mikros/rust
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
12,797 Commits 2 Branches 0 Tags 1.4 GiB
ba8750a550
Commit Graph

323 Commits

Author SHA1 Message Date
Brian Anderson
c42492e6aa rt: Rename config_valgrind_stack to register_valgrind_stack, etc 2012-02-10 12:58:35 -08:00
Brian Anderson
fc028c30a0 rt: Move the addition of the stack canary into create_stack 2012-02-10 12:58:35 -08:00
Brian Anderson
d493438c98 rt: Disable some expensive asserts 2012-02-09 22:58:40 -08:00
Brian Anderson
d90a9d3da0 rt: Inline everything on the C-stack-switching path 2012-02-09 22:15:15 -08:00
Brian Anderson
8ad9cf8087 rt: Add more stack-switching commentary 2012-02-09 19:00:16 -08:00
Brian Anderson
5d2442d89c rt: Add upcall_call_shim_on_rust_stack 2012-02-09 19:00:16 -08:00
Brian Anderson
79b1563abb rt: Use rust_task_thread's C-stack pool for native calls 2012-02-09 19:00:16 -08:00
Brian Anderson
8fe506bdca rt: Move more setup code into create_stack 2012-02-09 19:00:16 -08:00
Brian Anderson
ea76d3f454 rt: Add rust_task::call_on_c_stack 2012-02-09 19:00:16 -08:00
Brian Anderson
1e2fe4ab8f rt: Rename call_shim_on_c_stack to call_and_change_stacks 2012-02-09 19:00:15 -08:00
Brian Anderson
ae8ea4a78b rt: Add constructors and destructors for stacks 2012-02-09 19:00:15 -08:00
Brian Anderson
b98df86c09 rt: Move some stack manipulation functions into rust_task 2012-02-09 19:00:15 -08:00
Brian Anderson
4bd8f8d936 rt: Rename new_stack to next_stack, del_stack to prev_stack 2012-02-09 19:00:15 -08:00
Brian Anderson
889a783494 rt: Remove rust_task::rust_sp 2012-02-09 19:00:15 -08:00
Brian Anderson
d39ea47746 rt: Remove rust_task_user struct 2012-02-09 19:00:15 -08:00
Brian Anderson
d756b01619 rt: Add a function for configuring task notification 2012-02-09 19:00:15 -08:00
Brian Anderson
e0d5b92b84 rt: Begin moving stack-building functions to rust_stack.cpp 2012-02-09 19:00:15 -08:00
Brian Anderson
028af5cb6c rt: Change the lifecycle of tasks and schedulers for various reasons 
This is in preparation for giving schedulers their own life cycle separate
from the kernel.

Tasks must be deleted before their scheduler thread, so we can't let the
scheduler exit before all its tasks have been cleaned up. In this scheme,
the scheduler will unregister tasks with the kernel when they are reaped,
then drop their ref on the task (there may still be others). When the task
ref count hits zero, the task will request to be unregistered from the
scheduler, which is responsible for deleting the task.

