xref: /openbmc/qemu/tcg/s390x/tcg-target.c.inc (revision 1c7d05ff70f09367ab8b519cbbb69dd5491f85f1)
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_U32        (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#define TCG_CT_CONST_CMP        (1 << 15)
39
40#define ALL_GENERAL_REGS     MAKE_64BIT_MASK(0, 16)
41#define ALL_VECTOR_REGS      MAKE_64BIT_MASK(32, 32)
42
43/* Several places within the instruction set 0 means "no register"
44   rather than TCG_REG_R0.  */
45#define TCG_REG_NONE    0
46
47/* A scratch register that may be be used throughout the backend.  */
48#define TCG_TMP0        TCG_REG_R1
49#define TCG_VEC_TMP0    TCG_REG_V31
50
51#define TCG_GUEST_BASE_REG TCG_REG_R13
52
53/* All of the following instructions are prefixed with their instruction
54   format, and are defined as 8- or 16-bit quantities, even when the two
55   halves of the 16-bit quantity may appear 32 bits apart in the insn.
56   This makes it easy to copy the values from the tables in Appendix B.  */
57typedef enum S390Opcode {
58    RIL_AFI     = 0xc209,
59    RIL_AGFI    = 0xc208,
60    RIL_ALFI    = 0xc20b,
61    RIL_ALGFI   = 0xc20a,
62    RIL_BRASL   = 0xc005,
63    RIL_BRCL    = 0xc004,
64    RIL_CFI     = 0xc20d,
65    RIL_CGFI    = 0xc20c,
66    RIL_CLFI    = 0xc20f,
67    RIL_CLGFI   = 0xc20e,
68    RIL_CLRL    = 0xc60f,
69    RIL_CLGRL   = 0xc60a,
70    RIL_CRL     = 0xc60d,
71    RIL_CGRL    = 0xc608,
72    RIL_IIHF    = 0xc008,
73    RIL_IILF    = 0xc009,
74    RIL_LARL    = 0xc000,
75    RIL_LGFI    = 0xc001,
76    RIL_LGRL    = 0xc408,
77    RIL_LLIHF   = 0xc00e,
78    RIL_LLILF   = 0xc00f,
79    RIL_LRL     = 0xc40d,
80    RIL_MSFI    = 0xc201,
81    RIL_MSGFI   = 0xc200,
82    RIL_NIHF    = 0xc00a,
83    RIL_NILF    = 0xc00b,
84    RIL_OIHF    = 0xc00c,
85    RIL_OILF    = 0xc00d,
86    RIL_SLFI    = 0xc205,
87    RIL_SLGFI   = 0xc204,
88    RIL_XIHF    = 0xc006,
89    RIL_XILF    = 0xc007,
90
91    RI_AGHI     = 0xa70b,
92    RI_AHI      = 0xa70a,
93    RI_BRC      = 0xa704,
94    RI_CHI      = 0xa70e,
95    RI_CGHI     = 0xa70f,
96    RI_IIHH     = 0xa500,
97    RI_IIHL     = 0xa501,
98    RI_IILH     = 0xa502,
99    RI_IILL     = 0xa503,
100    RI_LGHI     = 0xa709,
101    RI_LLIHH    = 0xa50c,
102    RI_LLIHL    = 0xa50d,
103    RI_LLILH    = 0xa50e,
104    RI_LLILL    = 0xa50f,
105    RI_MGHI     = 0xa70d,
106    RI_MHI      = 0xa70c,
107    RI_NIHH     = 0xa504,
108    RI_NIHL     = 0xa505,
109    RI_NILH     = 0xa506,
110    RI_NILL     = 0xa507,
111    RI_OIHH     = 0xa508,
112    RI_OIHL     = 0xa509,
113    RI_OILH     = 0xa50a,
114    RI_OILL     = 0xa50b,
115    RI_TMLL     = 0xa701,
116    RI_TMLH     = 0xa700,
117    RI_TMHL     = 0xa703,
118    RI_TMHH     = 0xa702,
119
120    RIEb_CGRJ    = 0xec64,
121    RIEb_CLGRJ   = 0xec65,
122    RIEb_CLRJ    = 0xec77,
123    RIEb_CRJ     = 0xec76,
124
125    RIEc_CGIJ    = 0xec7c,
126    RIEc_CIJ     = 0xec7e,
127    RIEc_CLGIJ   = 0xec7d,
128    RIEc_CLIJ    = 0xec7f,
129
130    RIEf_RISBG   = 0xec55,
131
132    RIEg_LOCGHI  = 0xec46,
133
134    RRE_AGR     = 0xb908,
135    RRE_ALGR    = 0xb90a,
136    RRE_ALCR    = 0xb998,
137    RRE_ALCGR   = 0xb988,
138    RRE_ALGFR   = 0xb91a,
139    RRE_CGR     = 0xb920,
140    RRE_CLGR    = 0xb921,
141    RRE_DLGR    = 0xb987,
142    RRE_DLR     = 0xb997,
143    RRE_DSGFR   = 0xb91d,
144    RRE_DSGR    = 0xb90d,
145    RRE_FLOGR   = 0xb983,
146    RRE_LGBR    = 0xb906,
147    RRE_LCGR    = 0xb903,
148    RRE_LGFR    = 0xb914,
149    RRE_LGHR    = 0xb907,
150    RRE_LGR     = 0xb904,
151    RRE_LLGCR   = 0xb984,
152    RRE_LLGFR   = 0xb916,
153    RRE_LLGHR   = 0xb985,
154    RRE_LRVR    = 0xb91f,
155    RRE_LRVGR   = 0xb90f,
156    RRE_LTGR    = 0xb902,
157    RRE_MLGR    = 0xb986,
158    RRE_MSGR    = 0xb90c,
159    RRE_MSR     = 0xb252,
160    RRE_NGR     = 0xb980,
161    RRE_OGR     = 0xb981,
162    RRE_SGR     = 0xb909,
163    RRE_SLGR    = 0xb90b,
164    RRE_SLBR    = 0xb999,
165    RRE_SLBGR   = 0xb989,
166    RRE_XGR     = 0xb982,
167
168    RRFa_MGRK   = 0xb9ec,
169    RRFa_MSRKC  = 0xb9fd,
170    RRFa_MSGRKC = 0xb9ed,
171    RRFa_NCRK   = 0xb9f5,
172    RRFa_NCGRK  = 0xb9e5,
173    RRFa_NNRK   = 0xb974,
174    RRFa_NNGRK  = 0xb964,
175    RRFa_NORK   = 0xb976,
176    RRFa_NOGRK  = 0xb966,
177    RRFa_NRK    = 0xb9f4,
178    RRFa_NGRK   = 0xb9e4,
179    RRFa_NXRK   = 0xb977,
180    RRFa_NXGRK  = 0xb967,
181    RRFa_OCRK   = 0xb975,
182    RRFa_OCGRK  = 0xb965,
183    RRFa_ORK    = 0xb9f6,
184    RRFa_OGRK   = 0xb9e6,
185    RRFa_SRK    = 0xb9f9,
186    RRFa_SGRK   = 0xb9e9,
187    RRFa_SLRK   = 0xb9fb,
188    RRFa_SLGRK  = 0xb9eb,
189    RRFa_XRK    = 0xb9f7,
190    RRFa_XGRK   = 0xb9e7,
191
192    RRFam_SELGR = 0xb9e3,
193
194    RRFc_LOCR   = 0xb9f2,
195    RRFc_LOCGR  = 0xb9e2,
196    RRFc_POPCNT = 0xb9e1,
197
198    RR_AR       = 0x1a,
199    RR_ALR      = 0x1e,
200    RR_BASR     = 0x0d,
201    RR_BCR      = 0x07,
202    RR_CLR      = 0x15,
203    RR_CR       = 0x19,
204    RR_DR       = 0x1d,
205    RR_LCR      = 0x13,
206    RR_LR       = 0x18,
207    RR_LTR      = 0x12,
208    RR_NR       = 0x14,
209    RR_OR       = 0x16,
210    RR_SR       = 0x1b,
211    RR_SLR      = 0x1f,
212    RR_XR       = 0x17,
213
214    RSY_RLL     = 0xeb1d,
215    RSY_RLLG    = 0xeb1c,
216    RSY_SLLG    = 0xeb0d,
217    RSY_SLLK    = 0xebdf,
218    RSY_SRAG    = 0xeb0a,
219    RSY_SRAK    = 0xebdc,
220    RSY_SRLG    = 0xeb0c,
221    RSY_SRLK    = 0xebde,
222
223    RS_SLL      = 0x89,
224    RS_SRA      = 0x8a,
225    RS_SRL      = 0x88,
226
227    RXY_AG      = 0xe308,
228    RXY_AY      = 0xe35a,
229    RXY_CG      = 0xe320,
230    RXY_CLG     = 0xe321,
231    RXY_CLY     = 0xe355,
232    RXY_CY      = 0xe359,
233    RXY_LAY     = 0xe371,
234    RXY_LB      = 0xe376,
235    RXY_LG      = 0xe304,
236    RXY_LGB     = 0xe377,
237    RXY_LGF     = 0xe314,
238    RXY_LGH     = 0xe315,
239    RXY_LHY     = 0xe378,
240    RXY_LLGC    = 0xe390,
241    RXY_LLGF    = 0xe316,
242    RXY_LLGH    = 0xe391,
243    RXY_LMG     = 0xeb04,
244    RXY_LPQ     = 0xe38f,
245    RXY_LRV     = 0xe31e,
246    RXY_LRVG    = 0xe30f,
247    RXY_LRVH    = 0xe31f,
248    RXY_LY      = 0xe358,
249    RXY_NG      = 0xe380,
250    RXY_OG      = 0xe381,
251    RXY_STCY    = 0xe372,
252    RXY_STG     = 0xe324,
253    RXY_STHY    = 0xe370,
254    RXY_STMG    = 0xeb24,
255    RXY_STPQ    = 0xe38e,
256    RXY_STRV    = 0xe33e,
257    RXY_STRVG   = 0xe32f,
258    RXY_STRVH   = 0xe33f,
259    RXY_STY     = 0xe350,
260    RXY_XG      = 0xe382,
261
262    RX_A        = 0x5a,
263    RX_C        = 0x59,
264    RX_L        = 0x58,
265    RX_LA       = 0x41,
266    RX_LH       = 0x48,
267    RX_ST       = 0x50,
268    RX_STC      = 0x42,
269    RX_STH      = 0x40,
270
271    VRIa_VGBM   = 0xe744,
272    VRIa_VREPI  = 0xe745,
273    VRIb_VGM    = 0xe746,
274    VRIc_VREP   = 0xe74d,
275
276    VRRa_VLC    = 0xe7de,
277    VRRa_VLP    = 0xe7df,
278    VRRa_VLR    = 0xe756,
279    VRRc_VA     = 0xe7f3,
280    VRRc_VCEQ   = 0xe7f8,   /* we leave the m5 cs field 0 */
281    VRRc_VCH    = 0xe7fb,   /* " */
282    VRRc_VCHL   = 0xe7f9,   /* " */
283    VRRc_VERLLV = 0xe773,
284    VRRc_VESLV  = 0xe770,
285    VRRc_VESRAV = 0xe77a,
286    VRRc_VESRLV = 0xe778,
287    VRRc_VML    = 0xe7a2,
288    VRRc_VMN    = 0xe7fe,
289    VRRc_VMNL   = 0xe7fc,
290    VRRc_VMX    = 0xe7ff,
291    VRRc_VMXL   = 0xe7fd,
292    VRRc_VN     = 0xe768,
293    VRRc_VNC    = 0xe769,
294    VRRc_VNN    = 0xe76e,
295    VRRc_VNO    = 0xe76b,
296    VRRc_VNX    = 0xe76c,
297    VRRc_VO     = 0xe76a,
298    VRRc_VOC    = 0xe76f,
299    VRRc_VPKS   = 0xe797,   /* we leave the m5 cs field 0 */
300    VRRc_VS     = 0xe7f7,
301    VRRa_VUPH   = 0xe7d7,
302    VRRa_VUPL   = 0xe7d6,
303    VRRc_VX     = 0xe76d,
304    VRRe_VSEL   = 0xe78d,
305    VRRf_VLVGP  = 0xe762,
306
307    VRSa_VERLL  = 0xe733,
308    VRSa_VESL   = 0xe730,
309    VRSa_VESRA  = 0xe73a,
310    VRSa_VESRL  = 0xe738,
311    VRSb_VLVG   = 0xe722,
312    VRSc_VLGV   = 0xe721,
313
314    VRX_VL      = 0xe706,
315    VRX_VLLEZ   = 0xe704,
316    VRX_VLREP   = 0xe705,
317    VRX_VST     = 0xe70e,
318    VRX_VSTEF   = 0xe70b,
319    VRX_VSTEG   = 0xe70a,
320
321    NOP         = 0x0707,
322} S390Opcode;
323
324#ifdef CONFIG_DEBUG_TCG
325static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
326    "%r0",  "%r1",  "%r2",  "%r3",  "%r4",  "%r5",  "%r6",  "%r7",
327    "%r8",  "%r9",  "%r10", "%r11", "%r12", "%r13", "%r14", "%r15",
328    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
329    "%v0",  "%v1",  "%v2",  "%v3",  "%v4",  "%v5",  "%v6",  "%v7",
330    "%v8",  "%v9",  "%v10", "%v11", "%v12", "%v13", "%v14", "%v15",
331    "%v16", "%v17", "%v18", "%v19", "%v20", "%v21", "%v22", "%v23",
332    "%v24", "%v25", "%v26", "%v27", "%v28", "%v29", "%v30", "%v31",
333};
334#endif
335
336/* Since R6 is a potential argument register, choose it last of the
337   call-saved registers.  Likewise prefer the call-clobbered registers
338   in reverse order to maximize the chance of avoiding the arguments.  */
339static const int tcg_target_reg_alloc_order[] = {
340    /* Call saved registers.  */
341    TCG_REG_R13,
342    TCG_REG_R12,
343    TCG_REG_R11,
344    TCG_REG_R10,
345    TCG_REG_R9,
346    TCG_REG_R8,
347    TCG_REG_R7,
348    TCG_REG_R6,
349    /* Call clobbered registers.  */
350    TCG_REG_R14,
351    TCG_REG_R0,
352    TCG_REG_R1,
353    /* Argument registers, in reverse order of allocation.  */
354    TCG_REG_R5,
355    TCG_REG_R4,
356    TCG_REG_R3,
357    TCG_REG_R2,
358
359    /* V8-V15 are call saved, and omitted. */
360    TCG_REG_V0,
361    TCG_REG_V1,
362    TCG_REG_V2,
363    TCG_REG_V3,
364    TCG_REG_V4,
365    TCG_REG_V5,
366    TCG_REG_V6,
367    TCG_REG_V7,
368    TCG_REG_V16,
369    TCG_REG_V17,
370    TCG_REG_V18,
371    TCG_REG_V19,
372    TCG_REG_V20,
373    TCG_REG_V21,
374    TCG_REG_V22,
375    TCG_REG_V23,
376    TCG_REG_V24,
377    TCG_REG_V25,
378    TCG_REG_V26,
379    TCG_REG_V27,
380    TCG_REG_V28,
381    TCG_REG_V29,
382    TCG_REG_V30,
383    TCG_REG_V31,
384};
385
386static const int tcg_target_call_iarg_regs[] = {
387    TCG_REG_R2,
388    TCG_REG_R3,
389    TCG_REG_R4,
390    TCG_REG_R5,
391    TCG_REG_R6,
392};
393
394static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
395{
396    tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
397    tcg_debug_assert(slot == 0);
398    return TCG_REG_R2;
399}
400
401#define S390_CC_EQ      8
402#define S390_CC_LT      4
403#define S390_CC_GT      2
404#define S390_CC_OV      1
405#define S390_CC_NE      (S390_CC_LT | S390_CC_GT)
406#define S390_CC_LE      (S390_CC_LT | S390_CC_EQ)
407#define S390_CC_GE      (S390_CC_GT | S390_CC_EQ)
408#define S390_CC_NEVER   0
409#define S390_CC_ALWAYS  15
410
411#define S390_TM_EQ      8  /* CC == 0 */
412#define S390_TM_NE      7  /* CC in {1,2,3} */
413
414/* Condition codes that result from a COMPARE and COMPARE LOGICAL.  */
415static const uint8_t tcg_cond_to_s390_cond[16] = {
416    [TCG_COND_EQ]  = S390_CC_EQ,
417    [TCG_COND_NE]  = S390_CC_NE,
418    [TCG_COND_TSTEQ] = S390_CC_EQ,
419    [TCG_COND_TSTNE] = S390_CC_NE,
420    [TCG_COND_LT]  = S390_CC_LT,
421    [TCG_COND_LE]  = S390_CC_LE,
422    [TCG_COND_GT]  = S390_CC_GT,
423    [TCG_COND_GE]  = S390_CC_GE,
424    [TCG_COND_LTU] = S390_CC_LT,
425    [TCG_COND_LEU] = S390_CC_LE,
426    [TCG_COND_GTU] = S390_CC_GT,
427    [TCG_COND_GEU] = S390_CC_GE,
428};
429
430/* Condition codes that result from a LOAD AND TEST.  Here, we have no
431   unsigned instruction variation, however since the test is vs zero we
432   can re-map the outcomes appropriately.  */
433static const uint8_t tcg_cond_to_ltr_cond[16] = {
434    [TCG_COND_EQ]  = S390_CC_EQ,
435    [TCG_COND_NE]  = S390_CC_NE,
436    [TCG_COND_TSTEQ] = S390_CC_ALWAYS,
437    [TCG_COND_TSTNE] = S390_CC_NEVER,
438    [TCG_COND_LT]  = S390_CC_LT,
439    [TCG_COND_LE]  = S390_CC_LE,
440    [TCG_COND_GT]  = S390_CC_GT,
441    [TCG_COND_GE]  = S390_CC_GE,
442    [TCG_COND_LTU] = S390_CC_NEVER,
443    [TCG_COND_LEU] = S390_CC_EQ,
444    [TCG_COND_GTU] = S390_CC_NE,
445    [TCG_COND_GEU] = S390_CC_ALWAYS,
446};
447
448static const tcg_insn_unit *tb_ret_addr;
449uint64_t s390_facilities[3];
450
451static inline bool is_general_reg(TCGReg r)
452{
453    return r <= TCG_REG_R15;
454}
455
456static inline bool is_vector_reg(TCGReg r)
457{
458    return r >= TCG_REG_V0 && r <= TCG_REG_V31;
459}
460
461static bool patch_reloc(tcg_insn_unit *src_rw, int type,
462                        intptr_t value, intptr_t addend)
463{
464    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
465    intptr_t pcrel2;
466    uint32_t old;
467
468    value += addend;
469    pcrel2 = (tcg_insn_unit *)value - src_rx;
470
471    switch (type) {
472    case R_390_PC16DBL:
473        if (pcrel2 == (int16_t)pcrel2) {
474            tcg_patch16(src_rw, pcrel2);
475            return true;
476        }
477        break;
478    case R_390_PC32DBL:
479        if (pcrel2 == (int32_t)pcrel2) {
480            tcg_patch32(src_rw, pcrel2);
481            return true;
482        }
483        break;
484    case R_390_20:
485        if (value == sextract64(value, 0, 20)) {
486            old = *(uint32_t *)src_rw & 0xf00000ff;
487            old |= ((value & 0xfff) << 16) | ((value & 0xff000) >> 4);
488            tcg_patch32(src_rw, old);
489            return true;
490        }
491        break;
492    default:
493        g_assert_not_reached();
494    }
495    return false;
496}
497
498static int is_const_p16(uint64_t val)
499{
500    for (int i = 0; i < 4; ++i) {
501        uint64_t mask = 0xffffull << (i * 16);
502        if ((val & ~mask) == 0) {
503            return i;
504        }
505    }
506    return -1;
507}
508
509static int is_const_p32(uint64_t val)
510{
511    if ((val & 0xffffffff00000000ull) == 0) {
512        return 0;
513    }
514    if ((val & 0x00000000ffffffffull) == 0) {
515        return 1;
516    }
517    return -1;
518}
519
520/*
521 * Accept bit patterns like these:
522 *  0....01....1
523 *  1....10....0
524 *  1..10..01..1
525 *  0..01..10..0
526 * Copied from gcc sources.
