xref: /openbmc/qemu/tcg/s390x/tcg-target.c.inc (revision 5a5bb0a5a0b879c8f110c6a9bde9146181ef840c)
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 "elf.h"
28
29/* Used for function call generation. */
30#define TCG_TARGET_STACK_ALIGN          8
31#define TCG_TARGET_CALL_STACK_OFFSET    160
32#define TCG_TARGET_CALL_ARG_I32         TCG_CALL_ARG_EXTEND
33#define TCG_TARGET_CALL_ARG_I64         TCG_CALL_ARG_NORMAL
34#define TCG_TARGET_CALL_ARG_I128        TCG_CALL_ARG_BY_REF
35#define TCG_TARGET_CALL_RET_I128        TCG_CALL_RET_BY_REF
36
37#define TCG_CT_CONST_S16        (1 << 8)
38#define TCG_CT_CONST_S32        (1 << 9)
39#define TCG_CT_CONST_U32        (1 << 10)
40#define TCG_CT_CONST_ZERO       (1 << 11)
41#define TCG_CT_CONST_P32        (1 << 12)
42#define TCG_CT_CONST_INV        (1 << 13)
43#define TCG_CT_CONST_INVRISBG   (1 << 14)
44#define TCG_CT_CONST_CMP        (1 << 15)
45#define TCG_CT_CONST_M1         (1 << 16)
46
47#define ALL_GENERAL_REGS     MAKE_64BIT_MASK(0, 16)
48#define ALL_VECTOR_REGS      MAKE_64BIT_MASK(32, 32)
49
50/* Several places within the instruction set 0 means "no register"
51   rather than TCG_REG_R0.  */
52#define TCG_REG_NONE    0
53
54/* A scratch register that may be be used throughout the backend.  */
55#define TCG_TMP0        TCG_REG_R1
56#define TCG_VEC_TMP0    TCG_REG_V31
57
58#define TCG_GUEST_BASE_REG TCG_REG_R13
59
60/* All of the following instructions are prefixed with their instruction
61   format, and are defined as 8- or 16-bit quantities, even when the two
62   halves of the 16-bit quantity may appear 32 bits apart in the insn.
63   This makes it easy to copy the values from the tables in Appendix B.  */
64typedef enum S390Opcode {
65    RIL_AFI     = 0xc209,
66    RIL_AGFI    = 0xc208,
67    RIL_ALFI    = 0xc20b,
68    RIL_ALGFI   = 0xc20a,
69    RIL_BRASL   = 0xc005,
70    RIL_BRCL    = 0xc004,
71    RIL_CFI     = 0xc20d,
72    RIL_CGFI    = 0xc20c,
73    RIL_CLFI    = 0xc20f,
74    RIL_CLGFI   = 0xc20e,
75    RIL_CLRL    = 0xc60f,
76    RIL_CLGRL   = 0xc60a,
77    RIL_CRL     = 0xc60d,
78    RIL_CGRL    = 0xc608,
79    RIL_IIHF    = 0xc008,
80    RIL_IILF    = 0xc009,
81    RIL_LARL    = 0xc000,
82    RIL_LGFI    = 0xc001,
83    RIL_LGRL    = 0xc408,
84    RIL_LLIHF   = 0xc00e,
85    RIL_LLILF   = 0xc00f,
86    RIL_LRL     = 0xc40d,
87    RIL_MSFI    = 0xc201,
88    RIL_MSGFI   = 0xc200,
89    RIL_NIHF    = 0xc00a,
90    RIL_NILF    = 0xc00b,
91    RIL_OIHF    = 0xc00c,
92    RIL_OILF    = 0xc00d,
93    RIL_SLFI    = 0xc205,
94    RIL_SLGFI   = 0xc204,
95    RIL_XIHF    = 0xc006,
96    RIL_XILF    = 0xc007,
97
98    RI_AGHI     = 0xa70b,
99    RI_AHI      = 0xa70a,
100    RI_BRC      = 0xa704,
101    RI_CHI      = 0xa70e,
102    RI_CGHI     = 0xa70f,
103    RI_IIHH     = 0xa500,
104    RI_IIHL     = 0xa501,
105    RI_IILH     = 0xa502,
106    RI_IILL     = 0xa503,
107    RI_LGHI     = 0xa709,
108    RI_LLIHH    = 0xa50c,
109    RI_LLIHL    = 0xa50d,
110    RI_LLILH    = 0xa50e,
111    RI_LLILL    = 0xa50f,
112    RI_MGHI     = 0xa70d,
113    RI_MHI      = 0xa70c,
114    RI_NIHH     = 0xa504,
115    RI_NIHL     = 0xa505,
116    RI_NILH     = 0xa506,
117    RI_NILL     = 0xa507,
118    RI_OIHH     = 0xa508,
119    RI_OIHL     = 0xa509,
120    RI_OILH     = 0xa50a,
121    RI_OILL     = 0xa50b,
122    RI_TMLL     = 0xa701,
123    RI_TMLH     = 0xa700,
124    RI_TMHL     = 0xa703,
125    RI_TMHH     = 0xa702,
126
127    RIEb_CGRJ    = 0xec64,
128    RIEb_CLGRJ   = 0xec65,
129    RIEb_CLRJ    = 0xec77,
130    RIEb_CRJ     = 0xec76,
131
132    RIEc_CGIJ    = 0xec7c,
133    RIEc_CIJ     = 0xec7e,
134    RIEc_CLGIJ   = 0xec7d,
135    RIEc_CLIJ    = 0xec7f,
136
137    RIEf_RISBG   = 0xec55,
138
139    RIEg_LOCGHI  = 0xec46,
140
141    RRE_AGR     = 0xb908,
142    RRE_ALGR    = 0xb90a,
143    RRE_ALCR    = 0xb998,
144    RRE_ALCGR   = 0xb988,
145    RRE_ALGFR   = 0xb91a,
146    RRE_CGR     = 0xb920,
147    RRE_CLGR    = 0xb921,
148    RRE_DLGR    = 0xb987,
149    RRE_DLR     = 0xb997,
150    RRE_DSGFR   = 0xb91d,
151    RRE_DSGR    = 0xb90d,
152    RRE_FLOGR   = 0xb983,
153    RRE_LGBR    = 0xb906,
154    RRE_LCGR    = 0xb903,
155    RRE_LGFR    = 0xb914,
156    RRE_LGHR    = 0xb907,
157    RRE_LGR     = 0xb904,
158    RRE_LLGCR   = 0xb984,
159    RRE_LLGFR   = 0xb916,
160    RRE_LLGHR   = 0xb985,
161    RRE_LRVR    = 0xb91f,
162    RRE_LRVGR   = 0xb90f,
163    RRE_LTGR    = 0xb902,
164    RRE_MLGR    = 0xb986,
165    RRE_MSGR    = 0xb90c,
166    RRE_MSR     = 0xb252,
167    RRE_NGR     = 0xb980,
168    RRE_OGR     = 0xb981,
169    RRE_SGR     = 0xb909,
170    RRE_SLGR    = 0xb90b,
171    RRE_SLBR    = 0xb999,
172    RRE_SLBGR   = 0xb989,
173    RRE_XGR     = 0xb982,
174
175    RRFa_MGRK   = 0xb9ec,
176    RRFa_MSRKC  = 0xb9fd,
177    RRFa_MSGRKC = 0xb9ed,
178    RRFa_NCRK   = 0xb9f5,
179    RRFa_NCGRK  = 0xb9e5,
180    RRFa_NNRK   = 0xb974,
181    RRFa_NNGRK  = 0xb964,
182    RRFa_NORK   = 0xb976,
183    RRFa_NOGRK  = 0xb966,
184    RRFa_NRK    = 0xb9f4,
185    RRFa_NGRK   = 0xb9e4,
186    RRFa_NXRK   = 0xb977,
187    RRFa_NXGRK  = 0xb967,
188    RRFa_OCRK   = 0xb975,
189    RRFa_OCGRK  = 0xb965,
190    RRFa_ORK    = 0xb9f6,
191    RRFa_OGRK   = 0xb9e6,
192    RRFa_SRK    = 0xb9f9,
193    RRFa_SGRK   = 0xb9e9,
194    RRFa_SLRK   = 0xb9fb,
195    RRFa_SLGRK  = 0xb9eb,
196    RRFa_XRK    = 0xb9f7,
197    RRFa_XGRK   = 0xb9e7,
198
199    RRFam_SELGR = 0xb9e3,
200
201    RRFc_LOCR   = 0xb9f2,
202    RRFc_LOCGR  = 0xb9e2,
203    RRFc_POPCNT = 0xb9e1,
204
205    RR_AR       = 0x1a,
206    RR_ALR      = 0x1e,
207    RR_BASR     = 0x0d,
208    RR_BCR      = 0x07,
209    RR_CLR      = 0x15,
210    RR_CR       = 0x19,
211    RR_DR       = 0x1d,
212    RR_LCR      = 0x13,
213    RR_LR       = 0x18,
214    RR_LTR      = 0x12,
215    RR_NR       = 0x14,
216    RR_OR       = 0x16,
217    RR_SR       = 0x1b,
218    RR_SLR      = 0x1f,
219    RR_XR       = 0x17,
220
221    RSY_RLL     = 0xeb1d,
222    RSY_RLLG    = 0xeb1c,
223    RSY_SLLG    = 0xeb0d,
224    RSY_SLLK    = 0xebdf,
225    RSY_SRAG    = 0xeb0a,
226    RSY_SRAK    = 0xebdc,
227    RSY_SRLG    = 0xeb0c,
228    RSY_SRLK    = 0xebde,
229
230    RS_SLL      = 0x89,
231    RS_SRA      = 0x8a,
232    RS_SRL      = 0x88,
233
234    RXY_AG      = 0xe308,
235    RXY_AY      = 0xe35a,
236    RXY_CG      = 0xe320,
237    RXY_CLG     = 0xe321,
238    RXY_CLY     = 0xe355,
239    RXY_CY      = 0xe359,
240    RXY_LAY     = 0xe371,
241    RXY_LB      = 0xe376,
242    RXY_LG      = 0xe304,
243    RXY_LGB     = 0xe377,
244    RXY_LGF     = 0xe314,
245    RXY_LGH     = 0xe315,
246    RXY_LHY     = 0xe378,
247    RXY_LLGC    = 0xe390,
248    RXY_LLGF    = 0xe316,
249    RXY_LLGH    = 0xe391,
250    RXY_LMG     = 0xeb04,
251    RXY_LPQ     = 0xe38f,
252    RXY_LRV     = 0xe31e,
253    RXY_LRVG    = 0xe30f,
254    RXY_LRVH    = 0xe31f,
255    RXY_LY      = 0xe358,
256    RXY_NG      = 0xe380,
257    RXY_OG      = 0xe381,
258    RXY_STCY    = 0xe372,
259    RXY_STG     = 0xe324,
260    RXY_STHY    = 0xe370,
261    RXY_STMG    = 0xeb24,
262    RXY_STPQ    = 0xe38e,
263    RXY_STRV    = 0xe33e,
264    RXY_STRVG   = 0xe32f,
265    RXY_STRVH   = 0xe33f,
266    RXY_STY     = 0xe350,
267    RXY_XG      = 0xe382,
268
269    RX_A        = 0x5a,
270    RX_C        = 0x59,
271    RX_L        = 0x58,
272    RX_LA       = 0x41,
273    RX_LH       = 0x48,
274    RX_ST       = 0x50,
275    RX_STC      = 0x42,
276    RX_STH      = 0x40,
277
278    VRIa_VGBM   = 0xe744,
279    VRIa_VREPI  = 0xe745,
280    VRIb_VGM    = 0xe746,
281    VRIc_VREP   = 0xe74d,
282
283    VRRa_VLC    = 0xe7de,
284    VRRa_VLP    = 0xe7df,
285    VRRa_VLR    = 0xe756,
286    VRRc_VA     = 0xe7f3,
287    VRRc_VCEQ   = 0xe7f8,   /* we leave the m5 cs field 0 */
288    VRRc_VCH    = 0xe7fb,   /* " */
289    VRRc_VCHL   = 0xe7f9,   /* " */
290    VRRc_VERLLV = 0xe773,
291    VRRc_VESLV  = 0xe770,
292    VRRc_VESRAV = 0xe77a,
293    VRRc_VESRLV = 0xe778,
294    VRRc_VML    = 0xe7a2,
295    VRRc_VMN    = 0xe7fe,
296    VRRc_VMNL   = 0xe7fc,
297    VRRc_VMX    = 0xe7ff,
298    VRRc_VMXL   = 0xe7fd,
299    VRRc_VN     = 0xe768,
300    VRRc_VNC    = 0xe769,
301    VRRc_VNN    = 0xe76e,
302    VRRc_VNO    = 0xe76b,
303    VRRc_VNX    = 0xe76c,
304    VRRc_VO     = 0xe76a,
305    VRRc_VOC    = 0xe76f,
306    VRRc_VPKS   = 0xe797,   /* we leave the m5 cs field 0 */
307    VRRc_VS     = 0xe7f7,
308    VRRa_VUPH   = 0xe7d7,
309    VRRa_VUPL   = 0xe7d6,
310    VRRc_VX     = 0xe76d,
311    VRRe_VSEL   = 0xe78d,
312    VRRf_VLVGP  = 0xe762,
313
314    VRSa_VERLL  = 0xe733,
315    VRSa_VESL   = 0xe730,
316    VRSa_VESRA  = 0xe73a,
317    VRSa_VESRL  = 0xe738,
318    VRSb_VLVG   = 0xe722,
319    VRSc_VLGV   = 0xe721,
320
321    VRX_VL      = 0xe706,
322    VRX_VLLEZ   = 0xe704,
323    VRX_VLREP   = 0xe705,
324    VRX_VST     = 0xe70e,
325    VRX_VSTEF   = 0xe70b,
326    VRX_VSTEG   = 0xe70a,
327
328    NOP         = 0x0707,
329} S390Opcode;
330
331#ifdef CONFIG_DEBUG_TCG
332static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
333    "%r0",  "%r1",  "%r2",  "%r3",  "%r4",  "%r5",  "%r6",  "%r7",
334    "%r8",  "%r9",  "%r10", "%r11", "%r12", "%r13", "%r14", "%r15",
335    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
336    "%v0",  "%v1",  "%v2",  "%v3",  "%v4",  "%v5",  "%v6",  "%v7",
337    "%v8",  "%v9",  "%v10", "%v11", "%v12", "%v13", "%v14", "%v15",
338    "%v16", "%v17", "%v18", "%v19", "%v20", "%v21", "%v22", "%v23",
339    "%v24", "%v25", "%v26", "%v27", "%v28", "%v29", "%v30", "%v31",
340};
341#endif
342
343/* Since R6 is a potential argument register, choose it last of the
344   call-saved registers.  Likewise prefer the call-clobbered registers
345   in reverse order to maximize the chance of avoiding the arguments.  */
346static const int tcg_target_reg_alloc_order[] = {
347    /* Call saved registers.  */
348    TCG_REG_R13,
349    TCG_REG_R12,
350    TCG_REG_R11,
351    TCG_REG_R10,
352    TCG_REG_R9,
353    TCG_REG_R8,
354    TCG_REG_R7,
355    TCG_REG_R6,
356    /* Call clobbered registers.  */
357    TCG_REG_R14,
358    TCG_REG_R0,
359    TCG_REG_R1,
360    /* Argument registers, in reverse order of allocation.  */
361    TCG_REG_R5,
362    TCG_REG_R4,
363    TCG_REG_R3,
364    TCG_REG_R2,
365
366    /* V8-V15 are call saved, and omitted. */
367    TCG_REG_V0,
368    TCG_REG_V1,
369    TCG_REG_V2,
370    TCG_REG_V3,
371    TCG_REG_V4,
372    TCG_REG_V5,
373    TCG_REG_V6,
374    TCG_REG_V7,
375    TCG_REG_V16,
376    TCG_REG_V17,
377    TCG_REG_V18,
378    TCG_REG_V19,
379    TCG_REG_V20,
380    TCG_REG_V21,
381    TCG_REG_V22,
382    TCG_REG_V23,
383    TCG_REG_V24,
384    TCG_REG_V25,
385    TCG_REG_V26,
386    TCG_REG_V27,
387    TCG_REG_V28,
388    TCG_REG_V29,
389    TCG_REG_V30,
390    TCG_REG_V31,
391};
392
393static const int tcg_target_call_iarg_regs[] = {
394    TCG_REG_R2,
395    TCG_REG_R3,
396    TCG_REG_R4,
397    TCG_REG_R5,
398    TCG_REG_R6,
399};
400
401static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
402{
403    tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
404    tcg_debug_assert(slot == 0);
405    return TCG_REG_R2;
406}
407
408#define S390_CC_EQ      8
409#define S390_CC_LT      4
410#define S390_CC_GT      2
411#define S390_CC_OV      1
412#define S390_CC_NE      (S390_CC_LT | S390_CC_GT)
413#define S390_CC_LE      (S390_CC_LT | S390_CC_EQ)
414#define S390_CC_GE      (S390_CC_GT | S390_CC_EQ)
415#define S390_CC_NEVER   0
416#define S390_CC_ALWAYS  15
417
418#define S390_TM_EQ      8  /* CC == 0 */
419#define S390_TM_NE      7  /* CC in {1,2,3} */
420
421/* Condition codes that result from a COMPARE and COMPARE LOGICAL.  */
422static const uint8_t tcg_cond_to_s390_cond[16] = {
423    [TCG_COND_EQ]  = S390_CC_EQ,
424    [TCG_COND_NE]  = S390_CC_NE,
425    [TCG_COND_TSTEQ] = S390_CC_EQ,
426    [TCG_COND_TSTNE] = S390_CC_NE,
427    [TCG_COND_LT]  = S390_CC_LT,
428    [TCG_COND_LE]  = S390_CC_LE,
429    [TCG_COND_GT]  = S390_CC_GT,
430    [TCG_COND_GE]  = S390_CC_GE,
431    [TCG_COND_LTU] = S390_CC_LT,
432    [TCG_COND_LEU] = S390_CC_LE,
433    [TCG_COND_GTU] = S390_CC_GT,
434    [TCG_COND_GEU] = S390_CC_GE,
435};
436
437/* Condition codes that result from a LOAD AND TEST.  Here, we have no
438   unsigned instruction variation, however since the test is vs zero we
439   can re-map the outcomes appropriately.  */
440static const uint8_t tcg_cond_to_ltr_cond[16] = {
441    [TCG_COND_EQ]  = S390_CC_EQ,
442    [TCG_COND_NE]  = S390_CC_NE,
443    [TCG_COND_TSTEQ] = S390_CC_ALWAYS,
444    [TCG_COND_TSTNE] = S390_CC_NEVER,
445    [TCG_COND_LT]  = S390_CC_LT,
446    [TCG_COND_LE]  = S390_CC_LE,
447    [TCG_COND_GT]  = S390_CC_GT,
448    [TCG_COND_GE]  = S390_CC_GE,
449    [TCG_COND_LTU] = S390_CC_NEVER,
450    [TCG_COND_LEU] = S390_CC_EQ,
451    [TCG_COND_GTU] = S390_CC_NE,
452    [TCG_COND_GEU] = S390_CC_ALWAYS,
453};
454
455static const tcg_insn_unit *tb_ret_addr;
456uint64_t s390_facilities[3];
457
458static inline bool is_general_reg(TCGReg r)
459{
460    return r <= TCG_REG_R15;
461}
462
463static inline bool is_vector_reg(TCGReg r)
464{
465    return r >= TCG_REG_V0 && r <= TCG_REG_V31;
466}
467
468static bool patch_reloc(tcg_insn_unit *src_rw, int type,
469                        intptr_t value, intptr_t addend)
470{
471    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
472    intptr_t pcrel2;
473    uint32_t old;
474
475    value += addend;
476    pcrel2 = (tcg_insn_unit *)value - src_rx;
477
478    switch (type) {
479    case R_390_PC16DBL:
480        if (pcrel2 == (int16_t)pcrel2) {
481            tcg_patch16(src_rw, pcrel2);
482            return true;
483        }
484        break;
485    case R_390_PC32DBL:
486        if (pcrel2 == (int32_t)pcrel2) {
487            tcg_patch32(src_rw, pcrel2);
488            return true;
489        }
490        break;
491    case R_390_20:
492        if (value == sextract64(value, 0, 20)) {
493            old = *(uint32_t *)src_rw & 0xf00000ff;
494            old |= ((value & 0xfff) << 16) | ((value & 0xff000) >> 4);
495            tcg_patch32(src_rw, old);
496            return true;
497        }
498        break;
499    default:
500        g_assert_not_reached();
501    }
502    return false;
503}
504
505static int is_const_p16(uint64_t val)
506{
507    for (int i = 0; i < 4; ++i) {
508        uint64_t mask = 0xffffull << (i * 16);
509        if ((val & ~mask) == 0) {
510            return i;
511        }
512    }
513    return -1;
514}
515
516static int is_const_p32(uint64_t val)
517{
518    if ((val & 0xffffffff00000000ull) == 0) {
519        return 0;
520    }
521    if ((val & 0x00000000ffffffffull) == 0) {
522        return 1;
523    }
524    return -1;
525}
526
527/*
528 * Accept bit patterns like these:
529 *  0....01....1
530 *  1....10....0
531 *  1..10..01..1
532 *  0..01..10..0
533 * Copied from gcc sources.