Instead of having the kernel tell the scheduler to exit, let the scheduler
decide when to exit. For now it will exit when all of its tasks are
unregistered.
 2012-02-08 15:42:51 -08:00
Brian Anderson
c7777f4fd9 rt: Add sanity checks when we hit the max task/port id 2012-02-03 23:48:12 -08:00
Brian Anderson
12fa90888e rt: Clean up the way the kernel tracks tasks 2012-02-03 23:48:12 -08:00
Brian Anderson
3d0e7a7af3 rt: Add a field to rust_task to hold its scheduler 2012-02-03 23:48:12 -08:00
Brian Anderson
f94339cc1c rt: Rename rust_scheduler to rust_task_thread 2012-02-03 23:48:12 -08:00
Marijn Haverbeke
c1b075d042 Remove experimental GC code 
It's been sitting unused long enough to have bitrotted completely.
 2012-02-03 11:34:12 +01:00
Brian Anderson
5c3c8d454d rt: Do all signalling while holding a lock 
This will matter once the scheduler is changed to not wake up on a timer
 2012-02-02 18:10:24 -08:00
Brian Anderson
3978fbff8b rt: Remove rust_task::can_schedule. Does nothing 2012-02-02 18:10:24 -08:00
Brian Anderson
ed3a5ad5ad rt: Remove running_on flag. Does nothing 2012-02-02 18:10:24 -08:00
Brian Anderson
57cad61353 rt: Remove task pinning. Does nothing 2012-02-02 18:10:24 -08:00
Brian Anderson
18de0f2aeb rt: Rename task_sleep intrinsic to task_yield. Remove usec param 2012-02-02 18:10:24 -08:00
Brian Anderson
8693fcc2d7 rt: Remove yield timers 2012-02-02 18:10:24 -08:00
Niko Matsakis
196d69beb2 make boxes self-describing (fixes #1493)" (take 2) 
this will be used to generate a new snapshot.
 2012-02-01 18:52:08 -08:00
Niko Matsakis
3116643806 Revert "make boxes self-describing (fixes #1493)" until a new 
snapshot is prepared.
 2012-02-01 18:50:19 -08:00
Niko Matsakis
c36207bfb8 make boxes self-describing (fixes #1493) 2012-02-01 18:18:07 -08:00
Brian Anderson
4eb92d4177 rt: Don't access fields on null supervisor task 2012-02-01 16:46:11 -08:00
Brian Anderson
fa13fd9d64 rt: Remove set_min_stack 2012-01-29 21:27:37 -08:00
Brian Anderson
6548cdd59b rt: Make the initial segment of the main task's stack 1MB 
This is a trick to fool microbenchmarks. Closes #1681
 2012-01-29 21:20:39 -08:00
Brian Anderson
94c389a25b rt: Add RUST_MAX_STACK env var with 8MB default 
Closes #1489
 2012-01-11 13:57:11 -08:00
Niko Matsakis
110c3ccdca add rust_task_is_unwinding predicate and do not kill if already unwinding 2012-01-09 19:53:32 -08:00
Brian Anderson
7e8974305c rt: Set Linux red zone to 20K again 
With runtime logging on it smashes the stack
 2012-01-09 11:44:36 -08:00
Niko Matsakis
8506241f3a fix how we walk functions to match new closure fmt 2012-01-06 22:40:31 -08:00
Niko Matsakis
25e81e34ea rewrite task tests 2012-01-06 22:40:31 -08:00
Niko Matsakis
98f5109cde simplify task impl 2012-01-06 22:40:31 -08:00
Brian Anderson
4632358cbc Revert "rt: Reduce mac red zone to 6K" 
This reverts commit d499e739a1.
 2012-01-06 18:16:19 -08:00
Brian Anderson
d499e739a1 rt: Reduce mac red zone to 6K 2012-01-06 16:42:26 -08:00
Brian Anderson
2f4c9315af rt: Fail the scheduler immediately on task failure in win32 2012-01-06 16:18:56 -08:00
User Jyyou
a59c4b1b47 freebsd support 2012-01-01 20:18:55 -08:00
Brian Anderson
dedfef4c4c Revert "wip" 