527 */
528static bool risbg_mask(uint64_t c)
529{
530    uint64_t lsb;
531    /* We don't change the number of transitions by inverting,
532       so make sure we start with the LSB zero.  */
533    if (c & 1) {
534        c = ~c;
535    }
536    /* Reject all zeros or all ones.  */
537    if (c == 0) {
538        return false;
539    }
540    /* Find the first transition.  */
541    lsb = c & -c;
542    /* Invert to look for a second transition.  */
543    c = ~c;
544    /* Erase the first transition.  */
545    c &= -lsb;
546    /* Find the second transition, if any.  */
547    lsb = c & -c;
548    /* Match if all the bits are 1's, or if c is zero.  */
549    return c == -lsb;
550}
551
552/* Test if a constant matches the constraint. */
553static bool tcg_target_const_match(int64_t val, int ct,
554                                   TCGType type, TCGCond cond, int vece)
555{
556    uint64_t uval = val;
557
558    if (ct & TCG_CT_CONST) {
559        return true;
560    }
561    if (type == TCG_TYPE_I32) {
562        uval = (uint32_t)val;
563        val = (int32_t)val;
564    }
565
566    if (ct & TCG_CT_CONST_CMP) {
567        switch (cond) {
568        case TCG_COND_EQ:
569        case TCG_COND_NE:
570            ct |= TCG_CT_CONST_S32 | TCG_CT_CONST_U32;  /* CGFI or CLGFI */
571            break;
572        case TCG_COND_LT:
573        case TCG_COND_GE:
574        case TCG_COND_LE:
575        case TCG_COND_GT:
576            ct |= TCG_CT_CONST_S32;  /* CGFI */
577            break;
578        case TCG_COND_LTU:
579        case TCG_COND_GEU:
580        case TCG_COND_LEU:
581        case TCG_COND_GTU:
582            ct |= TCG_CT_CONST_U32;  /* CLGFI */
583            break;
584        case TCG_COND_TSTNE:
585        case TCG_COND_TSTEQ:
586            if (is_const_p16(uval) >= 0) {
587                return true;  /* TMxx */
588            }
589            if (risbg_mask(uval)) {
590                return true;  /* RISBG */
591            }
592            break;
593        default:
594            g_assert_not_reached();
595        }
596    }
597
598    if ((ct & TCG_CT_CONST_S32) && val == (int32_t)val) {
599        return true;
600    }
601    if ((ct & TCG_CT_CONST_U32) && val == (uint32_t)val) {
602        return true;
603    }
604    if ((ct & TCG_CT_CONST_S16) && val == (int16_t)val) {
605        return true;
606    }
607    if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
608        return true;
609    }
610
611    if (ct & TCG_CT_CONST_INV) {
612        val = ~val;
613    }
614    if ((ct & TCG_CT_CONST_P32) && is_const_p32(val) >= 0) {
615        return true;
616    }
617    if ((ct & TCG_CT_CONST_INVRISBG) && risbg_mask(~val)) {
618        return true;
619    }
620    return false;
621}
622
623/* Emit instructions according to the given instruction format.  */
624
625static void tcg_out_insn_RR(TCGContext *s, S390Opcode op, TCGReg r1, TCGReg r2)
626{
627    tcg_out16(s, (op << 8) | (r1 << 4) | r2);
628}
629
630static void tcg_out_insn_RRE(TCGContext *s, S390Opcode op,
631                             TCGReg r1, TCGReg r2)
632{
633    tcg_out32(s, (op << 16) | (r1 << 4) | r2);
634}
635
636/* RRF-a without the m4 field */
637static void tcg_out_insn_RRFa(TCGContext *s, S390Opcode op,
638                              TCGReg r1, TCGReg r2, TCGReg r3)
639{
640    tcg_out32(s, (op << 16) | (r3 << 12) | (r1 << 4) | r2);
641}
642
643/* RRF-a with the m4 field */
644static void tcg_out_insn_RRFam(TCGContext *s, S390Opcode op,
645                               TCGReg r1, TCGReg r2, TCGReg r3, int m4)
646{
647    tcg_out32(s, (op << 16) | (r3 << 12) | (m4 << 8) | (r1 << 4) | r2);
648}
649
650static void tcg_out_insn_RRFc(TCGContext *s, S390Opcode op,
651                              TCGReg r1, TCGReg r2, int m3)
652{
653    tcg_out32(s, (op << 16) | (m3 << 12) | (r1 << 4) | r2);
654}
655
656static void tcg_out_insn_RI(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
657{
658    tcg_out32(s, (op << 16) | (r1 << 20) | (i2 & 0xffff));
659}
660
661static void tcg_out_insn_RIEg(TCGContext *s, S390Opcode op, TCGReg r1,
662                             int i2, int m3)
663{
664    tcg_out16(s, (op & 0xff00) | (r1 << 4) | m3);
665    tcg_out32(s, (i2 << 16) | (op & 0xff));
666}
667
668static void tcg_out_insn_RIL(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
669{
670    tcg_out16(s, op | (r1 << 4));
671    tcg_out32(s, i2);
672}
673
674static void tcg_out_insn_RS(TCGContext *s, S390Opcode op, TCGReg r1,
675                            TCGReg b2, TCGReg r3, int disp)
676{
677    tcg_out32(s, (op << 24) | (r1 << 20) | (r3 << 16) | (b2 << 12)
678              | (disp & 0xfff));
679}
680
681static void tcg_out_insn_RSY(TCGContext *s, S390Opcode op, TCGReg r1,
682                             TCGReg b2, TCGReg r3, int disp)
683{
684    tcg_out16(s, (op & 0xff00) | (r1 << 4) | r3);
685    tcg_out32(s, (op & 0xff) | (b2 << 28)
686              | ((disp & 0xfff) << 16) | ((disp & 0xff000) >> 4));
687}
688
689#define tcg_out_insn_RX   tcg_out_insn_RS
690#define tcg_out_insn_RXY  tcg_out_insn_RSY
691
692static int RXB(TCGReg v1, TCGReg v2, TCGReg v3, TCGReg v4)
693{
694    /*
695     * Shift bit 4 of each regno to its corresponding bit of RXB.
696     * RXB itself begins at bit 8 of the instruction so 8 - 4 = 4
697     * is the left-shift of the 4th operand.
698     */
699    return ((v1 & 0x10) << (4 + 3))
700         | ((v2 & 0x10) << (4 + 2))
701         | ((v3 & 0x10) << (4 + 1))
702         | ((v4 & 0x10) << (4 + 0));
703}
704
705static void tcg_out_insn_VRIa(TCGContext *s, S390Opcode op,
706                              TCGReg v1, uint16_t i2, int m3)
707{
708    tcg_debug_assert(is_vector_reg(v1));
709    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4));
710    tcg_out16(s, i2);
711    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m3 << 12));
712}
713
714static void tcg_out_insn_VRIb(TCGContext *s, S390Opcode op,
715                              TCGReg v1, uint8_t i2, uint8_t i3, int m4)
716{
717    tcg_debug_assert(is_vector_reg(v1));
718    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4));
719    tcg_out16(s, (i2 << 8) | (i3 & 0xff));
720    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m4 << 12));
721}
722
723static void tcg_out_insn_VRIc(TCGContext *s, S390Opcode op,
724                              TCGReg v1, uint16_t i2, TCGReg v3, int m4)
725{
726    tcg_debug_assert(is_vector_reg(v1));
727    tcg_debug_assert(is_vector_reg(v3));
728    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v3 & 0xf));
729    tcg_out16(s, i2);
730    tcg_out16(s, (op & 0x00ff) | RXB(v1, v3, 0, 0) | (m4 << 12));
731}
732
733static void tcg_out_insn_VRRa(TCGContext *s, S390Opcode op,
734                              TCGReg v1, TCGReg v2, int m3)
735{
736    tcg_debug_assert(is_vector_reg(v1));
737    tcg_debug_assert(is_vector_reg(v2));
738    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
739    tcg_out32(s, (op & 0x00ff) | RXB(v1, v2, 0, 0) | (m3 << 12));
740}
741
742static void tcg_out_insn_VRRc(TCGContext *s, S390Opcode op,
743                              TCGReg v1, TCGReg v2, TCGReg v3, int m4)
744{
745    tcg_debug_assert(is_vector_reg(v1));
746    tcg_debug_assert(is_vector_reg(v2));
747    tcg_debug_assert(is_vector_reg(v3));
748    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
749    tcg_out16(s, v3 << 12);
750    tcg_out16(s, (op & 0x00ff) | RXB(v1, v2, v3, 0) | (m4 << 12));
751}
752
753static void tcg_out_insn_VRRe(TCGContext *s, S390Opcode op,
754                              TCGReg v1, TCGReg v2, TCGReg v3, TCGReg v4)
755{
756    tcg_debug_assert(is_vector_reg(v1));
757    tcg_debug_assert(is_vector_reg(v2));
758    tcg_debug_assert(is_vector_reg(v3));
759    tcg_debug_assert(is_vector_reg(v4));
760    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
761    tcg_out16(s, v3 << 12);
762    tcg_out16(s, (op & 0x00ff) | RXB(v1, v2, v3, v4) | (v4 << 12));
763}
764
765static void tcg_out_insn_VRRf(TCGContext *s, S390Opcode op,
766                              TCGReg v1, TCGReg r2, TCGReg r3)
767{
768    tcg_debug_assert(is_vector_reg(v1));
769    tcg_debug_assert(is_general_reg(r2));
770    tcg_debug_assert(is_general_reg(r3));
771    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | r2);
772    tcg_out16(s, r3 << 12);
773    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0));
774}
775
776static void tcg_out_insn_VRSa(TCGContext *s, S390Opcode op, TCGReg v1,
777                              intptr_t d2, TCGReg b2, TCGReg v3, int m4)
778{
779    tcg_debug_assert(is_vector_reg(v1));
780    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
781    tcg_debug_assert(is_general_reg(b2));
782    tcg_debug_assert(is_vector_reg(v3));
783    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v3 & 0xf));
784    tcg_out16(s, b2 << 12 | d2);
785    tcg_out16(s, (op & 0x00ff) | RXB(v1, v3, 0, 0) | (m4 << 12));
786}
787
788static void tcg_out_insn_VRSb(TCGContext *s, S390Opcode op, TCGReg v1,
789                              intptr_t d2, TCGReg b2, TCGReg r3, int m4)
790{
791    tcg_debug_assert(is_vector_reg(v1));
792    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
793    tcg_debug_assert(is_general_reg(b2));
794    tcg_debug_assert(is_general_reg(r3));
795    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | r3);
796    tcg_out16(s, b2 << 12 | d2);
797    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m4 << 12));
798}
799
800static void tcg_out_insn_VRSc(TCGContext *s, S390Opcode op, TCGReg r1,
801                              intptr_t d2, TCGReg b2, TCGReg v3, int m4)
802{
803    tcg_debug_assert(is_general_reg(r1));
804    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
805    tcg_debug_assert(is_general_reg(b2));
806    tcg_debug_assert(is_vector_reg(v3));
807    tcg_out16(s, (op & 0xff00) | (r1 << 4) | (v3 & 0xf));
808    tcg_out16(s, b2 << 12 | d2);
809    tcg_out16(s, (op & 0x00ff) | RXB(0, v3, 0, 0) | (m4 << 12));
810}
811
812static void tcg_out_insn_VRX(TCGContext *s, S390Opcode op, TCGReg v1,
813                             TCGReg b2, TCGReg x2, intptr_t d2, int m3)
814{
815    tcg_debug_assert(is_vector_reg(v1));
816    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
817    tcg_debug_assert(is_general_reg(x2));
818    tcg_debug_assert(is_general_reg(b2));
819    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | x2);
820    tcg_out16(s, (b2 << 12) | d2);
821    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m3 << 12));
822}
823
824/* Emit an opcode with "type-checking" of the format.  */
825#define tcg_out_insn(S, FMT, OP, ...) \
826    glue(tcg_out_insn_,FMT)(S, glue(glue(FMT,_),OP), ## __VA_ARGS__)
827
828
829/* emit 64-bit shifts */
830static void tcg_out_sh64(TCGContext* s, S390Opcode op, TCGReg dest,
831                         TCGReg src, TCGReg sh_reg, int sh_imm)
832{
833    tcg_out_insn_RSY(s, op, dest, sh_reg, src, sh_imm);
834}
835
836/* emit 32-bit shifts */
837static void tcg_out_sh32(TCGContext* s, S390Opcode op, TCGReg dest,
838                         TCGReg sh_reg, int sh_imm)
839{
840    tcg_out_insn_RS(s, op, dest, sh_reg, 0, sh_imm);
841}
842
843static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg dst, TCGReg src)
844{
845    if (src == dst) {
846        return true;
847    }
848    switch (type) {
849    case TCG_TYPE_I32:
850        if (likely(is_general_reg(dst) && is_general_reg(src))) {
851            tcg_out_insn(s, RR, LR, dst, src);
852            break;
853        }
854        /* fallthru */
855
856    case TCG_TYPE_I64:
857        if (likely(is_general_reg(dst))) {
858            if (likely(is_general_reg(src))) {
859                tcg_out_insn(s, RRE, LGR, dst, src);
860            } else {
861                tcg_out_insn(s, VRSc, VLGV, dst, 0, 0, src, 3);
862            }
863            break;
864        } else if (is_general_reg(src)) {
865            tcg_out_insn(s, VRSb, VLVG, dst, 0, 0, src, 3);
866            break;
867        }
868        /* fallthru */
869
870    case TCG_TYPE_V64:
871    case TCG_TYPE_V128:
872        tcg_out_insn(s, VRRa, VLR, dst, src, 0);
873        break;
874
875    default:
876        g_assert_not_reached();
877    }
878    return true;
879}
880
881static const S390Opcode li_insns[4] = {
882    RI_LLILL, RI_LLILH, RI_LLIHL, RI_LLIHH
883};
884static const S390Opcode oi_insns[4] = {
885    RI_OILL, RI_OILH, RI_OIHL, RI_OIHH
886};
887static const S390Opcode lif_insns[2] = {
888    RIL_LLILF, RIL_LLIHF,
889};
890static const S390Opcode tm_insns[4] = {
891    RI_TMLL, RI_TMLH, RI_TMHL, RI_TMHH
892};
893
894/* load a register with an immediate value */
895static void tcg_out_movi(TCGContext *s, TCGType type,
896                         TCGReg ret, tcg_target_long sval)
897{
898    tcg_target_ulong uval = sval;
899    ptrdiff_t pc_off;
900    int i;
901
902    if (type == TCG_TYPE_I32) {
903        uval = (uint32_t)sval;
904        sval = (int32_t)sval;
905    }
906
907    /* Try all 32-bit insns that can load it in one go.  */
908    if (sval >= -0x8000 && sval < 0x8000) {
909        tcg_out_insn(s, RI, LGHI, ret, sval);
910        return;
911    }
912
913    i = is_const_p16(uval);
914    if (i >= 0) {
915        tcg_out_insn_RI(s, li_insns[i], ret, uval >> (i * 16));
916        return;
917    }
918
919    /* Try all 48-bit insns that can load it in one go.  */
920    if (sval == (int32_t)sval) {
921        tcg_out_insn(s, RIL, LGFI, ret, sval);
922        return;
923    }
924
925    i = is_const_p32(uval);
926    if (i >= 0) {
927        tcg_out_insn_RIL(s, lif_insns[i], ret, uval >> (i * 32));
928        return;
929    }
930
931    /* Try for PC-relative address load.  For odd addresses, add one. */
932    pc_off = tcg_pcrel_diff(s, (void *)sval) >> 1;
933    if (pc_off == (int32_t)pc_off) {
934        tcg_out_insn(s, RIL, LARL, ret, pc_off);
935        if (sval & 1) {
936            tcg_out_insn(s, RI, AGHI, ret, 1);
937        }
938        return;
939    }
940
941    /* Otherwise, load it by parts. */
942    i = is_const_p16((uint32_t)uval);
943    if (i >= 0) {
944        tcg_out_insn_RI(s, li_insns[i], ret, uval >> (i * 16));
945    } else {
946        tcg_out_insn(s, RIL, LLILF, ret, uval);
947    }
948    uval >>= 32;
949    i = is_const_p16(uval);
950    if (i >= 0) {
951        tcg_out_insn_RI(s, oi_insns[i + 2], ret, uval >> (i * 16));
952    } else {
953        tcg_out_insn(s, RIL, OIHF, ret, uval);
954    }
955}
956
957/* Emit a load/store type instruction.  Inputs are:
958   DATA:     The register to be loaded or stored.
959   BASE+OFS: The effective address.
960   OPC_RX:   If the operation has an RX format opcode (e.g. STC), otherwise 0.
961   OPC_RXY:  The RXY format opcode for the operation (e.g. STCY).  */
962
963static void tcg_out_mem(TCGContext *s, S390Opcode opc_rx, S390Opcode opc_rxy,
964                        TCGReg data, TCGReg base, TCGReg index,
965                        tcg_target_long ofs)
966{
967    if (ofs < -0x80000 || ofs >= 0x80000) {
968        /* Combine the low 20 bits of the offset with the actual load insn;
969           the high 44 bits must come from an immediate load.  */
970        tcg_target_long low = ((ofs & 0xfffff) ^ 0x80000) - 0x80000;
971        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs - low);
972        ofs = low;
973
974        /* If we were already given an index register, add it in.  */
975        if (index != TCG_REG_NONE) {
976            tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
977        }
978        index = TCG_TMP0;
979    }
980
981    if (opc_rx && ofs >= 0 && ofs < 0x1000) {
982        tcg_out_insn_RX(s, opc_rx, data, base, index, ofs);
983    } else {
984        tcg_out_insn_RXY(s, opc_rxy, data, base, index, ofs);
985    }
986}
987
988static void tcg_out_vrx_mem(TCGContext *s, S390Opcode opc_vrx,
989                            TCGReg data, TCGReg base, TCGReg index,
990                            tcg_target_long ofs, int m3)
991{
992    if (ofs < 0 || ofs >= 0x1000) {
993        if (ofs >= -0x80000 && ofs < 0x80000) {
994            tcg_out_insn(s, RXY, LAY, TCG_TMP0, base, index, ofs);
995            base = TCG_TMP0;
996            index = TCG_REG_NONE;
997            ofs = 0;
998        } else {
999            tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs);
1000            if (index != TCG_REG_NONE) {
1001                tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
1002            }
1003            index = TCG_TMP0;
1004            ofs = 0;
1005        }
1006    }
1007    tcg_out_insn_VRX(s, opc_vrx, data, base, index, ofs, m3);
1008}
1009
1010/* load data without address translation or endianness conversion */
1011static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg data,
1012                       TCGReg base, intptr_t ofs)
1013{
1014    switch (type) {
1015    case TCG_TYPE_I32:
1016        if (likely(is_general_reg(data))) {
1017            tcg_out_mem(s, RX_L, RXY_LY, data, base, TCG_REG_NONE, ofs);
1018            break;
1019        }
1020        tcg_out_vrx_mem(s, VRX_VLLEZ, data, base, TCG_REG_NONE, ofs, MO_32);
1021        break;
1022
1023    case TCG_TYPE_I64:
1024        if (likely(is_general_reg(data))) {
1025            tcg_out_mem(s, 0, RXY_LG, data, base, TCG_REG_NONE, ofs);
1026            break;
1027        }
1028        /* fallthru */
1029
1030    case TCG_TYPE_V64:
1031        tcg_out_vrx_mem(s, VRX_VLLEZ, data, base, TCG_REG_NONE, ofs, MO_64);
1032        break;
1033
1034    case TCG_TYPE_V128:
1035        /* Hint quadword aligned.  */
1036        tcg_out_vrx_mem(s, VRX_VL, data, base, TCG_REG_NONE, ofs, 4);
1037        break;
1038
1039    default:
1040        g_assert_not_reached();
1041    }
1042}
1043
1044static void tcg_out_st(TCGContext *s, TCGType type, TCGReg data,
1045                       TCGReg base, intptr_t ofs)
1046{
1047    switch (type) {
1048    case TCG_TYPE_I32:
1049        if (likely(is_general_reg(data))) {
1050            tcg_out_mem(s, RX_ST, RXY_STY, data, base, TCG_REG_NONE, ofs);
1051        } else {
1052            tcg_out_vrx_mem(s, VRX_VSTEF, data, base, TCG_REG_NONE, ofs, 1);
1053        }
1054        break;
1055
1056    case TCG_TYPE_I64:
1057        if (likely(is_general_reg(data))) {
1058            tcg_out_mem(s, 0, RXY_STG, data, base, TCG_REG_NONE, ofs);
1059            break;
1060        }
1061        /* fallthru */
1062
1063    case TCG_TYPE_V64:
1064        tcg_out_vrx_mem(s, VRX_VSTEG, data, base, TCG_REG_NONE, ofs, 0);
1065        break;
1066
1067    case TCG_TYPE_V128:
1068        /* Hint quadword aligned.  */
1069        tcg_out_vrx_mem(s, VRX_VST, data, base, TCG_REG_NONE, ofs, 4);
1070        break;
1071
1072    default:
1073        g_assert_not_reached();
1074    }
1075}
1076
1077static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
1078                               TCGReg base, intptr_t ofs)
1079{
1080    return false;
1081}
1082
1083static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
1084{
1085    return false;
1086}
1087
1088static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
1089                             tcg_target_long imm)
1090{
1091    /* This function is only used for passing structs by reference. */
1092    tcg_out_mem(s, RX_LA, RXY_LAY, rd, rs, TCG_REG_NONE, imm);
1093}
1094
1095static inline void tcg_out_risbg(TCGContext *s, TCGReg dest, TCGReg src,
1096                                 int msb, int lsb, int ofs, int z)
1097{
1098    /* Format RIE-f */
1099    tcg_out16(s, (RIEf_RISBG & 0xff00) | (dest << 4) | src);
1100    tcg_out16(s, (msb << 8) | (z << 7) | lsb);
1101    tcg_out16(s, (ofs << 8) | (RIEf_RISBG & 0xff));
1102}
1103
1104static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1105{
1106    tcg_out_insn(s, RRE, LGBR, dest, src);
1107}
1108
1109static void tcg_out_ext8u(TCGContext *s, TCGReg dest, TCGReg src)
1110{
1111    tcg_out_insn(s, RRE, LLGCR, dest, src);
1112}
1113
1114static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1115{
1116    tcg_out_insn(s, RRE, LGHR, dest, src);
1117}
1118
1119static void tcg_out_ext16u(TCGContext *s, TCGReg dest, TCGReg src)
1120{
1121    tcg_out_insn(s, RRE, LLGHR, dest, src);
1122}
1123
1124static void tcg_out_ext32s(TCGContext *s, TCGReg dest, TCGReg src)
1125{
1126    tcg_out_insn(s, RRE, LGFR, dest, src);
1127}
1128
1129static void tcg_out_ext32u(TCGContext *s, TCGReg dest, TCGReg src)
1130{
1131    tcg_out_insn(s, RRE, LLGFR, dest, src);
1132}
1133
1134static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg dest, TCGReg src)
1135{
1136    tcg_out_ext32s(s, dest, src);
1137}
1138
1139static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg dest, TCGReg src)
1140{
1141    tcg_out_ext32u(s, dest, src);
1142}
1143
1144static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg dest, TCGReg src)
1145{
1146    tcg_out_mov(s, TCG_TYPE_I32, dest, src);
1147}
1148
1149static void tgen_andi_risbg(TCGContext *s, TCGReg out, TCGReg in, uint64_t val)
1150{
1151    int msb, lsb;
1152    if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
1153        /* Achieve wraparound by swapping msb and lsb.  */
1154        msb = 64 - ctz64(~val);
1155        lsb = clz64(~val) - 1;
1156    } else {
1157        msb = clz64(val);
1158        lsb = 63 - ctz64(val);
1159    }
1160    tcg_out_risbg(s, out, in, msb, lsb, 0, 1);
1161}
1162
1163static void tgen_andi(TCGContext *s, TCGType type, TCGReg dest, uint64_t val)
1164{
1165    static const S390Opcode ni_insns[4] = {
1166        RI_NILL, RI_NILH, RI_NIHL, RI_NIHH
1167    };
1168    static const S390Opcode nif_insns[2] = {
1169        RIL_NILF, RIL_NIHF
1170    };
1171    uint64_t valid = (type == TCG_TYPE_I32 ? 0xffffffffull : -1ull);
1172    int i;
1173
1174    /* Look for the zero-extensions.  */
1175    if ((val & valid) == 0xffffffff) {
1176        tcg_out_ext32u(s, dest, dest);
1177        return;
1178    }
1179    if ((val & valid) == 0xff) {
1180        tcg_out_ext8u(s, dest, dest);
1181        return;
1182    }
1183    if ((val & valid) == 0xffff) {
1184        tcg_out_ext16u(s, dest, dest);
1185        return;
1186    }
1187
1188    i = is_const_p16(~val & valid);
1189    if (i >= 0) {
1190        tcg_out_insn_RI(s, ni_insns[i], dest, val >> (i * 16));
1191        return;
1192    }
1193
1194    i = is_const_p32(~val & valid);
1195    tcg_debug_assert(i == 0 || type != TCG_TYPE_I32);
1196    if (i >= 0) {
1197        tcg_out_insn_RIL(s, nif_insns[i], dest, val >> (i * 32));
1198        return;
1199    }
1200
1201    if (risbg_mask(val)) {
1202        tgen_andi_risbg(s, dest, dest, val);
1203        return;
1204    }
1205
1206    g_assert_not_reached();
1207}
1208
1209static void tgen_ori(TCGContext *s, TCGReg dest, uint64_t val)
1210{
1211    static const S390Opcode oif_insns[2] = {
1212        RIL_OILF, RIL_OIHF
1213    };
1214
1215    int i;
1216
1217    i = is_const_p16(val);
1218    if (i >= 0) {
1219        tcg_out_insn_RI(s, oi_insns[i], dest, val >> (i * 16));
1220        return;
1221    }
1222
1223    i = is_const_p32(val);
1224    if (i >= 0) {
1225        tcg_out_insn_RIL(s, oif_insns[i], dest, val >> (i * 32));
1226        return;
1227    }
1228
1229    g_assert_not_reached();
1230}
1231
1232static void tgen_xori(TCGContext *s, TCGReg dest, uint64_t val)
1233{
1234    switch (is_const_p32(val)) {
1235    case 0:
1236        tcg_out_insn(s, RIL, XILF, dest, val);
1237        break;
1238    case 1:
1239        tcg_out_insn(s, RIL, XIHF, dest, val >> 32);
1240        break;
1241    default:
1242        g_assert_not_reached();
1243    }
1244}
1245
1246static int tgen_cmp2(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1247                     TCGArg c2, bool c2const, bool need_carry, int *inv_cc)
1248{
1249    bool is_unsigned = is_unsigned_cond(c);
1250    TCGCond inv_c = tcg_invert_cond(c);
1251    S390Opcode op;
1252
1253    if (is_tst_cond(c)) {
1254        tcg_debug_assert(!need_carry);
1255
1256        if (!c2const) {
1257            if (type == TCG_TYPE_I32) {
1258                tcg_out_insn(s, RRFa, NRK, TCG_REG_R0, r1, c2);
1259            } else {
1260                tcg_out_insn(s, RRFa, NGRK, TCG_REG_R0, r1, c2);
1261            }
1262            goto exit;
1263        }
1264
1265        if (type == TCG_TYPE_I32) {
1266            c2 = (uint32_t)c2;
1267        }
1268
1269        int i = is_const_p16(c2);
1270        if (i >= 0) {
1271            tcg_out_insn_RI(s, tm_insns[i], r1, c2 >> (i * 16));
1272            *inv_cc = c == TCG_COND_TSTEQ ? S390_TM_NE : S390_TM_EQ;
1273            return *inv_cc ^ 15;
1274        }
1275
1276        if (risbg_mask(c2)) {
1277            tgen_andi_risbg(s, TCG_REG_R0, r1, c2);
1278            goto exit;
1279        }
1280        g_assert_not_reached();
1281    }
1282
1283    if (c2const) {
1284        if (c2 == 0) {
1285            if (!(is_unsigned && need_carry)) {
1286                if (type == TCG_TYPE_I32) {
1287                    tcg_out_insn(s, RR, LTR, r1, r1);
1288                } else {
1289                    tcg_out_insn(s, RRE, LTGR, r1, r1);
1290                }
1291                *inv_cc = tcg_cond_to_ltr_cond[inv_c];
1292                return tcg_cond_to_ltr_cond[c];
1293            }
1294        }
1295
1296        if (!is_unsigned && c2 == (int16_t)c2) {
1297            op = (type == TCG_TYPE_I32 ? RI_CHI : RI_CGHI);
1298            tcg_out_insn_RI(s, op, r1, c2);
1299            goto exit;
1300        }
1301
1302        if (type == TCG_TYPE_I32) {
1303            op = (is_unsigned ? RIL_CLFI : RIL_CFI);
1304            tcg_out_insn_RIL(s, op, r1, c2);
1305            goto exit;
1306        }
1307
1308        /* Should match TCG_CT_CONST_CMP. */
1309        switch (c) {
1310        case TCG_COND_LT:
1311        case TCG_COND_GE:
1312        case TCG_COND_LE:
1313        case TCG_COND_GT:
1314            tcg_debug_assert(c2 == (int32_t)c2);
1315            op = RIL_CGFI;
1316            break;
1317        case TCG_COND_EQ:
1318        case TCG_COND_NE:
1319            if (c2 == (int32_t)c2) {
1320                op = RIL_CGFI;
1321                break;
1322            }
1323            /* fall through */
1324        case TCG_COND_LTU:
1325        case TCG_COND_GEU:
1326        case TCG_COND_LEU:
1327        case TCG_COND_GTU:
1328            tcg_debug_assert(c2 == (uint32_t)c2);
1329            op = RIL_CLGFI;
1330            break;
1331        default:
1332            g_assert_not_reached();
1333        }
1334        tcg_out_insn_RIL(s, op, r1, c2);
1335    } else if (type == TCG_TYPE_I32) {
1336        op = (is_unsigned ? RR_CLR : RR_CR);
1337        tcg_out_insn_RR(s, op, r1, c2);
1338    } else {
1339        op = (is_unsigned ? RRE_CLGR : RRE_CGR);
1340        tcg_out_insn_RRE(s, op, r1, c2);
1341    }
1342
1343 exit:
1344    *inv_cc = tcg_cond_to_s390_cond[inv_c];
1345    return tcg_cond_to_s390_cond[c];
1346}
1347
1348static int tgen_cmp(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1349                    TCGArg c2, bool c2const, bool need_carry)
1350{
1351    int inv_cc;
1352    return tgen_cmp2(s, type, c, r1, c2, c2const, need_carry, &inv_cc);
1353}
1354
1355static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond,
1356                         TCGReg dest, TCGReg c1, TCGArg c2,
1357                         bool c2const, bool neg)
1358{
1359    int cc;
1360
1361    /* With LOC2, we can always emit the minimum 3 insns.  */
1362    if (HAVE_FACILITY(LOAD_ON_COND2)) {
1363        /* Emit: d = 0, d = (cc ? 1 : d).  */
1364        cc = tgen_cmp(s, type, cond, c1, c2, c2const, false);
1365        tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1366        tcg_out_insn(s, RIEg, LOCGHI, dest, neg ? -1 : 1, cc);
1367        return;
1368    }
1369
1370    switch (cond) {
1371    case TCG_COND_GEU:
1372    case TCG_COND_LTU:
1373    case TCG_COND_LT:
1374    case TCG_COND_GE:
1375        /* Swap operands so that we can use LEU/GTU/GT/LE.  */
1376        if (!c2const) {
1377            TCGReg t = c1;
1378            c1 = c2;
1379            c2 = t;
1380            cond = tcg_swap_cond(cond);
1381        }
1382        break;
1383    default:
1384        break;
1385    }
1386
1387    switch (cond) {
1388    case TCG_COND_NE:
1389        /* X != 0 is X > 0.  */
1390        if (c2const && c2 == 0) {
1391            cond = TCG_COND_GTU;
1392        } else {
1393            break;
1394        }
1395        /* fallthru */
1396
1397    case TCG_COND_GTU:
1398    case TCG_COND_GT:
1399        /*
1400         * The result of a compare has CC=2 for GT and CC=3 unused.