534 */
535static bool risbg_mask(uint64_t c)
536{
537    uint64_t lsb;
538    /* We don't change the number of transitions by inverting,
539       so make sure we start with the LSB zero.  */
540    if (c & 1) {
541        c = ~c;
542    }
543    /* Reject all zeros or all ones.  */
544    if (c == 0) {
545        return false;
546    }
547    /* Find the first transition.  */
548    lsb = c & -c;
549    /* Invert to look for a second transition.  */
550    c = ~c;
551    /* Erase the first transition.  */
552    c &= -lsb;
553    /* Find the second transition, if any.  */
554    lsb = c & -c;
555    /* Match if all the bits are 1's, or if c is zero.  */
556    return c == -lsb;
557}
558
559/* Test if a constant matches the constraint. */
560static bool tcg_target_const_match(int64_t val, int ct,
561                                   TCGType type, TCGCond cond, int vece)
562{
563    uint64_t uval = val;
564
565    if (ct & TCG_CT_CONST) {
566        return true;
567    }
568    if (type == TCG_TYPE_I32) {
569        uval = (uint32_t)val;
570        val = (int32_t)val;
571    }
572
573    if (ct & TCG_CT_CONST_CMP) {
574        if (is_tst_cond(cond)) {
575            if (is_const_p16(uval) >= 0) {
576                return true;  /* TMxx */
577            }
578            if (risbg_mask(uval)) {
579                return true;  /* RISBG */
580            }
581            return false;
582        }
583
584        if (type == TCG_TYPE_I32) {
585            return true;
586        }
587
588        switch (cond) {
589        case TCG_COND_EQ:
590        case TCG_COND_NE:
591            ct |= TCG_CT_CONST_S32 | TCG_CT_CONST_U32;  /* CGFI or CLGFI */
592            break;
593        case TCG_COND_LT:
594        case TCG_COND_GE:
595        case TCG_COND_LE:
596        case TCG_COND_GT:
597            ct |= TCG_CT_CONST_S32;  /* CGFI */
598            break;
599        case TCG_COND_LTU:
600        case TCG_COND_GEU:
601        case TCG_COND_LEU:
602        case TCG_COND_GTU:
603            ct |= TCG_CT_CONST_U32;  /* CLGFI */
604            break;
605        case TCG_COND_TSTNE:
606        case TCG_COND_TSTEQ:
607            /* checked above, fallthru */
608        default:
609            g_assert_not_reached();
610        }
611    }
612
613    if ((ct & TCG_CT_CONST_S32) && val == (int32_t)val) {
614        return true;
615    }
616    if ((ct & TCG_CT_CONST_U32) && val == (uint32_t)val) {
617        return true;
618    }
619    if ((ct & TCG_CT_CONST_S16) && val == (int16_t)val) {
620        return true;
621    }
622    if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
623        return true;
624    }
625    if ((ct & TCG_CT_CONST_M1) && val == -1) {
626        return true;
627    }
628
629    if (ct & TCG_CT_CONST_INV) {
630        val = ~val;
631    }
632    if ((ct & TCG_CT_CONST_P32) && is_const_p32(val) >= 0) {
633        return true;
634    }
635    if ((ct & TCG_CT_CONST_INVRISBG) && risbg_mask(~val)) {
636        return true;
637    }
638    return false;
639}
640
641/* Emit instructions according to the given instruction format.  */
642
643static void tcg_out_insn_RR(TCGContext *s, S390Opcode op, TCGReg r1, TCGReg r2)
644{
645    tcg_out16(s, (op << 8) | (r1 << 4) | r2);
646}
647
648static void tcg_out_insn_RRE(TCGContext *s, S390Opcode op,
649                             TCGReg r1, TCGReg r2)
650{
651    tcg_out32(s, (op << 16) | (r1 << 4) | r2);
652}
653
654/* RRF-a without the m4 field */
655static void tcg_out_insn_RRFa(TCGContext *s, S390Opcode op,
656                              TCGReg r1, TCGReg r2, TCGReg r3)
657{
658    tcg_out32(s, (op << 16) | (r3 << 12) | (r1 << 4) | r2);
659}
660
661/* RRF-a with the m4 field */
662static void tcg_out_insn_RRFam(TCGContext *s, S390Opcode op,
663                               TCGReg r1, TCGReg r2, TCGReg r3, int m4)
664{
665    tcg_out32(s, (op << 16) | (r3 << 12) | (m4 << 8) | (r1 << 4) | r2);
666}
667
668static void tcg_out_insn_RRFc(TCGContext *s, S390Opcode op,
669                              TCGReg r1, TCGReg r2, int m3)
670{
671    tcg_out32(s, (op << 16) | (m3 << 12) | (r1 << 4) | r2);
672}
673
674static void tcg_out_insn_RI(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
675{
676    tcg_out32(s, (op << 16) | (r1 << 20) | (i2 & 0xffff));
677}
678
679static void tcg_out_insn_RIEg(TCGContext *s, S390Opcode op, TCGReg r1,
680                             int i2, int m3)
681{
682    tcg_out16(s, (op & 0xff00) | (r1 << 4) | m3);
683    tcg_out32(s, (i2 << 16) | (op & 0xff));
684}
685
686static void tcg_out_insn_RIL(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
687{
688    tcg_out16(s, op | (r1 << 4));
689    tcg_out32(s, i2);
690}
691
692static void tcg_out_insn_RS(TCGContext *s, S390Opcode op, TCGReg r1,
693                            TCGReg b2, TCGReg r3, int disp)
694{
695    tcg_out32(s, (op << 24) | (r1 << 20) | (r3 << 16) | (b2 << 12)
696              | (disp & 0xfff));
697}
698
699static void tcg_out_insn_RSY(TCGContext *s, S390Opcode op, TCGReg r1,
700                             TCGReg b2, TCGReg r3, int disp)
701{
702    tcg_out16(s, (op & 0xff00) | (r1 << 4) | r3);
703    tcg_out32(s, (op & 0xff) | (b2 << 28)
704              | ((disp & 0xfff) << 16) | ((disp & 0xff000) >> 4));
705}
706
707#define tcg_out_insn_RX   tcg_out_insn_RS
708#define tcg_out_insn_RXY  tcg_out_insn_RSY
709
710static int RXB(TCGReg v1, TCGReg v2, TCGReg v3, TCGReg v4)
711{
712    /*
713     * Shift bit 4 of each regno to its corresponding bit of RXB.
714     * RXB itself begins at bit 8 of the instruction so 8 - 4 = 4
715     * is the left-shift of the 4th operand.
716     */
717    return ((v1 & 0x10) << (4 + 3))
718         | ((v2 & 0x10) << (4 + 2))
719         | ((v3 & 0x10) << (4 + 1))
720         | ((v4 & 0x10) << (4 + 0));
721}
722
723static void tcg_out_insn_VRIa(TCGContext *s, S390Opcode op,
724                              TCGReg v1, uint16_t i2, int m3)
725{
726    tcg_debug_assert(is_vector_reg(v1));
727    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4));
728    tcg_out16(s, i2);
729    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m3 << 12));
730}
731
732static void tcg_out_insn_VRIb(TCGContext *s, S390Opcode op,
733                              TCGReg v1, uint8_t i2, uint8_t i3, int m4)
734{
735    tcg_debug_assert(is_vector_reg(v1));
736    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4));
737    tcg_out16(s, (i2 << 8) | (i3 & 0xff));
738    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m4 << 12));
739}
740
741static void tcg_out_insn_VRIc(TCGContext *s, S390Opcode op,
742                              TCGReg v1, uint16_t i2, TCGReg v3, int m4)
743{
744    tcg_debug_assert(is_vector_reg(v1));
745    tcg_debug_assert(is_vector_reg(v3));
746    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v3 & 0xf));
747    tcg_out16(s, i2);
748    tcg_out16(s, (op & 0x00ff) | RXB(v1, v3, 0, 0) | (m4 << 12));
749}
750
751static void tcg_out_insn_VRRa(TCGContext *s, S390Opcode op,
752                              TCGReg v1, TCGReg v2, int m3)
753{
754    tcg_debug_assert(is_vector_reg(v1));
755    tcg_debug_assert(is_vector_reg(v2));
756    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
757    tcg_out32(s, (op & 0x00ff) | RXB(v1, v2, 0, 0) | (m3 << 12));
758}
759
760static void tcg_out_insn_VRRc(TCGContext *s, S390Opcode op,
761                              TCGReg v1, TCGReg v2, TCGReg v3, int m4)
762{
763    tcg_debug_assert(is_vector_reg(v1));
764    tcg_debug_assert(is_vector_reg(v2));
765    tcg_debug_assert(is_vector_reg(v3));
766    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
767    tcg_out16(s, v3 << 12);
768    tcg_out16(s, (op & 0x00ff) | RXB(v1, v2, v3, 0) | (m4 << 12));
769}
770
771static void tcg_out_insn_VRRe(TCGContext *s, S390Opcode op,
772                              TCGReg v1, TCGReg v2, TCGReg v3, TCGReg v4)
773{
774    tcg_debug_assert(is_vector_reg(v1));
775    tcg_debug_assert(is_vector_reg(v2));
776    tcg_debug_assert(is_vector_reg(v3));
777    tcg_debug_assert(is_vector_reg(v4));
778    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
779    tcg_out16(s, v3 << 12);
780    tcg_out16(s, (op & 0x00ff) | RXB(v1, v2, v3, v4) | (v4 << 12));
781}
782
783static void tcg_out_insn_VRRf(TCGContext *s, S390Opcode op,
784                              TCGReg v1, TCGReg r2, TCGReg r3)
785{
786    tcg_debug_assert(is_vector_reg(v1));
787    tcg_debug_assert(is_general_reg(r2));
788    tcg_debug_assert(is_general_reg(r3));
789    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | r2);
790    tcg_out16(s, r3 << 12);
791    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0));
792}
793
794static void tcg_out_insn_VRSa(TCGContext *s, S390Opcode op, TCGReg v1,
795                              intptr_t d2, TCGReg b2, TCGReg v3, int m4)
796{
797    tcg_debug_assert(is_vector_reg(v1));
798    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
799    tcg_debug_assert(is_general_reg(b2));
800    tcg_debug_assert(is_vector_reg(v3));
801    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v3 & 0xf));
802    tcg_out16(s, b2 << 12 | d2);
803    tcg_out16(s, (op & 0x00ff) | RXB(v1, v3, 0, 0) | (m4 << 12));
804}
805
806static void tcg_out_insn_VRSb(TCGContext *s, S390Opcode op, TCGReg v1,
807                              intptr_t d2, TCGReg b2, TCGReg r3, int m4)
808{
809    tcg_debug_assert(is_vector_reg(v1));
810    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
811    tcg_debug_assert(is_general_reg(b2));
812    tcg_debug_assert(is_general_reg(r3));
813    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | r3);
814    tcg_out16(s, b2 << 12 | d2);
815    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m4 << 12));
816}
817
818static void tcg_out_insn_VRSc(TCGContext *s, S390Opcode op, TCGReg r1,
819                              intptr_t d2, TCGReg b2, TCGReg v3, int m4)
820{
821    tcg_debug_assert(is_general_reg(r1));
822    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
823    tcg_debug_assert(is_general_reg(b2));
824    tcg_debug_assert(is_vector_reg(v3));
825    tcg_out16(s, (op & 0xff00) | (r1 << 4) | (v3 & 0xf));
826    tcg_out16(s, b2 << 12 | d2);
827    tcg_out16(s, (op & 0x00ff) | RXB(0, v3, 0, 0) | (m4 << 12));
828}
829
830static void tcg_out_insn_VRX(TCGContext *s, S390Opcode op, TCGReg v1,
831                             TCGReg b2, TCGReg x2, intptr_t d2, int m3)
832{
833    tcg_debug_assert(is_vector_reg(v1));
834    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
835    tcg_debug_assert(is_general_reg(x2));
836    tcg_debug_assert(is_general_reg(b2));
837    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | x2);
838    tcg_out16(s, (b2 << 12) | d2);
839    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m3 << 12));
840}
841
842/* Emit an opcode with "type-checking" of the format.  */
843#define tcg_out_insn(S, FMT, OP, ...) \
844    glue(tcg_out_insn_,FMT)(S, glue(glue(FMT,_),OP), ## __VA_ARGS__)
845
846
847/* emit 64-bit shifts */
848static void tcg_out_sh64(TCGContext* s, S390Opcode op, TCGReg dest,
849                         TCGReg src, TCGReg sh_reg, int sh_imm)
850{
851    tcg_out_insn_RSY(s, op, dest, sh_reg, src, sh_imm);
852}
853
854/* emit 32-bit shifts */
855static void tcg_out_sh32(TCGContext* s, S390Opcode op, TCGReg dest,
856                         TCGReg sh_reg, int sh_imm)
857{
858    tcg_out_insn_RS(s, op, dest, sh_reg, 0, sh_imm);
859}
860
861static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg dst, TCGReg src)
862{
863    if (src == dst) {
864        return true;
865    }
866    switch (type) {
867    case TCG_TYPE_I32:
868        if (likely(is_general_reg(dst) && is_general_reg(src))) {
869            tcg_out_insn(s, RR, LR, dst, src);
870            break;
871        }
872        /* fallthru */
873
874    case TCG_TYPE_I64:
875        if (likely(is_general_reg(dst))) {
876            if (likely(is_general_reg(src))) {
877                tcg_out_insn(s, RRE, LGR, dst, src);
878            } else {
879                tcg_out_insn(s, VRSc, VLGV, dst, 0, 0, src, 3);
880            }
881            break;
882        } else if (is_general_reg(src)) {
883            tcg_out_insn(s, VRSb, VLVG, dst, 0, 0, src, 3);
884            break;
885        }
886        /* fallthru */
887
888    case TCG_TYPE_V64:
889    case TCG_TYPE_V128:
890        tcg_out_insn(s, VRRa, VLR, dst, src, 0);
891        break;
892
893    default:
894        g_assert_not_reached();
895    }
896    return true;
897}
898
899static const S390Opcode li_insns[4] = {
900    RI_LLILL, RI_LLILH, RI_LLIHL, RI_LLIHH
901};
902static const S390Opcode oi_insns[4] = {
903    RI_OILL, RI_OILH, RI_OIHL, RI_OIHH
904};
905static const S390Opcode lif_insns[2] = {
906    RIL_LLILF, RIL_LLIHF,
907};
908static const S390Opcode tm_insns[4] = {
909    RI_TMLL, RI_TMLH, RI_TMHL, RI_TMHH
910};
911
912/* load a register with an immediate value */
913static void tcg_out_movi(TCGContext *s, TCGType type,
914                         TCGReg ret, tcg_target_long sval)
915{
916    tcg_target_ulong uval = sval;
917    ptrdiff_t pc_off;
918    int i;
919
920    if (type == TCG_TYPE_I32) {
921        uval = (uint32_t)sval;
922        sval = (int32_t)sval;
923    }
924
925    /* Try all 32-bit insns that can load it in one go.  */
926    if (sval >= -0x8000 && sval < 0x8000) {
927        tcg_out_insn(s, RI, LGHI, ret, sval);
928        return;
929    }
930
931    i = is_const_p16(uval);
932    if (i >= 0) {
933        tcg_out_insn_RI(s, li_insns[i], ret, uval >> (i * 16));
934        return;
935    }
936
937    /* Try all 48-bit insns that can load it in one go.  */
938    if (sval == (int32_t)sval) {
939        tcg_out_insn(s, RIL, LGFI, ret, sval);
940        return;
941    }
942
943    i = is_const_p32(uval);
944    if (i >= 0) {
945        tcg_out_insn_RIL(s, lif_insns[i], ret, uval >> (i * 32));
946        return;
947    }
948
949    /* Try for PC-relative address load.  For odd addresses, add one. */
950    pc_off = tcg_pcrel_diff(s, (void *)sval) >> 1;
951    if (pc_off == (int32_t)pc_off) {
952        tcg_out_insn(s, RIL, LARL, ret, pc_off);
953        if (sval & 1) {
954            tcg_out_insn(s, RI, AGHI, ret, 1);
955        }
956        return;
957    }
958
959    /* Otherwise, load it by parts. */
960    i = is_const_p16((uint32_t)uval);
961    if (i >= 0) {
962        tcg_out_insn_RI(s, li_insns[i], ret, uval >> (i * 16));
963    } else {
964        tcg_out_insn(s, RIL, LLILF, ret, uval);
965    }
966    uval >>= 32;
967    i = is_const_p16(uval);
968    if (i >= 0) {
969        tcg_out_insn_RI(s, oi_insns[i + 2], ret, uval >> (i * 16));
970    } else {
971        tcg_out_insn(s, RIL, OIHF, ret, uval);
972    }
973}
974
975/* Emit a load/store type instruction.  Inputs are:
976   DATA:     The register to be loaded or stored.
977   BASE+OFS: The effective address.
978   OPC_RX:   If the operation has an RX format opcode (e.g. STC), otherwise 0.
979   OPC_RXY:  The RXY format opcode for the operation (e.g. STCY).  */
980
981static void tcg_out_mem(TCGContext *s, S390Opcode opc_rx, S390Opcode opc_rxy,
982                        TCGReg data, TCGReg base, TCGReg index,
983                        tcg_target_long ofs)
984{
985    if (ofs < -0x80000 || ofs >= 0x80000) {
986        /* Combine the low 20 bits of the offset with the actual load insn;
987           the high 44 bits must come from an immediate load.  */
988        tcg_target_long low = ((ofs & 0xfffff) ^ 0x80000) - 0x80000;
989        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs - low);
990        ofs = low;
991
992        /* If we were already given an index register, add it in.  */
993        if (index != TCG_REG_NONE) {
994            tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
995        }
996        index = TCG_TMP0;
997    }
998
999    if (opc_rx && ofs >= 0 && ofs < 0x1000) {
1000        tcg_out_insn_RX(s, opc_rx, data, base, index, ofs);
1001    } else {
1002        tcg_out_insn_RXY(s, opc_rxy, data, base, index, ofs);
1003    }
1004}
1005
1006static void tcg_out_vrx_mem(TCGContext *s, S390Opcode opc_vrx,
1007                            TCGReg data, TCGReg base, TCGReg index,
1008                            tcg_target_long ofs, int m3)
1009{
1010    if (ofs < 0 || ofs >= 0x1000) {
1011        if (ofs >= -0x80000 && ofs < 0x80000) {
1012            tcg_out_insn(s, RXY, LAY, TCG_TMP0, base, index, ofs);
1013            base = TCG_TMP0;
1014            index = TCG_REG_NONE;
1015            ofs = 0;
1016        } else {
1017            tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs);
1018            if (index != TCG_REG_NONE) {
1019                tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
1020            }
1021            index = TCG_TMP0;
1022            ofs = 0;
1023        }
1024    }
1025    tcg_out_insn_VRX(s, opc_vrx, data, base, index, ofs, m3);
1026}
1027
1028/* load data without address translation or endianness conversion */
1029static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg data,
1030                       TCGReg base, intptr_t ofs)
1031{
1032    switch (type) {
1033    case TCG_TYPE_I32:
1034        if (likely(is_general_reg(data))) {
1035            tcg_out_mem(s, RX_L, RXY_LY, data, base, TCG_REG_NONE, ofs);
1036            break;
1037        }
1038        tcg_out_vrx_mem(s, VRX_VLLEZ, data, base, TCG_REG_NONE, ofs, MO_32);
1039        break;
1040
1041    case TCG_TYPE_I64:
1042        if (likely(is_general_reg(data))) {
1043            tcg_out_mem(s, 0, RXY_LG, data, base, TCG_REG_NONE, ofs);
1044            break;
1045        }
1046        /* fallthru */
1047
1048    case TCG_TYPE_V64:
1049        tcg_out_vrx_mem(s, VRX_VLLEZ, data, base, TCG_REG_NONE, ofs, MO_64);
1050        break;
1051
1052    case TCG_TYPE_V128:
1053        /* Hint quadword aligned.  */
1054        tcg_out_vrx_mem(s, VRX_VL, data, base, TCG_REG_NONE, ofs, 4);
1055        break;
1056
1057    default:
1058        g_assert_not_reached();
1059    }
1060}
1061
1062static void tcg_out_st(TCGContext *s, TCGType type, TCGReg data,
1063                       TCGReg base, intptr_t ofs)
1064{
1065    switch (type) {
1066    case TCG_TYPE_I32:
1067        if (likely(is_general_reg(data))) {
1068            tcg_out_mem(s, RX_ST, RXY_STY, data, base, TCG_REG_NONE, ofs);
1069        } else {
1070            tcg_out_vrx_mem(s, VRX_VSTEF, data, base, TCG_REG_NONE, ofs, 1);
1071        }
1072        break;
1073
1074    case TCG_TYPE_I64:
1075        if (likely(is_general_reg(data))) {
1076            tcg_out_mem(s, 0, RXY_STG, data, base, TCG_REG_NONE, ofs);
1077            break;
1078        }
1079        /* fallthru */
1080
1081    case TCG_TYPE_V64:
1082        tcg_out_vrx_mem(s, VRX_VSTEG, data, base, TCG_REG_NONE, ofs, 0);
1083        break;
1084
1085    case TCG_TYPE_V128:
1086        /* Hint quadword aligned.  */
1087        tcg_out_vrx_mem(s, VRX_VST, data, base, TCG_REG_NONE, ofs, 4);
1088        break;
1089
1090    default:
1091        g_assert_not_reached();
1092    }
1093}
1094
1095static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
1096                               TCGReg base, intptr_t ofs)
1097{
1098    return false;
1099}
1100
1101static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
1102{
1103    return false;
1104}
1105
1106static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
1107                             tcg_target_long imm)
1108{
1109    /* This function is only used for passing structs by reference. */
1110    tcg_out_mem(s, RX_LA, RXY_LAY, rd, rs, TCG_REG_NONE, imm);
1111}
1112
1113static inline void tcg_out_risbg(TCGContext *s, TCGReg dest, TCGReg src,
1114                                 int msb, int lsb, int ofs, int z)
1115{
1116    /* Format RIE-f */
1117    tcg_out16(s, (RIEf_RISBG & 0xff00) | (dest << 4) | src);
1118    tcg_out16(s, (msb << 8) | (z << 7) | lsb);
1119    tcg_out16(s, (ofs << 8) | (RIEf_RISBG & 0xff));
1120}
1121
1122static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1123{
1124    tcg_out_insn(s, RRE, LGBR, dest, src);
1125}
1126
1127static void tcg_out_ext8u(TCGContext *s, TCGReg dest, TCGReg src)
1128{
1129    tcg_out_insn(s, RRE, LLGCR, dest, src);
1130}
1131
1132static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1133{
1134    tcg_out_insn(s, RRE, LGHR, dest, src);
1135}
1136
1137static void tcg_out_ext16u(TCGContext *s, TCGReg dest, TCGReg src)
1138{
1139    tcg_out_insn(s, RRE, LLGHR, dest, src);
1140}
1141
1142static void tcg_out_ext32s(TCGContext *s, TCGReg dest, TCGReg src)
1143{
1144    tcg_out_insn(s, RRE, LGFR, dest, src);
1145}
1146
1147static void tcg_out_ext32u(TCGContext *s, TCGReg dest, TCGReg src)
1148{
1149    tcg_out_insn(s, RRE, LLGFR, dest, src);
1150}
1151
1152static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg dest, TCGReg src)
1153{
1154    tcg_out_ext32s(s, dest, src);
1155}
1156
1157static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg dest, TCGReg src)
1158{
1159    tcg_out_ext32u(s, dest, src);
1160}
1161
1162static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg dest, TCGReg src)
1163{
1164    tcg_out_mov(s, TCG_TYPE_I32, dest, src);
1165}
1166
1167static void tgen_andi_risbg(TCGContext *s, TCGReg out, TCGReg in, uint64_t val)
1168{
1169    int msb, lsb;
1170    if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
1171        /* Achieve wraparound by swapping msb and lsb.  */
1172        msb = 64 - ctz64(~val);
1173        lsb = clz64(~val) - 1;
1174    } else {
1175        msb = clz64(val);
1176        lsb = 63 - ctz64(val);
1177    }
1178    tcg_out_risbg(s, out, in, msb, lsb, 0, 1);
1179}
1180
1181static void tgen_andi(TCGContext *s, TCGType type, TCGReg dest, uint64_t val)
1182{
1183    static const S390Opcode ni_insns[4] = {
1184        RI_NILL, RI_NILH, RI_NIHL, RI_NIHH
1185    };
1186    static const S390Opcode nif_insns[2] = {
1187        RIL_NILF, RIL_NIHF
1188    };
1189    uint64_t valid = (type == TCG_TYPE_I32 ? 0xffffffffull : -1ull);
1190    int i;
1191
1192    /* Look for the zero-extensions.  */
1193    if ((val & valid) == 0xffffffff) {
1194        tcg_out_ext32u(s, dest, dest);
1195        return;
1196    }
1197    if ((val & valid) == 0xff) {
1198        tcg_out_ext8u(s, dest, dest);
1199        return;
1200    }
1201    if ((val & valid) == 0xffff) {
1202        tcg_out_ext16u(s, dest, dest);
1203        return;
1204    }
1205
1206    i = is_const_p16(~val & valid);
1207    if (i >= 0) {
1208        tcg_out_insn_RI(s, ni_insns[i], dest, val >> (i * 16));
1209        return;
1210    }
1211
1212    i = is_const_p32(~val & valid);
1213    tcg_debug_assert(i == 0 || type != TCG_TYPE_I32);
1214    if (i >= 0) {
1215        tcg_out_insn_RIL(s, nif_insns[i], dest, val >> (i * 32));
1216        return;
1217    }
1218
1219    if (risbg_mask(val)) {
1220        tgen_andi_risbg(s, dest, dest, val);
1221        return;
1222    }
1223
1224    g_assert_not_reached();
1225}
1226
1227static void tgen_ori(TCGContext *s, TCGReg dest, uint64_t val)
1228{
1229    static const S390Opcode oif_insns[2] = {
1230        RIL_OILF, RIL_OIHF
1231    };
1232
1233    int i;
1234
1235    i = is_const_p16(val);
1236    if (i >= 0) {
1237        tcg_out_insn_RI(s, oi_insns[i], dest, val >> (i * 16));
1238        return;
1239    }
1240
1241    i = is_const_p32(val);
1242    if (i >= 0) {
1243        tcg_out_insn_RIL(s, oif_insns[i], dest, val >> (i * 32));
1244        return;
1245    }
1246
1247    g_assert_not_reached();
1248}
1249
1250static void tgen_xori(TCGContext *s, TCGReg dest, uint64_t val)
1251{
1252    switch (is_const_p32(val)) {
1253    case 0:
1254        tcg_out_insn(s, RIL, XILF, dest, val);
1255        break;
1256    case 1:
1257        tcg_out_insn(s, RIL, XIHF, dest, val >> 32);
1258        break;
1259    default:
1260        g_assert_not_reached();
1261    }
1262}
1263
1264static int tgen_cmp2(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1265                     TCGArg c2, bool c2const, bool need_carry, int *inv_cc)
1266{
1267    bool is_unsigned = is_unsigned_cond(c);
1268    TCGCond inv_c = tcg_invert_cond(c);
1269    S390Opcode op;
1270
1271    if (is_tst_cond(c)) {
1272        tcg_debug_assert(!need_carry);
1273
1274        if (!c2const) {
1275            if (type == TCG_TYPE_I32) {
1276                tcg_out_insn(s, RRFa, NRK, TCG_REG_R0, r1, c2);
1277            } else {
1278                tcg_out_insn(s, RRFa, NGRK, TCG_REG_R0, r1, c2);
1279            }
1280            goto exit;
1281        }
1282
1283        if (type == TCG_TYPE_I32) {
1284            c2 = (uint32_t)c2;
1285        }
1286
1287        int i = is_const_p16(c2);
1288        if (i >= 0) {
1289            tcg_out_insn_RI(s, tm_insns[i], r1, c2 >> (i * 16));
1290            *inv_cc = c == TCG_COND_TSTEQ ? S390_TM_NE : S390_TM_EQ;
1291            return *inv_cc ^ 15;
1292        }
1293
1294        if (risbg_mask(c2)) {
1295            tgen_andi_risbg(s, TCG_REG_R0, r1, c2);
1296            goto exit;
1297        }
1298        g_assert_not_reached();
1299    }
1300
1301    if (c2const) {
1302        if (c2 == 0) {
1303            if (!(is_unsigned && need_carry)) {
1304                if (type == TCG_TYPE_I32) {
1305                    tcg_out_insn(s, RR, LTR, r1, r1);
1306                } else {
1307                    tcg_out_insn(s, RRE, LTGR, r1, r1);
1308                }
1309                *inv_cc = tcg_cond_to_ltr_cond[inv_c];
1310                return tcg_cond_to_ltr_cond[c];
1311            }
1312        }
1313
1314        if (!is_unsigned && c2 == (int16_t)c2) {
1315            op = (type == TCG_TYPE_I32 ? RI_CHI : RI_CGHI);
1316            tcg_out_insn_RI(s, op, r1, c2);
1317            goto exit;
1318        }
1319
1320        if (type == TCG_TYPE_I32) {
1321            op = (is_unsigned ? RIL_CLFI : RIL_CFI);
1322            tcg_out_insn_RIL(s, op, r1, c2);
1323            goto exit;
1324        }
1325
1326        /* Should match TCG_CT_CONST_CMP. */
1327        switch (c) {
1328        case TCG_COND_LT:
1329        case TCG_COND_GE:
1330        case TCG_COND_LE:
1331        case TCG_COND_GT:
1332            tcg_debug_assert(c2 == (int32_t)c2);
1333            op = RIL_CGFI;
1334            break;
1335        case TCG_COND_EQ:
1336        case TCG_COND_NE:
1337            if (c2 == (int32_t)c2) {
1338                op = RIL_CGFI;
1339                break;
1340            }
1341            /* fall through */
1342        case TCG_COND_LTU:
1343        case TCG_COND_GEU:
1344        case TCG_COND_LEU:
1345        case TCG_COND_GTU:
1346            tcg_debug_assert(c2 == (uint32_t)c2);
1347            op = RIL_CLGFI;
1348            break;
1349        default:
1350            g_assert_not_reached();
1351        }
1352        tcg_out_insn_RIL(s, op, r1, c2);
1353    } else if (type == TCG_TYPE_I32) {
1354        op = (is_unsigned ? RR_CLR : RR_CR);
1355        tcg_out_insn_RR(s, op, r1, c2);
1356    } else {
1357        op = (is_unsigned ? RRE_CLGR : RRE_CGR);
1358        tcg_out_insn_RRE(s, op, r1, c2);
1359    }
1360
1361 exit:
1362    *inv_cc = tcg_cond_to_s390_cond[inv_c];
1363    return tcg_cond_to_s390_cond[c];
1364}
1365
1366static int tgen_cmp(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1367                    TCGArg c2, bool c2const, bool need_carry)
1368{
1369    int inv_cc;
1370    return tgen_cmp2(s, type, c, r1, c2, c2const, need_carry, &inv_cc);
1371}
1372
1373static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond,
1374                         TCGReg dest, TCGReg c1, TCGArg c2,
1375                         bool c2const, bool neg)
1376{
1377    int cc;
1378
1379    /* With LOC2, we can always emit the minimum 3 insns.  */
1380    if (HAVE_FACILITY(LOAD_ON_COND2)) {
1381        /* Emit: d = 0, d = (cc ? 1 : d).  */
1382        cc = tgen_cmp(s, type, cond, c1, c2, c2const, false);
1383        tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1384        tcg_out_insn(s, RIEg, LOCGHI, dest, neg ? -1 : 1, cc);
1385        return;
1386    }
1387
1388    switch (cond) {
1389    case TCG_COND_GEU:
1390    case TCG_COND_LTU:
1391    case TCG_COND_LT:
1392    case TCG_COND_GE:
1393        /* Swap operands so that we can use LEU/GTU/GT/LE.  */
1394        if (!c2const) {
1395            TCGReg t = c1;
1396            c1 = c2;
1397            c2 = t;
1398            cond = tcg_swap_cond(cond);
1399        }
1400        break;
1401    default:
1402        break;
1403    }
1404
1405    switch (cond) {
1406    case TCG_COND_NE:
1407        /* X != 0 is X > 0.  */
1408        if (c2const && c2 == 0) {
1409            cond = TCG_COND_GTU;
1410        } else {
1411            break;
1412        }
1413        /* fallthru */
1414
1415    case TCG_COND_GTU:
1416    case TCG_COND_GT:
1417        /*
1418         * The result of a compare has CC=2 for GT and CC=3 unused.
1419         * ADD LOGICAL WITH CARRY considers (CC & 2) the carry bit.
