xref: /openbmc/qemu/tcg/s390x/tcg-target.c.inc (revision 4d8722183932d9502e405ae86b1889e1d8a475e5)
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/* Used for function call generation. */
32#define TCG_TARGET_STACK_ALIGN          8
33#define TCG_TARGET_CALL_STACK_OFFSET    160
34#define TCG_TARGET_CALL_ARG_I32         TCG_CALL_ARG_EXTEND
35#define TCG_TARGET_CALL_ARG_I64         TCG_CALL_ARG_NORMAL
36#define TCG_TARGET_CALL_ARG_I128        TCG_CALL_ARG_BY_REF
37#define TCG_TARGET_CALL_RET_I128        TCG_CALL_RET_BY_REF
38
39#define TCG_CT_CONST_S16        (1 << 8)
40#define TCG_CT_CONST_S32        (1 << 9)
41#define TCG_CT_CONST_U32        (1 << 10)
42#define TCG_CT_CONST_ZERO       (1 << 11)
43#define TCG_CT_CONST_P32        (1 << 12)
44#define TCG_CT_CONST_INV        (1 << 13)
45#define TCG_CT_CONST_INVRISBG   (1 << 14)
46#define TCG_CT_CONST_CMP        (1 << 15)
47#define TCG_CT_CONST_M1         (1 << 16)
48
49#define ALL_GENERAL_REGS     MAKE_64BIT_MASK(0, 16)
50#define ALL_VECTOR_REGS      MAKE_64BIT_MASK(32, 32)
51
52/* Several places within the instruction set 0 means "no register"
53   rather than TCG_REG_R0.  */
54#define TCG_REG_NONE    0
55
56/* A scratch register that may be be used throughout the backend.  */
57#define TCG_TMP0        TCG_REG_R1
58#define TCG_VEC_TMP0    TCG_REG_V31
59
60#define TCG_GUEST_BASE_REG TCG_REG_R13
61
62/* All of the following instructions are prefixed with their instruction
63   format, and are defined as 8- or 16-bit quantities, even when the two
64   halves of the 16-bit quantity may appear 32 bits apart in the insn.
65   This makes it easy to copy the values from the tables in Appendix B.  */
66typedef enum S390Opcode {
67    RIL_AFI     = 0xc209,
68    RIL_AGFI    = 0xc208,
69    RIL_ALFI    = 0xc20b,
70    RIL_ALGFI   = 0xc20a,
71    RIL_BRASL   = 0xc005,
72    RIL_BRCL    = 0xc004,
73    RIL_CFI     = 0xc20d,
74    RIL_CGFI    = 0xc20c,
75    RIL_CLFI    = 0xc20f,
76    RIL_CLGFI   = 0xc20e,
77    RIL_CLRL    = 0xc60f,
78    RIL_CLGRL   = 0xc60a,
79    RIL_CRL     = 0xc60d,
80    RIL_CGRL    = 0xc608,
81    RIL_IIHF    = 0xc008,
82    RIL_IILF    = 0xc009,
83    RIL_LARL    = 0xc000,
84    RIL_LGFI    = 0xc001,
85    RIL_LGRL    = 0xc408,
86    RIL_LLIHF   = 0xc00e,
87    RIL_LLILF   = 0xc00f,
88    RIL_LRL     = 0xc40d,
89    RIL_MSFI    = 0xc201,
90    RIL_MSGFI   = 0xc200,
91    RIL_NIHF    = 0xc00a,
92    RIL_NILF    = 0xc00b,
93    RIL_OIHF    = 0xc00c,
94    RIL_OILF    = 0xc00d,
95    RIL_SLFI    = 0xc205,
96    RIL_SLGFI   = 0xc204,
97    RIL_XIHF    = 0xc006,
98    RIL_XILF    = 0xc007,
99
100    RI_AGHI     = 0xa70b,
101    RI_AHI      = 0xa70a,
102    RI_BRC      = 0xa704,
103    RI_CHI      = 0xa70e,
104    RI_CGHI     = 0xa70f,
105    RI_IIHH     = 0xa500,
106    RI_IIHL     = 0xa501,
107    RI_IILH     = 0xa502,
108    RI_IILL     = 0xa503,
109    RI_LGHI     = 0xa709,
110    RI_LLIHH    = 0xa50c,
111    RI_LLIHL    = 0xa50d,
112    RI_LLILH    = 0xa50e,
113    RI_LLILL    = 0xa50f,
114    RI_MGHI     = 0xa70d,
115    RI_MHI      = 0xa70c,
116    RI_NIHH     = 0xa504,
117    RI_NIHL     = 0xa505,
118    RI_NILH     = 0xa506,
119    RI_NILL     = 0xa507,
120    RI_OIHH     = 0xa508,
121    RI_OIHL     = 0xa509,
122    RI_OILH     = 0xa50a,
123    RI_OILL     = 0xa50b,
124    RI_TMLL     = 0xa701,
125    RI_TMLH     = 0xa700,
126    RI_TMHL     = 0xa703,
127    RI_TMHH     = 0xa702,
128
129    RIEb_CGRJ    = 0xec64,
130    RIEb_CLGRJ   = 0xec65,
131    RIEb_CLRJ    = 0xec77,
132    RIEb_CRJ     = 0xec76,
133
134    RIEc_CGIJ    = 0xec7c,
135    RIEc_CIJ     = 0xec7e,
136    RIEc_CLGIJ   = 0xec7d,
137    RIEc_CLIJ    = 0xec7f,
138
139    RIEf_RISBG   = 0xec55,
140
141    RIEg_LOCGHI  = 0xec46,
142
143    RRE_AGR     = 0xb908,
144    RRE_ALGR    = 0xb90a,
145    RRE_ALCR    = 0xb998,
146    RRE_ALCGR   = 0xb988,
147    RRE_ALGFR   = 0xb91a,
148    RRE_CGR     = 0xb920,
149    RRE_CLGR    = 0xb921,
150    RRE_DLGR    = 0xb987,
151    RRE_DLR     = 0xb997,
152    RRE_DSGFR   = 0xb91d,
153    RRE_DSGR    = 0xb90d,
154    RRE_FLOGR   = 0xb983,
155    RRE_LGBR    = 0xb906,
156    RRE_LCGR    = 0xb903,
157    RRE_LGFR    = 0xb914,
158    RRE_LGHR    = 0xb907,
159    RRE_LGR     = 0xb904,
160    RRE_LLGCR   = 0xb984,
161    RRE_LLGFR   = 0xb916,
162    RRE_LLGHR   = 0xb985,
163    RRE_LRVR    = 0xb91f,
164    RRE_LRVGR   = 0xb90f,
165    RRE_LTGR    = 0xb902,
166    RRE_MLGR    = 0xb986,
167    RRE_MSGR    = 0xb90c,
168    RRE_MSR     = 0xb252,
169    RRE_NGR     = 0xb980,
170    RRE_OGR     = 0xb981,
171    RRE_SGR     = 0xb909,
172    RRE_SLGR    = 0xb90b,
173    RRE_SLBR    = 0xb999,
174    RRE_SLBGR   = 0xb989,
175    RRE_XGR     = 0xb982,
176
177    RRFa_MGRK   = 0xb9ec,
178    RRFa_MSRKC  = 0xb9fd,
179    RRFa_MSGRKC = 0xb9ed,
180    RRFa_NCRK   = 0xb9f5,
181    RRFa_NCGRK  = 0xb9e5,
182    RRFa_NNRK   = 0xb974,
183    RRFa_NNGRK  = 0xb964,
184    RRFa_NORK   = 0xb976,
185    RRFa_NOGRK  = 0xb966,
186    RRFa_NRK    = 0xb9f4,
187    RRFa_NGRK   = 0xb9e4,
188    RRFa_NXRK   = 0xb977,
189    RRFa_NXGRK  = 0xb967,
190    RRFa_OCRK   = 0xb975,
191    RRFa_OCGRK  = 0xb965,
192    RRFa_ORK    = 0xb9f6,
193    RRFa_OGRK   = 0xb9e6,
194    RRFa_SRK    = 0xb9f9,
195    RRFa_SGRK   = 0xb9e9,
196    RRFa_SLRK   = 0xb9fb,
197    RRFa_SLGRK  = 0xb9eb,
198    RRFa_XRK    = 0xb9f7,
199    RRFa_XGRK   = 0xb9e7,
200
201    RRFam_SELGR = 0xb9e3,
202
203    RRFc_LOCR   = 0xb9f2,
204    RRFc_LOCGR  = 0xb9e2,
205    RRFc_POPCNT = 0xb9e1,
206
207    RR_AR       = 0x1a,
208    RR_ALR      = 0x1e,
209    RR_BASR     = 0x0d,
210    RR_BCR      = 0x07,
211    RR_CLR      = 0x15,
212    RR_CR       = 0x19,
213    RR_DR       = 0x1d,
214    RR_LCR      = 0x13,
215    RR_LR       = 0x18,
216    RR_LTR      = 0x12,
217    RR_NR       = 0x14,
218    RR_OR       = 0x16,
219    RR_SR       = 0x1b,
220    RR_SLR      = 0x1f,
221    RR_XR       = 0x17,
222
223    RSY_RLL     = 0xeb1d,
224    RSY_RLLG    = 0xeb1c,
225    RSY_SLLG    = 0xeb0d,
226    RSY_SLLK    = 0xebdf,
227    RSY_SRAG    = 0xeb0a,
228    RSY_SRAK    = 0xebdc,
229    RSY_SRLG    = 0xeb0c,
230    RSY_SRLK    = 0xebde,
231
232    RS_SLL      = 0x89,
233    RS_SRA      = 0x8a,
234    RS_SRL      = 0x88,
235
236    RXY_AG      = 0xe308,
237    RXY_AY      = 0xe35a,
238    RXY_CG      = 0xe320,
239    RXY_CLG     = 0xe321,
240    RXY_CLY     = 0xe355,
241    RXY_CY      = 0xe359,
242    RXY_LAY     = 0xe371,
243    RXY_LB      = 0xe376,
244    RXY_LG      = 0xe304,
245    RXY_LGB     = 0xe377,
246    RXY_LGF     = 0xe314,
247    RXY_LGH     = 0xe315,
248    RXY_LHY     = 0xe378,
249    RXY_LLGC    = 0xe390,
250    RXY_LLGF    = 0xe316,
251    RXY_LLGH    = 0xe391,
252    RXY_LMG     = 0xeb04,
253    RXY_LPQ     = 0xe38f,
254    RXY_LRV     = 0xe31e,
255    RXY_LRVG    = 0xe30f,
256    RXY_LRVH    = 0xe31f,
257    RXY_LY      = 0xe358,
258    RXY_NG      = 0xe380,
259    RXY_OG      = 0xe381,
260    RXY_STCY    = 0xe372,
261    RXY_STG     = 0xe324,
262    RXY_STHY    = 0xe370,
263    RXY_STMG    = 0xeb24,
264    RXY_STPQ    = 0xe38e,
265    RXY_STRV    = 0xe33e,
266    RXY_STRVG   = 0xe32f,
267    RXY_STRVH   = 0xe33f,
268    RXY_STY     = 0xe350,
269    RXY_XG      = 0xe382,
270
271    RX_A        = 0x5a,
272    RX_C        = 0x59,
273    RX_L        = 0x58,
274    RX_LA       = 0x41,
275    RX_LH       = 0x48,
276    RX_ST       = 0x50,
277    RX_STC      = 0x42,
278    RX_STH      = 0x40,
279
280    VRIa_VGBM   = 0xe744,
281    VRIa_VREPI  = 0xe745,
282    VRIb_VGM    = 0xe746,
283    VRIc_VREP   = 0xe74d,
284
285    VRRa_VLC    = 0xe7de,
286    VRRa_VLP    = 0xe7df,
287    VRRa_VLR    = 0xe756,
288    VRRc_VA     = 0xe7f3,
289    VRRc_VCEQ   = 0xe7f8,   /* we leave the m5 cs field 0 */
290    VRRc_VCH    = 0xe7fb,   /* " */
291    VRRc_VCHL   = 0xe7f9,   /* " */
292    VRRc_VERLLV = 0xe773,
293    VRRc_VESLV  = 0xe770,
294    VRRc_VESRAV = 0xe77a,
295    VRRc_VESRLV = 0xe778,
296    VRRc_VML    = 0xe7a2,
297    VRRc_VMN    = 0xe7fe,
298    VRRc_VMNL   = 0xe7fc,
299    VRRc_VMX    = 0xe7ff,
300    VRRc_VMXL   = 0xe7fd,
301    VRRc_VN     = 0xe768,
302    VRRc_VNC    = 0xe769,
303    VRRc_VNN    = 0xe76e,
304    VRRc_VNO    = 0xe76b,
305    VRRc_VNX    = 0xe76c,
306    VRRc_VO     = 0xe76a,
307    VRRc_VOC    = 0xe76f,
308    VRRc_VPKS   = 0xe797,   /* we leave the m5 cs field 0 */
309    VRRc_VS     = 0xe7f7,
310    VRRa_VUPH   = 0xe7d7,
311    VRRa_VUPL   = 0xe7d6,
312    VRRc_VX     = 0xe76d,
313    VRRe_VSEL   = 0xe78d,
314    VRRf_VLVGP  = 0xe762,
315
316    VRSa_VERLL  = 0xe733,
317    VRSa_VESL   = 0xe730,
318    VRSa_VESRA  = 0xe73a,
319    VRSa_VESRL  = 0xe738,
320    VRSb_VLVG   = 0xe722,
321    VRSc_VLGV   = 0xe721,
322
323    VRX_VL      = 0xe706,
324    VRX_VLLEZ   = 0xe704,
325    VRX_VLREP   = 0xe705,
326    VRX_VST     = 0xe70e,
327    VRX_VSTEF   = 0xe70b,
328    VRX_VSTEG   = 0xe70a,
329
330    NOP         = 0x0707,
331} S390Opcode;
332
333#ifdef CONFIG_DEBUG_TCG
334static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
335    "%r0",  "%r1",  "%r2",  "%r3",  "%r4",  "%r5",  "%r6",  "%r7",
336    "%r8",  "%r9",  "%r10", "%r11", "%r12", "%r13", "%r14", "%r15",
337    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
338    "%v0",  "%v1",  "%v2",  "%v3",  "%v4",  "%v5",  "%v6",  "%v7",
339    "%v8",  "%v9",  "%v10", "%v11", "%v12", "%v13", "%v14", "%v15",
340    "%v16", "%v17", "%v18", "%v19", "%v20", "%v21", "%v22", "%v23",
341    "%v24", "%v25", "%v26", "%v27", "%v28", "%v29", "%v30", "%v31",
342};
343#endif
344
345/* Since R6 is a potential argument register, choose it last of the
346   call-saved registers.  Likewise prefer the call-clobbered registers
347   in reverse order to maximize the chance of avoiding the arguments.  */
348static const int tcg_target_reg_alloc_order[] = {
349    /* Call saved registers.  */
350    TCG_REG_R13,
351    TCG_REG_R12,
352    TCG_REG_R11,
353    TCG_REG_R10,
354    TCG_REG_R9,
355    TCG_REG_R8,
356    TCG_REG_R7,
357    TCG_REG_R6,
358    /* Call clobbered registers.  */
359    TCG_REG_R14,
360    TCG_REG_R0,
361    TCG_REG_R1,
362    /* Argument registers, in reverse order of allocation.  */
363    TCG_REG_R5,
364    TCG_REG_R4,
365    TCG_REG_R3,
366    TCG_REG_R2,
367
368    /* V8-V15 are call saved, and omitted. */
369    TCG_REG_V0,
370    TCG_REG_V1,
371    TCG_REG_V2,
372    TCG_REG_V3,
373    TCG_REG_V4,
374    TCG_REG_V5,
375    TCG_REG_V6,
376    TCG_REG_V7,
377    TCG_REG_V16,
378    TCG_REG_V17,
379    TCG_REG_V18,
380    TCG_REG_V19,
381    TCG_REG_V20,
382    TCG_REG_V21,
383    TCG_REG_V22,
384    TCG_REG_V23,
385    TCG_REG_V24,
386    TCG_REG_V25,
387    TCG_REG_V26,
388    TCG_REG_V27,
389    TCG_REG_V28,
390    TCG_REG_V29,
391    TCG_REG_V30,
392    TCG_REG_V31,
393};
394
395static const int tcg_target_call_iarg_regs[] = {
396    TCG_REG_R2,
397    TCG_REG_R3,
398    TCG_REG_R4,
399    TCG_REG_R5,
400    TCG_REG_R6,
401};
402
403static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
404{
405    tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
406    tcg_debug_assert(slot == 0);
407    return TCG_REG_R2;
408}
409
410#define S390_CC_EQ      8
411#define S390_CC_LT      4
412#define S390_CC_GT      2
413#define S390_CC_OV      1
414#define S390_CC_NE      (S390_CC_LT | S390_CC_GT)
415#define S390_CC_LE      (S390_CC_LT | S390_CC_EQ)
416#define S390_CC_GE      (S390_CC_GT | S390_CC_EQ)
417#define S390_CC_NEVER   0
418#define S390_CC_ALWAYS  15
419
420#define S390_TM_EQ      8  /* CC == 0 */
421#define S390_TM_NE      7  /* CC in {1,2,3} */
422
423/* Condition codes that result from a COMPARE and COMPARE LOGICAL.  */
424static const uint8_t tcg_cond_to_s390_cond[16] = {
425    [TCG_COND_EQ]  = S390_CC_EQ,
426    [TCG_COND_NE]  = S390_CC_NE,
427    [TCG_COND_TSTEQ] = S390_CC_EQ,
428    [TCG_COND_TSTNE] = S390_CC_NE,
429    [TCG_COND_LT]  = S390_CC_LT,
430    [TCG_COND_LE]  = S390_CC_LE,
431    [TCG_COND_GT]  = S390_CC_GT,
432    [TCG_COND_GE]  = S390_CC_GE,
433    [TCG_COND_LTU] = S390_CC_LT,
434    [TCG_COND_LEU] = S390_CC_LE,
435    [TCG_COND_GTU] = S390_CC_GT,
436    [TCG_COND_GEU] = S390_CC_GE,
437};
438
439/* Condition codes that result from a LOAD AND TEST.  Here, we have no
440   unsigned instruction variation, however since the test is vs zero we
441   can re-map the outcomes appropriately.  */
442static const uint8_t tcg_cond_to_ltr_cond[16] = {
443    [TCG_COND_EQ]  = S390_CC_EQ,
444    [TCG_COND_NE]  = S390_CC_NE,
445    [TCG_COND_TSTEQ] = S390_CC_ALWAYS,
446    [TCG_COND_TSTNE] = S390_CC_NEVER,
447    [TCG_COND_LT]  = S390_CC_LT,
448    [TCG_COND_LE]  = S390_CC_LE,
449    [TCG_COND_GT]  = S390_CC_GT,
450    [TCG_COND_GE]  = S390_CC_GE,
451    [TCG_COND_LTU] = S390_CC_NEVER,
452    [TCG_COND_LEU] = S390_CC_EQ,
453    [TCG_COND_GTU] = S390_CC_NE,
454    [TCG_COND_GEU] = S390_CC_ALWAYS,
455};
456
457static const tcg_insn_unit *tb_ret_addr;
458uint64_t s390_facilities[3];
459
460static inline bool is_general_reg(TCGReg r)
461{
462    return r <= TCG_REG_R15;
463}
464
465static inline bool is_vector_reg(TCGReg r)
466{
467    return r >= TCG_REG_V0 && r <= TCG_REG_V31;
468}
469
470static bool patch_reloc(tcg_insn_unit *src_rw, int type,
471                        intptr_t value, intptr_t addend)
472{
473    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
474    intptr_t pcrel2;
475    uint32_t old;
476
477    value += addend;
478    pcrel2 = (tcg_insn_unit *)value - src_rx;
479
480    switch (type) {
481    case R_390_PC16DBL:
482        if (pcrel2 == (int16_t)pcrel2) {
483            tcg_patch16(src_rw, pcrel2);
484            return true;
485        }
486        break;
487    case R_390_PC32DBL:
488        if (pcrel2 == (int32_t)pcrel2) {
489            tcg_patch32(src_rw, pcrel2);
490            return true;
491        }
492        break;
493    case R_390_20:
494        if (value == sextract64(value, 0, 20)) {
495            old = *(uint32_t *)src_rw & 0xf00000ff;
496            old |= ((value & 0xfff) << 16) | ((value & 0xff000) >> 4);
497            tcg_patch32(src_rw, old);
498            return true;
499        }
500        break;
501    default:
502        g_assert_not_reached();
503    }
504    return false;
505}
506
507static int is_const_p16(uint64_t val)
508{
509    for (int i = 0; i < 4; ++i) {
510        uint64_t mask = 0xffffull << (i * 16);
511        if ((val & ~mask) == 0) {
512            return i;
513        }
514    }
515    return -1;
516}
517
518static int is_const_p32(uint64_t val)
519{
520    if ((val & 0xffffffff00000000ull) == 0) {
521        return 0;
522    }
523    if ((val & 0x00000000ffffffffull) == 0) {
524        return 1;
525    }
526    return -1;
527}
528
529/*
530 * Accept bit patterns like these:
531 *  0....01....1
532 *  1....10....0
533 *  1..10..01..1
534 *  0..01..10..0
535 * Copied from gcc sources.