This reverts commit aeadc6269e.
 2011-12-20 20:57:04 -08:00
Brian Anderson
aeadc6269e wip 2011-12-20 20:15:09 -08:00
Brian Anderson
128621be97 rt: Remove rust_task::grow. Obsolete 2011-12-20 20:02:57 -08:00
Brian Anderson
cab4da70c8 rt: Don't check the stack canary on most upcalls 2011-12-20 16:50:54 -08:00
Brian Anderson
2a490d1cf8 Revert "rt: Reduce the mac red zone to 2K" 
This reverts commit ba8f369f89.
 2011-12-20 15:17:30 -08:00
Brian Anderson
ba8f369f89 rt: Reduce the mac red zone to 2K 2011-12-20 14:24:13 -08:00
Brian Anderson
2e230b1cf9 rt: Reduce the linux red zone to 2K 2011-12-20 14:02:59 -08:00
Brian Anderson
93fef2e113 rt: Parenthesize the red zone macros 2011-12-20 11:57:22 -08:00
Brian Anderson
784713ec68 rt: Add a canary value to the end of every stack 
Check it on upcall entry and exit, and on stack deletion
 2011-12-20 11:48:47 -08:00
Brian Anderson
56ec9cb278 rt: Run yet more task_start_wrapper cleanup on the C stack 2011-12-20 10:29:40 -08:00
Brian Anderson
25f7c844df rt: Don't reuse stack segments when they are not big enough 2011-12-19 22:47:28 -08:00
Brian Anderson
bd6b80c972 rt: Get rid of the rethrow in upcall_fail 
Throwing in upcall_fail ends up running lots of code in the red zone. To avoid
it we have the personality function figure out which stack it's on and switch
as needed.
 2011-12-18 17:17:31 -08:00
Brian Anderson
f57fd8de1f rt: Run the cycle collector on the C stack 2011-12-18 15:23:44 -08:00
Brian Anderson
74d5faa86e rt: Give each platform its own red zone definition 2011-12-18 15:23:41 -08:00
Brian Anderson
762d7e43a6 rt: Trim the red zone to 20k on all platforms 2011-12-18 15:22:56 -08:00
Brian Anderson
3fa930f598 rt: Remove a bogus FIXME from ~rust_task 2011-12-18 02:09:41 -08:00
Brian Anderson
fe683dfb80 rt: Get rid of the valgrind guard bytes at the end of the stack 
Preventing us from writing beyond our allocations is _what valgrind does_,
so telling valgrind not to let us write to the end of the stack isn't
buying anything.
 2011-12-17 16:46:50 -08:00
Brian Anderson
4ac345939c rt: Cache an extra stack segment to avoid bad behavior at stack boundaries 2011-12-17 13:45:38 -08:00
Brian Anderson
0ed5117a9d rt: Set the default stack size to 768 bytes. Double on each alloc 2011-12-17 13:45:38 -08:00
Brian Anderson
23df4de86d rt: Fix alignment of new stack segments 2011-12-16 18:18:43 -08:00
Brian Anderson
81f932e7d7 rt: Fix alignment of stacks generated by __morestack 2011-12-07 14:06:46 -08:00
Brian Anderson
9656ceac60 rt: Put 16 guard bytes at the end of the stack 2011-12-06 22:35:42 -08:00
Brian Anderson
5d1a1dc420 rt: Rename stk_seg.limit to stk_seg.end 
rust_task is using the word limit it two ways, so one has to change.
 2011-12-06 21:45:53 -08:00
Brian Anderson
9a738fd61d rt: Various tweaks to make __morestack unwinding work on linux 
When unwinding through __morestack the stack limit in the TLS is invalidated
and must be reset. Instead of actually landing at __morestack we're
just going to make all our Rust landing pads call upcall_reset_stack_limit,
which will find the stack segment that corresponds to the current stack
pointer and put the limit in the TLS.