1420         */
1421        tgen_cmp(s, type, cond, c1, c2, c2const, true);
1422        tcg_out_movi(s, type, dest, 0);
1423        tcg_out_insn(s, RRE, ALCGR, dest, dest);
1424        if (neg) {
1425            if (type == TCG_TYPE_I32) {
1426                tcg_out_insn(s, RR, LCR, dest, dest);
1427            } else {
1428                tcg_out_insn(s, RRE, LCGR, dest, dest);
1429            }
1430        }
1431        return;
1432
1433    case TCG_COND_EQ:
1434        /* X == 0 is X <= 0.  */
1435        if (c2const && c2 == 0) {
1436            cond = TCG_COND_LEU;
1437        } else {
1438            break;
1439        }
1440        /* fallthru */
1441
1442    case TCG_COND_LEU:
1443    case TCG_COND_LE:
1444        /*
1445         * As above, but we're looking for borrow, or !carry.
1446         * The second insn computes d - d - borrow, or -1 for true
1447         * and 0 for false.  So we must mask to 1 bit afterward.
1448         */
1449        tgen_cmp(s, type, cond, c1, c2, c2const, true);
1450        tcg_out_insn(s, RRE, SLBGR, dest, dest);
1451        if (!neg) {
1452            tgen_andi(s, type, dest, 1);
1453        }
1454        return;
1455
1456    default:
1457        g_assert_not_reached();
1458    }
1459
1460    cc = tgen_cmp(s, type, cond, c1, c2, c2const, false);
1461    /* Emit: d = 0, t = 1, d = (cc ? t : d).  */
1462    tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1463    tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, neg ? -1 : 1);
1464    tcg_out_insn(s, RRFc, LOCGR, dest, TCG_TMP0, cc);
1465}
1466
1467static void tgen_movcond_int(TCGContext *s, TCGType type, TCGReg dest,
1468                             TCGArg v3, int v3const, TCGReg v4,
1469                             int cc, int inv_cc)
1470{
1471    TCGReg src;
1472
1473    if (v3const) {
1474        if (dest == v4) {
1475            if (HAVE_FACILITY(LOAD_ON_COND2)) {
1476                /* Emit: if (cc) dest = v3. */
1477                tcg_out_insn(s, RIEg, LOCGHI, dest, v3, cc);
1478                return;
1479            }
1480            tcg_out_insn(s, RI, LGHI, TCG_TMP0, v3);
1481            src = TCG_TMP0;
1482        } else {
1483            /* LGR+LOCGHI is larger than LGHI+LOCGR. */
1484            tcg_out_insn(s, RI, LGHI, dest, v3);
1485            cc = inv_cc;
1486            src = v4;
1487        }
1488    } else {
1489        if (HAVE_FACILITY(MISC_INSN_EXT3)) {
1490            /* Emit: dest = cc ? v3 : v4. */
1491            tcg_out_insn(s, RRFam, SELGR, dest, v3, v4, cc);
1492            return;
1493        }
1494        if (dest == v4) {
1495            src = v3;
1496        } else {
1497            tcg_out_mov(s, type, dest, v3);
1498            cc = inv_cc;
1499            src = v4;
1500        }
1501    }
1502
1503    /* Emit: if (cc) dest = src. */
1504    tcg_out_insn(s, RRFc, LOCGR, dest, src, cc);
1505}
1506
1507static void tgen_movcond(TCGContext *s, TCGType type, TCGCond c, TCGReg dest,
1508                         TCGReg c1, TCGArg c2, int c2const,
1509                         TCGArg v3, int v3const, TCGReg v4)
1510{
1511    int cc, inv_cc;
1512
1513    cc = tgen_cmp2(s, type, c, c1, c2, c2const, false, &inv_cc);
1514    tgen_movcond_int(s, type, dest, v3, v3const, v4, cc, inv_cc);
1515}
1516
1517static void tgen_ctpop(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1518{
1519    /* With MIE3, and bit 0 of m4 set, we get the complete result. */
1520    if (HAVE_FACILITY(MISC_INSN_EXT3)) {
1521        if (type == TCG_TYPE_I32) {
1522            tcg_out_ext32u(s, dest, src);
1523            src = dest;
1524        }
1525        tcg_out_insn(s, RRFc, POPCNT, dest, src, 8);
1526        return;
1527    }
1528
1529    /* Without MIE3, each byte gets the count of bits for the byte. */
1530    tcg_out_insn(s, RRFc, POPCNT, dest, src, 0);
1531
1532    /* Multiply to sum each byte at the top of the word. */
1533    if (type == TCG_TYPE_I32) {
1534        tcg_out_insn(s, RIL, MSFI, dest, 0x01010101);
1535        tcg_out_sh32(s, RS_SRL, dest, TCG_REG_NONE, 24);
1536    } else {
1537        tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, 0x0101010101010101ull);
1538        tcg_out_insn(s, RRE, MSGR, dest, TCG_TMP0);
1539        tcg_out_sh64(s, RSY_SRLG, dest, dest, TCG_REG_NONE, 56);
1540    }
1541}
1542
1543static void tgen_deposit(TCGContext *s, TCGReg dest, TCGReg src,
1544                         int ofs, int len, int z)
1545{
1546    int lsb = (63 - ofs);
1547    int msb = lsb - (len - 1);
1548    tcg_out_risbg(s, dest, src, msb, lsb, ofs, z);
1549}
1550
1551static void tgen_extract(TCGContext *s, TCGReg dest, TCGReg src,
1552                         int ofs, int len)
1553{
1554    if (ofs == 0) {
1555        switch (len) {
1556        case 8:
1557            tcg_out_ext8u(s, dest, src);
1558            return;
1559        case 16:
1560            tcg_out_ext16u(s, dest, src);
1561            return;
1562        case 32:
1563            tcg_out_ext32u(s, dest, src);
1564            return;
1565        }
1566    }
1567    tcg_out_risbg(s, dest, src, 64 - len, 63, 64 - ofs, 1);
1568}
1569
1570static void tgen_sextract(TCGContext *s, TCGReg dest, TCGReg src,
1571                          int ofs, int len)
1572{
1573    if (ofs == 0) {
1574        switch (len) {
1575        case 8:
1576            tcg_out_ext8s(s, TCG_TYPE_REG, dest, src);
1577            return;
1578        case 16:
1579            tcg_out_ext16s(s, TCG_TYPE_REG, dest, src);
1580            return;
1581        case 32:
1582            tcg_out_ext32s(s, dest, src);
1583            return;
1584        }
1585    }
1586    g_assert_not_reached();
1587}
1588
1589static void tgen_gotoi(TCGContext *s, int cc, const tcg_insn_unit *dest)
1590{
1591    ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1;
1592    if (off == (int16_t)off) {
1593        tcg_out_insn(s, RI, BRC, cc, off);
1594    } else if (off == (int32_t)off) {
1595        tcg_out_insn(s, RIL, BRCL, cc, off);
1596    } else {
1597        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1598        tcg_out_insn(s, RR, BCR, cc, TCG_TMP0);
1599    }
1600}
1601
1602static void tgen_branch(TCGContext *s, int cc, TCGLabel *l)
1603{
1604    if (l->has_value) {
1605        tgen_gotoi(s, cc, l->u.value_ptr);
1606    } else {
1607        tcg_out16(s, RI_BRC | (cc << 4));
1608        tcg_out_reloc(s, s->code_ptr, R_390_PC16DBL, l, 2);
1609        s->code_ptr += 1;
1610    }
1611}
1612
1613static void tgen_compare_branch(TCGContext *s, S390Opcode opc, int cc,
1614                                TCGReg r1, TCGReg r2, TCGLabel *l)
1615{
1616    tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1617    /* Format RIE-b */
1618    tcg_out16(s, (opc & 0xff00) | (r1 << 4) | r2);
1619    tcg_out16(s, 0);
1620    tcg_out16(s, cc << 12 | (opc & 0xff));
1621}
1622
1623static void tgen_compare_imm_branch(TCGContext *s, S390Opcode opc, int cc,
1624                                    TCGReg r1, int i2, TCGLabel *l)
1625{
1626    tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1627    /* Format RIE-c */
1628    tcg_out16(s, (opc & 0xff00) | (r1 << 4) | cc);
1629    tcg_out16(s, 0);
1630    tcg_out16(s, (i2 << 8) | (opc & 0xff));
1631}
1632
1633static void tgen_brcond(TCGContext *s, TCGType type, TCGCond c,
1634                        TCGReg r1, TCGArg c2, int c2const, TCGLabel *l)
1635{
1636    int cc;
1637
1638    if (!is_tst_cond(c)) {
1639        bool is_unsigned = is_unsigned_cond(c);
1640        bool in_range;
1641        S390Opcode opc;
1642
1643        cc = tcg_cond_to_s390_cond[c];
1644
1645        if (!c2const) {
1646            opc = (type == TCG_TYPE_I32
1647                   ? (is_unsigned ? RIEb_CLRJ : RIEb_CRJ)
1648                   : (is_unsigned ? RIEb_CLGRJ : RIEb_CGRJ));
1649            tgen_compare_branch(s, opc, cc, r1, c2, l);
1650            return;
1651        }
1652
1653        /*
1654         * COMPARE IMMEDIATE AND BRANCH RELATIVE has an 8-bit immediate field.
1655         * If the immediate we've been given does not fit that range, we'll
1656         * fall back to separate compare and branch instructions using the
1657         * larger comparison range afforded by COMPARE IMMEDIATE.
1658         */
1659        if (type == TCG_TYPE_I32) {
1660            if (is_unsigned) {
1661                opc = RIEc_CLIJ;
1662                in_range = (uint32_t)c2 == (uint8_t)c2;
1663            } else {
1664                opc = RIEc_CIJ;
1665                in_range = (int32_t)c2 == (int8_t)c2;
1666            }
1667        } else {
1668            if (is_unsigned) {
1669                opc = RIEc_CLGIJ;
1670                in_range = (uint64_t)c2 == (uint8_t)c2;
1671            } else {
1672                opc = RIEc_CGIJ;
1673                in_range = (int64_t)c2 == (int8_t)c2;
1674            }
1675        }
1676        if (in_range) {
1677            tgen_compare_imm_branch(s, opc, cc, r1, c2, l);
1678            return;
1679        }
1680    }
1681
1682    cc = tgen_cmp(s, type, c, r1, c2, c2const, false);
1683    tgen_branch(s, cc, l);
1684}
1685
1686static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *dest)
1687{
1688    ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1;
1689    if (off == (int32_t)off) {
1690        tcg_out_insn(s, RIL, BRASL, TCG_REG_R14, off);
1691    } else {
1692        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1693        tcg_out_insn(s, RR, BASR, TCG_REG_R14, TCG_TMP0);
1694    }
1695}
1696
1697static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest,
1698                         const TCGHelperInfo *info)
1699{
1700    tcg_out_call_int(s, dest);
1701}
1702
1703typedef struct {
1704    TCGReg base;
1705    TCGReg index;
1706    int disp;
1707    TCGAtomAlign aa;
1708} HostAddress;
1709
1710bool tcg_target_has_memory_bswap(MemOp memop)
1711{
1712    TCGAtomAlign aa;
1713
1714    if ((memop & MO_SIZE) <= MO_64) {
1715        return true;
1716    }
1717
1718    /*
1719     * Reject 16-byte memop with 16-byte atomicity,
1720     * but do allow a pair of 64-bit operations.
1721     */
1722    aa = atom_and_align_for_opc(tcg_ctx, memop, MO_ATOM_IFALIGN, true);
1723    return aa.atom <= MO_64;
1724}
1725
1726static void tcg_out_qemu_ld_direct(TCGContext *s, MemOp opc, TCGReg data,
1727                                   HostAddress h)
1728{
1729    switch (opc & (MO_SSIZE | MO_BSWAP)) {
1730    case MO_UB:
1731        tcg_out_insn(s, RXY, LLGC, data, h.base, h.index, h.disp);
1732        break;
1733    case MO_SB:
1734        tcg_out_insn(s, RXY, LGB, data, h.base, h.index, h.disp);
1735        break;
1736
1737    case MO_UW | MO_BSWAP:
1738        /* swapped unsigned halfword load with upper bits zeroed */
1739        tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp);
1740        tcg_out_ext16u(s, data, data);
1741        break;
1742    case MO_UW:
1743        tcg_out_insn(s, RXY, LLGH, data, h.base, h.index, h.disp);
1744        break;
1745
1746    case MO_SW | MO_BSWAP:
1747        /* swapped sign-extended halfword load */
1748        tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp);
1749        tcg_out_ext16s(s, TCG_TYPE_REG, data, data);
1750        break;
1751    case MO_SW:
1752        tcg_out_insn(s, RXY, LGH, data, h.base, h.index, h.disp);
1753        break;
1754
1755    case MO_UL | MO_BSWAP:
1756        /* swapped unsigned int load with upper bits zeroed */
1757        tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp);
1758        tcg_out_ext32u(s, data, data);
1759        break;
1760    case MO_UL:
1761        tcg_out_insn(s, RXY, LLGF, data, h.base, h.index, h.disp);
1762        break;
1763
1764    case MO_SL | MO_BSWAP:
1765        /* swapped sign-extended int load */
1766        tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp);
1767        tcg_out_ext32s(s, data, data);
1768        break;
1769    case MO_SL:
1770        tcg_out_insn(s, RXY, LGF, data, h.base, h.index, h.disp);
1771        break;
1772
1773    case MO_UQ | MO_BSWAP:
1774        tcg_out_insn(s, RXY, LRVG, data, h.base, h.index, h.disp);
1775        break;
1776    case MO_UQ:
1777        tcg_out_insn(s, RXY, LG, data, h.base, h.index, h.disp);
1778        break;
1779
1780    default:
1781        g_assert_not_reached();
1782    }
1783}
1784
1785static void tcg_out_qemu_st_direct(TCGContext *s, MemOp opc, TCGReg data,
1786                                   HostAddress h)
1787{
1788    switch (opc & (MO_SIZE | MO_BSWAP)) {
1789    case MO_UB:
1790        if (h.disp >= 0 && h.disp < 0x1000) {
1791            tcg_out_insn(s, RX, STC, data, h.base, h.index, h.disp);
1792        } else {
1793            tcg_out_insn(s, RXY, STCY, data, h.base, h.index, h.disp);
1794        }
1795        break;
1796
1797    case MO_UW | MO_BSWAP:
1798        tcg_out_insn(s, RXY, STRVH, data, h.base, h.index, h.disp);
1799        break;
1800    case MO_UW:
1801        if (h.disp >= 0 && h.disp < 0x1000) {
1802            tcg_out_insn(s, RX, STH, data, h.base, h.index, h.disp);
1803        } else {
1804            tcg_out_insn(s, RXY, STHY, data, h.base, h.index, h.disp);
1805        }
1806        break;
1807
1808    case MO_UL | MO_BSWAP:
1809        tcg_out_insn(s, RXY, STRV, data, h.base, h.index, h.disp);
1810        break;
1811    case MO_UL:
1812        if (h.disp >= 0 && h.disp < 0x1000) {
1813            tcg_out_insn(s, RX, ST, data, h.base, h.index, h.disp);
1814        } else {
1815            tcg_out_insn(s, RXY, STY, data, h.base, h.index, h.disp);
1816        }
1817        break;
1818
1819    case MO_UQ | MO_BSWAP:
1820        tcg_out_insn(s, RXY, STRVG, data, h.base, h.index, h.disp);
1821        break;
1822    case MO_UQ:
1823        tcg_out_insn(s, RXY, STG, data, h.base, h.index, h.disp);
1824        break;
1825
1826    default:
1827        g_assert_not_reached();
1828    }
1829}
1830
1831static const TCGLdstHelperParam ldst_helper_param = {
1832    .ntmp = 1, .tmp = { TCG_TMP0 }
1833};
1834
1835static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1836{
1837    MemOp opc = get_memop(lb->oi);
1838
1839    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1840                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) {
1841        return false;
1842    }
1843
1844    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
1845    tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SIZE]);
1846    tcg_out_ld_helper_ret(s, lb, false, &ldst_helper_param);
1847
1848    tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1849    return true;
1850}
1851
1852static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1853{
1854    MemOp opc = get_memop(lb->oi);
1855
1856    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1857                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) {
1858        return false;
1859    }
1860
1861    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
1862    tcg_out_call_int(s, qemu_st_helpers[opc & MO_SIZE]);
1863
1864    tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1865    return true;
1866}
1867
1868/* We're expecting to use a 20-bit negative offset on the tlb memory ops.  */
1869#define MIN_TLB_MASK_TABLE_OFS  -(1 << 19)
1870
1871/*
1872 * For system-mode, perform the TLB load and compare.
1873 * For user-mode, perform any required alignment tests.
1874 * In both cases, return a TCGLabelQemuLdst structure if the slow path
1875 * is required and fill in @h with the host address for the fast path.
1876 */
1877static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
1878                                           TCGReg addr_reg, MemOpIdx oi,
1879                                           bool is_ld)
1880{
1881    TCGType addr_type = s->addr_type;
1882    TCGLabelQemuLdst *ldst = NULL;
1883    MemOp opc = get_memop(oi);
1884    MemOp s_bits = opc & MO_SIZE;
1885    unsigned a_mask;
1886
1887    h->aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, s_bits == MO_128);
1888    a_mask = (1 << h->aa.align) - 1;
1889
1890    if (tcg_use_softmmu) {
1891        unsigned s_mask = (1 << s_bits) - 1;
1892        int mem_index = get_mmuidx(oi);
1893        int fast_off = tlb_mask_table_ofs(s, mem_index);
1894        int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1895        int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1896        int ofs, a_off;
1897        uint64_t tlb_mask;
1898
1899        ldst = new_ldst_label(s);
1900        ldst->is_ld = is_ld;
1901        ldst->oi = oi;
1902        ldst->addr_reg = addr_reg;
1903
1904        tcg_out_sh64(s, RSY_SRLG, TCG_TMP0, addr_reg, TCG_REG_NONE,
1905                     s->page_bits - CPU_TLB_ENTRY_BITS);
1906
1907        tcg_out_insn(s, RXY, NG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, mask_off);
1908        tcg_out_insn(s, RXY, AG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, table_off);
1909
1910        /*
1911         * For aligned accesses, we check the first byte and include the
1912         * alignment bits within the address.  For unaligned access, we
1913         * check that we don't cross pages using the address of the last
1914         * byte of the access.
