1 /* visemul.c: Emulation of VIS instructions.
3 * Copyright (C) 2006 David S. Miller (davem@davemloft.net)
5 #include <linux/kernel.h>
6 #include <linux/errno.h>
7 #include <linux/thread_info.h>
8 #include <linux/perf_event.h>
10 #include <asm/ptrace.h>
11 #include <asm/pstate.h>
12 #include <asm/fpumacro.h>
13 #include <asm/uaccess.h>
14 #include <asm/cacheflush.h>
16 /* OPF field of various VIS instructions. */
18 /* 000111011 - four 16-bit packs */
19 #define FPACK16_OPF 0x03b
21 /* 000111010 - two 32-bit packs */
22 #define FPACK32_OPF 0x03a
24 /* 000111101 - four 16-bit packs */
25 #define FPACKFIX_OPF 0x03d
27 /* 001001101 - four 16-bit expands */
28 #define FEXPAND_OPF 0x04d
30 /* 001001011 - two 32-bit merges */
31 #define FPMERGE_OPF 0x04b
33 /* 000110001 - 8-by-16-bit partitoned product */
34 #define FMUL8x16_OPF 0x031
36 /* 000110011 - 8-by-16-bit upper alpha partitioned product */
37 #define FMUL8x16AU_OPF 0x033
39 /* 000110101 - 8-by-16-bit lower alpha partitioned product */
40 #define FMUL8x16AL_OPF 0x035
42 /* 000110110 - upper 8-by-16-bit partitioned product */
43 #define FMUL8SUx16_OPF 0x036
45 /* 000110111 - lower 8-by-16-bit partitioned product */
46 #define FMUL8ULx16_OPF 0x037
48 /* 000111000 - upper 8-by-16-bit partitioned product */
49 #define FMULD8SUx16_OPF 0x038
51 /* 000111001 - lower unsigned 8-by-16-bit partitioned product */
52 #define FMULD8ULx16_OPF 0x039
54 /* 000101000 - four 16-bit compare; set rd if src1 > src2 */
55 #define FCMPGT16_OPF 0x028
57 /* 000101100 - two 32-bit compare; set rd if src1 > src2 */
58 #define FCMPGT32_OPF 0x02c
60 /* 000100000 - four 16-bit compare; set rd if src1 <= src2 */
61 #define FCMPLE16_OPF 0x020
63 /* 000100100 - two 32-bit compare; set rd if src1 <= src2 */
64 #define FCMPLE32_OPF 0x024
66 /* 000100010 - four 16-bit compare; set rd if src1 != src2 */
67 #define FCMPNE16_OPF 0x022
69 /* 000100110 - two 32-bit compare; set rd if src1 != src2 */
70 #define FCMPNE32_OPF 0x026
72 /* 000101010 - four 16-bit compare; set rd if src1 == src2 */
73 #define FCMPEQ16_OPF 0x02a
75 /* 000101110 - two 32-bit compare; set rd if src1 == src2 */
76 #define FCMPEQ32_OPF 0x02e
78 /* 000000000 - Eight 8-bit edge boundary processing */
79 #define EDGE8_OPF 0x000
81 /* 000000001 - Eight 8-bit edge boundary processing, no CC */
82 #define EDGE8N_OPF 0x001
84 /* 000000010 - Eight 8-bit edge boundary processing, little-endian */
85 #define EDGE8L_OPF 0x002
87 /* 000000011 - Eight 8-bit edge boundary processing, little-endian, no CC */
88 #define EDGE8LN_OPF 0x003
90 /* 000000100 - Four 16-bit edge boundary processing */
91 #define EDGE16_OPF 0x004
93 /* 000000101 - Four 16-bit edge boundary processing, no CC */
94 #define EDGE16N_OPF 0x005
96 /* 000000110 - Four 16-bit edge boundary processing, little-endian */
