/* * very-low-level utilities for runtime support */ /* * This software is part of the SBCL system. See the README file for * more information. * * This software is derived from the CMU CL system, which was * written at Carnegie Mellon University and released into the * public domain. The software is in the public domain and is * provided with absolutely no warranty. See the COPYING and CREDITS * files for more information. */ #define LANGUAGE_ASSEMBLY #include "sbcl.h" #include "validate.h" #include "genesis/closure.h" #include "genesis/fdefn.h" #include "genesis/static-symbols.h" #include "genesis/symbol.h" #include "genesis/thread.h" /* Minimize conditionalization for different OS naming schemes. * * (As of sbcl-0.8.10, this seems no longer to be much of an issue, * since everyone has converged on ELF. If this generality really * turns out not to matter, perhaps it's just clutter we could get * rid of? -- WHN 2004-04-18) * * (Except Win32, which is unlikely ever to be ELF, sorry. -- AB 2005-12-08) */ #if defined __linux__ || defined __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ || defined __sun #define GNAME(var) var #else #define GNAME(var) _##var #endif #if defined __linux__ || defined __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ || defined __sun #define GNAMEDOLLAR(var) $##var #else #define GNAMEDOLLAR(var) $_##var #endif #if defined __linux__ || defined __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ || defined __sun #define DOLLARLITERAL(var) $##var #else #define DOLLARLITERAL(var) $##(var) #endif /* Get the right type of alignment. Linux, FreeBSD and NetBSD (but not OpenBSD) * want alignment in bytes. * * (As in the GNAME() definitions above, as of sbcl-0.8.10, this seems * no longer to be much of an issue, since everyone has converged on * the same value. If this generality really turns out not to * matter any more, perhaps it's just clutter we could get * rid of? -- WHN 2004-04-18) */ #if defined(__linux__) || defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__sun) || defined(LISP_FEATURE_WIN32) #define align_4byte 4 #define align_8byte 8 #define align_16byte 16 #else #define align_4byte 2 #define align_8byte 3 #define align_16byte 4 #endif /* * The assembler used for win32 doesn't like .type or .size directives, * so we want to conditionally kill them out. So let's wrap them in macros * that are defined to be no-ops on win32. Hopefully this still works on * other platforms. */ #if !defined(LISP_FEATURE_WIN32) && !defined(LISP_FEATURE_DARWIN) #define TYPE(name) .type name,@function #define SIZE(name) .size name,.-name #else #define TYPE(name) #define SIZE(name) #endif /* * x86/darwin (as of MacOS X 10.4.5) doesn't reliably file signal * handlers (SIGTRAP or Mach exception handlers) for 0xCC, wo we have * to use ud2 instead. ud2 is an undefined opcode, #x0b0f, or * 0F 0B in low-endian notation, that causes SIGILL to fire. We check * for this instruction in the SIGILL handler and if we see it, we * advance the EIP by two bytes to skip over ud2 instruction and * call sigtrap_handler. */ #if defined(LISP_FEATURE_DARWIN) #define END() #define TRAP ud2 #else #define END() .end #define TRAP