aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorAlexander Graf <agraf@suse.de>2010-02-19 05:00:44 -0500
committerAvi Kivity <avi@redhat.com>2010-04-25 05:35:27 -0400
commit831317b605e7d7ce0bdadb3b0f50560fc13cecbf (patch)
tree3d1cb766faebc56404f3abd43af0b2f461f584fe
parente5c29e926cd29444d76657398801d49119851a56 (diff)
KVM: PPC: Implement Paired Single emulation
The one big thing about the Gekko is paired singles. Paired singles are an extension to the instruction set, that adds 32 single precision floating point registers (qprs), some SPRs to modify the behavior of paired singled operations and instructions to deal with qprs to the instruction set. Unfortunately, it also changes semantics of existing operations that affect single values in FPRs. In most cases they get mirrored to the coresponding QPR. Thanks to that we need to emulate all FPU operations and all the new paired single operations too. In order to achieve that, we use the just introduced FPU call helpers to call the real FPU whenever the guest wants to modify an FPR. Additionally we also fix up the QPR values along the way. That way we can execute paired single FPU operations without implementing a soft fpu. Signed-off-by: Alexander Graf <agraf@suse.de> Signed-off-by: Avi Kivity <avi@redhat.com>
Diffstat
-rw-r--r--arch/powerpc/include/asm/kvm_book3s.h1
-rw-r--r--arch/powerpc/kvm/Makefile1
-rw-r--r--arch/powerpc/kvm/book3s_64_emulate.c3
-rw-r--r--arch/powerpc/kvm/book3s_paired_singles.c1289
4 files changed, 1294 insertions, 0 deletions
diff --git a/arch/powerpc/include/asm/kvm_book3s.h b/arch/powerpc/include/asm/kvm_book3s.h
index fd432100f6db..e6ea974df44e 100644
--- a/arch/powerpc/include/asm/kvm_book3s.h
+++ b/arch/powerpc/include/asm/kvm_book3s.h
@@ -121,6 +121,7 @@ extern void kvmppc_book3s_queue_irqprio(struct kvm_vcpu *vcpu, unsigned int vec)
121extern void kvmppc_set_bat(struct kvm_vcpu *vcpu, struct kvmppc_bat *bat, 121extern void kvmppc_set_bat(struct kvm_vcpu *vcpu, struct kvmppc_bat *bat,
122 bool upper, u32 val); 122 bool upper, u32 val);
123extern void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr); 123extern void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr);
124extern int kvmppc_emulate_paired_single(struct kvm_run *run, struct kvm_vcpu *vcpu);
124 125
125extern u32 kvmppc_trampoline_lowmem; 126extern u32 kvmppc_trampoline_lowmem;
126extern u32 kvmppc_trampoline_enter; 127extern u32 kvmppc_trampoline_enter;
diff --git a/arch/powerpc/kvm/Makefile b/arch/powerpc/kvm/Makefile
index e575cfd015fe..eba721e39328 100644
--- a/arch/powerpc/kvm/Makefile
+++ b/arch/powerpc/kvm/Makefile
@@ -41,6 +41,7 @@ kvm-objs-$(CONFIG_KVM_E500) := $(kvm-e500-objs)
41kvm-book3s_64-objs := \ 41kvm-book3s_64-objs := \
42 $(common-objs-y) \ 42 $(common-objs-y) \
43 fpu.o \ 43 fpu.o \
44 book3s_paired_singles.o \
44 book3s.o \ 45 book3s.o \
45 book3s_64_emulate.o \ 46 book3s_64_emulate.o \
46 book3s_64_interrupts.o \ 47 book3s_64_interrupts.o \
diff --git a/arch/powerpc/kvm/book3s_64_emulate.c b/arch/powerpc/kvm/book3s_64_emulate.c
index 1d1b9524f0e4..c9892140dd02 100644
--- a/arch/powerpc/kvm/book3s_64_emulate.c
+++ b/arch/powerpc/kvm/book3s_64_emulate.c
@@ -200,6 +200,9 @@ int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
200 emulated = EMULATE_FAIL; 200 emulated = EMULATE_FAIL;
201 } 201 }
202 202
203 if (emulated == EMULATE_FAIL)
204 emulated = kvmppc_emulate_paired_single(run, vcpu);
205
203 return emulated; 206 return emulated;
204} 207}
205 208
diff --git a/arch/powerpc/kvm/book3s_paired_singles.c b/arch/powerpc/kvm/book3s_paired_singles.c
new file mode 100644
index 000000000000..7a27bac8c44a
--- /dev/null
+++ b/arch/powerpc/kvm/book3s_paired_singles.c
@@ -0,0 +1,1289 @@
1/*
2 * This program is free software; you can redistribute it and/or modify
3 * it under the terms of the GNU General Public License, version 2, as
4 * published by the Free Software Foundation.
5 *
6 * This program is distributed in the hope that it will be useful,
7 * but WITHOUT ANY WARRANTY; without even the implied warranty of
8 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
9 * GNU General Public License for more details.
10 *
11 * You should have received a copy of the GNU General Public License
12 * along with this program; if not, write to the Free Software
13 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
14 *
15 * Copyright Novell Inc 2010
16 *
17 * Authors: Alexander Graf <agraf@suse.de>
18 */
19
20#include <asm/kvm.h>
21#include <asm/kvm_ppc.h>
22#include <asm/disassemble.h>
23#include <asm/kvm_book3s.h>
24#include <asm/kvm_fpu.h>
25#include <asm/reg.h>
26#include <asm/cacheflush.h>
27#include <linux/vmalloc.h>
28
29/* #define DEBUG */
30
31#ifdef DEBUG
32#define dprintk printk
33#else
34#define dprintk(...) do { } while(0);
35#endif
36
37#define OP_LFS 48
