xref: /openbmc/qemu/tcg/s390x/tcg-target.c.inc (revision 89aafcf2)
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/* We only support generating code for 64-bit mode.  */
28#if TCG_TARGET_REG_BITS != 64
29#error "unsupported code generation mode"
30#endif
31
32#include "../tcg-ldst.c.inc"
33#include "../tcg-pool.c.inc"
34#include "elf.h"
35
36#define TCG_CT_CONST_S16        (1 << 8)
37#define TCG_CT_CONST_S32        (1 << 9)
38#define TCG_CT_CONST_S33        (1 << 10)
39#define TCG_CT_CONST_ZERO       (1 << 11)
40#define TCG_CT_CONST_P32        (1 << 12)
41#define TCG_CT_CONST_INV        (1 << 13)
42#define TCG_CT_CONST_INVRISBG   (1 << 14)
43
44#define ALL_GENERAL_REGS     MAKE_64BIT_MASK(0, 16)
45#define ALL_VECTOR_REGS      MAKE_64BIT_MASK(32, 32)
46
47/* Several places within the instruction set 0 means "no register"
48   rather than TCG_REG_R0.  */
49#define TCG_REG_NONE    0
50
51/* A scratch register that may be be used throughout the backend.  */
52#define TCG_TMP0        TCG_REG_R1
53
54#ifndef CONFIG_SOFTMMU
55#define TCG_GUEST_BASE_REG TCG_REG_R13
56#endif
57
58/* All of the following instructions are prefixed with their instruction
59   format, and are defined as 8- or 16-bit quantities, even when the two
60   halves of the 16-bit quantity may appear 32 bits apart in the insn.
61   This makes it easy to copy the values from the tables in Appendix B.  */
62typedef enum S390Opcode {
63    RIL_AFI     = 0xc209,
64    RIL_AGFI    = 0xc208,
65    RIL_ALFI    = 0xc20b,
66    RIL_ALGFI   = 0xc20a,
67    RIL_BRASL   = 0xc005,
68    RIL_BRCL    = 0xc004,
69    RIL_CFI     = 0xc20d,
70    RIL_CGFI    = 0xc20c,
71    RIL_CLFI    = 0xc20f,
72    RIL_CLGFI   = 0xc20e,
73    RIL_CLRL    = 0xc60f,
74    RIL_CLGRL   = 0xc60a,
75    RIL_CRL     = 0xc60d,
76    RIL_CGRL    = 0xc608,
77    RIL_IIHF    = 0xc008,
78    RIL_IILF    = 0xc009,
79    RIL_LARL    = 0xc000,
80    RIL_LGFI    = 0xc001,
81    RIL_LGRL    = 0xc408,
82    RIL_LLIHF   = 0xc00e,
83    RIL_LLILF   = 0xc00f,
84    RIL_LRL     = 0xc40d,
85    RIL_MSFI    = 0xc201,
86    RIL_MSGFI   = 0xc200,
87    RIL_NIHF    = 0xc00a,
88    RIL_NILF    = 0xc00b,
89    RIL_OIHF    = 0xc00c,
90    RIL_OILF    = 0xc00d,
91    RIL_SLFI    = 0xc205,
92    RIL_SLGFI   = 0xc204,
93    RIL_XIHF    = 0xc006,
94    RIL_XILF    = 0xc007,
95
96    RI_AGHI     = 0xa70b,
97    RI_AHI      = 0xa70a,
98    RI_BRC      = 0xa704,
99    RI_CHI      = 0xa70e,
100    RI_CGHI     = 0xa70f,
101    RI_IIHH     = 0xa500,
102    RI_IIHL     = 0xa501,
103    RI_IILH     = 0xa502,
104    RI_IILL     = 0xa503,
105    RI_LGHI     = 0xa709,
106    RI_LLIHH    = 0xa50c,
107    RI_LLIHL    = 0xa50d,
108    RI_LLILH    = 0xa50e,
