arm/nwfpe/entry.S

1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3     NetWinder Floating Point Emulator
7     Direct questions, comments to Scott Bambrough <scottb@netwinder.org>
13 /* This is the kernel's entry point into the floating point emulator.
14 It is called from the kernel with code similar to this:
23 	strb	r8, [r10, #TSK_USED_MATH]	@ set current->used_math
26 	ldr	pc, [r4]			@ Call FP emulator entry point
28 The kernel expects the emulator to return via one of two possible
29 points of return it passes to the emulator.  The emulator, if
30 successful in its emulation, jumps to ret_from_exception (passed in
31 r9) and the kernel takes care of returning control from the trap to
32 the user code.  If the emulator is unable to emulate the instruction,
36 On entry to the emulator r10 points to an area of private FP workspace
39 is used as a flag to detect the first time a process uses floating point,
48 2) It calls EmulateAll to emulate a floating point instruction.
52 the next instruction.  If it is a floating point instruction, it
53 executes the instruction, without returning to user space.  In this
54 way it repeatedly looks ahead and executes floating point instructions
55 until it encounters a non floating point instruction, at which time it
58 This is done to reduce the effect of the trap overhead on each
59 floating point instructions.  GCC attempts to group floating point
60 instructions to allow the emulator to spread the cost of the trap over
61 several floating point instructions.  */
63 #include <asm/asm-offsets.h>
67 	mov	r4, lr			@ save the failure-return addresses
77 	@ if condition code failed to match, next insn
101 	@ We need to be prepared for the instructions at .Lx1 and .Lx2
102 	@ to fault.  Emit the appropriate exception gunk to fix things up.