Also massively expand the stack segment red zone to make more room for the
dynamic linker. Will fix in the future.
 2011-12-06 16:32:41 -08:00
Austin Seipp
b513a5a500 Make valgrind usage more consistent and less error prone. 
I was still having issues with the build system somehow getting confused
as to which set of valgrind headers to use when compiling rt.

This commit moves all the valgrind headers into their own directory
under rt and makes the usage more consistent. The compiler is now passed
the -DNVALGRIND flag when valgrind is not installed, as opposed to
passing -DHAVE_VALGRIND.

We also pass -I src/rt to the compiler when building rt so you can more
easily import what you want. I also cleaned up some erroneous #includes
along the way.

It should be safe to always just import the local valgrind headers and use
them without question. NVALGRIND turns the operations to no-ops when it
is active, and the build and tests run cleanly with or without.
 2011-12-06 01:15:29 -06:00
Brian Anderson
58844aee42 rt: Make stack unwinding work more correctly with stack growth 2011-12-05 17:42:58 -08:00
Brian Anderson
52d7dc5e0a rt: Update 32-bit __morestack for recent LLVM changes 2011-12-04 20:40:34 -08:00
Brian Anderson
1974cf9a81 rt: Move RED_ZONE_SIZE to rust_task.cpp 2011-12-01 15:50:00 -08:00
Brian Anderson
249f017bf2 rt: Add FIXME's about future changes to LLVM's __morestack impl 2011-12-01 15:33:15 -08:00
Brian Anderson
6da1a3fcd6 rt: Refactor record_sp into task::record_stack_limit 2011-12-01 15:26:42 -08:00
Brian Anderson
f2de42c39a rt: Reorganize stack growth code 2011-11-30 20:31:33 -08:00
Brian Anderson
e6ef4d929c rt: Remove the stack pointer field of stk_seg 2011-11-30 20:31:33 -08:00
Brian Anderson
06f4cb43f6 rt: Delete the entire stack chain on task destruction 
Unwinding through __morestack on 64-bit Linux seems to be no big deal, and
all we have to do is free the stacks to make unwinding work with split stacks.
 2011-11-29 22:22:49 -08:00
Brian Anderson
cd75c9ce11 rt: Implement part of the 64-bit __morestack 2011-11-28 16:29:52 -08:00
Brian Anderson
6bdf347418 rt: Make __morestack (without unwinding) work on 32-bit linux 2011-11-22 18:02:10 -08:00
Brian Anderson
9aa67052e1 rt: Remove some stack-wasting macros from rust_task::yield 2011-11-18 15:59:10 -08:00
Brian Anderson
1e58049da2 rt: Remove fail calls from rust_task::yield 2011-11-18 15:36:48 -08:00
Brian Anderson
792068d871 rt: Remove unblock call from rust_task::yield 2011-11-18 14:45:48 -08:00
Brian Anderson
0f339b481a rt: Remove a lock from task_start_wrapper that does nothing 2011-11-18 13:38:20 -08:00
Brian Anderson
3cfcdb4bb2 rt: Replace two uses of yield with ctx->swap 
These uses aren't really doing a full yield. They are just giving up control
to the scheduler and will never return.
 2011-11-18 13:38:18 -08:00
Brian Anderson
57b43b53f9 rt: Add comments to rust_task::yield 2011-11-18 13:28:12 -08:00
Brian Anderson
d1ef29ab37 rt: rust_task::conclude_failure doesn't need to call unblock 2011-11-18 13:23:27 -08:00
Brian Anderson
93931311ff rt: Add FIXMEs about races in rust_task 2011-11-18 10:32:27 -08:00
Brian Anderson
0dfa1410d3 rt: Remove task::on_wakeup. Unused 2011-11-18 10:32:27 -08:00
Brian Anderson
63d40b8ab6 rt: Remove some unnecessary setting of rust_task::killed 
The value of this variable doesn't matter after the task fails.
 2011-11-18 10:32:27 -08:00
Brian Anderson
d52888f0ba rt: Remove rust_task::yield. Unused 2011-11-18 10:32:27 -08:00
Brian Anderson
5e9f9e8c48 rt: Add some comments about methods that run on the Rust stack 2011-11-18 10:32:27 -08:00
Niko Matsakis
60e93cd1ec Re-enable cycle coll. on x86_64, seems to work better now. 2011-11-17 14:40:03 -08:00
Patrick Walton
e6c3c4e48c rt: More work on morestack 2011-11-17 10:52:59 -08:00
Brian Anderson
02cebbb9f1 Revert "rt: More work on morestack" 
This reverts commit 68aff2ad6d.
 2011-11-16 18:49:19 -08:00
Patrick Walton
68aff2ad6d rt: More work on morestack 2011-11-16 18:00:57 -08:00
Brian Anderson
8199558443 rt: Remove an unnecessary check from rust_task::yield 2011-11-16 17:55:04 -08:00
Brian Anderson
bae817efe6 rt: Remove rust_task::notify_tasks_wating_to_join 2011-11-16 17:22:14 -08:00
Niko Matsakis
69911c2bc1 temp. disable cycle coll. on x86_64 2011-11-16 15:16:41 -08:00
Patrick Walton
b3cf0c4d1b Revert "rt: More work on morestack" 
This reverts commit ced0d4f15e.
 2011-11-14 16:20:53 -08:00
Patrick Walton
ced0d4f15e rt: More work on morestack 2011-11-14 16:17:08 -08:00
Brian Anderson
58e923de6e rt: Perform task notification before killing the parent task 2011-11-13 16:36:47 -08:00
Brian Anderson
3d9023fa4d rt: Take the task lock when dropping port refcounts 
Sucks, but otherwise there are races when one task drops the refcount to zero
followed by another bumping it again
 2011-11-11 16:11:31 -08:00
Brian Anderson
07771ec25b Fix run-fail/spawnfail 
Catch the case where a parent is killed immediately before it terminates
normally.
 2011-11-11 14:20:00 -08:00
Brian Anderson
2385deaa0d rt: Add locking invariants to rust_task 2011-11-11 12:11:21 -08:00
Brian Anderson
5d1e321ecb rt: Remove rust_chan 2011-11-11 12:11:21 -08:00
Brian Anderson
793da65a8e rt: Move rust_chan::send to rust_port::send 2011-11-10 17:53:19 -08:00
Patrick Walton
c9003d301f Stub a __morestack implementation and stack segment allocation. Untested. 2011-10-31 14:20:56 -07:00
Marijn Haverbeke
8124846b2c Get rid of taskpointer-passing throughout the compiler 
Only intrinsics still take a dummy taskptr. We'll have to do some
makefile stunts to snapshot a version without taskptrs-in-intrinsics.