38#define OP_LFSU 49
39#define OP_LFD 50
40#define OP_LFDU 51
41#define OP_STFS 52
42#define OP_STFSU 53
43#define OP_STFD 54
44#define OP_STFDU 55
45#define OP_PSQ_L 56
46#define OP_PSQ_LU 57
47#define OP_PSQ_ST 60
48#define OP_PSQ_STU 61
49
50#define OP_31_LFSX 535
51#define OP_31_LFSUX 567
52#define OP_31_LFDX 599
53#define OP_31_LFDUX 631
54#define OP_31_STFSX 663
55#define OP_31_STFSUX 695
56#define OP_31_STFX 727
57#define OP_31_STFUX 759
58#define OP_31_LWIZX 887
59#define OP_31_STFIWX 983
60
61#define OP_59_FADDS 21
62#define OP_59_FSUBS 20
63#define OP_59_FSQRTS 22
64#define OP_59_FDIVS 18
65#define OP_59_FRES 24
66#define OP_59_FMULS 25
67#define OP_59_FRSQRTES 26
68#define OP_59_FMSUBS 28
69#define OP_59_FMADDS 29
70#define OP_59_FNMSUBS 30
71#define OP_59_FNMADDS 31
72
73#define OP_63_FCMPU 0
74#define OP_63_FCPSGN 8
75#define OP_63_FRSP 12
76#define OP_63_FCTIW 14
77#define OP_63_FCTIWZ 15
78#define OP_63_FDIV 18
79#define OP_63_FADD 21
80#define OP_63_FSQRT 22
81#define OP_63_FSEL 23
82#define OP_63_FRE 24
83#define OP_63_FMUL 25
84#define OP_63_FRSQRTE 26
85#define OP_63_FMSUB 28
86#define OP_63_FMADD 29
87#define OP_63_FNMSUB 30
88#define OP_63_FNMADD 31
89#define OP_63_FCMPO 32
90#define OP_63_MTFSB1 38 // XXX
91#define OP_63_FSUB 20
92#define OP_63_FNEG 40
93#define OP_63_MCRFS 64
94#define OP_63_MTFSB0 70
95#define OP_63_FMR 72
96#define OP_63_MTFSFI 134
97#define OP_63_FABS 264
98#define OP_63_MFFS 583
99#define OP_63_MTFSF 711
100
101#define OP_4X_PS_CMPU0 0
102#define OP_4X_PSQ_LX 6
103#define OP_4XW_PSQ_STX 7
104#define OP_4A_PS_SUM0 10
105#define OP_4A_PS_SUM1 11
106#define OP_4A_PS_MULS0 12
107#define OP_4A_PS_MULS1 13
108#define OP_4A_PS_MADDS0 14
109#define OP_4A_PS_MADDS1 15
110#define OP_4A_PS_DIV 18
111#define OP_4A_PS_SUB 20
112#define OP_4A_PS_ADD 21
113#define OP_4A_PS_SEL 23
114#define OP_4A_PS_RES 24
115#define OP_4A_PS_MUL 25
116#define OP_4A_PS_RSQRTE 26
117#define OP_4A_PS_MSUB 28
118#define OP_4A_PS_MADD 29
119#define OP_4A_PS_NMSUB 30
120#define OP_4A_PS_NMADD 31
121#define OP_4X_PS_CMPO0 32
122#define OP_4X_PSQ_LUX 38
123#define OP_4XW_PSQ_STUX 39
124#define OP_4X_PS_NEG 40
125#define OP_4X_PS_CMPU1 64
126#define OP_4X_PS_MR 72
127#define OP_4X_PS_CMPO1 96
128#define OP_4X_PS_NABS 136
129#define OP_4X_PS_ABS 264
130#define OP_4X_PS_MERGE00 528
131#define OP_4X_PS_MERGE01 560
132#define OP_4X_PS_MERGE10 592
133#define OP_4X_PS_MERGE11 624
134
135#define SCALAR_NONE 0
136#define SCALAR_HIGH (1 << 0)
137#define SCALAR_LOW (1 << 1)
138#define SCALAR_NO_PS0 (1 << 2)
139#define SCALAR_NO_PS1 (1 << 3)
140
141#define GQR_ST_TYPE_MASK 0x00000007
142#define GQR_ST_TYPE_SHIFT 0
143#define GQR_ST_SCALE_MASK 0x00003f00
144#define GQR_ST_SCALE_SHIFT 8
145#define GQR_LD_TYPE_MASK 0x00070000
146#define GQR_LD_TYPE_SHIFT 16
147#define GQR_LD_SCALE_MASK 0x3f000000
148#define GQR_LD_SCALE_SHIFT 24
149
150#define GQR_QUANTIZE_FLOAT 0
151#define GQR_QUANTIZE_U8 4
152#define GQR_QUANTIZE_U16 5
153#define GQR_QUANTIZE_S8 6
154#define GQR_QUANTIZE_S16 7
155
156#define FPU_LS_SINGLE 0
157#define FPU_LS_DOUBLE 1
158#define FPU_LS_SINGLE_LOW 2
159
160static inline void kvmppc_sync_qpr(struct kvm_vcpu *vcpu, int rt)
161{
162 struct thread_struct t;
163
164 t.fpscr.val = vcpu->arch.fpscr;
165 cvt_df((double*)&vcpu->arch.fpr[rt], (float*)&vcpu->arch.qpr[rt], &t);
166}
167
168static void kvmppc_inject_pf(struct kvm_vcpu *vcpu, ulong eaddr, bool is_store)
169{
170 u64 dsisr;
171
172 vcpu->arch.msr = kvmppc_set_field(vcpu->arch.msr, 33, 36, 0);
173 vcpu->arch.msr = kvmppc_set_field(vcpu->arch.msr, 42, 47, 0);
174 vcpu->arch.dear = eaddr;
175 /* Page Fault */
176 dsisr = kvmppc_set_field(0, 33, 33, 1);
177 if (is_store)
178 to_book3s(vcpu)->dsisr = kvmppc_set_field(dsisr, 38, 38, 1);
179 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_DATA_STORAGE);
180}
181
182static int kvmppc_emulate_fpr_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
183 int rs, ulong addr, int ls_type)
184{
185 int emulated = EMULATE_FAIL;
186 struct thread_struct t;
187 int r;
188 char tmp[8];
189 int len = sizeof(u32);
190
191 if (ls_type == FPU_LS_DOUBLE)
192 len = sizeof(u64);
193
194 t.fpscr.val = vcpu->arch.fpscr;
195
196 /* read from memory */
197 r = kvmppc_ld(vcpu, &addr, len, tmp, true);
198 vcpu->arch.paddr_accessed = addr;
199
200 if (r < 0) {
201 kvmppc_inject_pf(vcpu, addr, false);
202 goto done_load;
203 } else if (r == EMULATE_DO_MMIO) {
204 emulated = kvmppc_handle_load(run, vcpu, KVM_REG_FPR | rs, len, 1);
205 goto done_load;
206 }
207
208 emulated = EMULATE_DONE;
209
210 /* put in registers */
211 switch (ls_type) {
212 case FPU_LS_SINGLE:
213 cvt_fd((float*)tmp, (double*)&vcpu->arch.fpr[rs], &t);
214 vcpu->arch.qpr[rs] = *((u32*)tmp);
215 break;
216 case FPU_LS_DOUBLE:
217 vcpu->arch.fpr[rs] = *((u64*)tmp);
218 break;
219 }
220