97 #define EDGE16L_OPF 0x006
99 /* 000000111 - Four 16-bit edge boundary processing, little-endian, no CC */
100 #define EDGE16LN_OPF 0x007
102 /* 000001000 - Two 32-bit edge boundary processing */
103 #define EDGE32_OPF 0x008
105 /* 000001001 - Two 32-bit edge boundary processing, no CC */
106 #define EDGE32N_OPF 0x009
108 /* 000001010 - Two 32-bit edge boundary processing, little-endian */
109 #define EDGE32L_OPF 0x00a
111 /* 000001011 - Two 32-bit edge boundary processing, little-endian, no CC */
112 #define EDGE32LN_OPF 0x00b
114 /* 000111110 - distance between 8 8-bit components */
115 #define PDIST_OPF 0x03e
117 /* 000010000 - convert 8-bit 3-D address to blocked byte address */
118 #define ARRAY8_OPF 0x010
120 /* 000010010 - convert 16-bit 3-D address to blocked byte address */
121 #define ARRAY16_OPF 0x012
123 /* 000010100 - convert 32-bit 3-D address to blocked byte address */
124 #define ARRAY32_OPF 0x014
126 /* 000011001 - Set the GSR.MASK field in preparation for a BSHUFFLE */
127 #define BMASK_OPF 0x019
129 /* 001001100 - Permute bytes as specified by GSR.MASK */
130 #define BSHUFFLE_OPF 0x04c
132 #define VIS_OPF_SHIFT 5
133 #define VIS_OPF_MASK (0x1ff << VIS_OPF_SHIFT)
135 #define RS1(INSN) (((INSN) >> 14) & 0x1f)
136 #define RS2(INSN) (((INSN) >> 0) & 0x1f)
137 #define RD(INSN) (((INSN) >> 25) & 0x1f)
139 static inline void maybe_flush_windows(unsigned int rs1, unsigned int rs2,
140 unsigned int rd, int from_kernel)
142 if (rs2 >= 16 || rs1 >= 16 || rd >= 16) {
143 if (from_kernel != 0)
144 __asm__ __volatile__("flushw");
150 static unsigned long fetch_reg(unsigned int reg, struct pt_regs *regs)
152 unsigned long value, fp;
155 return (!reg ? 0 : regs->u_regs[reg]);
157 fp = regs->u_regs[UREG_FP];
159 if (regs->tstate & TSTATE_PRIV) {
160 struct reg_window *win;
161 win = (struct reg_window *)(fp + STACK_BIAS);
162 value = win->locals[reg - 16];
163 } else if (!test_thread_64bit_stack(fp)) {
164 struct reg_window32 __user *win32;
165 win32 = (struct reg_window32 __user *)((unsigned long)((u32)fp));
166 get_user(value, &win32->locals[reg - 16]);
168 struct reg_window __user *win;
169 win = (struct reg_window __user *)(fp + STACK_BIAS);
170 get_user(value, &win->locals[reg - 16]);
175 static inline unsigned long __user *__fetch_reg_addr_user(unsigned int reg,
176 struct pt_regs *regs)
178 unsigned long fp = regs->u_regs[UREG_FP];
181 BUG_ON(regs->tstate & TSTATE_PRIV);
183 if (!test_thread_64bit_stack(fp)) {
184 struct reg_window32 __user *win32;
185 win32 = (struct reg_window32 __user *)((unsigned long)((u32)fp));
186 return (unsigned long __user *)&win32->locals[reg - 16];
188 struct reg_window __user *win;
189 win = (struct reg_window __user *)(fp + STACK_BIAS);
190 return &win->locals[reg - 16];
194 static inline unsigned long *__fetch_reg_addr_kern(unsigned int reg,