536 */
537static bool risbg_mask(uint64_t c)
538{
539    uint64_t lsb;
540    /* We don't change the number of transitions by inverting,
541       so make sure we start with the LSB zero.  */
542    if (c & 1) {
543        c = ~c;
544    }
545    /* Reject all zeros or all ones.  */
546    if (c == 0) {
547        return false;
548    }
549    /* Find the first transition.  */
550    lsb = c & -c;
551    /* Invert to look for a second transition.  */
552    c = ~c;
553    /* Erase the first transition.  */
554    c &= -lsb;
555    /* Find the second transition, if any.  */
556    lsb = c & -c;
557    /* Match if all the bits are 1's, or if c is zero.  */
558    return c == -lsb;
559}
560
561/* Test if a constant matches the constraint. */
562static bool tcg_target_const_match(int64_t val, int ct,
563                                   TCGType type, TCGCond cond, int vece)
564{
565    uint64_t uval = val;
566
567    if (ct & TCG_CT_CONST) {
568        return true;
569    }
570    if (type == TCG_TYPE_I32) {
571        uval = (uint32_t)val;
572        val = (int32_t)val;
573    }
574
575    if (ct & TCG_CT_CONST_CMP) {
576        if (is_tst_cond(cond)) {
577            if (is_const_p16(uval) >= 0) {
578                return true;  /* TMxx */
579            }
580            if (risbg_mask(uval)) {
581                return true;  /* RISBG */
582            }
583            return false;
584        }
585
586        if (type == TCG_TYPE_I32) {
587            return true;
588        }
589
590        switch (cond) {
591        case TCG_COND_EQ:
592        case TCG_COND_NE:
593            ct |= TCG_CT_CONST_S32 | TCG_CT_CONST_U32;  /* CGFI or CLGFI */
594            break;
595        case TCG_COND_LT:
596        case TCG_COND_GE:
597        case TCG_COND_LE:
598        case TCG_COND_GT:
599            ct |= TCG_CT_CONST_S32;  /* CGFI */
600            break;
601        case TCG_COND_LTU:
602        case TCG_COND_GEU:
603        case TCG_COND_LEU:
604        case TCG_COND_GTU:
605            ct |= TCG_CT_CONST_U32;  /* CLGFI */
606            break;
607        case TCG_COND_TSTNE:
608        case TCG_COND_TSTEQ:
609            /* checked above, fallthru */
610        default:
611            g_assert_not_reached();
612        }
613    }
614
615    if ((ct & TCG_CT_CONST_S32) && val == (int32_t)val) {
616        return true;
617    }
618    if ((ct & TCG_CT_CONST_U32) && val == (uint32_t)val) {
619        return true;
620    }
621    if ((ct & TCG_CT_CONST_S16) && val == (int16_t)val) {
622        return true;
623    }
624    if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
625        return true;
626    }
627    if ((ct & TCG_CT_CONST_M1) && val == -1) {
628        return true;
629    }
630
631    if (ct & TCG_CT_CONST_INV) {
632        val = ~val;
633    }
634    if ((ct & TCG_CT_CONST_P32) && is_const_p32(val) >= 0) {
635        return true;
636    }
637    if ((ct & TCG_CT_CONST_INVRISBG) && risbg_mask(~val)) {
638        return true;
639    }
640    return false;
641}
642
643/* Emit instructions according to the given instruction format.  */
644
645static void tcg_out_insn_RR(TCGContext *s, S390Opcode op, TCGReg r1, TCGReg r2)
646{
647    tcg_out16(s, (op << 8) | (r1 << 4) | r2);
648}
649
650static void tcg_out_insn_RRE(TCGContext *s, S390Opcode op,
651                             TCGReg r1, TCGReg r2)
652{
653    tcg_out32(s, (op << 16) | (r1 << 4) | r2);
654}
655
656/* RRF-a without the m4 field */
657static void tcg_out_insn_RRFa(TCGContext *s, S390Opcode op,
658                              TCGReg r1, TCGReg r2, TCGReg r3)
659{
660    tcg_out32(s, (op << 16) | (r3 << 12) | (r1 << 4) | r2);
661}
662
663/* RRF-a with the m4 field */
664static void tcg_out_insn_RRFam(TCGContext *s, S390Opcode op,
665                               TCGReg r1, TCGReg r2, TCGReg r3, int m4)
666{
667    tcg_out32(s, (op << 16) | (r3 << 12) | (m4 << 8) | (r1 << 4) | r2);
668}
669
670static void tcg_out_insn_RRFc(TCGContext *s, S390Opcode op,
671                              TCGReg r1, TCGReg r2, int m3)
672{
673    tcg_out32(s, (op << 16) | (m3 << 12) | (r1 << 4) | r2);
674}
675
676static void tcg_out_insn_RI(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
677{
678    tcg_out32(s, (op << 16) | (r1 << 20) | (i2 & 0xffff));
679}
680
681static void tcg_out_insn_RIEg(TCGContext *s, S390Opcode op, TCGReg r1,
682                             int i2, int m3)
683{
684    tcg_out16(s, (op & 0xff00) | (r1 << 4) | m3);
685    tcg_out32(s, (i2 << 16) | (op & 0xff));
686}
687
688static void tcg_out_insn_RIL(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
689{
690    tcg_out16(s, op | (r1 << 4));
691    tcg_out32(s, i2);
692}
693
694static void tcg_out_insn_RS(TCGContext *s, S390Opcode op, TCGReg r1,
695                            TCGReg b2, TCGReg r3, int disp)
696{
697    tcg_out32(s, (op << 24) | (r1 << 20) | (r3 << 16) | (b2 << 12)
698              | (disp & 0xfff));
699}
700
701static void tcg_out_insn_RSY(TCGContext *s, S390Opcode op, TCGReg r1,
702                             TCGReg b2, TCGReg r3, int disp)
703{
704    tcg_out16(s, (op & 0xff00) | (r1 << 4) | r3);
705    tcg_out32(s, (op & 0xff) | (b2 << 28)
706              | ((disp & 0xfff) << 16) | ((disp & 0xff000) >> 4));
707}
708
709#define tcg_out_insn_RX   tcg_out_insn_RS
710#define tcg_out_insn_RXY  tcg_out_insn_RSY
711
712static int RXB(TCGReg v1, TCGReg v2, TCGReg v3, TCGReg v4)
713{
714    /*
715     * Shift bit 4 of each regno to its corresponding bit of RXB.
716     * RXB itself begins at bit 8 of the instruction so 8 - 4 = 4
717     * is the left-shift of the 4th operand.
718     */
719    return ((v1 & 0x10) << (4 + 3))
720         | ((v2 & 0x10) << (4 + 2))
721         | ((v3 & 0x10) << (4 + 1))
722         | ((v4 & 0x10) << (4 + 0));
723}
724
725static void tcg_out_insn_VRIa(TCGContext *s, S390Opcode op,
726                              TCGReg v1, uint16_t i2, int m3)
727{
728    tcg_debug_assert(is_vector_reg(v1));
729    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4));
730    tcg_out16(s, i2);
731    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m3 << 12));
732}
733
734static void tcg_out_insn_VRIb(TCGContext *s, S390Opcode op,
735                              TCGReg v1, uint8_t i2, uint8_t i3, int m4)
736{
737    tcg_debug_assert(is_vector_reg(v1));
738    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4));
739    tcg_out16(s, (i2 << 8) | (i3 & 0xff));
740    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m4 << 12));
741}
742
743static void tcg_out_insn_VRIc(TCGContext *s, S390Opcode op,
744                              TCGReg v1, uint16_t i2, TCGReg v3, int m4)
745{
746    tcg_debug_assert(is_vector_reg(v1));
747    tcg_debug_assert(is_vector_reg(v3));
748    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v3 & 0xf));
749    tcg_out16(s, i2);
750    tcg_out16(s, (op & 0x00ff) | RXB(v1, v3, 0, 0) | (m4 << 12));
751}
752
753static void tcg_out_insn_VRRa(TCGContext *s, S390Opcode op,
754                              TCGReg v1, TCGReg v2, int m3)
755{
756    tcg_debug_assert(is_vector_reg(v1));
757    tcg_debug_assert(is_vector_reg(v2));
758    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
759    tcg_out32(s, (op & 0x00ff) | RXB(v1, v2, 0, 0) | (m3 << 12));
760}
761
762static void tcg_out_insn_VRRc(TCGContext *s, S390Opcode op,
763                              TCGReg v1, TCGReg v2, TCGReg v3, int m4)
764{
765    tcg_debug_assert(is_vector_reg(v1));
766    tcg_debug_assert(is_vector_reg(v2));
767    tcg_debug_assert(is_vector_reg(v3));
768    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
769    tcg_out16(s, v3 << 12);
770    tcg_out16(s, (op & 0x00ff) | RXB(v1, v2, v3, 0) | (m4 << 12));
771}
772
773static void tcg_out_insn_VRRe(TCGContext *s, S390Opcode op,
774                              TCGReg v1, TCGReg v2, TCGReg v3, TCGReg v4)
775{
776    tcg_debug_assert(is_vector_reg(v1));
777    tcg_debug_assert(is_vector_reg(v2));
778    tcg_debug_assert(is_vector_reg(v3));
779    tcg_debug_assert(is_vector_reg(v4));
780    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
781    tcg_out16(s, v3 << 12);
782    tcg_out16(s, (op & 0x00ff) | RXB(v1, v2, v3, v4) | (v4 << 12));
783}
784
785static void tcg_out_insn_VRRf(TCGContext *s, S390Opcode op,
786                              TCGReg v1, TCGReg r2, TCGReg r3)
787{
788    tcg_debug_assert(is_vector_reg(v1));
789    tcg_debug_assert(is_general_reg(r2));
790    tcg_debug_assert(is_general_reg(r3));
791    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | r2);
792    tcg_out16(s, r3 << 12);
793    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0));
794}
795
796static void tcg_out_insn_VRSa(TCGContext *s, S390Opcode op, TCGReg v1,
797                              intptr_t d2, TCGReg b2, TCGReg v3, int m4)
798{
799    tcg_debug_assert(is_vector_reg(v1));
800    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
801    tcg_debug_assert(is_general_reg(b2));
802    tcg_debug_assert(is_vector_reg(v3));
803    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v3 & 0xf));
804    tcg_out16(s, b2 << 12 | d2);
805    tcg_out16(s, (op & 0x00ff) | RXB(v1, v3, 0, 0) | (m4 << 12));
806}
807
808static void tcg_out_insn_VRSb(TCGContext *s, S390Opcode op, TCGReg v1,
809                              intptr_t d2, TCGReg b2, TCGReg r3, int m4)
810{
811    tcg_debug_assert(is_vector_reg(v1));
812    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
813    tcg_debug_assert(is_general_reg(b2));
814    tcg_debug_assert(is_general_reg(r3));
815    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | r3);
816    tcg_out16(s, b2 << 12 | d2);
817    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m4 << 12));
818}
819
820static void tcg_out_insn_VRSc(TCGContext *s, S390Opcode op, TCGReg r1,
821                              intptr_t d2, TCGReg b2, TCGReg v3, int m4)
822{
823    tcg_debug_assert(is_general_reg(r1));
824    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
825    tcg_debug_assert(is_general_reg(b2));
826    tcg_debug_assert(is_vector_reg(v3));
827    tcg_out16(s, (op & 0xff00) | (r1 << 4) | (v3 & 0xf));
828    tcg_out16(s, b2 << 12 | d2);
829    tcg_out16(s, (op & 0x00ff) | RXB(0, v3, 0, 0) | (m4 << 12));
830}
831
832static void tcg_out_insn_VRX(TCGContext *s, S390Opcode op, TCGReg v1,
833                             TCGReg b2, TCGReg x2, intptr_t d2, int m3)
834{
835    tcg_debug_assert(is_vector_reg(v1));
836    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
837    tcg_debug_assert(is_general_reg(x2));
838    tcg_debug_assert(is_general_reg(b2));
839    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | x2);
840    tcg_out16(s, (b2 << 12) | d2);
841    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m3 << 12));
842}
843
844/* Emit an opcode with "type-checking" of the format.  */
845#define tcg_out_insn(S, FMT, OP, ...) \
846    glue(tcg_out_insn_,FMT)(S, glue(glue(FMT,_),OP), ## __VA_ARGS__)
847
848
849/* emit 64-bit shifts */
850static void tcg_out_sh64(TCGContext* s, S390Opcode op, TCGReg dest,
851                         TCGReg src, TCGReg sh_reg, int sh_imm)
852{
853    tcg_out_insn_RSY(s, op, dest, sh_reg, src, sh_imm);
854}
855
856/* emit 32-bit shifts */
857static void tcg_out_sh32(TCGContext* s, S390Opcode op, TCGReg dest,
858                         TCGReg sh_reg, int sh_imm)
859{
860    tcg_out_insn_RS(s, op, dest, sh_reg, 0, sh_imm);
861}
862
863static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg dst, TCGReg src)
864{
865    if (src == dst) {
866        return true;
867    }
868    switch (type) {
869    case TCG_TYPE_I32:
870        if (likely(is_general_reg(dst) && is_general_reg(src))) {
871            tcg_out_insn(s, RR, LR, dst, src);
872            break;
873        }
874        /* fallthru */
875
876    case TCG_TYPE_I64:
877        if (likely(is_general_reg(dst))) {
878            if (likely(is_general_reg(src))) {
879                tcg_out_insn(s, RRE, LGR, dst, src);
880            } else {
881                tcg_out_insn(s, VRSc, VLGV, dst, 0, 0, src, 3);
882            }
883            break;
884        } else if (is_general_reg(src)) {
885            tcg_out_insn(s, VRSb, VLVG, dst, 0, 0, src, 3);
886            break;
887        }
888        /* fallthru */
889
890    case TCG_TYPE_V64:
891    case TCG_TYPE_V128:
892        tcg_out_insn(s, VRRa, VLR, dst, src, 0);
893        break;
894
895    default:
896        g_assert_not_reached();
897    }
898    return true;
899}
900
901static const S390Opcode li_insns[4] = {
902    RI_LLILL, RI_LLILH, RI_LLIHL, RI_LLIHH
903};
904static const S390Opcode oi_insns[4] = {
905    RI_OILL, RI_OILH, RI_OIHL, RI_OIHH
906};
907static const S390Opcode lif_insns[2] = {
908    RIL_LLILF, RIL_LLIHF,
909};
910static const S390Opcode tm_insns[4] = {
911    RI_TMLL, RI_TMLH, RI_TMHL, RI_TMHH
912};
913
914/* load a register with an immediate value */
915static void tcg_out_movi(TCGContext *s, TCGType type,
916                         TCGReg ret, tcg_target_long sval)
917{
918    tcg_target_ulong uval = sval;
919    ptrdiff_t pc_off;
920    int i;
921
922    if (type == TCG_TYPE_I32) {
923        uval = (uint32_t)sval;
924        sval = (int32_t)sval;
925    }
926
927    /* Try all 32-bit insns that can load it in one go.  */
928    if (sval >= -0x8000 && sval < 0x8000) {
929        tcg_out_insn(s, RI, LGHI, ret, sval);
930        return;
931    }
932
933    i = is_const_p16(uval);
934    if (i >= 0) {
935        tcg_out_insn_RI(s, li_insns[i], ret, uval >> (i * 16));
936        return;
937    }
938
939    /* Try all 48-bit insns that can load it in one go.  */
940    if (sval == (int32_t)sval) {
941        tcg_out_insn(s, RIL, LGFI, ret, sval);
942        return;
943    }
944
945    i = is_const_p32(uval);
946    if (i >= 0) {
947        tcg_out_insn_RIL(s, lif_insns[i], ret, uval >> (i * 32));
948        return;
949    }
950
951    /* Try for PC-relative address load.  For odd addresses, add one. */
952    pc_off = tcg_pcrel_diff(s, (void *)sval) >> 1;
953    if (pc_off == (int32_t)pc_off) {
954        tcg_out_insn(s, RIL, LARL, ret, pc_off);
955        if (sval & 1) {
956            tcg_out_insn(s, RI, AGHI, ret, 1);
957        }
958        return;
959    }
960
961    /* Otherwise, load it by parts. */
962    i = is_const_p16((uint32_t)uval);
963    if (i >= 0) {
964        tcg_out_insn_RI(s, li_insns[i], ret, uval >> (i * 16));
965    } else {
966        tcg_out_insn(s, RIL, LLILF, ret, uval);
967    }
968    uval >>= 32;
969    i = is_const_p16(uval);
970    if (i >= 0) {
971        tcg_out_insn_RI(s, oi_insns[i + 2], ret, uval >> (i * 16));
972    } else {
973        tcg_out_insn(s, RIL, OIHF, ret, uval);
974    }
975}
976
977/* Emit a load/store type instruction.  Inputs are:
978   DATA:     The register to be loaded or stored.
979   BASE+OFS: The effective address.
980   OPC_RX:   If the operation has an RX format opcode (e.g. STC), otherwise 0.
981   OPC_RXY:  The RXY format opcode for the operation (e.g. STCY).  */
982
983static void tcg_out_mem(TCGContext *s, S390Opcode opc_rx, S390Opcode opc_rxy,
984                        TCGReg data, TCGReg base, TCGReg index,
985                        tcg_target_long ofs)
986{
987    if (ofs < -0x80000 || ofs >= 0x80000) {
988        /* Combine the low 20 bits of the offset with the actual load insn;
989           the high 44 bits must come from an immediate load.  */
990        tcg_target_long low = ((ofs & 0xfffff) ^ 0x80000) - 0x80000;
991        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs - low);
992        ofs = low;
993
994        /* If we were already given an index register, add it in.  */
995        if (index != TCG_REG_NONE) {
996            tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
997        }
998        index = TCG_TMP0;
999    }
1000
1001    if (opc_rx && ofs >= 0 && ofs < 0x1000) {
1002        tcg_out_insn_RX(s, opc_rx, data, base, index, ofs);
1003    } else {
1004        tcg_out_insn_RXY(s, opc_rxy, data, base, index, ofs);
1005    }
1006}
1007
1008static void tcg_out_vrx_mem(TCGContext *s, S390Opcode opc_vrx,
1009                            TCGReg data, TCGReg base, TCGReg index,
1010                            tcg_target_long ofs, int m3)
1011{
1012    if (ofs < 0 || ofs >= 0x1000) {
1013        if (ofs >= -0x80000 && ofs < 0x80000) {
1014            tcg_out_insn(s, RXY, LAY, TCG_TMP0, base, index, ofs);
1015            base = TCG_TMP0;
1016            index = TCG_REG_NONE;
1017            ofs = 0;
1018        } else {
1019            tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs);
1020            if (index != TCG_REG_NONE) {
1021                tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
1022            }
1023            index = TCG_TMP0;
1024            ofs = 0;
1025        }
1026    }
1027    tcg_out_insn_VRX(s, opc_vrx, data, base, index, ofs, m3);
1028}
1029
1030/* load data without address translation or endianness conversion */
1031static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg data,
1032                       TCGReg base, intptr_t ofs)
1033{
1034    switch (type) {
1035    case TCG_TYPE_I32:
1036        if (likely(is_general_reg(data))) {
1037            tcg_out_mem(s, RX_L, RXY_LY, data, base, TCG_REG_NONE, ofs);
1038            break;
1039        }
1040        tcg_out_vrx_mem(s, VRX_VLLEZ, data, base, TCG_REG_NONE, ofs, MO_32);
1041        break;
1042
1043    case TCG_TYPE_I64:
1044        if (likely(is_general_reg(data))) {
1045            tcg_out_mem(s, 0, RXY_LG, data, base, TCG_REG_NONE, ofs);
1046            break;
1047        }
1048        /* fallthru */
1049
1050    case TCG_TYPE_V64:
1051        tcg_out_vrx_mem(s, VRX_VLLEZ, data, base, TCG_REG_NONE, ofs, MO_64);
1052        break;
1053
1054    case TCG_TYPE_V128:
1055        /* Hint quadword aligned.  */
1056        tcg_out_vrx_mem(s, VRX_VL, data, base, TCG_REG_NONE, ofs, 4);
1057        break;
1058
1059    default:
1060        g_assert_not_reached();
1061    }
1062}
1063
1064static void tcg_out_st(TCGContext *s, TCGType type, TCGReg data,
1065                       TCGReg base, intptr_t ofs)
1066{
1067    switch (type) {
1068    case TCG_TYPE_I32:
1069        if (likely(is_general_reg(data))) {
1070            tcg_out_mem(s, RX_ST, RXY_STY, data, base, TCG_REG_NONE, ofs);
1071        } else {
1072            tcg_out_vrx_mem(s, VRX_VSTEF, data, base, TCG_REG_NONE, ofs, 1);
1073        }
1074        break;
1075
1076    case TCG_TYPE_I64:
1077        if (likely(is_general_reg(data))) {
1078            tcg_out_mem(s, 0, RXY_STG, data, base, TCG_REG_NONE, ofs);
1079            break;
1080        }
1081        /* fallthru */
1082
1083    case TCG_TYPE_V64:
1084        tcg_out_vrx_mem(s, VRX_VSTEG, data, base, TCG_REG_NONE, ofs, 0);
1085        break;
1086
1087    case TCG_TYPE_V128:
1088        /* Hint quadword aligned.  */
1089        tcg_out_vrx_mem(s, VRX_VST, data, base, TCG_REG_NONE, ofs, 4);
1090        break;
1091
1092    default:
1093        g_assert_not_reached();
1094    }
1095}
1096
1097static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
1098                               TCGReg base, intptr_t ofs)
1099{
1100    return false;
1101}
1102
1103static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
1104{
1105    return false;
1106}
1107
1108static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
1109                             tcg_target_long imm)
1110{
1111    /* This function is only used for passing structs by reference. */
1112    tcg_out_mem(s, RX_LA, RXY_LAY, rd, rs, TCG_REG_NONE, imm);
1113}
1114
1115static inline void tcg_out_risbg(TCGContext *s, TCGReg dest, TCGReg src,
1116                                 int msb, int lsb, int ofs, int z)
1117{
1118    /* Format RIE-f */
1119    tcg_out16(s, (RIEf_RISBG & 0xff00) | (dest << 4) | src);
1120    tcg_out16(s, (msb << 8) | (z << 7) | lsb);
1121    tcg_out16(s, (ofs << 8) | (RIEf_RISBG & 0xff));
1122}
1123
1124static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1125{
1126    tcg_out_insn(s, RRE, LGBR, dest, src);
1127}
1128
1129static void tcg_out_ext8u(TCGContext *s, TCGReg dest, TCGReg src)
1130{
1131    tcg_out_insn(s, RRE, LLGCR, dest, src);
1132}
1133
1134static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1135{
1136    tcg_out_insn(s, RRE, LGHR, dest, src);
1137}
1138
1139static void tcg_out_ext16u(TCGContext *s, TCGReg dest, TCGReg src)
1140{
1141    tcg_out_insn(s, RRE, LLGHR, dest, src);
1142}
1143
1144static void tcg_out_ext32s(TCGContext *s, TCGReg dest, TCGReg src)
1145{
1146    tcg_out_insn(s, RRE, LGFR, dest, src);
1147}
1148
1149static void tcg_out_ext32u(TCGContext *s, TCGReg dest, TCGReg src)
1150{
1151    tcg_out_insn(s, RRE, LLGFR, dest, src);
1152}
1153
1154static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg dest, TCGReg src)
1155{
1156    tcg_out_ext32s(s, dest, src);
1157}
1158
1159static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg dest, TCGReg src)
1160{
1161    tcg_out_ext32u(s, dest, src);
1162}
1163
1164static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg dest, TCGReg src)
1165{
1166    tcg_out_mov(s, TCG_TYPE_I32, dest, src);
1167}
1168
1169static void tgen_andi_risbg(TCGContext *s, TCGReg out, TCGReg in, uint64_t val)
1170{
1171    int msb, lsb;
1172    if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
1173        /* Achieve wraparound by swapping msb and lsb.  */
1174        msb = 64 - ctz64(~val);
1175        lsb = clz64(~val) - 1;
1176    } else {
1177        msb = clz64(val);
1178        lsb = 63 - ctz64(val);
1179    }
1180    tcg_out_risbg(s, out, in, msb, lsb, 0, 1);
1181}
1182
1183static void tgen_andi(TCGContext *s, TCGType type, TCGReg dest, uint64_t val)
1184{
1185    static const S390Opcode ni_insns[4] = {
1186        RI_NILL, RI_NILH, RI_NIHL, RI_NIHH
1187    };
1188    static const S390Opcode nif_insns[2] = {
1189        RIL_NILF, RIL_NIHF
1190    };
1191    uint64_t valid = (type == TCG_TYPE_I32 ? 0xffffffffull : -1ull);
1192    int i;
1193
1194    /* Look for the zero-extensions.  */
1195    if ((val & valid) == 0xffffffff) {
1196        tcg_out_ext32u(s, dest, dest);
1197        return;
1198    }
1199    if ((val & valid) == 0xff) {
1200        tcg_out_ext8u(s, dest, dest);
1201        return;
1202    }
1203    if ((val & valid) == 0xffff) {
1204        tcg_out_ext16u(s, dest, dest);
1205        return;
1206    }
1207
1208    i = is_const_p16(~val & valid);
1209    if (i >= 0) {
1210        tcg_out_insn_RI(s, ni_insns[i], dest, val >> (i * 16));
1211        return;
1212    }
1213
1214    i = is_const_p32(~val & valid);
1215    tcg_debug_assert(i == 0 || type != TCG_TYPE_I32);
1216    if (i >= 0) {
1217        tcg_out_insn_RIL(s, nif_insns[i], dest, val >> (i * 32));
1218        return;
1219    }
1220
1221    if (risbg_mask(val)) {
1222        tgen_andi_risbg(s, dest, dest, val);
1223        return;
1224    }
1225
1226    g_assert_not_reached();
1227}
1228
1229static void tgen_ori(TCGContext *s, TCGReg dest, uint64_t val)
1230{
1231    static const S390Opcode oif_insns[2] = {
1232        RIL_OILF, RIL_OIHF
1233    };
1234
1235    int i;
1236
1237    i = is_const_p16(val);
1238    if (i >= 0) {
1239        tcg_out_insn_RI(s, oi_insns[i], dest, val >> (i * 16));
1240        return;
1241    }
1242
1243    i = is_const_p32(val);
1244    if (i >= 0) {
1245        tcg_out_insn_RIL(s, oif_insns[i], dest, val >> (i * 32));
1246        return;
1247    }
1248
1249    g_assert_not_reached();
1250}
1251
1252static void tgen_xori(TCGContext *s, TCGReg dest, uint64_t val)
1253{
1254    switch (is_const_p32(val)) {
1255    case 0:
1256        tcg_out_insn(s, RIL, XILF, dest, val);
1257        break;
1258    case 1:
1259        tcg_out_insn(s, RIL, XIHF, dest, val >> 32);
1260        break;
1261    default:
1262        g_assert_not_reached();
1263    }
1264}
1265
1266static int tgen_cmp2(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1267                     TCGArg c2, bool c2const, bool need_carry, int *inv_cc)
1268{
1269    bool is_unsigned = is_unsigned_cond(c);
1270    TCGCond inv_c = tcg_invert_cond(c);
1271    S390Opcode op;
1272
1273    if (is_tst_cond(c)) {
1274        tcg_debug_assert(!need_carry);
1275
1276        if (!c2const) {
1277            if (type == TCG_TYPE_I32) {
1278                tcg_out_insn(s, RRFa, NRK, TCG_REG_R0, r1, c2);
1279            } else {
1280                tcg_out_insn(s, RRFa, NGRK, TCG_REG_R0, r1, c2);
1281            }
1282            goto exit;
1283        }
1284
1285        if (type == TCG_TYPE_I32) {
1286            c2 = (uint32_t)c2;
1287        }
1288
1289        int i = is_const_p16(c2);
1290        if (i >= 0) {
1291            tcg_out_insn_RI(s, tm_insns[i], r1, c2 >> (i * 16));
1292            *inv_cc = c == TCG_COND_TSTEQ ? S390_TM_NE : S390_TM_EQ;
1293            return *inv_cc ^ 15;
1294        }
1295
1296        if (risbg_mask(c2)) {
1297            tgen_andi_risbg(s, TCG_REG_R0, r1, c2);
1298            goto exit;
1299        }
1300        g_assert_not_reached();
1301    }
1302
1303    if (c2const) {
1304        if (c2 == 0) {
1305            if (!(is_unsigned && need_carry)) {
1306                if (type == TCG_TYPE_I32) {
1307                    tcg_out_insn(s, RR, LTR, r1, r1);
1308                } else {
1309                    tcg_out_insn(s, RRE, LTGR, r1, r1);
1310                }
1311                *inv_cc = tcg_cond_to_ltr_cond[inv_c];
1312                return tcg_cond_to_ltr_cond[c];
1313            }
1314        }
1315
1316        if (!is_unsigned && c2 == (int16_t)c2) {
1317            op = (type == TCG_TYPE_I32 ? RI_CHI : RI_CGHI);
1318            tcg_out_insn_RI(s, op, r1, c2);
1319            goto exit;
1320        }
1321
1322        if (type == TCG_TYPE_I32) {
1323            op = (is_unsigned ? RIL_CLFI : RIL_CFI);
1324            tcg_out_insn_RIL(s, op, r1, c2);
1325            goto exit;
1326        }
1327
1328        /* Should match TCG_CT_CONST_CMP. */
1329        switch (c) {
1330        case TCG_COND_LT:
1331        case TCG_COND_GE:
1332        case TCG_COND_LE:
1333        case TCG_COND_GT:
1334            tcg_debug_assert(c2 == (int32_t)c2);
1335            op = RIL_CGFI;
1336            break;
1337        case TCG_COND_EQ:
1338        case TCG_COND_NE:
1339            if (c2 == (int32_t)c2) {
1340                op = RIL_CGFI;
1341                break;
1342            }
1343            /* fall through */
1344        case TCG_COND_LTU:
1345        case TCG_COND_GEU:
1346        case TCG_COND_LEU:
1347        case TCG_COND_GTU:
1348            tcg_debug_assert(c2 == (uint32_t)c2);
1349            op = RIL_CLGFI;
1350            break;
1351        default:
1352            g_assert_not_reached();
1353        }
1354        tcg_out_insn_RIL(s, op, r1, c2);
1355    } else if (type == TCG_TYPE_I32) {
1356        op = (is_unsigned ? RR_CLR : RR_CR);
1357        tcg_out_insn_RR(s, op, r1, c2);
1358    } else {
1359        op = (is_unsigned ? RRE_CLGR : RRE_CGR);
1360        tcg_out_insn_RRE(s, op, r1, c2);
1361    }
1362
1363 exit:
1364    *inv_cc = tcg_cond_to_s390_cond[inv_c];
1365    return tcg_cond_to_s390_cond[c];
1366}
1367
1368static int tgen_cmp(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1369                    TCGArg c2, bool c2const, bool need_carry)
1370{
1371    int inv_cc;
1372    return tgen_cmp2(s, type, c, r1, c2, c2const, need_carry, &inv_cc);
1373}
1374
1375static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond,
1376                         TCGReg dest, TCGReg c1, TCGArg c2,
1377                         bool c2const, bool neg)
1378{
1379    int cc;
1380
1381    /* With LOC2, we can always emit the minimum 3 insns.  */
1382    if (HAVE_FACILITY(LOAD_ON_COND2)) {
1383        /* Emit: d = 0, d = (cc ? 1 : d).  */
1384        cc = tgen_cmp(s, type, cond, c1, c2, c2const, false);
1385        tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1386        tcg_out_insn(s, RIEg, LOCGHI, dest, neg ? -1 : 1, cc);
1387        return;
1388    }
1389
1390    switch (cond) {
1391    case TCG_COND_GEU:
1392    case TCG_COND_LTU:
1393    case TCG_COND_LT:
1394    case TCG_COND_GE:
1395        /* Swap operands so that we can use LEU/GTU/GT/LE.  */
1396        if (!c2const) {
1397            TCGReg t = c1;
1398            c1 = c2;
1399            c2 = t;
1400            cond = tcg_swap_cond(cond);
1401        }
1402        break;
1403    default:
1404        break;
1405    }
1406
1407    switch (cond) {
1408    case TCG_COND_NE:
1409        /* X != 0 is X > 0.  */
1410        if (c2const && c2 == 0) {
1411            cond = TCG_COND_GTU;
1412        } else {
1413            break;
1414        }
1415        /* fallthru */
1416
1417    case TCG_COND_GTU:
1418    case TCG_COND_GT:
1419        /*
1420         * The result of a compare has CC=2 for GT and CC=3 unused.