1401         * ADD LOGICAL WITH CARRY considers (CC & 2) the carry bit.
1402         */
1403        tgen_cmp(s, type, cond, c1, c2, c2const, true);
1404        tcg_out_movi(s, type, dest, 0);
1405        tcg_out_insn(s, RRE, ALCGR, dest, dest);
1406        if (neg) {
1407            if (type == TCG_TYPE_I32) {
1408                tcg_out_insn(s, RR, LCR, dest, dest);
1409            } else {
1410                tcg_out_insn(s, RRE, LCGR, dest, dest);
1411            }
1412        }
1413        return;
1414
1415    case TCG_COND_EQ:
1416        /* X == 0 is X <= 0.  */
1417        if (c2const && c2 == 0) {
1418            cond = TCG_COND_LEU;
1419        } else {
1420            break;
1421        }
1422        /* fallthru */
1423
1424    case TCG_COND_LEU:
1425    case TCG_COND_LE:
1426        /*
1427         * As above, but we're looking for borrow, or !carry.
1428         * The second insn computes d - d - borrow, or -1 for true
1429         * and 0 for false.  So we must mask to 1 bit afterward.
1430         */
1431        tgen_cmp(s, type, cond, c1, c2, c2const, true);
1432        tcg_out_insn(s, RRE, SLBGR, dest, dest);
1433        if (!neg) {
1434            tgen_andi(s, type, dest, 1);
1435        }
1436        return;
1437
1438    default:
1439        g_assert_not_reached();
1440    }
1441
1442    cc = tgen_cmp(s, type, cond, c1, c2, c2const, false);
1443    /* Emit: d = 0, t = 1, d = (cc ? t : d).  */
1444    tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1445    tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, neg ? -1 : 1);
1446    tcg_out_insn(s, RRFc, LOCGR, dest, TCG_TMP0, cc);
1447}
1448
1449static void tgen_movcond_int(TCGContext *s, TCGType type, TCGReg dest,
1450                             TCGArg v3, int v3const, TCGReg v4,
1451                             int cc, int inv_cc)
1452{
1453    TCGReg src;
1454
1455    if (v3const) {
1456        if (dest == v4) {
1457            if (HAVE_FACILITY(LOAD_ON_COND2)) {
1458                /* Emit: if (cc) dest = v3. */
1459                tcg_out_insn(s, RIEg, LOCGHI, dest, v3, cc);
1460                return;
1461            }
1462            tcg_out_insn(s, RI, LGHI, TCG_TMP0, v3);
1463            src = TCG_TMP0;
1464        } else {
1465            /* LGR+LOCGHI is larger than LGHI+LOCGR. */
1466            tcg_out_insn(s, RI, LGHI, dest, v3);
1467            cc = inv_cc;
1468            src = v4;
1469        }
1470    } else {
1471        if (HAVE_FACILITY(MISC_INSN_EXT3)) {
1472            /* Emit: dest = cc ? v3 : v4. */
1473            tcg_out_insn(s, RRFam, SELGR, dest, v3, v4, cc);
1474            return;
1475        }
1476        if (dest == v4) {
1477            src = v3;
1478        } else {
1479            tcg_out_mov(s, type, dest, v3);
1480            cc = inv_cc;
1481            src = v4;
1482        }
1483    }
1484
1485    /* Emit: if (cc) dest = src. */
1486    tcg_out_insn(s, RRFc, LOCGR, dest, src, cc);
1487}
1488
1489static void tgen_movcond(TCGContext *s, TCGType type, TCGCond c, TCGReg dest,
1490                         TCGReg c1, TCGArg c2, int c2const,
1491                         TCGArg v3, int v3const, TCGReg v4)
1492{
1493    int cc, inv_cc;
1494
1495    cc = tgen_cmp2(s, type, c, c1, c2, c2const, false, &inv_cc);
1496    tgen_movcond_int(s, type, dest, v3, v3const, v4, cc, inv_cc);
1497}
1498
1499static void tgen_clz(TCGContext *s, TCGReg dest, TCGReg a1,
1500                     TCGArg a2, int a2const)
1501{
1502    /* Since this sets both R and R+1, we have no choice but to store the
1503       result into R0, allowing R1 == TCG_TMP0 to be clobbered as well.  */
1504    QEMU_BUILD_BUG_ON(TCG_TMP0 != TCG_REG_R1);
1505    tcg_out_insn(s, RRE, FLOGR, TCG_REG_R0, a1);
1506
1507    if (a2const && a2 == 64) {
1508        tcg_out_mov(s, TCG_TYPE_I64, dest, TCG_REG_R0);
1509        return;
1510    }
1511
1512    /*
1513     * Conditions from FLOGR are:
1514     *   2 -> one bit found
1515     *   8 -> no one bit found
1516     */
1517    tgen_movcond_int(s, TCG_TYPE_I64, dest, a2, a2const, TCG_REG_R0, 8, 2);
1518}
1519
1520static void tgen_ctpop(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1521{
1522    /* With MIE3, and bit 0 of m4 set, we get the complete result. */
1523    if (HAVE_FACILITY(MISC_INSN_EXT3)) {
1524        if (type == TCG_TYPE_I32) {
1525            tcg_out_ext32u(s, dest, src);
1526            src = dest;
1527        }
1528        tcg_out_insn(s, RRFc, POPCNT, dest, src, 8);
1529        return;
1530    }
1531
1532    /* Without MIE3, each byte gets the count of bits for the byte. */
1533    tcg_out_insn(s, RRFc, POPCNT, dest, src, 0);
1534
1535    /* Multiply to sum each byte at the top of the word. */
1536    if (type == TCG_TYPE_I32) {
1537        tcg_out_insn(s, RIL, MSFI, dest, 0x01010101);
1538        tcg_out_sh32(s, RS_SRL, dest, TCG_REG_NONE, 24);
1539    } else {
1540        tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, 0x0101010101010101ull);
1541        tcg_out_insn(s, RRE, MSGR, dest, TCG_TMP0);
1542        tcg_out_sh64(s, RSY_SRLG, dest, dest, TCG_REG_NONE, 56);
1543    }
1544}
1545
1546static void tgen_deposit(TCGContext *s, TCGReg dest, TCGReg src,
1547                         int ofs, int len, int z)
1548{
1549    int lsb = (63 - ofs);
1550    int msb = lsb - (len - 1);
1551    tcg_out_risbg(s, dest, src, msb, lsb, ofs, z);
1552}
1553
1554static void tgen_extract(TCGContext *s, TCGReg dest, TCGReg src,
1555                         int ofs, int len)
1556{
1557    tcg_out_risbg(s, dest, src, 64 - len, 63, 64 - ofs, 1);
1558}
1559
1560static void tgen_gotoi(TCGContext *s, int cc, const tcg_insn_unit *dest)
1561{
1562    ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1;
1563    if (off == (int16_t)off) {
1564        tcg_out_insn(s, RI, BRC, cc, off);
1565    } else if (off == (int32_t)off) {
1566        tcg_out_insn(s, RIL, BRCL, cc, off);
1567    } else {
1568        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1569        tcg_out_insn(s, RR, BCR, cc, TCG_TMP0);
1570    }
1571}
1572
1573static void tgen_branch(TCGContext *s, int cc, TCGLabel *l)
1574{
1575    if (l->has_value) {
1576        tgen_gotoi(s, cc, l->u.value_ptr);
1577    } else {
1578        tcg_out16(s, RI_BRC | (cc << 4));
1579        tcg_out_reloc(s, s->code_ptr, R_390_PC16DBL, l, 2);
1580        s->code_ptr += 1;
1581    }
1582}
1583
1584static void tgen_compare_branch(TCGContext *s, S390Opcode opc, int cc,
1585                                TCGReg r1, TCGReg r2, TCGLabel *l)
1586{
1587    tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1588    /* Format RIE-b */
1589    tcg_out16(s, (opc & 0xff00) | (r1 << 4) | r2);
1590    tcg_out16(s, 0);
1591    tcg_out16(s, cc << 12 | (opc & 0xff));
1592}
1593
1594static void tgen_compare_imm_branch(TCGContext *s, S390Opcode opc, int cc,
1595                                    TCGReg r1, int i2, TCGLabel *l)
1596{
1597    tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1598    /* Format RIE-c */
1599    tcg_out16(s, (opc & 0xff00) | (r1 << 4) | cc);
1600    tcg_out16(s, 0);
1601    tcg_out16(s, (i2 << 8) | (opc & 0xff));
1602}
1603
1604static void tgen_brcond(TCGContext *s, TCGType type, TCGCond c,
1605                        TCGReg r1, TCGArg c2, int c2const, TCGLabel *l)
1606{
1607    int cc;
1608
1609    if (!is_tst_cond(c)) {
1610        bool is_unsigned = is_unsigned_cond(c);
1611        bool in_range;
1612        S390Opcode opc;
1613
1614        cc = tcg_cond_to_s390_cond[c];
1615
1616        if (!c2const) {
1617            opc = (type == TCG_TYPE_I32
1618                   ? (is_unsigned ? RIEb_CLRJ : RIEb_CRJ)
1619                   : (is_unsigned ? RIEb_CLGRJ : RIEb_CGRJ));
1620            tgen_compare_branch(s, opc, cc, r1, c2, l);
1621            return;
1622        }
1623
1624        /*
1625         * COMPARE IMMEDIATE AND BRANCH RELATIVE has an 8-bit immediate field.
1626         * If the immediate we've been given does not fit that range, we'll
1627         * fall back to separate compare and branch instructions using the
1628         * larger comparison range afforded by COMPARE IMMEDIATE.
1629         */
1630        if (type == TCG_TYPE_I32) {
1631            if (is_unsigned) {
1632                opc = RIEc_CLIJ;
1633                in_range = (uint32_t)c2 == (uint8_t)c2;
1634            } else {
1635                opc = RIEc_CIJ;
1636                in_range = (int32_t)c2 == (int8_t)c2;
1637            }
1638        } else {
1639            if (is_unsigned) {
1640                opc = RIEc_CLGIJ;
1641                in_range = (uint64_t)c2 == (uint8_t)c2;
1642            } else {
1643                opc = RIEc_CGIJ;
1644                in_range = (int64_t)c2 == (int8_t)c2;
1645            }
1646        }
1647        if (in_range) {
1648            tgen_compare_imm_branch(s, opc, cc, r1, c2, l);
1649            return;
1650        }
1651    }
1652
1653    cc = tgen_cmp(s, type, c, r1, c2, c2const, false);
1654    tgen_branch(s, cc, l);
1655}
1656
1657static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *dest)
1658{
1659    ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1;
1660    if (off == (int32_t)off) {
1661        tcg_out_insn(s, RIL, BRASL, TCG_REG_R14, off);
1662    } else {
1663        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1664        tcg_out_insn(s, RR, BASR, TCG_REG_R14, TCG_TMP0);
1665    }
1666}
1667
1668static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest,
1669                         const TCGHelperInfo *info)
1670{
1671    tcg_out_call_int(s, dest);
1672}
1673
1674typedef struct {
1675    TCGReg base;
1676    TCGReg index;
1677    int disp;
1678    TCGAtomAlign aa;
1679} HostAddress;
1680
1681bool tcg_target_has_memory_bswap(MemOp memop)
1682{
1683    TCGAtomAlign aa;
1684
1685    if ((memop & MO_SIZE) <= MO_64) {
1686        return true;
1687    }
1688
1689    /*
1690     * Reject 16-byte memop with 16-byte atomicity,
1691     * but do allow a pair of 64-bit operations.
1692     */
1693    aa = atom_and_align_for_opc(tcg_ctx, memop, MO_ATOM_IFALIGN, true);
1694    return aa.atom <= MO_64;
1695}
1696
1697static void tcg_out_qemu_ld_direct(TCGContext *s, MemOp opc, TCGReg data,
1698                                   HostAddress h)
1699{
1700    switch (opc & (MO_SSIZE | MO_BSWAP)) {
1701    case MO_UB:
1702        tcg_out_insn(s, RXY, LLGC, data, h.base, h.index, h.disp);
1703        break;
1704    case MO_SB:
1705        tcg_out_insn(s, RXY, LGB, data, h.base, h.index, h.disp);
1706        break;
1707
1708    case MO_UW | MO_BSWAP:
1709        /* swapped unsigned halfword load with upper bits zeroed */
1710        tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp);
1711        tcg_out_ext16u(s, data, data);
1712        break;
1713    case MO_UW:
1714        tcg_out_insn(s, RXY, LLGH, data, h.base, h.index, h.disp);
1715        break;
1716
1717    case MO_SW | MO_BSWAP:
1718        /* swapped sign-extended halfword load */
1719        tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp);
1720        tcg_out_ext16s(s, TCG_TYPE_REG, data, data);
1721        break;
1722    case MO_SW:
1723        tcg_out_insn(s, RXY, LGH, data, h.base, h.index, h.disp);
1724        break;
1725
1726    case MO_UL | MO_BSWAP:
1727        /* swapped unsigned int load with upper bits zeroed */
1728        tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp);
1729        tcg_out_ext32u(s, data, data);
1730        break;
1731    case MO_UL:
1732        tcg_out_insn(s, RXY, LLGF, data, h.base, h.index, h.disp);
1733        break;
1734
1735    case MO_SL | MO_BSWAP:
1736        /* swapped sign-extended int load */
1737        tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp);
1738        tcg_out_ext32s(s, data, data);
1739        break;
1740    case MO_SL:
1741        tcg_out_insn(s, RXY, LGF, data, h.base, h.index, h.disp);
1742        break;
1743
1744    case MO_UQ | MO_BSWAP:
1745        tcg_out_insn(s, RXY, LRVG, data, h.base, h.index, h.disp);
1746        break;
1747    case MO_UQ:
1748        tcg_out_insn(s, RXY, LG, data, h.base, h.index, h.disp);
1749        break;
1750
1751    default:
1752        g_assert_not_reached();
1753    }
1754}
1755
1756static void tcg_out_qemu_st_direct(TCGContext *s, MemOp opc, TCGReg data,
1757                                   HostAddress h)
1758{
1759    switch (opc & (MO_SIZE | MO_BSWAP)) {
1760    case MO_UB:
1761        if (h.disp >= 0 && h.disp < 0x1000) {
1762            tcg_out_insn(s, RX, STC, data, h.base, h.index, h.disp);
1763        } else {
1764            tcg_out_insn(s, RXY, STCY, data, h.base, h.index, h.disp);
1765        }
1766        break;
1767
1768    case MO_UW | MO_BSWAP:
1769        tcg_out_insn(s, RXY, STRVH, data, h.base, h.index, h.disp);
1770        break;
1771    case MO_UW:
1772        if (h.disp >= 0 && h.disp < 0x1000) {
1773            tcg_out_insn(s, RX, STH, data, h.base, h.index, h.disp);
1774        } else {
1775            tcg_out_insn(s, RXY, STHY, data, h.base, h.index, h.disp);
1776        }
1777        break;
1778
1779    case MO_UL | MO_BSWAP:
1780        tcg_out_insn(s, RXY, STRV, data, h.base, h.index, h.disp);
1781        break;
1782    case MO_UL:
1783        if (h.disp >= 0 && h.disp < 0x1000) {
1784            tcg_out_insn(s, RX, ST, data, h.base, h.index, h.disp);
1785        } else {
1786            tcg_out_insn(s, RXY, STY, data, h.base, h.index, h.disp);
1787        }
1788        break;
1789
1790    case MO_UQ | MO_BSWAP:
1791        tcg_out_insn(s, RXY, STRVG, data, h.base, h.index, h.disp);
1792        break;
1793    case MO_UQ:
1794        tcg_out_insn(s, RXY, STG, data, h.base, h.index, h.disp);
1795        break;
1796
1797    default:
1798        g_assert_not_reached();
1799    }
1800}
1801
1802static const TCGLdstHelperParam ldst_helper_param = {
1803    .ntmp = 1, .tmp = { TCG_TMP0 }
1804};
1805
1806static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1807{
1808    MemOp opc = get_memop(lb->oi);
1809
1810    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1811                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) {
1812        return false;
1813    }
1814
1815    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
1816    tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SIZE]);
1817    tcg_out_ld_helper_ret(s, lb, false, &ldst_helper_param);
1818
1819    tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1820    return true;
1821}
1822
1823static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1824{
1825    MemOp opc = get_memop(lb->oi);
1826
1827    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1828                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) {
1829        return false;
1830    }
1831
1832    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
1833    tcg_out_call_int(s, qemu_st_helpers[opc & MO_SIZE]);
1834
1835    tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1836    return true;
1837}
1838
1839/* We're expecting to use a 20-bit negative offset on the tlb memory ops.  */
1840#define MIN_TLB_MASK_TABLE_OFS  -(1 << 19)
1841
1842/*
1843 * For system-mode, perform the TLB load and compare.
1844 * For user-mode, perform any required alignment tests.
1845 * In both cases, return a TCGLabelQemuLdst structure if the slow path
1846 * is required and fill in @h with the host address for the fast path.