195 struct pt_regs *regs)
198 BUG_ON(regs->tstate & TSTATE_PRIV);
200 return ®s->u_regs[reg];
203 static void store_reg(struct pt_regs *regs, unsigned long val, unsigned long rd)
206 unsigned long *rd_kern = __fetch_reg_addr_kern(rd, regs);
210 unsigned long __user *rd_user = __fetch_reg_addr_user(rd, regs);
212 if (!test_thread_64bit_stack(regs->u_regs[UREG_FP]))
213 __put_user((u32)val, (u32 __user *)rd_user);
215 __put_user(val, rd_user);
219 static inline unsigned long fpd_regval(struct fpustate *f,
220 unsigned int insn_regnum)
222 insn_regnum = (((insn_regnum & 1) << 5) |
223 (insn_regnum & 0x1e));
225 return *(unsigned long *) &f->regs[insn_regnum];
228 static inline unsigned long *fpd_regaddr(struct fpustate *f,
229 unsigned int insn_regnum)
231 insn_regnum = (((insn_regnum & 1) << 5) |
232 (insn_regnum & 0x1e));
234 return (unsigned long *) &f->regs[insn_regnum];
237 static inline unsigned int fps_regval(struct fpustate *f,
238 unsigned int insn_regnum)
240 return f->regs[insn_regnum];
243 static inline unsigned int *fps_regaddr(struct fpustate *f,
244 unsigned int insn_regnum)
246 return &f->regs[insn_regnum];
252 static struct edge_tab edge8_tab[8] = {
262 static struct edge_tab edge8_tab_l[8] = {
272 static struct edge_tab edge16_tab[4] = {
278 static struct edge_tab edge16_tab_l[4] = {
284 static struct edge_tab edge32_tab[2] = {
288 static struct edge_tab edge32_tab_l[2] = {
293 static void edge(struct pt_regs *regs, unsigned int insn, unsigned int opf)
295 unsigned long orig_rs1, rs1, orig_rs2, rs2, rd_val;
298 maybe_flush_windows(RS1(insn), RS2(insn), RD(insn), 0);
299 orig_rs1 = rs1 = fetch_reg(RS1(insn), regs);
300 orig_rs2 = rs2 = fetch_reg(RS2(insn), regs);
302 if (test_thread_flag(TIF_32BIT)) {
303 rs1 = rs1 & 0xffffffff;
304 rs2 = rs2 & 0xffffffff;
310 left = edge8_tab[rs1 & 0x7].left;
311 right = edge8_tab[rs2 & 0x7].right;
315 left = edge8_tab_l[rs1 & 0x7].left;
316 right = edge8_tab_l[rs2 & 0x7].right;
321 left = edge16_tab[(rs1 >> 1) & 0x3].left;
322 right = edge16_tab[(rs2 >> 1) & 0x3].right;
327 left = edge16_tab_l[(rs1 >> 1) & 0x3].left;
328 right = edge16_tab_l[(rs2 >> 1) & 0x3].right;
333 left = edge32_tab[(rs1 >> 2) & 0x1].left;
334 right = edge32_tab[(rs2 >> 2) & 0x1].right;
339 left = edge32_tab_l[(rs1 >> 2) & 0x1].left;
340 right = edge32_tab_l[(rs2 >> 2) & 0x1].right;
344 if ((rs1 & ~0x7UL) == (rs2 & ~0x7UL))
345 rd_val = right & left;
349 store_reg(regs, rd_val, RD(insn));
358 unsigned long ccr, tstate;
360 __asm__ __volatile__("subcc %1, %2, %%g0\n\t"
363 : "r" (orig_rs1), "r" (orig_rs2)
365 tstate = regs->tstate & ~(TSTATE_XCC | TSTATE_ICC);
366 regs->tstate = tstate | (ccr << 32UL);
371 static void array(struct pt_regs *regs, unsigned int insn, unsigned int opf)
373 unsigned long rs1, rs2, rd_val;
