This patch set modifies Cygwin's profiling support to sample PC values
of all an app's threads, not just the main thread. There is no change
to how profiling is requested: just compile and link the app with "-pg"
as usual. The profiling info is dumped into file gmon.out as always.
A new facility enabled via the environment variable GMON_OUT_PREFIX.
This facility is intended to match an undocumented Linux glibc feature.
Exporting the variable with a non-empty value such as "foo" causes the
profiling info to go to a file named foo.$pid instead of the default.
With that, both resulting processes of a fork() can have their profiling
data captured in separate files. gprof already knows how to accumulate
data from multiple files if they all pertain to the same app.
There is no change to the normal Cygwin execution paths if profiling is
not enabled. And when it is enabled, only the one profiling thread per
profiled app is doing more work than it used to.
* include/sys/cygwin.h: Add CW_CYGHEAP_PROFTHR_ALL.
* cygheap.cc (cygheap_profthr_all): New C-callable function that
runs cygheap's threadlist handing each pthread's thread handle in
turn to profthr_byhandle().
* external.cc (cygwin_internal): Add case CW_CYGHEAP_PROFTHR_ALL.
* gmon.c (_mcleanup): Add support for multiple simultaneous
gmon.out* files named via environment variable GMON_OUT_PREFIX.
* gmon.h (struct gmonparam): Make state decl volatile.
* mcount.c (_MCOUNT_DECL): Change stores into gmonparam.state to use
Interlocked operations. Add #include "winsup.h", update commentary.
* profil.c (profthr_byhandle): New function abstracting out the
updating of profile counters based on a thread handle.
(profthr_func): Update to call profthr_byhandle() to sample the main
thread then call cygheap_profthr_all() indirectly through
cygwin_internal(CW_CYGHEAP_PROFTHR_ALL) to sample all other threads.
(profile_off): Zero targthr to indicate profiling was turned off.
(profile_on): Fix handle leak on failure path.
(profile_child): New callback func to restart profiling in child
process after a fork if the parent was being profiled.
(profile_ctl): Call pthread_atfork() to set profile_child callback.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
According to https://cygwin.com/ml/cygwin/2016-03/msg00124.html it's a
problem to collect friendlyname info using AF_INET6 sockets. Fix problem
by exposing additional hardware info for all collected interfaces via the
pointer in the ifaddrs::ifa_data member.
* include/ifaddrs.h (struct ifaddrs_hwdata): Define as struct of
not yet exposed members of struct ifall, defined in net.cc.
* net.cc (struct ifall): Replace hardware dta members with struct
ifaddrs_hwdata. Accommodate throughout.
(get_ifs): Let ifaddrs ifa_data member point to ifall::ifa_hwdata
member.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
Calling open from acl_extended_file{_nofollow} indiscriminately may hang
if the file is a FIFO. Ultimately the FIFO implementation needs a thorough
rewrite, but for the time being we better do what stat(2) and friends do:
Just create an fhandler directly.
* sec_posixacl.cc (__acl_extended_fh): New static function calling
fhandler::facl.
(acl_extended_fd): Just call __acl_extended_fh.
(__acl_extended_file): Take just a path_conv as parameter and
create temporary fhandler to call __acl_extended_fh.
(acl_extended_file): Create path_conv from incoming path and
call __acl_extended_file on it.
(acl_extended_file_nofollow): Ditto.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
* ntdll.h (RtlGetNtVersionNumbers): Declare.
* wincap.cc (wincapc::init): Overwrite kernel version info
returned by RtlGetVersion with correct info returnd by
RtlGetNtVersionNumbers. Add comment.
Originally, using RtlGetVersion instead of GetVersionEx was supposed to
fix the fact that GetVersionInfo returns the wrong kernel version if the
executable has been built with an old manifest (or none at all), starting
with Windows 8.1. Either this never really worked as desired and our
testing was flawed, or this has been changed again with Windows 10, so
that RtlGetVersion does the kernel faking twist as well. Since we're
only reading the value in the first process in a process tree. the entire
process tree is running with a wrong OS version information in that case.