221 dprintk(KERN_INFO "KVM: FPR_LD [0x%llx] at 0x%lx (%d)\n", *(u64*)tmp,
222 addr, len);
223
224done_load:
225 return emulated;
226}
227
228static int kvmppc_emulate_fpr_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
229 int rs, ulong addr, int ls_type)
230{
231 int emulated = EMULATE_FAIL;
232 struct thread_struct t;
233 int r;
234 char tmp[8];
235 u64 val;
236 int len;
237
238 t.fpscr.val = vcpu->arch.fpscr;
239
240 switch (ls_type) {
241 case FPU_LS_SINGLE:
242 cvt_df((double*)&vcpu->arch.fpr[rs], (float*)tmp, &t);
243 val = *((u32*)tmp);
244 len = sizeof(u32);
245 break;
246 case FPU_LS_SINGLE_LOW:
247 *((u32*)tmp) = vcpu->arch.fpr[rs];
248 val = vcpu->arch.fpr[rs] & 0xffffffff;
249 len = sizeof(u32);
250 break;
251 case FPU_LS_DOUBLE:
252 *((u64*)tmp) = vcpu->arch.fpr[rs];
253 val = vcpu->arch.fpr[rs];
254 len = sizeof(u64);
255 break;
256 default:
257 val = 0;
258 len = 0;
259 }
260
261 r = kvmppc_st(vcpu, &addr, len, tmp, true);
262 vcpu->arch.paddr_accessed = addr;
263 if (r < 0) {
264 kvmppc_inject_pf(vcpu, addr, true);
265 } else if (r == EMULATE_DO_MMIO) {
266 emulated = kvmppc_handle_store(run, vcpu, val, len, 1);
267 } else {
268 emulated = EMULATE_DONE;
269 }
270
271 dprintk(KERN_INFO "KVM: FPR_ST [0x%llx] at 0x%lx (%d)\n",
272 val, addr, len);
273
274 return emulated;
275}
276
277static int kvmppc_emulate_psq_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
278 int rs, ulong addr, bool w, int i)
279{
280 int emulated = EMULATE_FAIL;
281 struct thread_struct t;
282 int r;
283 float one = 1.0;
284 u32 tmp[2];
285
286 t.fpscr.val = vcpu->arch.fpscr;
287
288 /* read from memory */
289 if (w) {
290 r = kvmppc_ld(vcpu, &addr, sizeof(u32), tmp, true);
291 memcpy(&tmp[1], &one, sizeof(u32));
292 } else {
293 r = kvmppc_ld(vcpu, &addr, sizeof(u32) * 2, tmp, true);
294 }
295 vcpu->arch.paddr_accessed = addr;
296 if (r < 0) {
297 kvmppc_inject_pf(vcpu, addr, false);
298 goto done_load;
299 } else if ((r == EMULATE_DO_MMIO) && w) {
300 emulated = kvmppc_handle_load(run, vcpu, KVM_REG_FPR | rs, 4, 1);
301 vcpu->arch.qpr[rs] = tmp[1];
302 goto done_load;
303 } else if (r == EMULATE_DO_MMIO) {
304 emulated = kvmppc_handle_load(run, vcpu, KVM_REG_FQPR | rs, 8, 1);
305 goto done_load;
306 }
307
308 emulated = EMULATE_DONE;
309
310 /* put in registers */
311 cvt_fd((float*)&tmp[0], (double*)&vcpu->arch.fpr[rs], &t);
312 vcpu->arch.qpr[rs] = tmp[1];
313
314 dprintk(KERN_INFO "KVM: PSQ_LD [0x%x, 0x%x] at 0x%lx (%d)\n", tmp[0],
315 tmp[1], addr, w ? 4 : 8);
316
317done_load:
318 return emulated;
319}
320
321static int kvmppc_emulate_psq_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
322 int rs, ulong addr, bool w, int i)
323{
324 int emulated = EMULATE_FAIL;
325 struct thread_struct t;
326 int r;
327 u32 tmp[2];
328 int len = w ? sizeof(u32) : sizeof(u64);
329
330 t.fpscr.val = vcpu->arch.fpscr;
331
332 cvt_df((double*)&vcpu->arch.fpr[rs], (float*)&tmp[0], &t);
333 tmp[1] = vcpu->arch.qpr[rs];
334
335 r = kvmppc_st(vcpu, &addr, len, tmp, true);
336 vcpu->arch.paddr_accessed = addr;
337 if (r < 0) {
338 kvmppc_inject_pf(vcpu, addr, true);
339 } else if ((r == EMULATE_DO_MMIO) && w) {
340 emulated = kvmppc_handle_store(run, vcpu, tmp[0], 4, 1);
341 } else if (r == EMULATE_DO_MMIO) {
342 u64 val = ((u64)tmp[0] << 32) | tmp[1];
343 emulated = kvmppc_handle_store(run, vcpu, val, 8, 1);
344 } else {
345 emulated = EMULATE_DONE;
346 }
347
348 dprintk(KERN_INFO "KVM: PSQ_ST [0x%x, 0x%x] at 0x%lx (%d)\n",
349 tmp[0], tmp[1], addr, len);
350
351 return emulated;
352}
353
354/*
355 * Cuts out inst bits with ordering according to spec.
356 * That means the leftmost bit is zero. All given bits are included.
357 */
358static inline u32 inst_get_field(u32 inst, int msb, int lsb)
359{
360 return kvmppc_get_field(inst, msb + 32, lsb + 32);
361}
362
363/*
364 * Replaces inst bits with ordering according to spec.
365 */
366static inline u32 inst_set_field(u32 inst, int msb, int lsb, int value)
367{
368 return kvmppc_set_field(inst, msb + 32, lsb + 32, value);
369}
370
371bool kvmppc_inst_is_paired_single(struct kvm_vcpu *vcpu, u32 inst)
372{
373 if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
374 return false;
375
376 switch (get_op(inst)) {
377 case OP_PSQ_L:
378 case OP_PSQ_LU:
379 case OP_PSQ_ST:
380 case OP_PSQ_STU:
381 case OP_LFS:
382 case OP_LFSU:
383 case OP_LFD:
384 case OP_LFDU:
385 case OP_STFS:
386 case OP_STFSU:
387 case OP_STFD:
388 case OP_STFDU:
389 return true;
390 case 4:
391 /* X form */
392 switch (inst_get_field(inst, 21, 30)) {
393 case OP_4X_PS_CMPU0:
394 case OP_4X_PSQ_LX:
395 case OP_4X_PS_CMPO0:
396 case OP_4X_PSQ_LUX:
397 case OP_4X_PS_NEG:
398 case OP_4X_PS_CMPU1:
399 case OP_4X_PS_MR:
400 case OP_4X_PS_CMPO1:
401 case OP_4X_PS_NABS:
402 case OP_4X_PS_ABS:
403 case OP_4X_PS_MERGE00:
404 case OP_4X_PS_MERGE01:
405 case OP_4X_PS_MERGE10:
406 case OP_4X_PS_MERGE11:
407 return true;
408 }
409 /* XW form */
410 switch (inst_get_field(inst, 25, 30)) {
411 case OP_4XW_PSQ_STX:
412 case OP_4XW_PSQ_STUX:
413 return true;
414 }
415 /* A form */
416 switch (inst_get_field(inst, 26, 30)) {
417 case OP_4A_PS_SUM1:
418 case OP_4A_PS_SUM0:
419 case OP_4A_PS_MULS0:
420 case OP_4A_PS_MULS1:
421 case OP_4A_PS_MADDS0:
422 case OP_4A_PS_MADDS1:
423 case OP_4A_PS_DIV:
424 case OP_4A_PS_SUB:
425 case OP_4A_PS_ADD:
426 case OP_4A_PS_SEL:
427 case OP_4A_PS_RES:
428 case OP_4A_PS_MUL:
429 case OP_4A_PS_RSQRTE:
430 case OP_4A_PS_MSUB:
431 case OP_4A_PS_MADD:
432 case OP_4A_PS_NMSUB:
433 case OP_4A_PS_NMADD:
434 return true;
435 }
436 break;
437 case 59:
438 switch (inst_get_field(inst, 21, 30)) {
439 case OP_59_FADDS:
440 case OP_59_FSUBS:
441 case OP_59_FDIVS:
442 case OP_59_FRES:
443 case OP_59_FRSQRTES:
444 return true;
445 }
446 switch (inst_get_field(inst, 26, 30)) {
447 case OP_59_FMULS:
448 case OP_59_FMSUBS:
449 case OP_59_FMADDS:
450 case OP_59_FNMSUBS:
451 case OP_59_FNMADDS:
452 return true;
453 }
454 break;
455 case 63:
456 switch (inst_get_field(inst, 21, 30)) {
457 case OP_63_MTFSB0:
458 case OP_63_MTFSB1:
459 case OP_63_MTFSF:
460 case OP_63_MTFSFI:
461 case OP_63_MCRFS:
462 case OP_63_MFFS:
463 case OP_63_FCMPU:
464 case OP_63_FCMPO:
465 case OP_63_FNEG:
466 case OP_63_FMR:
467 case OP_63_FABS:
468 case OP_63_FRSP:
469 case OP_63_FDIV:
470 case OP_63_FADD:
471 case OP_63_FSUB:
472 case OP_63_FCTIW:
473 case OP_63_FCTIWZ:
474 case OP_63_FRSQRTE:
475 case OP_63_FCPSGN:
476 return true;
477 }
478 switch (inst_get_field(inst, 26, 30)) {
479 case OP_63_FMUL:
480 case OP_63_FSEL:
481 case OP_63_FMSUB:
482 case OP_63_FMADD:
483 case OP_63_FNMSUB:
484 case OP_63_FNMADD:
485 return true;
486 }
487 break;
488 case 31:
489 switch (inst_get_field(inst, 21, 30)) {
490 case OP_31_LFSX:
491 case OP_31_LFSUX:
492 case OP_31_LFDX:
493 case OP_31_LFDUX:
494 case OP_31_STFSX:
495 case OP_31_STFSUX:
496 case OP_31_STFX:
497 case OP_31_STFUX:
498 case OP_31_STFIWX:
499 return true;
500 }
501 break;
502 }
503
504 return false;
505}
506
507static int get_d_signext(u32 inst)
508{
509 int d = inst & 0x8ff;
510
511 if (d & 0x800)
512 return -(d & 0x7ff);
513
514 return (d & 0x7ff);
515}
516
517static int kvmppc_ps_three_in(struct kvm_vcpu *vcpu, bool rc,
518 int reg_out, int reg_in1, int reg_in2,
519 int reg_in3, int scalar,
520 void (*func)(struct thread_struct *t,
521 u32 *dst, u32 *src1,
522 u32 *src2, u32 *src3))
523{
524 u32 *qpr = vcpu->arch.qpr;
525 u64 *fpr = vcpu->arch.fpr;
526 u32 ps0_out;
527 u32 ps0_in1, ps0_in2, ps0_in3;
528 u32 ps1_in1, ps1_in2, ps1_in3;
529 struct thread_struct t;
530 t.fpscr.val = vcpu->arch.fpscr;
531
532 /* RC */
533 WARN_ON(rc);
534
535 /* PS0 */
536 cvt_df((double*)&fpr[reg_in1], (float*)&ps0_in1, &t);
537 cvt_df((double*)&fpr[reg_in2], (float*)&ps0_in2, &t);
538 cvt_df((double*)&fpr[reg_in3], (float*)&ps0_in3, &t);
539
540 if (scalar & SCALAR_LOW)
541 ps0_in2 = qpr[reg_in2];
542
543 func(&t, &ps0_out, &ps0_in1, &ps0_in2, &ps0_in3);
544
545 dprintk(KERN_INFO "PS3 ps0 -> f(0x%x, 0x%x, 0x%x) = 0x%x\n",
546 ps0_in1, ps0_in2, ps0_in3, ps0_out);
547
548 if (!(scalar & SCALAR_NO_PS0))
549 cvt_fd((float*)&ps0_out, (double*)&fpr[reg_out], &t);
550
551 /* PS1 */
552 ps1_in1 = qpr[reg_in1];
553 ps1_in2 = qpr[reg_in2];
554 ps1_in3 = qpr[reg_in3];
555
556 if (scalar & SCALAR_HIGH)
557 ps1_in2 = ps0_in2;
558
559 if (!(scalar & SCALAR_NO_PS1))
560 func(&t, &qpr[reg_out], &ps1_in1, &ps1_in2, &ps1_in3);
561
562 dprintk(KERN_INFO "PS3 ps1 -> f(0x%x, 0x%x, 0x%x) = 0x%x\n",
563 ps1_in1, ps1_in2, ps1_in3, qpr[reg_out]);
564
565 return EMULATE_DONE;
566}
567
568static int kvmppc_ps_two_in(struct kvm_vcpu *vcpu, bool rc,
569 int reg_out, int reg_in1, int reg_in2,
570 int scalar,
571 void (*func)(struct thread_struct *t,
572 u32 *dst, u32 *src1,
573 u32 *src2))
574{
575 u32 *qpr = vcpu->arch.qpr;
576 u64 *fpr = vcpu->arch.fpr;
577 u32 ps0_out;
578 u32 ps0_in1, ps0_in2;
579 u32 ps1_out;
580 u32 ps1_in1, ps1_in2;
581 struct thread_struct t;
582 t.fpscr.val = vcpu->arch.fpscr;
583
584 /* RC */
585 WARN_ON(rc);
586
587 /* PS0 */
588 cvt_df((double*)&fpr[reg_in1], (float*)&ps0_in1, &t);
589
590 if (scalar & SCALAR_LOW)
591 ps0_in2 = qpr[reg_in2];
592 else
593 cvt_df((double*)&fpr[reg_in2], (float*)&ps0_in2, &t);
594
595 func(&t, &ps0_out, &ps0_in1, &ps0_in2);
596
597 if (!(scalar & SCALAR_NO_PS0)) {
598 dprintk(KERN_INFO "PS2 ps0 -> f(0x%x, 0x%x) = 0x%x\n",
599 ps0_in1, ps0_in2, ps0_out);
600
601 cvt_fd((float*)&ps0_out, (double*)&fpr[reg_out], &t);
602 }
603
604 /* PS1 */
605 ps1_in1 = qpr[reg_in1];
606 ps1_in2 = qpr[reg_in2];
607
608 if (scalar & SCALAR_HIGH)
609 ps1_in2 = ps0_in2;
610
611 func(&t, &ps1_out, &ps1_in1, &ps1_in2);
612
613 if (!(scalar & SCALAR_NO_PS1)) {