Issue #466
 2011-10-20 14:22:17 +02:00
Marijn Haverbeke
e927df17f7 Remove spawn_wrap and main_wrap kludges 
This isn't needed now that our functions are cdecl (and was apparently
only still working by accident).

Issue #992
 2011-10-20 13:15:09 +02:00
Patrick Walton
5c973142df rt: Turn on cycle collection at task death; add a test case 2011-09-26 16:59:15 -07:00
Patrick Walton
ad19ab4c6f rt: Make the logic that moves environments between tasks update the GC alloc chain correctly 2011-09-26 16:59:15 -07:00
Patrick Walton
307957710c rt: Remove the GC alloc chain 2011-09-20 14:20:16 -07:00
Brian Anderson
25394950ae Handle the case where a child task tries to kill a parent while it is dying 
Still looks pretty racy
 2011-09-16 10:05:12 -07:00
Brian Anderson
a0ad9a42cd Unsupervise tasks before the scheduler kills them. Unblock before yield->fail 2011-09-14 17:05:35 -07:00
Brian Anderson
103197bc42 Make failure propagation to dead parents work 
The failure will basically go 'through' the dead parent and continue
propagating the failure (as if the child was reparented).
 2011-09-14 15:48:14 -07:00
Brian Anderson
9505d70513 Make linked task failure work again 2011-09-14 14:20:41 -07:00
Brian Anderson
6f6f36172b Remove unused task_exit function 
Issue #236
 2011-09-11 17:31:40 -07:00
Brian Anderson
c047cfb710 Unwind the stack on task failure 
When a task fails, we will throw an exception, then catch it at the bottom of
the stack.

On Windows we don't do this yet because the exception doesn't propagate
correctly.

No cleanups yet.

Issue #236
 2011-09-07 10:32:58 -07:00
Brian Anderson
c337fd5467 Child tasks take a ref to their parents 
This is so that when a child dies after the parent, it still holds a valid
pointer and can call supervisor->kill() safely.
 2011-09-07 10:32:58 -07:00
Brian Anderson
25ae3d655c Rewrite spawn yet again 
The motivation here is that the bottom of each stack needs to contain a C++
try/catch block so that we can unwind. This is already the case for main, but
not spawned tasks.

Issue #236
 2011-09-07 10:32:58 -07:00
Brian Anderson
bb08ffbaf4 Refactor task failure a bit 
Issue #236
 2011-09-07 10:32:58 -07:00
Brian Anderson
abdb6cd71b Rewrite reap_dead_tasks to never grab the sched lock before a task lock 
Doing so contradicts the locking order used everywhere else and causes
deadlocks.

