuite/ld-plugin/pr26262c.c: Likewise.
* testsuite/ld-plugin/pr26267.err: Likewise.
* testsuite/ld-plugin/pr26267a.c: Likewise.
* testsuite/ld-plugin/pr26267b.c: Likewise.
* testsuite/ld-plugin/pr26267c.c: Likewise.
Max Filippov [Wed, 22 Jul 2020 09:17:38 +0000 (02:17 -0700)]
bfd: xtensa: pr26246: fix removed_literal_compare
2020-07-22 Max Filippov <jcmvbkbc@gmail.com>
bfd/
PR 26246
* elf32-xtensa.c (removed_literal_compare): Use correct pointer
type for the first function argument. Rename pointers to reflect
that they have distinct types.
John Baldwin [Wed, 22 Jul 2020 00:28:16 +0000 (17:28 -0700)]
Migrate the sparc64 ADI handle_segmentation_fault hook to report_signal_info.
gdb/ChangeLog:
* sparc64-linux-tdep.c (sparc64_linux_handle_segmentation_fault):
Rename to sparc64_linux_report_signal_info and add siggnal
argument.
(sparc64_linux_init_abi): Use sparc64_linux_report_signal_info
instead of sparc64_linux_handle_segmentation_fault.
John Baldwin [Wed, 22 Jul 2020 00:28:16 +0000 (17:28 -0700)]
Migrate the x86 MPX handle_segmentation_fault hook to report_signal_info.
gdb/ChangeLog:
* amd64-linux-tdep.c (amd64_linux_init_abi_common): Use
i386_linux_report_signal_info instead of
i386_linux_handle_segmentation_fault.
* i386-linux-tdep.c (i386_linux_handle_segmentation_fault): Rename
to i386_linux_report_signal_info and add siggnal argument.
(i386_linux_init_abi): Use i386_linux_report_signal_info instead
of i386_linux_handle_segmentation_fault.
* i386-linux-tdep.h (i386_linux_handle_segmentation_fault): Rename
to i386_linux_report_signal_info and add siggnal argument.
John Baldwin [Wed, 22 Jul 2020 00:28:16 +0000 (17:28 -0700)]
Report architecture-specific signal information for core files.
When opening a core file, if the process terminated due to a signal,
invoke the gdbarch report_signal_info hook to report
architecture-specific information about the signal.
gdb/ChangeLog:
* corelow.c (core_target_open): Invoke gdbarch report_signal_info
hook if present.
John Baldwin [Wed, 22 Jul 2020 00:28:16 +0000 (17:28 -0700)]
Add a new gdbarch hook to report additional signal information.
This is a more general version of the existing handle_segmentation_fault
hook that is able to report information for an arbitrary signal, not
just SIGSEGV.
gdb/ChangeLog:
* gdbarch.c: Regenerate.
* gdbarch.h: Regenerate.
* gdbarch.sh (report_signal_info): New method.
* infrun.c (print_signal_received_reason): Invoke gdbarch
report_signal_info hook if present.
Andrew Burgess [Tue, 7 Jul 2020 14:26:42 +0000 (15:26 +0100)]
gdb/python: Reuse gdb.RegisterGroup objects where possible
Only create one gdb.RegisterGroup Python object for each of GDB's
reggroup objects.
I could have added a field into the reggroup object to hold the Python
object pointer for each reggroup, however, as reggroups are never
deleted within GDB, and are global (not per-architecture) a simpler
solution seemed to be just to hold a single global map from reggroup
pointer to a Python object representing the reggroup. Then we can
reuse the objects out of this map.
After this commit it is possible for a user to tell that two
gdb.RegisterGroup objects are now identical when previously they were
unique, however, as both these objects are read-only I don't think
this should be a problem.
There should be no other user visible changes after this commit.
gdb/ChangeLog:
* python/py-registers.c : Add 'unordered_map' include.
(gdbpy_new_reggroup): Renamed to...
(gdbpy_get_reggroup): ...this. Update to only create register
group descriptors when needed.
(gdbpy_reggroup_iter_next): Update.
Andrew Burgess [Tue, 7 Jul 2020 14:00:30 +0000 (15:00 +0100)]
gdb/python: Reuse gdb.RegisterDescriptor objects where possible
Instead of having the gdb.RegisterDescriptorIterator creating new
gdb.RegisterDescriptor objects for each regnum, instead cache
gdb.RegisterDescriptor objects on the gdbarch object and reuse these.