1915         */
1916        a_off = (a_mask >= s_mask ? 0 : s_mask - a_mask);
1917        tlb_mask = (uint64_t)s->page_mask | a_mask;
1918        if (a_off == 0) {
1919            tgen_andi_risbg(s, TCG_REG_R0, addr_reg, tlb_mask);
1920        } else {
1921            tcg_out_insn(s, RX, LA, TCG_REG_R0, addr_reg, TCG_REG_NONE, a_off);
1922            tgen_andi(s, addr_type, TCG_REG_R0, tlb_mask);
1923        }
1924
1925        if (is_ld) {
1926            ofs = offsetof(CPUTLBEntry, addr_read);
1927        } else {
1928            ofs = offsetof(CPUTLBEntry, addr_write);
1929        }
1930        if (addr_type == TCG_TYPE_I32) {
1931            ofs += HOST_BIG_ENDIAN * 4;
1932            tcg_out_insn(s, RX, C, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs);
1933        } else {
1934            tcg_out_insn(s, RXY, CG, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs);
1935        }
1936
1937        tcg_out16(s, RI_BRC | (S390_CC_NE << 4));
1938        ldst->label_ptr[0] = s->code_ptr++;
1939
1940        h->index = TCG_TMP0;
1941        tcg_out_insn(s, RXY, LG, h->index, TCG_TMP0, TCG_REG_NONE,
1942                     offsetof(CPUTLBEntry, addend));
1943
1944        if (addr_type == TCG_TYPE_I32) {
1945            tcg_out_insn(s, RRE, ALGFR, h->index, addr_reg);
1946            h->base = TCG_REG_NONE;
1947        } else {
1948            h->base = addr_reg;
1949        }
1950        h->disp = 0;
1951    } else {
1952        if (a_mask) {
1953            ldst = new_ldst_label(s);
1954            ldst->is_ld = is_ld;
1955            ldst->oi = oi;
1956            ldst->addr_reg = addr_reg;
1957
1958            tcg_debug_assert(a_mask <= 0xffff);
1959            tcg_out_insn(s, RI, TMLL, addr_reg, a_mask);
1960
1961            tcg_out16(s, RI_BRC | (S390_TM_NE << 4));
1962            ldst->label_ptr[0] = s->code_ptr++;
1963        }
1964
1965        h->base = addr_reg;
1966        if (addr_type == TCG_TYPE_I32) {
1967            tcg_out_ext32u(s, TCG_TMP0, addr_reg);
1968            h->base = TCG_TMP0;
1969        }
1970        if (guest_base < 0x80000) {
1971            h->index = TCG_REG_NONE;
1972            h->disp = guest_base;
1973        } else {
1974            h->index = TCG_GUEST_BASE_REG;
1975            h->disp = 0;
1976        }
1977    }
1978
1979    return ldst;
1980}
1981
1982static void tcg_out_qemu_ld(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
1983                            MemOpIdx oi, TCGType data_type)
1984{
1985    TCGLabelQemuLdst *ldst;
1986    HostAddress h;
1987
1988    ldst = prepare_host_addr(s, &h, addr_reg, oi, true);
1989    tcg_out_qemu_ld_direct(s, get_memop(oi), data_reg, h);
1990
1991    if (ldst) {
1992        ldst->type = data_type;
1993        ldst->datalo_reg = data_reg;
1994        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1995    }
1996}
1997
1998static void tcg_out_qemu_st(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
1999                            MemOpIdx oi, TCGType data_type)
2000{
2001    TCGLabelQemuLdst *ldst;
2002    HostAddress h;
2003
2004    ldst = prepare_host_addr(s, &h, addr_reg, oi, false);
2005    tcg_out_qemu_st_direct(s, get_memop(oi), data_reg, h);
2006
2007    if (ldst) {
2008        ldst->type = data_type;
2009        ldst->datalo_reg = data_reg;
2010        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2011    }
2012}
2013
2014static void tcg_out_qemu_ldst_i128(TCGContext *s, TCGReg datalo, TCGReg datahi,
2015                                   TCGReg addr_reg, MemOpIdx oi, bool is_ld)
2016{
2017    TCGLabel *l1 = NULL, *l2 = NULL;
2018    TCGLabelQemuLdst *ldst;
2019    HostAddress h;
2020    bool need_bswap;
2021    bool use_pair;
2022    S390Opcode insn;
2023
2024    ldst = prepare_host_addr(s, &h, addr_reg, oi, is_ld);
2025
2026    use_pair = h.aa.atom < MO_128;
2027    need_bswap = get_memop(oi) & MO_BSWAP;
2028
2029    if (!use_pair) {
2030        /*
2031         * Atomicity requires we use LPQ.  If we've already checked for
2032         * 16-byte alignment, that's all we need.  If we arrive with
2033         * lesser alignment, we have determined that less than 16-byte
2034         * alignment can be satisfied with two 8-byte loads.
2035         */
2036        if (h.aa.align < MO_128) {
2037            use_pair = true;
2038            l1 = gen_new_label();
2039            l2 = gen_new_label();
2040
2041            tcg_out_insn(s, RI, TMLL, addr_reg, 15);
2042            tgen_branch(s, S390_TM_NE, l1);
2043        }
2044
2045        tcg_debug_assert(!need_bswap);
2046        tcg_debug_assert(datalo & 1);
2047        tcg_debug_assert(datahi == datalo - 1);
2048        insn = is_ld ? RXY_LPQ : RXY_STPQ;
2049        tcg_out_insn_RXY(s, insn, datahi, h.base, h.index, h.disp);
2050
2051        if (use_pair) {
2052            tgen_branch(s, S390_CC_ALWAYS, l2);
2053            tcg_out_label(s, l1);
2054        }
2055    }
2056    if (use_pair) {
2057        TCGReg d1, d2;
2058
2059        if (need_bswap) {
2060            d1 = datalo, d2 = datahi;
2061            insn = is_ld ? RXY_LRVG : RXY_STRVG;
2062        } else {
2063            d1 = datahi, d2 = datalo;
2064            insn = is_ld ? RXY_LG : RXY_STG;
2065        }
2066
2067        if (h.base == d1 || h.index == d1) {
2068            tcg_out_insn(s, RXY, LAY, TCG_TMP0, h.base, h.index, h.disp);
2069            h.base = TCG_TMP0;
2070            h.index = TCG_REG_NONE;
2071            h.disp = 0;
2072        }
2073        tcg_out_insn_RXY(s, insn, d1, h.base, h.index, h.disp);
2074        tcg_out_insn_RXY(s, insn, d2, h.base, h.index, h.disp + 8);
2075    }
2076    if (l2) {
2077        tcg_out_label(s, l2);
2078    }
2079
2080    if (ldst) {
2081        ldst->type = TCG_TYPE_I128;
2082        ldst->datalo_reg = datalo;
2083        ldst->datahi_reg = datahi;
2084        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
2085    }
2086}
2087
2088static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
2089{
2090    /* Reuse the zeroing that exists for goto_ptr.  */
2091    if (a0 == 0) {
2092        tgen_gotoi(s, S390_CC_ALWAYS, tcg_code_gen_epilogue);
2093    } else {
2094        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, a0);
2095        tgen_gotoi(s, S390_CC_ALWAYS, tb_ret_addr);
2096    }
2097}
2098
2099static void tcg_out_goto_tb(TCGContext *s, int which)
2100{
2101    /*
2102     * Branch displacement must be aligned for atomic patching;
2103     * see if we need to add extra nop before branch
2104     */
2105    if (!QEMU_PTR_IS_ALIGNED(s->code_ptr + 1, 4)) {
2106        tcg_out16(s, NOP);
2107    }
2108    tcg_out16(s, RIL_BRCL | (S390_CC_ALWAYS << 4));
2109    set_jmp_insn_offset(s, which);
2110    s->code_ptr += 2;
2111    set_jmp_reset_offset(s, which);
2112}
2113
2114void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
2115                              uintptr_t jmp_rx, uintptr_t jmp_rw)
2116{
2117    if (!HAVE_FACILITY(GEN_INST_EXT)) {
2118        return;
2119    }
2120    /* patch the branch destination */
2121    uintptr_t addr = tb->jmp_target_addr[n];
2122    intptr_t disp = addr - (jmp_rx - 2);
2123    qatomic_set((int32_t *)jmp_rw, disp / 2);
2124    /* no need to flush icache explicitly */
2125}
2126
2127
2128static void tgen_add(TCGContext *s, TCGType type,
2129                     TCGReg a0, TCGReg a1, TCGReg a2)
2130{
2131    if (a0 != a1) {
2132        tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
2133    } else if (type == TCG_TYPE_I32) {
2134        tcg_out_insn(s, RR, AR, a0, a2);
2135    } else {
2136        tcg_out_insn(s, RRE, AGR, a0, a2);
2137    }
2138}
2139
2140static void tgen_addi(TCGContext *s, TCGType type,
2141                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2142{
2143    if (a0 == a1) {
2144        if (type == TCG_TYPE_I32) {
2145            if (a2 == (int16_t)a2) {
2146                tcg_out_insn(s, RI, AHI, a0, a2);
2147            } else {
2148                tcg_out_insn(s, RIL, AFI, a0, a2);
2149            }
2150            return;
2151        }
2152        if (a2 == (int16_t)a2) {
2153            tcg_out_insn(s, RI, AGHI, a0, a2);
2154            return;
2155        }
2156        if (a2 == (int32_t)a2) {
2157            tcg_out_insn(s, RIL, AGFI, a0, a2);
2158            return;
2159        }
2160        if (a2 == (uint32_t)a2) {
2161            tcg_out_insn(s, RIL, ALGFI, a0, a2);
2162            return;
2163        }
2164        if (-a2 == (uint32_t)-a2) {
2165            tcg_out_insn(s, RIL, SLGFI, a0, -a2);
2166            return;
2167        }
2168    }
2169    tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
2170}
2171
2172static const TCGOutOpBinary outop_add = {
2173    .base.static_constraint = C_O1_I2(r, r, ri),
2174    .out_rrr = tgen_add,
2175    .out_rri = tgen_addi,
2176};
2177
2178static void tgen_and(TCGContext *s, TCGType type,
2179                     TCGReg a0, TCGReg a1, TCGReg a2)
2180{
2181    if (type != TCG_TYPE_I32) {
2182        tcg_out_insn(s, RRFa, NGRK, a0, a1, a2);
2183    } else if (a0 == a1) {
2184        tcg_out_insn(s, RR, NR, a0, a2);
2185    } else {
2186        tcg_out_insn(s, RRFa, NRK, a0, a1, a2);
2187    }
2188}
2189
2190static void tgen_andi_3(TCGContext *s, TCGType type,
2191                        TCGReg a0, TCGReg a1, tcg_target_long a2)
2192{
2193    tcg_out_mov(s, type, a0, a1);
2194    tgen_andi(s, type, a0, a2);
2195}
2196
2197static const TCGOutOpBinary outop_and = {
2198    .base.static_constraint = C_O1_I2(r, r, rNKR),
2199    .out_rrr = tgen_and,
2200    .out_rri = tgen_andi_3,
2201};
2202
2203static void tgen_andc(TCGContext *s, TCGType type,
2204                      TCGReg a0, TCGReg a1, TCGReg a2)
2205{
2206    if (type == TCG_TYPE_I32) {
2207        tcg_out_insn(s, RRFa, NCRK, a0, a1, a2);
2208    } else {
2209        tcg_out_insn(s, RRFa, NCGRK, a0, a1, a2);
2210    }
2211}
2212
2213static TCGConstraintSetIndex cset_misc3_rrr(TCGType type, unsigned flags)
2214{
2215    return HAVE_FACILITY(MISC_INSN_EXT3) ? C_O1_I2(r, r, r) : C_NotImplemented;
2216}
2217
2218static const TCGOutOpBinary outop_andc = {
2219    .base.static_constraint = C_Dynamic,
2220    .base.dynamic_constraint = cset_misc3_rrr,
2221    .out_rrr = tgen_andc,
2222};
2223
2224static void tgen_clz_int(TCGContext *s, TCGReg dest, TCGReg a1,
2225                         TCGArg a2, int a2const)
2226{
2227    /*
2228     * Since this sets both R and R+1, we have no choice but to store the
2229     * result into R0, allowing R1 == TCG_TMP0 to be clobbered as well.
2230     */
2231    QEMU_BUILD_BUG_ON(TCG_TMP0 != TCG_REG_R1);
2232    tcg_out_insn(s, RRE, FLOGR, TCG_REG_R0, a1);
2233
2234    if (a2const && a2 == 64) {
2235        tcg_out_mov(s, TCG_TYPE_I64, dest, TCG_REG_R0);
2236        return;
2237    }
2238
2239    /*
2240     * Conditions from FLOGR are:
2241     *   2 -> one bit found
2242     *   8 -> no one bit found
2243     */
2244    tgen_movcond_int(s, TCG_TYPE_I64, dest, a2, a2const, TCG_REG_R0, 8, 2);
2245}
2246
2247static void tgen_clz(TCGContext *s, TCGType type,
2248                     TCGReg a0, TCGReg a1, TCGReg a2)
2249{
2250    tgen_clz_int(s, a0, a1, a2, false);
2251}
2252
2253static void tgen_clzi(TCGContext *s, TCGType type,
2254                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2255{
2256    tgen_clz_int(s, a0, a1, a2, true);
2257}
2258
2259static TCGConstraintSetIndex cset_clz(TCGType type, unsigned flags)
2260{
2261    return type == TCG_TYPE_I64 ? C_O1_I2(r, r, rI) : C_NotImplemented;
2262}
2263
2264static const TCGOutOpBinary outop_clz = {
2265    .base.static_constraint = C_Dynamic,
2266    .base.dynamic_constraint = cset_clz,
2267    .out_rrr = tgen_clz,
2268    .out_rri = tgen_clzi,
2269};
2270
2271static const TCGOutOpBinary outop_divs = {
2272    .base.static_constraint = C_NotImplemented,
2273};
2274
2275static void tgen_divs2(TCGContext *s, TCGType type,
2276                       TCGReg a0, TCGReg a1, TCGReg a4)
2277{
2278    tcg_debug_assert((a1 & 1) == 0);
2279    tcg_debug_assert(a0 == a1 + 1);
2280    if (type == TCG_TYPE_I32) {
2281        tcg_out_insn(s, RR, DR, a1, a4);
2282    } else {
2283        /*
2284         * TODO: Move the sign-extend of the numerator from a2 into a3
2285         * into the tcg backend, instead of in early expansion.  It is
2286         * required for 32-bit DR, but not 64-bit DSGR.