1421         * ADD LOGICAL WITH CARRY considers (CC & 2) the carry bit.
1422         */
1423        tgen_cmp(s, type, cond, c1, c2, c2const, true);
1424        tcg_out_movi(s, type, dest, 0);
1425        tcg_out_insn(s, RRE, ALCGR, dest, dest);
1426        if (neg) {
1427            if (type == TCG_TYPE_I32) {
1428                tcg_out_insn(s, RR, LCR, dest, dest);
1429            } else {
1430                tcg_out_insn(s, RRE, LCGR, dest, dest);
1431            }
1432        }
1433        return;
1434
1435    case TCG_COND_EQ:
1436        /* X == 0 is X <= 0.  */
1437        if (c2const && c2 == 0) {
1438            cond = TCG_COND_LEU;
1439        } else {
1440            break;
1441        }
1442        /* fallthru */
1443
1444    case TCG_COND_LEU:
1445    case TCG_COND_LE:
1446        /*
1447         * As above, but we're looking for borrow, or !carry.
1448         * The second insn computes d - d - borrow, or -1 for true
1449         * and 0 for false.  So we must mask to 1 bit afterward.
1450         */
1451        tgen_cmp(s, type, cond, c1, c2, c2const, true);
1452        tcg_out_insn(s, RRE, SLBGR, dest, dest);
1453        if (!neg) {
1454            tgen_andi(s, type, dest, 1);
1455        }
1456        return;
1457
1458    default:
1459        g_assert_not_reached();
1460    }
1461
1462    cc = tgen_cmp(s, type, cond, c1, c2, c2const, false);
1463    /* Emit: d = 0, t = 1, d = (cc ? t : d).  */
1464    tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1465    tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, neg ? -1 : 1);
1466    tcg_out_insn(s, RRFc, LOCGR, dest, TCG_TMP0, cc);
1467}
1468
1469static void tgen_movcond_int(TCGContext *s, TCGType type, TCGReg dest,
1470                             TCGArg v3, int v3const, TCGReg v4,
1471                             int cc, int inv_cc)
1472{
1473    TCGReg src;
1474
1475    if (v3const) {
1476        if (dest == v4) {
1477            if (HAVE_FACILITY(LOAD_ON_COND2)) {
1478                /* Emit: if (cc) dest = v3. */
1479                tcg_out_insn(s, RIEg, LOCGHI, dest, v3, cc);
1480                return;
1481            }
1482            tcg_out_insn(s, RI, LGHI, TCG_TMP0, v3);
1483            src = TCG_TMP0;
1484        } else {
1485            /* LGR+LOCGHI is larger than LGHI+LOCGR. */
1486            tcg_out_insn(s, RI, LGHI, dest, v3);
1487            cc = inv_cc;
1488            src = v4;
1489        }
1490    } else {
1491        if (HAVE_FACILITY(MISC_INSN_EXT3)) {
1492            /* Emit: dest = cc ? v3 : v4. */
1493            tcg_out_insn(s, RRFam, SELGR, dest, v3, v4, cc);
1494            return;
1495        }
1496        if (dest == v4) {
1497            src = v3;
1498        } else {
1499            tcg_out_mov(s, type, dest, v3);
1500            cc = inv_cc;
1501            src = v4;
1502        }
1503    }
1504
1505    /* Emit: if (cc) dest = src. */
1506    tcg_out_insn(s, RRFc, LOCGR, dest, src, cc);
1507}
1508
1509static void tgen_movcond(TCGContext *s, TCGType type, TCGCond c, TCGReg dest,
1510                         TCGReg c1, TCGArg c2, int c2const,
1511                         TCGArg v3, int v3const, TCGReg v4)
1512{
1513    int cc, inv_cc;
1514
1515    cc = tgen_cmp2(s, type, c, c1, c2, c2const, false, &inv_cc);
1516    tgen_movcond_int(s, type, dest, v3, v3const, v4, cc, inv_cc);
1517}
1518
1519static void tgen_clz(TCGContext *s, TCGReg dest, TCGReg a1,
1520                     TCGArg a2, int a2const)
1521{
1522    /* Since this sets both R and R+1, we have no choice but to store the
1523       result into R0, allowing R1 == TCG_TMP0 to be clobbered as well.  */
1524    QEMU_BUILD_BUG_ON(TCG_TMP0 != TCG_REG_R1);
1525    tcg_out_insn(s, RRE, FLOGR, TCG_REG_R0, a1);
1526
1527    if (a2const && a2 == 64) {
1528        tcg_out_mov(s, TCG_TYPE_I64, dest, TCG_REG_R0);
1529        return;
1530    }
1531
1532    /*
1533     * Conditions from FLOGR are:
1534     *   2 -> one bit found
1535     *   8 -> no one bit found
1536     */
1537    tgen_movcond_int(s, TCG_TYPE_I64, dest, a2, a2const, TCG_REG_R0, 8, 2);
1538}
1539
1540static void tgen_ctpop(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1541{
1542    /* With MIE3, and bit 0 of m4 set, we get the complete result. */
1543    if (HAVE_FACILITY(MISC_INSN_EXT3)) {
1544        if (type == TCG_TYPE_I32) {
1545            tcg_out_ext32u(s, dest, src);
1546            src = dest;
1547        }
1548        tcg_out_insn(s, RRFc, POPCNT, dest, src, 8);
1549        return;
1550    }
1551
1552    /* Without MIE3, each byte gets the count of bits for the byte. */
1553    tcg_out_insn(s, RRFc, POPCNT, dest, src, 0);
1554
1555    /* Multiply to sum each byte at the top of the word. */
1556    if (type == TCG_TYPE_I32) {
1557        tcg_out_insn(s, RIL, MSFI, dest, 0x01010101);
1558        tcg_out_sh32(s, RS_SRL, dest, TCG_REG_NONE, 24);
1559    } else {
1560        tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, 0x0101010101010101ull);
1561        tcg_out_insn(s, RRE, MSGR, dest, TCG_TMP0);
1562        tcg_out_sh64(s, RSY_SRLG, dest, dest, TCG_REG_NONE, 56);
1563    }
1564}
1565
1566static void tgen_deposit(TCGContext *s, TCGReg dest, TCGReg src,
1567                         int ofs, int len, int z)
1568{
1569    int lsb = (63 - ofs);
1570    int msb = lsb - (len - 1);
1571    tcg_out_risbg(s, dest, src, msb, lsb, ofs, z);
1572}
1573
1574static void tgen_extract(TCGContext *s, TCGReg dest, TCGReg src,
1575                         int ofs, int len)
1576{
1577    tcg_out_risbg(s, dest, src, 64 - len, 63, 64 - ofs, 1);
1578}
1579
1580static void tgen_gotoi(TCGContext *s, int cc, const tcg_insn_unit *dest)
1581{
1582    ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1;
1583    if (off == (int16_t)off) {
1584        tcg_out_insn(s, RI, BRC, cc, off);
1585    } else if (off == (int32_t)off) {
1586        tcg_out_insn(s, RIL, BRCL, cc, off);
1587    } else {
1588        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1589        tcg_out_insn(s, RR, BCR, cc, TCG_TMP0);
1590    }
1591}
1592
1593static void tgen_branch(TCGContext *s, int cc, TCGLabel *l)
1594{
1595    if (l->has_value) {
1596        tgen_gotoi(s, cc, l->u.value_ptr);
1597    } else {
1598        tcg_out16(s, RI_BRC | (cc << 4));
1599        tcg_out_reloc(s, s->code_ptr, R_390_PC16DBL, l, 2);
1600        s->code_ptr += 1;
1601    }
1602}
1603
1604static void tgen_compare_branch(TCGContext *s, S390Opcode opc, int cc,
1605                                TCGReg r1, TCGReg r2, TCGLabel *l)
1606{
1607    tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1608    /* Format RIE-b */
1609    tcg_out16(s, (opc & 0xff00) | (r1 << 4) | r2);
1610    tcg_out16(s, 0);
1611    tcg_out16(s, cc << 12 | (opc & 0xff));
1612}
1613
1614static void tgen_compare_imm_branch(TCGContext *s, S390Opcode opc, int cc,
1615                                    TCGReg r1, int i2, TCGLabel *l)
1616{
1617    tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1618    /* Format RIE-c */
1619    tcg_out16(s, (opc & 0xff00) | (r1 << 4) | cc);
1620    tcg_out16(s, 0);
1621    tcg_out16(s, (i2 << 8) | (opc & 0xff));
1622}
1623
1624static void tgen_brcond(TCGContext *s, TCGType type, TCGCond c,
1625                        TCGReg r1, TCGArg c2, int c2const, TCGLabel *l)
1626{
1627    int cc;
1628
1629    if (!is_tst_cond(c)) {
1630        bool is_unsigned = is_unsigned_cond(c);
1631        bool in_range;
1632        S390Opcode opc;
1633
1634        cc = tcg_cond_to_s390_cond[c];
1635
1636        if (!c2const) {
1637            opc = (type == TCG_TYPE_I32
1638                   ? (is_unsigned ? RIEb_CLRJ : RIEb_CRJ)
1639                   : (is_unsigned ? RIEb_CLGRJ : RIEb_CGRJ));
1640            tgen_compare_branch(s, opc, cc, r1, c2, l);
1641            return;
1642        }
1643
1644        /*
1645         * COMPARE IMMEDIATE AND BRANCH RELATIVE has an 8-bit immediate field.
1646         * If the immediate we've been given does not fit that range, we'll
1647         * fall back to separate compare and branch instructions using the
1648         * larger comparison range afforded by COMPARE IMMEDIATE.
1649         */
1650        if (type == TCG_TYPE_I32) {
1651            if (is_unsigned) {
1652                opc = RIEc_CLIJ;
1653                in_range = (uint32_t)c2 == (uint8_t)c2;
1654            } else {
1655                opc = RIEc_CIJ;
1656                in_range = (int32_t)c2 == (int8_t)c2;
1657            }
1658        } else {
1659            if (is_unsigned) {
1660                opc = RIEc_CLGIJ;
1661                in_range = (uint64_t)c2 == (uint8_t)c2;
1662            } else {
1663                opc = RIEc_CGIJ;
1664                in_range = (int64_t)c2 == (int8_t)c2;
1665            }
1666        }
1667        if (in_range) {
1668            tgen_compare_imm_branch(s, opc, cc, r1, c2, l);
1669            return;
1670        }
1671    }
1672
1673    cc = tgen_cmp(s, type, c, r1, c2, c2const, false);
1674    tgen_branch(s, cc, l);
1675}
1676
1677static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *dest)
1678{
1679    ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1;
1680    if (off == (int32_t)off) {
1681        tcg_out_insn(s, RIL, BRASL, TCG_REG_R14, off);
1682    } else {
1683        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1684        tcg_out_insn(s, RR, BASR, TCG_REG_R14, TCG_TMP0);
1685    }
1686}
1687
1688static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest,
1689                         const TCGHelperInfo *info)
1690{
1691    tcg_out_call_int(s, dest);
1692}
1693
1694typedef struct {
1695    TCGReg base;
1696    TCGReg index;
1697    int disp;
1698    TCGAtomAlign aa;
1699} HostAddress;
1700
1701bool tcg_target_has_memory_bswap(MemOp memop)
1702{
1703    TCGAtomAlign aa;
1704
1705    if ((memop & MO_SIZE) <= MO_64) {
1706        return true;
1707    }
1708
1709    /*
1710     * Reject 16-byte memop with 16-byte atomicity,
1711     * but do allow a pair of 64-bit operations.
1712     */
1713    aa = atom_and_align_for_opc(tcg_ctx, memop, MO_ATOM_IFALIGN, true);
1714    return aa.atom <= MO_64;
1715}
1716
1717static void tcg_out_qemu_ld_direct(TCGContext *s, MemOp opc, TCGReg data,
1718                                   HostAddress h)
1719{
1720    switch (opc & (MO_SSIZE | MO_BSWAP)) {
1721    case MO_UB:
1722        tcg_out_insn(s, RXY, LLGC, data, h.base, h.index, h.disp);
1723        break;
1724    case MO_SB:
1725        tcg_out_insn(s, RXY, LGB, data, h.base, h.index, h.disp);
1726        break;
1727
1728    case MO_UW | MO_BSWAP:
1729        /* swapped unsigned halfword load with upper bits zeroed */
1730        tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp);
1731        tcg_out_ext16u(s, data, data);
1732        break;
1733    case MO_UW:
1734        tcg_out_insn(s, RXY, LLGH, data, h.base, h.index, h.disp);
1735        break;
1736
1737    case MO_SW | MO_BSWAP:
1738        /* swapped sign-extended halfword load */
1739        tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp);
1740        tcg_out_ext16s(s, TCG_TYPE_REG, data, data);
1741        break;
1742    case MO_SW:
1743        tcg_out_insn(s, RXY, LGH, data, h.base, h.index, h.disp);
1744        break;
1745
1746    case MO_UL | MO_BSWAP:
1747        /* swapped unsigned int load with upper bits zeroed */
1748        tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp);
1749        tcg_out_ext32u(s, data, data);
1750        break;
1751    case MO_UL:
1752        tcg_out_insn(s, RXY, LLGF, data, h.base, h.index, h.disp);
1753        break;
1754
1755    case MO_SL | MO_BSWAP:
1756        /* swapped sign-extended int load */
1757        tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp);
1758        tcg_out_ext32s(s, data, data);
1759        break;
1760    case MO_SL:
1761        tcg_out_insn(s, RXY, LGF, data, h.base, h.index, h.disp);
1762        break;
1763
1764    case MO_UQ | MO_BSWAP:
1765        tcg_out_insn(s, RXY, LRVG, data, h.base, h.index, h.disp);
1766        break;
1767    case MO_UQ:
1768        tcg_out_insn(s, RXY, LG, data, h.base, h.index, h.disp);
1769        break;
1770
1771    default:
1772        g_assert_not_reached();
1773    }
1774}
1775
1776static void tcg_out_qemu_st_direct(TCGContext *s, MemOp opc, TCGReg data,
1777                                   HostAddress h)
1778{
1779    switch (opc & (MO_SIZE | MO_BSWAP)) {
1780    case MO_UB:
1781        if (h.disp >= 0 && h.disp < 0x1000) {
1782            tcg_out_insn(s, RX, STC, data, h.base, h.index, h.disp);
1783        } else {
1784            tcg_out_insn(s, RXY, STCY, data, h.base, h.index, h.disp);
1785        }
1786        break;
1787
1788    case MO_UW | MO_BSWAP:
1789        tcg_out_insn(s, RXY, STRVH, data, h.base, h.index, h.disp);
1790        break;
1791    case MO_UW:
1792        if (h.disp >= 0 && h.disp < 0x1000) {
1793            tcg_out_insn(s, RX, STH, data, h.base, h.index, h.disp);
1794        } else {
1795            tcg_out_insn(s, RXY, STHY, data, h.base, h.index, h.disp);
1796        }
1797        break;
1798
1799    case MO_UL | MO_BSWAP:
1800        tcg_out_insn(s, RXY, STRV, data, h.base, h.index, h.disp);
1801        break;
1802    case MO_UL:
1803        if (h.disp >= 0 && h.disp < 0x1000) {
1804            tcg_out_insn(s, RX, ST, data, h.base, h.index, h.disp);
1805        } else {
1806            tcg_out_insn(s, RXY, STY, data, h.base, h.index, h.disp);
1807        }
1808        break;
1809
1810    case MO_UQ | MO_BSWAP:
1811        tcg_out_insn(s, RXY, STRVG, data, h.base, h.index, h.disp);
1812        break;
1813    case MO_UQ:
1814        tcg_out_insn(s, RXY, STG, data, h.base, h.index, h.disp);
1815        break;
1816
1817    default:
1818        g_assert_not_reached();
1819    }
1820}
1821
1822static const TCGLdstHelperParam ldst_helper_param = {
1823    .ntmp = 1, .tmp = { TCG_TMP0 }
1824};
1825
1826static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1827{
1828    MemOp opc = get_memop(lb->oi);
1829
1830    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1831                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) {
1832        return false;
1833    }
1834
1835    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
1836    tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SIZE]);
1837    tcg_out_ld_helper_ret(s, lb, false, &ldst_helper_param);
1838
1839    tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1840    return true;
1841}
1842
1843static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1844{
1845    MemOp opc = get_memop(lb->oi);
1846
1847    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1848                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) {
1849        return false;
1850    }
1851
1852    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
1853    tcg_out_call_int(s, qemu_st_helpers[opc & MO_SIZE]);
1854
1855    tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1856    return true;
1857}
1858
1859/* We're expecting to use a 20-bit negative offset on the tlb memory ops.  */
1860#define MIN_TLB_MASK_TABLE_OFS  -(1 << 19)
1861
1862/*
1863 * For system-mode, perform the TLB load and compare.
1864 * For user-mode, perform any required alignment tests.
1865 * In both cases, return a TCGLabelQemuLdst structure if the slow path
1866 * is required and fill in @h with the host address for the fast path.