614 qpr[reg_out] = ps1_out;
615
616 dprintk(KERN_INFO "PS2 ps1 -> f(0x%x, 0x%x) = 0x%x\n",
617 ps1_in1, ps1_in2, qpr[reg_out]);
618 }
619
620 return EMULATE_DONE;
621}
622
623static int kvmppc_ps_one_in(struct kvm_vcpu *vcpu, bool rc,
624 int reg_out, int reg_in,
625 void (*func)(struct thread_struct *t,
626 u32 *dst, u32 *src1))
627{
628 u32 *qpr = vcpu->arch.qpr;
629 u64 *fpr = vcpu->arch.fpr;
630 u32 ps0_out, ps0_in;
631 u32 ps1_in;
632 struct thread_struct t;
633 t.fpscr.val = vcpu->arch.fpscr;
634
635 /* RC */
636 WARN_ON(rc);
637
638 /* PS0 */
639 cvt_df((double*)&fpr[reg_in], (float*)&ps0_in, &t);
640 func(&t, &ps0_out, &ps0_in);
641
642 dprintk(KERN_INFO "PS1 ps0 -> f(0x%x) = 0x%x\n",
643 ps0_in, ps0_out);
644
645 cvt_fd((float*)&ps0_out, (double*)&fpr[reg_out], &t);
646
647 /* PS1 */
648 ps1_in = qpr[reg_in];
649 func(&t, &qpr[reg_out], &ps1_in);
650
651 dprintk(KERN_INFO "PS1 ps1 -> f(0x%x) = 0x%x\n",
652 ps1_in, qpr[reg_out]);
653
654 return EMULATE_DONE;
655}
656
657int kvmppc_emulate_paired_single(struct kvm_run *run, struct kvm_vcpu *vcpu)
658{
659 u32 inst = vcpu->arch.last_inst;
660 enum emulation_result emulated = EMULATE_DONE;
661
662 int ax_rd = inst_get_field(inst, 6, 10);
663 int ax_ra = inst_get_field(inst, 11, 15);
664 int ax_rb = inst_get_field(inst, 16, 20);
665 int ax_rc = inst_get_field(inst, 21, 25);
666 short full_d = inst_get_field(inst, 16, 31);
667
668 u64 *fpr_d = &vcpu->arch.fpr[ax_rd];
669 u64 *fpr_a = &vcpu->arch.fpr[ax_ra];
670 u64 *fpr_b = &vcpu->arch.fpr[ax_rb];
671 u64 *fpr_c = &vcpu->arch.fpr[ax_rc];
672
673 bool rcomp = (inst & 1) ? true : false;
674 u32 cr = kvmppc_get_cr(vcpu);
675 struct thread_struct t;
676#ifdef DEBUG
677 int i;
678#endif
679
680 t.fpscr.val = vcpu->arch.fpscr;
681
682 if (!kvmppc_inst_is_paired_single(vcpu, inst))
683 return EMULATE_FAIL;
684
685 if (!(vcpu->arch.msr & MSR_FP)) {
686 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL);
687 return EMULATE_AGAIN;
688 }
689
690 kvmppc_giveup_ext(vcpu, MSR_FP);
691 preempt_disable();
692 enable_kernel_fp();
693 /* Do we need to clear FE0 / FE1 here? Don't think so. */
694
695#ifdef DEBUG
696 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++) {
697 u32 f;
698 cvt_df((double*)&vcpu->arch.fpr[i], (float*)&f, &t);
699 dprintk(KERN_INFO "FPR[%d] = 0x%x / 0x%llx QPR[%d] = 0x%x\n",
700 i, f, vcpu->arch.fpr[i], i, vcpu->arch.qpr[i]);
701 }
702#endif
703
704 switch (get_op(inst)) {
705 case OP_PSQ_L:
706 {
707 ulong addr = ax_ra ? kvmppc_get_gpr(vcpu, ax_ra) : 0;
708 bool w = inst_get_field(inst, 16, 16) ? true : false;
709 int i = inst_get_field(inst, 17, 19);
710
711 addr += get_d_signext(inst);
712 emulated = kvmppc_emulate_psq_load(run, vcpu, ax_rd, addr, w, i);
713 break;
714 }
715 case OP_PSQ_LU:
716 {
717 ulong addr = kvmppc_get_gpr(vcpu, ax_ra);
718 bool w = inst_get_field(inst, 16, 16) ? true : false;
719 int i = inst_get_field(inst, 17, 19);
720
721 addr += get_d_signext(inst);
722 emulated = kvmppc_emulate_psq_load(run, vcpu, ax_rd, addr, w, i);
723
724 if (emulated == EMULATE_DONE)
725 kvmppc_set_gpr(vcpu, ax_ra, addr);
726 break;
727 }
728 case OP_PSQ_ST:
729 {
730 ulong addr = ax_ra ? kvmppc_get_gpr(vcpu, ax_ra) : 0;
731 bool w = inst_get_field(inst, 16, 16) ? true : false;
732 int i = inst_get_field(inst, 17, 19);
733
734 addr += get_d_signext(inst);
735 emulated = kvmppc_emulate_psq_store(run, vcpu, ax_rd, addr, w, i);
736 break;
737 }
738 case OP_PSQ_STU:
739 {
740 ulong addr = kvmppc_get_gpr(vcpu, ax_ra);
741 bool w = inst_get_field(inst, 16, 16) ? true : false;
742 int i = inst_get_field(inst, 17, 19);
743
744 addr += get_d_signext(inst);
745 emulated = kvmppc_emulate_psq_store(run, vcpu, ax_rd, addr, w, i);
746
747 if (emulated == EMULATE_DONE)
748 kvmppc_set_gpr(vcpu, ax_ra, addr);
749 break;
750 }
751 case 4:
752 /* X form */
753 switch (inst_get_field(inst, 21, 30)) {
754 case OP_4X_PS_CMPU0:
755 /* XXX */
756 emulated = EMULATE_FAIL;
757 break;
758 case OP_4X_PSQ_LX:
759 {
760 ulong addr = ax_ra ? kvmppc_get_gpr(vcpu, ax_ra) : 0;
761 bool w = inst_get_field(inst, 21, 21) ? true : false;
762 int i = inst_get_field(inst, 22, 24);
763
764 addr += kvmppc_get_gpr(vcpu, ax_rb);
765 emulated = kvmppc_emulate_psq_load(run, vcpu, ax_rd, addr, w, i);
766 break;
767 }
768 case OP_4X_PS_CMPO0:
769 /* XXX */
770 emulated = EMULATE_FAIL;
771 break;
772 case OP_4X_PSQ_LUX:
773 {
774 ulong addr = kvmppc_get_gpr(vcpu, ax_ra);
775 bool w = inst_get_field(inst, 21, 21) ? true : false;
776 int i = inst_get_field(inst, 22, 24);
777
778 addr += kvmppc_get_gpr(vcpu, ax_rb);
779 emulated = kvmppc_emulate_psq_load(run, vcpu, ax_rd, addr, w, i);
780
781 if (emulated == EMULATE_DONE)
782 kvmppc_set_gpr(vcpu, ax_ra, addr);
783 break;
784 }
785 case OP_4X_PS_NEG:
786 vcpu->arch.fpr[ax_rd] = vcpu->arch.fpr[ax_rb];
787 vcpu->arch.fpr[ax_rd] ^= 0x8000000000000000ULL;
