xref: /openbmc/qemu/tcg/s390x/tcg-target.c.inc (revision effd60c8)
1/*
2 * Tiny Code Generator for QEMU
3 *
4 * Copyright (c) 2009 Ulrich Hecht <uli@suse.de>
5 * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
6 * Copyright (c) 2010 Richard Henderson <rth@twiddle.net>
7 *
8 * Permission is hereby granted, free of charge, to any person obtaining a copy
9 * of this software and associated documentation files (the "Software"), to deal
10 * in the Software without restriction, including without limitation the rights
11 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12 * copies of the Software, and to permit persons to whom the Software is
13 * furnished to do so, subject to the following conditions:
14 *
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 * THE SOFTWARE.
25 */
26
27#include "../tcg-ldst.c.inc"
28#include "../tcg-pool.c.inc"
29#include "elf.h"
30
31#define TCG_CT_CONST_S16        (1 << 8)
32#define TCG_CT_CONST_S32        (1 << 9)
33#define TCG_CT_CONST_S33        (1 << 10)
34#define TCG_CT_CONST_ZERO       (1 << 11)
35#define TCG_CT_CONST_P32        (1 << 12)
36#define TCG_CT_CONST_INV        (1 << 13)
37#define TCG_CT_CONST_INVRISBG   (1 << 14)
38
39#define ALL_GENERAL_REGS     MAKE_64BIT_MASK(0, 16)
40#define ALL_VECTOR_REGS      MAKE_64BIT_MASK(32, 32)
41
42/* Several places within the instruction set 0 means "no register"
43   rather than TCG_REG_R0.  */
44#define TCG_REG_NONE    0
45
46/* A scratch register that may be be used throughout the backend.  */
47#define TCG_TMP0        TCG_REG_R1
48
49#define TCG_GUEST_BASE_REG TCG_REG_R13
50
51/* All of the following instructions are prefixed with their instruction
52   format, and are defined as 8- or 16-bit quantities, even when the two
53   halves of the 16-bit quantity may appear 32 bits apart in the insn.
54   This makes it easy to copy the values from the tables in Appendix B.  */
55typedef enum S390Opcode {
56    RIL_AFI     = 0xc209,
57    RIL_AGFI    = 0xc208,
58    RIL_ALFI    = 0xc20b,
59    RIL_ALGFI   = 0xc20a,
60    RIL_BRASL   = 0xc005,
61    RIL_BRCL    = 0xc004,
62    RIL_CFI     = 0xc20d,
63    RIL_CGFI    = 0xc20c,
64    RIL_CLFI    = 0xc20f,
65    RIL_CLGFI   = 0xc20e,
66    RIL_CLRL    = 0xc60f,
67    RIL_CLGRL   = 0xc60a,
68    RIL_CRL     = 0xc60d,
69    RIL_CGRL    = 0xc608,
70    RIL_IIHF    = 0xc008,
71    RIL_IILF    = 0xc009,
72    RIL_LARL    = 0xc000,
73    RIL_LGFI    = 0xc001,
74    RIL_LGRL    = 0xc408,
75    RIL_LLIHF   = 0xc00e,
76    RIL_LLILF   = 0xc00f,
77    RIL_LRL     = 0xc40d,
78    RIL_MSFI    = 0xc201,
79    RIL_MSGFI   = 0xc200,
80    RIL_NIHF    = 0xc00a,
81    RIL_NILF    = 0xc00b,
82    RIL_OIHF    = 0xc00c,
83    RIL_OILF    = 0xc00d,
84    RIL_SLFI    = 0xc205,
85    RIL_SLGFI   = 0xc204,
86    RIL_XIHF    = 0xc006,
87    RIL_XILF    = 0xc007,
88
89    RI_AGHI     = 0xa70b,
90    RI_AHI      = 0xa70a,
91    RI_BRC      = 0xa704,
92    RI_CHI      = 0xa70e,
93    RI_CGHI     = 0xa70f,
94    RI_IIHH     = 0xa500,
95    RI_IIHL     = 0xa501,
96    RI_IILH     = 0xa502,
97    RI_IILL     = 0xa503,
98    RI_LGHI     = 0xa709,
99    RI_LLIHH    = 0xa50c,
100    RI_LLIHL    = 0xa50d,
101    RI_LLILH    = 0xa50e,
102    RI_LLILL    = 0xa50f,
103    RI_MGHI     = 0xa70d,
104    RI_MHI      = 0xa70c,
105    RI_NIHH     = 0xa504,
106    RI_NIHL     = 0xa505,
107    RI_NILH     = 0xa506,
108    RI_NILL     = 0xa507,
109    RI_OIHH     = 0xa508,
110    RI_OIHL     = 0xa509,
111    RI_OILH     = 0xa50a,
112    RI_OILL     = 0xa50b,
113    RI_TMLL     = 0xa701,
114
115    RIEb_CGRJ    = 0xec64,
116    RIEb_CLGRJ   = 0xec65,
117    RIEb_CLRJ    = 0xec77,
118    RIEb_CRJ     = 0xec76,
119
120    RIEc_CGIJ    = 0xec7c,
121    RIEc_CIJ     = 0xec7e,
122    RIEc_CLGIJ   = 0xec7d,
123    RIEc_CLIJ    = 0xec7f,
124
125    RIEf_RISBG   = 0xec55,
126
127    RIEg_LOCGHI  = 0xec46,
128
129    RRE_AGR     = 0xb908,
130    RRE_ALGR    = 0xb90a,
131    RRE_ALCR    = 0xb998,
132    RRE_ALCGR   = 0xb988,
133    RRE_ALGFR   = 0xb91a,
134    RRE_CGR     = 0xb920,
135    RRE_CLGR    = 0xb921,
136    RRE_DLGR    = 0xb987,
137    RRE_DLR     = 0xb997,
138    RRE_DSGFR   = 0xb91d,
139    RRE_DSGR    = 0xb90d,
140    RRE_FLOGR   = 0xb983,
141    RRE_LGBR    = 0xb906,
142    RRE_LCGR    = 0xb903,
143    RRE_LGFR    = 0xb914,
144    RRE_LGHR    = 0xb907,
145    RRE_LGR     = 0xb904,
146    RRE_LLGCR   = 0xb984,
147    RRE_LLGFR   = 0xb916,
148    RRE_LLGHR   = 0xb985,
149    RRE_LRVR    = 0xb91f,
150    RRE_LRVGR   = 0xb90f,
151    RRE_LTGR    = 0xb902,
152    RRE_MLGR    = 0xb986,
153    RRE_MSGR    = 0xb90c,
154    RRE_MSR     = 0xb252,
155    RRE_NGR     = 0xb980,
156    RRE_OGR     = 0xb981,
157    RRE_SGR     = 0xb909,
158    RRE_SLGR    = 0xb90b,
159    RRE_SLBR    = 0xb999,
160    RRE_SLBGR   = 0xb989,
161    RRE_XGR     = 0xb982,
162
163    RRFa_MGRK   = 0xb9ec,
164    RRFa_MSRKC  = 0xb9fd,
165    RRFa_MSGRKC = 0xb9ed,
166    RRFa_NCRK   = 0xb9f5,
167    RRFa_NCGRK  = 0xb9e5,
168    RRFa_NNRK   = 0xb974,
169    RRFa_NNGRK  = 0xb964,
170    RRFa_NORK   = 0xb976,
171    RRFa_NOGRK  = 0xb966,
172    RRFa_NRK    = 0xb9f4,
173    RRFa_NGRK   = 0xb9e4,
174    RRFa_NXRK   = 0xb977,
175    RRFa_NXGRK  = 0xb967,
176    RRFa_OCRK   = 0xb975,
177    RRFa_OCGRK  = 0xb965,
178    RRFa_ORK    = 0xb9f6,
179    RRFa_OGRK   = 0xb9e6,
180    RRFa_SRK    = 0xb9f9,
181    RRFa_SGRK   = 0xb9e9,
182    RRFa_SLRK   = 0xb9fb,
183    RRFa_SLGRK  = 0xb9eb,
184    RRFa_XRK    = 0xb9f7,
185    RRFa_XGRK   = 0xb9e7,
186
187    RRFam_SELGR = 0xb9e3,
188
189    RRFc_LOCR   = 0xb9f2,
190    RRFc_LOCGR  = 0xb9e2,
191    RRFc_POPCNT = 0xb9e1,
192
193    RR_AR       = 0x1a,
194    RR_ALR      = 0x1e,
195    RR_BASR     = 0x0d,
196    RR_BCR      = 0x07,
197    RR_CLR      = 0x15,
198    RR_CR       = 0x19,
199    RR_DR       = 0x1d,
200    RR_LCR      = 0x13,
201    RR_LR       = 0x18,
202    RR_LTR      = 0x12,
203    RR_NR       = 0x14,
204    RR_OR       = 0x16,
205    RR_SR       = 0x1b,
206    RR_SLR      = 0x1f,
207    RR_XR       = 0x17,
208
209    RSY_RLL     = 0xeb1d,
210    RSY_RLLG    = 0xeb1c,
211    RSY_SLLG    = 0xeb0d,
212    RSY_SLLK    = 0xebdf,
213    RSY_SRAG    = 0xeb0a,
214    RSY_SRAK    = 0xebdc,
215    RSY_SRLG    = 0xeb0c,
216    RSY_SRLK    = 0xebde,
217
218    RS_SLL      = 0x89,
219    RS_SRA      = 0x8a,
220    RS_SRL      = 0x88,
221
222    RXY_AG      = 0xe308,
223    RXY_AY      = 0xe35a,
224    RXY_CG      = 0xe320,
225    RXY_CLG     = 0xe321,
226    RXY_CLY     = 0xe355,
227    RXY_CY      = 0xe359,
228    RXY_LAY     = 0xe371,
229    RXY_LB      = 0xe376,
230    RXY_LG      = 0xe304,
231    RXY_LGB     = 0xe377,
232    RXY_LGF     = 0xe314,
233    RXY_LGH     = 0xe315,
234    RXY_LHY     = 0xe378,
235    RXY_LLGC    = 0xe390,
236    RXY_LLGF    = 0xe316,
237    RXY_LLGH    = 0xe391,
238    RXY_LMG     = 0xeb04,
239    RXY_LPQ     = 0xe38f,
240    RXY_LRV     = 0xe31e,
241    RXY_LRVG    = 0xe30f,
242    RXY_LRVH    = 0xe31f,
243    RXY_LY      = 0xe358,
244    RXY_NG      = 0xe380,
245    RXY_OG      = 0xe381,
246    RXY_STCY    = 0xe372,
247    RXY_STG     = 0xe324,
248    RXY_STHY    = 0xe370,
249    RXY_STMG    = 0xeb24,
250    RXY_STPQ    = 0xe38e,
251    RXY_STRV    = 0xe33e,
252    RXY_STRVG   = 0xe32f,
253    RXY_STRVH   = 0xe33f,
254    RXY_STY     = 0xe350,
255    RXY_XG      = 0xe382,
256
257    RX_A        = 0x5a,
258    RX_C        = 0x59,
259    RX_L        = 0x58,
260    RX_LA       = 0x41,
261    RX_LH       = 0x48,
262    RX_ST       = 0x50,
263    RX_STC      = 0x42,
264    RX_STH      = 0x40,
265
266    VRIa_VGBM   = 0xe744,
267    VRIa_VREPI  = 0xe745,
268    VRIb_VGM    = 0xe746,
269    VRIc_VREP   = 0xe74d,
270
271    VRRa_VLC    = 0xe7de,
272    VRRa_VLP    = 0xe7df,
273    VRRa_VLR    = 0xe756,
274    VRRc_VA     = 0xe7f3,
275    VRRc_VCEQ   = 0xe7f8,   /* we leave the m5 cs field 0 */
276    VRRc_VCH    = 0xe7fb,   /* " */
277    VRRc_VCHL   = 0xe7f9,   /* " */
278    VRRc_VERLLV = 0xe773,
279    VRRc_VESLV  = 0xe770,
280    VRRc_VESRAV = 0xe77a,
281    VRRc_VESRLV = 0xe778,
282    VRRc_VML    = 0xe7a2,
283    VRRc_VMN    = 0xe7fe,
284    VRRc_VMNL   = 0xe7fc,
285    VRRc_VMX    = 0xe7ff,
286    VRRc_VMXL   = 0xe7fd,
287    VRRc_VN     = 0xe768,
288    VRRc_VNC    = 0xe769,
289    VRRc_VNN    = 0xe76e,
290    VRRc_VNO    = 0xe76b,
291    VRRc_VNX    = 0xe76c,
292    VRRc_VO     = 0xe76a,
293    VRRc_VOC    = 0xe76f,
294    VRRc_VPKS   = 0xe797,   /* we leave the m5 cs field 0 */
295    VRRc_VS     = 0xe7f7,
296    VRRa_VUPH   = 0xe7d7,
297    VRRa_VUPL   = 0xe7d6,
298    VRRc_VX     = 0xe76d,
299    VRRe_VSEL   = 0xe78d,
300    VRRf_VLVGP  = 0xe762,
301
302    VRSa_VERLL  = 0xe733,
303    VRSa_VESL   = 0xe730,
304    VRSa_VESRA  = 0xe73a,
305    VRSa_VESRL  = 0xe738,
306    VRSb_VLVG   = 0xe722,
307    VRSc_VLGV   = 0xe721,
308
309    VRX_VL      = 0xe706,
310    VRX_VLLEZ   = 0xe704,
311    VRX_VLREP   = 0xe705,
312    VRX_VST     = 0xe70e,
313    VRX_VSTEF   = 0xe70b,
314    VRX_VSTEG   = 0xe70a,
315
316    NOP         = 0x0707,
317} S390Opcode;
318
319#ifdef CONFIG_DEBUG_TCG
320static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
321    "%r0",  "%r1",  "%r2",  "%r3",  "%r4",  "%r5",  "%r6",  "%r7",
322    "%r8",  "%r9",  "%r10", "%r11", "%r12", "%r13", "%r14", "%r15",
323    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
324    "%v0",  "%v1",  "%v2",  "%v3",  "%v4",  "%v5",  "%v6",  "%v7",
325    "%v8",  "%v9",  "%v10", "%v11", "%v12", "%v13", "%v14", "%v15",
326    "%v16", "%v17", "%v18", "%v19", "%v20", "%v21", "%v22", "%v23",
327    "%v24", "%v25", "%v26", "%v27", "%v28", "%v29", "%v30", "%v31",
328};
329#endif
330
331/* Since R6 is a potential argument register, choose it last of the
332   call-saved registers.  Likewise prefer the call-clobbered registers
333   in reverse order to maximize the chance of avoiding the arguments.  */
334static const int tcg_target_reg_alloc_order[] = {
335    /* Call saved registers.  */
336    TCG_REG_R13,
337    TCG_REG_R12,
338    TCG_REG_R11,
339    TCG_REG_R10,
340    TCG_REG_R9,
341    TCG_REG_R8,
342    TCG_REG_R7,
343    TCG_REG_R6,
344    /* Call clobbered registers.  */
345    TCG_REG_R14,
346    TCG_REG_R0,
347    TCG_REG_R1,
348    /* Argument registers, in reverse order of allocation.  */
349    TCG_REG_R5,
350    TCG_REG_R4,
351    TCG_REG_R3,
352    TCG_REG_R2,
353
354    /* V8-V15 are call saved, and omitted. */
355    TCG_REG_V0,
356    TCG_REG_V1,
357    TCG_REG_V2,
358    TCG_REG_V3,
359    TCG_REG_V4,
360    TCG_REG_V5,
361    TCG_REG_V6,
362    TCG_REG_V7,
363    TCG_REG_V16,
364    TCG_REG_V17,
365    TCG_REG_V18,
366    TCG_REG_V19,
367    TCG_REG_V20,
368    TCG_REG_V21,
369    TCG_REG_V22,
370    TCG_REG_V23,
371    TCG_REG_V24,
372    TCG_REG_V25,
373    TCG_REG_V26,
374    TCG_REG_V27,
375    TCG_REG_V28,
376    TCG_REG_V29,
377    TCG_REG_V30,
378    TCG_REG_V31,
379};
380
381static const int tcg_target_call_iarg_regs[] = {
382    TCG_REG_R2,
383    TCG_REG_R3,
384    TCG_REG_R4,
385    TCG_REG_R5,
386    TCG_REG_R6,
387};
388
389static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
390{
391    tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
392    tcg_debug_assert(slot == 0);
393    return TCG_REG_R2;
394}
395
396#define S390_CC_EQ      8
397#define S390_CC_LT      4
398#define S390_CC_GT      2
399#define S390_CC_OV      1
400#define S390_CC_NE      (S390_CC_LT | S390_CC_GT)
401#define S390_CC_LE      (S390_CC_LT | S390_CC_EQ)
402#define S390_CC_GE      (S390_CC_GT | S390_CC_EQ)
403#define S390_CC_NEVER   0
404#define S390_CC_ALWAYS  15
405
406/* Condition codes that result from a COMPARE and COMPARE LOGICAL.  */
407static const uint8_t tcg_cond_to_s390_cond[] = {
408    [TCG_COND_EQ]  = S390_CC_EQ,
409    [TCG_COND_NE]  = S390_CC_NE,
410    [TCG_COND_LT]  = S390_CC_LT,
411    [TCG_COND_LE]  = S390_CC_LE,
412    [TCG_COND_GT]  = S390_CC_GT,
413    [TCG_COND_GE]  = S390_CC_GE,
414    [TCG_COND_LTU] = S390_CC_LT,
415    [TCG_COND_LEU] = S390_CC_LE,
416    [TCG_COND_GTU] = S390_CC_GT,
417    [TCG_COND_GEU] = S390_CC_GE,
418};
419
420/* Condition codes that result from a LOAD AND TEST.  Here, we have no
421   unsigned instruction variation, however since the test is vs zero we
422   can re-map the outcomes appropriately.  */
423static const uint8_t tcg_cond_to_ltr_cond[] = {
424    [TCG_COND_EQ]  = S390_CC_EQ,
425    [TCG_COND_NE]  = S390_CC_NE,
426    [TCG_COND_LT]  = S390_CC_LT,
427    [TCG_COND_LE]  = S390_CC_LE,
428    [TCG_COND_GT]  = S390_CC_GT,
429    [TCG_COND_GE]  = S390_CC_GE,
430    [TCG_COND_LTU] = S390_CC_NEVER,
431    [TCG_COND_LEU] = S390_CC_EQ,
432    [TCG_COND_GTU] = S390_CC_NE,
433    [TCG_COND_GEU] = S390_CC_ALWAYS,
434};
435
436static const tcg_insn_unit *tb_ret_addr;
437uint64_t s390_facilities[3];
438
439static inline bool is_general_reg(TCGReg r)
440{
441    return r <= TCG_REG_R15;
442}
443
444static inline bool is_vector_reg(TCGReg r)
445{
446    return r >= TCG_REG_V0 && r <= TCG_REG_V31;
447}
448
449static bool patch_reloc(tcg_insn_unit *src_rw, int type,
450                        intptr_t value, intptr_t addend)
451{
452    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
453    intptr_t pcrel2;
454    uint32_t old;
455
456    value += addend;
457    pcrel2 = (tcg_insn_unit *)value - src_rx;
458
459    switch (type) {
460    case R_390_PC16DBL:
461        if (pcrel2 == (int16_t)pcrel2) {
462            tcg_patch16(src_rw, pcrel2);
463            return true;
464        }
465        break;
466    case R_390_PC32DBL:
467        if (pcrel2 == (int32_t)pcrel2) {
468            tcg_patch32(src_rw, pcrel2);
469            return true;
470        }
471        break;
472    case R_390_20:
473        if (value == sextract64(value, 0, 20)) {
474            old = *(uint32_t *)src_rw & 0xf00000ff;
475            old |= ((value & 0xfff) << 16) | ((value & 0xff000) >> 4);
476            tcg_patch32(src_rw, old);
477            return true;
478        }
479        break;
480    default:
481        g_assert_not_reached();
482    }
483    return false;
484}
485
486static int is_const_p16(uint64_t val)
487{
488    for (int i = 0; i < 4; ++i) {
489        uint64_t mask = 0xffffull << (i * 16);
490        if ((val & ~mask) == 0) {
491            return i;
492        }
493    }
494    return -1;
495}
496
497static int is_const_p32(uint64_t val)
498{
499    if ((val & 0xffffffff00000000ull) == 0) {
500        return 0;
501    }
502    if ((val & 0x00000000ffffffffull) == 0) {
503        return 1;
504    }
505    return -1;
506}
507
508/*
509 * Accept bit patterns like these:
510 *  0....01....1
511 *  1....10....0
512 *  1..10..01..1
513 *  0..01..10..0
514 * Copied from gcc sources.