1847 */
1848static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
1849                                           TCGReg addr_reg, MemOpIdx oi,
1850                                           bool is_ld)
1851{
1852    TCGType addr_type = s->addr_type;
1853    TCGLabelQemuLdst *ldst = NULL;
1854    MemOp opc = get_memop(oi);
1855    MemOp s_bits = opc & MO_SIZE;
1856    unsigned a_mask;
1857
1858    h->aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, s_bits == MO_128);
1859    a_mask = (1 << h->aa.align) - 1;
1860
1861    if (tcg_use_softmmu) {
1862        unsigned s_mask = (1 << s_bits) - 1;
1863        int mem_index = get_mmuidx(oi);
1864        int fast_off = tlb_mask_table_ofs(s, mem_index);
1865        int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1866        int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1867        int ofs, a_off;
1868        uint64_t tlb_mask;
1869
1870        ldst = new_ldst_label(s);
1871        ldst->is_ld = is_ld;
1872        ldst->oi = oi;
1873        ldst->addrlo_reg = addr_reg;
1874
1875        tcg_out_sh64(s, RSY_SRLG, TCG_TMP0, addr_reg, TCG_REG_NONE,
1876                     s->page_bits - CPU_TLB_ENTRY_BITS);
1877
1878        tcg_out_insn(s, RXY, NG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, mask_off);
1879        tcg_out_insn(s, RXY, AG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, table_off);
1880
1881        /*
1882         * For aligned accesses, we check the first byte and include the
1883         * alignment bits within the address.  For unaligned access, we
1884         * check that we don't cross pages using the address of the last
1885         * byte of the access.
1886         */
1887        a_off = (a_mask >= s_mask ? 0 : s_mask - a_mask);
1888        tlb_mask = (uint64_t)s->page_mask | a_mask;
1889        if (a_off == 0) {
1890            tgen_andi_risbg(s, TCG_REG_R0, addr_reg, tlb_mask);
1891        } else {
1892            tcg_out_insn(s, RX, LA, TCG_REG_R0, addr_reg, TCG_REG_NONE, a_off);
1893            tgen_andi(s, addr_type, TCG_REG_R0, tlb_mask);
1894        }
1895
1896        if (is_ld) {
1897            ofs = offsetof(CPUTLBEntry, addr_read);
1898        } else {
1899            ofs = offsetof(CPUTLBEntry, addr_write);
1900        }
1901        if (addr_type == TCG_TYPE_I32) {
1902            ofs += HOST_BIG_ENDIAN * 4;
1903            tcg_out_insn(s, RX, C, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs);
1904        } else {
1905            tcg_out_insn(s, RXY, CG, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs);
1906        }
1907
1908        tcg_out16(s, RI_BRC | (S390_CC_NE << 4));
1909        ldst->label_ptr[0] = s->code_ptr++;
1910
1911        h->index = TCG_TMP0;
1912        tcg_out_insn(s, RXY, LG, h->index, TCG_TMP0, TCG_REG_NONE,
1913                     offsetof(CPUTLBEntry, addend));
1914
1915        if (addr_type == TCG_TYPE_I32) {
1916            tcg_out_insn(s, RRE, ALGFR, h->index, addr_reg);
1917            h->base = TCG_REG_NONE;
1918        } else {
1919            h->base = addr_reg;
1920        }
1921        h->disp = 0;
1922    } else {
1923        if (a_mask) {
1924            ldst = new_ldst_label(s);
1925            ldst->is_ld = is_ld;
1926            ldst->oi = oi;
1927            ldst->addrlo_reg = addr_reg;
1928
1929            tcg_debug_assert(a_mask <= 0xffff);
1930            tcg_out_insn(s, RI, TMLL, addr_reg, a_mask);
1931
1932            tcg_out16(s, RI_BRC | (S390_TM_NE << 4));
1933            ldst->label_ptr[0] = s->code_ptr++;
1934        }
1935
1936        h->base = addr_reg;
1937        if (addr_type == TCG_TYPE_I32) {
1938            tcg_out_ext32u(s, TCG_TMP0, addr_reg);
1939            h->base = TCG_TMP0;
1940        }
1941        if (guest_base < 0x80000) {
1942            h->index = TCG_REG_NONE;
1943            h->disp = guest_base;
1944        } else {
1945            h->index = TCG_GUEST_BASE_REG;
1946            h->disp = 0;
1947        }
1948    }
1949
1950    return ldst;
1951}
1952
1953static void tcg_out_qemu_ld(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
1954                            MemOpIdx oi, TCGType data_type)
1955{
1956    TCGLabelQemuLdst *ldst;
1957    HostAddress h;
1958
1959    ldst = prepare_host_addr(s, &h, addr_reg, oi, true);
1960    tcg_out_qemu_ld_direct(s, get_memop(oi), data_reg, h);
1961
1962    if (ldst) {
1963        ldst->type = data_type;
1964        ldst->datalo_reg = data_reg;
1965        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1966    }
1967}
1968
1969static void tcg_out_qemu_st(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
1970                            MemOpIdx oi, TCGType data_type)
1971{
1972    TCGLabelQemuLdst *ldst;
1973    HostAddress h;
1974
1975    ldst = prepare_host_addr(s, &h, addr_reg, oi, false);
1976    tcg_out_qemu_st_direct(s, get_memop(oi), data_reg, h);
1977
1978    if (ldst) {
1979        ldst->type = data_type;
1980        ldst->datalo_reg = data_reg;
1981        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1982    }
1983}
1984
1985static void tcg_out_qemu_ldst_i128(TCGContext *s, TCGReg datalo, TCGReg datahi,
1986                                   TCGReg addr_reg, MemOpIdx oi, bool is_ld)
1987{
1988    TCGLabel *l1 = NULL, *l2 = NULL;
1989    TCGLabelQemuLdst *ldst;
1990    HostAddress h;
1991    bool need_bswap;
1992    bool use_pair;
1993    S390Opcode insn;
1994
1995    ldst = prepare_host_addr(s, &h, addr_reg, oi, is_ld);
1996
1997    use_pair = h.aa.atom < MO_128;
1998    need_bswap = get_memop(oi) & MO_BSWAP;
1999
2000    if (!use_pair) {
2001        /*
2002         * Atomicity requires we use LPQ.  If we've already checked for
2003         * 16-byte alignment, that's all we need.  If we arrive with
2004         * lesser alignment, we have determined that less than 16-byte
2005         * alignment can be satisfied with two 8-byte loads.
2006         */
2007        if (h.aa.align < MO_128) {
2008            use_pair = true;
2009            l1 = gen_new_label();
2010            l2 = gen_new_label();
2011
2012            tcg_out_insn(s, RI, TMLL, addr_reg, 15);
2013            tgen_branch(s, S390_TM_NE, l1);
2014        }
2015
2016        tcg_debug_assert(!need_bswap);
2017        tcg_debug_assert(datalo & 1);
2018        tcg_debug_assert(datahi == datalo - 1);
2019        insn = is_ld ? RXY_LPQ : RXY_STPQ;
2020        tcg_out_insn_RXY(s, insn, datahi, h.base, h.index, h.disp);
2021
2022        if (use_pair) {
2023            tgen_branch(s, S390_CC_ALWAYS, l2);
2024            tcg_out_label(s, l1);
2025        }
2026    }
2027    if (use_pair) {
2028        TCGReg d1, d2;
2029
2030        if (need_bswap) {
2031            d1 = datalo, d2 = datahi;
2032            insn = is_ld ? RXY_LRVG : RXY_STRVG;
2033        } else {
2034            d1 = datahi, d2 = datalo;
2035            insn = is_ld ? RXY_LG : RXY_STG;
2036        }
2037
2038        if (h.base == d1 || h.index == d1) {
2039            tcg_out_insn(s, RXY, LAY, TCG_TMP0, h.base, h.index, h.disp);
2040            h.base = TCG_TMP0;
2041            h.index = TCG_REG_NONE;
2042            h.disp = 0;
2043        }
2044        tcg_out_insn_RXY(s, insn, d1, h.base, h.index, h.disp);
2045        tcg_out_insn_RXY(s, insn, d2, h.base, h.index, h.disp + 8);
2046    }
2047    if (l2) {
2048        tcg_out_label(s, l2);
2049    }
2050
2051    if (ldst) {
2052        ldst->type = TCG_TYPE_I128;
2053        ldst->datalo_reg = datalo;
2054        ldst->datahi_reg = datahi;
2055        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2056    }
2057}
2058
2059static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
2060{
2061    /* Reuse the zeroing that exists for goto_ptr.  */
2062    if (a0 == 0) {
2063        tgen_gotoi(s, S390_CC_ALWAYS, tcg_code_gen_epilogue);
2064    } else {
2065        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, a0);
2066        tgen_gotoi(s, S390_CC_ALWAYS, tb_ret_addr);
2067    }
2068}
2069
2070static void tcg_out_goto_tb(TCGContext *s, int which)
2071{
2072    /*
2073     * Branch displacement must be aligned for atomic patching;
2074     * see if we need to add extra nop before branch
2075     */
2076    if (!QEMU_PTR_IS_ALIGNED(s->code_ptr + 1, 4)) {
2077        tcg_out16(s, NOP);
2078    }
2079    tcg_out16(s, RIL_BRCL | (S390_CC_ALWAYS << 4));
2080    set_jmp_insn_offset(s, which);
2081    s->code_ptr += 2;
2082    set_jmp_reset_offset(s, which);
2083}
2084
2085void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
2086                              uintptr_t jmp_rx, uintptr_t jmp_rw)
2087{
2088    if (!HAVE_FACILITY(GEN_INST_EXT)) {
2089        return;
2090    }
2091    /* patch the branch destination */
2092    uintptr_t addr = tb->jmp_target_addr[n];
2093    intptr_t disp = addr - (jmp_rx - 2);
2094    qatomic_set((int32_t *)jmp_rw, disp / 2);
2095    /* no need to flush icache explicitly */
2096}
2097
2098# define OP_32_64(x) \
2099        case glue(glue(INDEX_op_,x),_i32): \
2100        case glue(glue(INDEX_op_,x),_i64)
2101
2102static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
2103                              const TCGArg args[TCG_MAX_OP_ARGS],
2104                              const int const_args[TCG_MAX_OP_ARGS])
2105{
2106    S390Opcode op, op2;
2107    TCGArg a0, a1, a2;
2108
2109    switch (opc) {
2110    case INDEX_op_goto_ptr:
2111        a0 = args[0];
2112        tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, a0);
2113        break;
2114
2115    OP_32_64(ld8u):
2116        /* ??? LLC (RXY format) is only present with the extended-immediate
2117           facility, whereas LLGC is always present.  */
2118        tcg_out_mem(s, 0, RXY_LLGC, args[0], args[1], TCG_REG_NONE, args[2]);
2119        break;
2120
2121    OP_32_64(ld8s):
2122        /* ??? LB is no smaller than LGB, so no point to using it.  */
2123        tcg_out_mem(s, 0, RXY_LGB, args[0], args[1], TCG_REG_NONE, args[2]);
2124        break;
2125
2126    OP_32_64(ld16u):
2127        /* ??? LLH (RXY format) is only present with the extended-immediate
2128           facility, whereas LLGH is always present.  */
2129        tcg_out_mem(s, 0, RXY_LLGH, args[0], args[1], TCG_REG_NONE, args[2]);
2130        break;
2131
2132    case INDEX_op_ld16s_i32:
2133        tcg_out_mem(s, RX_LH, RXY_LHY, args[0], args[1], TCG_REG_NONE, args[2]);
2134        break;
2135
2136    case INDEX_op_ld_i32:
2137        tcg_out_ld(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2138        break;
2139
2140    OP_32_64(st8):
2141        tcg_out_mem(s, RX_STC, RXY_STCY, args[0], args[1],
2142                    TCG_REG_NONE, args[2]);
2143        break;
2144
2145    OP_32_64(st16):
2146        tcg_out_mem(s, RX_STH, RXY_STHY, args[0], args[1],
2147                    TCG_REG_NONE, args[2]);
2148        break;
2149
2150    case INDEX_op_st_i32:
2151        tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2152        break;
2153
2154    case INDEX_op_add_i32:
2155        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2156        if (const_args[2]) {
2157        do_addi_32:
2158            if (a0 == a1) {
2159                if (a2 == (int16_t)a2) {
2160                    tcg_out_insn(s, RI, AHI, a0, a2);
2161                    break;
2162                }
2163                tcg_out_insn(s, RIL, AFI, a0, a2);
2164                break;
2165            }
2166            tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
2167        } else if (a0 == a1) {
2168            tcg_out_insn(s, RR, AR, a0, a2);
2169        } else {
2170            tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
2171        }
2172        break;
2173    case INDEX_op_sub_i32:
2174        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2175        if (const_args[2]) {
2176            a2 = -a2;
2177            goto do_addi_32;
2178        } else if (a0 == a1) {
2179            tcg_out_insn(s, RR, SR, a0, a2);
2180        } else {
2181            tcg_out_insn(s, RRFa, SRK, a0, a1, a2);
2182        }
2183        break;
2184
2185    case INDEX_op_and_i32:
2186        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2187        if (const_args[2]) {
2188            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2189            tgen_andi(s, TCG_TYPE_I32, a0, a2);
2190        } else if (a0 == a1) {
2191            tcg_out_insn(s, RR, NR, a0, a2);
2192        } else {
2193            tcg_out_insn(s, RRFa, NRK, a0, a1, a2);
2194        }
2195        break;
2196    case INDEX_op_or_i32:
2197        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2198        if (const_args[2]) {
2199            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2200            tgen_ori(s, a0, a2);
2201        } else if (a0 == a1) {
2202            tcg_out_insn(s, RR, OR, a0, a2);
2203        } else {
2204            tcg_out_insn(s, RRFa, ORK, a0, a1, a2);
2205        }
2206        break;
2207    case INDEX_op_xor_i32:
2208        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2209        if (const_args[2]) {
2210            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2211            tcg_out_insn(s, RIL, XILF, a0, a2);
2212        } else if (a0 == a1) {
2213            tcg_out_insn(s, RR, XR, args[0], args[2]);
2214        } else {
2215            tcg_out_insn(s, RRFa, XRK, a0, a1, a2);
2216        }
2217        break;
2218
2219    case INDEX_op_andc_i32:
2220        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2221        if (const_args[2]) {
2222            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2223            tgen_andi(s, TCG_TYPE_I32, a0, (uint32_t)~a2);
2224	} else {
2225            tcg_out_insn(s, RRFa, NCRK, a0, a1, a2);
2226	}
2227        break;
2228    case INDEX_op_orc_i32:
2229        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2230        if (const_args[2]) {
2231            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2232            tgen_ori(s, a0, (uint32_t)~a2);
2233        } else {
2234            tcg_out_insn(s, RRFa, OCRK, a0, a1, a2);
2235        }
2236        break;
2237    case INDEX_op_eqv_i32:
2238        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2239        if (const_args[2]) {
2240            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2241            tcg_out_insn(s, RIL, XILF, a0, ~a2);
2242        } else {
2243            tcg_out_insn(s, RRFa, NXRK, a0, a1, a2);
2244        }
2245        break;
2246    case INDEX_op_nand_i32:
2247        tcg_out_insn(s, RRFa, NNRK, args[0], args[1], args[2]);
2248        break;
2249    case INDEX_op_nor_i32:
2250        tcg_out_insn(s, RRFa, NORK, args[0], args[1], args[2]);
2251        break;
2252
2253    case INDEX_op_neg_i32:
2254        tcg_out_insn(s, RR, LCR, args[0], args[1]);
2255        break;
2256    case INDEX_op_not_i32:
2257        tcg_out_insn(s, RRFa, NORK, args[0], args[1], args[1]);
2258        break;
2259
2260    case INDEX_op_mul_i32:
2261        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2262        if (const_args[2]) {
2263            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2264            if (a2 == (int16_t)a2) {
2265                tcg_out_insn(s, RI, MHI, a0, a2);
2266            } else {
2267                tcg_out_insn(s, RIL, MSFI, a0, a2);
2268            }
2269        } else if (a0 == a1) {
2270            tcg_out_insn(s, RRE, MSR, a0, a2);
2271        } else {
2272            tcg_out_insn(s, RRFa, MSRKC, a0, a1, a2);
2273        }
2274        break;
2275
2276    case INDEX_op_div2_i32:
2277        tcg_debug_assert(args[0] == args[2]);
2278        tcg_debug_assert(args[1] == args[3]);
2279        tcg_debug_assert((args[1] & 1) == 0);
2280        tcg_debug_assert(args[0] == args[1] + 1);
2281        tcg_out_insn(s, RR, DR, args[1], args[4]);
2282        break;
2283    case INDEX_op_divu2_i32:
2284        tcg_debug_assert(args[0] == args[2]);
2285        tcg_debug_assert(args[1] == args[3]);
2286        tcg_debug_assert((args[1] & 1) == 0);
2287        tcg_debug_assert(args[0] == args[1] + 1);
2288        tcg_out_insn(s, RRE, DLR, args[1], args[4]);
2289        break;
2290
2291    case INDEX_op_shl_i32:
2292        op = RS_SLL;
2293        op2 = RSY_SLLK;
2294    do_shift32:
2295        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2296        if (a0 == a1) {
2297            if (const_args[2]) {
2298                tcg_out_sh32(s, op, a0, TCG_REG_NONE, a2);
2299            } else {
2300                tcg_out_sh32(s, op, a0, a2, 0);
2301            }
2302        } else {
2303            /* Using tcg_out_sh64 here for the format; it is a 32-bit shift.  */
2304            if (const_args[2]) {
2305                tcg_out_sh64(s, op2, a0, a1, TCG_REG_NONE, a2);
2306            } else {
2307                tcg_out_sh64(s, op2, a0, a1, a2, 0);
2308            }
2309        }
2310        break;
2311    case INDEX_op_shr_i32:
2312        op = RS_SRL;
2313        op2 = RSY_SRLK;
2314        goto do_shift32;
2315    case INDEX_op_sar_i32:
2316        op = RS_SRA;
2317        op2 = RSY_SRAK;
2318        goto do_shift32;
2319
2320    case INDEX_op_rotl_i32:
2321        /* ??? Using tcg_out_sh64 here for the format; it is a 32-bit rol.  */