109    RI_LLILL    = 0xa50f,
110    RI_MGHI     = 0xa70d,
111    RI_MHI      = 0xa70c,
112    RI_NIHH     = 0xa504,
113    RI_NIHL     = 0xa505,
114    RI_NILH     = 0xa506,
115    RI_NILL     = 0xa507,
116    RI_OIHH     = 0xa508,
117    RI_OIHL     = 0xa509,
118    RI_OILH     = 0xa50a,
119    RI_OILL     = 0xa50b,
120    RI_TMLL     = 0xa701,
121
122    RIEb_CGRJ    = 0xec64,
123    RIEb_CLGRJ   = 0xec65,
124    RIEb_CLRJ    = 0xec77,
125    RIEb_CRJ     = 0xec76,
126
127    RIEc_CGIJ    = 0xec7c,
128    RIEc_CIJ     = 0xec7e,
129    RIEc_CLGIJ   = 0xec7d,
130    RIEc_CLIJ    = 0xec7f,
131
132    RIEf_RISBG   = 0xec55,
133
134    RIEg_LOCGHI  = 0xec46,
135
136    RRE_AGR     = 0xb908,
137    RRE_ALGR    = 0xb90a,
138    RRE_ALCR    = 0xb998,
139    RRE_ALCGR   = 0xb988,
140    RRE_ALGFR   = 0xb91a,
141    RRE_CGR     = 0xb920,
142    RRE_CLGR    = 0xb921,
143    RRE_DLGR    = 0xb987,
144    RRE_DLR     = 0xb997,
145    RRE_DSGFR   = 0xb91d,
146    RRE_DSGR    = 0xb90d,
147    RRE_FLOGR   = 0xb983,
148    RRE_LGBR    = 0xb906,
149    RRE_LCGR    = 0xb903,
150    RRE_LGFR    = 0xb914,
151    RRE_LGHR    = 0xb907,
152    RRE_LGR     = 0xb904,
153    RRE_LLGCR   = 0xb984,
154    RRE_LLGFR   = 0xb916,
155    RRE_LLGHR   = 0xb985,
156    RRE_LRVR    = 0xb91f,
157    RRE_LRVGR   = 0xb90f,
158    RRE_LTGR    = 0xb902,
159    RRE_MLGR    = 0xb986,
160    RRE_MSGR    = 0xb90c,
161    RRE_MSR     = 0xb252,
162    RRE_NGR     = 0xb980,
163    RRE_OGR     = 0xb981,
164    RRE_SGR     = 0xb909,
165    RRE_SLGR    = 0xb90b,
166    RRE_SLBR    = 0xb999,
167    RRE_SLBGR   = 0xb989,
168    RRE_XGR     = 0xb982,
169
170    RRFa_MGRK   = 0xb9ec,
171    RRFa_MSRKC  = 0xb9fd,
172    RRFa_MSGRKC = 0xb9ed,
173    RRFa_NCRK   = 0xb9f5,
174    RRFa_NCGRK  = 0xb9e5,
175    RRFa_NNRK   = 0xb974,
176    RRFa_NNGRK  = 0xb964,
177    RRFa_NORK   = 0xb976,
178    RRFa_NOGRK  = 0xb966,
179    RRFa_NRK    = 0xb9f4,
180    RRFa_NGRK   = 0xb9e4,
181    RRFa_NXRK   = 0xb977,
182    RRFa_NXGRK  = 0xb967,
183    RRFa_OCRK   = 0xb975,
184    RRFa_OCGRK  = 0xb965,
185    RRFa_ORK    = 0xb9f6,
186    RRFa_OGRK   = 0xb9e6,
187    RRFa_SRK    = 0xb9f9,
188    RRFa_SGRK   = 0xb9e9,
189    RRFa_SLRK   = 0xb9fb,
190    RRFa_SLGRK  = 0xb9eb,
191    RRFa_XRK    = 0xb9f7,
192    RRFa_XGRK   = 0xb9e7,
193
194    RRFam_SELGR = 0xb9e3,
195
196    RRFc_LOCR   = 0xb9f2,
197    RRFc_LOCGR  = 0xb9e2,
198    RRFc_POPCNT = 0xb9e1,
199
200    RR_AR       = 0x1a,
201    RR_ALR      = 0x1e,
202    RR_BASR     = 0x0d,
203    RR_BCR      = 0x07,
204    RR_CLR      = 0x15,
205    RR_CR       = 0x19,
206    RR_DR       = 0x1d,
207    RR_LCR      = 0x13,
208    RR_LR       = 0x18,
209    RR_LTR      = 0x12,
210    RR_NR       = 0x14,
