b321c5656e
cpu.x86.64: define jit-load-vm as a no-op
...
This was the signal handler prolog code can be shared completely between x86.32 and x86.64.
2011-12-14 11:33:05 -08:00
7c9332e966
cpu.x86.*: clean up signal handler prolog code
...
Also bump the signal handler frame sizes up again since it looks like we shrank them too much.
2011-12-14 09:56:50 -08:00
079025559d
simplify function prologs
2011-12-14 09:56:49 -08:00
ee656387ea
cpu.*.bootstrap: return address for (set-context)
...
Instead of pushing zero as the fake return address for a set-context call, push an address that points back into the primitive so that callstack walking can match the frame back to the subprimitive.
2011-12-14 09:56:46 -08:00
1911905c6b
vm: frame size for signal handler code blocks
2011-12-14 09:56:46 -08:00
943596575a
use radix literals
2011-11-23 19:03:40 -08:00
e0dc236725
cpu.*.bootstrap: clear faulting flag directly
...
No reason to go through a safepoint in unwind-native-frames really.
2011-11-12 13:02:57 -08:00
e24400679f
vm: more defense against multi-faulting
...
* Clear faulting_p from a safepoint rather than inside general_error, because jumping into unwind-native-frames could blow up.
* Handle multiple faults from fatal_error by breakpointing. Is there anything else we can safely do at that point?
* Verify memory protection faults in the top half of the signal handlers because signal dispatch could fault. Treat memory faults during gc or fep as fatal errors.
* Add a function factor_vm::abort() that restores the default SIGABRT handler and ::abort()s. Use it from fatal_error() so we get useful context from gdb and so the user gets feedback from the system crash reporter that Factor blew up and didn't just disappear.
* In factorbug(), don't proceed with .s .r .c if it would be unsafe to do so.
* Don't pile on signals if we've already called fatal_error().
2011-11-12 13:02:57 -08:00
c736b04ada
cpu.x86.*.bootstrap: save flags in signal-handlers
2011-11-04 11:56:01 -07:00
d9293337bb
cpu.x86.64.bootstrap: save all the regs on signal
...
Appears to fix another sporadic crash from mason on x86-64. There's no reason to cut corners here, really.
2011-11-04 10:01:56 -07:00
6bb46a3f1e
vm: always generate safepoints in jit
...
Even if there's no stack frame we still need to safepoint before leaving the function. Fixes #332 .
2011-11-03 11:57:43 -07:00
2d5360778f
cleanups from code review
2011-10-27 21:18:23 -07:00
c70a1004f8
cpu.x86.64: also must save ctx-reg
2011-10-27 21:18:21 -07:00
ff2d480d98
cpu.x86.32: update for resumable signal handlers
2011-10-27 21:18:19 -07:00
d6eae2f42d
vm: suppress signal-handler stack frame generation
...
This is a hack; see #295
2011-10-27 21:18:19 -07:00
402e1155a5
vm: resumably handle signals from leaf procedures
2011-10-27 21:18:18 -07:00
1386212d23
vm: dispatch signal handlers through subprimitive
...
We also need to save C ABI volatile registers before calling the signal handler in order to be able to reliably resume. Add signal-handler and leaf-signal-handler subprimitives to preserve volatile registers before invoking the signal handler C function.
2011-10-27 21:18:18 -07:00
68518e741e
cpu.x86.(32,64).bootstrap: get rid of PUSH malarky
...
In the function prologue, move the stack pointer once after the stack frame is set up instead of PUSHing the stack frame piece by piece.
2011-10-27 21:14:50 -07:00
4be5eadb34
cpu.x86.bootstrap: add safepoint to jit-epilog
2011-10-27 21:14:45 -07:00
f8e475485a
Add RT_INLINE_CACHE_MISS relocation type to avoid frequent dlsym lookups when compiling PICs. Fixes #123
2011-10-03 00:26:21 -07:00
05f7419f2f
Rework non-optimizing compiler backend to use compiler.codegen.relocation
2011-10-03 00:26:21 -07:00
d2cf2d8f92
Rename get-fpu-state entry point to fpu-state, and fix stack-checker unit tests
2010-09-04 12:58:59 -07:00
cdc9538540
cpu.x86.64: fix bootstrap
2010-09-02 21:17:14 -07:00
1818bcb780
Fix three problems discovered by running math.floats.env tests in a loop:
...
