1 # x86/x86_64 support for -fsplit-stack.
2 # Copyright (C) 2009, 2010 Free Software Foundation, Inc.
3 # Contributed by Ian Lance Taylor <iant@google.com>.
5 # This file is part of GCC.
7 # GCC is free software; you can redistribute it and/or modify it under
8 # the terms of the GNU General Public License as published by the Free
9 # Software Foundation; either version 3, or (at your option) any later
12 # GCC is distributed in the hope that it will be useful, but WITHOUT ANY
13 # WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 # FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
17 # Under Section 7 of GPL version 3, you are granted additional
18 # permissions described in the GCC Runtime Library Exception, version
19 # 3.1, as published by the Free Software Foundation.
21 # You should have received a copy of the GNU General Public License and
22 # a copy of the GCC Runtime Library Exception along with this program;
23 # see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
24 # <http://www.gnu.org/licenses/>.
27 # Support for allocating more stack space when using -fsplit-stack.
28 # When a function discovers that it needs more stack space, it will
29 # call __morestack with the size of the stack frame and the size of
30 # the parameters to copy from the old stack frame to the new one.
31 # The __morestack function preserves the parameter registers and
32 # calls __generic_morestack to actually allocate the stack space.
34 # When this is called stack space is very low, but we ensure that
35 # there is enough space to push the parameter registers and to call
36 # __generic_morestack.
38 # When calling __generic_morestack, FRAME_SIZE points to the size of
39 # the desired frame when the function is called, and the function
40 # sets it to the size of the allocated stack. OLD_STACK points to
41 # the parameters on the old stack and PARAM_SIZE is the number of
42 # bytes of parameters to copy to the new stack. These are the
43 # parameters of the function that called __morestack. The
44 # __generic_morestack function returns the new stack pointer,
45 # pointing to the address of the first copied parameter. The return
46 # value minus the returned *FRAME_SIZE will be the first address on
47 # the stack which we should not use.
49 # void *__generic_morestack (size_t *frame_size, void *old_stack,
52 # The __morestack routine has to arrange for the caller to return to a
53 # stub on the new stack. The stub is responsible for restoring the
54 # old stack pointer and returning to the caller's caller. This calls
55 # __generic_releasestack to retrieve the old stack pointer and release
56 # the newly allocated stack.
58 # void *__generic_releasestack (size_t *available);
60 # We do a little dance so that the processor's call/return return
61 # address prediction works out. The compiler arranges for the caller
63 # call __generic_morestack
66 # // carry on with function
67 # After we allocate more stack, we call L, which is in our caller.
68 # When that returns (to the predicted instruction), we release the
69 # stack segment and reset the stack pointer. We then return to the
70 # predicted instruction, namely the ret instruction immediately after
71 # the call to __generic_morestack. That then returns to the caller of
72 # the original caller.
75 # The amount of extra space we ask for. In general this has to be
76 # enough for the dynamic loader to find a symbol and for a signal
80 #define BACKOFF (1024)
82 #define BACKOFF (1536)
86 # This entry point is for split-stack code which calls non-split-stack
87 # code. When the linker sees this case, it converts the call to
88 # __morestack to call __morestack_non_split instead. We just bump the
89 # requested stack space by 16K.
91 .global __morestack_non_split
92 .hidden __morestack_non_split
95 .type __morestack_non_split,@function
98 __morestack_non_split:
107 .size __morestack_non_split, . - __morestack_non_split
110 # __morestack_non_split falls through into __morestack.
113 # The __morestack function.
119 .type __morestack,@function
130 # The 32-bit __morestack function.
132 # We use a cleanup to restore the stack guard if an exception
133 # is thrown through this code.
135 .cfi_personality 0,__gcc_personality_v0
138 .cfi_personality 0x9b,DW.ref.__gcc_personality_v0
139 .cfi_lsda 0x1b,.LLSDA1
142 # Set up a normal backtrace.
144 .cfi_def_cfa_offset 8
147 .cfi_def_cfa_register %ebp
149 # We return below with a ret $8. We will return to a single
150 # return instruction, which will return to the caller of our
151 # caller. We let the unwinder skip that single return
152 # instruction, and just return to the real caller.