2287         */
2288        tcg_out_insn(s, RRE, DSGR, a1, a4);
2289    }
2290}
2291
2292static const TCGOutOpDivRem outop_divs2 = {
2293    .base.static_constraint = C_O2_I3(o, m, 0, 1, r),
2294    .out_rr01r = tgen_divs2,
2295};
2296
2297static const TCGOutOpBinary outop_divu = {
2298    .base.static_constraint = C_NotImplemented,
2299};
2300
2301static void tgen_divu2(TCGContext *s, TCGType type,
2302                       TCGReg a0, TCGReg a1, TCGReg a4)
2303{
2304    tcg_debug_assert((a1 & 1) == 0);
2305    tcg_debug_assert(a0 == a1 + 1);
2306    if (type == TCG_TYPE_I32) {
2307        tcg_out_insn(s, RRE, DLR, a1, a4);
2308    } else {
2309        tcg_out_insn(s, RRE, DLGR, a1, a4);
2310    }
2311}
2312
2313static const TCGOutOpDivRem outop_divu2 = {
2314    .base.static_constraint = C_O2_I3(o, m, 0, 1, r),
2315    .out_rr01r = tgen_divu2,
2316};
2317
2318static void tgen_eqv(TCGContext *s, TCGType type,
2319                      TCGReg a0, TCGReg a1, TCGReg a2)
2320{
2321    if (type == TCG_TYPE_I32) {
2322        tcg_out_insn(s, RRFa, NXRK, a0, a1, a2);
2323    } else {
2324        tcg_out_insn(s, RRFa, NXGRK, a0, a1, a2);
2325    }
2326}
2327
2328static const TCGOutOpBinary outop_eqv = {
2329    .base.static_constraint = C_Dynamic,
2330    .base.dynamic_constraint = cset_misc3_rrr,
2331    .out_rrr = tgen_eqv,
2332};
2333
2334static void tgen_mul(TCGContext *s, TCGType type,
2335                     TCGReg a0, TCGReg a1, TCGReg a2)
2336{
2337    if (type == TCG_TYPE_I32) {
2338        if (a0 == a1) {
2339            tcg_out_insn(s, RRE, MSR, a0, a2);
2340        } else {
2341            tcg_out_insn(s, RRFa, MSRKC, a0, a1, a2);
2342        }
2343    } else {
2344        if (a0 == a1) {
2345            tcg_out_insn(s, RRE, MSGR, a0, a2);
2346        } else {
2347            tcg_out_insn(s, RRFa, MSGRKC, a0, a1, a2);
2348        }
2349    }
2350}
2351
2352static void tgen_muli(TCGContext *s, TCGType type,
2353                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2354{
2355    tcg_out_mov(s, type, a0, a1);
2356    if (type == TCG_TYPE_I32) {
2357        if (a2 == (int16_t)a2) {
2358            tcg_out_insn(s, RI, MHI, a0, a2);
2359        } else {
2360            tcg_out_insn(s, RIL, MSFI, a0, a2);
2361        }
2362    } else {
2363        if (a2 == (int16_t)a2) {
2364            tcg_out_insn(s, RI, MGHI, a0, a2);
2365        } else {
2366            tcg_out_insn(s, RIL, MSGFI, a0, a2);
2367        }
2368    }
2369}
2370
2371static TCGConstraintSetIndex cset_mul(TCGType type, unsigned flags)
2372{
2373    return (HAVE_FACILITY(MISC_INSN_EXT2)
2374            ? C_O1_I2(r, r, rJ)
2375            : C_O1_I2(r, 0, rJ));
2376}
2377
2378static const TCGOutOpBinary outop_mul = {
2379    .base.static_constraint = C_Dynamic,
2380    .base.dynamic_constraint = cset_mul,
2381    .out_rrr = tgen_mul,
2382    .out_rri = tgen_muli,
2383};
2384
2385static const TCGOutOpBinary outop_mulsh = {
2386    .base.static_constraint = C_NotImplemented,
2387};
2388
2389static const TCGOutOpBinary outop_muluh = {
2390    .base.static_constraint = C_NotImplemented,
2391};
2392
2393static void tgen_nand(TCGContext *s, TCGType type,
2394                      TCGReg a0, TCGReg a1, TCGReg a2)
2395{
2396    if (type == TCG_TYPE_I32) {
2397        tcg_out_insn(s, RRFa, NNRK, a0, a1, a2);
2398    } else {
2399        tcg_out_insn(s, RRFa, NNGRK, a0, a1, a2);
2400    }
2401}
2402
2403static const TCGOutOpBinary outop_nand = {
2404    .base.static_constraint = C_Dynamic,
2405    .base.dynamic_constraint = cset_misc3_rrr,
2406    .out_rrr = tgen_nand,
2407};
2408
2409static void tgen_nor(TCGContext *s, TCGType type,
2410                      TCGReg a0, TCGReg a1, TCGReg a2)
2411{
2412    if (type == TCG_TYPE_I32) {
2413        tcg_out_insn(s, RRFa, NORK, a0, a1, a2);
2414    } else {
2415        tcg_out_insn(s, RRFa, NOGRK, a0, a1, a2);
2416    }
2417}
2418
2419static const TCGOutOpBinary outop_nor = {
2420    .base.static_constraint = C_Dynamic,
2421    .base.dynamic_constraint = cset_misc3_rrr,
2422    .out_rrr = tgen_nor,
2423};
2424
2425static void tgen_or(TCGContext *s, TCGType type,
2426                     TCGReg a0, TCGReg a1, TCGReg a2)
2427{
2428    if (type != TCG_TYPE_I32) {
2429        tcg_out_insn(s, RRFa, OGRK, a0, a1, a2);
2430    } else if (a0 == a1) {
2431        tcg_out_insn(s, RR, OR, a0, a2);
2432    } else {
2433        tcg_out_insn(s, RRFa, ORK, a0, a1, a2);
2434    }
2435}
2436
2437static void tgen_ori_3(TCGContext *s, TCGType type,
2438                        TCGReg a0, TCGReg a1, tcg_target_long a2)
2439{
2440    tcg_out_mov(s, type, a0, a1);
2441    tgen_ori(s, a0, type == TCG_TYPE_I32 ? (uint32_t)a2 : a2);
2442}
2443
2444static const TCGOutOpBinary outop_or = {
2445    .base.static_constraint = C_O1_I2(r, r, rK),
2446    .out_rrr = tgen_or,
2447    .out_rri = tgen_ori_3,
2448};
2449
2450static void tgen_orc(TCGContext *s, TCGType type,
2451                     TCGReg a0, TCGReg a1, TCGReg a2)
2452{
2453    if (type == TCG_TYPE_I32) {
2454        tcg_out_insn(s, RRFa, OCRK, a0, a1, a2);
2455    } else {
2456        tcg_out_insn(s, RRFa, OCGRK, a0, a1, a2);
2457    }
2458}
2459
2460static const TCGOutOpBinary outop_orc = {
2461    .base.static_constraint = C_Dynamic,
2462    .base.dynamic_constraint = cset_misc3_rrr,
2463    .out_rrr = tgen_orc,
2464};
2465
2466static const TCGOutOpBinary outop_rems = {
2467    .base.static_constraint = C_NotImplemented,
2468};
2469
2470static const TCGOutOpBinary outop_remu = {
2471    .base.static_constraint = C_NotImplemented,
2472};
2473
2474static void tgen_rotl_int(TCGContext *s, TCGType type, TCGReg dst,
2475                          TCGReg src, TCGReg v, tcg_target_long i)
2476{
2477    S390Opcode insn = type == TCG_TYPE_I32 ? RSY_RLL : RSY_RLLG;
2478    tcg_out_sh64(s, insn, dst, src, v, i);
2479}
2480
2481static void tgen_rotl(TCGContext *s, TCGType type,
2482                      TCGReg a0, TCGReg a1, TCGReg a2)
2483{
2484    tgen_rotl_int(s, type, a0, a1, a2, 0);
2485}
2486
2487static void tgen_rotli(TCGContext *s, TCGType type,
2488                       TCGReg a0, TCGReg a1, tcg_target_long a2)
2489{
2490    tgen_rotl_int(s, type, a0, a1, TCG_REG_NONE, a2);
2491}
2492
2493static const TCGOutOpBinary outop_rotl = {
2494    .base.static_constraint = C_O1_I2(r, r, ri),
2495    .out_rrr = tgen_rotl,
2496    .out_rri = tgen_rotli,
2497};
2498
2499static const TCGOutOpBinary outop_rotr = {
2500    .base.static_constraint = C_NotImplemented,
2501};
2502
2503static void tgen_sar_int(TCGContext *s, TCGType type, TCGReg dst,
2504                         TCGReg src, TCGReg v, tcg_target_long i)
2505{
2506    if (type != TCG_TYPE_I32) {
2507        tcg_out_sh64(s, RSY_SRAG, dst, src, v, i);
2508    } else if (dst == src) {
2509        tcg_out_sh32(s, RS_SRA, dst, v, i);
2510    } else {
2511        tcg_out_sh64(s, RSY_SRAK, dst, src, v, i);
2512    }
2513}
2514
2515static void tgen_sar(TCGContext *s, TCGType type,
2516                     TCGReg a0, TCGReg a1, TCGReg a2)
2517{
2518    tgen_sar_int(s, type, a0, a1, a2, 0);
2519}
2520
2521static void tgen_sari(TCGContext *s, TCGType type,
2522                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2523{
2524    tgen_sar_int(s, type, a0, a1, TCG_REG_NONE, a2);
2525}
2526
2527static const TCGOutOpBinary outop_sar = {
2528    .base.static_constraint = C_O1_I2(r, r, ri),
2529    .out_rrr = tgen_sar,
2530    .out_rri = tgen_sari,
2531};
2532
2533static void tgen_shl_int(TCGContext *s, TCGType type, TCGReg dst,
2534                         TCGReg src, TCGReg v, tcg_target_long i)
2535{
2536    if (type != TCG_TYPE_I32) {
2537        tcg_out_sh64(s, RSY_SLLG, dst, src, v, i);
2538    } else if (dst == src) {
2539        tcg_out_sh32(s, RS_SLL, dst, v, i);
2540    } else {
2541        tcg_out_sh64(s, RSY_SLLK, dst, src, v, i);
2542    }
2543}
2544
2545static void tgen_shl(TCGContext *s, TCGType type,
2546                     TCGReg a0, TCGReg a1, TCGReg a2)
2547{
2548    tgen_shl_int(s, type, a0, a1, a2, 0);
2549}
2550
2551static void tgen_shli(TCGContext *s, TCGType type,
2552                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2553{
2554    tgen_shl_int(s, type, a0, a1, TCG_REG_NONE, a2);
2555}
2556
2557static const TCGOutOpBinary outop_shl = {
2558    .base.static_constraint = C_O1_I2(r, r, ri),
2559    .out_rrr = tgen_shl,
2560    .out_rri = tgen_shli,
2561};
2562
2563static void tgen_shr_int(TCGContext *s, TCGType type, TCGReg dst,
2564                         TCGReg src, TCGReg v, tcg_target_long i)
2565{
2566    if (type != TCG_TYPE_I32) {
2567        tcg_out_sh64(s, RSY_SRLG, dst, src, v, i);
2568    } else if (dst == src) {
2569        tcg_out_sh32(s, RS_SRL, dst, v, i);
2570    } else {
2571        tcg_out_sh64(s, RSY_SRLK, dst, src, v, i);
2572    }
2573}
2574
2575static void tgen_shr(TCGContext *s, TCGType type,
2576                     TCGReg a0, TCGReg a1, TCGReg a2)
2577{
2578    tgen_shr_int(s, type, a0, a1, a2, 0);
2579}
2580
2581static void tgen_shri(TCGContext *s, TCGType type,
2582                      TCGReg a0, TCGReg a1, tcg_target_long a2)
2583{
2584    tgen_shr_int(s, type, a0, a1, TCG_REG_NONE, a2);
2585}
2586
2587static const TCGOutOpBinary outop_shr = {
2588    .base.static_constraint = C_O1_I2(r, r, ri),
2589    .out_rrr = tgen_shr,
2590    .out_rri = tgen_shri,
2591};
2592
2593static void tgen_sub(TCGContext *s, TCGType type,
2594                     TCGReg a0, TCGReg a1, TCGReg a2)
2595{
2596    if (type != TCG_TYPE_I32) {
2597        tcg_out_insn(s, RRFa, SGRK, a0, a1, a2);
2598    } else if (a0 == a1) {
2599        tcg_out_insn(s, RR, SR, a0, a2);
2600    } else {
2601        tcg_out_insn(s, RRFa, SRK, a0, a1, a2);
2602    }
2603}
2604
2605static const TCGOutOpSubtract outop_sub = {
2606    .base.static_constraint = C_O1_I2(r, r, r),
2607    .out_rrr = tgen_sub,
2608};
2609
2610static void tgen_xor(TCGContext *s, TCGType type,
2611                     TCGReg a0, TCGReg a1, TCGReg a2)
2612{
2613    if (type != TCG_TYPE_I32) {
2614        tcg_out_insn(s, RRFa, XGRK, a0, a1, a2);
2615    } else if (a0 == a1) {
2616        tcg_out_insn(s, RR, XR, a0, a2);
2617    } else {
2618        tcg_out_insn(s, RRFa, XRK, a0, a1, a2);
2619    }
2620}
2621
2622static void tgen_xori_3(TCGContext *s, TCGType type,
2623                        TCGReg a0, TCGReg a1, tcg_target_long a2)
2624{
2625    tcg_out_mov(s, type, a0, a1);
2626    tgen_xori(s, a0, type == TCG_TYPE_I32 ? (uint32_t)a2 : a2);
2627}
2628
2629static const TCGOutOpBinary outop_xor = {
2630    .base.static_constraint = C_O1_I2(r, r, rK),
2631    .out_rrr = tgen_xor,
2632    .out_rri = tgen_xori_3,
2633};
2634
2635static void tgen_neg(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2636{
2637    if (type == TCG_TYPE_I32) {
2638        tcg_out_insn(s, RR, LCR, a0, a1);
2639    } else {
2640        tcg_out_insn(s, RRE, LCGR, a0, a1);
2641    }
2642}
2643
2644static const TCGOutOpUnary outop_neg = {
2645    .base.static_constraint = C_O1_I1(r, r),
2646    .out_rr = tgen_neg,
2647};
2648
2649static void tgen_not(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
2650{
2651    tgen_nor(s, type, a0, a1, a1);
2652}
2653
2654static TCGConstraintSetIndex cset_not(TCGType type, unsigned flags)
2655{
2656    return HAVE_FACILITY(MISC_INSN_EXT3) ? C_O1_I1(r, r) : C_NotImplemented;
2657}
2658
2659static const TCGOutOpUnary outop_not = {
2660    .base.static_constraint = C_Dynamic,
2661    .base.dynamic_constraint = cset_not,
2662    .out_rr = tgen_not,
2663};
2664
2665
2666# define OP_32_64(x) \
2667        case glue(glue(INDEX_op_,x),_i32): \
2668        case glue(glue(INDEX_op_,x),_i64)
2669
2670static void tcg_out_op(TCGContext *s, TCGOpcode opc, TCGType type,
2671                       const TCGArg args[TCG_MAX_OP_ARGS],
2672                       const int const_args[TCG_MAX_OP_ARGS])
2673{
2674    TCGArg a0, a1, a2;
2675
2676    switch (opc) {
2677    case INDEX_op_goto_ptr:
2678        a0 = args[0];
2679        tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, a0);
2680        break;
2681
2682    OP_32_64(ld8u):
2683        /* ??? LLC (RXY format) is only present with the extended-immediate
2684           facility, whereas LLGC is always present.  */
2685        tcg_out_mem(s, 0, RXY_LLGC, args[0], args[1], TCG_REG_NONE, args[2]);
2686        break;
2687
2688    OP_32_64(ld8s):
2689        /* ??? LB is no smaller than LGB, so no point to using it.  */
2690        tcg_out_mem(s, 0, RXY_LGB, args[0], args[1], TCG_REG_NONE, args[2]);
2691        break;
2692
2693    OP_32_64(ld16u):
2694        /* ??? LLH (RXY format) is only present with the extended-immediate
2695           facility, whereas LLGH is always present.  */
2696        tcg_out_mem(s, 0, RXY_LLGH, args[0], args[1], TCG_REG_NONE, args[2]);
2697        break;
2698
2699    case INDEX_op_ld16s_i32:
2700        tcg_out_mem(s, RX_LH, RXY_LHY, args[0], args[1], TCG_REG_NONE, args[2]);
2701        break;
2702
2703    case INDEX_op_ld_i32:
2704        tcg_out_ld(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2705        break;
2706
2707    OP_32_64(st8):
2708        tcg_out_mem(s, RX_STC, RXY_STCY, args[0], args[1],
2709                    TCG_REG_NONE, args[2]);
2710        break;
2711
2712    OP_32_64(st16):
2713        tcg_out_mem(s, RX_STH, RXY_STHY, args[0], args[1],
2714                    TCG_REG_NONE, args[2]);
2715        break;
2716
2717    case INDEX_op_st_i32:
2718        tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2719        break;
2720
2721    case INDEX_op_bswap16_i32:
2722        a0 = args[0], a1 = args[1], a2 = args[2];
2723        tcg_out_insn(s, RRE, LRVR, a0, a1);
2724        if (a2 & TCG_BSWAP_OS) {
2725            tcg_out_sh32(s, RS_SRA, a0, TCG_REG_NONE, 16);
2726        } else {
2727            tcg_out_sh32(s, RS_SRL, a0, TCG_REG_NONE, 16);
2728        }
2729        break;
2730    case INDEX_op_bswap16_i64:
2731        a0 = args[0], a1 = args[1], a2 = args[2];
2732        tcg_out_insn(s, RRE, LRVGR, a0, a1);
2733        if (a2 & TCG_BSWAP_OS) {
2734            tcg_out_sh64(s, RSY_SRAG, a0, a0, TCG_REG_NONE, 48);
2735        } else {
2736            tcg_out_sh64(s, RSY_SRLG, a0, a0, TCG_REG_NONE, 48);
2737        }
2738        break;
2739
2740    case INDEX_op_bswap32_i32:
2741        tcg_out_insn(s, RRE, LRVR, args[0], args[1]);
2742        break;
2743    case INDEX_op_bswap32_i64:
2744        a0 = args[0], a1 = args[1], a2 = args[2];
2745        tcg_out_insn(s, RRE, LRVR, a0, a1);
2746        if (a2 & TCG_BSWAP_OS) {
2747            tcg_out_ext32s(s, a0, a0);
2748        } else if ((a2 & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
2749            tcg_out_ext32u(s, a0, a0);
2750        }
2751        break;
2752
2753    case INDEX_op_add2_i32:
2754        if (const_args[4]) {
2755            tcg_out_insn(s, RIL, ALFI, args[0], args[4]);
2756        } else {
2757            tcg_out_insn(s, RR, ALR, args[0], args[4]);
2758        }
2759        tcg_out_insn(s, RRE, ALCR, args[1], args[5]);
2760        break;
2761    case INDEX_op_sub2_i32:
2762        if (const_args[4]) {
2763            tcg_out_insn(s, RIL, SLFI, args[0], args[4]);
2764        } else {
2765            tcg_out_insn(s, RR, SLR, args[0], args[4]);
2766        }
2767        tcg_out_insn(s, RRE, SLBR, args[1], args[5]);
2768        break;
2769
2770    case INDEX_op_br:
2771        tgen_branch(s, S390_CC_ALWAYS, arg_label(args[0]));
2772        break;
2773
2774    case INDEX_op_brcond_i32:
2775        tgen_brcond(s, TCG_TYPE_I32, args[2], args[0],
2776                    args[1], const_args[1], arg_label(args[3]));
2777        break;
2778    case INDEX_op_setcond_i32:
2779        tgen_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1],
2780                     args[2], const_args[2], false);
2781        break;
2782    case INDEX_op_negsetcond_i32:
2783        tgen_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1],
2784                     args[2], const_args[2], true);
2785        break;
2786    case INDEX_op_movcond_i32:
2787        tgen_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1],
2788                     args[2], const_args[2], args[3], const_args[3], args[4]);
2789        break;
2790
2791    case INDEX_op_qemu_ld_i32:
2792        tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I32);
2793        break;
2794    case INDEX_op_qemu_ld_i64:
2795        tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I64);
2796        break;
2797    case INDEX_op_qemu_st_i32:
2798        tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I32);
2799        break;
2800    case INDEX_op_qemu_st_i64:
2801        tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I64);
2802        break;
2803    case INDEX_op_qemu_ld_i128:
2804        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], true);
2805        break;
2806    case INDEX_op_qemu_st_i128:
2807        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], false);
2808        break;
2809
2810    case INDEX_op_ld16s_i64:
2811        tcg_out_mem(s, 0, RXY_LGH, args[0], args[1], TCG_REG_NONE, args[2]);
2812        break;
2813    case INDEX_op_ld32u_i64:
2814        tcg_out_mem(s, 0, RXY_LLGF, args[0], args[1], TCG_REG_NONE, args[2]);
2815        break;
2816    case INDEX_op_ld32s_i64:
2817        tcg_out_mem(s, 0, RXY_LGF, args[0], args[1], TCG_REG_NONE, args[2]);
2818        break;
2819    case INDEX_op_ld_i64:
2820        tcg_out_ld(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2821        break;
2822
2823    case INDEX_op_st32_i64:
2824        tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2825        break;
2826    case INDEX_op_st_i64:
2827        tcg_out_st(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2828        break;
2829
2830    case INDEX_op_bswap64_i64:
2831        tcg_out_insn(s, RRE, LRVGR, args[0], args[1]);
2832        break;
2833
2834    case INDEX_op_mulu2_i64:
2835        tcg_debug_assert(args[0] == args[2]);
2836        tcg_debug_assert((args[1] & 1) == 0);
2837        tcg_debug_assert(args[0] == args[1] + 1);
2838        tcg_out_insn(s, RRE, MLGR, args[1], args[3]);
2839        break;
2840    case INDEX_op_muls2_i64:
2841        tcg_debug_assert((args[1] & 1) == 0);
2842        tcg_debug_assert(args[0] == args[1] + 1);
2843        tcg_out_insn(s, RRFa, MGRK, args[1], args[2], args[3]);
2844        break;
2845
2846    case INDEX_op_add2_i64:
2847        if (const_args[4]) {
2848            if ((int64_t)args[4] >= 0) {
2849                tcg_out_insn(s, RIL, ALGFI, args[0], args[4]);
2850            } else {
2851                tcg_out_insn(s, RIL, SLGFI, args[0], -args[4]);
2852            }
2853        } else {
2854            tcg_out_insn(s, RRE, ALGR, args[0], args[4]);
2855        }
2856        tcg_out_insn(s, RRE, ALCGR, args[1], args[5]);
2857        break;
2858    case INDEX_op_sub2_i64:
2859        if (const_args[4]) {
2860            if ((int64_t)args[4] >= 0) {
2861                tcg_out_insn(s, RIL, SLGFI, args[0], args[4]);
2862            } else {
2863                tcg_out_insn(s, RIL, ALGFI, args[0], -args[4]);
2864            }
2865        } else {
2866            tcg_out_insn(s, RRE, SLGR, args[0], args[4]);
2867        }
2868        tcg_out_insn(s, RRE, SLBGR, args[1], args[5]);
2869        break;
2870
2871    case INDEX_op_brcond_i64:
2872        tgen_brcond(s, TCG_TYPE_I64, args[2], args[0],
2873                    args[1], const_args[1], arg_label(args[3]));
2874        break;
2875    case INDEX_op_setcond_i64:
2876        tgen_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1],
2877                     args[2], const_args[2], false);
2878        break;
2879    case INDEX_op_negsetcond_i64:
2880        tgen_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1],
2881                     args[2], const_args[2], true);
2882        break;
2883    case INDEX_op_movcond_i64:
2884        tgen_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1],
2885                     args[2], const_args[2], args[3], const_args[3], args[4]);
2886        break;
2887
2888    OP_32_64(deposit):
2889        a0 = args[0], a1 = args[1], a2 = args[2];
2890        if (const_args[1]) {
2891            tgen_deposit(s, a0, a2, args[3], args[4], 1);
2892        } else {
2893            /* Since we can't support "0Z" as a constraint, we allow a1 in
2894               any register.  Fix things up as if a matching constraint.  */
2895            if (a0 != a1) {
2896                if (a0 == a2) {
2897                    tcg_out_mov(s, type, TCG_TMP0, a2);
2898                    a2 = TCG_TMP0;
2899                }
2900                tcg_out_mov(s, type, a0, a1);
2901            }
2902            tgen_deposit(s, a0, a2, args[3], args[4], 0);
2903        }
2904        break;
2905
2906    OP_32_64(extract):
2907        tgen_extract(s, args[0], args[1], args[2], args[3]);
2908        break;
2909    OP_32_64(sextract):
2910        tgen_sextract(s, args[0], args[1], args[2], args[3]);
2911        break;
2912
2913    case INDEX_op_ctpop_i32:
2914        tgen_ctpop(s, TCG_TYPE_I32, args[0], args[1]);
2915        break;
2916    case INDEX_op_ctpop_i64:
2917        tgen_ctpop(s, TCG_TYPE_I64, args[0], args[1]);
2918        break;
2919
2920    case INDEX_op_mb:
2921        /* The host memory model is quite strong, we simply need to
2922           serialize the instruction stream.  */
2923        if (args[0] & TCG_MO_ST_LD) {
2924            /* fast-bcr-serialization facility (45) is present */
2925            tcg_out_insn(s, RR, BCR, 14, 0);
2926        }
2927        break;
2928
2929    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
2930    case INDEX_op_exit_tb:  /* Always emitted via tcg_out_exit_tb.  */
2931    case INDEX_op_goto_tb:  /* Always emitted via tcg_out_goto_tb.  */
2932    case INDEX_op_ext_i32_i64:  /* Always emitted via tcg_reg_alloc_op.  */
2933    case INDEX_op_extu_i32_i64:
2934    case INDEX_op_extrl_i64_i32:
2935    default:
2936        g_assert_not_reached();
2937    }
2938}
2939
2940static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
2941                            TCGReg dst, TCGReg src)
2942{
2943    if (is_general_reg(src)) {
2944        /* Replicate general register into two MO_64. */
2945        tcg_out_insn(s, VRRf, VLVGP, dst, src, src);
2946        if (vece == MO_64) {
2947            return true;
2948        }
2949        src = dst;
2950    }
2951
2952    /*
2953     * Recall that the "standard" integer, within a vector, is the
2954     * rightmost element of the leftmost doubleword, a-la VLLEZ.