374 unsigned int bits, bits_mask;
376 maybe_flush_windows(RS1(insn), RS2(insn), RD(insn), 0);
377 rs1 = fetch_reg(RS1(insn), regs);
378 rs2 = fetch_reg(RS2(insn), regs);
380 bits = (rs2 > 5 ? 5 : rs2);
381 bits_mask = (1UL << bits) - 1UL;
383 rd_val = ((((rs1 >> 11) & 0x3) << 0) |
384 (((rs1 >> 33) & 0x3) << 2) |
385 (((rs1 >> 55) & 0x1) << 4) |
386 (((rs1 >> 13) & 0xf) << 5) |
387 (((rs1 >> 35) & 0xf) << 9) |
388 (((rs1 >> 56) & 0xf) << 13) |
389 (((rs1 >> 17) & bits_mask) << 17) |
390 (((rs1 >> 39) & bits_mask) << (17 + bits)) |
391 (((rs1 >> 60) & 0xf) << (17 + (2*bits))));
402 store_reg(regs, rd_val, RD(insn));
405 static void bmask(struct pt_regs *regs, unsigned int insn)
407 unsigned long rs1, rs2, rd_val, gsr;
409 maybe_flush_windows(RS1(insn), RS2(insn), RD(insn), 0);
410 rs1 = fetch_reg(RS1(insn), regs);
411 rs2 = fetch_reg(RS2(insn), regs);
414 store_reg(regs, rd_val, RD(insn));
416 gsr = current_thread_info()->gsr[0] & 0xffffffff;
417 gsr |= rd_val << 32UL;
418 current_thread_info()->gsr[0] = gsr;
421 static void bshuffle(struct pt_regs *regs, unsigned int insn)
423 struct fpustate *f = FPUSTATE;
424 unsigned long rs1, rs2, rd_val;
425 unsigned long bmask, i;
427 bmask = current_thread_info()->gsr[0] >> 32UL;
429 rs1 = fpd_regval(f, RS1(insn));
430 rs2 = fpd_regval(f, RS2(insn));
433 for (i = 0; i < 8; i++) {
434 unsigned long which = (bmask >> (i * 4)) & 0xf;
438 byte = (rs1 >> (which * 8)) & 0xff;
440 byte = (rs2 >> ((which-8)*8)) & 0xff;
441 rd_val |= (byte << (i * 8));
444 *fpd_regaddr(f, RD(insn)) = rd_val;
447 static void pdist(struct pt_regs *regs, unsigned int insn)
449 struct fpustate *f = FPUSTATE;
450 unsigned long rs1, rs2, *rd, rd_val;
453 rs1 = fpd_regval(f, RS1(insn));
454 rs2 = fpd_regval(f, RS2(insn));
455 rd = fpd_regaddr(f, RD(insn));
459 for (i = 0; i < 8; i++) {
462 s1 = (rs1 >> (56 - (i * 8))) & 0xff;
463 s2 = (rs2 >> (56 - (i * 8))) & 0xff;
465 /* Absolute value of difference. */
476 static void pformat(struct pt_regs *regs, unsigned int insn, unsigned int opf)
478 struct fpustate *f = FPUSTATE;
479 unsigned long rs1, rs2, gsr, scale, rd_val;
481 gsr = current_thread_info()->gsr[0];
482 scale = (gsr >> 3) & (opf == FPACK16_OPF ? 0xf : 0x1f);
487 rs2 = fpd_regval(f, RS2(insn));
489 for (byte = 0; byte < 4; byte++) {
491 s16 src = (rs2 >> (byte * 16UL)) & 0xffffUL;
492 int scaled = src << scale;
493 int from_fixed = scaled >> 7;
495 val = ((from_fixed < 0) ?
500 rd_val |= (val << (8 * byte));
502 *fps_regaddr(f, RD(insn)) = rd_val;
509 rs1 = fpd_regval(f, RS1(insn));
510 rs2 = fpd_regval(f, RS2(insn));
511 rd_val = (rs1 << 8) & ~(0x000000ff000000ffUL);
512 for (word = 0; word < 2; word++) {
514 s32 src = (rs2 >> (word * 32UL));
515 s64 scaled = src << scale;
516 s64 from_fixed = scaled >> 23;
518 val = ((from_fixed < 0) ?