211    RR_OR       = 0x16,
212    RR_SR       = 0x1b,
213    RR_SLR      = 0x1f,
214    RR_XR       = 0x17,
215
216    RSY_RLL     = 0xeb1d,
217    RSY_RLLG    = 0xeb1c,
218    RSY_SLLG    = 0xeb0d,
219    RSY_SLLK    = 0xebdf,
220    RSY_SRAG    = 0xeb0a,
221    RSY_SRAK    = 0xebdc,
222    RSY_SRLG    = 0xeb0c,
223    RSY_SRLK    = 0xebde,
224
225    RS_SLL      = 0x89,
226    RS_SRA      = 0x8a,
227    RS_SRL      = 0x88,
228
229    RXY_AG      = 0xe308,
230    RXY_AY      = 0xe35a,
231    RXY_CG      = 0xe320,
232    RXY_CLG     = 0xe321,
233    RXY_CLY     = 0xe355,
234    RXY_CY      = 0xe359,
235    RXY_LAY     = 0xe371,
236    RXY_LB      = 0xe376,
237    RXY_LG      = 0xe304,
238    RXY_LGB     = 0xe377,
239    RXY_LGF     = 0xe314,
240    RXY_LGH     = 0xe315,
241    RXY_LHY     = 0xe378,
242    RXY_LLGC    = 0xe390,
243    RXY_LLGF    = 0xe316,
244    RXY_LLGH    = 0xe391,
245    RXY_LMG     = 0xeb04,
246    RXY_LPQ     = 0xe38f,
247    RXY_LRV     = 0xe31e,
248    RXY_LRVG    = 0xe30f,
249    RXY_LRVH    = 0xe31f,
250    RXY_LY      = 0xe358,
251    RXY_NG      = 0xe380,
252    RXY_OG      = 0xe381,
253    RXY_STCY    = 0xe372,
254    RXY_STG     = 0xe324,
255    RXY_STHY    = 0xe370,
256    RXY_STMG    = 0xeb24,
257    RXY_STPQ    = 0xe38e,
258    RXY_STRV    = 0xe33e,
259    RXY_STRVG   = 0xe32f,
260    RXY_STRVH   = 0xe33f,
261    RXY_STY     = 0xe350,
262    RXY_XG      = 0xe382,
263
264    RX_A        = 0x5a,
265    RX_C        = 0x59,
266    RX_L        = 0x58,
267    RX_LA       = 0x41,
268    RX_LH       = 0x48,
269    RX_ST       = 0x50,
270    RX_STC      = 0x42,
271    RX_STH      = 0x40,
272
273    VRIa_VGBM   = 0xe744,
274    VRIa_VREPI  = 0xe745,
275    VRIb_VGM    = 0xe746,
276    VRIc_VREP   = 0xe74d,
277
278    VRRa_VLC    = 0xe7de,
279    VRRa_VLP    = 0xe7df,
280    VRRa_VLR    = 0xe756,
281    VRRc_VA     = 0xe7f3,
282    VRRc_VCEQ   = 0xe7f8,   /* we leave the m5 cs field 0 */
283    VRRc_VCH    = 0xe7fb,   /* " */
284    VRRc_VCHL   = 0xe7f9,   /* " */
285    VRRc_VERLLV = 0xe773,
286    VRRc_VESLV  = 0xe770,
287    VRRc_VESRAV = 0xe77a,
288    VRRc_VESRLV = 0xe778,
289    VRRc_VML    = 0xe7a2,
290    VRRc_VMN    = 0xe7fe,
291    VRRc_VMNL   = 0xe7fc,
292    VRRc_VMX    = 0xe7ff,
293    VRRc_VMXL   = 0xe7fd,
294    VRRc_VN     = 0xe768,
295    VRRc_VNC    = 0xe769,
296    VRRc_VNN    = 0xe76e,
297    VRRc_VNO    = 0xe76b,
298    VRRc_VNX    = 0xe76c,
299    VRRc_VO     = 0xe76a,
300    VRRc_VOC    = 0xe76f,
301    VRRc_VPKS   = 0xe797,   /* we leave the m5 cs field 0 */
302    VRRc_VS     = 0xe7f7,
303    VRRa_VUPH   = 0xe7d7,
304    VRRa_VUPL   = 0xe7d6,
305    VRRc_VX     = 0xe76d,
306    VRRe_VSEL   = 0xe78d,
