1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * fpu.c - save/restore of Floating Point Unit Registers on task switch
4 *
5 * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
6 */
7
8 #include <linux/sched.h>
9 #include <asm/switch_to.h>
10
11 /*
12 * To save/restore FPU regs, simplest scheme would use LR/SR insns.
13 * However since SR serializes the pipeline, an alternate "hack" can be used
14 * which uses the FPU Exchange insn (DEXCL) to r/w FPU regs.
15 *
16 * Store to 64bit dpfp1 reg from a pair of core regs:
17 * dexcl1 0, r1, r0 ; where r1:r0 is the 64 bit val
18 *
19 * Read from dpfp1 into pair of core regs (w/o clobbering dpfp1)
20 * mov_s r3, 0
21 * daddh11 r1, r3, r3 ; get "hi" into r1 (dpfp1 unchanged)
22 * dexcl1 r0, r1, r3 ; get "low" into r0 (dpfp1 low clobbered)
23 * dexcl1 0, r1, r0 ; restore dpfp1 to orig value
24 *
25 * However we can tweak the read, so that read-out of outgoing task's FPU regs
26 * and write of incoming task's regs happen in one shot. So all the work is
27 * done before context switch
28 */
29
30 void fpu_save_restore(struct task_struct *prev, struct task_struct *next)
31 {
32 unsigned int *saveto = &prev->thread.fpu.aux_dpfp[0].l;
33 unsigned int *readfrom = &next->thread.fpu.aux_dpfp[0].l;
34
35 const unsigned int zero = 0;
36
37 __asm__ __volatile__(
38 "daddh11 %0, %2, %2\n"
39 "dexcl1 %1, %3, %4\n"
40 : "=&r" (*(saveto + 1)), /* early clobber must here */
41 "=&r" (*(saveto))
42 : "r" (zero), "r" (*(readfrom + 1)), "r" (*(readfrom))
43 );
44
45 __asm__ __volatile__(
46 "daddh22 %0, %2, %2\n"
47 "dexcl2 %1, %3, %4\n"
48 : "=&r"(*(saveto + 3)), /* early clobber must here */
49 "=&r"(*(saveto + 2))
50 : "r" (zero), "r" (*(readfrom + 3)), "r" (*(readfrom + 2))
51 );
52 }