Fortunately, the (undocumented) RtlGetNtVersionNumbers function is not
affected by this nonsense, so we simply override the OS version info
fields with the correct values now.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
The problem this patch fixes showed up after updating to gcc-5.3.0. The
cuplrit is a change in gcc when emitting section attributes. It only
shows up when building without optimization. Effect in Cygwin: ws2_32
functions failed to load.
In the original code the definition of "NO_COPY wsadata" was preceeding
an __asm__ block (the definition of the _wsock_init wrapper), while the
definition of "NO_COPY here" immediately follows the same assembler
block. When gcc-5.3.0 emits assembler code for the wsadata definition,
it emits the .data_cygwin_nocopy section attribute.
Next it emits the assembler output for the __asm_ block, entirely ignoring
its content. The __asm__ block adds a .text section definition.
Eventually gcc emits assembler code for the here definition. However,
apparently gcc still "knows" that it just emitted the .data_cygwin_nocopy
section attribute and so doesn't redefine it. Remember the __asm__? It
changed the section to .text.
So with gcc-4.9.3 we got:
.section .data_cygwin_nocopy,"w"
wsadata:
__asm__ block:
.text
.section .data_cygwin_nocopy,"w"
here:
With gcc 5.3.0 we now get:
.section .data_cygwin_nocopy,"w"
wsadata:
__asm__ block:
.text
here:
So "here" is now in the .text segment which is read-only. Hilarity ensues.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
* faq-using.xml (bloda): Alphabetically sort BLODA list for ease
of finding things in it.
(fixing-fork-failures): Update to suggest rebase-trigger rather
than running rebaseall via dash yourself. Mention detect_bloda
CYGWIN token.
Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
* sec_posixacl .cc (__acl_dup): Remove.
(acl_dup): Fold __acl_dup functionality into this function.
(acl_create_entry): Don't create new acl_t. Just realloc
acl->entry to make room for new aclent_t.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
* Makefile.in (DLL_OFILES): Add sec_posixacl.o.
(SUBLIBS): Add libacl.a
(libacl.a): New rule to create libacl.a.
* common.din: Export POSIX ACL functions as well as most libacl.a
extensions.
* fhandler.h (fhander_base::acl_get): New prototype.
(fhander_base::acl_set): Ditto.
(fhandler_disk_file::acl_get): Ditto.
(fhandler_disk_file::acl_set): Ditto.
* include/acl/libacl.h: New file.
* include/cygwin/version.h: Bump API minor version.
* include/sys/acl.h: Drop including cygwin/acl.h. Accommodate
throughout Cygwin. Add POSIX ACL definitions.
* sec_acl.cc: Include sec_posixacl.h. Replace ILLEGAL_UID and
ILLEGAL_GID with ACL_UNDEFINED_ID where sensible.
(__aclcheck): New internal acl check function to be used for
Solaris and POSIX ACLs.
(aclcheck32): Call __aclcheck.
(__aclcalcmask): New function to compute ACL_MASK value.
(__aclsort): New internal acl sort function to be used for Solaris
and POSIX ACLs.
(aclsort32): Call __aclsort.
(permtostr): Work directly on provided buffer.
(__acltotext): New internal acltotext function to be used for
Solaris and POSIX ACLs.
(acltotext32): Call __acltotext.
(__aclfromtext): New internal aclfromtext function to be used for
Solaris and POSIX ACLs.
(aclfromtext32): Call __aclfromtext.
* sec_posixacl.cc: New file implemeting POSIX ACL functions.
* sec_posixacl.h: New internal header.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
after some discussion on irc and the list i'm resubmitting a simpler
version of the patch.
setting the locale on cygwin to 'C.UTF-8' is not needed, so i'm always
setting it to 'C' which is sufficient for the build process and the
most simple fix.
short explanation: after setting up cygwin on my systems the default
locale is set to "de_DE.UTF-8". this leads to ccwrap not picking up
certain "-isystem" arguments, which in turn leads to "stddef.h: no such
file or directory". this breaks the build process for systems having non
english locale.
ccwrap scans the output of the first compiler invocation (line 21) for
some specific english output on and around line 43.
i changed the patch to check uname -o for cygwin string and set the
locale to either C or C.UTF-8
* pthread.h: Add prototypes for pthread_condattr_getclock(),
pthread_condattr_setclock(), pthread_setschedpri(),
pthread_getcpuclockid(), pthread_getconcurrency(), and
pthread_setconcurrency(). Also cleaned up file header to
remove CVS Id string.