307    VRRf_VLVGP  = 0xe762,
308
309    VRSa_VERLL  = 0xe733,
310    VRSa_VESL   = 0xe730,
311    VRSa_VESRA  = 0xe73a,
312    VRSa_VESRL  = 0xe738,
313    VRSb_VLVG   = 0xe722,
314    VRSc_VLGV   = 0xe721,
315
316    VRX_VL      = 0xe706,
317    VRX_VLLEZ   = 0xe704,
318    VRX_VLREP   = 0xe705,
319    VRX_VST     = 0xe70e,
320    VRX_VSTEF   = 0xe70b,
321    VRX_VSTEG   = 0xe70a,
322
323    NOP         = 0x0707,
324} S390Opcode;
325
326#ifdef CONFIG_DEBUG_TCG
327static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
328    "%r0",  "%r1",  "%r2",  "%r3",  "%r4",  "%r5",  "%r6",  "%r7",
329    "%r8",  "%r9",  "%r10", "%r11", "%r12", "%r13", "%r14", "%r15",
330    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
331    "%v0",  "%v1",  "%v2",  "%v3",  "%v4",  "%v5",  "%v6",  "%v7",
332    "%v8",  "%v9",  "%v10", "%v11", "%v12", "%v13", "%v14", "%v15",
333    "%v16", "%v17", "%v18", "%v19", "%v20", "%v21", "%v22", "%v23",
334    "%v24", "%v25", "%v26", "%v27", "%v28", "%v29", "%v30", "%v31",
335};
336#endif
337
338/* Since R6 is a potential argument register, choose it last of the
339   call-saved registers.  Likewise prefer the call-clobbered registers
340   in reverse order to maximize the chance of avoiding the arguments.  */
341static const int tcg_target_reg_alloc_order[] = {
342    /* Call saved registers.  */
343    TCG_REG_R13,
344    TCG_REG_R12,
345    TCG_REG_R11,
346    TCG_REG_R10,
347    TCG_REG_R9,
348    TCG_REG_R8,
349    TCG_REG_R7,
350    TCG_REG_R6,
351    /* Call clobbered registers.  */
352    TCG_REG_R14,
353    TCG_REG_R0,
354    TCG_REG_R1,
355    /* Argument registers, in reverse order of allocation.  */
356    TCG_REG_R5,
357    TCG_REG_R4,
358    TCG_REG_R3,
359    TCG_REG_R2,
360
361    /* V8-V15 are call saved, and omitted. */
362    TCG_REG_V0,
363    TCG_REG_V1,
364    TCG_REG_V2,
365    TCG_REG_V3,
366    TCG_REG_V4,
367    TCG_REG_V5,
368    TCG_REG_V6,
369    TCG_REG_V7,
370    TCG_REG_V16,
371    TCG_REG_V17,
372    TCG_REG_V18,
373    TCG_REG_V19,
374    TCG_REG_V20,
375    TCG_REG_V21,
376    TCG_REG_V22,
377    TCG_REG_V23,
378    TCG_REG_V24,
379    TCG_REG_V25,
380    TCG_REG_V26,
381    TCG_REG_V27,
382    TCG_REG_V28,
383    TCG_REG_V29,
384    TCG_REG_V30,
385    TCG_REG_V31,
386};
387
388static const int tcg_target_call_iarg_regs[] = {
389    TCG_REG_R2,
390    TCG_REG_R3,
391    TCG_REG_R4,
392    TCG_REG_R5,
393    TCG_REG_R6,
394};
395
396static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
397{
398    tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
399    tcg_debug_assert(slot == 0);
400    return TCG_REG_R2;
401}
402
403#define S390_CC_EQ      8
404#define S390_CC_LT      4
405#define S390_CC_GT      2