788 vcpu->arch.qpr[ax_rd] = vcpu->arch.qpr[ax_rb];
789 vcpu->arch.qpr[ax_rd] ^= 0x80000000;
790 break;
791 case OP_4X_PS_CMPU1:
792 /* XXX */
793 emulated = EMULATE_FAIL;
794 break;
795 case OP_4X_PS_MR:
796 WARN_ON(rcomp);
797 vcpu->arch.fpr[ax_rd] = vcpu->arch.fpr[ax_rb];
798 vcpu->arch.qpr[ax_rd] = vcpu->arch.qpr[ax_rb];
799 break;
800 case OP_4X_PS_CMPO1:
801 /* XXX */
802 emulated = EMULATE_FAIL;
803 break;
804 case OP_4X_PS_NABS:
805 WARN_ON(rcomp);
806 vcpu->arch.fpr[ax_rd] = vcpu->arch.fpr[ax_rb];
807 vcpu->arch.fpr[ax_rd] |= 0x8000000000000000ULL;
808 vcpu->arch.qpr[ax_rd] = vcpu->arch.qpr[ax_rb];
809 vcpu->arch.qpr[ax_rd] |= 0x80000000;
810 break;
811 case OP_4X_PS_ABS:
812 WARN_ON(rcomp);
813 vcpu->arch.fpr[ax_rd] = vcpu->arch.fpr[ax_rb];
814 vcpu->arch.fpr[ax_rd] &= ~0x8000000000000000ULL;
815 vcpu->arch.qpr[ax_rd] = vcpu->arch.qpr[ax_rb];
816 vcpu->arch.qpr[ax_rd] &= ~0x80000000;
817 break;
818 case OP_4X_PS_MERGE00:
819 WARN_ON(rcomp);
820 vcpu->arch.fpr[ax_rd] = vcpu->arch.fpr[ax_ra];
821 /* vcpu->arch.qpr[ax_rd] = vcpu->arch.fpr[ax_rb]; */
822 cvt_df((double*)&vcpu->arch.fpr[ax_rb],
823 (float*)&vcpu->arch.qpr[ax_rd], &t);
824 break;
825 case OP_4X_PS_MERGE01:
826 WARN_ON(rcomp);
827 vcpu->arch.fpr[ax_rd] = vcpu->arch.fpr[ax_ra];
828 vcpu->arch.qpr[ax_rd] = vcpu->arch.qpr[ax_rb];
829 break;
830 case OP_4X_PS_MERGE10:
831 WARN_ON(rcomp);
832 /* vcpu->arch.fpr[ax_rd] = vcpu->arch.qpr[ax_ra]; */
833 cvt_fd((float*)&vcpu->arch.qpr[ax_ra],
834 (double*)&vcpu->arch.fpr[ax_rd], &t);
835 /* vcpu->arch.qpr[ax_rd] = vcpu->arch.fpr[ax_rb]; */
836 cvt_df((double*)&vcpu->arch.fpr[ax_rb],
837 (float*)&vcpu->arch.qpr[ax_rd], &t);
838 break;
839 case OP_4X_PS_MERGE11:
840 WARN_ON(rcomp);
841 /* vcpu->arch.fpr[ax_rd] = vcpu->arch.qpr[ax_ra]; */
842 cvt_fd((float*)&vcpu->arch.qpr[ax_ra],
843 (double*)&vcpu->arch.fpr[ax_rd], &t);
844 vcpu->arch.qpr[ax_rd] = vcpu->arch.qpr[ax_rb];
845 break;
846 }
847 /* XW form */
848 switch (inst_get_field(inst, 25, 30)) {
849 case OP_4XW_PSQ_STX:
850 {
851 ulong addr = ax_ra ? kvmppc_get_gpr(vcpu, ax_ra) : 0;
852 bool w = inst_get_field(inst, 21, 21) ? true : false;
853 int i = inst_get_field(inst, 22, 24);
854
855 addr += kvmppc_get_gpr(vcpu, ax_rb);
856 emulated = kvmppc_emulate_psq_store(run, vcpu, ax_rd, addr, w, i);
857 break;
858 }
859 case OP_4XW_PSQ_STUX:
860 {
861 ulong addr = kvmppc_get_gpr(vcpu, ax_ra);
862 bool w = inst_get_field(inst, 21, 21) ? true : false;
863 int i = inst_get_field(inst, 22, 24);
864
865 addr += kvmppc_get_gpr(vcpu, ax_rb);
866 emulated = kvmppc_emulate_psq_store(run, vcpu, ax_rd, addr, w, i);
867
868 if (emulated == EMULATE_DONE)
869 kvmppc_set_gpr(vcpu, ax_ra, addr);
870 break;
871 }
872 }
873 /* A form */
874 switch (inst_get_field(inst, 26, 30)) {
875 case OP_4A_PS_SUM1:
876 emulated = kvmppc_ps_two_in(vcpu, rcomp, ax_rd,
877 ax_rb, ax_ra, SCALAR_NO_PS0 | SCALAR_HIGH, fps_fadds);
878 vcpu->arch.fpr[ax_rd] = vcpu->arch.fpr[ax_rc];
879 break;
880 case OP_4A_PS_SUM0:
881 emulated = kvmppc_ps_two_in(vcpu, rcomp, ax_rd,
882 ax_ra, ax_rb, SCALAR_NO_PS1 | SCALAR_LOW, fps_fadds);
883 vcpu->arch.qpr[ax_rd] = vcpu->arch.qpr[ax_rc];
884 break;
885 case OP_4A_PS_MULS0:
886 emulated = kvmppc_ps_two_in(vcpu, rcomp, ax_rd,
887 ax_ra, ax_rc, SCALAR_HIGH, fps_fmuls);
888 break;
889 case OP_4A_PS_MULS1:
890 emulated = kvmppc_ps_two_in(vcpu, rcomp, ax_rd,
891 ax_ra, ax_rc, SCALAR_LOW, fps_fmuls);
892 break;
893 case OP_4A_PS_MADDS0:
894 emulated = kvmppc_ps_three_in(vcpu, rcomp, ax_rd,
895 ax_ra, ax_rc, ax_rb, SCALAR_HIGH, fps_fmadds);
896 break;
897 case OP_4A_PS_MADDS1:
898 emulated = kvmppc_ps_three_in(vcpu, rcomp, ax_rd,
899 ax_ra, ax_rc, ax_rb, SCALAR_LOW, fps_fmadds);
900 break;
901 case OP_4A_PS_DIV:
902 emulated = kvmppc_ps_two_in(vcpu, rcomp, ax_rd,
903 ax_ra, ax_rb, SCALAR_NONE, fps_fdivs);
904 break;
905 case OP_4A_PS_SUB:
906 emulated = kvmppc_ps_two_in(vcpu, rcomp, ax_rd,
907 ax_ra, ax_rb, SCALAR_NONE, fps_fsubs);
908 break;
909 case OP_4A_PS_ADD:
910 emulated = kvmppc_ps_two_in(vcpu, rcomp, ax_rd,
911 ax_ra, ax_rb, SCALAR_NONE, fps_fadds);
912 break;
913 case OP_4A_PS_SEL:
914 emulated = kvmppc_ps_three_in(vcpu, rcomp, ax_rd,
915 ax_ra, ax_rc, ax_rb, SCALAR_NONE, fps_fsel);
916 break;
917 case OP_4A_PS_RES:
918 emulated = kvmppc_ps_one_in(vcpu, rcomp, ax_rd,
919 ax_rb, fps_fres);
920 break;
921 case OP_4A_PS_MUL:
922 emulated = kvmppc_ps_two_in(vcpu, rcomp, ax_rd,
923 ax_ra, ax_rc, SCALAR_NONE, fps_fmuls);
924 break;
925 case OP_4A_PS_RSQRTE:
926 emulated = kvmppc_ps_one_in(vcpu, rcomp, ax_rd,
927 ax_rb, fps_frsqrte);
928 break;
929 case OP_4A_PS_MSUB:
930 emulated = kvmppc_ps_three_in(vcpu, rcomp, ax_rd,
931 ax_ra, ax_rc, ax_rb, SCALAR_NONE, fps_fmsubs);
932 break;
933 case OP_4A_PS_MADD:
934 emulated = kvmppc_ps_three_in(vcpu, rcomp, ax_rd,