1867 */
1868static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
1869                                           TCGReg addr_reg, MemOpIdx oi,
1870                                           bool is_ld)
1871{
1872    TCGType addr_type = s->addr_type;
1873    TCGLabelQemuLdst *ldst = NULL;
1874    MemOp opc = get_memop(oi);
1875    MemOp s_bits = opc & MO_SIZE;
1876    unsigned a_mask;
1877
1878    h->aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, s_bits == MO_128);
1879    a_mask = (1 << h->aa.align) - 1;
1880
1881    if (tcg_use_softmmu) {
1882        unsigned s_mask = (1 << s_bits) - 1;
1883        int mem_index = get_mmuidx(oi);
1884        int fast_off = tlb_mask_table_ofs(s, mem_index);
1885        int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1886        int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1887        int ofs, a_off;
1888        uint64_t tlb_mask;
1889
1890        ldst = new_ldst_label(s);
1891        ldst->is_ld = is_ld;
1892        ldst->oi = oi;
1893        ldst->addrlo_reg = addr_reg;
1894
1895        tcg_out_sh64(s, RSY_SRLG, TCG_TMP0, addr_reg, TCG_REG_NONE,
1896                     s->page_bits - CPU_TLB_ENTRY_BITS);
1897
1898        tcg_out_insn(s, RXY, NG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, mask_off);
1899        tcg_out_insn(s, RXY, AG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, table_off);
1900
1901        /*
1902         * For aligned accesses, we check the first byte and include the
1903         * alignment bits within the address.  For unaligned access, we
1904         * check that we don't cross pages using the address of the last
1905         * byte of the access.
1906         */
1907        a_off = (a_mask >= s_mask ? 0 : s_mask - a_mask);
1908        tlb_mask = (uint64_t)s->page_mask | a_mask;
1909        if (a_off == 0) {
1910            tgen_andi_risbg(s, TCG_REG_R0, addr_reg, tlb_mask);
1911        } else {
1912            tcg_out_insn(s, RX, LA, TCG_REG_R0, addr_reg, TCG_REG_NONE, a_off);
1913            tgen_andi(s, addr_type, TCG_REG_R0, tlb_mask);
1914        }
1915
1916        if (is_ld) {
1917            ofs = offsetof(CPUTLBEntry, addr_read);
1918        } else {
1919            ofs = offsetof(CPUTLBEntry, addr_write);
1920        }
1921        if (addr_type == TCG_TYPE_I32) {
1922            ofs += HOST_BIG_ENDIAN * 4;
1923            tcg_out_insn(s, RX, C, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs);
1924        } else {
1925            tcg_out_insn(s, RXY, CG, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs);
1926        }
1927
1928        tcg_out16(s, RI_BRC | (S390_CC_NE << 4));
1929        ldst->label_ptr[0] = s->code_ptr++;
1930
1931        h->index = TCG_TMP0;
1932        tcg_out_insn(s, RXY, LG, h->index, TCG_TMP0, TCG_REG_NONE,
1933                     offsetof(CPUTLBEntry, addend));
1934
1935        if (addr_type == TCG_TYPE_I32) {
1936            tcg_out_insn(s, RRE, ALGFR, h->index, addr_reg);
1937            h->base = TCG_REG_NONE;
1938        } else {
1939            h->base = addr_reg;
1940        }
1941        h->disp = 0;
1942    } else {
1943        if (a_mask) {
1944            ldst = new_ldst_label(s);
1945            ldst->is_ld = is_ld;
1946            ldst->oi = oi;
1947            ldst->addrlo_reg = addr_reg;
1948
1949            tcg_debug_assert(a_mask <= 0xffff);
1950            tcg_out_insn(s, RI, TMLL, addr_reg, a_mask);
1951
1952            tcg_out16(s, RI_BRC | (S390_TM_NE << 4));
1953            ldst->label_ptr[0] = s->code_ptr++;
1954        }
1955
1956        h->base = addr_reg;
1957        if (addr_type == TCG_TYPE_I32) {
1958            tcg_out_ext32u(s, TCG_TMP0, addr_reg);
1959            h->base = TCG_TMP0;
1960        }
1961        if (guest_base < 0x80000) {
1962            h->index = TCG_REG_NONE;
1963            h->disp = guest_base;
1964        } else {
1965            h->index = TCG_GUEST_BASE_REG;
1966            h->disp = 0;
1967        }
1968    }
1969
1970    return ldst;
1971}
1972
1973static void tcg_out_qemu_ld(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
1974                            MemOpIdx oi, TCGType data_type)
1975{
1976    TCGLabelQemuLdst *ldst;
1977    HostAddress h;
1978
1979    ldst = prepare_host_addr(s, &h, addr_reg, oi, true);
1980    tcg_out_qemu_ld_direct(s, get_memop(oi), data_reg, h);
1981
1982    if (ldst) {
1983        ldst->type = data_type;
1984        ldst->datalo_reg = data_reg;
1985        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1986    }
1987}
1988
1989static void tcg_out_qemu_st(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
1990                            MemOpIdx oi, TCGType data_type)
1991{
1992    TCGLabelQemuLdst *ldst;
1993    HostAddress h;
1994
1995    ldst = prepare_host_addr(s, &h, addr_reg, oi, false);
1996    tcg_out_qemu_st_direct(s, get_memop(oi), data_reg, h);
1997
1998    if (ldst) {
1999        ldst->type = data_type;
2000        ldst->datalo_reg = data_reg;
2001        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2002    }
2003}
2004
2005static void tcg_out_qemu_ldst_i128(TCGContext *s, TCGReg datalo, TCGReg datahi,
2006                                   TCGReg addr_reg, MemOpIdx oi, bool is_ld)
2007{
2008    TCGLabel *l1 = NULL, *l2 = NULL;
2009    TCGLabelQemuLdst *ldst;
2010    HostAddress h;
2011    bool need_bswap;
2012    bool use_pair;
2013    S390Opcode insn;
2014
2015    ldst = prepare_host_addr(s, &h, addr_reg, oi, is_ld);
2016
2017    use_pair = h.aa.atom < MO_128;
2018    need_bswap = get_memop(oi) & MO_BSWAP;
2019
2020    if (!use_pair) {
2021        /*
2022         * Atomicity requires we use LPQ.  If we've already checked for
2023         * 16-byte alignment, that's all we need.  If we arrive with
2024         * lesser alignment, we have determined that less than 16-byte
2025         * alignment can be satisfied with two 8-byte loads.
2026         */
2027        if (h.aa.align < MO_128) {
2028            use_pair = true;
2029            l1 = gen_new_label();
2030            l2 = gen_new_label();
2031
2032            tcg_out_insn(s, RI, TMLL, addr_reg, 15);
2033            tgen_branch(s, S390_TM_NE, l1);
2034        }
2035
2036        tcg_debug_assert(!need_bswap);
2037        tcg_debug_assert(datalo & 1);
2038        tcg_debug_assert(datahi == datalo - 1);
2039        insn = is_ld ? RXY_LPQ : RXY_STPQ;
2040        tcg_out_insn_RXY(s, insn, datahi, h.base, h.index, h.disp);
2041
2042        if (use_pair) {
2043            tgen_branch(s, S390_CC_ALWAYS, l2);
2044            tcg_out_label(s, l1);
2045        }
2046    }
2047    if (use_pair) {
2048        TCGReg d1, d2;
2049
2050        if (need_bswap) {
2051            d1 = datalo, d2 = datahi;
2052            insn = is_ld ? RXY_LRVG : RXY_STRVG;
2053        } else {
2054            d1 = datahi, d2 = datalo;
2055            insn = is_ld ? RXY_LG : RXY_STG;
2056        }
2057
2058        if (h.base == d1 || h.index == d1) {
2059            tcg_out_insn(s, RXY, LAY, TCG_TMP0, h.base, h.index, h.disp);
2060            h.base = TCG_TMP0;
2061            h.index = TCG_REG_NONE;
2062            h.disp = 0;
2063        }
2064        tcg_out_insn_RXY(s, insn, d1, h.base, h.index, h.disp);
2065        tcg_out_insn_RXY(s, insn, d2, h.base, h.index, h.disp + 8);
2066    }
2067    if (l2) {
2068        tcg_out_label(s, l2);
2069    }
2070
2071    if (ldst) {
2072        ldst->type = TCG_TYPE_I128;
2073        ldst->datalo_reg = datalo;
2074        ldst->datahi_reg = datahi;
2075        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2076    }
2077}
2078
2079static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
2080{
2081    /* Reuse the zeroing that exists for goto_ptr.  */
2082    if (a0 == 0) {
2083        tgen_gotoi(s, S390_CC_ALWAYS, tcg_code_gen_epilogue);
2084    } else {
2085        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, a0);
2086        tgen_gotoi(s, S390_CC_ALWAYS, tb_ret_addr);
2087    }
2088}
2089
2090static void tcg_out_goto_tb(TCGContext *s, int which)
2091{
2092    /*
2093     * Branch displacement must be aligned for atomic patching;
2094     * see if we need to add extra nop before branch
2095     */
2096    if (!QEMU_PTR_IS_ALIGNED(s->code_ptr + 1, 4)) {
2097        tcg_out16(s, NOP);
2098    }
2099    tcg_out16(s, RIL_BRCL | (S390_CC_ALWAYS << 4));
2100    set_jmp_insn_offset(s, which);
2101    s->code_ptr += 2;
2102    set_jmp_reset_offset(s, which);
2103}
2104
2105void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
2106                              uintptr_t jmp_rx, uintptr_t jmp_rw)
2107{
2108    if (!HAVE_FACILITY(GEN_INST_EXT)) {
2109        return;
2110    }
2111    /* patch the branch destination */
2112    uintptr_t addr = tb->jmp_target_addr[n];
2113    intptr_t disp = addr - (jmp_rx - 2);
2114    qatomic_set((int32_t *)jmp_rw, disp / 2);
2115    /* no need to flush icache explicitly */
2116}
2117
2118# define OP_32_64(x) \
2119        case glue(glue(INDEX_op_,x),_i32): \
2120        case glue(glue(INDEX_op_,x),_i64)
2121
2122static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
2123                              const TCGArg args[TCG_MAX_OP_ARGS],
2124                              const int const_args[TCG_MAX_OP_ARGS])
2125{
2126    S390Opcode op, op2;
2127    TCGArg a0, a1, a2;
2128
2129    switch (opc) {
2130    case INDEX_op_goto_ptr:
2131        a0 = args[0];
2132        tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, a0);
2133        break;
2134
2135    OP_32_64(ld8u):
2136        /* ??? LLC (RXY format) is only present with the extended-immediate
2137           facility, whereas LLGC is always present.  */
2138        tcg_out_mem(s, 0, RXY_LLGC, args[0], args[1], TCG_REG_NONE, args[2]);
2139        break;
2140
2141    OP_32_64(ld8s):
2142        /* ??? LB is no smaller than LGB, so no point to using it.  */
2143        tcg_out_mem(s, 0, RXY_LGB, args[0], args[1], TCG_REG_NONE, args[2]);
2144        break;
2145
2146    OP_32_64(ld16u):
2147        /* ??? LLH (RXY format) is only present with the extended-immediate
2148           facility, whereas LLGH is always present.  */
2149        tcg_out_mem(s, 0, RXY_LLGH, args[0], args[1], TCG_REG_NONE, args[2]);
2150        break;
2151
2152    case INDEX_op_ld16s_i32:
2153        tcg_out_mem(s, RX_LH, RXY_LHY, args[0], args[1], TCG_REG_NONE, args[2]);
2154        break;
2155
2156    case INDEX_op_ld_i32:
2157        tcg_out_ld(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2158        break;
2159
2160    OP_32_64(st8):
2161        tcg_out_mem(s, RX_STC, RXY_STCY, args[0], args[1],
2162                    TCG_REG_NONE, args[2]);
2163        break;
2164
2165    OP_32_64(st16):
2166        tcg_out_mem(s, RX_STH, RXY_STHY, args[0], args[1],
2167                    TCG_REG_NONE, args[2]);
2168        break;
2169
2170    case INDEX_op_st_i32:
2171        tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2172        break;
2173
2174    case INDEX_op_add_i32:
2175        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2176        if (const_args[2]) {
2177        do_addi_32:
2178            if (a0 == a1) {
2179                if (a2 == (int16_t)a2) {
2180                    tcg_out_insn(s, RI, AHI, a0, a2);
2181                    break;
2182                }
2183                tcg_out_insn(s, RIL, AFI, a0, a2);
2184                break;
2185            }
2186            tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
2187        } else if (a0 == a1) {
2188            tcg_out_insn(s, RR, AR, a0, a2);
2189        } else {
2190            tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
2191        }
2192        break;
2193    case INDEX_op_sub_i32:
2194        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2195        if (const_args[2]) {
2196            a2 = -a2;
2197            goto do_addi_32;
2198        } else if (a0 == a1) {
2199            tcg_out_insn(s, RR, SR, a0, a2);
2200        } else {
2201            tcg_out_insn(s, RRFa, SRK, a0, a1, a2);
2202        }
2203        break;
2204
2205    case INDEX_op_and_i32:
2206        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2207        if (const_args[2]) {
2208            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2209            tgen_andi(s, TCG_TYPE_I32, a0, a2);
2210        } else if (a0 == a1) {
2211            tcg_out_insn(s, RR, NR, a0, a2);
2212        } else {
2213            tcg_out_insn(s, RRFa, NRK, a0, a1, a2);
2214        }
2215        break;
2216    case INDEX_op_or_i32:
2217        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2218        if (const_args[2]) {
2219            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2220            tgen_ori(s, a0, a2);
2221        } else if (a0 == a1) {
2222            tcg_out_insn(s, RR, OR, a0, a2);
2223        } else {
2224            tcg_out_insn(s, RRFa, ORK, a0, a1, a2);
2225        }
2226        break;
2227    case INDEX_op_xor_i32:
2228        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2229        if (const_args[2]) {
2230            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2231            tcg_out_insn(s, RIL, XILF, a0, a2);
2232        } else if (a0 == a1) {
2233            tcg_out_insn(s, RR, XR, args[0], args[2]);
2234        } else {
2235            tcg_out_insn(s, RRFa, XRK, a0, a1, a2);
2236        }
2237        break;
2238
2239    case INDEX_op_andc_i32:
2240        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2241        if (const_args[2]) {
2242            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2243            tgen_andi(s, TCG_TYPE_I32, a0, (uint32_t)~a2);
2244	} else {
2245            tcg_out_insn(s, RRFa, NCRK, a0, a1, a2);
2246	}
2247        break;
2248    case INDEX_op_orc_i32:
2249        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2250        if (const_args[2]) {
2251            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2252            tgen_ori(s, a0, (uint32_t)~a2);
2253        } else {
2254            tcg_out_insn(s, RRFa, OCRK, a0, a1, a2);
2255        }
2256        break;
2257    case INDEX_op_eqv_i32:
2258        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2259        if (const_args[2]) {
2260            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2261            tcg_out_insn(s, RIL, XILF, a0, ~a2);
2262        } else {
2263            tcg_out_insn(s, RRFa, NXRK, a0, a1, a2);
2264        }
2265        break;
2266    case INDEX_op_nand_i32:
2267        tcg_out_insn(s, RRFa, NNRK, args[0], args[1], args[2]);
2268        break;
2269    case INDEX_op_nor_i32:
2270        tcg_out_insn(s, RRFa, NORK, args[0], args[1], args[2]);
2271        break;
2272
2273    case INDEX_op_neg_i32:
2274        tcg_out_insn(s, RR, LCR, args[0], args[1]);
2275        break;
2276    case INDEX_op_not_i32:
2277        tcg_out_insn(s, RRFa, NORK, args[0], args[1], args[1]);
2278        break;
2279
2280    case INDEX_op_mul_i32:
2281        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2282        if (const_args[2]) {
2283            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2284            if (a2 == (int16_t)a2) {
2285                tcg_out_insn(s, RI, MHI, a0, a2);
2286            } else {
2287                tcg_out_insn(s, RIL, MSFI, a0, a2);
2288            }
2289        } else if (a0 == a1) {
2290            tcg_out_insn(s, RRE, MSR, a0, a2);
2291        } else {
2292            tcg_out_insn(s, RRFa, MSRKC, a0, a1, a2);
2293        }
2294        break;
2295
2296    case INDEX_op_div2_i32:
2297        tcg_debug_assert(args[0] == args[2]);
2298        tcg_debug_assert(args[1] == args[3]);
2299        tcg_debug_assert((args[1] & 1) == 0);
2300        tcg_debug_assert(args[0] == args[1] + 1);
2301        tcg_out_insn(s, RR, DR, args[1], args[4]);
2302        break;
2303    case INDEX_op_divu2_i32:
2304        tcg_debug_assert(args[0] == args[2]);
2305        tcg_debug_assert(args[1] == args[3]);
2306        tcg_debug_assert((args[1] & 1) == 0);
2307        tcg_debug_assert(args[0] == args[1] + 1);
2308        tcg_out_insn(s, RRE, DLR, args[1], args[4]);
2309        break;
2310
2311    case INDEX_op_shl_i32:
2312        op = RS_SLL;
2313        op2 = RSY_SLLK;
2314    do_shift32:
2315        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2316        if (a0 == a1) {
2317            if (const_args[2]) {
2318                tcg_out_sh32(s, op, a0, TCG_REG_NONE, a2);
2319            } else {
2320                tcg_out_sh32(s, op, a0, a2, 0);
2321            }
2322        } else {
2323            /* Using tcg_out_sh64 here for the format; it is a 32-bit shift.  */
2324            if (const_args[2]) {
2325                tcg_out_sh64(s, op2, a0, a1, TCG_REG_NONE, a2);
2326            } else {
2327                tcg_out_sh64(s, op2, a0, a1, a2, 0);
2328            }
2329        }
2330        break;
2331    case INDEX_op_shr_i32:
2332        op = RS_SRL;
2333        op2 = RSY_SRLK;
2334        goto do_shift32;
2335    case INDEX_op_sar_i32:
2336        op = RS_SRA;
2337        op2 = RSY_SRAK;
2338        goto do_shift32;
2339
2340    case INDEX_op_rotl_i32:
2341        /* ??? Using tcg_out_sh64 here for the format; it is a 32-bit rol.  */