523 rd_val |= (val << (32 * word));
525 *fpd_regaddr(f, RD(insn)) = rd_val;
532 rs2 = fpd_regval(f, RS2(insn));
535 for (word = 0; word < 2; word++) {
537 s32 src = (rs2 >> (word * 32UL));
538 s64 scaled = src << scale;
539 s64 from_fixed = scaled >> 16;
541 val = ((from_fixed < -32768) ?
543 (from_fixed > 32767) ?
546 rd_val |= ((val & 0xffff) << (word * 16));
548 *fps_regaddr(f, RD(insn)) = rd_val;
555 rs2 = fps_regval(f, RS2(insn));
558 for (byte = 0; byte < 4; byte++) {
560 u8 src = (rs2 >> (byte * 8)) & 0xff;
564 rd_val |= (val << (byte * 16));
566 *fpd_regaddr(f, RD(insn)) = rd_val;
571 rs1 = fps_regval(f, RS1(insn));
572 rs2 = fps_regval(f, RS2(insn));
574 rd_val = (((rs2 & 0x000000ff) << 0) |
575 ((rs1 & 0x000000ff) << 8) |
576 ((rs2 & 0x0000ff00) << 8) |
577 ((rs1 & 0x0000ff00) << 16) |
578 ((rs2 & 0x00ff0000) << 16) |
579 ((rs1 & 0x00ff0000) << 24) |
580 ((rs2 & 0xff000000) << 24) |
581 ((rs1 & 0xff000000) << 32));
582 *fpd_regaddr(f, RD(insn)) = rd_val;
588 static void pmul(struct pt_regs *regs, unsigned int insn, unsigned int opf)
590 struct fpustate *f = FPUSTATE;
591 unsigned long rs1, rs2, rd_val;
597 rs1 = fps_regval(f, RS1(insn));
598 rs2 = fpd_regval(f, RS2(insn));
601 for (byte = 0; byte < 4; byte++) {
602 u16 src1 = (rs1 >> (byte * 8)) & 0x00ff;
603 s16 src2 = (rs2 >> (byte * 16)) & 0xffff;
604 u32 prod = src1 * src2;
605 u16 scaled = ((prod & 0x00ffff00) >> 8);
610 rd_val |= ((scaled & 0xffffUL) << (byte * 16UL));
613 *fpd_regaddr(f, RD(insn)) = rd_val;
618 case FMUL8x16AL_OPF: {
622 rs1 = fps_regval(f, RS1(insn));
623 rs2 = fps_regval(f, RS2(insn));
626 src2 = rs2 >> (opf == FMUL8x16AU_OPF ? 16 : 0);
627 for (byte = 0; byte < 4; byte++) {
628 u16 src1 = (rs1 >> (byte * 8)) & 0x00ff;
629 u32 prod = src1 * src2;
630 u16 scaled = ((prod & 0x00ffff00) >> 8);
635 rd_val |= ((scaled & 0xffffUL) << (byte * 16UL));
638 *fpd_regaddr(f, RD(insn)) = rd_val;
643 case FMUL8ULx16_OPF: {
644 unsigned long byte, ushift;
646 rs1 = fpd_regval(f, RS1(insn));
647 rs2 = fpd_regval(f, RS2(insn));
650 ushift = (opf == FMUL8SUx16_OPF) ? 8 : 0;
651 for (byte = 0; byte < 4; byte++) {
657 src1 = ((rs1 >> ((16 * byte) + ushift)) & 0x00ff);
658 src2 = ((rs2 >> (16 * byte)) & 0xffff);
660 scaled = ((prod & 0x00ffff00) >> 8);
665 rd_val |= ((scaled & 0xffffUL) << (byte * 16UL));
668 *fpd_regaddr(f, RD(insn)) = rd_val;
672 case FMULD8SUx16_OPF:
673 case FMULD8ULx16_OPF: {
674 unsigned long byte, ushift;
676 rs1 = fps_regval(f, RS1(insn));
677 rs2 = fps_regval(f, RS2(insn));