* sys/types.h: Add clock ID to pthread_condattr_t.
Fix various 32/64-bit portability issues in ssp, the single-step profiler, and
also build it for 64-bit.
This didn't turn out to actually be very useful for what I wanted to use it for,
so it's only been lightly tested.
It appears that on x86_64, single-step exceptions occur for much more of the
code in system DLLs, unlike x86, so ssp may take much, much longer to profile
some programs. There is existing code to use breakpoints to mitigate this, but
that is currently disabled.
Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
* libc/include/sys/features.h: Set POSIX option macros from 200112L
to 200809L for Cygwin.
(_POSIX_SPAWN): Set to 200809L for Cygwin.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
* sec_auth.cc (get_token_group_sidlist): Drop auth_luid and
auth_pos parameter. Remove code adding a logon SID.
(get_initgroups_sidlist): Drop auth_luid and auth_pos parameter.
Drop in call to get_token_group_sidlist. Accommodate in callers.
(get_setgroups_sidlist): Ditto.
(create_token): Explicitely set auth_luid to ANONYMOUS_LOGON_LUID
or LOCALSERVICE_LUID depending on OS. Explain why.
Remove handling of logon SID since we don't generate one anymore.
(lsaauth): Drop now unused local variable auth_luid and auth_pos.
* wincap.h (wincaps::has_broken_whoami): New element.
* wincap.cc: Implement above element throughout.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
Currently, the newlib version information needs to be updated in two places:
- newlib/acinclude.m4
- newlib/libc/include/sys/features.h
The goal of this patch is to:
- supply a single location for defining the newlib version
information: newlib/acinclude.m4
- define __NEWLIB__, __NEWLIB_MINOR__ and __NEWLIB_PATCHLEVEL__
This is in line with what gcc does for its version macros. See:
https://gcc.gnu.org/onlinedocs/cpp/Common-Predefined-Macros.html
This patch moves the definition of the _NEWLIB_VERSION, __NEWLIB__
and __NEWLIB_MINOR__ macros from newlib/libc/include/sys/features.h,
to the newly generated newlib/_newlib_version.h file. Additionally,
the __NEWLIB_PATCHLEVEL__ macro was created, for completeness.
In order to stay backwards compatible, newlib/_newlib_version.h gets
included by newlib/newlib.h and newlib/libc/include/sys/features.h.
Note: This patch does _not_ include the modifications to the following
files, as these should all be generated any way.
*Makefile.in,
*aclocal.m4,
*configure
stamp-* files
Signed-off-by: Pieter du Preez <pdupreez@gmail.com>
* cyglsa.c (must_create_logon_sid): Remove.
(LsaApInitializePackage): Drop checking version and setting
must_create_logon_sid accordingly.
(LsaApLogonUserEx): Drop code setting login SID only required on
Windows 2000 and earlier.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
This follows up from my msg re GMP-ECM failing its 'make check' on the
main list https://cygwin.com/ml/cygwin/2016-02/msg00147.html .
There's an error that ought to be reported during dynamic linking if the
linked-to address is too far from the relocation site. However the error
is not reported if __OPTIMIZE__ was #defined when building the Cygwin DLL.
I can't see why optimization settings should affect this.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
* mount.cc (mount_info::conv_to_posix_path): Account for scenarios
where the path is identical to a mount point in terms of preserving
the trailing slash in the output.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
GCC 5 adds #line directives (and hence extra newlines) for macros
expansions, which confuses cygmagic. Using the -P flag avoids
them entirely.
https://cygwin.com/ml/cygwin-patches/2016-q1/msg00016.html
Signed-off-by: Yaakov Selkowitz <yselkowi@redhat.com>
GCC 5 switched from C89 to C11 by default. This implies a change from
GNU to C99 inline by default, which have very different meanings of
extern inline vs. static inline:
https://gcc.gnu.org/onlinedocs/gcc/Inline.html
Marking these as gnu_inline retains the previous behaviour.
winsup/cygwin/
* exceptions.cc (exception::handle): Change debugging to int to fix
an always-true boolean comparison warning.
