blob: d18dde95b8aae6a12dd319e71747ff58dfad8d84 [file] [log] [blame]
NetWinder Floating Point Emulator
(c) Rebel.COM, 1998
(c) 1998, 1999 Philip Blundell
Direct questions, comments to Scott Bambrough <>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
#include <asm/opcodes.h>
/* This is the kernel's entry point into the floating point emulator.
It is called from the kernel with code similar to this:
sub r4, r5, #4
ldrt r0, [r4] @ r0 = instruction
adrsvc al, r9, ret_from_exception @ r9 = normal FP return
adrsvc al, lr, fpundefinstr @ lr = undefined instr return
get_current_task r10
mov r8, #1
strb r8, [r10, #TSK_USED_MATH] @ set current->used_math
add r10, r10, #TSS_FPESAVE @ r10 = workspace
ldr r4, .LC2
ldr pc, [r4] @ Call FP emulator entry point
The kernel expects the emulator to return via one of two possible
points of return it passes to the emulator. The emulator, if
successful in its emulation, jumps to ret_from_exception (passed in
r9) and the kernel takes care of returning control from the trap to
the user code. If the emulator is unable to emulate the instruction,
it returns via _fpundefinstr (passed via lr) and the kernel halts the
user program with a core dump.
On entry to the emulator r10 points to an area of private FP workspace
reserved in the thread structure for this process. This is where the
emulator saves its registers across calls. The first word of this area
is used as a flag to detect the first time a process uses floating point,
so that the emulator startup cost can be avoided for tasks that don't
want it.
This routine does three things:
1) The kernel has created a struct pt_regs on the stack and saved the
user registers into it. See /usr/include/asm/proc/ptrace.h for details.
2) It calls EmulateAll to emulate a floating point instruction.
EmulateAll returns 1 if the emulation was successful, or 0 if not.
3) If an instruction has been emulated successfully, it looks ahead at
the next instruction. If it is a floating point instruction, it
executes the instruction, without returning to user space. In this
way it repeatedly looks ahead and executes floating point instructions
until it encounters a non floating point instruction, at which time it
returns via _fpreturn.
This is done to reduce the effect of the trap overhead on each
floating point instructions. GCC attempts to group floating point
instructions to allow the emulator to spread the cost of the trap over
several floating point instructions. */
#include <asm/asm-offsets.h>
.globl nwfpe_enter
mov r4, lr @ save the failure-return addresses
mov sl, sp @ we access the registers via 'sl'
ldr r5, [sp, #S_PC] @ get contents of PC;
mov r6, r0 @ save the opcode
ldr r1, [sp, #S_PSR] @ fetch the PSR
bl arm_check_condition @ check the condition
cmp r0, #ARM_OPCODE_CONDTEST_PASS @ condition passed?
@ if condition code failed to match, next insn
bne next @ get the next instruction;
mov r0, r6 @ prepare for EmulateAll()
bl EmulateAll @ emulate the instruction
cmp r0, #0 @ was emulation successful
moveq pc, r4 @ no, return failure
.Lx1: ldrt r6, [r5], #4 @ get the next instruction and
@ increment PC
and r2, r6, #0x0F000000 @ test for FP insns
teq r2, #0x0C000000
teqne r2, #0x0D000000
teqne r2, #0x0E000000
movne pc, r9 @ return ok if not a fp insn
str r5, [sp, #S_PC] @ update PC copy in regs
mov r0, r6 @ save a copy
b emulate @ check condition and emulate
@ We need to be prepared for the instructions at .Lx1 and .Lx2
@ to fault. Emit the appropriate exception gunk to fix things up.
@ ??? For some reason, faults can happen at .Lx2 even with a
@ plain LDR instruction. Weird, but it seems harmless.
.pushsection .fixup,"ax"
.align 2
.Lfix: mov pc, r9 @ let the user eat segfaults
.pushsection __ex_table,"a"
.align 3
.long .Lx1, .Lfix