680 ushift = (opf == FMULD8SUx16_OPF) ? 8 : 0;
681 for (byte = 0; byte < 2; byte++) {
687 src1 = ((rs1 >> ((16 * byte) + ushift)) & 0x00ff);
688 src2 = ((rs2 >> (16 * byte)) & 0xffff);
690 scaled = ((prod & 0x00ffff00) >> 8);
695 rd_val |= ((scaled & 0xffffUL) <<
696 ((byte * 32UL) + 7UL));
698 *fpd_regaddr(f, RD(insn)) = rd_val;
704 static void pcmp(struct pt_regs *regs, unsigned int insn, unsigned int opf)
706 struct fpustate *f = FPUSTATE;
707 unsigned long rs1, rs2, rd_val, i;
709 rs1 = fpd_regval(f, RS1(insn));
710 rs2 = fpd_regval(f, RS2(insn));
716 for (i = 0; i < 4; i++) {
717 s16 a = (rs1 >> (i * 16)) & 0xffff;
718 s16 b = (rs2 >> (i * 16)) & 0xffff;
726 for (i = 0; i < 2; i++) {
727 s32 a = (rs1 >> (i * 32)) & 0xffffffff;
728 s32 b = (rs2 >> (i * 32)) & 0xffffffff;
736 for (i = 0; i < 4; i++) {
737 s16 a = (rs1 >> (i * 16)) & 0xffff;
738 s16 b = (rs2 >> (i * 16)) & 0xffff;
746 for (i = 0; i < 2; i++) {
747 s32 a = (rs1 >> (i * 32)) & 0xffffffff;
748 s32 b = (rs2 >> (i * 32)) & 0xffffffff;
756 for (i = 0; i < 4; i++) {
757 s16 a = (rs1 >> (i * 16)) & 0xffff;
758 s16 b = (rs2 >> (i * 16)) & 0xffff;
766 for (i = 0; i < 2; i++) {
767 s32 a = (rs1 >> (i * 32)) & 0xffffffff;
768 s32 b = (rs2 >> (i * 32)) & 0xffffffff;
776 for (i = 0; i < 4; i++) {
777 s16 a = (rs1 >> (i * 16)) & 0xffff;
778 s16 b = (rs2 >> (i * 16)) & 0xffff;
786 for (i = 0; i < 2; i++) {
787 s32 a = (rs1 >> (i * 32)) & 0xffffffff;
788 s32 b = (rs2 >> (i * 32)) & 0xffffffff;
796 maybe_flush_windows(0, 0, RD(insn), 0);
797 store_reg(regs, rd_val, RD(insn));
800 /* Emulate the VIS instructions which are not implemented in
801 * hardware on Niagara.
803 int vis_emul(struct pt_regs *regs, unsigned int insn)
805 unsigned long pc = regs->tpc;
808 BUG_ON(regs->tstate & TSTATE_PRIV);
810 perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS, 1, regs, 0);
812 if (test_thread_flag(TIF_32BIT))
815 if (get_user(insn, (u32 __user *) pc))
818 save_and_clear_fpu();
820 opf = (insn & VIS_OPF_MASK) >> VIS_OPF_SHIFT;
825 /* Pixel Formatting Instructions. */
831 pformat(regs, insn, opf);
834 /* Partitioned Multiply Instructions */
840 case FMULD8SUx16_OPF:
841 case FMULD8ULx16_OPF:
842 pmul(regs, insn, opf);
845 /* Pixel Compare Instructions */
854 pcmp(regs, insn, opf);
857 /* Edge Handling Instructions */
870 edge(regs, insn, opf);
873 /* Pixel Component Distance */
878 /* Three-Dimensional Array Addressing Instructions */
882 array(regs, insn, opf);
885 /* Byte Mask and Shuffle Instructions */
891 bshuffle(regs, insn);
895 regs->tpc = regs->tnpc;