515 */
516static bool risbg_mask(uint64_t c)
517{
518    uint64_t lsb;
519    /* We don't change the number of transitions by inverting,
520       so make sure we start with the LSB zero.  */
521    if (c & 1) {
522        c = ~c;
523    }
524    /* Reject all zeros or all ones.  */
525    if (c == 0) {
526        return false;
527    }
528    /* Find the first transition.  */
529    lsb = c & -c;
530    /* Invert to look for a second transition.  */
531    c = ~c;
532    /* Erase the first transition.  */
533    c &= -lsb;
534    /* Find the second transition, if any.  */
535    lsb = c & -c;
536    /* Match if all the bits are 1's, or if c is zero.  */
537    return c == -lsb;
538}
539
540/* Test if a constant matches the constraint. */
541static bool tcg_target_const_match(int64_t val, TCGType type, int ct, int vece)
542{
543    if (ct & TCG_CT_CONST) {
544        return 1;
545    }
546
547    if (type == TCG_TYPE_I32) {
548        val = (int32_t)val;
549    }
550
551    /* The following are mutually exclusive.  */
552    if (ct & TCG_CT_CONST_S16) {
553        return val == (int16_t)val;
554    } else if (ct & TCG_CT_CONST_S32) {
555        return val == (int32_t)val;
556    } else if (ct & TCG_CT_CONST_S33) {
557        return val >= -0xffffffffll && val <= 0xffffffffll;
558    } else if (ct & TCG_CT_CONST_ZERO) {
559        return val == 0;
560    }
561
562    if (ct & TCG_CT_CONST_INV) {
563        val = ~val;
564    }
565    /*
566     * Note that is_const_p16 is a subset of is_const_p32,
567     * so we don't need both constraints.
568     */
569    if ((ct & TCG_CT_CONST_P32) && is_const_p32(val) >= 0) {
570        return true;
571    }
572    if ((ct & TCG_CT_CONST_INVRISBG) && risbg_mask(~val)) {
573        return true;
574    }
575
576    return 0;
577}
578
579/* Emit instructions according to the given instruction format.  */
580
581static void tcg_out_insn_RR(TCGContext *s, S390Opcode op, TCGReg r1, TCGReg r2)
582{
583    tcg_out16(s, (op << 8) | (r1 << 4) | r2);
584}
585
586static void tcg_out_insn_RRE(TCGContext *s, S390Opcode op,
587                             TCGReg r1, TCGReg r2)
588{
589    tcg_out32(s, (op << 16) | (r1 << 4) | r2);
590}
591
592/* RRF-a without the m4 field */
593static void tcg_out_insn_RRFa(TCGContext *s, S390Opcode op,
594                              TCGReg r1, TCGReg r2, TCGReg r3)
595{
596    tcg_out32(s, (op << 16) | (r3 << 12) | (r1 << 4) | r2);
597}
598
599/* RRF-a with the m4 field */
600static void tcg_out_insn_RRFam(TCGContext *s, S390Opcode op,
601                               TCGReg r1, TCGReg r2, TCGReg r3, int m4)
602{
603    tcg_out32(s, (op << 16) | (r3 << 12) | (m4 << 8) | (r1 << 4) | r2);
604}
605
606static void tcg_out_insn_RRFc(TCGContext *s, S390Opcode op,
607                              TCGReg r1, TCGReg r2, int m3)
608{
609    tcg_out32(s, (op << 16) | (m3 << 12) | (r1 << 4) | r2);
610}
611
612static void tcg_out_insn_RI(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
613{
614    tcg_out32(s, (op << 16) | (r1 << 20) | (i2 & 0xffff));
615}
616
617static void tcg_out_insn_RIEg(TCGContext *s, S390Opcode op, TCGReg r1,
618                             int i2, int m3)
619{
620    tcg_out16(s, (op & 0xff00) | (r1 << 4) | m3);
621    tcg_out32(s, (i2 << 16) | (op & 0xff));
622}
623
624static void tcg_out_insn_RIL(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
625{
626    tcg_out16(s, op | (r1 << 4));
627    tcg_out32(s, i2);
628}
629
630static void tcg_out_insn_RS(TCGContext *s, S390Opcode op, TCGReg r1,
631                            TCGReg b2, TCGReg r3, int disp)
632{
633    tcg_out32(s, (op << 24) | (r1 << 20) | (r3 << 16) | (b2 << 12)
634              | (disp & 0xfff));
635}
636
637static void tcg_out_insn_RSY(TCGContext *s, S390Opcode op, TCGReg r1,
638                             TCGReg b2, TCGReg r3, int disp)
639{
640    tcg_out16(s, (op & 0xff00) | (r1 << 4) | r3);
641    tcg_out32(s, (op & 0xff) | (b2 << 28)
642              | ((disp & 0xfff) << 16) | ((disp & 0xff000) >> 4));
643}
644
645#define tcg_out_insn_RX   tcg_out_insn_RS
646#define tcg_out_insn_RXY  tcg_out_insn_RSY
647
648static int RXB(TCGReg v1, TCGReg v2, TCGReg v3, TCGReg v4)
649{
650    /*
651     * Shift bit 4 of each regno to its corresponding bit of RXB.
652     * RXB itself begins at bit 8 of the instruction so 8 - 4 = 4
653     * is the left-shift of the 4th operand.
654     */
655    return ((v1 & 0x10) << (4 + 3))
656         | ((v2 & 0x10) << (4 + 2))
657         | ((v3 & 0x10) << (4 + 1))
658         | ((v4 & 0x10) << (4 + 0));
659}
660
661static void tcg_out_insn_VRIa(TCGContext *s, S390Opcode op,
662                              TCGReg v1, uint16_t i2, int m3)
663{
664    tcg_debug_assert(is_vector_reg(v1));
665    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4));
666    tcg_out16(s, i2);
667    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m3 << 12));
668}
669
670static void tcg_out_insn_VRIb(TCGContext *s, S390Opcode op,
671                              TCGReg v1, uint8_t i2, uint8_t i3, int m4)
672{
673    tcg_debug_assert(is_vector_reg(v1));
674    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4));
675    tcg_out16(s, (i2 << 8) | (i3 & 0xff));
676    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m4 << 12));
677}
678
679static void tcg_out_insn_VRIc(TCGContext *s, S390Opcode op,
680                              TCGReg v1, uint16_t i2, TCGReg v3, int m4)
681{
682    tcg_debug_assert(is_vector_reg(v1));
683    tcg_debug_assert(is_vector_reg(v3));
684    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v3 & 0xf));
685    tcg_out16(s, i2);
686    tcg_out16(s, (op & 0x00ff) | RXB(v1, v3, 0, 0) | (m4 << 12));
687}
688
689static void tcg_out_insn_VRRa(TCGContext *s, S390Opcode op,
690                              TCGReg v1, TCGReg v2, int m3)
691{
692    tcg_debug_assert(is_vector_reg(v1));
693    tcg_debug_assert(is_vector_reg(v2));
694    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
695    tcg_out32(s, (op & 0x00ff) | RXB(v1, v2, 0, 0) | (m3 << 12));
696}
697
698static void tcg_out_insn_VRRc(TCGContext *s, S390Opcode op,
699                              TCGReg v1, TCGReg v2, TCGReg v3, int m4)
700{
701    tcg_debug_assert(is_vector_reg(v1));
702    tcg_debug_assert(is_vector_reg(v2));
703    tcg_debug_assert(is_vector_reg(v3));
704    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
705    tcg_out16(s, v3 << 12);
706    tcg_out16(s, (op & 0x00ff) | RXB(v1, v2, v3, 0) | (m4 << 12));
707}
708
709static void tcg_out_insn_VRRe(TCGContext *s, S390Opcode op,
710                              TCGReg v1, TCGReg v2, TCGReg v3, TCGReg v4)
711{
712    tcg_debug_assert(is_vector_reg(v1));
713    tcg_debug_assert(is_vector_reg(v2));
714    tcg_debug_assert(is_vector_reg(v3));
715    tcg_debug_assert(is_vector_reg(v4));
716    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
717    tcg_out16(s, v3 << 12);
718    tcg_out16(s, (op & 0x00ff) | RXB(v1, v2, v3, v4) | (v4 << 12));
719}
720
721static void tcg_out_insn_VRRf(TCGContext *s, S390Opcode op,
722                              TCGReg v1, TCGReg r2, TCGReg r3)
723{
724    tcg_debug_assert(is_vector_reg(v1));
725    tcg_debug_assert(is_general_reg(r2));
726    tcg_debug_assert(is_general_reg(r3));
727    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | r2);
728    tcg_out16(s, r3 << 12);
729    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0));
730}
731
732static void tcg_out_insn_VRSa(TCGContext *s, S390Opcode op, TCGReg v1,
733                              intptr_t d2, TCGReg b2, TCGReg v3, int m4)
734{
735    tcg_debug_assert(is_vector_reg(v1));
736    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
737    tcg_debug_assert(is_general_reg(b2));
738    tcg_debug_assert(is_vector_reg(v3));
739    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v3 & 0xf));
740    tcg_out16(s, b2 << 12 | d2);
741    tcg_out16(s, (op & 0x00ff) | RXB(v1, v3, 0, 0) | (m4 << 12));
742}
743
744static void tcg_out_insn_VRSb(TCGContext *s, S390Opcode op, TCGReg v1,
745                              intptr_t d2, TCGReg b2, TCGReg r3, int m4)
746{
747    tcg_debug_assert(is_vector_reg(v1));
748    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
749    tcg_debug_assert(is_general_reg(b2));
750    tcg_debug_assert(is_general_reg(r3));
751    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | r3);
752    tcg_out16(s, b2 << 12 | d2);
753    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m4 << 12));
754}
755
756static void tcg_out_insn_VRSc(TCGContext *s, S390Opcode op, TCGReg r1,
757                              intptr_t d2, TCGReg b2, TCGReg v3, int m4)
758{
759    tcg_debug_assert(is_general_reg(r1));
760    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
761    tcg_debug_assert(is_general_reg(b2));
762    tcg_debug_assert(is_vector_reg(v3));
763    tcg_out16(s, (op & 0xff00) | (r1 << 4) | (v3 & 0xf));
764    tcg_out16(s, b2 << 12 | d2);
765    tcg_out16(s, (op & 0x00ff) | RXB(0, v3, 0, 0) | (m4 << 12));
766}
767
768static void tcg_out_insn_VRX(TCGContext *s, S390Opcode op, TCGReg v1,
769                             TCGReg b2, TCGReg x2, intptr_t d2, int m3)
770{
771    tcg_debug_assert(is_vector_reg(v1));
772    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
773    tcg_debug_assert(is_general_reg(x2));
774    tcg_debug_assert(is_general_reg(b2));
775    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | x2);
776    tcg_out16(s, (b2 << 12) | d2);
777    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m3 << 12));
778}
779
780/* Emit an opcode with "type-checking" of the format.  */
781#define tcg_out_insn(S, FMT, OP, ...) \
782    glue(tcg_out_insn_,FMT)(S, glue(glue(FMT,_),OP), ## __VA_ARGS__)
783
784
785/* emit 64-bit shifts */
786static void tcg_out_sh64(TCGContext* s, S390Opcode op, TCGReg dest,
787                         TCGReg src, TCGReg sh_reg, int sh_imm)
788{
789    tcg_out_insn_RSY(s, op, dest, sh_reg, src, sh_imm);
790}
791
792/* emit 32-bit shifts */
793static void tcg_out_sh32(TCGContext* s, S390Opcode op, TCGReg dest,
794                         TCGReg sh_reg, int sh_imm)
795{
796    tcg_out_insn_RS(s, op, dest, sh_reg, 0, sh_imm);
797}
798
799static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg dst, TCGReg src)
800{
801    if (src == dst) {
802        return true;
803    }
804    switch (type) {
805    case TCG_TYPE_I32:
806        if (likely(is_general_reg(dst) && is_general_reg(src))) {
807            tcg_out_insn(s, RR, LR, dst, src);
808            break;
809        }
810        /* fallthru */
811
812    case TCG_TYPE_I64:
813        if (likely(is_general_reg(dst))) {
814            if (likely(is_general_reg(src))) {
815                tcg_out_insn(s, RRE, LGR, dst, src);
816            } else {
817                tcg_out_insn(s, VRSc, VLGV, dst, 0, 0, src, 3);
818            }
819            break;
820        } else if (is_general_reg(src)) {
821            tcg_out_insn(s, VRSb, VLVG, dst, 0, 0, src, 3);
822            break;
823        }
824        /* fallthru */
825
826    case TCG_TYPE_V64:
827    case TCG_TYPE_V128:
828        tcg_out_insn(s, VRRa, VLR, dst, src, 0);
829        break;
830
831    default:
832        g_assert_not_reached();
833    }
834    return true;
835}
836
837static const S390Opcode li_insns[4] = {
838    RI_LLILL, RI_LLILH, RI_LLIHL, RI_LLIHH
839};
840static const S390Opcode oi_insns[4] = {
841    RI_OILL, RI_OILH, RI_OIHL, RI_OIHH
842};
843static const S390Opcode lif_insns[2] = {
844    RIL_LLILF, RIL_LLIHF,
845};
846
847/* load a register with an immediate value */
848static void tcg_out_movi(TCGContext *s, TCGType type,
849                         TCGReg ret, tcg_target_long sval)
850{
851    tcg_target_ulong uval = sval;
852    ptrdiff_t pc_off;
853    int i;
854
855    if (type == TCG_TYPE_I32) {
856        uval = (uint32_t)sval;
857        sval = (int32_t)sval;
858    }
859
860    /* Try all 32-bit insns that can load it in one go.  */
861    if (sval >= -0x8000 && sval < 0x8000) {
862        tcg_out_insn(s, RI, LGHI, ret, sval);
863        return;
864    }
865
866    i = is_const_p16(uval);
867    if (i >= 0) {
868        tcg_out_insn_RI(s, li_insns[i], ret, uval >> (i * 16));
869        return;
870    }
871
872    /* Try all 48-bit insns that can load it in one go.  */
873    if (sval == (int32_t)sval) {
874        tcg_out_insn(s, RIL, LGFI, ret, sval);
875        return;
876    }
877
878    i = is_const_p32(uval);
879    if (i >= 0) {
880        tcg_out_insn_RIL(s, lif_insns[i], ret, uval >> (i * 32));
881        return;
882    }
883
884    /* Try for PC-relative address load.  For odd addresses, add one. */
885    pc_off = tcg_pcrel_diff(s, (void *)sval) >> 1;
886    if (pc_off == (int32_t)pc_off) {
887        tcg_out_insn(s, RIL, LARL, ret, pc_off);
888        if (sval & 1) {
889            tcg_out_insn(s, RI, AGHI, ret, 1);
890        }
891        return;
892    }
893
894    /* Otherwise, load it by parts. */
895    i = is_const_p16((uint32_t)uval);
896    if (i >= 0) {
897        tcg_out_insn_RI(s, li_insns[i], ret, uval >> (i * 16));
898    } else {
899        tcg_out_insn(s, RIL, LLILF, ret, uval);
900    }
901    uval >>= 32;
902    i = is_const_p16(uval);
903    if (i >= 0) {
904        tcg_out_insn_RI(s, oi_insns[i + 2], ret, uval >> (i * 16));
905    } else {
906        tcg_out_insn(s, RIL, OIHF, ret, uval);
907    }
908}
909
910/* Emit a load/store type instruction.  Inputs are:
911   DATA:     The register to be loaded or stored.
912   BASE+OFS: The effective address.
913   OPC_RX:   If the operation has an RX format opcode (e.g. STC), otherwise 0.
914   OPC_RXY:  The RXY format opcode for the operation (e.g. STCY).  */
915
916static void tcg_out_mem(TCGContext *s, S390Opcode opc_rx, S390Opcode opc_rxy,
917                        TCGReg data, TCGReg base, TCGReg index,
918                        tcg_target_long ofs)
919{
920    if (ofs < -0x80000 || ofs >= 0x80000) {
921        /* Combine the low 20 bits of the offset with the actual load insn;
922           the high 44 bits must come from an immediate load.  */
923        tcg_target_long low = ((ofs & 0xfffff) ^ 0x80000) - 0x80000;
924        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs - low);
925        ofs = low;
926
927        /* If we were already given an index register, add it in.  */
928        if (index != TCG_REG_NONE) {
929            tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
930        }
931        index = TCG_TMP0;
932    }
933
934    if (opc_rx && ofs >= 0 && ofs < 0x1000) {
935        tcg_out_insn_RX(s, opc_rx, data, base, index, ofs);
936    } else {
937        tcg_out_insn_RXY(s, opc_rxy, data, base, index, ofs);
938    }
939}
940
941static void tcg_out_vrx_mem(TCGContext *s, S390Opcode opc_vrx,
942                            TCGReg data, TCGReg base, TCGReg index,
943                            tcg_target_long ofs, int m3)
944{
945    if (ofs < 0 || ofs >= 0x1000) {
946        if (ofs >= -0x80000 && ofs < 0x80000) {
947            tcg_out_insn(s, RXY, LAY, TCG_TMP0, base, index, ofs);
948            base = TCG_TMP0;
949            index = TCG_REG_NONE;
950            ofs = 0;
951        } else {
952            tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs);
953            if (index != TCG_REG_NONE) {
954                tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
955            }
956            index = TCG_TMP0;
957            ofs = 0;
958        }
959    }
960    tcg_out_insn_VRX(s, opc_vrx, data, base, index, ofs, m3);
961}
962
963/* load data without address translation or endianness conversion */
964static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg data,
965                       TCGReg base, intptr_t ofs)
966{
967    switch (type) {
968    case TCG_TYPE_I32:
969        if (likely(is_general_reg(data))) {
970            tcg_out_mem(s, RX_L, RXY_LY, data, base, TCG_REG_NONE, ofs);
971            break;
972        }
973        tcg_out_vrx_mem(s, VRX_VLLEZ, data, base, TCG_REG_NONE, ofs, MO_32);
974        break;
975
976    case TCG_TYPE_I64:
977        if (likely(is_general_reg(data))) {
978            tcg_out_mem(s, 0, RXY_LG, data, base, TCG_REG_NONE, ofs);
979            break;
980        }
981        /* fallthru */
982
983    case TCG_TYPE_V64:
984        tcg_out_vrx_mem(s, VRX_VLLEZ, data, base, TCG_REG_NONE, ofs, MO_64);
985        break;
986
987    case TCG_TYPE_V128:
988        /* Hint quadword aligned.  */
989        tcg_out_vrx_mem(s, VRX_VL, data, base, TCG_REG_NONE, ofs, 4);
990        break;
991
992    default:
993        g_assert_not_reached();
994    }
995}
996
997static void tcg_out_st(TCGContext *s, TCGType type, TCGReg data,
998                       TCGReg base, intptr_t ofs)
999{
1000    switch (type) {
1001    case TCG_TYPE_I32:
1002        if (likely(is_general_reg(data))) {
1003            tcg_out_mem(s, RX_ST, RXY_STY, data, base, TCG_REG_NONE, ofs);
1004        } else {
1005            tcg_out_vrx_mem(s, VRX_VSTEF, data, base, TCG_REG_NONE, ofs, 1);
1006        }
1007        break;
1008
1009    case TCG_TYPE_I64:
1010        if (likely(is_general_reg(data))) {
1011            tcg_out_mem(s, 0, RXY_STG, data, base, TCG_REG_NONE, ofs);
1012            break;
1013        }
1014        /* fallthru */
1015
1016    case TCG_TYPE_V64:
1017        tcg_out_vrx_mem(s, VRX_VSTEG, data, base, TCG_REG_NONE, ofs, 0);
1018        break;
1019
1020    case TCG_TYPE_V128:
1021        /* Hint quadword aligned.  */
1022        tcg_out_vrx_mem(s, VRX_VST, data, base, TCG_REG_NONE, ofs, 4);
1023        break;
1024
1025    default:
1026        g_assert_not_reached();
1027    }
1028}
1029
1030static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
1031                               TCGReg base, intptr_t ofs)
1032{
1033    return false;
1034}
1035
1036static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
1037{
1038    return false;
1039}
1040
1041static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
1042                             tcg_target_long imm)
1043{
1044    /* This function is only used for passing structs by reference. */
1045    tcg_out_mem(s, RX_LA, RXY_LAY, rd, rs, TCG_REG_NONE, imm);
1046}
1047
1048static inline void tcg_out_risbg(TCGContext *s, TCGReg dest, TCGReg src,
1049                                 int msb, int lsb, int ofs, int z)
1050{
1051    /* Format RIE-f */
1052    tcg_out16(s, (RIEf_RISBG & 0xff00) | (dest << 4) | src);
1053    tcg_out16(s, (msb << 8) | (z << 7) | lsb);
1054    tcg_out16(s, (ofs << 8) | (RIEf_RISBG & 0xff));
1055}
1056
1057static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1058{
1059    tcg_out_insn(s, RRE, LGBR, dest, src);
1060}
1061
1062static void tcg_out_ext8u(TCGContext *s, TCGReg dest, TCGReg src)
1063{
1064    tcg_out_insn(s, RRE, LLGCR, dest, src);
1065}
1066
1067static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1068{
1069    tcg_out_insn(s, RRE, LGHR, dest, src);
1070}
1071
1072static void tcg_out_ext16u(TCGContext *s, TCGReg dest, TCGReg src)
1073{
1074    tcg_out_insn(s, RRE, LLGHR, dest, src);
1075}
1076
1077static void tcg_out_ext32s(TCGContext *s, TCGReg dest, TCGReg src)
1078{
1079    tcg_out_insn(s, RRE, LGFR, dest, src);
1080}
1081
1082static void tcg_out_ext32u(TCGContext *s, TCGReg dest, TCGReg src)
1083{
1084    tcg_out_insn(s, RRE, LLGFR, dest, src);
1085}
1086
1087static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg dest, TCGReg src)
1088{
1089    tcg_out_ext32s(s, dest, src);
1090}
1091
1092static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg dest, TCGReg src)
1093{
1094    tcg_out_ext32u(s, dest, src);
1095}
1096
1097static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg dest, TCGReg src)
1098{
1099    tcg_out_mov(s, TCG_TYPE_I32, dest, src);
1100}
1101
1102static void tgen_andi_risbg(TCGContext *s, TCGReg out, TCGReg in, uint64_t val)
1103{
1104    int msb, lsb;
1105    if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
1106        /* Achieve wraparound by swapping msb and lsb.  */
1107        msb = 64 - ctz64(~val);
1108        lsb = clz64(~val) - 1;
1109    } else {
1110        msb = clz64(val);
1111        lsb = 63 - ctz64(val);
1112    }
1113    tcg_out_risbg(s, out, in, msb, lsb, 0, 1);
1114}
1115
1116static void tgen_andi(TCGContext *s, TCGType type, TCGReg dest, uint64_t val)
1117{
1118    static const S390Opcode ni_insns[4] = {
1119        RI_NILL, RI_NILH, RI_NIHL, RI_NIHH
1120    };
1121    static const S390Opcode nif_insns[2] = {
1122        RIL_NILF, RIL_NIHF
1123    };
1124    uint64_t valid = (type == TCG_TYPE_I32 ? 0xffffffffull : -1ull);
1125    int i;
1126
1127    /* Look for the zero-extensions.  */
1128    if ((val & valid) == 0xffffffff) {
1129        tcg_out_ext32u(s, dest, dest);
1130        return;
1131    }
1132    if ((val & valid) == 0xff) {
1133        tcg_out_ext8u(s, dest, dest);
1134        return;
1135    }
1136    if ((val & valid) == 0xffff) {
1137        tcg_out_ext16u(s, dest, dest);
1138        return;
1139    }
1140
1141    i = is_const_p16(~val & valid);
1142    if (i >= 0) {
1143        tcg_out_insn_RI(s, ni_insns[i], dest, val >> (i * 16));
1144        return;
1145    }
1146
1147    i = is_const_p32(~val & valid);
1148    tcg_debug_assert(i == 0 || type != TCG_TYPE_I32);
1149    if (i >= 0) {
1150        tcg_out_insn_RIL(s, nif_insns[i], dest, val >> (i * 32));
1151        return;
1152    }
1153
1154    if (risbg_mask(val)) {
1155        tgen_andi_risbg(s, dest, dest, val);
1156        return;
1157    }
1158
1159    g_assert_not_reached();
1160}
1161
1162static void tgen_ori(TCGContext *s, TCGReg dest, uint64_t val)
1163{
1164    static const S390Opcode oif_insns[2] = {
1165        RIL_OILF, RIL_OIHF
1166    };
1167
1168    int i;
1169
1170    i = is_const_p16(val);
1171    if (i >= 0) {
1172        tcg_out_insn_RI(s, oi_insns[i], dest, val >> (i * 16));
1173        return;
1174    }
1175
1176    i = is_const_p32(val);
1177    if (i >= 0) {
1178        tcg_out_insn_RIL(s, oif_insns[i], dest, val >> (i * 32));
1179        return;
1180    }
1181
1182    g_assert_not_reached();
1183}
1184
1185static void tgen_xori(TCGContext *s, TCGReg dest, uint64_t val)
1186{
1187    switch (is_const_p32(val)) {
1188    case 0:
1189        tcg_out_insn(s, RIL, XILF, dest, val);
1190        break;
1191    case 1:
1192        tcg_out_insn(s, RIL, XIHF, dest, val >> 32);
1193        break;
1194    default:
1195        g_assert_not_reached();
1196    }
1197}
1198
1199static int tgen_cmp2(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1200                     TCGArg c2, bool c2const, bool need_carry, int *inv_cc)
1201{
1202    bool is_unsigned = is_unsigned_cond(c);
1203    TCGCond inv_c = tcg_invert_cond(c);
1204    S390Opcode op;
1205
1206    if (c2const) {
1207        if (c2 == 0) {
1208            if (!(is_unsigned && need_carry)) {
1209                if (type == TCG_TYPE_I32) {
1210                    tcg_out_insn(s, RR, LTR, r1, r1);
1211                } else {
1212                    tcg_out_insn(s, RRE, LTGR, r1, r1);
1213                }
1214                *inv_cc = tcg_cond_to_ltr_cond[inv_c];
1215                return tcg_cond_to_ltr_cond[c];
1216            }
1217        }
1218
1219        if (!is_unsigned && c2 == (int16_t)c2) {
1220            op = (type == TCG_TYPE_I32 ? RI_CHI : RI_CGHI);
1221            tcg_out_insn_RI(s, op, r1, c2);
1222            goto exit;
1223        }
1224
1225        if (type == TCG_TYPE_I32) {
1226            op = (is_unsigned ? RIL_CLFI : RIL_CFI);
1227            tcg_out_insn_RIL(s, op, r1, c2);
1228            goto exit;
1229        }
1230
1231        /*
1232         * Constraints are for a signed 33-bit operand, which is a
1233         * convenient superset of this signed/unsigned test.
1234         */
1235        if (c2 == (is_unsigned ? (TCGArg)(uint32_t)c2 : (TCGArg)(int32_t)c2)) {
1236            op = (is_unsigned ? RIL_CLGFI : RIL_CGFI);
1237            tcg_out_insn_RIL(s, op, r1, c2);
1238            goto exit;
1239        }
1240
1241        /* Load everything else into a register. */
1242        tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, c2);
1243        c2 = TCG_TMP0;
1244    }
1245
1246    if (type == TCG_TYPE_I32) {
1247        op = (is_unsigned ? RR_CLR : RR_CR);
1248        tcg_out_insn_RR(s, op, r1, c2);
1249    } else {
1250        op = (is_unsigned ? RRE_CLGR : RRE_CGR);
1251        tcg_out_insn_RRE(s, op, r1, c2);
1252    }
1253
1254 exit:
1255    *inv_cc = tcg_cond_to_s390_cond[inv_c];
1256    return tcg_cond_to_s390_cond[c];
1257}
1258
1259static int tgen_cmp(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1260                    TCGArg c2, bool c2const, bool need_carry)
1261{
1262    int inv_cc;
1263    return tgen_cmp2(s, type, c, r1, c2, c2const, need_carry, &inv_cc);
1264}
1265
1266static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond,
1267                         TCGReg dest, TCGReg c1, TCGArg c2,
1268                         bool c2const, bool neg)
1269{
1270    int cc;
1271
1272    /* With LOC2, we can always emit the minimum 3 insns.  */
1273    if (HAVE_FACILITY(LOAD_ON_COND2)) {
1274        /* Emit: d = 0, d = (cc ? 1 : d).  */
1275        cc = tgen_cmp(s, type, cond, c1, c2, c2const, false);
1276        tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1277        tcg_out_insn(s, RIEg, LOCGHI, dest, neg ? -1 : 1, cc);
1278        return;
1279    }
1280
1281    switch (cond) {
1282    case TCG_COND_GEU:
1283    case TCG_COND_LTU:
1284    case TCG_COND_LT:
1285    case TCG_COND_GE:
1286        /* Swap operands so that we can use LEU/GTU/GT/LE.  */
1287        if (!c2const) {
1288            TCGReg t = c1;
1289            c1 = c2;
1290            c2 = t;
1291            cond = tcg_swap_cond(cond);
1292        }
1293        break;
1294    default:
1295        break;
1296    }
1297
1298    switch (cond) {
1299    case TCG_COND_NE:
1300        /* X != 0 is X > 0.  */
1301        if (c2const && c2 == 0) {
1302            cond = TCG_COND_GTU;
1303        } else {
1304            break;
1305        }
1306        /* fallthru */
1307
1308    case TCG_COND_GTU:
1309    case TCG_COND_GT:
1310        /*
1311         * The result of a compare has CC=2 for GT and CC=3 unused.
1312         * ADD LOGICAL WITH CARRY considers (CC & 2) the carry bit.
1313         */
1314        tgen_cmp(s, type, cond, c1, c2, c2const, true);
1315        tcg_out_movi(s, type, dest, 0);
1316        tcg_out_insn(s, RRE, ALCGR, dest, dest);
1317        if (neg) {
1318            if (type == TCG_TYPE_I32) {
1319                tcg_out_insn(s, RR, LCR, dest, dest);
1320            } else {
1321                tcg_out_insn(s, RRE, LCGR, dest, dest);
1322            }
1323        }
1324        return;
1325
1326    case TCG_COND_EQ:
1327        /* X == 0 is X <= 0.  */
1328        if (c2const && c2 == 0) {
1329            cond = TCG_COND_LEU;
1330        } else {
1331            break;
1332        }
1333        /* fallthru */
1334
1335    case TCG_COND_LEU:
1336    case TCG_COND_LE:
1337        /*
1338         * As above, but we're looking for borrow, or !carry.
1339         * The second insn computes d - d - borrow, or -1 for true
1340         * and 0 for false.  So we must mask to 1 bit afterward.