int3 #endif .text .globl GNAME(foreign_function_call_active) .globl GNAME(all_threads) /* * A call to call_into_c preserves esi, edi, and ebp. * (The C function will preserve ebx, esi, edi, and ebp across its * function call, but we trash ebx ourselves by using it to save the * return Lisp address.) * * Return values are in eax and maybe edx for quads, or st(0) for * floats. * * This should work for Lisp calls C calls Lisp calls C.. */ .text .align align_16byte,0x90 .globl GNAME(call_into_c) TYPE(GNAME(call_into_c)) GNAME(call_into_c): movl $1,GNAME(foreign_function_call_active) /* Save the return Lisp address in ebx. */ popl %ebx /* Setup the NPX for C */ fstp %st(0) fstp %st(0) fstp %st(0) fstp %st(0) fstp %st(0) fstp %st(0) fstp %st(0) fstp %st(0) #ifdef LISP_FEATURE_WIN32 cld #endif #ifdef LISP_FEATURE_DARWIN andl $0xfffffff0,%esp # align stack to 16-byte boundary before calling C #endif call *%eax # normal callout using Lisp stack movl %eax,%ecx # remember integer return value /* Check for a return FP value. */ fxam fnstsw %eax andl $0x4500,%eax cmpl $0x4100,%eax jne Lfp_rtn_value /* The return value is in eax, or eax,edx? */ /* Set up the NPX stack for Lisp. */ fldz # Ensure no regs are empty. fldz fldz fldz fldz fldz fldz fldz /* Restore the return value. */ movl %ecx,%eax # maybe return value movl $0,GNAME(foreign_function_call_active) /* Return. */ jmp *%ebx Lfp_rtn_value: /* The return result is in st(0). */ /* Set up the NPX stack for Lisp, placing the result in st(0). */ fldz # Ensure no regs are empty. fldz fldz fldz fldz fldz fldz fxch %st(7) # Move the result back to st(0). /* We don't need to restore eax, because the result is in st(0). */ movl $0,GNAME(foreign_function_call_active) /* Return. */ jmp *%ebx SIZE(GNAME(call_into_c)) .text .globl GNAME(call_into_lisp_first_time) TYPE(GNAME(call_into_lisp_first_time)) /* The *ALIEN-STACK* pointer is set up on the first call_into_lisp when * the stack changes. We don't worry too much about saving registers * here, because we never expect to return from the initial call to lisp * anyway */ .align align_16byte,0x90 GNAME(call_into_lisp_first_time): pushl %ebp # Save old frame pointer. movl %esp,%ebp # Establish new frame. #ifndef LISP_FEATURE_WIN32 movl %esp,ALIEN_STACK + SYMBOL_VALUE_OFFSET movl GNAME(all_threads),%eax movl THREAD_CONTROL_STACK_START_OFFSET(%eax) ,%esp /* don't think too hard about what happens if we get interrupted * here */ addl DOLLARLITERAL(THREAD_CONTROL_STACK_SIZE),%esp #else /* Win32 -really- doesn't like you switching stacks out from under it. */ movl GNAME(all_threads),%eax #endif jmp Lstack .text .globl GNAME(call_into_lisp) TYPE(GNAME(call_into_lisp)) /* The C conventions require that ebx, esi, edi, and ebp be preserved * across function calls. */ .align align_16byte,0x90 GNAME(call_into_lisp): pushl %ebp # Save old frame pointer. movl %esp,%ebp # Establish new frame. Lstack: /* Save the NPX state */ fwait # Catch any pending NPX exceptions. subl $108,%esp # Make room for the NPX state. fnsave (%esp) # save and reset NPX movl (%esp),%eax # Load NPX control word. andl $0xfffff2ff,%eax # Set rounding mode to nearest. orl $0x00000200,%eax # Set