This means that for every gdbarch/regnum pair there is a single unique
gdb.RegisterDescriptor, this feels like a neater implementation than
the existing one.
It is possible for a user to see (in Python code) that the descriptors
are now identical, but as the descriptors are read-only this should
make no real difference.
There should be no other user visible changes.
gdb/ChangeLog:
* python/py-registers.c (gdbpy_register_object_data): New static
global.
(gdbpy_register_object_data_init): New function.
(gdbpy_new_register_descriptor): Renamed to...
(gdbpy_get_register_descriptor): ...this, and update to reuse
existing register descriptors where possible.
(gdbpy_register_descriptor_iter_next): Update.
(gdbpy_initialize_registers): Register new gdbarch data.
Simon Marchi [Tue, 21 Jul 2020 19:12:56 +0000 (15:12 -0400)]
gdb: handle undefined properties in ada_discrete_type_{low,high}_bound
This patch fixes a failure in test `gdb.ada/access_to_packed_array.exp`.
The failure was introduced by 8c2e4e0689ea24 ("gdb: add accessors to
struct dynamic_prop"), but I think it in fact exposed a latent buglet.
Note that to reproduce it, I had to use AdaCore's Ada "distribution"
[1]. The one that comes with my distro doesn't have debug info for the
standard library stuff, so the bug wouldn't trigger.
The bug is that while executing the `maint print symbols` command, we
are accessing the value of a range type's high bound dynamic prop as a
"const" value (PROP_CONST), when it is actually undefined
(PROP_UNDEFINED). It results in this failed assertion:
`ada_discrete_type_high_bound` calls `resolve_dynamic_type`, which
eventually calls `resolve_dynamic_range`. This one is responsible for
evaluating a range type's dynamic bounds in the current context and
returning static values. It returns a new range type with these static
bounds.
The resulting bounds are typically properties of the PROP_CONST kind.
But when it's not possible to evaluate the properties, the properties
are PROP_UNDEFINED. In the case we are looking at, it's not possible to
evaluate the dynamic high bound, which is of type PROP_LOCLIST. It
would require a target with registers and a frame, but we run `maint
print symbols` without a live process.
`ada_discrete_type_high_bound` then accesses the high bound
unconditionally as a const value, which triggers the assert.
Note that the previous code in resolve_dynamic_range (before commit 8c2e4e0689ea24) did this:
That did not really made sense, setting the kind to `PROP_UNDEFINED` but
also setting the `const_val` field. The `const_val` field is only
meaningful if the kind if `PROP_CONST`. The new code
(post-8c2e4e0689ea24) simply calls `set_undefined ()`.
Fix this by making the caller, `ada_discrete_type_high_bound`, consider
that a range high bound could be of kind `PROP_UNDEFINED`, and return
0 in this case. I made the same change in ada_discrete_type_low_bound.
I didn't encounter a problem with this function, but the same could in
theory happen there.
Returning 0 here is kind of a lie, but the goal here is just to restore
the behavior of pre-8c2e4e0689ea24.
The output of `maint print symbols` is:
typedef <ada__exceptions__exception_data__append_info_basic_exception_information__TTnameSP1: range 1 .. 0;
record
ada__exceptions__exception_data__append_info_basic_exception_information__TTnameSP1: range 1 .. 0;
end record;
Instead of `1 .. 0`, which does not make sense, we could say something
like `1 .. <dynamic>`. But that would require more changes than I'm
willing to do at the moment.
Tom de Vries [Tue, 21 Jul 2020 14:22:34 +0000 (16:22 +0200)]
[gdb/testsuite] Fix gdb.reverse/solib-{precsave,reverse}.exp with gcc-8
With gcc-8, we have the following FAILs, which are not there for gcc-7:
...
FAIL: gdb.reverse/solib-precsave.exp: reverse-step into solib function one
FAIL: gdb.reverse/solib-precsave.exp: reverse-step within solib function one
FAIL: gdb.reverse/solib-precsave.exp: reverse-step back to main one
FAIL: gdb.reverse/solib-precsave.exp: reverse-step into solib function two
FAIL: gdb.reverse/solib-precsave.exp: reverse-step within solib function two
FAIL: gdb.reverse/solib-precsave.exp: reverse-step back to main two
FAIL: gdb.reverse/solib-precsave.exp: run until end part two
FAIL: gdb.reverse/solib-precsave.exp: reverse-next over solib function one
FAIL: gdb.reverse/solib-reverse.exp: reverse-step into solib function one
FAIL: gdb.reverse/solib-reverse.exp: reverse-step within solib function one
FAIL: gdb.reverse/solib-reverse.exp: reverse-step back to main one
FAIL: gdb.reverse/solib-reverse.exp: reverse-step into solib function two
FAIL: gdb.reverse/solib-reverse.exp: reverse-step within solib function two
FAIL: gdb.reverse/solib-reverse.exp: reverse-step back to main two
FAIL: gdb.reverse/solib-reverse.exp: run until end part two
FAIL: gdb.reverse/solib-reverse.exp: reverse-next over solib function one
...