2955     */
2956    tcg_out_insn(s, VRIc, VREP, dst, (8 >> vece) - 1, src, vece);
2957    return true;
2958}
2959
2960static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
2961                             TCGReg dst, TCGReg base, intptr_t offset)
2962{
2963    tcg_out_vrx_mem(s, VRX_VLREP, dst, base, TCG_REG_NONE, offset, vece);
2964    return true;
2965}
2966
2967static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
2968                             TCGReg dst, int64_t val)
2969{
2970    int i, mask, msb, lsb;
2971
2972    /* Look for int16_t elements.  */
2973    if (vece <= MO_16 ||
2974        (vece == MO_32 ? (int32_t)val : val) == (int16_t)val) {
2975        tcg_out_insn(s, VRIa, VREPI, dst, val, vece);
2976        return;
2977    }
2978
2979    /* Look for bit masks.  */
2980    if (vece == MO_32) {
2981        if (risbg_mask((int32_t)val)) {
2982            /* Handle wraparound by swapping msb and lsb.  */
2983            if ((val & 0x80000001u) == 0x80000001u) {
2984                msb = 32 - ctz32(~val);
2985                lsb = clz32(~val) - 1;
2986            } else {
2987                msb = clz32(val);
2988                lsb = 31 - ctz32(val);
2989            }
2990            tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_32);
2991            return;
2992        }
2993    } else {
2994        if (risbg_mask(val)) {
2995            /* Handle wraparound by swapping msb and lsb.  */
2996            if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
2997                /* Handle wraparound by swapping msb and lsb.  */
2998                msb = 64 - ctz64(~val);
2999                lsb = clz64(~val) - 1;
3000            } else {
3001                msb = clz64(val);
3002                lsb = 63 - ctz64(val);
3003            }
3004            tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_64);
3005            return;
3006        }
3007    }
3008
3009    /* Look for all bytes 0x00 or 0xff.  */
3010    for (i = mask = 0; i < 8; i++) {
3011        uint8_t byte = val >> (i * 8);
3012        if (byte == 0xff) {
3013            mask |= 1 << i;
3014        } else if (byte != 0) {
3015            break;
3016        }
3017    }
3018    if (i == 8) {
3019        tcg_out_insn(s, VRIa, VGBM, dst, mask * 0x0101, 0);
3020        return;
3021    }
3022
3023    /* Otherwise, stuff it in the constant pool.  */
3024    tcg_out_insn(s, RIL, LARL, TCG_TMP0, 0);
3025    new_pool_label(s, val, R_390_PC32DBL, s->code_ptr - 2, 2);
3026    tcg_out_insn(s, VRX, VLREP, dst, TCG_TMP0, TCG_REG_NONE, 0, MO_64);
3027}
3028
3029static bool tcg_out_cmp_vec_noinv(TCGContext *s, unsigned vece, TCGReg a0,
3030                                  TCGReg a1, TCGReg a2, TCGCond cond)
3031{
3032    bool need_swap = false, need_inv = false;
3033
3034    switch (cond) {
3035    case TCG_COND_EQ:
3036    case TCG_COND_GT:
3037    case TCG_COND_GTU:
3038        break;
3039    case TCG_COND_NE:
3040    case TCG_COND_LE:
3041    case TCG_COND_LEU:
3042        need_inv = true;
3043        break;
3044    case TCG_COND_LT:
3045    case TCG_COND_LTU:
3046        need_swap = true;
3047        break;
3048    case TCG_COND_GE:
3049    case TCG_COND_GEU:
3050        need_swap = need_inv = true;
3051        break;
3052    default:
3053        g_assert_not_reached();
3054    }
3055
3056    if (need_inv) {
3057        cond = tcg_invert_cond(cond);
3058    }
3059    if (need_swap) {
3060        TCGReg swap = a1;
3061        a1 = a2;
3062        a2 = swap;
3063        cond = tcg_swap_cond(cond);
3064    }
3065
3066    switch (cond) {
3067    case TCG_COND_EQ:
3068        tcg_out_insn(s, VRRc, VCEQ, a0, a1, a2, vece);
3069        break;
3070    case TCG_COND_GT:
3071        tcg_out_insn(s, VRRc, VCH, a0, a1, a2, vece);
3072        break;
3073    case TCG_COND_GTU:
3074        tcg_out_insn(s, VRRc, VCHL, a0, a1, a2, vece);
3075        break;
3076    default:
3077        g_assert_not_reached();
3078    }
3079    return need_inv;
3080}
3081
3082static void tcg_out_cmp_vec(TCGContext *s, unsigned vece, TCGReg a0,
3083                            TCGReg a1, TCGReg a2, TCGCond cond)
3084{
3085    if (tcg_out_cmp_vec_noinv(s, vece, a0, a1, a2, cond)) {
3086        tcg_out_insn(s, VRRc, VNO, a0, a0, a0, 0);
3087    }
3088}
3089
3090static void tcg_out_cmpsel_vec(TCGContext *s, unsigned vece, TCGReg a0,
3091                               TCGReg c1, TCGReg c2, TCGArg v3,
3092                               int const_v3, TCGReg v4, TCGCond cond)
3093{
3094    bool inv = tcg_out_cmp_vec_noinv(s, vece, TCG_VEC_TMP0, c1, c2, cond);
3095
3096    if (!const_v3) {
3097        if (inv) {
3098            tcg_out_insn(s, VRRe, VSEL, a0, v4, v3, TCG_VEC_TMP0);
3099        } else {
3100            tcg_out_insn(s, VRRe, VSEL, a0, v3, v4, TCG_VEC_TMP0);
3101        }
3102    } else if (v3) {
3103        if (inv) {
3104            tcg_out_insn(s, VRRc, VOC, a0, v4, TCG_VEC_TMP0, 0);
3105        } else {
3106            tcg_out_insn(s, VRRc, VO, a0, v4, TCG_VEC_TMP0, 0);
3107        }
3108    } else {
3109        if (inv) {
3110            tcg_out_insn(s, VRRc, VN, a0, v4, TCG_VEC_TMP0, 0);
3111        } else {
3112            tcg_out_insn(s, VRRc, VNC, a0, v4, TCG_VEC_TMP0, 0);
3113        }
3114    }
3115}
3116
3117static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
3118                           unsigned vecl, unsigned vece,
3119                           const TCGArg args[TCG_MAX_OP_ARGS],
3120                           const int const_args[TCG_MAX_OP_ARGS])
3121{
3122    TCGType type = vecl + TCG_TYPE_V64;
3123    TCGArg a0 = args[0], a1 = args[1], a2 = args[2];
3124
3125    switch (opc) {
3126    case INDEX_op_ld_vec:
3127        tcg_out_ld(s, type, a0, a1, a2);
3128        break;
3129    case INDEX_op_st_vec:
3130        tcg_out_st(s, type, a0, a1, a2);
3131        break;
3132    case INDEX_op_dupm_vec:
3133        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
3134        break;
3135
3136    case INDEX_op_abs_vec:
3137        tcg_out_insn(s, VRRa, VLP, a0, a1, vece);
3138        break;
3139    case INDEX_op_neg_vec:
3140        tcg_out_insn(s, VRRa, VLC, a0, a1, vece);
3141        break;
3142    case INDEX_op_not_vec:
3143        tcg_out_insn(s, VRRc, VNO, a0, a1, a1, 0);
3144        break;
3145
3146    case INDEX_op_add_vec:
3147        tcg_out_insn(s, VRRc, VA, a0, a1, a2, vece);
3148        break;
3149    case INDEX_op_sub_vec:
3150        tcg_out_insn(s, VRRc, VS, a0, a1, a2, vece);
3151        break;
3152    case INDEX_op_and_vec:
3153        tcg_out_insn(s, VRRc, VN, a0, a1, a2, 0);
3154        break;
3155    case INDEX_op_andc_vec:
3156        tcg_out_insn(s, VRRc, VNC, a0, a1, a2, 0);
3157        break;
3158    case INDEX_op_mul_vec:
3159        tcg_out_insn(s, VRRc, VML, a0, a1, a2, vece);
3160        break;
3161    case INDEX_op_or_vec:
3162        tcg_out_insn(s, VRRc, VO, a0, a1, a2, 0);
3163        break;
3164    case INDEX_op_orc_vec:
3165        tcg_out_insn(s, VRRc, VOC, a0, a1, a2, 0);
3166        break;
3167    case INDEX_op_xor_vec:
3168        tcg_out_insn(s, VRRc, VX, a0, a1, a2, 0);
3169        break;
3170    case INDEX_op_nand_vec:
3171        tcg_out_insn(s, VRRc, VNN, a0, a1, a2, 0);
3172        break;
3173    case INDEX_op_nor_vec:
3174        tcg_out_insn(s, VRRc, VNO, a0, a1, a2, 0);
3175        break;
3176    case INDEX_op_eqv_vec:
3177        tcg_out_insn(s, VRRc, VNX, a0, a1, a2, 0);
3178        break;
3179
3180    case INDEX_op_shli_vec:
3181        tcg_out_insn(s, VRSa, VESL, a0, a2, TCG_REG_NONE, a1, vece);
3182        break;
3183    case INDEX_op_shri_vec:
3184        tcg_out_insn(s, VRSa, VESRL, a0, a2, TCG_REG_NONE, a1, vece);
3185        break;
3186    case INDEX_op_sari_vec:
3187        tcg_out_insn(s, VRSa, VESRA, a0, a2, TCG_REG_NONE, a1, vece);
3188        break;
3189    case INDEX_op_rotli_vec:
3190        tcg_out_insn(s, VRSa, VERLL, a0, a2, TCG_REG_NONE, a1, vece);
3191        break;
3192    case INDEX_op_shls_vec:
3193        tcg_out_insn(s, VRSa, VESL, a0, 0, a2, a1, vece);
3194        break;
3195    case INDEX_op_shrs_vec:
3196        tcg_out_insn(s, VRSa, VESRL, a0, 0, a2, a1, vece);
3197        break;
3198    case INDEX_op_sars_vec:
3199        tcg_out_insn(s, VRSa, VESRA, a0, 0, a2, a1, vece);
3200        break;
3201    case INDEX_op_rotls_vec:
3202        tcg_out_insn(s, VRSa, VERLL, a0, 0, a2, a1, vece);
3203        break;
3204    case INDEX_op_shlv_vec:
3205        tcg_out_insn(s, VRRc, VESLV, a0, a1, a2, vece);
3206        break;
3207    case INDEX_op_shrv_vec:
3208        tcg_out_insn(s, VRRc, VESRLV, a0, a1, a2, vece);
3209        break;
3210    case INDEX_op_sarv_vec:
3211        tcg_out_insn(s, VRRc, VESRAV, a0, a1, a2, vece);
3212        break;
3213    case INDEX_op_rotlv_vec:
3214        tcg_out_insn(s, VRRc, VERLLV, a0, a1, a2, vece);
3215        break;
3216
3217    case INDEX_op_smin_vec:
3218        tcg_out_insn(s, VRRc, VMN, a0, a1, a2, vece);
3219        break;
3220    case INDEX_op_smax_vec:
3221        tcg_out_insn(s, VRRc, VMX, a0, a1, a2, vece);
3222        break;
3223    case INDEX_op_umin_vec:
3224        tcg_out_insn(s, VRRc, VMNL, a0, a1, a2, vece);
3225        break;
3226    case INDEX_op_umax_vec:
3227        tcg_out_insn(s, VRRc, VMXL, a0, a1, a2, vece);
3228        break;
3229
3230    case INDEX_op_bitsel_vec:
3231        tcg_out_insn(s, VRRe, VSEL, a0, a2, args[3], a1);
3232        break;
3233
3234    case INDEX_op_cmp_vec:
3235        tcg_out_cmp_vec(s, vece, a0, a1, a2, args[3]);
3236        break;
3237    case INDEX_op_cmpsel_vec:
3238        tcg_out_cmpsel_vec(s, vece, a0, a1, a2, args[3], const_args[3],
3239                           args[4], args[5]);
3240        break;
3241
3242    case INDEX_op_s390_vuph_vec:
3243        tcg_out_insn(s, VRRa, VUPH, a0, a1, vece);
3244        break;
3245    case INDEX_op_s390_vupl_vec:
3246        tcg_out_insn(s, VRRa, VUPL, a0, a1, vece);
3247        break;
3248    case INDEX_op_s390_vpks_vec:
3249        tcg_out_insn(s, VRRc, VPKS, a0, a1, a2, vece);
3250        break;
3251
3252    case INDEX_op_mov_vec:   /* Always emitted via tcg_out_mov.  */
3253    case INDEX_op_dup_vec:   /* Always emitted via tcg_out_dup_vec.  */
3254    default:
3255        g_assert_not_reached();
3256    }
3257}
3258
3259int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
3260{
3261    switch (opc) {
3262    case INDEX_op_abs_vec:
3263    case INDEX_op_add_vec:
3264    case INDEX_op_and_vec:
3265    case INDEX_op_andc_vec:
3266    case INDEX_op_bitsel_vec:
3267    case INDEX_op_eqv_vec:
3268    case INDEX_op_nand_vec:
3269    case INDEX_op_neg_vec:
3270    case INDEX_op_nor_vec:
3271    case INDEX_op_not_vec:
3272    case INDEX_op_or_vec:
3273    case INDEX_op_orc_vec:
3274    case INDEX_op_rotli_vec:
3275    case INDEX_op_rotls_vec:
3276    case INDEX_op_rotlv_vec:
3277    case INDEX_op_sari_vec:
3278    case INDEX_op_sars_vec:
3279    case INDEX_op_sarv_vec:
3280    case INDEX_op_shli_vec:
3281    case INDEX_op_shls_vec:
3282    case INDEX_op_shlv_vec:
3283    case INDEX_op_shri_vec:
3284    case INDEX_op_shrs_vec:
3285    case INDEX_op_shrv_vec:
3286    case INDEX_op_smax_vec:
3287    case INDEX_op_smin_vec:
3288    case INDEX_op_sub_vec:
3289    case INDEX_op_umax_vec:
3290    case INDEX_op_umin_vec:
3291    case INDEX_op_xor_vec:
3292    case INDEX_op_cmp_vec:
3293    case INDEX_op_cmpsel_vec:
3294        return 1;
3295    case INDEX_op_rotrv_vec:
3296        return -1;
3297    case INDEX_op_mul_vec:
3298        return vece < MO_64;
3299    case INDEX_op_ssadd_vec:
3300    case INDEX_op_sssub_vec:
3301        return vece < MO_64 ? -1 : 0;
3302    default:
3303        return 0;
3304    }
3305}
3306
3307static void expand_vec_sat(TCGType type, unsigned vece, TCGv_vec v0,
3308                           TCGv_vec v1, TCGv_vec v2, TCGOpcode add_sub_opc)
3309{
3310    TCGv_vec h1 = tcg_temp_new_vec(type);
3311    TCGv_vec h2 = tcg_temp_new_vec(type);
3312    TCGv_vec l1 = tcg_temp_new_vec(type);
3313    TCGv_vec l2 = tcg_temp_new_vec(type);
3314
3315    tcg_debug_assert (vece < MO_64);
3316
3317    /* Unpack with sign-extension. */
3318    vec_gen_2(INDEX_op_s390_vuph_vec, type, vece,
3319              tcgv_vec_arg(h1), tcgv_vec_arg(v1));
3320    vec_gen_2(INDEX_op_s390_vuph_vec, type, vece,
3321              tcgv_vec_arg(h2), tcgv_vec_arg(v2));
3322
3323    vec_gen_2(INDEX_op_s390_vupl_vec, type, vece,
3324              tcgv_vec_arg(l1), tcgv_vec_arg(v1));
3325    vec_gen_2(INDEX_op_s390_vupl_vec, type, vece,
3326              tcgv_vec_arg(l2), tcgv_vec_arg(v2));
3327
3328    /* Arithmetic on a wider element size. */
3329    vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(h1),
3330              tcgv_vec_arg(h1), tcgv_vec_arg(h2));
3331    vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(l1),
3332              tcgv_vec_arg(l1), tcgv_vec_arg(l2));
3333
3334    /* Pack with saturation. */
3335    vec_gen_3(INDEX_op_s390_vpks_vec, type, vece + 1,
3336              tcgv_vec_arg(v0), tcgv_vec_arg(h1), tcgv_vec_arg(l1));
3337
3338    tcg_temp_free_vec(h1);
3339    tcg_temp_free_vec(h2);
3340    tcg_temp_free_vec(l1);
3341    tcg_temp_free_vec(l2);