* include/cygwin/config.h (__getreent): Mark gnu_inline.
* winbase.h (ilockcmpexch, ilockcmpexch64): Ditto.
Signed-off-by: Yaakov Selkowitz <yselkowi@redhat.com>
Any attempt to use isnanf, isinff, or finitef from <ieeefp.h> with
GCC 5 on platforms other than SPU result in a "lvalue required as
unary '&' operand" error.
newlib/libc/
* include/ieeefp.h (__ieeefp_isnanf): Remove broken macro.
(__ieeefp_isinff, __ieeefp_finitef): Ditto.
Signed-off-by: Yaakov Selkowitz <yselkowi@redhat.com>
This patch fixes the problem reported in
https://cygwin.com/ml/cygwin/2016-02/msg00014.html
The 2009 changes to handle multibyte decimal point and thousands
separator missed to take the length of a multibyte decimal point into
account when computing the field size.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
Current mesa expects stdio.h to prototype fileno() when compiling with gcc
-std=c99 -D_XOPEN_SOURCE
Fix the 'POSIX 1003.1:2001' block in stdio.h to prototype functions if not
__STRICT_ANSI__, or _BSD_SOURCE, _POSIX_C_SOURCE or _XOPEN_SOURCE are defined
appropriately
As far as I can tell, despite being in this block, setbuffer() and setlinebuf()
are BSDisms which aren't in POSIX 1003.1:2001
Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Non-reentrant system calls version implies both MISSING_SYSCALL_NAMES
and REENTRANT_SYSCALL_PROVIDED macros to be defined.
Being coupled with --enable-newlib-nano-malloc knob it breaks the build:
bash-4.3$ ../newlib-2.3.0.20160104/configure CC_FOR_TARGET=gcc
AR_FOR_TARGET=ar RANLIB_FOR_TARGET=ranlib CFLAGS_FOR_TARGET="-m32
-DMISSING_SYSCALL_NAMES -DREENTRANT_SYSCALLS_PROVIDED"
--target=i386-elf --enable-newlib-nano-malloc && make
<...omitted output...>
../../../../../../newlib-2.3.0.20160104/newlib/libc/stdlib/nano-mallocr.c:
In function ‘_mallinfo_r’:
../../../../../../newlib-2.3.0.20160104/newlib/libc/stdlib/nano-mallocr.c:489:35:
error: macro "_sbrk_r" requires 2 arguments, but only 1 given
sbrk_now = _sbrk_r(RCALL 0);
^
../../../../../../newlib-2.3.0.20160104/newlib/libc/stdlib/nano-mallocr.c:489:20:
error: ‘_sbrk_r’ undeclared (first use in this function)
sbrk_now = _sbrk_r(RCALL 0);
^
../../../../../../newlib-2.3.0.20160104/newlib/libc/stdlib/nano-mallocr.c:489:20:
note: each undeclared identifier is reported only once for each
function it appears in
Makefile:1512: recipe for target 'lib_a-nano-mallinfor.o' failed
make[8]: *** [lib_a-nano-mallinfor.o] Error 1
In case of non-reentrant system calls _sbrk_r became a macro with TWO
args (defined in reent.h):
#define _sbrk_r(__reent, __incr) sbrk(__incr)
But in our case only one argument is present. (RCALL 0) is considered
as a single argument despite RCALL itself is a macro:)
So intermediate one-arg macro will be enough to expand args before
final _sbrk_r expansion:
#define _SBRK_R(X) _sbrk_r(X)
Here is a patch:
Mark Geisert
Corinna Vinschen
Nick Clifton
Jon Turney
Alexey Sokolov
Patrick Bendorf
Joel Sherrill
Jeff Johnston
Yaakov Selkowitz
Pieter du Preez
John Hood
Václav Haisman
Игорь Веневцев