Looking at the first FAIL for gdb.reverse/solib-precsave.exp, we have:
...
(gdb) PASS: reverse-next first shr1
reverse-next^M
40 b[0] = 6; b[1] = 9; /* generic statement, end part two */^M
(gdb) PASS: reverse-next generic
reverse-step^M
-shr2 (x=17) at gdb.reverse/shr2.c:23^M
-23 }^M
-(gdb) PASS: reverse-step into solib function one
+38 b[1] = shr2(17); /* middle part two */^M
+(gdb) FAIL: reverse-step into solib function one
...
There's a difference in line number info for line 38, where for gcc-7 we have:
...
Line number Starting address View Stmt
38 0x4005c6 x
...
and for gcc-8:
...
38 0x4005c1 x
38 0x4005cb x
...
which explains why we don't step directly into "solib function one".
Fix this by recognizing the extra "recommended breakpoint location" and
issuing an additional reverse-next/step.
Tom de Vries [Tue, 21 Jul 2020 13:27:18 +0000 (15:27 +0200)]
[gdb/testsuite] Fix step-reverse.c with gcc-10
The file gdb.reverse/step-reverse.c is used in test-cases:
- gdb.reverse/step-reverse.exp
- gdb.reverse/next-reverse-bkpt-over-sr.exp
- gdb.reverse/step-precsave.exp
With gcc-7, there are only PASSes (apart from one KFAIL), but with gcc-10, we
have the following FAILs:
...
FAIL: gdb.reverse/step-reverse.exp: reverse stepi from a function call \
(start statement)
FAIL: gdb.reverse/step-reverse.exp: simple reverse stepi
FAIL: gdb.reverse/step-reverse.exp: reverse step out of called fn
FAIL: gdb.reverse/step-reverse.exp: reverse next over call
FAIL: gdb.reverse/step-reverse.exp: reverse step test 1
FAIL: gdb.reverse/step-reverse.exp: reverse next test 1
FAIL: gdb.reverse/step-reverse.exp: reverse step test 2
FAIL: gdb.reverse/step-reverse.exp: reverse next test 2
FAIL: gdb.reverse/step-precsave.exp: reverse stepi from a function call \
(start statement)
FAIL: gdb.reverse/step-precsave.exp: simple reverse stepi
FAIL: gdb.reverse/step-precsave.exp: reverse step out of called fn
FAIL: gdb.reverse/step-precsave.exp: reverse next over call
FAIL: gdb.reverse/step-precsave.exp: reverse step test 1
FAIL: gdb.reverse/step-precsave.exp: reverse next test 1
FAIL: gdb.reverse/step-precsave.exp: reverse step test 2
FAIL: gdb.reverse/step-precsave.exp: reverse next test 2
...
Looking at the first step-precsave.exp FAIL, we have:
...
(gdb) stepi^M
26 myglob++; return 0; /* ARRIVED IN CALLEE */^M
(gdb) PASS: gdb.reverse/step-precsave.exp: reverse stepi thru function return
stepi^M
0x000000000040055f 26 myglob++; return 0; /* ARRIVED IN CALLEE */^M
(gdb) FAIL: gdb.reverse/step-precsave.exp: reverse stepi from a function call \
(start statement)
...
There's a difference in line info for callee:
...
25 int callee() { /* ENTER CALLEE */
26 myglob++; return 0; /* ARRIVED IN CALLEE */
27 } /* RETURN FROM CALLEE */
...
between gcc-7:
...
Line number Starting address View Stmt
25 0x400557 x
26 0x40055b x
27 0x40056f x
...
and gcc-10:
...
25 0x400552 x
26 0x400556 x
26 0x400565 x
27 0x40056a x
...
The two "recommend breakpoint location" entries at line 26 are for the two
statements ("myglob++" and "return 0"), but the test-case expects to hit line
26 only once.
Fix this by rewriting the two statements into a single statement:
...
- myglob++; return 0; /* ARRIVED IN CALLEE */
+ return myglob++; /* ARRIVED IN CALLEE */
...