406#define S390_CC_OV      1
407#define S390_CC_NE      (S390_CC_LT | S390_CC_GT)
408#define S390_CC_LE      (S390_CC_LT | S390_CC_EQ)
409#define S390_CC_GE      (S390_CC_GT | S390_CC_EQ)
410#define S390_CC_NEVER   0
411#define S390_CC_ALWAYS  15
412
413/* Condition codes that result from a COMPARE and COMPARE LOGICAL.  */
414static const uint8_t tcg_cond_to_s390_cond[] = {
415    [TCG_COND_EQ]  = S390_CC_EQ,
416    [TCG_COND_NE]  = S390_CC_NE,
417    [TCG_COND_LT]  = S390_CC_LT,
418    [TCG_COND_LE]  = S390_CC_LE,
419    [TCG_COND_GT]  = S390_CC_GT,
420    [TCG_COND_GE]  = S390_CC_GE,
421    [TCG_COND_LTU] = S390_CC_LT,
422    [TCG_COND_LEU] = S390_CC_LE,
423    [TCG_COND_GTU] = S390_CC_GT,
424    [TCG_COND_GEU] = S390_CC_GE,
425};
426
427/* Condition codes that result from a LOAD AND TEST.  Here, we have no
428   unsigned instruction variation, however since the test is vs zero we
429   can re-map the outcomes appropriately.  */
430static const uint8_t tcg_cond_to_ltr_cond[] = {
431    [TCG_COND_EQ]  = S390_CC_EQ,
432    [TCG_COND_NE]  = S390_CC_NE,
433    [TCG_COND_LT]  = S390_CC_LT,
434    [TCG_COND_LE]  = S390_CC_LE,
435    [TCG_COND_GT]  = S390_CC_GT,
436    [TCG_COND_GE]  = S390_CC_GE,
437    [TCG_COND_LTU] = S390_CC_NEVER,
438    [TCG_COND_LEU] = S390_CC_EQ,
439    [TCG_COND_GTU] = S390_CC_NE,
440    [TCG_COND_GEU] = S390_CC_ALWAYS,
441};
442
443static const tcg_insn_unit *tb_ret_addr;
444uint64_t s390_facilities[3];
445
446static inline bool is_general_reg(TCGReg r)
447{
448    return r <= TCG_REG_R15;
449}
450
451static inline bool is_vector_reg(TCGReg r)
452{
453    return r >= TCG_REG_V0 && r <= TCG_REG_V31;
454}
455
456static bool patch_reloc(tcg_insn_unit *src_rw, int type,
457                        intptr_t value, intptr_t addend)
458{
459    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
460    intptr_t pcrel2;
461    uint32_t old;
462
463    value += addend;
464    pcrel2 = (tcg_insn_unit *)value - src_rx;
465
466    switch (type) {
467    case R_390_PC16DBL:
468        if (pcrel2 == (int16_t)pcrel2) {
469            tcg_patch16(src_rw, pcrel2);
470            return true;
471        }
472        break;
473    case R_390_PC32DBL:
474        if (pcrel2 == (int32_t)pcrel2) {
475            tcg_patch32(src_rw, pcrel2);
476            return true;
477        }
478        break;
479    case R_390_20:
480        if (value == sextract64(value, 0, 20)) {
481            old = *(uint32_t *)src_rw & 0xf00000ff;
482            old |= ((value & 0xfff) << 16) | ((value & 0xff000) >> 4);
483            tcg_patch32(src_rw, old);
484            return true;
485        }
486        break;
487    default:
488        g_assert_not_reached();
489    }