precision to 64 bits. (53-bit mantissa) pushl %eax fldcw (%esp) # Recover modes. popl %eax fldz # Ensure no FP regs are empty. fldz fldz fldz fldz fldz fldz fldz /* Save C regs: ebx esi edi. */ pushl %ebx pushl %esi pushl %edi /* Clear descriptor regs. */ xorl %eax,%eax # lexenv xorl %ebx,%ebx # available xorl %ecx,%ecx # arg count xorl %edx,%edx # first arg xorl %edi,%edi # second arg xorl %esi,%esi # third arg /* no longer in function call */ movl %eax, GNAME(foreign_function_call_active) movl %esp,%ebx # remember current stack pushl %ebx # Save entry stack on (maybe) new stack. /* Establish Lisp args. */ movl 8(%ebp),%eax # lexenv? movl 12(%ebp),%ebx # address of arg vec movl 16(%ebp),%ecx # num args shll $2,%ecx # Make num args into fixnum. cmpl $0,%ecx je Ldone movl (%ebx),%edx # arg0 cmpl $4,%ecx je Ldone movl 4(%ebx),%edi # arg1 cmpl $8,%ecx je Ldone movl 8(%ebx),%esi # arg2 Ldone: /* Registers eax, ecx, edx, edi, and esi are now live. */ /* Alloc new frame. */ mov %esp,%ebx # The current sp marks start of new frame. push %ebp # fp in save location S0 sub $8,%esp # Ensure 3 slots are allocated, one above. mov %ebx,%ebp # Switch to new frame. call *CLOSURE_FUN_OFFSET(%eax) /* If the function returned multiple values, it will return to this point. Lose them */ jnc LsingleValue mov %ebx, %esp LsingleValue: /* A singled value function returns here */ /* Restore the stack, in case there was a stack change. */ popl %esp # c-sp /* Restore C regs: ebx esi edi. */ popl %edi popl %esi popl %ebx /* Restore the NPX state. */ frstor (%esp) addl $108, %esp popl %ebp # c-sp movl %edx,%eax # c-val ret SIZE(GNAME(call_into_lisp)) /* support for saving and restoring the NPX state from C */ .text .globl GNAME(fpu_save) TYPE(GNAME(fpu_save)) .align 2,0x90 GNAME(fpu_save): movl 4(%esp),%eax fnsave (%eax) # Save the NPX state. (resets NPX) ret SIZE(GNAME(fpu_save)) .globl GNAME(fpu_restore) TYPE(GNAME(fpu_restore)) .align 2,0x90 GNAME(fpu_restore): movl 4(%esp),%eax frstor (%eax) # Restore the NPX state. ret SIZE(GNAME(fpu_restore)) /* * the undefined-function trampoline */ .text .align align_4byte,0x90 .globl GNAME(undefined_tramp) TYPE(GNAME(undefined_tramp)) .byte 0, 0, 0, SIMPLE_FUN_HEADER_WIDETAG GNAME(undefined_tramp): TRAP .byte trap_Error .byte 2 .byte UNDEFINED_FUN_ERROR .byte sc_DescriptorReg # eax in the Descriptor-reg SC ret SIZE(GNAME(undefined_tramp)) /* * the closure trampoline */ .text .align align_4byte,0x90 .globl GNAME(closure_tramp) TYPE(GNAME(closure_tramp)) .byte 0, 0, 0, SIMPLE_FUN_HEADER_WIDETAG GNAME(closure_tramp): movl FDEFN_FUN_OFFSET(%eax),%eax /* FIXME: The '*' after "jmp" in the next line is from PVE's * patch posted to the CMU CL mailing list Oct 6, 1999. It looks * reasonable, and it certainly seems as though if CMU CL needs it, * SBCL needs it too, but I haven't actually verified that it's * right. It would be good to find a way to force the flow of * control through here to test it. */ jmp *CLOSURE_FUN_OFFSET(%eax) SIZE(GNAME(closure_tramp)) /* * fun-end breakpoint magic */ .text .globl GNAME(fun_end_breakpoint_guts) .align align_4byte GNAME(fun_end_breakpoint_guts): /* Multiple Value return */ jc multiple_value_return /* Single value return: The eventual return will now use the multiple values return convention but with a return values count of one. */ movl %esp,%ebx # Setup ebx - the ofp. subl $4,%esp # Allocate one stack slot for the return value movl $4,%ecx # Setup ecx for one return value. movl DOLLARLITERAL(NIL),%edi # default second value movl DOLLARLITERAL(NIL),%esi # default third value multiple_value_return: .globl GNAME(fun_end_breakpoint_trap) GNAME(fun_end_breakpoint_trap): TRAP .byte trap_FunEndBreakpoint hlt # We should never return here. .globl GNAME(fun_end_breakpoint_end) GNAME(fun_end_breakpoint_end): .globl GNAME(do_pending_interrupt) TYPE(GNAME(do_pending_interrupt)) .align align_4byte,0x90 GNAME(do_pending_interrupt): TRAP .byte trap_PendingInterrupt ret SIZE(GNAME(do_pending_interrupt)) /* * Allocate bytes and return the start of the allocated space * in the specified destination register. * * In the general case the size will be in the destination register. * * All registers must be preserved except the destination. * The C conventions will preserve ebx, esi, edi, and ebp. * So only eax, ecx, and edx need special care here. */ .globl GNAME(alloc_to_eax) TYPE(GNAME(alloc_to_eax)) .align align_4byte,0x90 GNAME(alloc_to_eax): pushl %ecx # Save ecx and edx as C could destroy them. pushl %edx pushl %eax # Push the size. call GNAME(alloc) addl $4,%esp # Pop the size arg. popl %edx # Restore ecx and edx. popl %ecx ret SIZE(GNAME(alloc_to_eax)) .globl GNAME(alloc_8_to_eax) TYPE(GNAME(alloc_8_to_eax)) .align align_4byte,0x90 GNAME(alloc_8_to_eax): pushl %ecx # Save ecx and edx as C could destroy them. pushl %edx pushl $8 # Push the size. call GNAME(alloc) addl $4,%esp # Pop the size arg. popl %edx # Restore ecx and edx. popl %ecx ret SIZE(GNAME(alloc_8_to_eax)) .globl GNAME(alloc_8_to_eax) TYPE(GNAME(alloc_8_to_eax)) .align align_4byte,0x90 .globl GNAME(alloc_16_to_eax) TYPE(GNAME(alloc_16_to_eax)) .align align_4byte,0x90 GNAME(alloc_16_to_eax): pushl %ecx # Save ecx and edx as C could destroy them. pushl %edx pushl $16 # Push the size. call GNAME(alloc) addl $4,%esp # Pop the size arg. popl %edx # Restore ecx and edx. popl %ecx ret SIZE(GNAME(alloc_16_to_eax)) .globl GNAME(alloc_to_ecx) TYPE(GNAME(alloc_to_ecx)) .align align_4byte,0x90 GNAME(alloc_to_ecx): pushl %eax # Save eax and edx as C could destroy them. pushl %edx pushl %ecx # Push the size. call GNAME(alloc) addl $4,%esp # Pop the size arg. movl %eax,%ecx # Set up the destination. popl %edx # Restore eax and edx. popl %eax ret SIZE(GNAME(alloc_to_ecx)) .globl GNAME(alloc_8_to_ecx) TYPE(GNAME(alloc_8_to_ecx)) .align align_4byte,0x90 GNAME(alloc_8_to_ecx): pushl %eax # Save eax and edx as C could destroy them. pushl %edx pushl $8 # Push the size. call GNAME(alloc) addl $4,%esp # Pop the size arg. movl %eax,%ecx # Set up the destination. popl %edx # Restore eax and edx. popl %eax ret SIZE(GNAME(alloc_8_to_ecx)) .globl GNAME(alloc_16_to_ecx) TYPE(GNAME(alloc_16_to_ecx)) .align align_4byte,0x90 GNAME(alloc_16_to_ecx): pushl %eax # Save eax and edx as C could destroy them. pushl %edx pushl $16 # Push the size. call GNAME(alloc) addl $4,%esp # Pop the size arg. movl %eax,%ecx # Set up the destination. popl %edx # Restore eax and edx. popl %eax ret SIZE(GNAME(alloc_16_to_ecx)) .globl GNAME(alloc_to_edx) TYPE(GNAME(alloc_to_edx)) .align align_4byte,0x90 GNAME(alloc_to_edx): pushl %eax # Save eax and ecx as C could destroy them. pushl %ecx pushl %edx # Push the size. call GNAME(alloc) addl $4,%esp # Pop the size arg. movl %eax,%edx # Set up the destination. popl %ecx # Restore eax and ecx. popl %eax ret SIZE(GNAME(alloc_to_edx)) .globl GNAME(alloc_8_to_edx) TYPE(GNAME(alloc_8_to_edx)) .align align_4byte,0x90 GNAME(alloc_8_to_edx): pushl %eax # Save eax and ecx as C could destroy them. pushl %ecx pushl $8 # Push the size. call GNAME(alloc) addl $4,%esp # Pop the size arg. movl %eax,%edx # Set up the destination. popl %ecx # Restore eax and ecx. popl %eax ret SIZE(GNAME(alloc_8_to_edx)) .globl GNAME(alloc_16_to_edx) TYPE(GNAME(alloc_16_to_edx)) .align align_4byte,0x90 GNAME(alloc_16_to_edx): pushl %eax # Save eax and ecx as C could destroy them. pushl %ecx pushl $16 # Push the size. call GNAME(alloc) addl $4,%esp # Pop the size arg. movl %eax,%edx # Set up the destination. popl %ecx # Restore eax and ecx. popl %eax ret SIZE(GNAME(alloc_16_to_edx)) .globl GNAME(alloc_to_ebx) TYPE(GNAME(alloc_to_ebx)) .align align_4byte,0x90 GNAME(alloc_to_ebx): pushl %eax # Save eax, ecx, and edx as C could destroy them. pushl %ecx pushl %edx pushl %ebx # Push the size. call GNAME(alloc) addl $4,%esp # Pop the size arg. movl %eax,%ebx # Set up the destination. popl %edx # Restore eax, ecx and edx. popl %ecx popl %eax ret SIZE(GNAME(alloc_to_ebx)) .globl GNAME(alloc_8_to_ebx) TYPE(GNAME(alloc_8_to_ebx)) .align align_4byte,0x90 GNAME(alloc_8_to_ebx): pushl %eax # Save eax, ecx, and edx as C could destroy them. pushl %ecx pushl %edx pushl $8 # Push the size. call GNAME(alloc) addl $4,%esp # Pop the size arg. movl %eax,%ebx # Set up the destination. popl %edx # Restore eax, ecx and edx. popl %ecx popl %eax ret SIZE(GNAME(alloc_8_to_ebx)) .globl GNAME(alloc_16_to_ebx) TYPE(GNAME(alloc_16_to_ebx)) .align align_4byte,0x90 GNAME(alloc_16_to_ebx): pushl %eax # Save eax, ecx, and edx as C could destroy them. pushl %ecx pushl %edx pushl $16 # Push the size call GNAME(alloc) addl $4,%esp # pop the size arg. movl %eax,%ebx # setup the destination. popl %edx # Restore eax, ecx and edx. popl %ecx popl %eax ret SIZE(GNAME(alloc_16_to_ebx)) .globl GNAME(alloc_to_esi) TYPE(GNAME(alloc_to_esi)) .align align_4byte,0x90 GNAME(alloc_to_esi): pushl %eax # Save eax, ecx, and edx as C could destroy them. pushl %ecx pushl %edx pushl %esi # Push the size call GNAME(alloc) addl $4,%esp # pop the size arg. movl %eax,%esi # setup the destination. popl %edx # Restore eax, ecx and edx. popl %ecx popl %eax ret SIZE(GNAME(alloc_to_esi)) .globl GNAME(alloc_8_to_esi) TYPE(GNAME(alloc_8_to_esi)) .align align_4byte,0x90 GNAME(alloc_8_to_esi): pushl %eax # Save eax, ecx, and edx as C could destroy them. pushl %ecx pushl %edx pushl $8 # Push the size call GNAME(alloc) addl $4,%esp # pop the size arg. movl %eax,%esi # setup the destination. popl %edx # Restore eax, ecx and edx. popl %ecx popl %eax ret SIZE(GNAME(alloc_8_to_esi)) .globl