2322        if (const_args[2]) {
2323            tcg_out_sh64(s, RSY_RLL, args[0], args[1], TCG_REG_NONE, args[2]);
2324        } else {
2325            tcg_out_sh64(s, RSY_RLL, args[0], args[1], args[2], 0);
2326        }
2327        break;
2328    case INDEX_op_rotr_i32:
2329        if (const_args[2]) {
2330            tcg_out_sh64(s, RSY_RLL, args[0], args[1],
2331                         TCG_REG_NONE, (32 - args[2]) & 31);
2332        } else {
2333            tcg_out_insn(s, RR, LCR, TCG_TMP0, args[2]);
2334            tcg_out_sh64(s, RSY_RLL, args[0], args[1], TCG_TMP0, 0);
2335        }
2336        break;
2337
2338    case INDEX_op_bswap16_i32:
2339        a0 = args[0], a1 = args[1], a2 = args[2];
2340        tcg_out_insn(s, RRE, LRVR, a0, a1);
2341        if (a2 & TCG_BSWAP_OS) {
2342            tcg_out_sh32(s, RS_SRA, a0, TCG_REG_NONE, 16);
2343        } else {
2344            tcg_out_sh32(s, RS_SRL, a0, TCG_REG_NONE, 16);
2345        }
2346        break;
2347    case INDEX_op_bswap16_i64:
2348        a0 = args[0], a1 = args[1], a2 = args[2];
2349        tcg_out_insn(s, RRE, LRVGR, a0, a1);
2350        if (a2 & TCG_BSWAP_OS) {
2351            tcg_out_sh64(s, RSY_SRAG, a0, a0, TCG_REG_NONE, 48);
2352        } else {
2353            tcg_out_sh64(s, RSY_SRLG, a0, a0, TCG_REG_NONE, 48);
2354        }
2355        break;
2356
2357    case INDEX_op_bswap32_i32:
2358        tcg_out_insn(s, RRE, LRVR, args[0], args[1]);
2359        break;
2360    case INDEX_op_bswap32_i64:
2361        a0 = args[0], a1 = args[1], a2 = args[2];
2362        tcg_out_insn(s, RRE, LRVR, a0, a1);
2363        if (a2 & TCG_BSWAP_OS) {
2364            tcg_out_ext32s(s, a0, a0);
2365        } else if ((a2 & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
2366            tcg_out_ext32u(s, a0, a0);
2367        }
2368        break;
2369
2370    case INDEX_op_add2_i32:
2371        if (const_args[4]) {
2372            tcg_out_insn(s, RIL, ALFI, args[0], args[4]);
2373        } else {
2374            tcg_out_insn(s, RR, ALR, args[0], args[4]);
2375        }
2376        tcg_out_insn(s, RRE, ALCR, args[1], args[5]);
2377        break;
2378    case INDEX_op_sub2_i32:
2379        if (const_args[4]) {
2380            tcg_out_insn(s, RIL, SLFI, args[0], args[4]);
2381        } else {
2382            tcg_out_insn(s, RR, SLR, args[0], args[4]);
2383        }
2384        tcg_out_insn(s, RRE, SLBR, args[1], args[5]);
2385        break;
2386
2387    case INDEX_op_br:
2388        tgen_branch(s, S390_CC_ALWAYS, arg_label(args[0]));
2389        break;
2390
2391    case INDEX_op_brcond_i32:
2392        tgen_brcond(s, TCG_TYPE_I32, args[2], args[0],
2393                    args[1], const_args[1], arg_label(args[3]));
2394        break;
2395    case INDEX_op_setcond_i32:
2396        tgen_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1],
2397                     args[2], const_args[2], false);
2398        break;
2399    case INDEX_op_negsetcond_i32:
2400        tgen_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1],
2401                     args[2], const_args[2], true);
2402        break;
2403    case INDEX_op_movcond_i32:
2404        tgen_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1],
2405                     args[2], const_args[2], args[3], const_args[3], args[4]);
2406        break;
2407
2408    case INDEX_op_qemu_ld_a32_i32:
2409    case INDEX_op_qemu_ld_a64_i32:
2410        tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I32);
2411        break;
2412    case INDEX_op_qemu_ld_a32_i64:
2413    case INDEX_op_qemu_ld_a64_i64:
2414        tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I64);
2415        break;
2416    case INDEX_op_qemu_st_a32_i32:
2417    case INDEX_op_qemu_st_a64_i32:
2418        tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I32);
2419        break;
2420    case INDEX_op_qemu_st_a32_i64:
2421    case INDEX_op_qemu_st_a64_i64:
2422        tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I64);
2423        break;
2424    case INDEX_op_qemu_ld_a32_i128:
2425    case INDEX_op_qemu_ld_a64_i128:
2426        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], true);
2427        break;
2428    case INDEX_op_qemu_st_a32_i128:
2429    case INDEX_op_qemu_st_a64_i128:
2430        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], false);
2431        break;
2432
2433    case INDEX_op_ld16s_i64:
2434        tcg_out_mem(s, 0, RXY_LGH, args[0], args[1], TCG_REG_NONE, args[2]);
2435        break;
2436    case INDEX_op_ld32u_i64:
2437        tcg_out_mem(s, 0, RXY_LLGF, args[0], args[1], TCG_REG_NONE, args[2]);
2438        break;
2439    case INDEX_op_ld32s_i64:
2440        tcg_out_mem(s, 0, RXY_LGF, args[0], args[1], TCG_REG_NONE, args[2]);
2441        break;
2442    case INDEX_op_ld_i64:
2443        tcg_out_ld(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2444        break;
2445
2446    case INDEX_op_st32_i64:
2447        tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2448        break;
2449    case INDEX_op_st_i64:
2450        tcg_out_st(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2451        break;
2452
2453    case INDEX_op_add_i64:
2454        a0 = args[0], a1 = args[1], a2 = args[2];
2455        if (const_args[2]) {
2456        do_addi_64:
2457            if (a0 == a1) {
2458                if (a2 == (int16_t)a2) {
2459                    tcg_out_insn(s, RI, AGHI, a0, a2);
2460                    break;
2461                }
2462                if (a2 == (int32_t)a2) {
2463                    tcg_out_insn(s, RIL, AGFI, a0, a2);
2464                    break;
2465                }
2466                if (a2 == (uint32_t)a2) {
2467                    tcg_out_insn(s, RIL, ALGFI, a0, a2);
2468                    break;
2469                }
2470                if (-a2 == (uint32_t)-a2) {
2471                    tcg_out_insn(s, RIL, SLGFI, a0, -a2);
2472                    break;
2473                }
2474            }
2475            tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
2476        } else if (a0 == a1) {
2477            tcg_out_insn(s, RRE, AGR, a0, a2);
2478        } else {
2479            tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
2480        }
2481        break;
2482    case INDEX_op_sub_i64:
2483        a0 = args[0], a1 = args[1], a2 = args[2];
2484        if (const_args[2]) {
2485            a2 = -a2;
2486            goto do_addi_64;
2487        } else {
2488            tcg_out_insn(s, RRFa, SGRK, a0, a1, a2);
2489        }
2490        break;
2491
2492    case INDEX_op_and_i64:
2493        a0 = args[0], a1 = args[1], a2 = args[2];
2494        if (const_args[2]) {
2495            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2496            tgen_andi(s, TCG_TYPE_I64, args[0], args[2]);
2497        } else {
2498            tcg_out_insn(s, RRFa, NGRK, a0, a1, a2);
2499        }
2500        break;
2501    case INDEX_op_or_i64:
2502        a0 = args[0], a1 = args[1], a2 = args[2];
2503        if (const_args[2]) {
2504            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2505            tgen_ori(s, a0, a2);
2506        } else {
2507            tcg_out_insn(s, RRFa, OGRK, a0, a1, a2);
2508        }
2509        break;
2510    case INDEX_op_xor_i64:
2511        a0 = args[0], a1 = args[1], a2 = args[2];
2512        if (const_args[2]) {
2513            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2514            tgen_xori(s, a0, a2);
2515        } else {
2516            tcg_out_insn(s, RRFa, XGRK, a0, a1, a2);
2517        }
2518        break;
2519
2520    case INDEX_op_andc_i64:
2521        a0 = args[0], a1 = args[1], a2 = args[2];
2522        if (const_args[2]) {
2523            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2524            tgen_andi(s, TCG_TYPE_I64, a0, ~a2);
2525        } else {
2526            tcg_out_insn(s, RRFa, NCGRK, a0, a1, a2);
2527        }
2528        break;
2529    case INDEX_op_orc_i64:
2530        a0 = args[0], a1 = args[1], a2 = args[2];
2531        if (const_args[2]) {
2532            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2533            tgen_ori(s, a0, ~a2);
2534        } else {
2535            tcg_out_insn(s, RRFa, OCGRK, a0, a1, a2);
2536        }
2537        break;
2538    case INDEX_op_eqv_i64:
2539        a0 = args[0], a1 = args[1], a2 = args[2];
2540        if (const_args[2]) {
2541            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2542            tgen_xori(s, a0, ~a2);
2543        } else {
2544            tcg_out_insn(s, RRFa, NXGRK, a0, a1, a2);
2545        }
2546        break;
2547    case INDEX_op_nand_i64:
2548        tcg_out_insn(s, RRFa, NNGRK, args[0], args[1], args[2]);
2549        break;
2550    case INDEX_op_nor_i64:
2551        tcg_out_insn(s, RRFa, NOGRK, args[0], args[1], args[2]);
2552        break;
2553
2554    case INDEX_op_neg_i64:
2555        tcg_out_insn(s, RRE, LCGR, args[0], args[1]);
2556        break;
2557    case INDEX_op_not_i64:
2558        tcg_out_insn(s, RRFa, NOGRK, args[0], args[1], args[1]);
2559        break;
2560    case INDEX_op_bswap64_i64:
2561        tcg_out_insn(s, RRE, LRVGR, args[0], args[1]);
2562        break;
2563
2564    case INDEX_op_mul_i64:
2565        a0 = args[0], a1 = args[1], a2 = args[2];
2566        if (const_args[2]) {
2567            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2568            if (a2 == (int16_t)a2) {
2569                tcg_out_insn(s, RI, MGHI, a0, a2);
2570            } else {
2571                tcg_out_insn(s, RIL, MSGFI, a0, a2);
2572            }
2573        } else if (a0 == a1) {
2574            tcg_out_insn(s, RRE, MSGR, a0, a2);
2575        } else {
2576            tcg_out_insn(s, RRFa, MSGRKC, a0, a1, a2);
2577        }
2578        break;
2579
2580    case INDEX_op_div2_i64:
2581        /*
2582         * ??? We get an unnecessary sign-extension of the dividend
2583         * into op0 with this definition, but as we do in fact always
2584         * produce both quotient and remainder using INDEX_op_div_i64
2585         * instead requires jumping through even more hoops.
2586         */
2587        tcg_debug_assert(args[0] == args[2]);
2588        tcg_debug_assert(args[1] == args[3]);
2589        tcg_debug_assert((args[1] & 1) == 0);
2590        tcg_debug_assert(args[0] == args[1] + 1);
2591        tcg_out_insn(s, RRE, DSGR, args[1], args[4]);
2592        break;
2593    case INDEX_op_divu2_i64:
2594        tcg_debug_assert(args[0] == args[2]);
2595        tcg_debug_assert(args[1] == args[3]);
2596        tcg_debug_assert((args[1] & 1) == 0);
2597        tcg_debug_assert(args[0] == args[1] + 1);
2598        tcg_out_insn(s, RRE, DLGR, args[1], args[4]);
2599        break;
2600    case INDEX_op_mulu2_i64:
2601        tcg_debug_assert(args[0] == args[2]);
2602        tcg_debug_assert((args[1] & 1) == 0);
2603        tcg_debug_assert(args[0] == args[1] + 1);
2604        tcg_out_insn(s, RRE, MLGR, args[1], args[3]);
2605        break;
2606    case INDEX_op_muls2_i64:
2607        tcg_debug_assert((args[1] & 1) == 0);
2608        tcg_debug_assert(args[0] == args[1] + 1);
2609        tcg_out_insn(s, RRFa, MGRK, args[1], args[2], args[3]);
2610        break;
2611
2612    case INDEX_op_shl_i64:
2613        op = RSY_SLLG;
2614    do_shift64:
2615        if (const_args[2]) {
2616            tcg_out_sh64(s, op, args[0], args[1], TCG_REG_NONE, args[2]);
2617        } else {
2618            tcg_out_sh64(s, op, args[0], args[1], args[2], 0);
2619        }
2620        break;
2621    case INDEX_op_shr_i64:
2622        op = RSY_SRLG;
2623        goto do_shift64;
2624    case INDEX_op_sar_i64:
2625        op = RSY_SRAG;
2626        goto do_shift64;
2627
2628    case INDEX_op_rotl_i64:
2629        if (const_args[2]) {
2630            tcg_out_sh64(s, RSY_RLLG, args[0], args[1],
2631                         TCG_REG_NONE, args[2]);
2632        } else {
2633            tcg_out_sh64(s, RSY_RLLG, args[0], args[1], args[2], 0);
2634        }
2635        break;
2636    case INDEX_op_rotr_i64:
2637        if (const_args[2]) {
2638            tcg_out_sh64(s, RSY_RLLG, args[0], args[1],
2639                         TCG_REG_NONE, (64 - args[2]) & 63);
2640        } else {
2641            /* We can use the smaller 32-bit negate because only the
2642               low 6 bits are examined for the rotate.  */
2643            tcg_out_insn(s, RR, LCR, TCG_TMP0, args[2]);
2644            tcg_out_sh64(s, RSY_RLLG, args[0], args[1], TCG_TMP0, 0);
2645        }
2646        break;
2647
2648    case INDEX_op_add2_i64:
2649        if (const_args[4]) {
2650            if ((int64_t)args[4] >= 0) {
2651                tcg_out_insn(s, RIL, ALGFI, args[0], args[4]);
2652            } else {
2653                tcg_out_insn(s, RIL, SLGFI, args[0], -args[4]);
2654            }
2655        } else {
2656            tcg_out_insn(s, RRE, ALGR, args[0], args[4]);
2657        }
2658        tcg_out_insn(s, RRE, ALCGR, args[1], args[5]);
2659        break;
2660    case INDEX_op_sub2_i64:
2661        if (const_args[4]) {
2662            if ((int64_t)args[4] >= 0) {
2663                tcg_out_insn(s, RIL, SLGFI, args[0], args[4]);
2664            } else {
2665                tcg_out_insn(s, RIL, ALGFI, args[0], -args[4]);
2666            }
2667        } else {
2668            tcg_out_insn(s, RRE, SLGR, args[0], args[4]);
2669        }
2670        tcg_out_insn(s, RRE, SLBGR, args[1], args[5]);
2671        break;
2672
2673    case INDEX_op_brcond_i64:
2674        tgen_brcond(s, TCG_TYPE_I64, args[2], args[0],
2675                    args[1], const_args[1], arg_label(args[3]));
2676        break;
2677    case INDEX_op_setcond_i64:
2678        tgen_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1],
2679                     args[2], const_args[2], false);
2680        break;
2681    case INDEX_op_negsetcond_i64:
2682        tgen_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1],
2683                     args[2], const_args[2], true);
2684        break;
2685    case INDEX_op_movcond_i64:
2686        tgen_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1],
2687                     args[2], const_args[2], args[3], const_args[3], args[4]);
2688        break;
2689
2690    OP_32_64(deposit):
2691        a0 = args[0], a1 = args[1], a2 = args[2];
2692        if (const_args[1]) {
2693            tgen_deposit(s, a0, a2, args[3], args[4], 1);
2694        } else {
2695            /* Since we can't support "0Z" as a constraint, we allow a1 in
2696               any register.  Fix things up as if a matching constraint.  */
2697            if (a0 != a1) {
2698                TCGType type = (opc == INDEX_op_deposit_i64);
2699                if (a0 == a2) {
2700                    tcg_out_mov(s, type, TCG_TMP0, a2);
2701                    a2 = TCG_TMP0;
2702                }
2703                tcg_out_mov(s, type, a0, a1);
2704            }
2705            tgen_deposit(s, a0, a2, args[3], args[4], 0);
2706        }
2707        break;
2708
2709    OP_32_64(extract):
2710        tgen_extract(s, args[0], args[1], args[2], args[3]);
2711        break;
2712
2713    case INDEX_op_clz_i64:
2714        tgen_clz(s, args[0], args[1], args[2], const_args[2]);
2715        break;
2716
2717    case INDEX_op_ctpop_i32:
2718        tgen_ctpop(s, TCG_TYPE_I32, args[0], args[1]);
2719        break;
2720    case INDEX_op_ctpop_i64:
2721        tgen_ctpop(s, TCG_TYPE_I64, args[0], args[1]);
2722        break;
2723
2724    case INDEX_op_mb:
2725        /* The host memory model is quite strong, we simply need to
2726           serialize the instruction stream.  */
2727        if (args[0] & TCG_MO_ST_LD) {
2728            /* fast-bcr-serialization facility (45) is present */
2729            tcg_out_insn(s, RR, BCR, 14, 0);
2730        }
2731        break;
2732
2733    case INDEX_op_mov_i32:  /* Always emitted via tcg_out_mov.  */
2734    case INDEX_op_mov_i64:
2735    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
2736    case INDEX_op_exit_tb:  /* Always emitted via tcg_out_exit_tb.  */
2737    case INDEX_op_goto_tb:  /* Always emitted via tcg_out_goto_tb.  */
2738    case INDEX_op_ext8s_i32:  /* Always emitted via tcg_reg_alloc_op.  */
2739    case INDEX_op_ext8s_i64:
2740    case INDEX_op_ext8u_i32:
2741    case INDEX_op_ext8u_i64:
2742    case INDEX_op_ext16s_i32:
2743    case INDEX_op_ext16s_i64:
2744    case INDEX_op_ext16u_i32:
2745    case INDEX_op_ext16u_i64:
2746    case INDEX_op_ext32s_i64:
2747    case INDEX_op_ext32u_i64:
2748    case INDEX_op_ext_i32_i64:
2749    case INDEX_op_extu_i32_i64:
2750    case INDEX_op_extrl_i64_i32:
2751    default:
2752        g_assert_not_reached();
2753    }
2754}
2755
2756static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
2757                            TCGReg dst, TCGReg src)
2758{
2759    if (is_general_reg(src)) {
2760        /* Replicate general register into two MO_64. */
2761        tcg_out_insn(s, VRRf, VLVGP, dst, src, src);
2762        if (vece == MO_64) {
2763            return true;
2764        }
2765        src = dst;
2766    }
2767
2768    /*
2769     * Recall that the "standard" integer, within a vector, is the
2770     * rightmost element of the leftmost doubleword, a-la VLLEZ.