- Crash if allocating error triggers a GC from a signal/SEH handler
- Crash if GC runs with floating point traps enabled on Windows
- Floating point traps didn't prettyprint properly
2010-09-02 22:57:14 -05:00
451237a028
Fix start-context-and-delete sub-primitive (reported by Sascha Matzke)
2010-08-02 20:28:23 -04:00
f2646fc92c
cpu.x86: save context before calling new_context() since it can trigger GC (bug reported by Sascha Matzke)
2010-07-30 01:07:53 -04:00
a55c8ee671
FFI rewrite part 6: deconcatenatize
2010-07-02 15:44:12 -04:00
70b8a74208
vm: have to be extra careful when messing with return addresses
2010-06-13 18:12:57 -04:00
3f28c0e7f5
vm: context switching needs to scrub the return address
2010-06-12 19:54:31 -04:00
1434a305c8
vm: add a new rc-absolute-1 relocation class to allow storing values in 8-bit operands, and optimized code sequences for inline caches using this
2010-04-12 14:22:41 -07:00
f828f9fdaa
Fiddle with register assignments in non-optimizing x86-32 backend
2010-04-10 00:10:05 -07:00
430a05dcea
Store stack bounds in TIB on win64 to make C++ exceptions work
2010-04-05 19:07:10 -04:00
c0af678c5b
cpu.x86.assembler: add support for absolute addressing on x86-64; [RIP+] now behaves like [] did, and [] now does absolute addressing just like in 32-bit mode
2010-04-04 19:42:57 -04:00
b16d91576c
cpu.x86.64: fix typo that caused bootstrap crash
2010-04-03 21:11:04 -04:00
74640b7f71
Get green threads working on Windows
...
- store stack base and limit in TIB
- set up a frame-based structured exception handler in each context's callstack
- boot.x86.32.image has now been replaced by boot.winnt-x86.32.image and boot.unix-x86.32.image
2010-04-03 20:24:33 -04:00
0faa3bcf4a
vm: pre-allocate context alien
2010-04-02 00:02:21 -04:00
ef53e6ecd5
cpu.x86.64: eliminate useless instruction from primitive call sequence for a marginal to non-existent gain
2010-03-31 19:32:36 -04:00
8f0487f1c3
cpu.x86: remove useless crap from c-to-factor sub-primitive
2010-03-31 15:19:00 -04:00
fb2ecab614
threads: delete old contexts immediately instead of handing them off to a 'context recycler' thread
2010-03-30 21:47:48 -04:00
b6dfdcb909
set-context and start-context primitives can now pass parameters around
2010-03-27 08:13:57 -04:00
1717b8d0f7
Implement start-context and set-context primitives
2010-03-27 02:56:11 -04:00
560c119cd2
vm: actually use context callstacks when running code
2010-03-26 22:44:43 -04:00
d5a25f99bf
cpu.x86.64: inline cache miss blocks have a prolog generated in the middle of a procedure; undo a recent change to non-optimizing backend to fix random crashes
2010-02-06 18:45:27 +13:00
22fbe01c17
Compiled code size optimizations for x86-64 resulting in development image size reduction of about 350kb
...
- RIP-relative LEA is now used to load program counter in prolog and tail calls; saves 3 bytes of code and 4 bytes of relocation information over an immediate MOV
- VM field accesses no longer involve a redundant addition
2010-02-03 23:11:35 +13:00
5606825e8b
vm: remove primitive table, non-optimizing compiler now looks up primitives with dlsym()
2010-01-19 20:00:33 +13:00
637a77c018
Stop wearing monocle and use the term "entry point" instead of "XT" throughout VM and compiler; also remove two unused relocation types
2010-01-18 20:54:00 +13:00
85a45d63d9
unwind-stack-frames didn't load VM pointer into the VM register on PowerPC and x86-64, and so if C code had clobbered this register it would crash
2010-01-12 16:31:10 +13:00
8d34a0f3c1
Store VM object in a register on x86-64
2010-01-11 01:20:32 +13:00
5b58117be1
cpu.x86: c-to-factor needs to deal with the fact that vm-reg might have been clobbered
2010-01-09 19:09:48 +13:00
1f88a925b4
Fix set-callstack primitive on Windows
2010-01-07 19:31:15 +13:00
Joe Groff
Slava Pestov
Slava Pestov