154 .cfi_escape 0x15, 4, 0x7d # DW_CFA_val_offset_sf, %esp, 12/-4
156 # In 32-bit mode the parameters are pushed on the stack. The
157 # argument size is pushed then the new stack frame size is
160 # In 32-bit mode the registers %eax, %edx, and %ecx may be
161 # used for parameters, depending on the regparm and fastcall
168 call __morestack_block_signals
170 pushl 12(%ebp) # The size of the parameters.
171 leal 20(%ebp),%eax # Address of caller's parameters.
173 addl $BACKOFF,8(%ebp) # Ask for backoff bytes.
174 leal 8(%ebp),%eax # The address of the new frame size.
177 # Note that %esp is exactly 32 bytes below the CFA -- perfect for
178 # a 16-byte aligned stack. That said, we still ought to compile
179 # generic-morestack.c with -mpreferred-stack-boundary=2. FIXME.
180 call __generic_morestack
182 movl %eax,%esp # Switch to the new stack.
183 subl 8(%ebp),%eax # The end of the stack space.
184 addl $BACKOFF,%eax # Back off 512 bytes.
187 # FIXME: The offset must match
188 # TARGET_THREAD_SPLIT_STACK_OFFSET in
189 # gcc/config/i386/linux.h.
190 movl %eax,%gs:0x30 # Save the new stack boundary.
192 call __morestack_unblock_signals
194 movl -8(%ebp),%edx # Restore registers.
197 movl 4(%ebp),%eax # Increment the return address
198 cmpb $0xc3,(%eax) # to skip the ret instruction;
203 movl %eax,-8(%ebp) # Store return address in an
206 movl -4(%ebp),%eax # Restore the last register.
208 call *-8(%ebp) # Call our caller!
210 # The caller will return here, as predicted.
212 # Save the registers which may hold a return value. We
213 # assume that __generic_releasestack does not touch any
214 # floating point or vector registers.
218 # Push the arguments to __generic_releasestack now so that the
219 # stack is at a 16-byte boundary for
220 # __morestack_block_signals.
221 pushl $0 # Where the available space is returned.
222 leal 0(%esp),%eax # Push its address.
225 call __morestack_block_signals
227 call __generic_releasestack
229 subl 4(%esp),%eax # Subtract available space.
230 addl $BACKOFF,%eax # Back off 512 bytes.
232 movl %eax,%gs:0x30 # Save the new stack boundary.
234 addl $8,%esp # Remove values from stack.
236 # We need to restore the old stack pointer, which is in %rbp,
237 # before we unblock signals. We also need to restore %eax and
238 # %edx after we unblock signals but before we return. Do this
239 # by moving %eax and %edx from the current stack to the old
242 popl %edx # Pop return value from current stack.
245 movl %ebp,%esp # Restore stack pointer.
247 pushl %eax # Push return value on old stack.
249 subl $8,%esp # Align stack to 16-byte boundary.
251 call __morestack_unblock_signals
254 popl %edx # Restore return value.
260 .cfi_def_cfa %esp, 12
261 ret $8 # Return to caller, which will
262 # immediately return. Pop
263 # arguments as we go.
265 # This is the cleanup code called by the stack unwinder when unwinding
266 # through the code between .LEHB0 and .LEHE0 above.
270 subl $16,%esp # Maintain 16 byte alignment.
271 movl %eax,4(%esp) # Save exception header.
272 movl %ebp,(%esp) # Stack pointer after resume.
273 call __generic_findstack
274 movl %ebp,%ecx # Get the stack pointer.
275 subl %eax,%ecx # Subtract available space.
276 addl $BACKOFF,%ecx # Back off 512 bytes.
277 movl %ecx,%gs:0x30 # Save new stack boundary.
278 movl 4(%esp),%eax # Function argument.
281 call __x86.get_pc_thunk.bx # %ebx may not be set up for us.
282 addl $_GLOBAL_OFFSET_TABLE_, %ebx
283 call _Unwind_Resume@PLT # Resume unwinding.
288 #else /* defined(__x86_64__) */
291 # The 64-bit __morestack function.