1341         */
1342        tgen_cmp(s, type, cond, c1, c2, c2const, true);
1343        tcg_out_insn(s, RRE, SLBGR, dest, dest);
1344        if (!neg) {
1345            tgen_andi(s, type, dest, 1);
1346        }
1347        return;
1348
1349    default:
1350        g_assert_not_reached();
1351    }
1352
1353    cc = tgen_cmp(s, type, cond, c1, c2, c2const, false);
1354    /* Emit: d = 0, t = 1, d = (cc ? t : d).  */
1355    tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1356    tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, neg ? -1 : 1);
1357    tcg_out_insn(s, RRFc, LOCGR, dest, TCG_TMP0, cc);
1358}
1359
1360static void tgen_movcond_int(TCGContext *s, TCGType type, TCGReg dest,
1361                             TCGArg v3, int v3const, TCGReg v4,
1362                             int cc, int inv_cc)
1363{
1364    TCGReg src;
1365
1366    if (v3const) {
1367        if (dest == v4) {
1368            if (HAVE_FACILITY(LOAD_ON_COND2)) {
1369                /* Emit: if (cc) dest = v3. */
1370                tcg_out_insn(s, RIEg, LOCGHI, dest, v3, cc);
1371                return;
1372            }
1373            tcg_out_insn(s, RI, LGHI, TCG_TMP0, v3);
1374            src = TCG_TMP0;
1375        } else {
1376            /* LGR+LOCGHI is larger than LGHI+LOCGR. */
1377            tcg_out_insn(s, RI, LGHI, dest, v3);
1378            cc = inv_cc;
1379            src = v4;
1380        }
1381    } else {
1382        if (HAVE_FACILITY(MISC_INSN_EXT3)) {
1383            /* Emit: dest = cc ? v3 : v4. */
1384            tcg_out_insn(s, RRFam, SELGR, dest, v3, v4, cc);
1385            return;
1386        }
1387        if (dest == v4) {
1388            src = v3;
1389        } else {
1390            tcg_out_mov(s, type, dest, v3);
1391            cc = inv_cc;
1392            src = v4;
1393        }
1394    }
1395
1396    /* Emit: if (cc) dest = src. */
1397    tcg_out_insn(s, RRFc, LOCGR, dest, src, cc);
1398}
1399
1400static void tgen_movcond(TCGContext *s, TCGType type, TCGCond c, TCGReg dest,
1401                         TCGReg c1, TCGArg c2, int c2const,
1402                         TCGArg v3, int v3const, TCGReg v4)
1403{
1404    int cc, inv_cc;
1405
1406    cc = tgen_cmp2(s, type, c, c1, c2, c2const, false, &inv_cc);
1407    tgen_movcond_int(s, type, dest, v3, v3const, v4, cc, inv_cc);
1408}
1409
1410static void tgen_clz(TCGContext *s, TCGReg dest, TCGReg a1,
1411                     TCGArg a2, int a2const)
1412{
1413    /* Since this sets both R and R+1, we have no choice but to store the
1414       result into R0, allowing R1 == TCG_TMP0 to be clobbered as well.  */
1415    QEMU_BUILD_BUG_ON(TCG_TMP0 != TCG_REG_R1);
1416    tcg_out_insn(s, RRE, FLOGR, TCG_REG_R0, a1);
1417
1418    if (a2const && a2 == 64) {
1419        tcg_out_mov(s, TCG_TYPE_I64, dest, TCG_REG_R0);
1420        return;
1421    }
1422
1423    /*
1424     * Conditions from FLOGR are:
1425     *   2 -> one bit found
1426     *   8 -> no one bit found
1427     */
1428    tgen_movcond_int(s, TCG_TYPE_I64, dest, a2, a2const, TCG_REG_R0, 8, 2);
1429}
1430
1431static void tgen_ctpop(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1432{
1433    /* With MIE3, and bit 0 of m4 set, we get the complete result. */
1434    if (HAVE_FACILITY(MISC_INSN_EXT3)) {
1435        if (type == TCG_TYPE_I32) {
1436            tcg_out_ext32u(s, dest, src);
1437            src = dest;
1438        }
1439        tcg_out_insn(s, RRFc, POPCNT, dest, src, 8);
1440        return;
1441    }
1442
1443    /* Without MIE3, each byte gets the count of bits for the byte. */
1444    tcg_out_insn(s, RRFc, POPCNT, dest, src, 0);
1445
1446    /* Multiply to sum each byte at the top of the word. */
1447    if (type == TCG_TYPE_I32) {
1448        tcg_out_insn(s, RIL, MSFI, dest, 0x01010101);
1449        tcg_out_sh32(s, RS_SRL, dest, TCG_REG_NONE, 24);
1450    } else {
1451        tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, 0x0101010101010101ull);
1452        tcg_out_insn(s, RRE, MSGR, dest, TCG_TMP0);
1453        tcg_out_sh64(s, RSY_SRLG, dest, dest, TCG_REG_NONE, 56);
1454    }
1455}
1456
1457static void tgen_deposit(TCGContext *s, TCGReg dest, TCGReg src,
1458                         int ofs, int len, int z)
1459{
1460    int lsb = (63 - ofs);
1461    int msb = lsb - (len - 1);
1462    tcg_out_risbg(s, dest, src, msb, lsb, ofs, z);
1463}
1464
1465static void tgen_extract(TCGContext *s, TCGReg dest, TCGReg src,
1466                         int ofs, int len)
1467{
1468    tcg_out_risbg(s, dest, src, 64 - len, 63, 64 - ofs, 1);
1469}
1470
1471static void tgen_gotoi(TCGContext *s, int cc, const tcg_insn_unit *dest)
1472{
1473    ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1;
1474    if (off == (int16_t)off) {
1475        tcg_out_insn(s, RI, BRC, cc, off);
1476    } else if (off == (int32_t)off) {
1477        tcg_out_insn(s, RIL, BRCL, cc, off);
1478    } else {
1479        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1480        tcg_out_insn(s, RR, BCR, cc, TCG_TMP0);
1481    }
1482}
1483
1484static void tgen_branch(TCGContext *s, int cc, TCGLabel *l)
1485{
1486    if (l->has_value) {
1487        tgen_gotoi(s, cc, l->u.value_ptr);
1488    } else {
1489        tcg_out16(s, RI_BRC | (cc << 4));
1490        tcg_out_reloc(s, s->code_ptr, R_390_PC16DBL, l, 2);
1491        s->code_ptr += 1;
1492    }
1493}
1494
1495static void tgen_compare_branch(TCGContext *s, S390Opcode opc, int cc,
1496                                TCGReg r1, TCGReg r2, TCGLabel *l)
1497{
1498    tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1499    /* Format RIE-b */
1500    tcg_out16(s, (opc & 0xff00) | (r1 << 4) | r2);
1501    tcg_out16(s, 0);
1502    tcg_out16(s, cc << 12 | (opc & 0xff));
1503}
1504
1505static void tgen_compare_imm_branch(TCGContext *s, S390Opcode opc, int cc,
1506                                    TCGReg r1, int i2, TCGLabel *l)
1507{
1508    tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1509    /* Format RIE-c */
1510    tcg_out16(s, (opc & 0xff00) | (r1 << 4) | cc);
1511    tcg_out16(s, 0);
1512    tcg_out16(s, (i2 << 8) | (opc & 0xff));
1513}
1514
1515static void tgen_brcond(TCGContext *s, TCGType type, TCGCond c,
1516                        TCGReg r1, TCGArg c2, int c2const, TCGLabel *l)
1517{
1518    int cc;
1519    bool is_unsigned = is_unsigned_cond(c);
1520    bool in_range;
1521    S390Opcode opc;
1522
1523    cc = tcg_cond_to_s390_cond[c];
1524
1525    if (!c2const) {
1526        opc = (type == TCG_TYPE_I32
1527               ? (is_unsigned ? RIEb_CLRJ : RIEb_CRJ)
1528               : (is_unsigned ? RIEb_CLGRJ : RIEb_CGRJ));
1529        tgen_compare_branch(s, opc, cc, r1, c2, l);
1530        return;
1531    }
1532
1533    /*
1534     * COMPARE IMMEDIATE AND BRANCH RELATIVE has an 8-bit immediate field.
1535     * If the immediate we've been given does not fit that range, we'll
1536     * fall back to separate compare and branch instructions using the
1537     * larger comparison range afforded by COMPARE IMMEDIATE.
1538     */
1539    if (type == TCG_TYPE_I32) {
1540        if (is_unsigned) {
1541            opc = RIEc_CLIJ;
1542            in_range = (uint32_t)c2 == (uint8_t)c2;
1543        } else {
1544            opc = RIEc_CIJ;
1545            in_range = (int32_t)c2 == (int8_t)c2;
1546        }
1547    } else {
1548        if (is_unsigned) {
1549            opc = RIEc_CLGIJ;
1550            in_range = (uint64_t)c2 == (uint8_t)c2;
1551        } else {
1552            opc = RIEc_CGIJ;
1553            in_range = (int64_t)c2 == (int8_t)c2;
1554        }
1555    }
1556    if (in_range) {
1557        tgen_compare_imm_branch(s, opc, cc, r1, c2, l);
1558        return;
1559    }
1560
1561    cc = tgen_cmp(s, type, c, r1, c2, c2const, false);
1562    tgen_branch(s, cc, l);
1563}
1564
1565static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *dest)
1566{
1567    ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1;
1568    if (off == (int32_t)off) {
1569        tcg_out_insn(s, RIL, BRASL, TCG_REG_R14, off);
1570    } else {
1571        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1572        tcg_out_insn(s, RR, BASR, TCG_REG_R14, TCG_TMP0);
1573    }
1574}
1575
1576static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest,
1577                         const TCGHelperInfo *info)
1578{
1579    tcg_out_call_int(s, dest);
1580}
1581
1582typedef struct {
1583    TCGReg base;
1584    TCGReg index;
1585    int disp;
1586    TCGAtomAlign aa;
1587} HostAddress;
1588
1589bool tcg_target_has_memory_bswap(MemOp memop)
1590{
1591    TCGAtomAlign aa;
1592
1593    if ((memop & MO_SIZE) <= MO_64) {
1594        return true;
1595    }
1596
1597    /*
1598     * Reject 16-byte memop with 16-byte atomicity,
1599     * but do allow a pair of 64-bit operations.
1600     */
1601    aa = atom_and_align_for_opc(tcg_ctx, memop, MO_ATOM_IFALIGN, true);
1602    return aa.atom <= MO_64;
1603}
1604
1605static void tcg_out_qemu_ld_direct(TCGContext *s, MemOp opc, TCGReg data,
1606                                   HostAddress h)
1607{
1608    switch (opc & (MO_SSIZE | MO_BSWAP)) {
1609    case MO_UB:
1610        tcg_out_insn(s, RXY, LLGC, data, h.base, h.index, h.disp);
1611        break;
1612    case MO_SB:
1613        tcg_out_insn(s, RXY, LGB, data, h.base, h.index, h.disp);
1614        break;
1615
1616    case MO_UW | MO_BSWAP:
1617        /* swapped unsigned halfword load with upper bits zeroed */
1618        tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp);
1619        tcg_out_ext16u(s, data, data);
1620        break;
1621    case MO_UW:
1622        tcg_out_insn(s, RXY, LLGH, data, h.base, h.index, h.disp);
1623        break;
1624
1625    case MO_SW | MO_BSWAP:
1626        /* swapped sign-extended halfword load */
1627        tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp);
1628        tcg_out_ext16s(s, TCG_TYPE_REG, data, data);
1629        break;
1630    case MO_SW:
1631        tcg_out_insn(s, RXY, LGH, data, h.base, h.index, h.disp);
1632        break;
1633
1634    case MO_UL | MO_BSWAP:
1635        /* swapped unsigned int load with upper bits zeroed */
1636        tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp);
1637        tcg_out_ext32u(s, data, data);
1638        break;
1639    case MO_UL:
1640        tcg_out_insn(s, RXY, LLGF, data, h.base, h.index, h.disp);
1641        break;
1642
1643    case MO_SL | MO_BSWAP:
1644        /* swapped sign-extended int load */
1645        tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp);
1646        tcg_out_ext32s(s, data, data);
1647        break;
1648    case MO_SL:
1649        tcg_out_insn(s, RXY, LGF, data, h.base, h.index, h.disp);
1650        break;
1651
1652    case MO_UQ | MO_BSWAP:
1653        tcg_out_insn(s, RXY, LRVG, data, h.base, h.index, h.disp);
1654        break;
1655    case MO_UQ:
1656        tcg_out_insn(s, RXY, LG, data, h.base, h.index, h.disp);
1657        break;
1658
1659    default:
1660        g_assert_not_reached();
1661    }
1662}
1663
1664static void tcg_out_qemu_st_direct(TCGContext *s, MemOp opc, TCGReg data,
1665                                   HostAddress h)
1666{
1667    switch (opc & (MO_SIZE | MO_BSWAP)) {
1668    case MO_UB:
1669        if (h.disp >= 0 && h.disp < 0x1000) {
1670            tcg_out_insn(s, RX, STC, data, h.base, h.index, h.disp);
1671        } else {
1672            tcg_out_insn(s, RXY, STCY, data, h.base, h.index, h.disp);
1673        }
1674        break;
1675
1676    case MO_UW | MO_BSWAP:
1677        tcg_out_insn(s, RXY, STRVH, data, h.base, h.index, h.disp);
1678        break;
1679    case MO_UW:
1680        if (h.disp >= 0 && h.disp < 0x1000) {
1681            tcg_out_insn(s, RX, STH, data, h.base, h.index, h.disp);
1682        } else {
1683            tcg_out_insn(s, RXY, STHY, data, h.base, h.index, h.disp);
1684        }
1685        break;
1686
1687    case MO_UL | MO_BSWAP:
1688        tcg_out_insn(s, RXY, STRV, data, h.base, h.index, h.disp);
1689        break;
1690    case MO_UL:
1691        if (h.disp >= 0 && h.disp < 0x1000) {
1692            tcg_out_insn(s, RX, ST, data, h.base, h.index, h.disp);
1693        } else {
1694            tcg_out_insn(s, RXY, STY, data, h.base, h.index, h.disp);
1695        }
1696        break;
1697
1698    case MO_UQ | MO_BSWAP:
1699        tcg_out_insn(s, RXY, STRVG, data, h.base, h.index, h.disp);
1700        break;
1701    case MO_UQ:
1702        tcg_out_insn(s, RXY, STG, data, h.base, h.index, h.disp);
1703        break;
1704
1705    default:
1706        g_assert_not_reached();
1707    }
1708}
1709
1710static const TCGLdstHelperParam ldst_helper_param = {
1711    .ntmp = 1, .tmp = { TCG_TMP0 }
1712};
1713
1714static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1715{
1716    MemOp opc = get_memop(lb->oi);
1717
1718    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1719                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) {
1720        return false;
1721    }
1722
1723    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
1724    tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SIZE]);
1725    tcg_out_ld_helper_ret(s, lb, false, &ldst_helper_param);
1726
1727    tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1728    return true;
1729}
1730
1731static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1732{
1733    MemOp opc = get_memop(lb->oi);
1734
1735    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1736                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) {
1737        return false;
1738    }
1739
1740    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
1741    tcg_out_call_int(s, qemu_st_helpers[opc & MO_SIZE]);
1742
1743    tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1744    return true;
1745}
1746
1747/* We're expecting to use a 20-bit negative offset on the tlb memory ops.  */
1748#define MIN_TLB_MASK_TABLE_OFS  -(1 << 19)
1749
1750/*
1751 * For system-mode, perform the TLB load and compare.
1752 * For user-mode, perform any required alignment tests.
1753 * In both cases, return a TCGLabelQemuLdst structure if the slow path
1754 * is required and fill in @h with the host address for the fast path.
1755 */
1756static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
1757                                           TCGReg addr_reg, MemOpIdx oi,
1758                                           bool is_ld)
1759{
1760    TCGType addr_type = s->addr_type;
1761    TCGLabelQemuLdst *ldst = NULL;
1762    MemOp opc = get_memop(oi);
1763    MemOp s_bits = opc & MO_SIZE;
1764    unsigned a_mask;
1765
1766    h->aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, s_bits == MO_128);
1767    a_mask = (1 << h->aa.align) - 1;
1768
1769    if (tcg_use_softmmu) {
1770        unsigned s_mask = (1 << s_bits) - 1;
1771        int mem_index = get_mmuidx(oi);
1772        int fast_off = tlb_mask_table_ofs(s, mem_index);
1773        int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1774        int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1775        int ofs, a_off;
1776        uint64_t tlb_mask;
1777
1778        ldst = new_ldst_label(s);
1779        ldst->is_ld = is_ld;
1780        ldst->oi = oi;
1781        ldst->addrlo_reg = addr_reg;
1782
1783        tcg_out_sh64(s, RSY_SRLG, TCG_TMP0, addr_reg, TCG_REG_NONE,
1784                     s->page_bits - CPU_TLB_ENTRY_BITS);
1785
1786        tcg_out_insn(s, RXY, NG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, mask_off);
1787        tcg_out_insn(s, RXY, AG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, table_off);
1788
1789        /*
1790         * For aligned accesses, we check the first byte and include the
1791         * alignment bits within the address.  For unaligned access, we
1792         * check that we don't cross pages using the address of the last
1793         * byte of the access.