Jan Beulich [Tue, 21 Jul 2020 12:20:11 +0000 (14:20 +0200)]
Revert "x86: Don't display eiz with no scale"
This reverts commit 04c662e2b66bedd050f97adec19afe0fcfce9ea7.
In my underlying suggestion I neglected the fact that in those
cases (,%eiz,1) is the only visible indication that 32-bit
addressing is in effect.
Tom de Vries [Tue, 21 Jul 2020 09:37:17 +0000 (11:37 +0200)]
[gdb/testsuite] Fix gdb.fortran/info-modules.exp with gcc-8
When using test-case gdb.fortran/info-modules.exp with gcc 8.4.0, I run into:
...
FAIL: gdb.fortran/info-modules.exp: info module variables: check for entry \
'info-types.f90', '35', 'Type m1t1 mod1::__def_init_mod1_M1t1;'
FAIL: gdb.fortran/info-modules.exp: info module variables: check for entry \
'info-types.f90', '35', 'Type __vtype_mod1_M1t1 mod1::__vtab_mod1_M1t1;'
...
This is caused by this change in gdb output:
...
(gdb) info module variables
...
File gdb.fortran/info-types.f90:
-35: Type m1t1 mod1::__def_init_mod1_M1t1;
+ Type m1t1 mod1::__def_init_mod1_M1t1;
-35: Type __vtype_mod1_M1t1 mod1::__vtab_mod1_M1t1;
+ Type __vtype_mod1_M1t1 mod1::__vtab_mod1_M1t1;
21: real(kind=4) mod1::mod1_var_1;
22: integer(kind=4) mod1::mod1_var_2;
...
caused by a change in debug info.
Fix this by allowing those entries without line number.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2020-07-21 Tom de Vries <tdevries@suse.de>
* gdb.fortran/info-modules.exp (info module variables): Allow missing
line numbers for some variables.
Jan Beulich [Tue, 21 Jul 2020 09:34:40 +0000 (11:34 +0200)]
Revert "x86: Replace evex-no-scale.s with evex-no-scale-[32|64].s"
This reverts commit 19449d7c67690c641b1ec9c13ff3531677a5afcc, addressing
the issue that was run into back then: There was no relationship to i686-*
and/or cross builds on 64-bit hosts. The sole problem was the use of / as
as comment character in certain ELF targets. Instead of division, use a
comparison operation.
At the same time also revert the ELF related part of 99c2d522f7a7 ("x86:
Update assembler tests for non-ELF targets") by replacing the construct
that's problematic for non-ELF, and by adding the "#pass" patterns to
the expected output files to cover for the tail padding generated into
COFF output.
Tom de Vries [Tue, 21 Jul 2020 07:40:21 +0000 (09:40 +0200)]
[gdb/testsuite] Don't leak env vars in gdb.debuginfod/fetch_src_and_symbols.exp
Test-case gdb.debuginfod/fetch_src_and_symbols.exp leaks env vars
DEBUGINFOD_URLS, DEBUGINFOD_TIMEOUT and DEBUGINFOD_CACHE_PATH, causing
timeouts in subsequent tests.
Fix this by using save_vars. Also, fix PATH and DUPLICATE errors. Finally,
cleanup whitespace.
gdb/testsuite/ChangeLog:
2020-07-21 Tom de Vries <tdevries@suse.de>
* gdb.debuginfod/fetch_src_and_symbols.exp: Use save_vars for env
vars. Fix PATH and DUPLICATE errors. Cleanup whitespace.
MIPS/GAS: Remove stale `prev_reloc_op_frag' variable
Ever since commit 4d7206a284ee ("Rework MIPS macro relaxation, fix string
merging bug"), <https://sourceware.org/ml/binutils/2004-01/msg00248.html>,
`prev_reloc_op_frag' has only been set and never used. Remove it then.
John Baldwin [Mon, 20 Jul 2020 16:09:58 +0000 (09:09 -0700)]
Implement the skip_solib_resolver gdbarch hook for FreeBSD architectures.
The ELF runtime linker on all FreeBSD architectures uses the
"_rtld_bind" entry point for unresolved PTL entries. FreeBSD/mips has
an additional entry point called "_mips_rtld_bind".
gdb/ChangeLog:
* fbsd-tdep.c (fbsd_skip_solib_resolver): New function.
(fbsd_init_abi): Install gdbarch "skip_solib_resolver" method.
* fbsd-tdep.h (fbsd_skip_solib_resolver): New prototype.