935 ax_ra, ax_rc, ax_rb, SCALAR_NONE, fps_fmadds);
936 break;
937 case OP_4A_PS_NMSUB:
938 emulated = kvmppc_ps_three_in(vcpu, rcomp, ax_rd,
939 ax_ra, ax_rc, ax_rb, SCALAR_NONE, fps_fnmsubs);
940 break;
941 case OP_4A_PS_NMADD:
942 emulated = kvmppc_ps_three_in(vcpu, rcomp, ax_rd,
943 ax_ra, ax_rc, ax_rb, SCALAR_NONE, fps_fnmadds);
944 break;
945 }
946 break;
947
948 /* Real FPU operations */
949
950 case OP_LFS:
951 {
952 ulong addr = (ax_ra ? kvmppc_get_gpr(vcpu, ax_ra) : 0) + full_d;
953
954 emulated = kvmppc_emulate_fpr_load(run, vcpu, ax_rd, addr,
955 FPU_LS_SINGLE);
956 break;
957 }
958 case OP_LFSU:
959 {
960 ulong addr = kvmppc_get_gpr(vcpu, ax_ra) + full_d;
961
962 emulated = kvmppc_emulate_fpr_load(run, vcpu, ax_rd, addr,
963 FPU_LS_SINGLE);
964
965 if (emulated == EMULATE_DONE)
966 kvmppc_set_gpr(vcpu, ax_ra, addr);
967 break;
968 }
969 case OP_LFD:
970 {
971 ulong addr = (ax_ra ? kvmppc_get_gpr(vcpu, ax_ra) : 0) + full_d;
972
973 emulated = kvmppc_emulate_fpr_load(run, vcpu, ax_rd, addr,
974 FPU_LS_DOUBLE);
975 break;
976 }
977 case OP_LFDU:
978 {
979 ulong addr = kvmppc_get_gpr(vcpu, ax_ra) + full_d;
980
981 emulated = kvmppc_emulate_fpr_load(run, vcpu, ax_rd, addr,
982 FPU_LS_DOUBLE);
983
984 if (emulated == EMULATE_DONE)
985 kvmppc_set_gpr(vcpu, ax_ra, addr);
986 break;
987 }
988 case OP_STFS:
989 {
990 ulong addr = (ax_ra ? kvmppc_get_gpr(vcpu, ax_ra) : 0) + full_d;
991
992 emulated = kvmppc_emulate_fpr_store(run, vcpu, ax_rd, addr,
993 FPU_LS_SINGLE);
994 break;
995 }
996 case OP_STFSU:
997 {
998 ulong addr = kvmppc_get_gpr(vcpu, ax_ra) + full_d;
999
1000 emulated = kvmppc_emulate_fpr_store(run, vcpu, ax_rd, addr,
1001 FPU_LS_SINGLE);
1002
1003 if (emulated == EMULATE_DONE)
1004 kvmppc_set_gpr(vcpu, ax_ra, addr);
1005 break;
1006 }
1007 case OP_STFD:
1008 {
1009 ulong addr = (ax_ra ? kvmppc_get_gpr(vcpu, ax_ra) : 0) + full_d;
1010
1011 emulated = kvmppc_emulate_fpr_store(run, vcpu, ax_rd, addr,
1012 FPU_LS_DOUBLE);
1013 break;
1014 }
1015 case OP_STFDU:
1016 {
1017 ulong addr = kvmppc_get_gpr(vcpu, ax_ra) + full_d;
1018
1019 emulated = kvmppc_emulate_fpr_store(run, vcpu, ax_rd, addr,
1020 FPU_LS_DOUBLE);
1021
1022 if (emulated == EMULATE_DONE)
1023 kvmppc_set_gpr(vcpu, ax_ra, addr);
1024 break;
1025 }
1026 case 31:
1027 switch (inst_get_field(inst, 21, 30)) {
1028 case OP_31_LFSX:
1029 {
1030 ulong addr = ax_ra ? kvmppc_get_gpr(vcpu, ax_ra) : 0;
1031
1032 addr += kvmppc_get_gpr(vcpu, ax_rb);
1033 emulated = kvmppc_emulate_fpr_load(run, vcpu, ax_rd,
1034 addr, FPU_LS_SINGLE);
1035 break;
1036 }
1037 case OP_31_LFSUX:
1038 {
1039 ulong addr = kvmppc_get_gpr(vcpu, ax_ra) +
1040 kvmppc_get_gpr(vcpu, ax_rb);
1041
1042 emulated = kvmppc_emulate_fpr_load(run, vcpu, ax_rd,
1043 addr, FPU_LS_SINGLE);
1044
1045 if (emulated == EMULATE_DONE)
1046 kvmppc_set_gpr(vcpu, ax_ra, addr);
1047 break;
1048 }
1049 case OP_31_LFDX:
1050 {
1051 ulong addr = (ax_ra ? kvmppc_get_gpr(vcpu, ax_ra) : 0) +
1052 kvmppc_get_gpr(vcpu, ax_rb);
1053
1054 emulated = kvmppc_emulate_fpr_load(run, vcpu, ax_rd,
1055 addr, FPU_LS_DOUBLE);
1056 break;
1057 }
1058 case OP_31_LFDUX:
1059 {
1060 ulong addr = kvmppc_get_gpr(vcpu, ax_ra) +
1061 kvmppc_get_gpr(vcpu, ax_rb);
1062
1063 emulated = kvmppc_emulate_fpr_load(run, vcpu, ax_rd,
1064 addr, FPU_LS_DOUBLE);
1065
1066 if (emulated == EMULATE_DONE)
1067 kvmppc_set_gpr(vcpu, ax_ra, addr);
1068 break;
1069 }
1070 case OP_31_STFSX:
1071 {
1072 ulong addr = (ax_ra ? kvmppc_get_gpr(vcpu, ax_ra) : 0) +
1073 kvmppc_get_gpr(vcpu, ax_rb);
1074
1075 emulated = kvmppc_emulate_fpr_store(run, vcpu, ax_rd,
1076 addr, FPU_LS_SINGLE);
1077 break;
1078 }
1079 case OP_31_STFSUX:
1080 {
1081 ulong addr = kvmppc_get_gpr(vcpu, ax_ra) +
1082 kvmppc_get_gpr(vcpu, ax_rb);
1083
1084 emulated = kvmppc_emulate_fpr_store(run, vcpu, ax_rd,
1085 addr, FPU_LS_SINGLE);
1086
1087 if (emulated == EMULATE_DONE)
1088 kvmppc_set_gpr(vcpu, ax_ra, addr);
1089 break;
1090 }
1091 case OP_31_STFX:
1092 {
1093 ulong addr = (ax_ra ? kvmppc_get_gpr(vcpu, ax_ra) : 0) +
1094 kvmppc_get_gpr(vcpu, ax_rb);
1095
1096 emulated = kvmppc_emulate_fpr_store(run, vcpu, ax_rd,
1097 addr, FPU_LS_DOUBLE);
1098 break;
1099 }
1100 case OP_31_STFUX:
1101 {
1102 ulong addr = kvmppc_get_gpr(vcpu, ax_ra) +
1103 kvmppc_get_gpr(vcpu, ax_rb);
1104
1105 emulated = kvmppc_emulate_fpr_store(run, vcpu, ax_rd,
1106 addr, FPU_LS_DOUBLE);
1107
1108 if (emulated == EMULATE_DONE)
1109 kvmppc_set_gpr(vcpu, ax_ra, addr);
1110 break;
1111 }
1112 case OP_31_STFIWX:
1113 {
1114 ulong addr = (ax_ra ? kvmppc_get_gpr(vcpu, ax_ra) : 0) +
1115 kvmppc_get_gpr(vcpu, ax_rb);
1116
1117 emulated = kvmppc_emulate_fpr_store(run, vcpu, ax_rd,
1118 addr,
1119 FPU_LS_SINGLE_LOW);
1120 break;
1121 }
1122 break;
1123 }
1124 break;
1125 case 59:
1126 switch (inst_get_field(inst, 21, 30)) {
1127 case OP_59_FADDS:
1128 fpd_fadds(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_b);
1129 kvmppc_sync_qpr(vcpu, ax_rd);
1130 break;
1131 case OP_59_FSUBS:
1132 fpd_fsubs(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_b);
1133 kvmppc_sync_qpr(vcpu, ax_rd);
1134 break;
1135 case OP_59_FDIVS:
1136 fpd_fdivs(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_b);
1137 kvmppc_sync_qpr(vcpu, ax_rd);
1138 break;
1139 case OP_59_FRES:
1140 fpd_fres(&vcpu->arch.fpscr, &cr, fpr_d, fpr_b);
1141 kvmppc_sync_qpr(vcpu, ax_rd);
1142 break;
1143 case OP_59_FRSQRTES:
1144 fpd_frsqrtes(&vcpu->arch.fpscr, &cr, fpr_d, fpr_b);
1145 kvmppc_sync_qpr(vcpu, ax_rd);
1146 break;
1147 }
1148 switch (inst_get_field(inst, 26, 30)) {
1149 case OP_59_FMULS:
1150 fpd_fmuls(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_c);
1151 kvmppc_sync_qpr(vcpu, ax_rd);
1152 break;
1153 case OP_59_FMSUBS:
1154 fpd_fmsubs(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_c, fpr_b);
1155 kvmppc_sync_qpr(vcpu, ax_rd);
1156 break;
1157 case OP_59_FMADDS:
1158 fpd_fmadds(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_c, fpr_b);
1159 kvmppc_sync_qpr(vcpu, ax_rd);
1160 break;
1161 case OP_59_FNMSUBS:
1162 fpd_fnmsubs(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_c, fpr_b);
1163 kvmppc_sync_qpr(vcpu, ax_rd);
1164 break;
1165 case OP_59_FNMADDS:
1166 fpd_fnmadds(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_c, fpr_b);
1167 kvmppc_sync_qpr(vcpu, ax_rd);
1168 break;
1169 }
1170 break;
1171 case 63:
1172 switch (inst_get_field(inst, 21, 30)) {
1173 case OP_63_MTFSB0:
1174 case OP_63_MTFSB1:
1175 case OP_63_MCRFS:
1176 case OP_63_MTFSFI:
1177 /* XXX need to implement */
1178 break;
1179 case OP_63_MFFS:
1180 /* XXX missing CR */
1181 *fpr_d = vcpu->arch.fpscr;
1182 break;
1183 case OP_63_MTFSF:
1184 /* XXX missing fm bits */
1185 /* XXX missing CR */
1186 vcpu->arch.fpscr = *fpr_b;
1187 break;
1188 case OP_63_FCMPU:
1189 {
1190 u32 tmp_cr;
1191 u32 cr0_mask = 0xf0000000;
1192 u32 cr_shift = inst_get_field(inst, 6, 8) * 4;
1193
1194 fpd_fcmpu(&vcpu->arch.fpscr, &tmp_cr, fpr_a, fpr_b);
1195 cr &= ~(cr0_mask >> cr_shift);
1196 cr |= (cr & cr0_mask) >> cr_shift;
1197 break;
1198 }
1199 case OP_63_FCMPO:
1200 {
1201 u32 tmp_cr;
1202 u32 cr0_mask = 0xf0000000;
1203 u32 cr_shift = inst_get_field(inst, 6, 8) * 4;
1204
1205 fpd_fcmpo(&vcpu->arch.fpscr, &tmp_cr, fpr_a, fpr_b);
1206 cr &= ~(cr0_mask >> cr_shift);
1207 cr |= (cr & cr0_mask) >> cr_shift;
1208 break;
1209 }
1210 case OP_63_FNEG:
1211 fpd_fneg(&vcpu->arch.fpscr, &cr, fpr_d, fpr_b);
1212 break;
1213 case OP_63_FMR:
1214 *fpr_d = *fpr_b;
1215 break;
1216 case OP_63_FABS:
1217 fpd_fabs(&vcpu->arch.fpscr, &cr, fpr_d, fpr_b);
1218 break;
1219 case OP_63_FCPSGN:
1220 fpd_fcpsgn(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_b);
1221 break;
1222 case OP_63_FDIV:
1223 fpd_fdiv(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_b);
1224 break;
1225 case OP_63_FADD:
1226 fpd_fadd(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_b);
1227 break;
1228 case OP_63_FSUB:
1229 fpd_fsub(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_b);
1230 break;
1231 case OP_63_FCTIW:
1232 fpd_fctiw(&vcpu->arch.fpscr, &cr, fpr_d, fpr_b);
1233 break;
1234 case OP_63_FCTIWZ:
1235 fpd_fctiwz(&vcpu->arch.fpscr, &cr, fpr_d, fpr_b);
1236 break;
1237 case OP_63_FRSP:
1238 fpd_frsp(&vcpu->arch.fpscr, &cr, fpr_d, fpr_b);
1239 kvmppc_sync_qpr(vcpu, ax_rd);
1240 break;
1241 case OP_63_FRSQRTE:
1242 {
1243 double one = 1.0f;
1244
1245 /* fD = sqrt(fB) */
1246 fpd_fsqrt(&vcpu->arch.fpscr, &cr, fpr_d, fpr_b);
1247 /* fD = 1.0f / fD */
1248 fpd_fdiv(&vcpu->arch.fpscr, &cr, fpr_d, (u64*)&one, fpr_d);
1249 break;
1250 }
1251 }
1252 switch (inst_get_field(inst, 26, 30)) {
1253 case OP_63_FMUL:
1254 fpd_fmul(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_c);
1255 break;
1256 case OP_63_FSEL:
1257 fpd_fsel(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_c, fpr_b);
1258 break;
1259 case OP_63_FMSUB:
1260 fpd_fmsub(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_c, fpr_b);
1261 break;
1262 case OP_63_FMADD:
1263 fpd_fmadd(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_c, fpr_b);
1264 break;
1265 case OP_63_FNMSUB:
1266 fpd_fnmsub(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_c, fpr_b);
1267 break;
1268 case OP_63_FNMADD:
1269 fpd_fnmadd(&vcpu->arch.fpscr, &cr, fpr_d, fpr_a, fpr_c, fpr_b);
1270 break;
1271 }
1272 break;
1273 }
1274
1275#ifdef DEBUG
1276 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++) {
1277 u32 f;
1278 cvt_df((double*)&vcpu->arch.fpr[i], (float*)&f, &t);
1279 dprintk(KERN_INFO "FPR[%d] = 0x%x\n", i, f);
1280 }
1281#endif
1282
1283 if (rcomp)
1284 kvmppc_set_cr(vcpu, cr);
1285
1286 preempt_enable();
1287
1288 return emulated;
1289}
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-28 22:27:14 -0500