2342        if (const_args[2]) {
2343            tcg_out_sh64(s, RSY_RLL, args[0], args[1], TCG_REG_NONE, args[2]);
2344        } else {
2345            tcg_out_sh64(s, RSY_RLL, args[0], args[1], args[2], 0);
2346        }
2347        break;
2348    case INDEX_op_rotr_i32:
2349        if (const_args[2]) {
2350            tcg_out_sh64(s, RSY_RLL, args[0], args[1],
2351                         TCG_REG_NONE, (32 - args[2]) & 31);
2352        } else {
2353            tcg_out_insn(s, RR, LCR, TCG_TMP0, args[2]);
2354            tcg_out_sh64(s, RSY_RLL, args[0], args[1], TCG_TMP0, 0);
2355        }
2356        break;
2357
2358    case INDEX_op_bswap16_i32:
2359        a0 = args[0], a1 = args[1], a2 = args[2];
2360        tcg_out_insn(s, RRE, LRVR, a0, a1);
2361        if (a2 & TCG_BSWAP_OS) {
2362            tcg_out_sh32(s, RS_SRA, a0, TCG_REG_NONE, 16);
2363        } else {
2364            tcg_out_sh32(s, RS_SRL, a0, TCG_REG_NONE, 16);
2365        }
2366        break;
2367    case INDEX_op_bswap16_i64:
2368        a0 = args[0], a1 = args[1], a2 = args[2];
2369        tcg_out_insn(s, RRE, LRVGR, a0, a1);
2370        if (a2 & TCG_BSWAP_OS) {
2371            tcg_out_sh64(s, RSY_SRAG, a0, a0, TCG_REG_NONE, 48);
2372        } else {
2373            tcg_out_sh64(s, RSY_SRLG, a0, a0, TCG_REG_NONE, 48);
2374        }
2375        break;
2376
2377    case INDEX_op_bswap32_i32:
2378        tcg_out_insn(s, RRE, LRVR, args[0], args[1]);
2379        break;
2380    case INDEX_op_bswap32_i64:
2381        a0 = args[0], a1 = args[1], a2 = args[2];
2382        tcg_out_insn(s, RRE, LRVR, a0, a1);
2383        if (a2 & TCG_BSWAP_OS) {
2384            tcg_out_ext32s(s, a0, a0);
2385        } else if ((a2 & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
2386            tcg_out_ext32u(s, a0, a0);
2387        }
2388        break;
2389
2390    case INDEX_op_add2_i32:
2391        if (const_args[4]) {
2392            tcg_out_insn(s, RIL, ALFI, args[0], args[4]);
2393        } else {
2394            tcg_out_insn(s, RR, ALR, args[0], args[4]);
2395        }
2396        tcg_out_insn(s, RRE, ALCR, args[1], args[5]);
2397        break;
2398    case INDEX_op_sub2_i32:
2399        if (const_args[4]) {
2400            tcg_out_insn(s, RIL, SLFI, args[0], args[4]);
2401        } else {
2402            tcg_out_insn(s, RR, SLR, args[0], args[4]);
2403        }
2404        tcg_out_insn(s, RRE, SLBR, args[1], args[5]);
2405        break;
2406
2407    case INDEX_op_br:
2408        tgen_branch(s, S390_CC_ALWAYS, arg_label(args[0]));
2409        break;
2410
2411    case INDEX_op_brcond_i32:
2412        tgen_brcond(s, TCG_TYPE_I32, args[2], args[0],
2413                    args[1], const_args[1], arg_label(args[3]));
2414        break;
2415    case INDEX_op_setcond_i32:
2416        tgen_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1],
2417                     args[2], const_args[2], false);
2418        break;
2419    case INDEX_op_negsetcond_i32:
2420        tgen_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1],
2421                     args[2], const_args[2], true);
2422        break;
2423    case INDEX_op_movcond_i32:
2424        tgen_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1],
2425                     args[2], const_args[2], args[3], const_args[3], args[4]);
2426        break;
2427
2428    case INDEX_op_qemu_ld_a32_i32:
2429    case INDEX_op_qemu_ld_a64_i32:
2430        tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I32);
2431        break;
2432    case INDEX_op_qemu_ld_a32_i64:
2433    case INDEX_op_qemu_ld_a64_i64:
2434        tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I64);
2435        break;
2436    case INDEX_op_qemu_st_a32_i32:
2437    case INDEX_op_qemu_st_a64_i32:
2438        tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I32);
2439        break;
2440    case INDEX_op_qemu_st_a32_i64:
2441    case INDEX_op_qemu_st_a64_i64:
2442        tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I64);
2443        break;
2444    case INDEX_op_qemu_ld_a32_i128:
2445    case INDEX_op_qemu_ld_a64_i128:
2446        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], true);
2447        break;
2448    case INDEX_op_qemu_st_a32_i128:
2449    case INDEX_op_qemu_st_a64_i128:
2450        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], false);
2451        break;
2452
2453    case INDEX_op_ld16s_i64:
2454        tcg_out_mem(s, 0, RXY_LGH, args[0], args[1], TCG_REG_NONE, args[2]);
2455        break;
2456    case INDEX_op_ld32u_i64:
2457        tcg_out_mem(s, 0, RXY_LLGF, args[0], args[1], TCG_REG_NONE, args[2]);
2458        break;
2459    case INDEX_op_ld32s_i64:
2460        tcg_out_mem(s, 0, RXY_LGF, args[0], args[1], TCG_REG_NONE, args[2]);
2461        break;
2462    case INDEX_op_ld_i64:
2463        tcg_out_ld(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2464        break;
2465
2466    case INDEX_op_st32_i64:
2467        tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2468        break;
2469    case INDEX_op_st_i64:
2470        tcg_out_st(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2471        break;
2472
2473    case INDEX_op_add_i64:
2474        a0 = args[0], a1 = args[1], a2 = args[2];
2475        if (const_args[2]) {
2476        do_addi_64:
2477            if (a0 == a1) {
2478                if (a2 == (int16_t)a2) {
2479                    tcg_out_insn(s, RI, AGHI, a0, a2);
2480                    break;
2481                }
2482                if (a2 == (int32_t)a2) {
2483                    tcg_out_insn(s, RIL, AGFI, a0, a2);
2484                    break;
2485                }
2486                if (a2 == (uint32_t)a2) {
2487                    tcg_out_insn(s, RIL, ALGFI, a0, a2);
2488                    break;
2489                }
2490                if (-a2 == (uint32_t)-a2) {
2491                    tcg_out_insn(s, RIL, SLGFI, a0, -a2);
2492                    break;
2493                }
2494            }
2495            tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
2496        } else if (a0 == a1) {
2497            tcg_out_insn(s, RRE, AGR, a0, a2);
2498        } else {
2499            tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
2500        }
2501        break;
2502    case INDEX_op_sub_i64:
2503        a0 = args[0], a1 = args[1], a2 = args[2];
2504        if (const_args[2]) {
2505            a2 = -a2;
2506            goto do_addi_64;
2507        } else {
2508            tcg_out_insn(s, RRFa, SGRK, a0, a1, a2);
2509        }
2510        break;
2511
2512    case INDEX_op_and_i64:
2513        a0 = args[0], a1 = args[1], a2 = args[2];
2514        if (const_args[2]) {
2515            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2516            tgen_andi(s, TCG_TYPE_I64, args[0], args[2]);
2517        } else {
2518            tcg_out_insn(s, RRFa, NGRK, a0, a1, a2);
2519        }
2520        break;
2521    case INDEX_op_or_i64:
2522        a0 = args[0], a1 = args[1], a2 = args[2];
2523        if (const_args[2]) {
2524            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2525            tgen_ori(s, a0, a2);
2526        } else {
2527            tcg_out_insn(s, RRFa, OGRK, a0, a1, a2);
2528        }
2529        break;
2530    case INDEX_op_xor_i64:
2531        a0 = args[0], a1 = args[1], a2 = args[2];
2532        if (const_args[2]) {
2533            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2534            tgen_xori(s, a0, a2);
2535        } else {
2536            tcg_out_insn(s, RRFa, XGRK, a0, a1, a2);
2537        }
2538        break;
2539
2540    case INDEX_op_andc_i64:
2541        a0 = args[0], a1 = args[1], a2 = args[2];
2542        if (const_args[2]) {
2543            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2544            tgen_andi(s, TCG_TYPE_I64, a0, ~a2);
2545        } else {
2546            tcg_out_insn(s, RRFa, NCGRK, a0, a1, a2);
2547        }
2548        break;
2549    case INDEX_op_orc_i64:
2550        a0 = args[0], a1 = args[1], a2 = args[2];
2551        if (const_args[2]) {
2552            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2553            tgen_ori(s, a0, ~a2);
2554        } else {
2555            tcg_out_insn(s, RRFa, OCGRK, a0, a1, a2);
2556        }
2557        break;
2558    case INDEX_op_eqv_i64:
2559        a0 = args[0], a1 = args[1], a2 = args[2];
2560        if (const_args[2]) {
2561            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2562            tgen_xori(s, a0, ~a2);
2563        } else {
2564            tcg_out_insn(s, RRFa, NXGRK, a0, a1, a2);
2565        }
2566        break;
2567    case INDEX_op_nand_i64:
2568        tcg_out_insn(s, RRFa, NNGRK, args[0], args[1], args[2]);
2569        break;
2570    case INDEX_op_nor_i64:
2571        tcg_out_insn(s, RRFa, NOGRK, args[0], args[1], args[2]);
2572        break;
2573
2574    case INDEX_op_neg_i64:
2575        tcg_out_insn(s, RRE, LCGR, args[0], args[1]);
2576        break;
2577    case INDEX_op_not_i64:
2578        tcg_out_insn(s, RRFa, NOGRK, args[0], args[1], args[1]);
2579        break;
2580    case INDEX_op_bswap64_i64:
2581        tcg_out_insn(s, RRE, LRVGR, args[0], args[1]);
2582        break;
2583
2584    case INDEX_op_mul_i64:
2585        a0 = args[0], a1 = args[1], a2 = args[2];
2586        if (const_args[2]) {
2587            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2588            if (a2 == (int16_t)a2) {
2589                tcg_out_insn(s, RI, MGHI, a0, a2);
2590            } else {
2591                tcg_out_insn(s, RIL, MSGFI, a0, a2);
2592            }
2593        } else if (a0 == a1) {
2594            tcg_out_insn(s, RRE, MSGR, a0, a2);
2595        } else {
2596            tcg_out_insn(s, RRFa, MSGRKC, a0, a1, a2);
2597        }
2598        break;
2599
2600    case INDEX_op_div2_i64:
2601        /*
2602         * ??? We get an unnecessary sign-extension of the dividend
2603         * into op0 with this definition, but as we do in fact always
2604         * produce both quotient and remainder using INDEX_op_div_i64
2605         * instead requires jumping through even more hoops.
2606         */
2607        tcg_debug_assert(args[0] == args[2]);
2608        tcg_debug_assert(args[1] == args[3]);
2609        tcg_debug_assert((args[1] & 1) == 0);
2610        tcg_debug_assert(args[0] == args[1] + 1);
2611        tcg_out_insn(s, RRE, DSGR, args[1], args[4]);
2612        break;
2613    case INDEX_op_divu2_i64:
2614        tcg_debug_assert(args[0] == args[2]);
2615        tcg_debug_assert(args[1] == args[3]);
2616        tcg_debug_assert((args[1] & 1) == 0);
2617        tcg_debug_assert(args[0] == args[1] + 1);
2618        tcg_out_insn(s, RRE, DLGR, args[1], args[4]);
2619        break;
2620    case INDEX_op_mulu2_i64:
2621        tcg_debug_assert(args[0] == args[2]);
2622        tcg_debug_assert((args[1] & 1) == 0);
2623        tcg_debug_assert(args[0] == args[1] + 1);
2624        tcg_out_insn(s, RRE, MLGR, args[1], args[3]);
2625        break;
2626    case INDEX_op_muls2_i64:
2627        tcg_debug_assert((args[1] & 1) == 0);
2628        tcg_debug_assert(args[0] == args[1] + 1);
2629        tcg_out_insn(s, RRFa, MGRK, args[1], args[2], args[3]);
2630        break;
2631
2632    case INDEX_op_shl_i64:
2633        op = RSY_SLLG;
2634    do_shift64:
2635        if (const_args[2]) {
2636            tcg_out_sh64(s, op, args[0], args[1], TCG_REG_NONE, args[2]);
2637        } else {
2638            tcg_out_sh64(s, op, args[0], args[1], args[2], 0);
2639        }
2640        break;
2641    case INDEX_op_shr_i64:
2642        op = RSY_SRLG;
2643        goto do_shift64;
2644    case INDEX_op_sar_i64:
2645        op = RSY_SRAG;
2646        goto do_shift64;
2647
2648    case INDEX_op_rotl_i64:
2649        if (const_args[2]) {
2650            tcg_out_sh64(s, RSY_RLLG, args[0], args[1],
2651                         TCG_REG_NONE, args[2]);
2652        } else {
2653            tcg_out_sh64(s, RSY_RLLG, args[0], args[1], args[2], 0);
2654        }
2655        break;
2656    case INDEX_op_rotr_i64:
2657        if (const_args[2]) {
2658            tcg_out_sh64(s, RSY_RLLG, args[0], args[1],
2659                         TCG_REG_NONE, (64 - args[2]) & 63);
2660        } else {
2661            /* We can use the smaller 32-bit negate because only the
2662               low 6 bits are examined for the rotate.  */
2663            tcg_out_insn(s, RR, LCR, TCG_TMP0, args[2]);
2664            tcg_out_sh64(s, RSY_RLLG, args[0], args[1], TCG_TMP0, 0);
2665        }
2666        break;
2667
2668    case INDEX_op_add2_i64:
2669        if (const_args[4]) {
2670            if ((int64_t)args[4] >= 0) {
2671                tcg_out_insn(s, RIL, ALGFI, args[0], args[4]);
2672            } else {
2673                tcg_out_insn(s, RIL, SLGFI, args[0], -args[4]);
2674            }
2675        } else {
2676            tcg_out_insn(s, RRE, ALGR, args[0], args[4]);
2677        }
2678        tcg_out_insn(s, RRE, ALCGR, args[1], args[5]);
2679        break;
2680    case INDEX_op_sub2_i64:
2681        if (const_args[4]) {
2682            if ((int64_t)args[4] >= 0) {
2683                tcg_out_insn(s, RIL, SLGFI, args[0], args[4]);
2684            } else {
2685                tcg_out_insn(s, RIL, ALGFI, args[0], -args[4]);
2686            }
2687        } else {
2688            tcg_out_insn(s, RRE, SLGR, args[0], args[4]);
2689        }
2690        tcg_out_insn(s, RRE, SLBGR, args[1], args[5]);
2691        break;
2692
2693    case INDEX_op_brcond_i64:
2694        tgen_brcond(s, TCG_TYPE_I64, args[2], args[0],
2695                    args[1], const_args[1], arg_label(args[3]));
2696        break;
2697    case INDEX_op_setcond_i64:
2698        tgen_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1],
2699                     args[2], const_args[2], false);
2700        break;
2701    case INDEX_op_negsetcond_i64:
2702        tgen_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1],
2703                     args[2], const_args[2], true);
2704        break;
2705    case INDEX_op_movcond_i64:
2706        tgen_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1],
2707                     args[2], const_args[2], args[3], const_args[3], args[4]);
2708        break;
2709
2710    OP_32_64(deposit):
2711        a0 = args[0], a1 = args[1], a2 = args[2];
2712        if (const_args[1]) {
2713            tgen_deposit(s, a0, a2, args[3], args[4], 1);
2714        } else {
2715            /* Since we can't support "0Z" as a constraint, we allow a1 in
2716               any register.  Fix things up as if a matching constraint.  */
2717            if (a0 != a1) {
2718                TCGType type = (opc == INDEX_op_deposit_i64);
2719                if (a0 == a2) {
2720                    tcg_out_mov(s, type, TCG_TMP0, a2);
2721                    a2 = TCG_TMP0;
2722                }
2723                tcg_out_mov(s, type, a0, a1);
2724            }
2725            tgen_deposit(s, a0, a2, args[3], args[4], 0);
2726        }
2727        break;
2728
2729    OP_32_64(extract):
2730        tgen_extract(s, args[0], args[1], args[2], args[3]);
2731        break;
2732
2733    case INDEX_op_clz_i64:
2734        tgen_clz(s, args[0], args[1], args[2], const_args[2]);
2735        break;
2736
2737    case INDEX_op_ctpop_i32:
2738        tgen_ctpop(s, TCG_TYPE_I32, args[0], args[1]);
2739        break;
2740    case INDEX_op_ctpop_i64:
2741        tgen_ctpop(s, TCG_TYPE_I64, args[0], args[1]);
2742        break;
2743
2744    case INDEX_op_mb:
2745        /* The host memory model is quite strong, we simply need to
2746           serialize the instruction stream.  */
2747        if (args[0] & TCG_MO_ST_LD) {
2748            /* fast-bcr-serialization facility (45) is present */
2749            tcg_out_insn(s, RR, BCR, 14, 0);
2750        }
2751        break;
2752
2753    case INDEX_op_mov_i32:  /* Always emitted via tcg_out_mov.  */
2754    case INDEX_op_mov_i64:
2755    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
2756    case INDEX_op_exit_tb:  /* Always emitted via tcg_out_exit_tb.  */
2757    case INDEX_op_goto_tb:  /* Always emitted via tcg_out_goto_tb.  */
2758    case INDEX_op_ext8s_i32:  /* Always emitted via tcg_reg_alloc_op.  */
2759    case INDEX_op_ext8s_i64:
2760    case INDEX_op_ext8u_i32:
2761    case INDEX_op_ext8u_i64:
2762    case INDEX_op_ext16s_i32:
2763    case INDEX_op_ext16s_i64:
2764    case INDEX_op_ext16u_i32:
2765    case INDEX_op_ext16u_i64:
2766    case INDEX_op_ext32s_i64:
2767    case INDEX_op_ext32u_i64:
2768    case INDEX_op_ext_i32_i64:
2769    case INDEX_op_extu_i32_i64:
2770    case INDEX_op_extrl_i64_i32:
2771    default:
2772        g_assert_not_reached();
2773    }
2774}
2775
2776static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
2777                            TCGReg dst, TCGReg src)
2778{
2779    if (is_general_reg(src)) {
2780        /* Replicate general register into two MO_64. */
2781        tcg_out_insn(s, VRRf, VLVGP, dst, src, src);
2782        if (vece == MO_64) {
2783            return true;
2784        }
2785        src = dst;
2786    }
2787
2788    /*
2789     * Recall that the "standard" integer, within a vector, is the
2790     * rightmost element of the leftmost doubleword, a-la VLLEZ.
