with all the libpthread symbols in it.
This makes -lpthread behave like to -Wl,--whole-archive -lpthread.
This avoids a situation where threaded static binaries use some libc thread
stubs, which are racy.
Fixes PR lib/54001: call_once2_32, call_once2_static test cases failing on
amd64 since gcc7 import.
Suggested by Jonathan Wakely, thanks!
vcpu_run. This saves a few syscalls and copyins.
For example on Windows 10, moving the mouse from the left to right sides of
the screen generates ~500 events, which now don't result in syscalls.
The error handling is done in vcpu_run and it is less precise, but this
doesn't matter a lot, and will be solved with future NVMM error codes.
Use an intermediate function trampoline to workaround different function
pointer prototypes.
While there, correct scenario returning thrd_nomem from thrd_create(3).
In practice ENOMEM is rarely returned from pthread(3).
Older code worked on tested ports, but was depending on unneeded UB.
The F spec is one of those, it should be terminated with \0 just like all
the others (irrelevant that it has no extra data to delimit).
Fix <sys/mman.h> to define the snprintb() format string correctly (include
the missing \0's). Fix the copy of that definition included into
snprintb(3) to match the updated mman.h version (ride the date bump
from the day before yesterday .. this is the same change, just corrected).
Undo the previous snprintb.c change ("off by one" fix) which was an
attempt to make the broken mman.h usage work (and did, but not the way
it should be done). Also, after using the new * format (instead of only
when something has already matched) skip the associated data so we don't
attempt to interpret it as more field specifiers. This func needs lots of TLC!
Fix the ATF tests for snprintb() to not assume that F format is really
exactly like f format, and has data after the field specifier. It doesn't.
Add several more tests (including testing the '*' field operator
recently added).
introduce a bidirectionnal "comm page", a page of memory shared between
the kernel and userland, and used to transfer data in and out in a more
performant manner than ioctls.
The comm page contains the VCPU state, plus three flags:
- "wanted": the states the kernel must get/set when requested via ioctls
- "cached": the states that are in the comm page
- "commit": the states the kernel must set in vcpu_run
The idea is to avoid performing expensive syscalls, by using the VCPU
state cached, either explicitly or speculatively, in the comm page. For
example, if the state is cached we do a direct 1->5 with no syscall:
+---------------------------------------------+
| Qemu |
+---------------------------------------------+
| ^
| (0) nvmm_vcpu_getstate | (6) Done
| |
V |
+---------------------------------------+
| libnvmm |
+---------------------------------------+
| ^ | ^
(1) State | | (2) No | (3) Ioctl: | (5) Ok, state
cached? | | | "please cache | fetched
| | | the state" |
V | | |
+-----------+ | |
| Comm Page |------+---------------+
+-----------+ |
^ |
(4) "Alright | V
babe" | +--------+
+-----| Kernel |
+--------+
The main changes in behavior are:
- nvmm_vcpu_getstate(): won't emit a syscall if the state is already
cached in the comm page, will just fetch from the comm page directly
- nvmm_vcpu_setstate(): won't emit a syscall at all, will just cache
the wanted state in the comm page
- nvmm_vcpu_run(): will commit the to-be-set state in the comm page,
as previously requested by nvmm_vcpu_setstate()
In addition to this, the kernel NVMM driver is changed to speculatively
cache certain states known to be of interest, so that the future
nvmm_vcpu_getstate() calls libnvmm or the emulator will perform will use
the comm page rather than expensive syscalls. For example, if an I/O
VMEXIT occurs, the I/O Assist in libnvmm will want GPRS+SEGS+CRS+MSRS,
and now the kernel caches all of that in the comm page before returning
to userland.
Overall, in a normal run of Windows 10, this saves several millions of
syscalls. Eg on a 4CPU Intel with 4VCPUs, booting the Win10 install ISO
goes from taking 1min35 to taking 1min16.
The libnvmm API is not changed, but the ABI is. If we changed the API it
would be possible to save expensive memcpys on libnvmm's side. This will
be avoided in a future version. The comm page can also be extended to
implement future services.
These were removed from the build for being wrong, but the implementation
stayed around. This is confusing, we have the attic for old code,
let's delete the unused implementations.
(Unclear why they were different, it was only the comments that
differed)
Add a comment describing why the generic version works for both 80-bit
and 128-bit double.
The original implementation of C11 threads(3) contained check for error
paths, but it was stripped in the calls that are documented to return
no status from an operation. Do the same in call_once(3).
Switch tss_t type from int to pthread_key_t (no functional change as
pthread_key_t was already typedefed as int).
Noted by <enh from Google>.
Use C11 _Noreturn in thrd_exit(3) instead of NetBSD specific __dead.
The former is documented in the standard as an attribute of thrd_exit(3),
the latter is more portable to pre-C11 compilers, however C11 thread
support library needs C11 compiler for TLS anyway. __dead made a little bit
more point 3 years ago than today as 3 years ago pre-C11 compilers were
more common.
C11 Thread support library is a portable threading C API between OSs,
similar to std::threads in the C++ world.
The library is implemented as a thin shim over POSIX interfaces.
NetBSD implements the API as a part of the POSIX threading library
(libpthread(3)).