* mips-fbsd-tdep.c (mips_fbsd_skip_solib_resolver): New function.
(mips_fbsd_init_abi): Install gdbarch "skip_solib_resolver"
method.
* guile/scm-math.c (vlscm_integer_fits_p): Use 'uintmax_t'
and 'intmax_t' instead of 'scm_t_uintmax' and 'scm_t_intmax',
which are deprecated in Guile 3.0.
* configure.ac (try_guile_versions): Add "guile-3.0".
* configure (try_guile_versions): Regenerate.
* NEWS: Update entry.
* gdb.guile/source2.scm: Add #f first argument to 'format'.
* gdb.guile/types-module.exp: Remove "ERROR:" from
regexps since Guile 3.0 no longer prints that.
* guile.texi (Memory Ports in Guile): Mark
'memory-port-read-buffer-size',
'set-memory-port-read-buffer-size!',
'memory-port-write-buffer-size',
'set-memory-port-read-buffer-size!' as deprecated.
* doc/guile.texi (Guile Introduction): Clarify which Guile
versions are supported.
Gary Benson [Mon, 20 Jul 2020 14:01:04 +0000 (15:01 +0100)]
Skip tests requiring "alignof (void)" when compiling using clang
As an extension, GCC allows void pointer arithmetic, with sizeof(void)
and alignof(void) both 1. GDB supports this extension, but clang does
not, and fails to compile the generated output of gdb.cp/align.exp
with the following error:
gdb compile failed, /gdbtest/build/gdb/testsuite/outputs/gdb.cp/align/align.cc:28:23:
error: invalid application of 'alignof' to an incomplete type 'void'
unsigned a_void = alignof (void);
^ ~~~~~~
1 error generated.
This commit adds preprocessor conditionals to the generated output, to
omit the unsupported code when using clang, and supplies the expected
value so the test can complete.
gdb/testsuite/ChangeLog:
* gdb.cp/align.exp: Fix "alignof (void)" tests when compiling
with clang.
Alan Modra [Mon, 20 Jul 2020 04:02:34 +0000 (13:32 +0930)]
PR26265, Spurious 'gc-sections requires ... when --init or --fini
bfd/
* elflink.c (_bfd_elf_gc_keep): Use bfd_is_const_section.
ld/
PR 26265
* ldlang.c (undef_from_cmdline): Delete.
(ldlang_add_undef): Mark "cmdline" param unused.
(lang_end): Traverse gc_sym_list to determine whether a symbol root
has been specified. Update error message.
* testsuite/ld-gc/noent.d: Adjust for changed error message.
Tom de Vries [Mon, 20 Jul 2020 12:40:52 +0000 (14:40 +0200)]
[gdb/testsuite] Stabilize execution order in omp-par-scope.c
In openmp test-case gdb.threads/omp-par-scope.exp we xfail and kfail dependent
on omp_get_thread_num (). Since execution order of the threads can vary from
execution to execution, this can cause changes in test results.
F.i., we can see this difference between two test runs:
...
-KFAIL: single_scope: first thread: print i3 (PRMS: gdb/22214)
+PASS: single_scope: first thread: print i3
-PASS: single_scope: second thread: print i3
+KFAIL: single_scope: second thread: print i3 (PRMS: gdb/22214)
...
In both cases, the KFAIL is for omp_get_thread_num () == 1, but in one case
that corresponds to the first thread executing that bit of code, and in the
other case to the second thread.
Get rid of this difference by stabilizing execution order.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2020-07-20 Tom de Vries <tdevries@suse.de>
* gdb.threads/omp-par-scope.c (lock, lock2): New variable.
(omp_set_lock_in_order): New function.
(single_scope, multi_scope, nested_func, nested_parallel): Use
omp_set_lock_in_order and omp_unset_lock.
(main): Init and destroy lock and lock2.
H.J. Lu [Mon, 20 Jul 2020 11:53:58 +0000 (04:53 -0700)]
x86: Update PR gas/26263 linker tests
Update and run PR gas/26263 linker tests for all x86 ELF targets to
accept any program header layout.
PR gas/26263
* testsuite/ld-i386/pr26263.d: Updated.
* testsuite/ld-x86-64/pr26263.d: Likewise.
* testsuite/ld-x86-64/x86-64.exp: Run gas/26263 test for all ELF
targets.
Tom de Vries [Mon, 20 Jul 2020 09:19:51 +0000 (11:19 +0200)]
[gdb/testsuite] Fix valgrind-infcall-2.exp without libc debug info
When running test-case gdb.base/valgrind-infcall-2.exp on a system without
libc debug info installed, I run into:
...