2791     */
2792    tcg_out_insn(s, VRIc, VREP, dst, (8 >> vece) - 1, src, vece);
2793    return true;
2794}
2795
2796static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
2797                             TCGReg dst, TCGReg base, intptr_t offset)
2798{
2799    tcg_out_vrx_mem(s, VRX_VLREP, dst, base, TCG_REG_NONE, offset, vece);
2800    return true;
2801}
2802
2803static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
2804                             TCGReg dst, int64_t val)
2805{
2806    int i, mask, msb, lsb;
2807
2808    /* Look for int16_t elements.  */
2809    if (vece <= MO_16 ||
2810        (vece == MO_32 ? (int32_t)val : val) == (int16_t)val) {
2811        tcg_out_insn(s, VRIa, VREPI, dst, val, vece);
2812        return;
2813    }
2814
2815    /* Look for bit masks.  */
2816    if (vece == MO_32) {
2817        if (risbg_mask((int32_t)val)) {
2818            /* Handle wraparound by swapping msb and lsb.  */
2819            if ((val & 0x80000001u) == 0x80000001u) {
2820                msb = 32 - ctz32(~val);
2821                lsb = clz32(~val) - 1;
2822            } else {
2823                msb = clz32(val);
2824                lsb = 31 - ctz32(val);
2825            }
2826            tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_32);
2827            return;
2828        }
2829    } else {
2830        if (risbg_mask(val)) {
2831            /* Handle wraparound by swapping msb and lsb.  */
2832            if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
2833                /* Handle wraparound by swapping msb and lsb.  */
2834                msb = 64 - ctz64(~val);
2835                lsb = clz64(~val) - 1;
2836            } else {
2837                msb = clz64(val);
2838                lsb = 63 - ctz64(val);
2839            }
2840            tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_64);
2841            return;
2842        }
2843    }
2844
2845    /* Look for all bytes 0x00 or 0xff.  */
2846    for (i = mask = 0; i < 8; i++) {
2847        uint8_t byte = val >> (i * 8);
2848        if (byte == 0xff) {
2849            mask |= 1 << i;
2850        } else if (byte != 0) {
2851            break;
2852        }
2853    }
2854    if (i == 8) {
2855        tcg_out_insn(s, VRIa, VGBM, dst, mask * 0x0101, 0);
2856        return;
2857    }
2858
2859    /* Otherwise, stuff it in the constant pool.  */
2860    tcg_out_insn(s, RIL, LARL, TCG_TMP0, 0);
2861    new_pool_label(s, val, R_390_PC32DBL, s->code_ptr - 2, 2);
2862    tcg_out_insn(s, VRX, VLREP, dst, TCG_TMP0, TCG_REG_NONE, 0, MO_64);
2863}
2864
2865static bool tcg_out_cmp_vec_noinv(TCGContext *s, unsigned vece, TCGReg a0,
2866                                  TCGReg a1, TCGReg a2, TCGCond cond)
2867{
2868    bool need_swap = false, need_inv = false;
2869
2870    switch (cond) {
2871    case TCG_COND_EQ:
2872    case TCG_COND_GT:
2873    case TCG_COND_GTU:
2874        break;
2875    case TCG_COND_NE:
2876    case TCG_COND_LE:
2877    case TCG_COND_LEU:
2878        need_inv = true;
2879        break;
2880    case TCG_COND_LT:
2881    case TCG_COND_LTU:
2882        need_swap = true;
2883        break;
2884    case TCG_COND_GE:
2885    case TCG_COND_GEU:
2886        need_swap = need_inv = true;
2887        break;
2888    default:
2889        g_assert_not_reached();
2890    }
2891
2892    if (need_inv) {
2893        cond = tcg_invert_cond(cond);
2894    }
2895    if (need_swap) {
2896        TCGReg swap = a1;
2897        a1 = a2;
2898        a2 = swap;
2899        cond = tcg_swap_cond(cond);
2900    }
2901
2902    switch (cond) {
2903    case TCG_COND_EQ:
2904        tcg_out_insn(s, VRRc, VCEQ, a0, a1, a2, vece);
2905        break;
2906    case TCG_COND_GT:
2907        tcg_out_insn(s, VRRc, VCH, a0, a1, a2, vece);
2908        break;
2909    case TCG_COND_GTU:
2910        tcg_out_insn(s, VRRc, VCHL, a0, a1, a2, vece);
2911        break;
2912    default:
2913        g_assert_not_reached();
2914    }
2915    return need_inv;
2916}
2917
2918static void tcg_out_cmp_vec(TCGContext *s, unsigned vece, TCGReg a0,
2919                            TCGReg a1, TCGReg a2, TCGCond cond)
2920{
2921    if (tcg_out_cmp_vec_noinv(s, vece, a0, a1, a2, cond)) {
2922        tcg_out_insn(s, VRRc, VNO, a0, a0, a0, 0);
2923    }
2924}
2925
2926static void tcg_out_cmpsel_vec(TCGContext *s, unsigned vece, TCGReg a0,
2927                               TCGReg c1, TCGReg c2, TCGArg v3,
2928                               int const_v3, TCGReg v4, TCGCond cond)
2929{
2930    bool inv = tcg_out_cmp_vec_noinv(s, vece, TCG_VEC_TMP0, c1, c2, cond);
2931
2932    if (!const_v3) {
2933        if (inv) {
2934            tcg_out_insn(s, VRRe, VSEL, a0, v4, v3, TCG_VEC_TMP0);
2935        } else {
2936            tcg_out_insn(s, VRRe, VSEL, a0, v3, v4, TCG_VEC_TMP0);
2937        }
2938    } else if (v3) {
2939        if (inv) {
2940            tcg_out_insn(s, VRRc, VOC, a0, v4, TCG_VEC_TMP0, 0);
2941        } else {
2942            tcg_out_insn(s, VRRc, VO, a0, v4, TCG_VEC_TMP0, 0);
2943        }
2944    } else {
2945        if (inv) {
2946            tcg_out_insn(s, VRRc, VN, a0, v4, TCG_VEC_TMP0, 0);
2947        } else {
2948            tcg_out_insn(s, VRRc, VNC, a0, v4, TCG_VEC_TMP0, 0);
2949        }
2950    }
2951}
2952
2953static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
2954                           unsigned vecl, unsigned vece,
2955                           const TCGArg args[TCG_MAX_OP_ARGS],
2956                           const int const_args[TCG_MAX_OP_ARGS])
2957{
2958    TCGType type = vecl + TCG_TYPE_V64;
2959    TCGArg a0 = args[0], a1 = args[1], a2 = args[2];
2960
2961    switch (opc) {
2962    case INDEX_op_ld_vec:
2963        tcg_out_ld(s, type, a0, a1, a2);
2964        break;
2965    case INDEX_op_st_vec:
2966        tcg_out_st(s, type, a0, a1, a2);
2967        break;
2968    case INDEX_op_dupm_vec:
2969        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
2970        break;
2971
2972    case INDEX_op_abs_vec:
2973        tcg_out_insn(s, VRRa, VLP, a0, a1, vece);
2974        break;
2975    case INDEX_op_neg_vec:
2976        tcg_out_insn(s, VRRa, VLC, a0, a1, vece);
2977        break;
2978    case INDEX_op_not_vec:
2979        tcg_out_insn(s, VRRc, VNO, a0, a1, a1, 0);
2980        break;
2981
2982    case INDEX_op_add_vec:
2983        tcg_out_insn(s, VRRc, VA, a0, a1, a2, vece);
2984        break;
2985    case INDEX_op_sub_vec:
2986        tcg_out_insn(s, VRRc, VS, a0, a1, a2, vece);
2987        break;
2988    case INDEX_op_and_vec:
2989        tcg_out_insn(s, VRRc, VN, a0, a1, a2, 0);
2990        break;
2991    case INDEX_op_andc_vec:
2992        tcg_out_insn(s, VRRc, VNC, a0, a1, a2, 0);
2993        break;
2994    case INDEX_op_mul_vec:
2995        tcg_out_insn(s, VRRc, VML, a0, a1, a2, vece);
2996        break;
2997    case INDEX_op_or_vec:
2998        tcg_out_insn(s, VRRc, VO, a0, a1, a2, 0);
2999        break;
3000    case INDEX_op_orc_vec:
3001        tcg_out_insn(s, VRRc, VOC, a0, a1, a2, 0);
3002        break;
3003    case INDEX_op_xor_vec:
3004        tcg_out_insn(s, VRRc, VX, a0, a1, a2, 0);
3005        break;
3006    case INDEX_op_nand_vec:
3007        tcg_out_insn(s, VRRc, VNN, a0, a1, a2, 0);
3008        break;
3009    case INDEX_op_nor_vec:
3010        tcg_out_insn(s, VRRc, VNO, a0, a1, a2, 0);
3011        break;
3012    case INDEX_op_eqv_vec:
3013        tcg_out_insn(s, VRRc, VNX, a0, a1, a2, 0);
3014        break;
3015
3016    case INDEX_op_shli_vec:
3017        tcg_out_insn(s, VRSa, VESL, a0, a2, TCG_REG_NONE, a1, vece);
3018        break;
3019    case INDEX_op_shri_vec:
3020        tcg_out_insn(s, VRSa, VESRL, a0, a2, TCG_REG_NONE, a1, vece);
3021        break;
3022    case INDEX_op_sari_vec:
3023        tcg_out_insn(s, VRSa, VESRA, a0, a2, TCG_REG_NONE, a1, vece);
3024        break;
3025    case INDEX_op_rotli_vec:
3026        tcg_out_insn(s, VRSa, VERLL, a0, a2, TCG_REG_NONE, a1, vece);
3027        break;
3028    case INDEX_op_shls_vec:
3029        tcg_out_insn(s, VRSa, VESL, a0, 0, a2, a1, vece);
3030        break;
3031    case INDEX_op_shrs_vec:
3032        tcg_out_insn(s, VRSa, VESRL, a0, 0, a2, a1, vece);
3033        break;
3034    case INDEX_op_sars_vec:
3035        tcg_out_insn(s, VRSa, VESRA, a0, 0, a2, a1, vece);
3036        break;
3037    case INDEX_op_rotls_vec:
3038        tcg_out_insn(s, VRSa, VERLL, a0, 0, a2, a1, vece);
3039        break;
3040    case INDEX_op_shlv_vec:
3041        tcg_out_insn(s, VRRc, VESLV, a0, a1, a2, vece);
3042        break;
3043    case INDEX_op_shrv_vec:
3044        tcg_out_insn(s, VRRc, VESRLV, a0, a1, a2, vece);
3045        break;
3046    case INDEX_op_sarv_vec:
3047        tcg_out_insn(s, VRRc, VESRAV, a0, a1, a2, vece);
3048        break;
3049    case INDEX_op_rotlv_vec:
3050        tcg_out_insn(s, VRRc, VERLLV, a0, a1, a2, vece);
3051        break;
3052
3053    case INDEX_op_smin_vec:
3054        tcg_out_insn(s, VRRc, VMN, a0, a1, a2, vece);
3055        break;
3056    case INDEX_op_smax_vec:
3057        tcg_out_insn(s, VRRc, VMX, a0, a1, a2, vece);
3058        break;
3059    case INDEX_op_umin_vec:
3060        tcg_out_insn(s, VRRc, VMNL, a0, a1, a2, vece);
3061        break;
3062    case INDEX_op_umax_vec:
3063        tcg_out_insn(s, VRRc, VMXL, a0, a1, a2, vece);
3064        break;
3065
3066    case INDEX_op_bitsel_vec:
3067        tcg_out_insn(s, VRRe, VSEL, a0, a2, args[3], a1);
3068        break;
3069
3070    case INDEX_op_cmp_vec:
3071        tcg_out_cmp_vec(s, vece, a0, a1, a2, args[3]);
3072        break;
3073    case INDEX_op_cmpsel_vec:
3074        tcg_out_cmpsel_vec(s, vece, a0, a1, a2, args[3], const_args[3],
3075                           args[4], args[5]);
3076        break;
3077
3078    case INDEX_op_s390_vuph_vec:
3079        tcg_out_insn(s, VRRa, VUPH, a0, a1, vece);
3080        break;
3081    case INDEX_op_s390_vupl_vec:
3082        tcg_out_insn(s, VRRa, VUPL, a0, a1, vece);
3083        break;
3084    case INDEX_op_s390_vpks_vec:
3085        tcg_out_insn(s, VRRc, VPKS, a0, a1, a2, vece);
3086        break;
3087
3088    case INDEX_op_mov_vec:   /* Always emitted via tcg_out_mov.  */
3089    case INDEX_op_dup_vec:   /* Always emitted via tcg_out_dup_vec.  */
3090    default:
3091        g_assert_not_reached();
3092    }
3093}
3094
3095int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
3096{
3097    switch (opc) {
3098    case INDEX_op_abs_vec:
3099    case INDEX_op_add_vec:
3100    case INDEX_op_and_vec:
3101    case INDEX_op_andc_vec:
3102    case INDEX_op_bitsel_vec:
3103    case INDEX_op_eqv_vec:
3104    case INDEX_op_nand_vec:
3105    case INDEX_op_neg_vec:
3106    case INDEX_op_nor_vec:
3107    case INDEX_op_not_vec:
3108    case INDEX_op_or_vec:
3109    case INDEX_op_orc_vec:
3110    case INDEX_op_rotli_vec:
3111    case INDEX_op_rotls_vec:
3112    case INDEX_op_rotlv_vec:
3113    case INDEX_op_sari_vec:
3114    case INDEX_op_sars_vec:
3115    case INDEX_op_sarv_vec:
3116    case INDEX_op_shli_vec:
3117    case INDEX_op_shls_vec:
3118    case INDEX_op_shlv_vec:
3119    case INDEX_op_shri_vec:
3120    case INDEX_op_shrs_vec:
3121    case INDEX_op_shrv_vec:
3122    case INDEX_op_smax_vec:
3123    case INDEX_op_smin_vec:
3124    case INDEX_op_sub_vec:
3125    case INDEX_op_umax_vec:
3126    case INDEX_op_umin_vec:
3127    case INDEX_op_xor_vec:
3128    case INDEX_op_cmp_vec:
3129    case INDEX_op_cmpsel_vec:
3130        return 1;
3131    case INDEX_op_rotrv_vec:
3132        return -1;
3133    case INDEX_op_mul_vec:
3134        return vece < MO_64;
3135    case INDEX_op_ssadd_vec:
3136    case INDEX_op_sssub_vec:
3137        return vece < MO_64 ? -1 : 0;
3138    default:
3139        return 0;
3140    }
3141}
3142
3143static void expand_vec_sat(TCGType type, unsigned vece, TCGv_vec v0,
3144                           TCGv_vec v1, TCGv_vec v2, TCGOpcode add_sub_opc)
3145{
3146    TCGv_vec h1 = tcg_temp_new_vec(type);
3147    TCGv_vec h2 = tcg_temp_new_vec(type);
3148    TCGv_vec l1 = tcg_temp_new_vec(type);
3149    TCGv_vec l2 = tcg_temp_new_vec(type);
3150
3151    tcg_debug_assert (vece < MO_64);
3152
3153    /* Unpack with sign-extension. */
3154    vec_gen_2(INDEX_op_s390_vuph_vec, type, vece,
3155              tcgv_vec_arg(h1), tcgv_vec_arg(v1));
3156    vec_gen_2(INDEX_op_s390_vuph_vec, type, vece,
3157              tcgv_vec_arg(h2), tcgv_vec_arg(v2));
3158
3159    vec_gen_2(INDEX_op_s390_vupl_vec, type, vece,
3160              tcgv_vec_arg(l1), tcgv_vec_arg(v1));
3161    vec_gen_2(INDEX_op_s390_vupl_vec, type, vece,
3162              tcgv_vec_arg(l2), tcgv_vec_arg(v2));
3163
3164    /* Arithmetic on a wider element size. */
3165    vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(h1),
3166              tcgv_vec_arg(h1), tcgv_vec_arg(h2));
3167    vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(l1),
3168              tcgv_vec_arg(l1), tcgv_vec_arg(l2));
3169
3170    /* Pack with saturation. */
3171    vec_gen_3(INDEX_op_s390_vpks_vec, type, vece + 1,
3172              tcgv_vec_arg(v0), tcgv_vec_arg(h1), tcgv_vec_arg(l1));
3173
3174    tcg_temp_free_vec(h1);
3175    tcg_temp_free_vec(h2);
3176    tcg_temp_free_vec(l1);
3177    tcg_temp_free_vec(l2);