GNAME(alloc_16_to_esi) TYPE(GNAME(alloc_16_to_esi)) .align align_4byte,0x90 GNAME(alloc_16_to_esi): pushl %eax # Save eax, ecx, and edx as C could destroy them. pushl %ecx pushl %edx pushl $16 # Push the size call GNAME(alloc) addl $4,%esp # pop the size arg. movl %eax,%esi # setup the destination. popl %edx # Restore eax, ecx and edx. popl %ecx popl %eax ret SIZE(GNAME(alloc_16_to_esi)) .globl GNAME(alloc_to_edi) TYPE(GNAME(alloc_to_edi)) .align align_4byte,0x90 GNAME(alloc_to_edi): pushl %eax # Save eax, ecx, and edx as C could destroy them. pushl %ecx pushl %edx pushl %edi # Push the size call GNAME(alloc) addl $4,%esp # pop the size arg. movl %eax,%edi # setup the destination. popl %edx # Restore eax, ecx and edx. popl %ecx popl %eax ret SIZE(GNAME(alloc_to_edi)) .globl GNAME(alloc_8_to_edi) TYPE(GNAME(alloc_8_to_edi)) .align align_4byte,0x90 GNAME(alloc_8_to_edi): pushl %eax # Save eax, ecx, and edx as C could destroy them. pushl %ecx pushl %edx pushl $8 # Push the size call GNAME(alloc) addl $4,%esp # pop the size arg. movl %eax,%edi # setup the destination. popl %edx # Restore eax, ecx and edx. popl %ecx popl %eax ret SIZE(GNAME(alloc_8_to_edi)) .globl GNAME(alloc_16_to_edi) TYPE(GNAME(alloc_16_to_edi)) .align align_4byte,0x90 GNAME(alloc_16_to_edi): pushl %eax # Save eax, ecx, and edx as C could destroy them. pushl %ecx pushl %edx pushl $16 # Push the size call GNAME(alloc) addl $4,%esp # pop the size arg. movl %eax,%edi # setup the destination. popl %edx # Restore eax, ecx and edx. popl %ecx popl %eax ret SIZE(GNAME(alloc_16_to_edi)) /* Called from lisp when an inline allocation overflows. Every register except the result needs to be preserved. We depend on C to preserve ebx, esi, edi, and ebp. But where necessary must save eax, ecx, edx. */ #ifdef LISP_FEATURE_SB_THREAD #define START_REGION %fs:THREAD_ALLOC_REGION_OFFSET #else #define START_REGION GNAME(boxed_region) #endif /* This routine handles an overflow with eax=crfp+size. So the size=eax-crfp. */ .align align_4byte .globl GNAME(alloc_overflow_eax) TYPE(GNAME(alloc_overflow_eax)) GNAME(alloc_overflow_eax): pushl %ecx # Save ecx pushl %edx # Save edx /* Calculate the size for the allocation. */ subl START_REGION,%eax pushl %eax # Push the size call GNAME(alloc) addl $4,%esp # pop the size arg. popl %edx # Restore edx. popl %ecx # Restore ecx. ret SIZE(GNAME(alloc_overflow_eax)) .align align_4byte .globl GNAME(alloc_overflow_ecx) TYPE(GNAME(alloc_overflow_ecx)) GNAME(alloc_overflow_ecx): pushl %eax # Save eax pushl %edx # Save edx /* Calculate the size for the allocation. */ subl START_REGION,%ecx pushl %ecx # Push the size call GNAME(alloc) addl $4,%esp # pop the size arg. movl %eax,%ecx # setup the destination. popl %edx # Restore edx. popl %eax # Restore eax. ret SIZE(GNAME(alloc_overflow_ecx)) .align align_4byte .globl GNAME(alloc_overflow_edx) TYPE(GNAME(alloc_overflow_edx)) GNAME(alloc_overflow_edx): pushl %eax # Save eax pushl %ecx # Save ecx /* Calculate the size for the allocation. */ subl START_REGION,%edx pushl %edx # Push the size call GNAME(alloc) addl $4,%esp # pop the size arg. movl %eax,%edx # setup the destination. popl %ecx # Restore ecx. popl %eax # Restore eax. ret