1794         */
1795        a_off = (a_mask >= s_mask ? 0 : s_mask - a_mask);
1796        tlb_mask = (uint64_t)s->page_mask | a_mask;
1797        if (a_off == 0) {
1798            tgen_andi_risbg(s, TCG_REG_R0, addr_reg, tlb_mask);
1799        } else {
1800            tcg_out_insn(s, RX, LA, TCG_REG_R0, addr_reg, TCG_REG_NONE, a_off);
1801            tgen_andi(s, addr_type, TCG_REG_R0, tlb_mask);
1802        }
1803
1804        if (is_ld) {
1805            ofs = offsetof(CPUTLBEntry, addr_read);
1806        } else {
1807            ofs = offsetof(CPUTLBEntry, addr_write);
1808        }
1809        if (addr_type == TCG_TYPE_I32) {
1810            ofs += HOST_BIG_ENDIAN * 4;
1811            tcg_out_insn(s, RX, C, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs);
1812        } else {
1813            tcg_out_insn(s, RXY, CG, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs);
1814        }
1815
1816        tcg_out16(s, RI_BRC | (S390_CC_NE << 4));
1817        ldst->label_ptr[0] = s->code_ptr++;
1818
1819        h->index = TCG_TMP0;
1820        tcg_out_insn(s, RXY, LG, h->index, TCG_TMP0, TCG_REG_NONE,
1821                     offsetof(CPUTLBEntry, addend));
1822
1823        if (addr_type == TCG_TYPE_I32) {
1824            tcg_out_insn(s, RRE, ALGFR, h->index, addr_reg);
1825            h->base = TCG_REG_NONE;
1826        } else {
1827            h->base = addr_reg;
1828        }
1829        h->disp = 0;
1830    } else {
1831        if (a_mask) {
1832            ldst = new_ldst_label(s);
1833            ldst->is_ld = is_ld;
1834            ldst->oi = oi;
1835            ldst->addrlo_reg = addr_reg;
1836
1837            /* We are expecting a_bits to max out at 7, much lower than TMLL. */
1838            tcg_debug_assert(a_mask <= 0xffff);
1839            tcg_out_insn(s, RI, TMLL, addr_reg, a_mask);
1840
1841            tcg_out16(s, RI_BRC | (7 << 4)); /* CC in {1,2,3} */
1842            ldst->label_ptr[0] = s->code_ptr++;
1843        }
1844
1845        h->base = addr_reg;
1846        if (addr_type == TCG_TYPE_I32) {
1847            tcg_out_ext32u(s, TCG_TMP0, addr_reg);
1848            h->base = TCG_TMP0;
1849        }
1850        if (guest_base < 0x80000) {
1851            h->index = TCG_REG_NONE;
1852            h->disp = guest_base;
1853        } else {
1854            h->index = TCG_GUEST_BASE_REG;
1855            h->disp = 0;
1856        }
1857    }
1858
1859    return ldst;
1860}
1861
1862static void tcg_out_qemu_ld(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
1863                            MemOpIdx oi, TCGType data_type)
1864{
1865    TCGLabelQemuLdst *ldst;
1866    HostAddress h;
1867
1868    ldst = prepare_host_addr(s, &h, addr_reg, oi, true);
1869    tcg_out_qemu_ld_direct(s, get_memop(oi), data_reg, h);
1870
1871    if (ldst) {
1872        ldst->type = data_type;
1873        ldst->datalo_reg = data_reg;
1874        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1875    }
1876}
1877
1878static void tcg_out_qemu_st(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
1879                            MemOpIdx oi, TCGType data_type)
1880{
1881    TCGLabelQemuLdst *ldst;
1882    HostAddress h;
1883
1884    ldst = prepare_host_addr(s, &h, addr_reg, oi, false);
1885    tcg_out_qemu_st_direct(s, get_memop(oi), data_reg, h);
1886
1887    if (ldst) {
1888        ldst->type = data_type;
1889        ldst->datalo_reg = data_reg;
1890        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1891    }
1892}
1893
1894static void tcg_out_qemu_ldst_i128(TCGContext *s, TCGReg datalo, TCGReg datahi,
1895                                   TCGReg addr_reg, MemOpIdx oi, bool is_ld)
1896{
1897    TCGLabel *l1 = NULL, *l2 = NULL;
1898    TCGLabelQemuLdst *ldst;
1899    HostAddress h;
1900    bool need_bswap;
1901    bool use_pair;
1902    S390Opcode insn;
1903
1904    ldst = prepare_host_addr(s, &h, addr_reg, oi, is_ld);
1905
1906    use_pair = h.aa.atom < MO_128;
1907    need_bswap = get_memop(oi) & MO_BSWAP;
1908
1909    if (!use_pair) {
1910        /*
1911         * Atomicity requires we use LPQ.  If we've already checked for
1912         * 16-byte alignment, that's all we need.  If we arrive with
1913         * lesser alignment, we have determined that less than 16-byte
1914         * alignment can be satisfied with two 8-byte loads.
1915         */
1916        if (h.aa.align < MO_128) {
1917            use_pair = true;
1918            l1 = gen_new_label();
1919            l2 = gen_new_label();
1920
1921            tcg_out_insn(s, RI, TMLL, addr_reg, 15);
1922            tgen_branch(s, 7, l1); /* CC in {1,2,3} */
1923        }
1924
1925        tcg_debug_assert(!need_bswap);
1926        tcg_debug_assert(datalo & 1);
1927        tcg_debug_assert(datahi == datalo - 1);
1928        insn = is_ld ? RXY_LPQ : RXY_STPQ;
1929        tcg_out_insn_RXY(s, insn, datahi, h.base, h.index, h.disp);
1930
1931        if (use_pair) {
1932            tgen_branch(s, S390_CC_ALWAYS, l2);
1933            tcg_out_label(s, l1);
1934        }
1935    }
1936    if (use_pair) {
1937        TCGReg d1, d2;
1938
1939        if (need_bswap) {
1940            d1 = datalo, d2 = datahi;
1941            insn = is_ld ? RXY_LRVG : RXY_STRVG;
1942        } else {
1943            d1 = datahi, d2 = datalo;
1944            insn = is_ld ? RXY_LG : RXY_STG;
1945        }
1946
1947        if (h.base == d1 || h.index == d1) {
1948            tcg_out_insn(s, RXY, LAY, TCG_TMP0, h.base, h.index, h.disp);
1949            h.base = TCG_TMP0;
1950            h.index = TCG_REG_NONE;
1951            h.disp = 0;
1952        }
1953        tcg_out_insn_RXY(s, insn, d1, h.base, h.index, h.disp);
1954        tcg_out_insn_RXY(s, insn, d2, h.base, h.index, h.disp + 8);
1955    }
1956    if (l2) {
1957        tcg_out_label(s, l2);
1958    }
1959
1960    if (ldst) {
1961        ldst->type = TCG_TYPE_I128;
1962        ldst->datalo_reg = datalo;
1963        ldst->datahi_reg = datahi;
1964        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1965    }
1966}
1967
1968static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
1969{
1970    /* Reuse the zeroing that exists for goto_ptr.  */
1971    if (a0 == 0) {
1972        tgen_gotoi(s, S390_CC_ALWAYS, tcg_code_gen_epilogue);
1973    } else {
1974        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, a0);
1975        tgen_gotoi(s, S390_CC_ALWAYS, tb_ret_addr);
1976    }
1977}
1978
1979static void tcg_out_goto_tb(TCGContext *s, int which)
1980{
1981    /*
1982     * Branch displacement must be aligned for atomic patching;
1983     * see if we need to add extra nop before branch
1984     */
1985    if (!QEMU_PTR_IS_ALIGNED(s->code_ptr + 1, 4)) {
1986        tcg_out16(s, NOP);
1987    }
1988    tcg_out16(s, RIL_BRCL | (S390_CC_ALWAYS << 4));
1989    set_jmp_insn_offset(s, which);
1990    s->code_ptr += 2;
1991    set_jmp_reset_offset(s, which);
1992}
1993
1994void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
1995                              uintptr_t jmp_rx, uintptr_t jmp_rw)
1996{
1997    if (!HAVE_FACILITY(GEN_INST_EXT)) {
1998        return;
1999    }
2000    /* patch the branch destination */
2001    uintptr_t addr = tb->jmp_target_addr[n];
2002    intptr_t disp = addr - (jmp_rx - 2);
2003    qatomic_set((int32_t *)jmp_rw, disp / 2);
2004    /* no need to flush icache explicitly */
2005}
2006
2007# define OP_32_64(x) \
2008        case glue(glue(INDEX_op_,x),_i32): \
2009        case glue(glue(INDEX_op_,x),_i64)
2010
2011static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
2012                              const TCGArg args[TCG_MAX_OP_ARGS],
2013                              const int const_args[TCG_MAX_OP_ARGS])
2014{
2015    S390Opcode op, op2;
2016    TCGArg a0, a1, a2;
2017
2018    switch (opc) {
2019    case INDEX_op_goto_ptr:
2020        a0 = args[0];
2021        tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, a0);
2022        break;
2023
2024    OP_32_64(ld8u):
2025        /* ??? LLC (RXY format) is only present with the extended-immediate
2026           facility, whereas LLGC is always present.  */
2027        tcg_out_mem(s, 0, RXY_LLGC, args[0], args[1], TCG_REG_NONE, args[2]);
2028        break;
2029
2030    OP_32_64(ld8s):
2031        /* ??? LB is no smaller than LGB, so no point to using it.  */
2032        tcg_out_mem(s, 0, RXY_LGB, args[0], args[1], TCG_REG_NONE, args[2]);
2033        break;
2034
2035    OP_32_64(ld16u):
2036        /* ??? LLH (RXY format) is only present with the extended-immediate
2037           facility, whereas LLGH is always present.  */
2038        tcg_out_mem(s, 0, RXY_LLGH, args[0], args[1], TCG_REG_NONE, args[2]);
2039        break;
2040
2041    case INDEX_op_ld16s_i32:
2042        tcg_out_mem(s, RX_LH, RXY_LHY, args[0], args[1], TCG_REG_NONE, args[2]);
2043        break;
2044
2045    case INDEX_op_ld_i32:
2046        tcg_out_ld(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2047        break;
2048
2049    OP_32_64(st8):
2050        tcg_out_mem(s, RX_STC, RXY_STCY, args[0], args[1],
2051                    TCG_REG_NONE, args[2]);
2052        break;
2053
2054    OP_32_64(st16):
2055        tcg_out_mem(s, RX_STH, RXY_STHY, args[0], args[1],
2056                    TCG_REG_NONE, args[2]);
2057        break;
2058
2059    case INDEX_op_st_i32:
2060        tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2061        break;
2062
2063    case INDEX_op_add_i32:
2064        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2065        if (const_args[2]) {
2066        do_addi_32:
2067            if (a0 == a1) {
2068                if (a2 == (int16_t)a2) {
2069                    tcg_out_insn(s, RI, AHI, a0, a2);
2070                    break;
2071                }
2072                tcg_out_insn(s, RIL, AFI, a0, a2);
2073                break;
2074            }
2075            tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
2076        } else if (a0 == a1) {
2077            tcg_out_insn(s, RR, AR, a0, a2);
2078        } else {
2079            tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
2080        }
2081        break;
2082    case INDEX_op_sub_i32:
2083        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2084        if (const_args[2]) {
2085            a2 = -a2;
2086            goto do_addi_32;
2087        } else if (a0 == a1) {
2088            tcg_out_insn(s, RR, SR, a0, a2);
2089        } else {
2090            tcg_out_insn(s, RRFa, SRK, a0, a1, a2);
2091        }
2092        break;
2093
2094    case INDEX_op_and_i32:
2095        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2096        if (const_args[2]) {
2097            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2098            tgen_andi(s, TCG_TYPE_I32, a0, a2);
2099        } else if (a0 == a1) {
2100            tcg_out_insn(s, RR, NR, a0, a2);
2101        } else {
2102            tcg_out_insn(s, RRFa, NRK, a0, a1, a2);
2103        }
2104        break;
2105    case INDEX_op_or_i32:
2106        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2107        if (const_args[2]) {
2108            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2109            tgen_ori(s, a0, a2);
2110        } else if (a0 == a1) {
2111            tcg_out_insn(s, RR, OR, a0, a2);
2112        } else {
2113            tcg_out_insn(s, RRFa, ORK, a0, a1, a2);
2114        }
2115        break;
2116    case INDEX_op_xor_i32:
2117        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2118        if (const_args[2]) {
2119            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2120            tcg_out_insn(s, RIL, XILF, a0, a2);
2121        } else if (a0 == a1) {
2122            tcg_out_insn(s, RR, XR, args[0], args[2]);
2123        } else {
2124            tcg_out_insn(s, RRFa, XRK, a0, a1, a2);
2125        }
2126        break;
2127
2128    case INDEX_op_andc_i32:
2129        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2130        if (const_args[2]) {
2131            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2132            tgen_andi(s, TCG_TYPE_I32, a0, (uint32_t)~a2);
2133	} else {
2134            tcg_out_insn(s, RRFa, NCRK, a0, a1, a2);
2135	}
2136        break;
2137    case INDEX_op_orc_i32:
2138        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2139        if (const_args[2]) {
2140            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2141            tgen_ori(s, a0, (uint32_t)~a2);
2142        } else {
2143            tcg_out_insn(s, RRFa, OCRK, a0, a1, a2);
2144        }
2145        break;
2146    case INDEX_op_eqv_i32:
2147        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2148        if (const_args[2]) {
2149            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2150            tcg_out_insn(s, RIL, XILF, a0, ~a2);
2151        } else {
2152            tcg_out_insn(s, RRFa, NXRK, a0, a1, a2);
2153        }
2154        break;
2155    case INDEX_op_nand_i32:
2156        tcg_out_insn(s, RRFa, NNRK, args[0], args[1], args[2]);
2157        break;
2158    case INDEX_op_nor_i32:
2159        tcg_out_insn(s, RRFa, NORK, args[0], args[1], args[2]);
2160        break;
2161
2162    case INDEX_op_neg_i32:
2163        tcg_out_insn(s, RR, LCR, args[0], args[1]);
2164        break;
2165    case INDEX_op_not_i32:
2166        tcg_out_insn(s, RRFa, NORK, args[0], args[1], args[1]);
2167        break;
2168
2169    case INDEX_op_mul_i32:
2170        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2171        if (const_args[2]) {
2172            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2173            if (a2 == (int16_t)a2) {
2174                tcg_out_insn(s, RI, MHI, a0, a2);
2175            } else {
2176                tcg_out_insn(s, RIL, MSFI, a0, a2);
2177            }
2178        } else if (a0 == a1) {
2179            tcg_out_insn(s, RRE, MSR, a0, a2);
2180        } else {
2181            tcg_out_insn(s, RRFa, MSRKC, a0, a1, a2);
2182        }
2183        break;
2184
2185    case INDEX_op_div2_i32:
2186        tcg_debug_assert(args[0] == args[2]);
2187        tcg_debug_assert(args[1] == args[3]);
2188        tcg_debug_assert((args[1] & 1) == 0);
2189        tcg_debug_assert(args[0] == args[1] + 1);
2190        tcg_out_insn(s, RR, DR, args[1], args[4]);
2191        break;
2192    case INDEX_op_divu2_i32:
2193        tcg_debug_assert(args[0] == args[2]);
2194        tcg_debug_assert(args[1] == args[3]);
2195        tcg_debug_assert((args[1] & 1) == 0);
2196        tcg_debug_assert(args[0] == args[1] + 1);
2197        tcg_out_insn(s, RRE, DLR, args[1], args[4]);
2198        break;
2199
2200    case INDEX_op_shl_i32:
2201        op = RS_SLL;
2202        op2 = RSY_SLLK;
2203    do_shift32:
2204        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2205        if (a0 == a1) {
2206            if (const_args[2]) {
2207                tcg_out_sh32(s, op, a0, TCG_REG_NONE, a2);
2208            } else {
2209                tcg_out_sh32(s, op, a0, a2, 0);
2210            }
2211        } else {
2212            /* Using tcg_out_sh64 here for the format; it is a 32-bit shift.  */
2213            if (const_args[2]) {
2214                tcg_out_sh64(s, op2, a0, a1, TCG_REG_NONE, a2);
2215            } else {
2216                tcg_out_sh64(s, op2, a0, a1, a2, 0);
2217            }
2218        }
2219        break;
2220    case INDEX_op_shr_i32:
2221        op = RS_SRL;
2222        op2 = RSY_SRLK;
2223        goto do_shift32;
2224    case INDEX_op_sar_i32:
2225        op = RS_SRA;
2226        op2 = RSY_SRAK;
2227        goto do_shift32;
2228
2229    case INDEX_op_rotl_i32:
2230        /* ??? Using tcg_out_sh64 here for the format; it is a 32-bit rol.  */
2231        if (const_args[2]) {
2232            tcg_out_sh64(s, RSY_RLL, args[0], args[1], TCG_REG_NONE, args[2]);
2233        } else {
2234            tcg_out_sh64(s, RSY_RLL, args[0], args[1], args[2], 0);
2235        }
2236        break;
2237    case INDEX_op_rotr_i32:
2238        if (const_args[2]) {
2239            tcg_out_sh64(s, RSY_RLL, args[0], args[1],
2240                         TCG_REG_NONE, (32 - args[2]) & 31);
2241        } else {
2242            tcg_out_insn(s, RR, LCR, TCG_TMP0, args[2]);
2243            tcg_out_sh64(s, RSY_RLL, args[0], args[1], TCG_TMP0, 0);
2244        }
2245        break;
2246
2247    case INDEX_op_bswap16_i32:
2248        a0 = args[0], a1 = args[1], a2 = args[2];
2249        tcg_out_insn(s, RRE, LRVR, a0, a1);
2250        if (a2 & TCG_BSWAP_OS) {