2771     */
2772    tcg_out_insn(s, VRIc, VREP, dst, (8 >> vece) - 1, src, vece);
2773    return true;
2774}
2775
2776static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
2777                             TCGReg dst, TCGReg base, intptr_t offset)
2778{
2779    tcg_out_vrx_mem(s, VRX_VLREP, dst, base, TCG_REG_NONE, offset, vece);
2780    return true;
2781}
2782
2783static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
2784                             TCGReg dst, int64_t val)
2785{
2786    int i, mask, msb, lsb;
2787
2788    /* Look for int16_t elements.  */
2789    if (vece <= MO_16 ||
2790        (vece == MO_32 ? (int32_t)val : val) == (int16_t)val) {
2791        tcg_out_insn(s, VRIa, VREPI, dst, val, vece);
2792        return;
2793    }
2794
2795    /* Look for bit masks.  */
2796    if (vece == MO_32) {
2797        if (risbg_mask((int32_t)val)) {
2798            /* Handle wraparound by swapping msb and lsb.  */
2799            if ((val & 0x80000001u) == 0x80000001u) {
2800                msb = 32 - ctz32(~val);
2801                lsb = clz32(~val) - 1;
2802            } else {
2803                msb = clz32(val);
2804                lsb = 31 - ctz32(val);
2805            }
2806            tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_32);
2807            return;
2808        }
2809    } else {
2810        if (risbg_mask(val)) {
2811            /* Handle wraparound by swapping msb and lsb.  */
2812            if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
2813                /* Handle wraparound by swapping msb and lsb.  */
2814                msb = 64 - ctz64(~val);
2815                lsb = clz64(~val) - 1;
2816            } else {
2817                msb = clz64(val);
2818                lsb = 63 - ctz64(val);
2819            }
2820            tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_64);
2821            return;
2822        }
2823    }
2824
2825    /* Look for all bytes 0x00 or 0xff.  */
2826    for (i = mask = 0; i < 8; i++) {
2827        uint8_t byte = val >> (i * 8);
2828        if (byte == 0xff) {
2829            mask |= 1 << i;
2830        } else if (byte != 0) {
2831            break;
2832        }
2833    }
2834    if (i == 8) {
2835        tcg_out_insn(s, VRIa, VGBM, dst, mask * 0x0101, 0);
2836        return;
2837    }
2838
2839    /* Otherwise, stuff it in the constant pool.  */
2840    tcg_out_insn(s, RIL, LARL, TCG_TMP0, 0);
2841    new_pool_label(s, val, R_390_PC32DBL, s->code_ptr - 2, 2);
2842    tcg_out_insn(s, VRX, VLREP, dst, TCG_TMP0, TCG_REG_NONE, 0, MO_64);
2843}
2844
2845static bool tcg_out_cmp_vec_noinv(TCGContext *s, unsigned vece, TCGReg a0,
2846                                  TCGReg a1, TCGReg a2, TCGCond cond)
2847{
2848    bool need_swap = false, need_inv = false;
2849
2850    switch (cond) {
2851    case TCG_COND_EQ:
2852    case TCG_COND_GT:
2853    case TCG_COND_GTU:
2854        break;
2855    case TCG_COND_NE:
2856    case TCG_COND_LE:
2857    case TCG_COND_LEU:
2858        need_inv = true;
2859        break;
2860    case TCG_COND_LT:
2861    case TCG_COND_LTU:
2862        need_swap = true;
2863        break;
2864    case TCG_COND_GE:
2865    case TCG_COND_GEU:
2866        need_swap = need_inv = true;
2867        break;
2868    default:
2869        g_assert_not_reached();
2870    }
2871
2872    if (need_inv) {
2873        cond = tcg_invert_cond(cond);
2874    }
2875    if (need_swap) {
2876        TCGReg swap = a1;
2877        a1 = a2;
2878        a2 = swap;
2879        cond = tcg_swap_cond(cond);
2880    }
2881
2882    switch (cond) {
2883    case TCG_COND_EQ:
2884        tcg_out_insn(s, VRRc, VCEQ, a0, a1, a2, vece);
2885        break;
2886    case TCG_COND_GT:
2887        tcg_out_insn(s, VRRc, VCH, a0, a1, a2, vece);
2888        break;
2889    case TCG_COND_GTU:
2890        tcg_out_insn(s, VRRc, VCHL, a0, a1, a2, vece);
2891        break;
2892    default:
2893        g_assert_not_reached();
2894    }
2895    return need_inv;
2896}
2897
2898static void tcg_out_cmp_vec(TCGContext *s, unsigned vece, TCGReg a0,
2899                            TCGReg a1, TCGReg a2, TCGCond cond)
2900{
2901    if (tcg_out_cmp_vec_noinv(s, vece, a0, a1, a2, cond)) {
2902        tcg_out_insn(s, VRRc, VNO, a0, a0, a0, 0);
2903    }
2904}
2905
2906static void tcg_out_cmpsel_vec(TCGContext *s, unsigned vece, TCGReg a0,
2907                               TCGReg c1, TCGReg c2,
2908                               TCGReg v3, TCGReg v4, TCGCond cond)
2909{
2910    if (tcg_out_cmp_vec_noinv(s, vece, TCG_VEC_TMP0, c1, c2, cond)) {
2911        TCGReg swap = v3;
2912        v3 = v4;
2913        v4 = swap;
2914    }
2915    tcg_out_insn(s, VRRe, VSEL, a0, v3, v4, TCG_VEC_TMP0);
2916}
2917
2918static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
2919                           unsigned vecl, unsigned vece,
2920                           const TCGArg args[TCG_MAX_OP_ARGS],
2921                           const int const_args[TCG_MAX_OP_ARGS])
2922{
2923    TCGType type = vecl + TCG_TYPE_V64;
2924    TCGArg a0 = args[0], a1 = args[1], a2 = args[2];
2925
2926    switch (opc) {
2927    case INDEX_op_ld_vec:
2928        tcg_out_ld(s, type, a0, a1, a2);
2929        break;
2930    case INDEX_op_st_vec:
2931        tcg_out_st(s, type, a0, a1, a2);
2932        break;
2933    case INDEX_op_dupm_vec:
2934        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
2935        break;
2936
2937    case INDEX_op_abs_vec:
2938        tcg_out_insn(s, VRRa, VLP, a0, a1, vece);
2939        break;
2940    case INDEX_op_neg_vec:
2941        tcg_out_insn(s, VRRa, VLC, a0, a1, vece);
2942        break;
2943    case INDEX_op_not_vec:
2944        tcg_out_insn(s, VRRc, VNO, a0, a1, a1, 0);
2945        break;
2946
2947    case INDEX_op_add_vec:
2948        tcg_out_insn(s, VRRc, VA, a0, a1, a2, vece);
2949        break;
2950    case INDEX_op_sub_vec:
2951        tcg_out_insn(s, VRRc, VS, a0, a1, a2, vece);
2952        break;
2953    case INDEX_op_and_vec:
2954        tcg_out_insn(s, VRRc, VN, a0, a1, a2, 0);
2955        break;
2956    case INDEX_op_andc_vec:
2957        tcg_out_insn(s, VRRc, VNC, a0, a1, a2, 0);
2958        break;
2959    case INDEX_op_mul_vec:
2960        tcg_out_insn(s, VRRc, VML, a0, a1, a2, vece);
2961        break;
2962    case INDEX_op_or_vec:
2963        tcg_out_insn(s, VRRc, VO, a0, a1, a2, 0);
2964        break;
2965    case INDEX_op_orc_vec:
2966        tcg_out_insn(s, VRRc, VOC, a0, a1, a2, 0);
2967        break;
2968    case INDEX_op_xor_vec:
2969        tcg_out_insn(s, VRRc, VX, a0, a1, a2, 0);
2970        break;
2971    case INDEX_op_nand_vec:
2972        tcg_out_insn(s, VRRc, VNN, a0, a1, a2, 0);
2973        break;
2974    case INDEX_op_nor_vec:
2975        tcg_out_insn(s, VRRc, VNO, a0, a1, a2, 0);
2976        break;
2977    case INDEX_op_eqv_vec:
2978        tcg_out_insn(s, VRRc, VNX, a0, a1, a2, 0);
2979        break;
2980
2981    case INDEX_op_shli_vec:
2982        tcg_out_insn(s, VRSa, VESL, a0, a2, TCG_REG_NONE, a1, vece);
2983        break;
2984    case INDEX_op_shri_vec:
2985        tcg_out_insn(s, VRSa, VESRL, a0, a2, TCG_REG_NONE, a1, vece);
2986        break;
2987    case INDEX_op_sari_vec:
2988        tcg_out_insn(s, VRSa, VESRA, a0, a2, TCG_REG_NONE, a1, vece);
2989        break;
2990    case INDEX_op_rotli_vec:
2991        tcg_out_insn(s, VRSa, VERLL, a0, a2, TCG_REG_NONE, a1, vece);
2992        break;
2993    case INDEX_op_shls_vec:
2994        tcg_out_insn(s, VRSa, VESL, a0, 0, a2, a1, vece);
2995        break;
2996    case INDEX_op_shrs_vec:
2997        tcg_out_insn(s, VRSa, VESRL, a0, 0, a2, a1, vece);
2998        break;
2999    case INDEX_op_sars_vec:
3000        tcg_out_insn(s, VRSa, VESRA, a0, 0, a2, a1, vece);
3001        break;
3002    case INDEX_op_rotls_vec:
3003        tcg_out_insn(s, VRSa, VERLL, a0, 0, a2, a1, vece);
3004        break;
3005    case INDEX_op_shlv_vec:
3006        tcg_out_insn(s, VRRc, VESLV, a0, a1, a2, vece);
3007        break;
3008    case INDEX_op_shrv_vec:
3009        tcg_out_insn(s, VRRc, VESRLV, a0, a1, a2, vece);
3010        break;
3011    case INDEX_op_sarv_vec:
3012        tcg_out_insn(s, VRRc, VESRAV, a0, a1, a2, vece);
3013        break;
3014    case INDEX_op_rotlv_vec:
3015        tcg_out_insn(s, VRRc, VERLLV, a0, a1, a2, vece);
3016        break;
3017
3018    case INDEX_op_smin_vec:
3019        tcg_out_insn(s, VRRc, VMN, a0, a1, a2, vece);
3020        break;
3021    case INDEX_op_smax_vec:
3022        tcg_out_insn(s, VRRc, VMX, a0, a1, a2, vece);
3023        break;
3024    case INDEX_op_umin_vec:
3025        tcg_out_insn(s, VRRc, VMNL, a0, a1, a2, vece);
3026        break;
3027    case INDEX_op_umax_vec:
3028        tcg_out_insn(s, VRRc, VMXL, a0, a1, a2, vece);
3029        break;
3030
3031    case INDEX_op_bitsel_vec:
3032        tcg_out_insn(s, VRRe, VSEL, a0, a2, args[3], a1);
3033        break;
3034
3035    case INDEX_op_cmp_vec:
3036        tcg_out_cmp_vec(s, vece, a0, a1, a2, args[3]);
3037        break;
3038    case INDEX_op_cmpsel_vec:
3039        tcg_out_cmpsel_vec(s, vece, a0, a1, a2, args[3], args[4], args[5]);
3040        break;
3041
3042    case INDEX_op_s390_vuph_vec:
3043        tcg_out_insn(s, VRRa, VUPH, a0, a1, vece);
3044        break;
3045    case INDEX_op_s390_vupl_vec:
3046        tcg_out_insn(s, VRRa, VUPL, a0, a1, vece);
3047        break;
3048    case INDEX_op_s390_vpks_vec:
3049        tcg_out_insn(s, VRRc, VPKS, a0, a1, a2, vece);
3050        break;
3051
3052    case INDEX_op_mov_vec:   /* Always emitted via tcg_out_mov.  */
3053    case INDEX_op_dup_vec:   /* Always emitted via tcg_out_dup_vec.  */
3054    default:
3055        g_assert_not_reached();
3056    }
3057}
3058
3059int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
3060{
3061    switch (opc) {
3062    case INDEX_op_abs_vec:
3063    case INDEX_op_add_vec:
3064    case INDEX_op_and_vec:
3065    case INDEX_op_andc_vec:
3066    case INDEX_op_bitsel_vec:
3067    case INDEX_op_eqv_vec:
3068    case INDEX_op_nand_vec:
3069    case INDEX_op_neg_vec:
3070    case INDEX_op_nor_vec:
3071    case INDEX_op_not_vec:
3072    case INDEX_op_or_vec:
3073    case INDEX_op_orc_vec:
3074    case INDEX_op_rotli_vec:
3075    case INDEX_op_rotls_vec:
3076    case INDEX_op_rotlv_vec:
3077    case INDEX_op_sari_vec:
3078    case INDEX_op_sars_vec:
3079    case INDEX_op_sarv_vec:
3080    case INDEX_op_shli_vec:
3081    case INDEX_op_shls_vec:
3082    case INDEX_op_shlv_vec:
3083    case INDEX_op_shri_vec:
3084    case INDEX_op_shrs_vec:
3085    case INDEX_op_shrv_vec:
3086    case INDEX_op_smax_vec:
3087    case INDEX_op_smin_vec:
3088    case INDEX_op_sub_vec:
3089    case INDEX_op_umax_vec:
3090    case INDEX_op_umin_vec:
3091    case INDEX_op_xor_vec:
3092    case INDEX_op_cmp_vec:
3093    case INDEX_op_cmpsel_vec:
3094        return 1;
3095    case INDEX_op_rotrv_vec:
3096        return -1;
3097    case INDEX_op_mul_vec:
3098        return vece < MO_64;
3099    case INDEX_op_ssadd_vec:
3100    case INDEX_op_sssub_vec:
3101        return vece < MO_64 ? -1 : 0;
3102    default:
3103        return 0;
3104    }
3105}
3106
3107static void expand_vec_sat(TCGType type, unsigned vece, TCGv_vec v0,
3108                           TCGv_vec v1, TCGv_vec v2, TCGOpcode add_sub_opc)
3109{
3110    TCGv_vec h1 = tcg_temp_new_vec(type);
3111    TCGv_vec h2 = tcg_temp_new_vec(type);
3112    TCGv_vec l1 = tcg_temp_new_vec(type);
3113    TCGv_vec l2 = tcg_temp_new_vec(type);
3114
3115    tcg_debug_assert (vece < MO_64);
3116
3117    /* Unpack with sign-extension. */
3118    vec_gen_2(INDEX_op_s390_vuph_vec, type, vece,
3119              tcgv_vec_arg(h1), tcgv_vec_arg(v1));
3120    vec_gen_2(INDEX_op_s390_vuph_vec, type, vece,
3121              tcgv_vec_arg(h2), tcgv_vec_arg(v2));
3122
3123    vec_gen_2(INDEX_op_s390_vupl_vec, type, vece,
3124              tcgv_vec_arg(l1), tcgv_vec_arg(v1));
3125    vec_gen_2(INDEX_op_s390_vupl_vec, type, vece,
3126              tcgv_vec_arg(l2), tcgv_vec_arg(v2));
3127
3128    /* Arithmetic on a wider element size. */
3129    vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(h1),
3130              tcgv_vec_arg(h1), tcgv_vec_arg(h2));
3131    vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(l1),
3132              tcgv_vec_arg(l1), tcgv_vec_arg(l2));
3133
3134    /* Pack with saturation. */
3135    vec_gen_3(INDEX_op_s390_vpks_vec, type, vece + 1,
3136              tcgv_vec_arg(v0), tcgv_vec_arg(h1), tcgv_vec_arg(l1));
3137
3138    tcg_temp_free_vec(h1);
3139    tcg_temp_free_vec(h2);
3140    tcg_temp_free_vec(l1);
3141    tcg_temp_free_vec(l2);