293 # We use a cleanup to restore the stack guard if an exception
294 # is thrown through this code.
296 .cfi_personality 0x3,__gcc_personality_v0
297 .cfi_lsda 0x3,.LLSDA1
299 .cfi_personality 0x9b,DW.ref.__gcc_personality_v0
300 .cfi_lsda 0x1b,.LLSDA1
303 # Set up a normal backtrace.
305 .cfi_def_cfa_offset 16
306 .cfi_offset %rbp, -16
308 .cfi_def_cfa_register %rbp
310 # We will return a single return instruction, which will
311 # return to the caller of our caller. Let the unwinder skip
312 # that single return instruction, and just return to the real
315 .cfi_escape 0x15, 7, 0x7f # DW_CFA_val_offset_sf, %esp, 8/-8
317 # In 64-bit mode the new stack frame size is passed in r10
318 # and the argument size is passed in r11.
320 addq $BACKOFF,%r10 # Ask for backoff bytes.
321 pushq %r10 # Save new frame size.
323 # In 64-bit mode the registers %rdi, %rsi, %rdx, %rcx, %r8,
324 # and %r9 may be used for parameters. We also preserve %rax
325 # which the caller may use to hold %r10.
336 pushq $0 # For alignment.
338 call __morestack_block_signals
340 leaq -8(%rbp),%rdi # Address of new frame size.
341 leaq 24(%rbp),%rsi # The caller's parameters.
343 popq %rdx # The size of the parameters.
345 call __generic_morestack
347 movq -8(%rbp),%r10 # Reload modified frame size
348 movq %rax,%rsp # Switch to the new stack.
349 subq %r10,%rax # The end of the stack space.
350 addq $BACKOFF,%rax # Back off 1024 bytes.
353 # FIXME: The offset must match
354 # TARGET_THREAD_SPLIT_STACK_OFFSET in
355 # gcc/config/i386/linux64.h.
356 movq %rax,%fs:0x70 # Save the new stack boundary.
358 call __morestack_unblock_signals
360 movq -24(%rbp),%rdi # Restore registers.
367 movq 8(%rbp),%r10 # Increment the return address
368 incq %r10 # to skip the ret instruction;
371 movq -16(%rbp),%rax # Restore caller's %rax.
373 call *%r10 # Call our caller!
375 # The caller will return here, as predicted.
377 # Save the registers which may hold a return value. We
378 # assume that __generic_releasestack does not touch any
379 # floating point or vector registers.
383 call __morestack_block_signals
385 pushq $0 # For alignment.
386 pushq $0 # Where the available space is returned.
387 leaq 0(%rsp),%rdi # Pass its address.
389 call __generic_releasestack
391 subq 0(%rsp),%rax # Subtract available space.
392 addq $BACKOFF,%rax # Back off 1024 bytes.
394 movq %rax,%fs:0x70 # Save the new stack boundary.
396 addq $16,%rsp # Remove values from stack.
398 # We need to restore the old stack pointer, which is in %rbp,
399 # before we unblock signals. We also need to restore %rax and
400 # %rdx after we unblock signals but before we return. Do this
401 # by moving %rax and %rdx from the current stack to the old
404 popq %rdx # Pop return value from current stack.
407 movq %rbp,%rsp # Restore stack pointer.
409 pushq %rax # Push return value on old stack.
412 call __morestack_unblock_signals
414 popq %rdx # Restore return value.
421 ret # Return to caller, which will
422 # immediately return.
424 # This is the cleanup code called by the stack unwinder when unwinding
425 # through the code between .LEHB0 and .LEHE0 above.
429 subq $16,%rsp # Maintain 16 byte alignment.
430 movq %rax,(%rsp) # Save exception header.
431 movq %rbp,%rdi # Stack pointer after resume.
432 call __generic_findstack
433 movq %rbp,%rcx # Get the stack pointer.
434 subq %rax,%rcx # Subtract available space.
435 addq $BACKOFF,%rcx # Back off 1024 bytes.
436 movq %rcx,%fs:0x70 # Save new stack boundary.