2251            tcg_out_sh32(s, RS_SRA, a0, TCG_REG_NONE, 16);
2252        } else {
2253            tcg_out_sh32(s, RS_SRL, a0, TCG_REG_NONE, 16);
2254        }
2255        break;
2256    case INDEX_op_bswap16_i64:
2257        a0 = args[0], a1 = args[1], a2 = args[2];
2258        tcg_out_insn(s, RRE, LRVGR, a0, a1);
2259        if (a2 & TCG_BSWAP_OS) {
2260            tcg_out_sh64(s, RSY_SRAG, a0, a0, TCG_REG_NONE, 48);
2261        } else {
2262            tcg_out_sh64(s, RSY_SRLG, a0, a0, TCG_REG_NONE, 48);
2263        }
2264        break;
2265
2266    case INDEX_op_bswap32_i32:
2267        tcg_out_insn(s, RRE, LRVR, args[0], args[1]);
2268        break;
2269    case INDEX_op_bswap32_i64:
2270        a0 = args[0], a1 = args[1], a2 = args[2];
2271        tcg_out_insn(s, RRE, LRVR, a0, a1);
2272        if (a2 & TCG_BSWAP_OS) {
2273            tcg_out_ext32s(s, a0, a0);
2274        } else if ((a2 & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
2275            tcg_out_ext32u(s, a0, a0);
2276        }
2277        break;
2278
2279    case INDEX_op_add2_i32:
2280        if (const_args[4]) {
2281            tcg_out_insn(s, RIL, ALFI, args[0], args[4]);
2282        } else {
2283            tcg_out_insn(s, RR, ALR, args[0], args[4]);
2284        }
2285        tcg_out_insn(s, RRE, ALCR, args[1], args[5]);
2286        break;
2287    case INDEX_op_sub2_i32:
2288        if (const_args[4]) {
2289            tcg_out_insn(s, RIL, SLFI, args[0], args[4]);
2290        } else {
2291            tcg_out_insn(s, RR, SLR, args[0], args[4]);
2292        }
2293        tcg_out_insn(s, RRE, SLBR, args[1], args[5]);
2294        break;
2295
2296    case INDEX_op_br:
2297        tgen_branch(s, S390_CC_ALWAYS, arg_label(args[0]));
2298        break;
2299
2300    case INDEX_op_brcond_i32:
2301        tgen_brcond(s, TCG_TYPE_I32, args[2], args[0],
2302                    args[1], const_args[1], arg_label(args[3]));
2303        break;
2304    case INDEX_op_setcond_i32:
2305        tgen_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1],
2306                     args[2], const_args[2], false);
2307        break;
2308    case INDEX_op_negsetcond_i32:
2309        tgen_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1],
2310                     args[2], const_args[2], true);
2311        break;
2312    case INDEX_op_movcond_i32:
2313        tgen_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1],
2314                     args[2], const_args[2], args[3], const_args[3], args[4]);
2315        break;
2316
2317    case INDEX_op_qemu_ld_a32_i32:
2318    case INDEX_op_qemu_ld_a64_i32:
2319        tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I32);
2320        break;
2321    case INDEX_op_qemu_ld_a32_i64:
2322    case INDEX_op_qemu_ld_a64_i64:
2323        tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I64);
2324        break;
2325    case INDEX_op_qemu_st_a32_i32:
2326    case INDEX_op_qemu_st_a64_i32:
2327        tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I32);
2328        break;
2329    case INDEX_op_qemu_st_a32_i64:
2330    case INDEX_op_qemu_st_a64_i64:
2331        tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I64);
2332        break;
2333    case INDEX_op_qemu_ld_a32_i128:
2334    case INDEX_op_qemu_ld_a64_i128:
2335        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], true);
2336        break;
2337    case INDEX_op_qemu_st_a32_i128:
2338    case INDEX_op_qemu_st_a64_i128:
2339        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], false);
2340        break;
2341
2342    case INDEX_op_ld16s_i64:
2343        tcg_out_mem(s, 0, RXY_LGH, args[0], args[1], TCG_REG_NONE, args[2]);
2344        break;
2345    case INDEX_op_ld32u_i64:
2346        tcg_out_mem(s, 0, RXY_LLGF, args[0], args[1], TCG_REG_NONE, args[2]);
2347        break;
2348    case INDEX_op_ld32s_i64:
2349        tcg_out_mem(s, 0, RXY_LGF, args[0], args[1], TCG_REG_NONE, args[2]);
2350        break;
2351    case INDEX_op_ld_i64:
2352        tcg_out_ld(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2353        break;
2354
2355    case INDEX_op_st32_i64:
2356        tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2357        break;
2358    case INDEX_op_st_i64:
2359        tcg_out_st(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2360        break;
2361
2362    case INDEX_op_add_i64:
2363        a0 = args[0], a1 = args[1], a2 = args[2];
2364        if (const_args[2]) {
2365        do_addi_64:
2366            if (a0 == a1) {
2367                if (a2 == (int16_t)a2) {
2368                    tcg_out_insn(s, RI, AGHI, a0, a2);
2369                    break;
2370                }
2371                if (a2 == (int32_t)a2) {
2372                    tcg_out_insn(s, RIL, AGFI, a0, a2);
2373                    break;
2374                }
2375                if (a2 == (uint32_t)a2) {
2376                    tcg_out_insn(s, RIL, ALGFI, a0, a2);
2377                    break;
2378                }
2379                if (-a2 == (uint32_t)-a2) {
2380                    tcg_out_insn(s, RIL, SLGFI, a0, -a2);
2381                    break;
2382                }
2383            }
2384            tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
2385        } else if (a0 == a1) {
2386            tcg_out_insn(s, RRE, AGR, a0, a2);
2387        } else {
2388            tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
2389        }
2390        break;
2391    case INDEX_op_sub_i64:
2392        a0 = args[0], a1 = args[1], a2 = args[2];
2393        if (const_args[2]) {
2394            a2 = -a2;
2395            goto do_addi_64;
2396        } else {
2397            tcg_out_insn(s, RRFa, SGRK, a0, a1, a2);
2398        }
2399        break;
2400
2401    case INDEX_op_and_i64:
2402        a0 = args[0], a1 = args[1], a2 = args[2];
2403        if (const_args[2]) {
2404            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2405            tgen_andi(s, TCG_TYPE_I64, args[0], args[2]);
2406        } else {
2407            tcg_out_insn(s, RRFa, NGRK, a0, a1, a2);
2408        }
2409        break;
2410    case INDEX_op_or_i64:
2411        a0 = args[0], a1 = args[1], a2 = args[2];
2412        if (const_args[2]) {
2413            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2414            tgen_ori(s, a0, a2);
2415        } else {
2416            tcg_out_insn(s, RRFa, OGRK, a0, a1, a2);
2417        }
2418        break;
2419    case INDEX_op_xor_i64:
2420        a0 = args[0], a1 = args[1], a2 = args[2];
2421        if (const_args[2]) {
2422            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2423            tgen_xori(s, a0, a2);
2424        } else {
2425            tcg_out_insn(s, RRFa, XGRK, a0, a1, a2);
2426        }
2427        break;
2428
2429    case INDEX_op_andc_i64:
2430        a0 = args[0], a1 = args[1], a2 = args[2];
2431        if (const_args[2]) {
2432            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2433            tgen_andi(s, TCG_TYPE_I64, a0, ~a2);
2434        } else {
2435            tcg_out_insn(s, RRFa, NCGRK, a0, a1, a2);
2436        }
2437        break;
2438    case INDEX_op_orc_i64:
2439        a0 = args[0], a1 = args[1], a2 = args[2];
2440        if (const_args[2]) {
2441            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2442            tgen_ori(s, a0, ~a2);
2443        } else {
2444            tcg_out_insn(s, RRFa, OCGRK, a0, a1, a2);
2445        }
2446        break;
2447    case INDEX_op_eqv_i64:
2448        a0 = args[0], a1 = args[1], a2 = args[2];
2449        if (const_args[2]) {
2450            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2451            tgen_xori(s, a0, ~a2);
2452        } else {
2453            tcg_out_insn(s, RRFa, NXGRK, a0, a1, a2);
2454        }
2455        break;
2456    case INDEX_op_nand_i64:
2457        tcg_out_insn(s, RRFa, NNGRK, args[0], args[1], args[2]);
2458        break;
2459    case INDEX_op_nor_i64:
2460        tcg_out_insn(s, RRFa, NOGRK, args[0], args[1], args[2]);
2461        break;
2462
2463    case INDEX_op_neg_i64:
2464        tcg_out_insn(s, RRE, LCGR, args[0], args[1]);
2465        break;
2466    case INDEX_op_not_i64:
2467        tcg_out_insn(s, RRFa, NOGRK, args[0], args[1], args[1]);
2468        break;
2469    case INDEX_op_bswap64_i64:
2470        tcg_out_insn(s, RRE, LRVGR, args[0], args[1]);
2471        break;
2472
2473    case INDEX_op_mul_i64:
2474        a0 = args[0], a1 = args[1], a2 = args[2];
2475        if (const_args[2]) {
2476            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2477            if (a2 == (int16_t)a2) {
2478                tcg_out_insn(s, RI, MGHI, a0, a2);
2479            } else {
2480                tcg_out_insn(s, RIL, MSGFI, a0, a2);
2481            }
2482        } else if (a0 == a1) {
2483            tcg_out_insn(s, RRE, MSGR, a0, a2);
2484        } else {
2485            tcg_out_insn(s, RRFa, MSGRKC, a0, a1, a2);
2486        }
2487        break;
2488
2489    case INDEX_op_div2_i64:
2490        /*
2491         * ??? We get an unnecessary sign-extension of the dividend
2492         * into op0 with this definition, but as we do in fact always
2493         * produce both quotient and remainder using INDEX_op_div_i64
2494         * instead requires jumping through even more hoops.
2495         */
2496        tcg_debug_assert(args[0] == args[2]);
2497        tcg_debug_assert(args[1] == args[3]);
2498        tcg_debug_assert((args[1] & 1) == 0);
2499        tcg_debug_assert(args[0] == args[1] + 1);
2500        tcg_out_insn(s, RRE, DSGR, args[1], args[4]);
2501        break;
2502    case INDEX_op_divu2_i64:
2503        tcg_debug_assert(args[0] == args[2]);
2504        tcg_debug_assert(args[1] == args[3]);
2505        tcg_debug_assert((args[1] & 1) == 0);
2506        tcg_debug_assert(args[0] == args[1] + 1);
2507        tcg_out_insn(s, RRE, DLGR, args[1], args[4]);
2508        break;
2509    case INDEX_op_mulu2_i64:
2510        tcg_debug_assert(args[0] == args[2]);
2511        tcg_debug_assert((args[1] & 1) == 0);
2512        tcg_debug_assert(args[0] == args[1] + 1);
2513        tcg_out_insn(s, RRE, MLGR, args[1], args[3]);
2514        break;
2515    case INDEX_op_muls2_i64:
2516        tcg_debug_assert((args[1] & 1) == 0);
2517        tcg_debug_assert(args[0] == args[1] + 1);
2518        tcg_out_insn(s, RRFa, MGRK, args[1], args[2], args[3]);
2519        break;
2520
2521    case INDEX_op_shl_i64:
2522        op = RSY_SLLG;
2523    do_shift64:
2524        if (const_args[2]) {
2525            tcg_out_sh64(s, op, args[0], args[1], TCG_REG_NONE, args[2]);
2526        } else {
2527            tcg_out_sh64(s, op, args[0], args[1], args[2], 0);
2528        }
2529        break;
2530    case INDEX_op_shr_i64:
2531        op = RSY_SRLG;
2532        goto do_shift64;
2533    case INDEX_op_sar_i64:
2534        op = RSY_SRAG;
2535        goto do_shift64;
2536
2537    case INDEX_op_rotl_i64:
2538        if (const_args[2]) {
2539            tcg_out_sh64(s, RSY_RLLG, args[0], args[1],
2540                         TCG_REG_NONE, args[2]);
2541        } else {
2542            tcg_out_sh64(s, RSY_RLLG, args[0], args[1], args[2], 0);
2543        }
2544        break;
2545    case INDEX_op_rotr_i64:
2546        if (const_args[2]) {
2547            tcg_out_sh64(s, RSY_RLLG, args[0], args[1],
2548                         TCG_REG_NONE, (64 - args[2]) & 63);
2549        } else {
2550            /* We can use the smaller 32-bit negate because only the
2551               low 6 bits are examined for the rotate.  */
2552            tcg_out_insn(s, RR, LCR, TCG_TMP0, args[2]);
2553            tcg_out_sh64(s, RSY_RLLG, args[0], args[1], TCG_TMP0, 0);
2554        }
2555        break;
2556
2557    case INDEX_op_add2_i64:
2558        if (const_args[4]) {
2559            if ((int64_t)args[4] >= 0) {
2560                tcg_out_insn(s, RIL, ALGFI, args[0], args[4]);
2561            } else {
2562                tcg_out_insn(s, RIL, SLGFI, args[0], -args[4]);
2563            }
2564        } else {
2565            tcg_out_insn(s, RRE, ALGR, args[0], args[4]);
2566        }
2567        tcg_out_insn(s, RRE, ALCGR, args[1], args[5]);
2568        break;
2569    case INDEX_op_sub2_i64:
2570        if (const_args[4]) {
2571            if ((int64_t)args[4] >= 0) {
2572                tcg_out_insn(s, RIL, SLGFI, args[0], args[4]);
2573            } else {
2574                tcg_out_insn(s, RIL, ALGFI, args[0], -args[4]);
2575            }
2576        } else {
2577            tcg_out_insn(s, RRE, SLGR, args[0], args[4]);
2578        }
2579        tcg_out_insn(s, RRE, SLBGR, args[1], args[5]);
2580        break;
2581
2582    case INDEX_op_brcond_i64:
2583        tgen_brcond(s, TCG_TYPE_I64, args[2], args[0],
2584                    args[1], const_args[1], arg_label(args[3]));
2585        break;
2586    case INDEX_op_setcond_i64:
2587        tgen_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1],
2588                     args[2], const_args[2], false);
2589        break;
2590    case INDEX_op_negsetcond_i64:
2591        tgen_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1],
2592                     args[2], const_args[2], true);
2593        break;
2594    case INDEX_op_movcond_i64:
2595        tgen_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1],
2596                     args[2], const_args[2], args[3], const_args[3], args[4]);
2597        break;
2598
2599    OP_32_64(deposit):
2600        a0 = args[0], a1 = args[1], a2 = args[2];
2601        if (const_args[1]) {
2602            tgen_deposit(s, a0, a2, args[3], args[4], 1);
2603        } else {
2604            /* Since we can't support "0Z" as a constraint, we allow a1 in
2605               any register.  Fix things up as if a matching constraint.  */
2606            if (a0 != a1) {
2607                TCGType type = (opc == INDEX_op_deposit_i64);
2608                if (a0 == a2) {
2609                    tcg_out_mov(s, type, TCG_TMP0, a2);
2610                    a2 = TCG_TMP0;
2611                }
2612                tcg_out_mov(s, type, a0, a1);
2613            }
2614            tgen_deposit(s, a0, a2, args[3], args[4], 0);
2615        }
2616        break;
2617
2618    OP_32_64(extract):
2619        tgen_extract(s, args[0], args[1], args[2], args[3]);
2620        break;
2621
2622    case INDEX_op_clz_i64:
2623        tgen_clz(s, args[0], args[1], args[2], const_args[2]);
2624        break;
2625
2626    case INDEX_op_ctpop_i32:
2627        tgen_ctpop(s, TCG_TYPE_I32, args[0], args[1]);
2628        break;
2629    case INDEX_op_ctpop_i64:
2630        tgen_ctpop(s, TCG_TYPE_I64, args[0], args[1]);
2631        break;
2632
2633    case INDEX_op_mb:
2634        /* The host memory model is quite strong, we simply need to
2635           serialize the instruction stream.  */
2636        if (args[0] & TCG_MO_ST_LD) {
2637            /* fast-bcr-serialization facility (45) is present */
2638            tcg_out_insn(s, RR, BCR, 14, 0);
2639        }
2640        break;
2641
2642    case INDEX_op_mov_i32:  /* Always emitted via tcg_out_mov.  */
2643    case INDEX_op_mov_i64:
2644    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
2645    case INDEX_op_exit_tb:  /* Always emitted via tcg_out_exit_tb.  */
2646    case INDEX_op_goto_tb:  /* Always emitted via tcg_out_goto_tb.  */
2647    case INDEX_op_ext8s_i32:  /* Always emitted via tcg_reg_alloc_op.  */
2648    case INDEX_op_ext8s_i64:
2649    case INDEX_op_ext8u_i32:
2650    case INDEX_op_ext8u_i64:
2651    case INDEX_op_ext16s_i32:
2652    case INDEX_op_ext16s_i64:
2653    case INDEX_op_ext16u_i32:
2654    case INDEX_op_ext16u_i64:
2655    case INDEX_op_ext32s_i64:
2656    case INDEX_op_ext32u_i64:
2657    case INDEX_op_ext_i32_i64:
2658    case INDEX_op_extu_i32_i64:
2659    case INDEX_op_extrl_i64_i32:
2660    default:
2661        g_assert_not_reached();
2662    }
2663}
2664
2665static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
2666                            TCGReg dst, TCGReg src)
2667{
2668    if (is_general_reg(src)) {
2669        /* Replicate general register into two MO_64. */
2670        tcg_out_insn(s, VRRf, VLVGP, dst, src, src);
2671        if (vece == MO_64) {
2672            return true;
2673        }
2674        src = dst;
2675    }
2676
2677    /*
2678     * Recall that the "standard" integer, within a vector, is the
2679     * rightmost element of the leftmost doubleword, a-la VLLEZ.