SIZE(GNAME(alloc_overflow_edx)) /* This routine handles an overflow with ebx=crfp+size. So the size=ebx-crfp. */ .align align_4byte .globl GNAME(alloc_overflow_ebx) TYPE(GNAME(alloc_overflow_ebx)) GNAME(alloc_overflow_ebx): pushl %eax # Save eax pushl %ecx # Save ecx pushl %edx # Save edx /* Calculate the size for the allocation. */ subl START_REGION,%ebx pushl %ebx # Push the size call GNAME(alloc) addl $4,%esp # pop the size arg. movl %eax,%ebx # setup the destination. popl %edx # Restore edx. popl %ecx # Restore ecx. popl %eax # Restore eax. ret SIZE(GNAME(alloc_overflow_ebx)) /* This routine handles an overflow with esi=crfp+size. So the size=esi-crfp. */ .align align_4byte .globl GNAME(alloc_overflow_esi) TYPE(GNAME(alloc_overflow_esi)) GNAME(alloc_overflow_esi): pushl %eax # Save eax pushl %ecx # Save ecx pushl %edx # Save edx /* Calculate the size for the allocation. */ subl START_REGION,%esi pushl %esi # Push the size call GNAME(alloc) addl $4,%esp # pop the size arg. movl %eax,%esi # setup the destination. popl %edx # Restore edx. popl %ecx # Restore ecx. popl %eax # Restore eax. ret SIZE(GNAME(alloc_overflow_esi)) .align align_4byte .globl GNAME(alloc_overflow_edi) TYPE(GNAME(alloc_overflow_edi)) GNAME(alloc_overflow_edi): pushl %eax # Save eax pushl %ecx # Save ecx pushl %edx # Save edx /* Calculate the size for the allocation. */ subl START_REGION,%edi pushl %edi # Push the size call GNAME(alloc) addl $4,%esp # pop the size arg. movl %eax,%edi # setup the destination. popl %edx # Restore edx. popl %ecx # Restore ecx. popl %eax # Restore eax. ret SIZE(GNAME(alloc_overflow_edi)) .align align_4byte,0x90 .globl GNAME(post_signal_tramp) TYPE(GNAME(post_signal_tramp)) GNAME(post_signal_tramp): /* this is notionally the second half of a function whose first half * doesn't exist. This is where call_into_lisp returns when called * using return_to_lisp_function */ addl $12,%esp /* clear call_into_lisp args from stack */ popal /* restore registers */ popfl leave ret SIZE(GNAME(post_signal_tramp)) #ifdef LISP_FEATURE_WIN32 /* * This is part of the funky magic for exception handling on win32. * see sigtrap_emulator() in win32-os.c for details. */ .globl GNAME(sigtrap_trampoline) GNAME(sigtrap_trampoline): pushl %eax pushl %ebp movl %esp, %ebp call GNAME(sigtrap_wrapper) pop %eax pop %eax TRAP .byte trap_ContextRestore hlt # We should never return here. /* * This is part of the funky magic for exception handling on win32. * see handle_exception() in win32-os.c for details. */ .globl GNAME(exception_trampoline) GNAME(exception_trampoline): pushl %eax pushl %ebp movl %esp, %ebp call GNAME(handle_win32_exception_wrapper) pop %eax pop %eax TRAP .byte trap_ContextRestore hlt # We should never return here. #endif /* fast_bzero implementations and code to detect which implementation * to use. */ .globl GNAME(fast_bzero_pointer) .data .align align_4byte GNAME(fast_bzero_pointer): /* Variable containing a pointer to the bzero function to use. * Initially points to a basic function. Change this variable * to fast_bzero_detect if OS supports SSE. */ .long GNAME(fast_bzero_base) .text .align align_8byte,0x90 .globl