(gdb) p printf ("bla")^M
'printf' has unknown return type; cast the call to its declared return type^M
(gdb) FAIL: gdb.base/valgrind-infcall-2.exp: do printf
...
Fix this by casting the result of the printf call to int.
Tom de Vries [Mon, 20 Jul 2020 08:54:31 +0000 (10:54 +0200)]
[gdb/testsuite] Bail out after gdb_start error in gdb.threads/attach-slow-waitpid.exp
When building gdb using CFLAGS/CXXFLAGS+=-fsanitizer=address and
LDFLAGS+=-lasan, and running test-case gdb.threads/attach-slow-waitpid.exp,
we get:
...
spawn gdb -nw -nx -data-directory data-directory^M
==16079==ASan runtime does not come first in initial library list; \
you should either link runtime to your application or manually preload \
it with LD_PRELOAD.^M
ERROR: (eof) GDB never initialized.
ERROR: : spawn id exp10 not open
while executing
"expect {
-i exp10 -timeout 120
-re "Kill the program being debugged. .y or n. $" {
send_gdb "y\n" answer
verbose "\t\tKilling previous pro..."
("uplevel" body line 1)
invoked from within
"uplevel $body" NONE : spawn id exp10 not open
WARNING: remote_expect statement without a default case
ERROR: : spawn id exp10 not open
while executing
"expect {
-i exp10 -timeout 120
-re "Reading symbols from.*LZMA support was disabled.*$gdb_prompt $" {
verbose "\t\tLoaded $arg into $GDB; .gnu_..."
("uplevel" body line 1)
invoked from within
"uplevel $body" NONE : spawn id exp10 not open
ERROR: Couldn't load attach-slow-waitpid into GDB (eof).
ERROR: Couldn't send attach 16070 to GDB.
UNRESOLVED: gdb.threads/attach-slow-waitpid.exp: attach to target
...
Bail out at the first ERROR, such that we have instead:
...
ERROR: (eof) GDB never initialized.
UNTESTED: gdb.threads/attach-slow-waitpid.exp: \
Couldn't start GDB with preloaded lib
...
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2020-07-20 Tom de Vries <tdevries@suse.de>
* gdb.threads/attach-slow-waitpid.exp: Bail out if gdb_start fails.
Jan Beulich [Mon, 20 Jul 2020 06:57:18 +0000 (08:57 +0200)]
x86: handle SVR4 escaped binary operators
PR gas/4572
When / is a comment character, its use as binary "divide" operator needs
escaping by a backslash. Besides the scrubber needing to support this
(addressed in an earlier change), there are also a few provisions needed
in target specific operator handling.
As the spec calls for % and * to also be escaped because of being
"overloaded", also recognize these, despite the overloading there not
really preventing their use as operators in most (%) or all (*) cases,
given the way how the rest of the assembler works.
To bring source and testsuite in line, also drop the TE_I386AIX part of
the respective conditional, as i?86-*-aix* support had been removed a
while ago.
Jan Beulich [Mon, 20 Jul 2020 06:56:23 +0000 (08:56 +0200)]
gas: generalize comment character escaping recognition
PR gas/4572
Generalize what ab1fadc6b2f0 ("PR22714, Assembler preprocessor loses
track of \@") did to always honor escaped comment chars. Use this then
to support escaped /, %, and * operators on x86, when / is a comment
char (to match the Sun assembler's behavior).
Jan Beulich [Mon, 20 Jul 2020 06:55:48 +0000 (08:55 +0200)]
x86: honor absolute section when emitting code
Various provisions exist for insns to be placed in the absolute section,
yet actually trying to do so didn't work. While data emission (of non-
zero values) is not allowed by generic code, I think this functionality
is useful for the programmer to be able to determine the size of insns.
Therefore, rather than turning the silnet failure into a verbose one,
make things mostly work; the one class of insns not supported (yet) are
branches (JMP and Jcc) with dynamically determined displacement widths.
In this one case, an error now gets reported instead of silently
ignoring the code.
Also avoid recording ISA / feature usage for insns emitted to the
absolute section.
H.J. Lu [Sun, 19 Jul 2020 13:51:19 +0000 (06:51 -0700)]
x86: Change PLT32 reloc against section to PC32
Commit 292676c1 resolved PLT32 reloc aganst local symbol to section.