2680     */
2681    tcg_out_insn(s, VRIc, VREP, dst, (8 >> vece) - 1, src, vece);
2682    return true;
2683}
2684
2685static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
2686                             TCGReg dst, TCGReg base, intptr_t offset)
2687{
2688    tcg_out_vrx_mem(s, VRX_VLREP, dst, base, TCG_REG_NONE, offset, vece);
2689    return true;
2690}
2691
2692static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
2693                             TCGReg dst, int64_t val)
2694{
2695    int i, mask, msb, lsb;
2696
2697    /* Look for int16_t elements.  */
2698    if (vece <= MO_16 ||
2699        (vece == MO_32 ? (int32_t)val : val) == (int16_t)val) {
2700        tcg_out_insn(s, VRIa, VREPI, dst, val, vece);
2701        return;
2702    }
2703
2704    /* Look for bit masks.  */
2705    if (vece == MO_32) {
2706        if (risbg_mask((int32_t)val)) {
2707            /* Handle wraparound by swapping msb and lsb.  */
2708            if ((val & 0x80000001u) == 0x80000001u) {
2709                msb = 32 - ctz32(~val);
2710                lsb = clz32(~val) - 1;
2711            } else {
2712                msb = clz32(val);
2713                lsb = 31 - ctz32(val);
2714            }
2715            tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_32);
2716            return;
2717        }
2718    } else {
2719        if (risbg_mask(val)) {
2720            /* Handle wraparound by swapping msb and lsb.  */
2721            if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
2722                /* Handle wraparound by swapping msb and lsb.  */
2723                msb = 64 - ctz64(~val);
2724                lsb = clz64(~val) - 1;
2725            } else {
2726                msb = clz64(val);
2727                lsb = 63 - ctz64(val);
2728            }
2729            tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_64);
2730            return;
2731        }
2732    }
2733
2734    /* Look for all bytes 0x00 or 0xff.  */
2735    for (i = mask = 0; i < 8; i++) {
2736        uint8_t byte = val >> (i * 8);
2737        if (byte == 0xff) {
2738            mask |= 1 << i;
2739        } else if (byte != 0) {
2740            break;
2741        }
2742    }
2743    if (i == 8) {
2744        tcg_out_insn(s, VRIa, VGBM, dst, mask * 0x0101, 0);
2745        return;
2746    }
2747
2748    /* Otherwise, stuff it in the constant pool.  */
2749    tcg_out_insn(s, RIL, LARL, TCG_TMP0, 0);
2750    new_pool_label(s, val, R_390_PC32DBL, s->code_ptr - 2, 2);
2751    tcg_out_insn(s, VRX, VLREP, dst, TCG_TMP0, TCG_REG_NONE, 0, MO_64);
2752}
2753
2754static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
2755                           unsigned vecl, unsigned vece,
2756                           const TCGArg args[TCG_MAX_OP_ARGS],
2757                           const int const_args[TCG_MAX_OP_ARGS])
2758{
2759    TCGType type = vecl + TCG_TYPE_V64;
2760    TCGArg a0 = args[0], a1 = args[1], a2 = args[2];
2761
2762    switch (opc) {
2763    case INDEX_op_ld_vec:
2764        tcg_out_ld(s, type, a0, a1, a2);
2765        break;
2766    case INDEX_op_st_vec:
2767        tcg_out_st(s, type, a0, a1, a2);
2768        break;
2769    case INDEX_op_dupm_vec:
2770        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
2771        break;
2772
2773    case INDEX_op_abs_vec:
2774        tcg_out_insn(s, VRRa, VLP, a0, a1, vece);
2775        break;
2776    case INDEX_op_neg_vec:
2777        tcg_out_insn(s, VRRa, VLC, a0, a1, vece);
2778        break;
2779    case INDEX_op_not_vec:
2780        tcg_out_insn(s, VRRc, VNO, a0, a1, a1, 0);
2781        break;
2782
2783    case INDEX_op_add_vec:
2784        tcg_out_insn(s, VRRc, VA, a0, a1, a2, vece);
2785        break;
2786    case INDEX_op_sub_vec:
2787        tcg_out_insn(s, VRRc, VS, a0, a1, a2, vece);
2788        break;
2789    case INDEX_op_and_vec:
2790        tcg_out_insn(s, VRRc, VN, a0, a1, a2, 0);
2791        break;
2792    case INDEX_op_andc_vec:
2793        tcg_out_insn(s, VRRc, VNC, a0, a1, a2, 0);
2794        break;
2795    case INDEX_op_mul_vec:
2796        tcg_out_insn(s, VRRc, VML, a0, a1, a2, vece);
2797        break;
2798    case INDEX_op_or_vec:
2799        tcg_out_insn(s, VRRc, VO, a0, a1, a2, 0);
2800        break;
2801    case INDEX_op_orc_vec:
2802        tcg_out_insn(s, VRRc, VOC, a0, a1, a2, 0);
2803        break;
2804    case INDEX_op_xor_vec:
2805        tcg_out_insn(s, VRRc, VX, a0, a1, a2, 0);
2806        break;
2807    case INDEX_op_nand_vec:
2808        tcg_out_insn(s, VRRc, VNN, a0, a1, a2, 0);
2809        break;
2810    case INDEX_op_nor_vec:
2811        tcg_out_insn(s, VRRc, VNO, a0, a1, a2, 0);
2812        break;
2813    case INDEX_op_eqv_vec:
2814        tcg_out_insn(s, VRRc, VNX, a0, a1, a2, 0);
2815        break;
2816
2817    case INDEX_op_shli_vec:
2818        tcg_out_insn(s, VRSa, VESL, a0, a2, TCG_REG_NONE, a1, vece);
2819        break;
2820    case INDEX_op_shri_vec:
2821        tcg_out_insn(s, VRSa, VESRL, a0, a2, TCG_REG_NONE, a1, vece);
2822        break;
2823    case INDEX_op_sari_vec:
2824        tcg_out_insn(s, VRSa, VESRA, a0, a2, TCG_REG_NONE, a1, vece);
2825        break;
2826    case INDEX_op_rotli_vec:
2827        tcg_out_insn(s, VRSa, VERLL, a0, a2, TCG_REG_NONE, a1, vece);
2828        break;
2829    case INDEX_op_shls_vec:
2830        tcg_out_insn(s, VRSa, VESL, a0, 0, a2, a1, vece);
2831        break;
2832    case INDEX_op_shrs_vec:
2833        tcg_out_insn(s, VRSa, VESRL, a0, 0, a2, a1, vece);
2834        break;
2835    case INDEX_op_sars_vec:
2836        tcg_out_insn(s, VRSa, VESRA, a0, 0, a2, a1, vece);
2837        break;
2838    case INDEX_op_rotls_vec:
2839        tcg_out_insn(s, VRSa, VERLL, a0, 0, a2, a1, vece);
2840        break;
2841    case INDEX_op_shlv_vec:
2842        tcg_out_insn(s, VRRc, VESLV, a0, a1, a2, vece);
2843        break;
2844    case INDEX_op_shrv_vec:
2845        tcg_out_insn(s, VRRc, VESRLV, a0, a1, a2, vece);
2846        break;
2847    case INDEX_op_sarv_vec:
2848        tcg_out_insn(s, VRRc, VESRAV, a0, a1, a2, vece);
2849        break;
2850    case INDEX_op_rotlv_vec:
2851        tcg_out_insn(s, VRRc, VERLLV, a0, a1, a2, vece);
2852        break;
2853
2854    case INDEX_op_smin_vec:
2855        tcg_out_insn(s, VRRc, VMN, a0, a1, a2, vece);
2856        break;
2857    case INDEX_op_smax_vec:
2858        tcg_out_insn(s, VRRc, VMX, a0, a1, a2, vece);
2859        break;
2860    case INDEX_op_umin_vec:
2861        tcg_out_insn(s, VRRc, VMNL, a0, a1, a2, vece);
2862        break;
2863    case INDEX_op_umax_vec:
2864        tcg_out_insn(s, VRRc, VMXL, a0, a1, a2, vece);
2865        break;
2866
2867    case INDEX_op_bitsel_vec:
2868        tcg_out_insn(s, VRRe, VSEL, a0, a2, args[3], a1);
2869        break;
2870
2871    case INDEX_op_cmp_vec:
2872        switch ((TCGCond)args[3]) {
2873        case TCG_COND_EQ:
2874            tcg_out_insn(s, VRRc, VCEQ, a0, a1, a2, vece);
2875            break;
2876        case TCG_COND_GT:
2877            tcg_out_insn(s, VRRc, VCH, a0, a1, a2, vece);
2878            break;
2879        case TCG_COND_GTU:
2880            tcg_out_insn(s, VRRc, VCHL, a0, a1, a2, vece);
2881            break;
2882        default:
2883            g_assert_not_reached();
2884        }
2885        break;
2886
2887    case INDEX_op_s390_vuph_vec:
2888        tcg_out_insn(s, VRRa, VUPH, a0, a1, vece);
2889        break;
2890    case INDEX_op_s390_vupl_vec:
2891        tcg_out_insn(s, VRRa, VUPL, a0, a1, vece);
2892        break;
2893    case INDEX_op_s390_vpks_vec:
2894        tcg_out_insn(s, VRRc, VPKS, a0, a1, a2, vece);
2895        break;
2896
2897    case INDEX_op_mov_vec:   /* Always emitted via tcg_out_mov.  */
2898    case INDEX_op_dup_vec:   /* Always emitted via tcg_out_dup_vec.  */
2899    default:
2900        g_assert_not_reached();
2901    }
2902}
2903
2904int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
2905{
2906    switch (opc) {
2907    case INDEX_op_abs_vec:
2908    case INDEX_op_add_vec:
2909    case INDEX_op_and_vec:
2910    case INDEX_op_andc_vec:
2911    case INDEX_op_bitsel_vec:
2912    case INDEX_op_eqv_vec:
2913    case INDEX_op_nand_vec:
2914    case INDEX_op_neg_vec:
2915    case INDEX_op_nor_vec:
2916    case INDEX_op_not_vec:
2917    case INDEX_op_or_vec:
2918    case INDEX_op_orc_vec:
2919    case INDEX_op_rotli_vec:
2920    case INDEX_op_rotls_vec:
2921    case INDEX_op_rotlv_vec:
2922    case INDEX_op_sari_vec:
2923    case INDEX_op_sars_vec:
2924    case INDEX_op_sarv_vec:
2925    case INDEX_op_shli_vec:
2926    case INDEX_op_shls_vec:
2927    case INDEX_op_shlv_vec:
2928    case INDEX_op_shri_vec:
2929    case INDEX_op_shrs_vec:
2930    case INDEX_op_shrv_vec:
2931    case INDEX_op_smax_vec:
2932    case INDEX_op_smin_vec:
2933    case INDEX_op_sub_vec:
2934    case INDEX_op_umax_vec:
2935    case INDEX_op_umin_vec:
2936    case INDEX_op_xor_vec:
2937        return 1;
2938    case INDEX_op_cmp_vec:
2939    case INDEX_op_cmpsel_vec:
2940    case INDEX_op_rotrv_vec:
2941        return -1;
2942    case INDEX_op_mul_vec:
2943        return vece < MO_64;
2944    case INDEX_op_ssadd_vec:
2945    case INDEX_op_sssub_vec:
2946        return vece < MO_64 ? -1 : 0;
2947    default:
2948        return 0;
2949    }
2950}
2951
2952static bool expand_vec_cmp_noinv(TCGType type, unsigned vece, TCGv_vec v0,
2953                                 TCGv_vec v1, TCGv_vec v2, TCGCond cond)
2954{
2955    bool need_swap = false, need_inv = false;
2956
2957    switch (cond) {
2958    case TCG_COND_EQ:
2959    case TCG_COND_GT:
2960    case TCG_COND_GTU:
2961        break;
2962    case TCG_COND_NE:
2963    case TCG_COND_LE:
2964    case TCG_COND_LEU:
2965        need_inv = true;
2966        break;
2967    case TCG_COND_LT:
2968    case TCG_COND_LTU:
2969        need_swap = true;
2970        break;
2971    case TCG_COND_GE:
2972    case TCG_COND_GEU:
2973        need_swap = need_inv = true;
2974        break;
2975    default:
2976        g_assert_not_reached();
2977    }
2978
2979    if (need_inv) {
2980        cond = tcg_invert_cond(cond);
2981    }
2982    if (need_swap) {
2983        TCGv_vec t1;
2984        t1 = v1, v1 = v2, v2 = t1;
2985        cond = tcg_swap_cond(cond);
2986    }
2987
2988    vec_gen_4(INDEX_op_cmp_vec, type, vece, tcgv_vec_arg(v0),
2989              tcgv_vec_arg(v1), tcgv_vec_arg(v2), cond);
2990
2991    return need_inv;
2992}
2993
2994static void expand_vec_cmp(TCGType type, unsigned vece, TCGv_vec v0,
2995                           TCGv_vec v1, TCGv_vec v2, TCGCond cond)
2996{
2997    if (expand_vec_cmp_noinv(type, vece, v0, v1, v2, cond)) {
2998        tcg_gen_not_vec(vece, v0, v0);
2999    }
3000}
3001
3002static void expand_vec_cmpsel(TCGType type, unsigned vece, TCGv_vec v0,
3003                              TCGv_vec c1, TCGv_vec c2,
3004                              TCGv_vec v3, TCGv_vec v4, TCGCond cond)
3005{
3006    TCGv_vec t = tcg_temp_new_vec(type);
3007
3008    if (expand_vec_cmp_noinv(type, vece, t, c1, c2, cond)) {
3009        /* Invert the sense of the compare by swapping arguments.  */
3010        tcg_gen_bitsel_vec(vece, v0, t, v4, v3);
3011    } else {
3012        tcg_gen_bitsel_vec(vece, v0, t, v3, v4);
3013    }
3014    tcg_temp_free_vec(t);
3015}
3016
3017static void expand_vec_sat(TCGType type, unsigned vece, TCGv_vec v0,
3018                           TCGv_vec v1, TCGv_vec v2, TCGOpcode add_sub_opc)
3019{
3020    TCGv_vec h1 = tcg_temp_new_vec(type);
3021    TCGv_vec h2 = tcg_temp_new_vec(type);
3022    TCGv_vec l1 = tcg_temp_new_vec(type);
3023    TCGv_vec l2 = tcg_temp_new_vec(type);
3024
3025    tcg_debug_assert (vece < MO_64);
3026
3027    /* Unpack with sign-extension. */
3028    vec_gen_2(INDEX_op_s390_vuph_vec, type, vece,
3029              tcgv_vec_arg(h1), tcgv_vec_arg(v1));
3030    vec_gen_2(INDEX_op_s390_vuph_vec, type, vece,
3031              tcgv_vec_arg(h2), tcgv_vec_arg(v2));
3032
3033    vec_gen_2(INDEX_op_s390_vupl_vec, type, vece,
3034              tcgv_vec_arg(l1), tcgv_vec_arg(v1));
3035    vec_gen_2(INDEX_op_s390_vupl_vec, type, vece,
3036              tcgv_vec_arg(l2), tcgv_vec_arg(v2));
3037
3038    /* Arithmetic on a wider element size. */
3039    vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(h1),
3040              tcgv_vec_arg(h1), tcgv_vec_arg(h2));
3041    vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(l1),
3042              tcgv_vec_arg(l1), tcgv_vec_arg(l2));
3043
3044    /* Pack with saturation. */
3045    vec_gen_3(INDEX_op_s390_vpks_vec, type, vece + 1,
3046              tcgv_vec_arg(v0), tcgv_vec_arg(h1), tcgv_vec_arg(l1));
3047
3048    tcg_temp_free_vec(h1);
3049    tcg_temp_free_vec(h2);
3050    tcg_temp_free_vec(l1);
3051    tcg_temp_free_vec(l2);