437 movq (%rsp),%rdi # Restore exception data for call.
439 call _Unwind_Resume@PLT # Resume unwinding.
441 call _Unwind_Resume # Resume unwinding.
444 #endif /* defined(__x86_64__) */
448 .size __morestack, . - __morestack
451 #if !defined(__x86_64__) && defined(__PIC__)
452 # Output the thunk to get PC into bx, since we use it above.
453 .section .text.__x86.get_pc_thunk.bx,"axG",@progbits,__x86.get_pc_thunk.bx,comdat
454 .globl __x86.get_pc_thunk.bx
455 .hidden __x86.get_pc_thunk.bx
457 .type __x86.get_pc_thunk.bx, @function
459 __x86.get_pc_thunk.bx:
465 .size __x86.get_pc_thunk.bx, . - __x86.get_pc_thunk.bx
469 # The exception table. This tells the personality routine to execute
470 # the exception handler.
472 .section .gcc_except_table,"a",@progbits
475 .byte 0xff # @LPStart format (omit)
476 .byte 0xff # @TType format (omit)
477 .byte 0x1 # call-site format (uleb128)
478 .uleb128 .LLSDACSE1-.LLSDACSB1 # Call-site table length
480 .uleb128 .LEHB0-.LFB1 # region 0 start
481 .uleb128 .LEHE0-.LEHB0 # length
482 .uleb128 .L1-.LFB1 # landing pad
487 .global __gcc_personality_v0
489 # Build a position independent reference to the basic
490 # personality function.
491 .hidden DW.ref.__gcc_personality_v0
492 .weak DW.ref.__gcc_personality_v0
493 .section .data.DW.ref.__gcc_personality_v0,"awG",@progbits,DW.ref.__gcc_personality_v0,comdat
494 .type DW.ref.__gcc_personality_v0, @object
495 DW.ref.__gcc_personality_v0:
498 .size DW.ref.__gcc_personality_v0, 4
499 .long __gcc_personality_v0
502 .size DW.ref.__gcc_personality_v0, 8
503 .quad __gcc_personality_v0
509 # This entry point is used for the large model. With this entry point
510 # the upper 32 bits of %r10 hold the argument size and the lower 32
511 # bits hold the new stack frame size. There doesn't seem to be a way
512 # to know in the assembler code that we are assembling for the large
513 # model, and there doesn't seem to be a large model multilib anyhow.
514 # If one is developed, then the non-PIC code is probably OK since we
515 # will probably be close to the morestack code, but the PIC code
516 # almost certainly needs to be changed. FIXME.
519 .global __morestack_large_model
520 .hidden __morestack_large_model
523 .type __morestack_large_model,@function
526 __morestack_large_model:
531 andl $0xffffffff, %r10d
537 .size __morestack_large_model, . - __morestack_large_model
540 #endif /* __x86_64__ */
542 # Initialize the stack test value when the program starts or when a
543 # new thread starts. We don't know how large the main stack is, so we
544 # guess conservatively. We might be able to use getrlimit here.
547 .global __stack_split_initialize
548 .hidden __stack_split_initialize
551 .type __stack_split_initialize, @function
554 __stack_split_initialize:
558 leal -16000(%esp),%eax # We should have at least 16K.
563 call __generic_morestack_set_initial_sp@PLT
565 call __generic_morestack_set_initial_sp
570 #else /* defined(__x86_64__) */
572 leaq -16000(%rsp),%rax # We should have at least 16K.
577 call __generic_morestack_set_initial_sp@PLT
579 call __generic_morestack_set_initial_sp
583 #endif /* defined(__x86_64__) */
586 .size __stack_split_initialize, . - __stack_split_initialize
590 # Make __stack_split_initialize a high priority constructor. FIXME:
591 # This is ELF specific.
593 .section .ctors.65535,"aw",@progbits
597 .long __stack_split_initialize
598 .long __morestack_load_mmap
601 .quad __stack_split_initialize
602 .quad __morestack_load_mmap
606 .section .note.GNU-stack,"",@progbits
607 .section .note.GNU-split-stack,"",@progbits
608 .section .note.GNU-no-split-stack,"",@progbits