Since PLT32 relocation must be against symbols, turn such PLT32
relocation into PC32 relocation.
gas/
PR gas/26263
* config/tc-i386.c (i386_validate_fix): Change PLT32 reloc
against section to PC32 reloc.
* testsuite/gas/i386/relax-5.d: Updated.
* testsuite/gas/i386/x86-64-relax-4.d: Likewise.
So, here's my suggestion for making _init .. __etext cover .text +
.rodata (including things like the read-only exception tables) for
elf64mmix. A quick web search gives that __etext (and friends) isn't
well defined, so each target can interpret the "end of text segment"
to their own liking. It seems likely this change is also a better fit
than the default for other ports, at least those with .rodata after
.text in the same segment.
The presence of a separate rodata-segment is optional (and not true
for elf64mmix). This is reflected in the name as SEPARATE_TEXT /
SEPARATE_CODE isn't considered, to keep it simple; each target has to
make sure their settings of variables make sense.
Alan Modra [Fri, 17 Jul 2020 07:17:28 +0000 (16:47 +0930)]
Power10 stub selection
This patch better supports mixing of power10 and non-power10 code,
as might be seen in a cpu-optimized library using ifuncs to select
functions optimized for a given cpu. Using -Wl,--no-power10-stubs
isn't that good in this situation since non-power10 notoc stubs are
slower and larger than the power10 variants, which you'd like to use
on power10 code paths.
With this change, power10 pc-relative code that makes calls marked
@notoc uses power10 stubs if stubs are necessary, and other calls use
non-power10 instructions in stubs. This will mean that if gcc is
generating code for -mcpu=power10 but with pc-rel disabled then you'll
get the older stubs even on power10 (unless you force with
-Wl,--power10-stubs). That shouldn't be too big a problem: stubs that
use r2 are reasonable. It's just the ones that set up addressing
using "mflr 12; bcl 20,31,.+4; mflr 11; mtlr 12" that should be
avoided if possible.
bfd/
* elf64-ppc.c (struct ppc_link_hash_table): Add has_power10_relocs.
(select_alt_stub): New function.
(ppc_get_stub_entry): Use it here.
(ppc64_elf_check_relocs): Set had_power10_relocs rather than
power10_stubs.
(ppc64_elf_size_stubs): Clear power10_stubs here instead. Don't
merge notoc stubs with other varieties when power10_stubs is "auto".
Instead dup the stub hash table entry.
(plt_stub_size, ppc_build_one_stub, ppc_size_one_stub): Adjust
tests of power10_stubs.
ld/
* emultempl/ppc64elf.em (power10-stubs): Accept optional "auto" arg.
* ld.texi (power10-stubs): Update.
* testsuite/ld-powerpc/callstub-1.d: Force --power10-stubs.
* testsuite/ld-powerpc/callstub-2.d: Relax branch offset comparison.
* testsuite/ld-powerpc/callstub-4.d: New test.
* testsuite/ld-powerpc/notoc.d: Force --no-power10-stubs.
* testsuite/ld-powerpc/notoc3.d,
* testsuite/ld-powerpc/notoc3.s,
* testsuite/ld-powerpc/notoc3.wf: New test.
* testsuite/ld-powerpc/powerpc.exp: Run new tests. Pass
--no-power10-stubs for notoc link.
Tom Tromey [Sat, 18 Jul 2020 16:43:16 +0000 (10:43 -0600)]
Remove "linux_multi_process" global
The "linux_multi_process" is initialized but never modified. I
discussed this with Pedro on irc, and he said that, while it was
useful when developing this feature, it is now no longer needed. So,
this removes it.
gdb/ChangeLog
2020-07-18 Tom Tromey <tom@tromey.com>
Andrew Burgess [Thu, 16 Jul 2020 16:48:12 +0000 (17:48 +0100)]
gdb/riscv: delete target descriptions when gdb exits
It was pointed out on IRC that the RISC-V target allocates target
descriptions and stores them in a global map, and doesn't delete these
target descriptions when GDB shuts down.
This isn't a particular problem, the total number of target
descriptions we can create is very limited so creating these on demand
and holding them for the entire run on GDB seems reasonable.
However, not deleting these objects on GDB exit means extra warnings
are printed from tools like valgrind, and the address sanitiser,
making it harder to spot real issues. As it's reasonably easy to have
GDB correctly delete these objects on exit, lets just do that.
I started by noticing that we already have a target_desc_up type, a
wrapper around unique_ptr that calls a function that will correctly
delete target descriptions, so I want to use that, but....