3052}
3053
3054void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
3055                       TCGArg a0, ...)
3056{
3057    va_list va;
3058    TCGv_vec v0, v1, v2, v3, v4, t0;
3059
3060    va_start(va, a0);
3061    v0 = temp_tcgv_vec(arg_temp(a0));
3062    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3063    v2 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3064
3065    switch (opc) {
3066    case INDEX_op_cmp_vec:
3067        expand_vec_cmp(type, vece, v0, v1, v2, va_arg(va, TCGArg));
3068        break;
3069
3070    case INDEX_op_cmpsel_vec:
3071        v3 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3072        v4 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3073        expand_vec_cmpsel(type, vece, v0, v1, v2, v3, v4, va_arg(va, TCGArg));
3074        break;
3075
3076    case INDEX_op_rotrv_vec:
3077        t0 = tcg_temp_new_vec(type);
3078        tcg_gen_neg_vec(vece, t0, v2);
3079        tcg_gen_rotlv_vec(vece, v0, v1, t0);
3080        tcg_temp_free_vec(t0);
3081        break;
3082
3083    case INDEX_op_ssadd_vec:
3084        expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_add_vec);
3085        break;
3086    case INDEX_op_sssub_vec:
3087        expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_sub_vec);
3088        break;
3089
3090    default:
3091        g_assert_not_reached();
3092    }
3093    va_end(va);
3094}
3095
3096static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
3097{
3098    switch (op) {
3099    case INDEX_op_goto_ptr:
3100        return C_O0_I1(r);
3101
3102    case INDEX_op_ld8u_i32:
3103    case INDEX_op_ld8u_i64:
3104    case INDEX_op_ld8s_i32:
3105    case INDEX_op_ld8s_i64:
3106    case INDEX_op_ld16u_i32:
3107    case INDEX_op_ld16u_i64:
3108    case INDEX_op_ld16s_i32:
3109    case INDEX_op_ld16s_i64:
3110    case INDEX_op_ld_i32:
3111    case INDEX_op_ld32u_i64:
3112    case INDEX_op_ld32s_i64:
3113    case INDEX_op_ld_i64:
3114        return C_O1_I1(r, r);
3115
3116    case INDEX_op_st8_i32:
3117    case INDEX_op_st8_i64:
3118    case INDEX_op_st16_i32:
3119    case INDEX_op_st16_i64:
3120    case INDEX_op_st_i32:
3121    case INDEX_op_st32_i64:
3122    case INDEX_op_st_i64:
3123        return C_O0_I2(r, r);
3124
3125    case INDEX_op_add_i32:
3126    case INDEX_op_add_i64:
3127    case INDEX_op_shl_i64:
3128    case INDEX_op_shr_i64:
3129    case INDEX_op_sar_i64:
3130    case INDEX_op_rotl_i32:
3131    case INDEX_op_rotl_i64:
3132    case INDEX_op_rotr_i32:
3133    case INDEX_op_rotr_i64:
3134    case INDEX_op_setcond_i32:
3135    case INDEX_op_negsetcond_i32:
3136        return C_O1_I2(r, r, ri);
3137    case INDEX_op_setcond_i64:
3138    case INDEX_op_negsetcond_i64:
3139        return C_O1_I2(r, r, rA);
3140
3141    case INDEX_op_clz_i64:
3142        return C_O1_I2(r, r, rI);
3143
3144    case INDEX_op_sub_i32:
3145    case INDEX_op_sub_i64:
3146    case INDEX_op_and_i32:
3147    case INDEX_op_or_i32:
3148    case INDEX_op_xor_i32:
3149        return C_O1_I2(r, r, ri);
3150    case INDEX_op_and_i64:
3151        return C_O1_I2(r, r, rNKR);
3152    case INDEX_op_or_i64:
3153    case INDEX_op_xor_i64:
3154        return C_O1_I2(r, r, rK);
3155
3156    case INDEX_op_andc_i32:
3157    case INDEX_op_orc_i32:
3158    case INDEX_op_eqv_i32:
3159        return C_O1_I2(r, r, ri);
3160    case INDEX_op_andc_i64:
3161        return C_O1_I2(r, r, rKR);
3162    case INDEX_op_orc_i64:
3163    case INDEX_op_eqv_i64:
3164        return C_O1_I2(r, r, rNK);
3165
3166    case INDEX_op_nand_i32:
3167    case INDEX_op_nand_i64:
3168    case INDEX_op_nor_i32:
3169    case INDEX_op_nor_i64:
3170        return C_O1_I2(r, r, r);
3171
3172    case INDEX_op_mul_i32:
3173        return (HAVE_FACILITY(MISC_INSN_EXT2)
3174                ? C_O1_I2(r, r, ri)
3175                : C_O1_I2(r, 0, ri));
3176    case INDEX_op_mul_i64:
3177        return (HAVE_FACILITY(MISC_INSN_EXT2)
3178                ? C_O1_I2(r, r, rJ)
3179                : C_O1_I2(r, 0, rJ));
3180
3181    case INDEX_op_shl_i32:
3182    case INDEX_op_shr_i32:
3183    case INDEX_op_sar_i32:
3184        return C_O1_I2(r, r, ri);
3185
3186    case INDEX_op_brcond_i32:
3187        return C_O0_I2(r, ri);
3188    case INDEX_op_brcond_i64:
3189        return C_O0_I2(r, rA);
3190
3191    case INDEX_op_bswap16_i32:
3192    case INDEX_op_bswap16_i64:
3193    case INDEX_op_bswap32_i32:
3194    case INDEX_op_bswap32_i64:
3195    case INDEX_op_bswap64_i64:
3196    case INDEX_op_neg_i32:
3197    case INDEX_op_neg_i64:
3198    case INDEX_op_not_i32:
3199    case INDEX_op_not_i64:
3200    case INDEX_op_ext8s_i32:
3201    case INDEX_op_ext8s_i64:
3202    case INDEX_op_ext8u_i32:
3203    case INDEX_op_ext8u_i64:
3204    case INDEX_op_ext16s_i32:
3205    case INDEX_op_ext16s_i64:
3206    case INDEX_op_ext16u_i32:
3207    case INDEX_op_ext16u_i64:
3208    case INDEX_op_ext32s_i64:
3209    case INDEX_op_ext32u_i64:
3210    case INDEX_op_ext_i32_i64:
3211    case INDEX_op_extu_i32_i64:
3212    case INDEX_op_extract_i32:
3213    case INDEX_op_extract_i64:
3214    case INDEX_op_ctpop_i32:
3215    case INDEX_op_ctpop_i64:
3216        return C_O1_I1(r, r);
3217
3218    case INDEX_op_qemu_ld_a32_i32:
3219    case INDEX_op_qemu_ld_a64_i32:
3220    case INDEX_op_qemu_ld_a32_i64:
3221    case INDEX_op_qemu_ld_a64_i64:
3222        return C_O1_I1(r, r);
3223    case INDEX_op_qemu_st_a32_i64:
3224    case INDEX_op_qemu_st_a64_i64:
3225    case INDEX_op_qemu_st_a32_i32:
3226    case INDEX_op_qemu_st_a64_i32:
3227        return C_O0_I2(r, r);
3228    case INDEX_op_qemu_ld_a32_i128:
3229    case INDEX_op_qemu_ld_a64_i128:
3230        return C_O2_I1(o, m, r);
3231    case INDEX_op_qemu_st_a32_i128:
3232    case INDEX_op_qemu_st_a64_i128:
3233        return C_O0_I3(o, m, r);
3234
3235    case INDEX_op_deposit_i32:
3236    case INDEX_op_deposit_i64:
3237        return C_O1_I2(r, rZ, r);
3238
3239    case INDEX_op_movcond_i32:
3240        return C_O1_I4(r, r, ri, rI, r);
3241    case INDEX_op_movcond_i64:
3242        return C_O1_I4(r, r, rA, rI, r);
3243
3244    case INDEX_op_div2_i32:
3245    case INDEX_op_div2_i64:
3246    case INDEX_op_divu2_i32:
3247    case INDEX_op_divu2_i64:
3248        return C_O2_I3(o, m, 0, 1, r);
3249
3250    case INDEX_op_mulu2_i64:
3251        return C_O2_I2(o, m, 0, r);
3252    case INDEX_op_muls2_i64:
3253        return C_O2_I2(o, m, r, r);
3254
3255    case INDEX_op_add2_i32:
3256    case INDEX_op_sub2_i32:
3257        return C_N1_O1_I4(r, r, 0, 1, ri, r);
3258
3259    case INDEX_op_add2_i64:
3260    case INDEX_op_sub2_i64:
3261        return C_N1_O1_I4(r, r, 0, 1, rA, r);
3262
3263    case INDEX_op_st_vec:
3264        return C_O0_I2(v, r);
3265    case INDEX_op_ld_vec:
3266    case INDEX_op_dupm_vec:
3267        return C_O1_I1(v, r);
3268    case INDEX_op_dup_vec:
3269        return C_O1_I1(v, vr);
3270    case INDEX_op_abs_vec:
3271    case INDEX_op_neg_vec:
3272    case INDEX_op_not_vec:
3273    case INDEX_op_rotli_vec:
3274    case INDEX_op_sari_vec:
3275    case INDEX_op_shli_vec:
3276    case INDEX_op_shri_vec:
3277    case INDEX_op_s390_vuph_vec:
3278    case INDEX_op_s390_vupl_vec:
3279        return C_O1_I1(v, v);
3280    case INDEX_op_add_vec:
3281    case INDEX_op_sub_vec:
3282    case INDEX_op_and_vec:
3283    case INDEX_op_andc_vec:
3284    case INDEX_op_or_vec:
3285    case INDEX_op_orc_vec:
3286    case INDEX_op_xor_vec:
3287    case INDEX_op_nand_vec:
3288    case INDEX_op_nor_vec:
3289    case INDEX_op_eqv_vec:
3290    case INDEX_op_cmp_vec:
3291    case INDEX_op_mul_vec:
3292    case INDEX_op_rotlv_vec:
3293    case INDEX_op_rotrv_vec:
3294    case INDEX_op_shlv_vec:
3295    case INDEX_op_shrv_vec:
3296    case INDEX_op_sarv_vec:
3297    case INDEX_op_smax_vec:
3298    case INDEX_op_smin_vec:
3299    case INDEX_op_umax_vec:
3300    case INDEX_op_umin_vec:
3301    case INDEX_op_s390_vpks_vec:
3302        return C_O1_I2(v, v, v);
3303    case INDEX_op_rotls_vec:
3304    case INDEX_op_shls_vec:
3305    case INDEX_op_shrs_vec:
3306    case INDEX_op_sars_vec:
3307        return C_O1_I2(v, v, r);
3308    case INDEX_op_bitsel_vec:
3309        return C_O1_I3(v, v, v, v);
3310
3311    default:
3312        g_assert_not_reached();
3313    }
3314}
3315
3316/*
3317 * Mainline glibc added HWCAP_S390_VX before it was kernel abi.
3318 * Some distros have fixed this up locally, others have not.
3319 */
3320#ifndef HWCAP_S390_VXRS
3321#define HWCAP_S390_VXRS 2048
3322#endif
3323
3324static void query_s390_facilities(void)
3325{
3326    unsigned long hwcap = qemu_getauxval(AT_HWCAP);
3327    const char *which;
3328
3329    /* Is STORE FACILITY LIST EXTENDED available?  Honestly, I believe this
3330       is present on all 64-bit systems, but let's check for it anyway.  */
3331    if (hwcap & HWCAP_S390_STFLE) {
3332        register int r0 __asm__("0") = ARRAY_SIZE(s390_facilities) - 1;
3333        register void *r1 __asm__("1") = s390_facilities;
3334
3335        /* stfle 0(%r1) */
3336        asm volatile(".word 0xb2b0,0x1000"
3337                     : "=r"(r0) : "r"(r0), "r"(r1) : "memory", "cc");
3338    }
3339
3340    /*
3341     * Use of vector registers requires os support beyond the facility bit.
3342     * If the kernel does not advertise support, disable the facility bits.
3343     * There is nothing else we currently care about in the 3rd word, so
3344     * disable VECTOR with one store.
3345     */
3346    if (!(hwcap & HWCAP_S390_VXRS)) {
3347        s390_facilities[2] = 0;
3348    }
3349
3350    /*
3351     * Minimum supported cpu revision is z196.
3352     * Check for all required facilities.
3353     * ZARCH_ACTIVE is done via preprocessor check for 64-bit.
3354     */
3355    if (!HAVE_FACILITY(LONG_DISP)) {
3356        which = "long-displacement";
3357        goto fail;
3358    }
3359    if (!HAVE_FACILITY(EXT_IMM)) {
3360        which = "extended-immediate";
3361        goto fail;
3362    }
3363    if (!HAVE_FACILITY(GEN_INST_EXT)) {
3364        which = "general-instructions-extension";
3365        goto fail;
3366    }
3367    /*
3368     * Facility 45 is a big bin that contains: distinct-operands,
3369     * fast-BCR-serialization, high-word, population-count,
3370     * interlocked-access-1, and load/store-on-condition-1
3371     */
3372    if (!HAVE_FACILITY(45)) {
3373        which = "45";
3374        goto fail;
3375    }
3376    return;
3377
3378 fail:
3379    error_report("%s: missing required facility %s", __func__, which);
3380    exit(EXIT_FAILURE);
3381}
3382
3383static void tcg_target_init(TCGContext *s)
3384{
3385    query_s390_facilities();
3386
3387    tcg_target_available_regs[TCG_TYPE_I32] = 0xffff;
3388    tcg_target_available_regs[TCG_TYPE_I64] = 0xffff;
3389    if (HAVE_FACILITY(VECTOR)) {
3390        tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull;
3391        tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull;
3392    }
3393
3394    tcg_target_call_clobber_regs = 0;
3395    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
3396    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1);
3397    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
3398    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
3399    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
3400    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
3401    /* The r6 register is technically call-saved, but it's also a parameter
3402       register, so it can get killed by setup for the qemu_st helper.  */
3403    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
3404    /* The return register can be considered call-clobbered.  */
3405    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14);
3406
3407    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0);
3408    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1);
3409    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V2);
3410    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V3);
3411    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V4);
3412    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V5);
3413    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V6);
3414    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V7);
3415    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V16);
3416    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V17);
3417    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V18);
3418    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V19);
3419    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V20);
3420    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V21);
3421    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V22);
3422    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V23);
3423    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V24);
3424    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V25);
3425    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V26);
3426    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V27);
3427    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V28);
3428    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V29);
3429    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V30);
3430    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V31);
3431
3432    s->reserved_regs = 0;
3433    tcg_regset_set_reg(s->reserved_regs, TCG_TMP0);
3434    /* XXX many insns can't be used with R0, so we better avoid it for now */
3435    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0);
3436    tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
3437}
3438
3439#define FRAME_SIZE  ((int)(TCG_TARGET_CALL_STACK_OFFSET          \
3440                           + TCG_STATIC_CALL_ARGS_SIZE           \
3441                           + CPU_TEMP_BUF_NLONGS * sizeof(long)))
3442
3443static void tcg_target_qemu_prologue(TCGContext *s)
3444{
3445    /* stmg %r6,%r15,48(%r15) (save registers) */
3446    tcg_out_insn(s, RXY, STMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15, 48);
3447
3448    /* aghi %r15,-frame_size */
3449    tcg_out_insn(s, RI, AGHI, TCG_REG_R15, -FRAME_SIZE);
3450
3451    tcg_set_frame(s, TCG_REG_CALL_STACK,
3452                  TCG_STATIC_CALL_ARGS_SIZE + TCG_TARGET_CALL_STACK_OFFSET,
3453                  CPU_TEMP_BUF_NLONGS * sizeof(long));
3454
3455    if (!tcg_use_softmmu && guest_base >= 0x80000) {
3456        tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base);
3457        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
3458    }
3459
3460    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
3461
3462    /* br %r3 (go to TB) */
3463    tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, tcg_target_call_iarg_regs[1]);
3464
3465    /*
3466     * Return path for goto_ptr. Set return value to 0, a-la exit_tb,
3467     * and fall through to the rest of the epilogue.
3468     */
3469    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
3470    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, 0);
3471
3472    /* TB epilogue */
3473    tb_ret_addr = tcg_splitwx_to_rx(s->code_ptr);
3474
3475    /* lmg %r6,%r15,fs+48(%r15) (restore registers) */
3476    tcg_out_insn(s, RXY, LMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15,
3477                 FRAME_SIZE + 48);
3478
3479    /* br %r14 (return) */
3480    tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_R14);
3481}
3482
3483static void tcg_out_tb_start(TCGContext *s)
3484{
3485    /* nothing to do */
3486}
3487
3488static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
3489{
3490    memset(p, 0x07, count * sizeof(tcg_insn_unit));
3491}
3492
3493typedef struct {
3494    DebugFrameHeader h;
3495    uint8_t fde_def_cfa[4];
3496    uint8_t fde_reg_ofs[18];
3497} DebugFrame;
3498
3499/* We're expecting a 2 byte uleb128 encoded value.  */
3500QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3501
3502#define ELF_HOST_MACHINE  EM_S390
3503
3504static const DebugFrame debug_frame = {
3505    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3506    .h.cie.id = -1,
3507    .h.cie.version = 1,
3508    .h.cie.code_align = 1,
3509    .h.cie.data_align = 8,                /* sleb128 8 */
3510    .h.cie.return_column = TCG_REG_R14,
3511
3512    /* Total FDE size does not include the "len" member.  */
3513    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
3514
3515    .fde_def_cfa = {
3516        12, TCG_REG_CALL_STACK,         /* DW_CFA_def_cfa %r15, ... */
3517        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3518        (FRAME_SIZE >> 7)
3519    },
3520    .fde_reg_ofs = {
3521        0x86, 6,                        /* DW_CFA_offset, %r6, 48 */
3522        0x87, 7,                        /* DW_CFA_offset, %r7, 56 */
3523        0x88, 8,                        /* DW_CFA_offset, %r8, 64 */
3524        0x89, 9,                        /* DW_CFA_offset, %r92, 72 */
3525        0x8a, 10,                       /* DW_CFA_offset, %r10, 80 */
3526        0x8b, 11,                       /* DW_CFA_offset, %r11, 88 */
3527        0x8c, 12,                       /* DW_CFA_offset, %r12, 96 */
3528        0x8d, 13,                       /* DW_CFA_offset, %r13, 104 */
3529        0x8e, 14,                       /* DW_CFA_offset, %r14, 112 */
3530    }
3531};
3532
3533void tcg_register_jit(const void *buf, size_t buf_size)
3534{
3535    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3536}
3537