3342}
3343
3344void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
3345                       TCGArg a0, ...)
3346{
3347    va_list va;
3348    TCGv_vec v0, v1, v2, t0;
3349
3350    va_start(va, a0);
3351    v0 = temp_tcgv_vec(arg_temp(a0));
3352    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3353    v2 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3354
3355    switch (opc) {
3356    case INDEX_op_rotrv_vec:
3357        t0 = tcg_temp_new_vec(type);
3358        tcg_gen_neg_vec(vece, t0, v2);
3359        tcg_gen_rotlv_vec(vece, v0, v1, t0);
3360        tcg_temp_free_vec(t0);
3361        break;
3362
3363    case INDEX_op_ssadd_vec:
3364        expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_add_vec);
3365        break;
3366    case INDEX_op_sssub_vec:
3367        expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_sub_vec);
3368        break;
3369
3370    default:
3371        g_assert_not_reached();
3372    }
3373    va_end(va);
3374}
3375
3376static TCGConstraintSetIndex
3377tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags)
3378{
3379    switch (op) {
3380    case INDEX_op_goto_ptr:
3381        return C_O0_I1(r);
3382
3383    case INDEX_op_ld8u_i32:
3384    case INDEX_op_ld8u_i64:
3385    case INDEX_op_ld8s_i32:
3386    case INDEX_op_ld8s_i64:
3387    case INDEX_op_ld16u_i32:
3388    case INDEX_op_ld16u_i64:
3389    case INDEX_op_ld16s_i32:
3390    case INDEX_op_ld16s_i64:
3391    case INDEX_op_ld_i32:
3392    case INDEX_op_ld32u_i64:
3393    case INDEX_op_ld32s_i64:
3394    case INDEX_op_ld_i64:
3395        return C_O1_I1(r, r);
3396
3397    case INDEX_op_st8_i32:
3398    case INDEX_op_st8_i64:
3399    case INDEX_op_st16_i32:
3400    case INDEX_op_st16_i64:
3401    case INDEX_op_st_i32:
3402    case INDEX_op_st32_i64:
3403    case INDEX_op_st_i64:
3404        return C_O0_I2(r, r);
3405
3406    case INDEX_op_setcond_i32:
3407    case INDEX_op_negsetcond_i32:
3408    case INDEX_op_setcond_i64:
3409    case INDEX_op_negsetcond_i64:
3410        return C_O1_I2(r, r, rC);
3411
3412    case INDEX_op_brcond_i32:
3413        return C_O0_I2(r, ri);
3414    case INDEX_op_brcond_i64:
3415        return C_O0_I2(r, rC);
3416
3417    case INDEX_op_bswap16_i32:
3418    case INDEX_op_bswap16_i64:
3419    case INDEX_op_bswap32_i32:
3420    case INDEX_op_bswap32_i64:
3421    case INDEX_op_bswap64_i64:
3422    case INDEX_op_ext_i32_i64:
3423    case INDEX_op_extu_i32_i64:
3424    case INDEX_op_extract_i32:
3425    case INDEX_op_extract_i64:
3426    case INDEX_op_sextract_i32:
3427    case INDEX_op_sextract_i64:
3428    case INDEX_op_ctpop_i32:
3429    case INDEX_op_ctpop_i64:
3430        return C_O1_I1(r, r);
3431
3432    case INDEX_op_qemu_ld_i32:
3433    case INDEX_op_qemu_ld_i64:
3434        return C_O1_I1(r, r);
3435    case INDEX_op_qemu_st_i64:
3436    case INDEX_op_qemu_st_i32:
3437        return C_O0_I2(r, r);
3438    case INDEX_op_qemu_ld_i128:
3439        return C_O2_I1(o, m, r);
3440    case INDEX_op_qemu_st_i128:
3441        return C_O0_I3(o, m, r);
3442
3443    case INDEX_op_deposit_i32:
3444    case INDEX_op_deposit_i64:
3445        return C_O1_I2(r, rZ, r);
3446
3447    case INDEX_op_movcond_i32:
3448        return C_O1_I4(r, r, ri, rI, r);
3449    case INDEX_op_movcond_i64:
3450        return C_O1_I4(r, r, rC, rI, r);
3451
3452    case INDEX_op_mulu2_i64:
3453        return C_O2_I2(o, m, 0, r);
3454    case INDEX_op_muls2_i64:
3455        return C_O2_I2(o, m, r, r);
3456
3457    case INDEX_op_add2_i32:
3458    case INDEX_op_sub2_i32:
3459        return C_N1_O1_I4(r, r, 0, 1, ri, r);
3460
3461    case INDEX_op_add2_i64:
3462    case INDEX_op_sub2_i64:
3463        return C_N1_O1_I4(r, r, 0, 1, rJU, r);
3464
3465    case INDEX_op_st_vec:
3466        return C_O0_I2(v, r);
3467    case INDEX_op_ld_vec:
3468    case INDEX_op_dupm_vec:
3469        return C_O1_I1(v, r);
3470    case INDEX_op_dup_vec:
3471        return C_O1_I1(v, vr);
3472    case INDEX_op_abs_vec:
3473    case INDEX_op_neg_vec:
3474    case INDEX_op_not_vec:
3475    case INDEX_op_rotli_vec:
3476    case INDEX_op_sari_vec:
3477    case INDEX_op_shli_vec:
3478    case INDEX_op_shri_vec:
3479    case INDEX_op_s390_vuph_vec:
3480    case INDEX_op_s390_vupl_vec:
3481        return C_O1_I1(v, v);
3482    case INDEX_op_add_vec:
3483    case INDEX_op_sub_vec:
3484    case INDEX_op_and_vec:
3485    case INDEX_op_andc_vec:
3486    case INDEX_op_or_vec:
3487    case INDEX_op_orc_vec:
3488    case INDEX_op_xor_vec:
3489    case INDEX_op_nand_vec:
3490    case INDEX_op_nor_vec:
3491    case INDEX_op_eqv_vec:
3492    case INDEX_op_cmp_vec:
3493    case INDEX_op_mul_vec:
3494    case INDEX_op_rotlv_vec:
3495    case INDEX_op_rotrv_vec:
3496    case INDEX_op_shlv_vec:
3497    case INDEX_op_shrv_vec:
3498    case INDEX_op_sarv_vec:
3499    case INDEX_op_smax_vec:
3500    case INDEX_op_smin_vec:
3501    case INDEX_op_umax_vec:
3502    case INDEX_op_umin_vec:
3503    case INDEX_op_s390_vpks_vec:
3504        return C_O1_I2(v, v, v);
3505    case INDEX_op_rotls_vec:
3506    case INDEX_op_shls_vec:
3507    case INDEX_op_shrs_vec:
3508    case INDEX_op_sars_vec:
3509        return C_O1_I2(v, v, r);
3510    case INDEX_op_bitsel_vec:
3511        return C_O1_I3(v, v, v, v);
3512    case INDEX_op_cmpsel_vec:
3513        return (TCG_TARGET_HAS_orc_vec
3514                ? C_O1_I4(v, v, v, vZM, v)
3515                : C_O1_I4(v, v, v, vZ, v));
3516
3517    default:
3518        return C_NotImplemented;
3519    }
3520}
3521
3522/*
3523 * Mainline glibc added HWCAP_S390_VX before it was kernel abi.
3524 * Some distros have fixed this up locally, others have not.
3525 */
3526#ifndef HWCAP_S390_VXRS
3527#define HWCAP_S390_VXRS 2048
3528#endif
3529
3530static void query_s390_facilities(void)
3531{
3532    unsigned long hwcap = qemu_getauxval(AT_HWCAP);
3533    const char *which;
3534
3535    /* Is STORE FACILITY LIST EXTENDED available?  Honestly, I believe this
3536       is present on all 64-bit systems, but let's check for it anyway.  */
3537    if (hwcap & HWCAP_S390_STFLE) {
3538        register int r0 __asm__("0") = ARRAY_SIZE(s390_facilities) - 1;
3539        register void *r1 __asm__("1") = s390_facilities;
3540
3541        /* stfle 0(%r1) */
3542        asm volatile(".word 0xb2b0,0x1000"
3543                     : "=r"(r0) : "r"(r0), "r"(r1) : "memory", "cc");
3544    }
3545
3546    /*
3547     * Use of vector registers requires os support beyond the facility bit.
3548     * If the kernel does not advertise support, disable the facility bits.
3549     * There is nothing else we currently care about in the 3rd word, so
3550     * disable VECTOR with one store.
3551     */
3552    if (!(hwcap & HWCAP_S390_VXRS)) {
3553        s390_facilities[2] = 0;
3554    }
3555
3556    /*
3557     * Minimum supported cpu revision is z196.
3558     * Check for all required facilities.
3559     * ZARCH_ACTIVE is done via preprocessor check for 64-bit.
3560     */
3561    if (!HAVE_FACILITY(LONG_DISP)) {
3562        which = "long-displacement";
3563        goto fail;
3564    }
3565    if (!HAVE_FACILITY(EXT_IMM)) {
3566        which = "extended-immediate";
3567        goto fail;
3568    }
3569    if (!HAVE_FACILITY(GEN_INST_EXT)) {
3570        which = "general-instructions-extension";
3571        goto fail;
3572    }
3573    /*
3574     * Facility 45 is a big bin that contains: distinct-operands,
3575     * fast-BCR-serialization, high-word, population-count,
3576     * interlocked-access-1, and load/store-on-condition-1
3577     */
3578    if (!HAVE_FACILITY(45)) {
3579        which = "45";
3580        goto fail;
3581    }
3582    return;
3583
3584 fail:
3585    error_report("%s: missing required facility %s", __func__, which);
3586    exit(EXIT_FAILURE);
3587}
3588
3589static void tcg_target_init(TCGContext *s)
3590{
3591    query_s390_facilities();
3592
3593    tcg_target_available_regs[TCG_TYPE_I32] = 0xffff;
3594    tcg_target_available_regs[TCG_TYPE_I64] = 0xffff;
3595    if (HAVE_FACILITY(VECTOR)) {
3596        tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull;
3597        tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull;
3598    }
3599
3600    tcg_target_call_clobber_regs = 0;
3601    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
3602    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1);
3603    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
3604    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
3605    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
3606    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
3607    /* The r6 register is technically call-saved, but it's also a parameter
3608       register, so it can get killed by setup for the qemu_st helper.  */
3609    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
3610    /* The return register can be considered call-clobbered.  */
3611    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14);
3612
3613    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0);
3614    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1);
3615    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V2);
3616    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V3);
3617    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V4);
3618    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V5);
3619    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V6);
3620    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V7);
3621    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V16);
3622    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V17);
3623    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V18);
3624    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V19);
3625    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V20);
3626    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V21);
3627    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V22);
3628    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V23);
3629    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V24);
3630    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V25);
3631    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V26);
3632    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V27);
3633    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V28);
3634    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V29);
3635    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V30);
3636    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V31);
3637
3638    s->reserved_regs = 0;
3639    tcg_regset_set_reg(s->reserved_regs, TCG_TMP0);
3640    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP0);
3641    /* XXX many insns can't be used with R0, so we better avoid it for now */
3642    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0);
3643    tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
3644}
3645
3646#define FRAME_SIZE  ((int)(TCG_TARGET_CALL_STACK_OFFSET          \
3647                           + TCG_STATIC_CALL_ARGS_SIZE           \
3648                           + CPU_TEMP_BUF_NLONGS * sizeof(long)))
3649
3650static void tcg_target_qemu_prologue(TCGContext *s)
3651{
3652    /* stmg %r6,%r15,48(%r15) (save registers) */
3653    tcg_out_insn(s, RXY, STMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15, 48);
3654
3655    /* aghi %r15,-frame_size */
3656    tcg_out_insn(s, RI, AGHI, TCG_REG_R15, -FRAME_SIZE);
3657
3658    tcg_set_frame(s, TCG_REG_CALL_STACK,
3659                  TCG_STATIC_CALL_ARGS_SIZE + TCG_TARGET_CALL_STACK_OFFSET,
3660                  CPU_TEMP_BUF_NLONGS * sizeof(long));
3661
3662    if (!tcg_use_softmmu && guest_base >= 0x80000) {
3663        tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base);
3664        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
3665    }
3666
3667    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
3668
3669    /* br %r3 (go to TB) */
3670    tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, tcg_target_call_iarg_regs[1]);
3671
3672    /*
3673     * Return path for goto_ptr. Set return value to 0, a-la exit_tb,
3674     * and fall through to the rest of the epilogue.
3675     */
3676    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
3677    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, 0);
3678
3679    /* TB epilogue */
3680    tb_ret_addr = tcg_splitwx_to_rx(s->code_ptr);
3681
3682    /* lmg %r6,%r15,fs+48(%r15) (restore registers) */
3683    tcg_out_insn(s, RXY, LMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15,
3684                 FRAME_SIZE + 48);
3685
3686    /* br %r14 (return) */
3687    tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_R14);
3688}
3689
3690static void tcg_out_tb_start(TCGContext *s)
3691{
3692    /* nothing to do */
3693}
3694
3695static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
3696{
3697    memset(p, 0x07, count * sizeof(tcg_insn_unit));
3698}
3699
3700typedef struct {
3701    DebugFrameHeader h;
3702    uint8_t fde_def_cfa[4];
3703    uint8_t fde_reg_ofs[18];
3704} DebugFrame;
3705
3706/* We're expecting a 2 byte uleb128 encoded value.  */
3707QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3708
3709#define ELF_HOST_MACHINE  EM_S390
3710
3711static const DebugFrame debug_frame = {
3712    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3713    .h.cie.id = -1,
3714    .h.cie.version = 1,
3715    .h.cie.code_align = 1,
3716    .h.cie.data_align = 8,                /* sleb128 8 */
3717    .h.cie.return_column = TCG_REG_R14,
3718
3719    /* Total FDE size does not include the "len" member.  */
3720    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
3721
3722    .fde_def_cfa = {
3723        12, TCG_REG_CALL_STACK,         /* DW_CFA_def_cfa %r15, ... */
3724        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3725        (FRAME_SIZE >> 7)
3726    },
3727    .fde_reg_ofs = {
3728        0x86, 6,                        /* DW_CFA_offset, %r6, 48 */
3729        0x87, 7,                        /* DW_CFA_offset, %r7, 56 */
3730        0x88, 8,                        /* DW_CFA_offset, %r8, 64 */
3731        0x89, 9,                        /* DW_CFA_offset, %r92, 72 */
3732        0x8a, 10,                       /* DW_CFA_offset, %r10, 80 */
3733        0x8b, 11,                       /* DW_CFA_offset, %r11, 88 */
3734        0x8c, 12,                       /* DW_CFA_offset, %r12, 96 */
3735        0x8d, 13,                       /* DW_CFA_offset, %r13, 104 */
3736        0x8e, 14,                       /* DW_CFA_offset, %r14, 112 */
3737    }
3738};
3739
3740void tcg_register_jit(const void *buf, size_t buf_size)
3741{
3742    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3743}
3744