C11 threads(3) are in the process of making them an integral part
of the POSIX standard. The interface has been implemented in major
OSs and used with stopgap libraries for older versions of them.
C11 threading library is already used (with a stopgap implementation)
in the NetBSD distribution in MESA.
Original implementation by myself from 2016.
ATF tests are new and cover almost all interfaces.
Proposed on tech-userlevel@.
This fixes:
env TZ=Australia/Melbourne date
printing
Thu Apr 18 03:32:32 LMT 2019
instead of:
Thu Apr 18 03:32:32 AEST 2019
XXX: To be strictly correct, the tzgetname() call should probably take a time
reference point in order to return the appropriate zone name for the time
given.
The types array can have 0 elements (i.e. timecnt == 0), but the ttis array
needs to have one element (typecnt >= 1). Need to revisit and take into
account the types array.
then we try the non-dst name and finally we just print ??? if that did not
work. This is not needed to handle the "TZ= date" case which was crashing
before because the default timezone was not consistently initialized, but
should handle any timezone now.
Changes to code
zic now has an -r option to limit the time range of output data.
For example, 'zic -r @1000000000' limits the output data to
timestamps starting 1000000000 seconds after the Epoch.
This helps shrink output size and can be useful for applications
not needing the full timestamp history, such as TZDIST truncation;
see Internet RFC 8536 section 5.1. (Inspired by a feature request
from Christopher Wong, helped along by bug reports from Wong and
from Tim Parenti.)
Changes to documentation
Mention Internet RFC 8536 (February 2019), which documents TZif.
tz-link.html now cites tzdata-meta
<https://tzdata-meta.timtimeonline.com/>.
This makes it more portable as open_memstream is POSIX and fixes a
potential issue with wide characters not fully being printed
due to any buffer overflow.
Quoting of special characters in filename completion was implemented for single match
case, this enables it for multiple matches as well. For example:
$ touch 'foo bar'
$ touch 'foo baz'
$ ls fo<TAB>
autocompletes to =>
$ ls foo\ ba
hitting <TAB> again shows:
foo bar foo baz
This required unescaping escape sequences generated during last completion
in order to find the word to complete.
While there, also update the test to include cases for multiple matches.
Reviewed by christos
during refresh.
Ensure the character width is not negative when advancing during refresh
(unlikely) and we actually have something to insert in the lower right
corner depending on terminal caps.
Fixes PR lib/54085
doesn't work that well. addstr() the whole buffer instead. This is
still not enirely correct b/c printf can run out of stdio buffer
mid-character for very long output, but deal with it later.
If the user supplies a value for the attempted_completion_function parameter
then we cannot be sure if the completion is for filename or something else, in such
a case don't attempt to quote the completion matches.
Reviewed by christos
This should address PR lib/54067
For some reason I had initially concluded that it wasn't doable; verily it
is, so let's do it.
The reserved 'flags' argument of nvmm_gpa_map() becomes 'prot' and takes
mmap-like protection codes.
Until now, if application happens to have a global variable of the same
name, it was overridden by curses routines. This is the scenario in
which aspell crashes when linked to our curses, reported in pkg/44005.
We need to wipe out global/static variables like "_cursesi_state" or
"wstate" for thread safety. But it would be a future task...
XXX pullup to netbsd-8 and netbsd-7
When auto scaling, and the buffer is bigger than big enough
for the biggest possible number, don't try and calculate
the max value that will fit in the buffer - that calc
will overflow (guaranteed) and is useless, the value
we're formatting cannot possibly be bigger. So simply
use the unscaled value (the raw number).
While here, also avoid returning values that are larger
than the buffer len ... while it would be nice to be able
to find out how big the buffer should be so the data will
fit, the interface doesn't really allow that (the buffer
length passed in controls the scaling - at least when
auto scaling) and the code already does "return -1" when
it detects the buffer length is too small, even before
it works out how much would have been needed. So, rather
than returning a value > len (while truncating the result
to fit in len ... all courtesy of snprintf()) return -1
in this case as well.
Also, allow suffix==NULL (meaning "") - there's no reason
not to, and requiring users to pass in an explicit "" is
not useful.
- Compress x86_rexpref, x86_regmodrm, x86_opcode and x86_instr.
- Cache-align the register, opcode and group tables.
- Modify the opcode tables to have 256 entries, and avoid a lookup.
to the pthread tsd implementation when the main thread is created.
This corrects a problem where a process created keys before libpthread
was loaded (either from the libc constructor or because libpthread
was dlopened later). This fixes a problem with jemalloc which creates
keys in the constructor.
This allows things that occur after lib (like do-x11) to depend on llvm
libraries already existing.
We need libexecinfo, so adding llvm after the 2nd library barrier.
- Reorder it, to match the CPU encoding. This is the universal order,
also used by Qemu. Drop the seg_to_nvmm[] tables.
- Compress it. This divides its size by two.
- Rename some of its fields, to better match the x86 spec. Also, take S
out of Type, this was a NetBSD-ism that was likely confusing to other
people.
The long-standing GCC default is to not respect rounding mode.
it looks like GCC 7 optimizes rint to a builtin, causing our few
rounding mode tests to fail.
Fixes PR port-amd64/54000: FP tests failing on amd64 since gcc7 import
maya