Un-XFAIL task-perf-spawnalot

Closes #854
 2011-08-20 16:21:27 -07:00
Patrick Walton
d0171913aa rt: Remove rustboot's GC infrastructure 2011-08-19 19:17:05 -07:00
Eric Holk
ae89ea223d Making more of the rust_task structure directly accessible from Rust. 2011-08-17 14:42:40 -07:00
Patrick Walton
d8c5bd6195 rt: Implement obstacks, untested as of yet 2011-08-16 19:49:25 -07:00
Eric Holk
8686645aad New channel-based task status notifications. 2011-08-16 16:47:40 -07:00
Eric Holk
cf2def46c1 Removed trans_comm.rs from the compiler. Updating aio/sio to work with the new chan and port system, started on a networking module for the standard library. 2011-08-16 09:36:29 -07:00
Eric Holk
be7325073a Removed spawn and task from the parser. Updated all the tests except for the benchmarks. 2011-08-15 09:26:52 -07:00
Eric Holk
b9f1f77622 Fixed memory accounting and task stack creation bugs. 2011-08-15 09:26:51 -07:00
Eric Holk
b2dad8af31 Added a library version of spawn. Before long, we can remove the old version. 2011-08-15 09:26:51 -07:00
Eric Holk
04af99ecb0 First step towards port handles. 2011-08-15 09:26:51 -07:00
unknown
44bef5f2cb Introduced task handles. 
This is the new way to refer to tasks in rust-land. Currently all they
do is serve as a key to look up the old rust_task structure. Ideally
they won't be ref counted, but baby steps.
 2011-08-08 16:55:38 -07:00
Eric Holk
d1dbb99984 Removing proxies and message queues. 2011-07-29 18:54:59 -07:00
Brian Anderson
4ef1ec580a Do all runtime calls to getenv at initialization 
getenv is not threadsafe and (maybe as a result) it's randomly crashing with
CFLAGS=-g and RUST_THREADS=32. Calls from rust code are still on their
own.
 2011-07-28 12:23:01 -07:00
Eric Holk
279844ce9f Atomic reference counting for tasks. 2011-07-28 10:47:28 -07:00
Eric Holk
5302cde188 Made task threads wait instead of sleep, so they can be woken up. This appears to give us much better parallel performance. 
Also, commented out one more unsafe log and updated rust_kernel.cpp to compile under g++
 2011-07-28 10:47:28 -07:00
Eric Holk
e697a52359 Adding a function to stdlib to set the min stack size, for programs 
that absolutely will not succeed with a large default stack. This
should be removed once we have stack grown working.

Also updated word-count to succeed under the new test framework.
 2011-07-28 10:47:28 -07:00
Eric Holk
62bc6b5113 Per-thread scheduling. Closes #682. 
Tasks are spawned on a random thread. Currently they stay there, but
we should add task migration and load balancing in the future. This
should drammatically improve our task performance benchmarks.
 2011-07-28 10:47:28 -07:00
Eric Holk
b51f5c395c Made root_task no longer special. 2011-07-28 10:47:28 -07:00
Michael Sullivan
f33309502a Bump the stack size and make the RUST_MIN_STACK env variable accept hex values. 2011-07-21 14:05:51 -07:00
Eric Holk
3ae4dcd41e Lots of work on memory tracking and channels. 
We're trying to get closer to doing correct move semantics for channel
operations. This involves a lot of cleanup (such as removing the
unused sched parameter from rust_vec constructor) and making
circular_buffer kernel_owned.

Added tagging for memory allocations. This means we give a string tag
to everything we allocate. If we leak something and TRACK_ALLOCATIONS
is enabled, then it's much easier now to tell exactly what is leaking.
 2011-07-21 11:51:22 -07:00
Brian Anderson
ced8393f20 Modify task::join to indicate how the task terminated 
This involves sticking yet another field into the task structure
 2011-07-15 10:07:29 -07:00
Brian Anderson
4738cf83b0 Move the responsibility for process failure from tasks to the scheduler 
When the root task fails the process fails. Failures on other tasks propagate
up the task tree. Failures on non-root tasks without parents just
(theoretically) unwind and disappear.
 2011-07-14 17:44:39 -07:00
Graydon Hoare
39151f2ad8 Prohibit trailing whitespace under 'tidy' script. Clean up all caught cases. 2011-07-13 15:44:09 -07:00
Graydon Hoare
49116adbe9 Fix compile-command lines in rt. 2011-07-13 13:51:20 -07:00
Graydon Hoare
2e2951305d Remove obsolete nargs counts from runtime. 2011-07-13 13:43:35 -07:00
Graydon Hoare
01fc165517 Attempt to correct buggy win32 timer code (causing tinderbox failures). 2011-07-13 12:25:36 -07:00
Eric Holk
d7db25e8f6 Added an environment variable to override the minimum stack size. Closes #637. 2011-07-08 11:36:56 -07:00
Eric Holk
4739953b84 Fixed two races. 
The first is that the memory_region destructor would complain there is
still an outstanding allocation. This is because circular_buffer from
rust_chan wasn't refing its task, so the task was being destructed too
soon.

The second was where the program could deadlock while joining a
task. The target task would die in the time between checking whether
the task should block and then actually blocking. The fix is to use
the target task's lock.
 2011-07-07 18:22:27 -07:00