3178}
3179
3180void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
3181                       TCGArg a0, ...)
3182{
3183    va_list va;
3184    TCGv_vec v0, v1, v2, t0;
3185
3186    va_start(va, a0);
3187    v0 = temp_tcgv_vec(arg_temp(a0));
3188    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3189    v2 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3190
3191    switch (opc) {
3192    case INDEX_op_rotrv_vec:
3193        t0 = tcg_temp_new_vec(type);
3194        tcg_gen_neg_vec(vece, t0, v2);
3195        tcg_gen_rotlv_vec(vece, v0, v1, t0);
3196        tcg_temp_free_vec(t0);
3197        break;
3198
3199    case INDEX_op_ssadd_vec:
3200        expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_add_vec);
3201        break;
3202    case INDEX_op_sssub_vec:
3203        expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_sub_vec);
3204        break;
3205
3206    default:
3207        g_assert_not_reached();
3208    }
3209    va_end(va);
3210}
3211
3212static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
3213{
3214    switch (op) {
3215    case INDEX_op_goto_ptr:
3216        return C_O0_I1(r);
3217
3218    case INDEX_op_ld8u_i32:
3219    case INDEX_op_ld8u_i64:
3220    case INDEX_op_ld8s_i32:
3221    case INDEX_op_ld8s_i64:
3222    case INDEX_op_ld16u_i32:
3223    case INDEX_op_ld16u_i64:
3224    case INDEX_op_ld16s_i32:
3225    case INDEX_op_ld16s_i64:
3226    case INDEX_op_ld_i32:
3227    case INDEX_op_ld32u_i64:
3228    case INDEX_op_ld32s_i64:
3229    case INDEX_op_ld_i64:
3230        return C_O1_I1(r, r);
3231
3232    case INDEX_op_st8_i32:
3233    case INDEX_op_st8_i64:
3234    case INDEX_op_st16_i32:
3235    case INDEX_op_st16_i64:
3236    case INDEX_op_st_i32:
3237    case INDEX_op_st32_i64:
3238    case INDEX_op_st_i64:
3239        return C_O0_I2(r, r);
3240
3241    case INDEX_op_add_i32:
3242    case INDEX_op_add_i64:
3243    case INDEX_op_shl_i64:
3244    case INDEX_op_shr_i64:
3245    case INDEX_op_sar_i64:
3246    case INDEX_op_rotl_i32:
3247    case INDEX_op_rotl_i64:
3248    case INDEX_op_rotr_i32:
3249    case INDEX_op_rotr_i64:
3250        return C_O1_I2(r, r, ri);
3251    case INDEX_op_setcond_i32:
3252    case INDEX_op_negsetcond_i32:
3253    case INDEX_op_setcond_i64:
3254    case INDEX_op_negsetcond_i64:
3255        return C_O1_I2(r, r, rC);
3256
3257    case INDEX_op_clz_i64:
3258        return C_O1_I2(r, r, rI);
3259
3260    case INDEX_op_sub_i32:
3261    case INDEX_op_sub_i64:
3262    case INDEX_op_and_i32:
3263    case INDEX_op_or_i32:
3264    case INDEX_op_xor_i32:
3265        return C_O1_I2(r, r, ri);
3266    case INDEX_op_and_i64:
3267        return C_O1_I2(r, r, rNKR);
3268    case INDEX_op_or_i64:
3269    case INDEX_op_xor_i64:
3270        return C_O1_I2(r, r, rK);
3271
3272    case INDEX_op_andc_i32:
3273    case INDEX_op_orc_i32:
3274    case INDEX_op_eqv_i32:
3275        return C_O1_I2(r, r, ri);
3276    case INDEX_op_andc_i64:
3277        return C_O1_I2(r, r, rKR);
3278    case INDEX_op_orc_i64:
3279    case INDEX_op_eqv_i64:
3280        return C_O1_I2(r, r, rNK);
3281
3282    case INDEX_op_nand_i32:
3283    case INDEX_op_nand_i64:
3284    case INDEX_op_nor_i32:
3285    case INDEX_op_nor_i64:
3286        return C_O1_I2(r, r, r);
3287
3288    case INDEX_op_mul_i32:
3289        return (HAVE_FACILITY(MISC_INSN_EXT2)
3290                ? C_O1_I2(r, r, ri)
3291                : C_O1_I2(r, 0, ri));
3292    case INDEX_op_mul_i64:
3293        return (HAVE_FACILITY(MISC_INSN_EXT2)
3294                ? C_O1_I2(r, r, rJ)
3295                : C_O1_I2(r, 0, rJ));
3296
3297    case INDEX_op_shl_i32:
3298    case INDEX_op_shr_i32:
3299    case INDEX_op_sar_i32:
3300        return C_O1_I2(r, r, ri);
3301
3302    case INDEX_op_brcond_i32:
3303        return C_O0_I2(r, ri);
3304    case INDEX_op_brcond_i64:
3305        return C_O0_I2(r, rC);
3306
3307    case INDEX_op_bswap16_i32:
3308    case INDEX_op_bswap16_i64:
3309    case INDEX_op_bswap32_i32:
3310    case INDEX_op_bswap32_i64:
3311    case INDEX_op_bswap64_i64:
3312    case INDEX_op_neg_i32:
3313    case INDEX_op_neg_i64:
3314    case INDEX_op_not_i32:
3315    case INDEX_op_not_i64:
3316    case INDEX_op_ext8s_i32:
3317    case INDEX_op_ext8s_i64:
3318    case INDEX_op_ext8u_i32:
3319    case INDEX_op_ext8u_i64:
3320    case INDEX_op_ext16s_i32:
3321    case INDEX_op_ext16s_i64:
3322    case INDEX_op_ext16u_i32:
3323    case INDEX_op_ext16u_i64:
3324    case INDEX_op_ext32s_i64:
3325    case INDEX_op_ext32u_i64:
3326    case INDEX_op_ext_i32_i64:
3327    case INDEX_op_extu_i32_i64:
3328    case INDEX_op_extract_i32:
3329    case INDEX_op_extract_i64:
3330    case INDEX_op_ctpop_i32:
3331    case INDEX_op_ctpop_i64:
3332        return C_O1_I1(r, r);
3333
3334    case INDEX_op_qemu_ld_a32_i32:
3335    case INDEX_op_qemu_ld_a64_i32:
3336    case INDEX_op_qemu_ld_a32_i64:
3337    case INDEX_op_qemu_ld_a64_i64:
3338        return C_O1_I1(r, r);
3339    case INDEX_op_qemu_st_a32_i64:
3340    case INDEX_op_qemu_st_a64_i64:
3341    case INDEX_op_qemu_st_a32_i32:
3342    case INDEX_op_qemu_st_a64_i32:
3343        return C_O0_I2(r, r);
3344    case INDEX_op_qemu_ld_a32_i128:
3345    case INDEX_op_qemu_ld_a64_i128:
3346        return C_O2_I1(o, m, r);
3347    case INDEX_op_qemu_st_a32_i128:
3348    case INDEX_op_qemu_st_a64_i128:
3349        return C_O0_I3(o, m, r);
3350
3351    case INDEX_op_deposit_i32:
3352    case INDEX_op_deposit_i64:
3353        return C_O1_I2(r, rZ, r);
3354
3355    case INDEX_op_movcond_i32:
3356        return C_O1_I4(r, r, ri, rI, r);
3357    case INDEX_op_movcond_i64:
3358        return C_O1_I4(r, r, rC, rI, r);
3359
3360    case INDEX_op_div2_i32:
3361    case INDEX_op_div2_i64:
3362    case INDEX_op_divu2_i32:
3363    case INDEX_op_divu2_i64:
3364        return C_O2_I3(o, m, 0, 1, r);
3365
3366    case INDEX_op_mulu2_i64:
3367        return C_O2_I2(o, m, 0, r);
3368    case INDEX_op_muls2_i64:
3369        return C_O2_I2(o, m, r, r);
3370
3371    case INDEX_op_add2_i32:
3372    case INDEX_op_sub2_i32:
3373        return C_N1_O1_I4(r, r, 0, 1, ri, r);
3374
3375    case INDEX_op_add2_i64:
3376    case INDEX_op_sub2_i64:
3377        return C_N1_O1_I4(r, r, 0, 1, rJU, r);
3378
3379    case INDEX_op_st_vec:
3380        return C_O0_I2(v, r);
3381    case INDEX_op_ld_vec:
3382    case INDEX_op_dupm_vec:
3383        return C_O1_I1(v, r);
3384    case INDEX_op_dup_vec:
3385        return C_O1_I1(v, vr);
3386    case INDEX_op_abs_vec:
3387    case INDEX_op_neg_vec:
3388    case INDEX_op_not_vec:
3389    case INDEX_op_rotli_vec:
3390    case INDEX_op_sari_vec:
3391    case INDEX_op_shli_vec:
3392    case INDEX_op_shri_vec:
3393    case INDEX_op_s390_vuph_vec:
3394    case INDEX_op_s390_vupl_vec:
3395        return C_O1_I1(v, v);
3396    case INDEX_op_add_vec:
3397    case INDEX_op_sub_vec:
3398    case INDEX_op_and_vec:
3399    case INDEX_op_andc_vec:
3400    case INDEX_op_or_vec:
3401    case INDEX_op_orc_vec:
3402    case INDEX_op_xor_vec:
3403    case INDEX_op_nand_vec:
3404    case INDEX_op_nor_vec:
3405    case INDEX_op_eqv_vec:
3406    case INDEX_op_cmp_vec:
3407    case INDEX_op_mul_vec:
3408    case INDEX_op_rotlv_vec:
3409    case INDEX_op_rotrv_vec:
3410    case INDEX_op_shlv_vec:
3411    case INDEX_op_shrv_vec:
3412    case INDEX_op_sarv_vec:
3413    case INDEX_op_smax_vec:
3414    case INDEX_op_smin_vec:
3415    case INDEX_op_umax_vec:
3416    case INDEX_op_umin_vec:
3417    case INDEX_op_s390_vpks_vec:
3418        return C_O1_I2(v, v, v);
3419    case INDEX_op_rotls_vec:
3420    case INDEX_op_shls_vec:
3421    case INDEX_op_shrs_vec:
3422    case INDEX_op_sars_vec:
3423        return C_O1_I2(v, v, r);
3424    case INDEX_op_bitsel_vec:
3425        return C_O1_I3(v, v, v, v);
3426    case INDEX_op_cmpsel_vec:
3427        return (TCG_TARGET_HAS_orc_vec
3428                ? C_O1_I4(v, v, v, vZM, v)
3429                : C_O1_I4(v, v, v, vZ, v));
3430
3431    default:
3432        g_assert_not_reached();
3433    }
3434}
3435
3436/*
3437 * Mainline glibc added HWCAP_S390_VX before it was kernel abi.
3438 * Some distros have fixed this up locally, others have not.
3439 */
3440#ifndef HWCAP_S390_VXRS
3441#define HWCAP_S390_VXRS 2048
3442#endif
3443
3444static void query_s390_facilities(void)
3445{
3446    unsigned long hwcap = qemu_getauxval(AT_HWCAP);
3447    const char *which;
3448
3449    /* Is STORE FACILITY LIST EXTENDED available?  Honestly, I believe this
3450       is present on all 64-bit systems, but let's check for it anyway.  */
3451    if (hwcap & HWCAP_S390_STFLE) {
3452        register int r0 __asm__("0") = ARRAY_SIZE(s390_facilities) - 1;
3453        register void *r1 __asm__("1") = s390_facilities;
3454
3455        /* stfle 0(%r1) */
3456        asm volatile(".word 0xb2b0,0x1000"
3457                     : "=r"(r0) : "r"(r0), "r"(r1) : "memory", "cc");
3458    }
3459
3460    /*
3461     * Use of vector registers requires os support beyond the facility bit.
3462     * If the kernel does not advertise support, disable the facility bits.
3463     * There is nothing else we currently care about in the 3rd word, so
3464     * disable VECTOR with one store.
3465     */
3466    if (!(hwcap & HWCAP_S390_VXRS)) {
3467        s390_facilities[2] = 0;
3468    }
3469
3470    /*
3471     * Minimum supported cpu revision is z196.
3472     * Check for all required facilities.
3473     * ZARCH_ACTIVE is done via preprocessor check for 64-bit.
3474     */
3475    if (!HAVE_FACILITY(LONG_DISP)) {
3476        which = "long-displacement";
3477        goto fail;
3478    }
3479    if (!HAVE_FACILITY(EXT_IMM)) {
3480        which = "extended-immediate";
3481        goto fail;
3482    }
3483    if (!HAVE_FACILITY(GEN_INST_EXT)) {
3484        which = "general-instructions-extension";
3485        goto fail;
3486    }
3487    /*
3488     * Facility 45 is a big bin that contains: distinct-operands,
3489     * fast-BCR-serialization, high-word, population-count,
3490     * interlocked-access-1, and load/store-on-condition-1
3491     */
3492    if (!HAVE_FACILITY(45)) {
3493        which = "45";
3494        goto fail;
3495    }
3496    return;
3497
3498 fail:
3499    error_report("%s: missing required facility %s", __func__, which);
3500    exit(EXIT_FAILURE);
3501}
3502
3503static void tcg_target_init(TCGContext *s)
3504{
3505    query_s390_facilities();
3506
3507    tcg_target_available_regs[TCG_TYPE_I32] = 0xffff;
3508    tcg_target_available_regs[TCG_TYPE_I64] = 0xffff;
3509    if (HAVE_FACILITY(VECTOR)) {
3510        tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull;
3511        tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull;
3512    }
3513
3514    tcg_target_call_clobber_regs = 0;
3515    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
3516    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1);
3517    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
3518    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
3519    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
3520    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
3521    /* The r6 register is technically call-saved, but it's also a parameter
3522       register, so it can get killed by setup for the qemu_st helper.  */
3523    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
3524    /* The return register can be considered call-clobbered.  */
3525    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14);
3526
3527    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0);
3528    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1);
3529    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V2);
3530    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V3);
3531    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V4);
3532    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V5);
3533    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V6);
3534    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V7);
3535    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V16);
3536    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V17);
3537    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V18);
3538    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V19);
3539    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V20);
3540    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V21);
3541    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V22);
3542    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V23);
3543    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V24);
3544    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V25);
3545    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V26);
3546    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V27);
3547    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V28);
3548    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V29);
3549    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V30);
3550    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V31);
3551
3552    s->reserved_regs = 0;
3553    tcg_regset_set_reg(s->reserved_regs, TCG_TMP0);
3554    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP0);
3555    /* XXX many insns can't be used with R0, so we better avoid it for now */
3556    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0);
3557    tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
3558}
3559
3560#define FRAME_SIZE  ((int)(TCG_TARGET_CALL_STACK_OFFSET          \
3561                           + TCG_STATIC_CALL_ARGS_SIZE           \
3562                           + CPU_TEMP_BUF_NLONGS * sizeof(long)))
3563
3564static void tcg_target_qemu_prologue(TCGContext *s)
3565{
3566    /* stmg %r6,%r15,48(%r15) (save registers) */
3567    tcg_out_insn(s, RXY, STMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15, 48);
3568
3569    /* aghi %r15,-frame_size */
3570    tcg_out_insn(s, RI, AGHI, TCG_REG_R15, -FRAME_SIZE);
3571
3572    tcg_set_frame(s, TCG_REG_CALL_STACK,
3573                  TCG_STATIC_CALL_ARGS_SIZE + TCG_TARGET_CALL_STACK_OFFSET,
3574                  CPU_TEMP_BUF_NLONGS * sizeof(long));
3575
3576    if (!tcg_use_softmmu && guest_base >= 0x80000) {
3577        tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base);
3578        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
3579    }
3580
3581    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
3582
3583    /* br %r3 (go to TB) */
3584    tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, tcg_target_call_iarg_regs[1]);
3585
3586    /*
3587     * Return path for goto_ptr. Set return value to 0, a-la exit_tb,
3588     * and fall through to the rest of the epilogue.
3589     */
3590    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
3591    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, 0);
3592
3593    /* TB epilogue */
3594    tb_ret_addr = tcg_splitwx_to_rx(s->code_ptr);
3595
3596    /* lmg %r6,%r15,fs+48(%r15) (restore registers) */
3597    tcg_out_insn(s, RXY, LMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15,
3598                 FRAME_SIZE + 48);
3599
3600    /* br %r14 (return) */
3601    tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_R14);
3602}
3603
3604static void tcg_out_tb_start(TCGContext *s)
3605{
3606    /* nothing to do */
3607}
3608
3609static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
3610{
3611    memset(p, 0x07, count * sizeof(tcg_insn_unit));
3612}
3613
3614typedef struct {
3615    DebugFrameHeader h;
3616    uint8_t fde_def_cfa[4];
3617    uint8_t fde_reg_ofs[18];
3618} DebugFrame;
3619
3620/* We're expecting a 2 byte uleb128 encoded value.  */
3621QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3622
3623#define ELF_HOST_MACHINE  EM_S390
3624
3625static const DebugFrame debug_frame = {
3626    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3627    .h.cie.id = -1,
3628    .h.cie.version = 1,
3629    .h.cie.code_align = 1,
3630    .h.cie.data_align = 8,                /* sleb128 8 */
3631    .h.cie.return_column = TCG_REG_R14,
3632
3633    /* Total FDE size does not include the "len" member.  */
3634    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
3635
3636    .fde_def_cfa = {
3637        12, TCG_REG_CALL_STACK,         /* DW_CFA_def_cfa %r15, ... */
3638        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3639        (FRAME_SIZE >> 7)
3640    },
3641    .fde_reg_ofs = {
3642        0x86, 6,                        /* DW_CFA_offset, %r6, 48 */
3643        0x87, 7,                        /* DW_CFA_offset, %r7, 56 */
3644        0x88, 8,                        /* DW_CFA_offset, %r8, 64 */
3645        0x89, 9,                        /* DW_CFA_offset, %r92, 72 */
3646        0x8a, 10,                       /* DW_CFA_offset, %r10, 80 */
3647        0x8b, 11,                       /* DW_CFA_offset, %r11, 88 */
3648        0x8c, 12,                       /* DW_CFA_offset, %r12, 96 */
3649        0x8d, 13,                       /* DW_CFA_offset, %r13, 104 */
3650        0x8e, 14,                       /* DW_CFA_offset, %r14, 112 */
3651    }
3652};
3653
3654void tcg_register_jit(const void *buf, size_t buf_size)
3655{
3656    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3657}
3658