...that type is declared in gdb/target-descriptions.h. If I try to
include that file in gdb/arch/riscv.c I run into a problem, that file
is compiled into both GDB and GDBServer.
OK, I could guard the include with #ifdef, but surely we can do
better.
So then I decided to move the target_desc_up type into
gdbsupport/tdesc.h, this is the interface file for generic code shared
between GDB and GDBserver (relating to target descriptions). The
actual implementation for the delete function still lives in
gdb/target-description.c, but now gdb/arch/riscv.c can see the
declaration. Problem solved....
... but, though RISC-V doesn't use it I've now exposed the
target_desc_up type to gdbserver, so in future someone _might_ start
using it, which is fine, except right now there's no definition of the
delete function - remember the delete I used is only defined in GDB
code.
No problem, I add an implementation of the delete operator into
gdbserver/tdesc.cc, and all is good..... except....
I start getting this error from GCC:
tdesc.cc:109:10: error: deleting object of polymorphic class type ‘target_desc’ which has non-virtual destructor might cause undefined behavior [-Werror=delete-non-virtual-dtor]
Which is caused because gdbserver's target_desc type inherits from
tdesc_element which has a virtual method, and so GCC worries that
target_desc might be used as a base class.
The solution is to declare gdbserver's target_desc class as final.
This is fine so long as we never intent to inherit from
target_desc (in gdbserver). But if we did then we'd want to make
target_desc's destructor virtual anyway, so the error above would be
resolved, and there wouldn't be an issue.
gdb/ChangeLog:
* arch/riscv.c (riscv_tdesc_cache): Change map type.
(riscv_lookup_target_description): Return pointer out of
unique_ptr.
* target-descriptions.c (allocate_target_description): Add
comment.
(target_desc_deleter::operator()): Likewise.
* target-descriptions.h (struct target_desc_deleter): Moved to
gdbsupport/tdesc.h.
(target_desc_up): Likewise.
gdbserver/ChangeLog:
* tdesc.cc (allocate_target_description): Add header comment.
(target_desc_deleter::operator()): New function.
* tdesc.h (struct target_desc): Declare as final.
gdbsupport/ChangeLog:
* tdesc.h (struct target_desc_deleter): Moved here
from gdb/target-descriptions.h, extend comment.
(target_desc_up): Likewise.
In commit ee3c5f8968 "Fix GDB crash when registers cannot be modified", we
fix a GDB crash:
...
$ valgrind /usr/bin/sleep 10000
==31595== Memcheck, a memory error detector
==31595== Command: /usr/bin/sleep 10000
==31595==
$ gdb /usr/bin/sleep
(gdb) target remote | vgdb --pid=31595
Remote debugging using | vgdb --pid=31595
...
$hex in __GI___nanosleep () at nanosleep.c:27
27 return SYSCALL_CANCEL (nanosleep, requested_time, remaining);
(gdb) p printf ("bla")
terminate called after throwing an instance of 'gdb_exception_error'
Aborted (core dumped)
...
This patch adds a test-case for it.
Unfortunately, I was not able to trigger the error condition using a regular
vgdb_start, so I've added a parameter active_at_startup, and when set to 0
this causes valgrind to be started without --vgdb-error=0.
Tested on x86_64-linux.
Tested with the commit mentioned above reverted, resulting in:
...
(gdb) p printf ("bla")^M
terminate called after throwing an instance of 'gdb_exception_error'^M
ERROR: GDB process no longer exists
GDB process exited with wait status 6152 exp10 0 0 CHILDKILLED SIGABRT SIGABRT
UNRESOLVED: gdb.base/valgrind-infcall-2.exp: do printf
...
gdb/testsuite/ChangeLog:
2020-07-17 Tom de Vries <tdevries@suse.de>
* gdb.base/valgrind-infcall-2.c: New test.
* gdb.base/valgrind-infcall-2.exp: New file.
* lib/valgrind.exp (vgdb_start): Add and handle active_at_startup.
Tom Tromey [Fri, 17 Jul 2020 15:20:33 +0000 (09:20 -0600)]
Use boolean literals in linux-nat.c
I noticed a couple of spots in linux-nat.c that use 0/1 where boolean
literals would be more idiomatic. This patch makes this change.
gdb/ChangeLog
2020-07-17 Tom Tromey <tromey@adacore.com>
* linux-nat.c (linux_nat_target::supports_non_stop)
(linux_nat_target::always_non_stop_p): Use "true".
(linux_nat_target::supports_disable_randomization): Use "true" and
"false".
projects / thirdparty / binutils-gdb.git / log