GNAME(fast_bzero) TYPE(GNAME(fast_bzero)) GNAME(fast_bzero): /* Indirect function call */ jmp *GNAME(fast_bzero_pointer) SIZE(GNAME(fast_bzero)) .text .align align_8byte,0x90 .globl GNAME(fast_bzero_detect) TYPE(GNAME(fast_bzero_detect)) GNAME(fast_bzero_detect): /* Decide whether to use SSE, MMX or REP version */ push %eax /* CPUID uses EAX-EDX */ push %ebx push %ecx push %edx mov $1, %eax cpuid test $0x04000000, %edx /* SSE2 needed for MOVNTDQ */ jnz Lsse2 /* Originally there was another case here for using the * MOVNTQ instruction for processors that supported MMX but * not SSE2. This turned out to be a loss especially on * Athlons (where this instruction is apparently microcoded * somewhat slowly). So for simplicity revert to REP STOSL * for all non-SSE2 processors. */ Lbase: movl GNAMEDOLLAR(fast_bzero_base), GNAME(fast_bzero_pointer) jmp Lrestore Lsse2: movl GNAMEDOLLAR(fast_bzero_sse), GNAME(fast_bzero_pointer) jmp Lrestore Lrestore: pop %edx pop %ecx pop %ebx pop %eax jmp *GNAME(fast_bzero_pointer) SIZE(GNAME(fast_bzero_detect)) .text .align align_8byte,0x90 .globl GNAME(fast_bzero_sse) TYPE(GNAME(fast_bzero_sse)) GNAME(fast_bzero_sse): /* A fast routine for zero-filling blocks of memory that are * guaranteed to start and end at a 4096-byte aligned address. */ push %esi /* Save temporary registers */ push %edi mov 16(%esp), %esi /* Parameter: amount of bytes to fill */ mov 12(%esp), %edi /* Parameter: start address */ shr $6, %esi /* Amount of 64-byte blocks to copy */ jz Lend_sse /* If none, stop */ movups %xmm7, -16(%esp) /* Save XMM register */ xorps %xmm7, %xmm7 /* Zero the XMM register */ jmp Lloop_sse .align align_16byte Lloop_sse: /* Copy the 16 zeroes from xmm7 to memory, 4 times. MOVNTDQ is the * non-caching double-quadword moving variant, i.e. the memory areas * we're touching are not fetched into the L1 cache, since we're just * going to overwrite the memory soon anyway. */ movntdq %xmm7, 0(%edi) movntdq %xmm7, 16(%edi) movntdq %xmm7, 32(%edi) movntdq %xmm7, 48(%edi) add $64, %edi /* Advance pointer */ dec %esi /* Decrement 64-byte block count */ jnz Lloop_sse movups -16(%esp), %xmm7 /* Restore the XMM register */ sfence /* Ensure that weakly ordered writes are flushed. */ Lend_sse: mov 12(%esp), %esi /* Parameter: start address */ prefetcht0 0(%esi) /* Prefetch the start of the block into cache, * since it's likely to be used immediately. */ pop %edi /* Restore temp registers */ pop %esi ret SIZE(GNAME(fast_bzero_sse)) .text .align align_8byte,0x90 .globl GNAME(fast_bzero_base) TYPE(GNAME(fast_bzero_base)) GNAME(fast_bzero_base): /* A fast routine for zero-filling blocks of memory that are * guaranteed to start and end at a 4096-byte aligned address. */ push %eax /* Save temporary registers */ push %ecx push %edi mov 20(%esp), %ecx /* Parameter: amount of bytes to fill */ mov 16(%esp), %edi /* Parameter: start address */ xor %eax, %eax /* Zero EAX */ shr $2, %ecx /* Amount of 4-byte blocks to copy */ jz Lend_base cld /* Set direction of STOSL to increment */ rep stosl /* Store EAX to *EDI, ECX times, incrementing * EDI by 4 after each store */ Lend_base: pop %edi /* Restore temp registers */ pop %ecx pop %eax ret SIZE(GNAME(fast_bzero_base)) END()