3142}
3143
3144void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
3145                       TCGArg a0, ...)
3146{
3147    va_list va;
3148    TCGv_vec v0, v1, v2, t0;
3149
3150    va_start(va, a0);
3151    v0 = temp_tcgv_vec(arg_temp(a0));
3152    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3153    v2 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3154
3155    switch (opc) {
3156    case INDEX_op_rotrv_vec:
3157        t0 = tcg_temp_new_vec(type);
3158        tcg_gen_neg_vec(vece, t0, v2);
3159        tcg_gen_rotlv_vec(vece, v0, v1, t0);
3160        tcg_temp_free_vec(t0);
3161        break;
3162
3163    case INDEX_op_ssadd_vec:
3164        expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_add_vec);
3165        break;
3166    case INDEX_op_sssub_vec:
3167        expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_sub_vec);
3168        break;
3169
3170    default:
3171        g_assert_not_reached();
3172    }
3173    va_end(va);
3174}
3175
3176static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
3177{
3178    switch (op) {
3179    case INDEX_op_goto_ptr:
3180        return C_O0_I1(r);
3181
3182    case INDEX_op_ld8u_i32:
3183    case INDEX_op_ld8u_i64:
3184    case INDEX_op_ld8s_i32:
3185    case INDEX_op_ld8s_i64:
3186    case INDEX_op_ld16u_i32:
3187    case INDEX_op_ld16u_i64:
3188    case INDEX_op_ld16s_i32:
3189    case INDEX_op_ld16s_i64:
3190    case INDEX_op_ld_i32:
3191    case INDEX_op_ld32u_i64:
3192    case INDEX_op_ld32s_i64:
3193    case INDEX_op_ld_i64:
3194        return C_O1_I1(r, r);
3195
3196    case INDEX_op_st8_i32:
3197    case INDEX_op_st8_i64:
3198    case INDEX_op_st16_i32:
3199    case INDEX_op_st16_i64:
3200    case INDEX_op_st_i32:
3201    case INDEX_op_st32_i64:
3202    case INDEX_op_st_i64:
3203        return C_O0_I2(r, r);
3204
3205    case INDEX_op_add_i32:
3206    case INDEX_op_add_i64:
3207    case INDEX_op_shl_i64:
3208    case INDEX_op_shr_i64:
3209    case INDEX_op_sar_i64:
3210    case INDEX_op_rotl_i32:
3211    case INDEX_op_rotl_i64:
3212    case INDEX_op_rotr_i32:
3213    case INDEX_op_rotr_i64:
3214    case INDEX_op_setcond_i32:
3215    case INDEX_op_negsetcond_i32:
3216        return C_O1_I2(r, r, ri);
3217    case INDEX_op_setcond_i64:
3218    case INDEX_op_negsetcond_i64:
3219        return C_O1_I2(r, r, rC);
3220
3221    case INDEX_op_clz_i64:
3222        return C_O1_I2(r, r, rI);
3223
3224    case INDEX_op_sub_i32:
3225    case INDEX_op_sub_i64:
3226    case INDEX_op_and_i32:
3227    case INDEX_op_or_i32:
3228    case INDEX_op_xor_i32:
3229        return C_O1_I2(r, r, ri);
3230    case INDEX_op_and_i64:
3231        return C_O1_I2(r, r, rNKR);
3232    case INDEX_op_or_i64:
3233    case INDEX_op_xor_i64:
3234        return C_O1_I2(r, r, rK);
3235
3236    case INDEX_op_andc_i32:
3237    case INDEX_op_orc_i32:
3238    case INDEX_op_eqv_i32:
3239        return C_O1_I2(r, r, ri);
3240    case INDEX_op_andc_i64:
3241        return C_O1_I2(r, r, rKR);
3242    case INDEX_op_orc_i64:
3243    case INDEX_op_eqv_i64:
3244        return C_O1_I2(r, r, rNK);
3245
3246    case INDEX_op_nand_i32:
3247    case INDEX_op_nand_i64:
3248    case INDEX_op_nor_i32:
3249    case INDEX_op_nor_i64:
3250        return C_O1_I2(r, r, r);
3251
3252    case INDEX_op_mul_i32:
3253        return (HAVE_FACILITY(MISC_INSN_EXT2)
3254                ? C_O1_I2(r, r, ri)
3255                : C_O1_I2(r, 0, ri));
3256    case INDEX_op_mul_i64:
3257        return (HAVE_FACILITY(MISC_INSN_EXT2)
3258                ? C_O1_I2(r, r, rJ)
3259                : C_O1_I2(r, 0, rJ));
3260
3261    case INDEX_op_shl_i32:
3262    case INDEX_op_shr_i32:
3263    case INDEX_op_sar_i32:
3264        return C_O1_I2(r, r, ri);
3265
3266    case INDEX_op_brcond_i32:
3267        return C_O0_I2(r, ri);
3268    case INDEX_op_brcond_i64:
3269        return C_O0_I2(r, rC);
3270
3271    case INDEX_op_bswap16_i32:
3272    case INDEX_op_bswap16_i64:
3273    case INDEX_op_bswap32_i32:
3274    case INDEX_op_bswap32_i64:
3275    case INDEX_op_bswap64_i64:
3276    case INDEX_op_neg_i32:
3277    case INDEX_op_neg_i64:
3278    case INDEX_op_not_i32:
3279    case INDEX_op_not_i64:
3280    case INDEX_op_ext8s_i32:
3281    case INDEX_op_ext8s_i64:
3282    case INDEX_op_ext8u_i32:
3283    case INDEX_op_ext8u_i64:
3284    case INDEX_op_ext16s_i32:
3285    case INDEX_op_ext16s_i64:
3286    case INDEX_op_ext16u_i32:
3287    case INDEX_op_ext16u_i64:
3288    case INDEX_op_ext32s_i64:
3289    case INDEX_op_ext32u_i64:
3290    case INDEX_op_ext_i32_i64:
3291    case INDEX_op_extu_i32_i64:
3292    case INDEX_op_extract_i32:
3293    case INDEX_op_extract_i64:
3294    case INDEX_op_ctpop_i32:
3295    case INDEX_op_ctpop_i64:
3296        return C_O1_I1(r, r);
3297
3298    case INDEX_op_qemu_ld_a32_i32:
3299    case INDEX_op_qemu_ld_a64_i32:
3300    case INDEX_op_qemu_ld_a32_i64:
3301    case INDEX_op_qemu_ld_a64_i64:
3302        return C_O1_I1(r, r);
3303    case INDEX_op_qemu_st_a32_i64:
3304    case INDEX_op_qemu_st_a64_i64:
3305    case INDEX_op_qemu_st_a32_i32:
3306    case INDEX_op_qemu_st_a64_i32:
3307        return C_O0_I2(r, r);
3308    case INDEX_op_qemu_ld_a32_i128:
3309    case INDEX_op_qemu_ld_a64_i128:
3310        return C_O2_I1(o, m, r);
3311    case INDEX_op_qemu_st_a32_i128:
3312    case INDEX_op_qemu_st_a64_i128:
3313        return C_O0_I3(o, m, r);
3314
3315    case INDEX_op_deposit_i32:
3316    case INDEX_op_deposit_i64:
3317        return C_O1_I2(r, rZ, r);
3318
3319    case INDEX_op_movcond_i32:
3320        return C_O1_I4(r, r, ri, rI, r);
3321    case INDEX_op_movcond_i64:
3322        return C_O1_I4(r, r, rC, rI, r);
3323
3324    case INDEX_op_div2_i32:
3325    case INDEX_op_div2_i64:
3326    case INDEX_op_divu2_i32:
3327    case INDEX_op_divu2_i64:
3328        return C_O2_I3(o, m, 0, 1, r);
3329
3330    case INDEX_op_mulu2_i64:
3331        return C_O2_I2(o, m, 0, r);
3332    case INDEX_op_muls2_i64:
3333        return C_O2_I2(o, m, r, r);
3334
3335    case INDEX_op_add2_i32:
3336    case INDEX_op_sub2_i32:
3337        return C_N1_O1_I4(r, r, 0, 1, ri, r);
3338
3339    case INDEX_op_add2_i64:
3340    case INDEX_op_sub2_i64:
3341        return C_N1_O1_I4(r, r, 0, 1, rJU, r);
3342
3343    case INDEX_op_st_vec:
3344        return C_O0_I2(v, r);
3345    case INDEX_op_ld_vec:
3346    case INDEX_op_dupm_vec:
3347        return C_O1_I1(v, r);
3348    case INDEX_op_dup_vec:
3349        return C_O1_I1(v, vr);
3350    case INDEX_op_abs_vec:
3351    case INDEX_op_neg_vec:
3352    case INDEX_op_not_vec:
3353    case INDEX_op_rotli_vec:
3354    case INDEX_op_sari_vec:
3355    case INDEX_op_shli_vec:
3356    case INDEX_op_shri_vec:
3357    case INDEX_op_s390_vuph_vec:
3358    case INDEX_op_s390_vupl_vec:
3359        return C_O1_I1(v, v);
3360    case INDEX_op_add_vec:
3361    case INDEX_op_sub_vec:
3362    case INDEX_op_and_vec:
3363    case INDEX_op_andc_vec:
3364    case INDEX_op_or_vec:
3365    case INDEX_op_orc_vec:
3366    case INDEX_op_xor_vec:
3367    case INDEX_op_nand_vec:
3368    case INDEX_op_nor_vec:
3369    case INDEX_op_eqv_vec:
3370    case INDEX_op_cmp_vec:
3371    case INDEX_op_mul_vec:
3372    case INDEX_op_rotlv_vec:
3373    case INDEX_op_rotrv_vec:
3374    case INDEX_op_shlv_vec:
3375    case INDEX_op_shrv_vec:
3376    case INDEX_op_sarv_vec:
3377    case INDEX_op_smax_vec:
3378    case INDEX_op_smin_vec:
3379    case INDEX_op_umax_vec:
3380    case INDEX_op_umin_vec:
3381    case INDEX_op_s390_vpks_vec:
3382        return C_O1_I2(v, v, v);
3383    case INDEX_op_rotls_vec:
3384    case INDEX_op_shls_vec:
3385    case INDEX_op_shrs_vec:
3386    case INDEX_op_sars_vec:
3387        return C_O1_I2(v, v, r);
3388    case INDEX_op_bitsel_vec:
3389        return C_O1_I3(v, v, v, v);
3390    case INDEX_op_cmpsel_vec:
3391        return C_O1_I4(v, v, v, v, v);
3392
3393    default:
3394        g_assert_not_reached();
3395    }
3396}
3397
3398/*
3399 * Mainline glibc added HWCAP_S390_VX before it was kernel abi.
3400 * Some distros have fixed this up locally, others have not.
3401 */
3402#ifndef HWCAP_S390_VXRS
3403#define HWCAP_S390_VXRS 2048
3404#endif
3405
3406static void query_s390_facilities(void)
3407{
3408    unsigned long hwcap = qemu_getauxval(AT_HWCAP);
3409    const char *which;
3410
3411    /* Is STORE FACILITY LIST EXTENDED available?  Honestly, I believe this
3412       is present on all 64-bit systems, but let's check for it anyway.  */
3413    if (hwcap & HWCAP_S390_STFLE) {
3414        register int r0 __asm__("0") = ARRAY_SIZE(s390_facilities) - 1;
3415        register void *r1 __asm__("1") = s390_facilities;
3416
3417        /* stfle 0(%r1) */
3418        asm volatile(".word 0xb2b0,0x1000"
3419                     : "=r"(r0) : "r"(r0), "r"(r1) : "memory", "cc");
3420    }
3421
3422    /*
3423     * Use of vector registers requires os support beyond the facility bit.
3424     * If the kernel does not advertise support, disable the facility bits.
3425     * There is nothing else we currently care about in the 3rd word, so
3426     * disable VECTOR with one store.
3427     */
3428    if (!(hwcap & HWCAP_S390_VXRS)) {
3429        s390_facilities[2] = 0;
3430    }
3431
3432    /*
3433     * Minimum supported cpu revision is z196.
3434     * Check for all required facilities.
3435     * ZARCH_ACTIVE is done via preprocessor check for 64-bit.
3436     */
3437    if (!HAVE_FACILITY(LONG_DISP)) {
3438        which = "long-displacement";
3439        goto fail;
3440    }
3441    if (!HAVE_FACILITY(EXT_IMM)) {
3442        which = "extended-immediate";
3443        goto fail;
3444    }
3445    if (!HAVE_FACILITY(GEN_INST_EXT)) {
3446        which = "general-instructions-extension";
3447        goto fail;
3448    }
3449    /*
3450     * Facility 45 is a big bin that contains: distinct-operands,
3451     * fast-BCR-serialization, high-word, population-count,
3452     * interlocked-access-1, and load/store-on-condition-1
3453     */
3454    if (!HAVE_FACILITY(45)) {
3455        which = "45";
3456        goto fail;
3457    }
3458    return;
3459
3460 fail:
3461    error_report("%s: missing required facility %s", __func__, which);
3462    exit(EXIT_FAILURE);
3463}
3464
3465static void tcg_target_init(TCGContext *s)
3466{
3467    query_s390_facilities();
3468
3469    tcg_target_available_regs[TCG_TYPE_I32] = 0xffff;
3470    tcg_target_available_regs[TCG_TYPE_I64] = 0xffff;
3471    if (HAVE_FACILITY(VECTOR)) {
3472        tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull;
3473        tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull;
3474    }
3475
3476    tcg_target_call_clobber_regs = 0;
3477    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
3478    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1);
3479    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
3480    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
3481    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
3482    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
3483    /* The r6 register is technically call-saved, but it's also a parameter
3484       register, so it can get killed by setup for the qemu_st helper.  */
3485    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
3486    /* The return register can be considered call-clobbered.  */
3487    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14);
3488
3489    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0);
3490    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1);
3491    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V2);
3492    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V3);
3493    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V4);
3494    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V5);
3495    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V6);
3496    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V7);
3497    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V16);
3498    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V17);
3499    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V18);
3500    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V19);
3501    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V20);
3502    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V21);
3503    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V22);
3504    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V23);
3505    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V24);
3506    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V25);
3507    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V26);
3508    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V27);
3509    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V28);
3510    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V29);
3511    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V30);
3512    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V31);
3513
3514    s->reserved_regs = 0;
3515    tcg_regset_set_reg(s->reserved_regs, TCG_TMP0);
3516    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP0);
3517    /* XXX many insns can't be used with R0, so we better avoid it for now */
3518    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0);
3519    tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
3520}
3521
3522#define FRAME_SIZE  ((int)(TCG_TARGET_CALL_STACK_OFFSET          \
3523                           + TCG_STATIC_CALL_ARGS_SIZE           \
3524                           + CPU_TEMP_BUF_NLONGS * sizeof(long)))
3525
3526static void tcg_target_qemu_prologue(TCGContext *s)
3527{
3528    /* stmg %r6,%r15,48(%r15) (save registers) */
3529    tcg_out_insn(s, RXY, STMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15, 48);
3530
3531    /* aghi %r15,-frame_size */
3532    tcg_out_insn(s, RI, AGHI, TCG_REG_R15, -FRAME_SIZE);
3533
3534    tcg_set_frame(s, TCG_REG_CALL_STACK,
3535                  TCG_STATIC_CALL_ARGS_SIZE + TCG_TARGET_CALL_STACK_OFFSET,
3536                  CPU_TEMP_BUF_NLONGS * sizeof(long));
3537
3538    if (!tcg_use_softmmu && guest_base >= 0x80000) {
3539        tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base);
3540        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
3541    }
3542
3543    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
3544
3545    /* br %r3 (go to TB) */
3546    tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, tcg_target_call_iarg_regs[1]);
3547
3548    /*
3549     * Return path for goto_ptr. Set return value to 0, a-la exit_tb,
3550     * and fall through to the rest of the epilogue.
3551     */
3552    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
3553    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, 0);
3554
3555    /* TB epilogue */
3556    tb_ret_addr = tcg_splitwx_to_rx(s->code_ptr);
3557
3558    /* lmg %r6,%r15,fs+48(%r15) (restore registers) */
3559    tcg_out_insn(s, RXY, LMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15,
3560                 FRAME_SIZE + 48);
3561
3562    /* br %r14 (return) */
3563    tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_R14);
3564}
3565
3566static void tcg_out_tb_start(TCGContext *s)
3567{
3568    /* nothing to do */
3569}
3570
3571static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
3572{
3573    memset(p, 0x07, count * sizeof(tcg_insn_unit));
3574}
3575
3576typedef struct {
3577    DebugFrameHeader h;
3578    uint8_t fde_def_cfa[4];
3579    uint8_t fde_reg_ofs[18];
3580} DebugFrame;
3581
3582/* We're expecting a 2 byte uleb128 encoded value.  */
3583QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3584
3585#define ELF_HOST_MACHINE  EM_S390
3586
3587static const DebugFrame debug_frame = {
3588    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3589    .h.cie.id = -1,
3590    .h.cie.version = 1,
3591    .h.cie.code_align = 1,
3592    .h.cie.data_align = 8,                /* sleb128 8 */
3593    .h.cie.return_column = TCG_REG_R14,
3594
3595    /* Total FDE size does not include the "len" member.  */
3596    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
3597
3598    .fde_def_cfa = {
3599        12, TCG_REG_CALL_STACK,         /* DW_CFA_def_cfa %r15, ... */
3600        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3601        (FRAME_SIZE >> 7)
3602    },
3603    .fde_reg_ofs = {
3604        0x86, 6,                        /* DW_CFA_offset, %r6, 48 */
3605        0x87, 7,                        /* DW_CFA_offset, %r7, 56 */
3606        0x88, 8,                        /* DW_CFA_offset, %r8, 64 */
3607        0x89, 9,                        /* DW_CFA_offset, %r92, 72 */
3608        0x8a, 10,                       /* DW_CFA_offset, %r10, 80 */
3609        0x8b, 11,                       /* DW_CFA_offset, %r11, 88 */
3610        0x8c, 12,                       /* DW_CFA_offset, %r12, 96 */
3611        0x8d, 13,                       /* DW_CFA_offset, %r13, 104 */
3612        0x8e, 14,                       /* DW_CFA_offset, %r14, 112 */
3613    }
3614};
3615
3616void tcg_register_jit(const void *buf, size_t buf_size)
3617{
3618    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3619}
3620