490    return false;
491}
492
493static int is_const_p16(uint64_t val)
494{
495    for (int i = 0; i < 4; ++i) {
496        uint64_t mask = 0xffffull << (i * 16);
497        if ((val & ~mask) == 0) {
498            return i;
499        }
500    }
501    return -1;
502}
503
504static int is_const_p32(uint64_t val)
505{
506    if ((val & 0xffffffff00000000ull) == 0) {
507        return 0;
508    }
509    if ((val & 0x00000000ffffffffull) == 0) {
510        return 1;
511    }
512    return -1;
513}
514
515/*
516 * Accept bit patterns like these:
517 *  0....01....1
518 *  1....10....0
519 *  1..10..01..1
520 *  0..01..10..0
521 * Copied from gcc sources.
522 */
523static bool risbg_mask(uint64_t c)
524{
525    uint64_t lsb;
526    /* We don't change the number of transitions by inverting,
527       so make sure we start with the LSB zero.  */
528    if (c & 1) {
529        c = ~c;
530    }
531    /* Reject all zeros or all ones.  */
532    if (c == 0) {
533        return false;
534    }
535    /* Find the first transition.  */
536    lsb = c & -c;
537    /* Invert to look for a second transition.  */
538    c = ~c;
539    /* Erase the first transition.  */
540    c &= -lsb;
541    /* Find the second transition, if any.  */
542    lsb = c & -c;
543    /* Match if all the bits are 1's, or if c is zero.  */
544    return c == -lsb;
545}
546
547/* Test if a constant matches the constraint. */
548static bool tcg_target_const_match(int64_t val, TCGType type, int ct)
549{
550    if (ct & TCG_CT_CONST) {
551        return 1;
552    }
553
554    if (type == TCG_TYPE_I32) {
555        val = (int32_t)val;
556    }
557
558    /* The following are mutually exclusive.  */
559    if (ct & TCG_CT_CONST_S16) {
560        return val == (int16_t)val;
561    } else if (ct & TCG_CT_CONST_S32) {
562        return val == (int32_t)val;
563    } else if (ct & TCG_CT_CONST_S33) {
564        return val >= -0xffffffffll && val <= 0xffffffffll;
565    } else if (ct & TCG_CT_CONST_ZERO) {
566        return val == 0;
567    }
568
569    if (ct & TCG_CT_CONST_INV) {
570        val = ~val;
571    }
572    /*
573     * Note that is_const_p16 is a subset of is_const_p32,
574     * so we don't need both constraints.
575     */
576    if ((ct & TCG_CT_CONST_P32) && is_const_p32(val) >= 0) {
577        return true;
578    }
579    if ((ct & TCG_CT_CONST_INVRISBG) && risbg_mask(~val)) {
580        return true;
581    }
582
583    return 0;
584}
585
586/* Emit instructions according to the given instruction format.  */
587
588static void tcg_out_insn_RR(TCGContext *s, S390Opcode op, TCGReg r1, TCGReg r2)
589{
590    tcg_out16(s, (op << 8) | (r1 << 4) | r2);
591}
592
593static void tcg_out_insn_RRE(TCGContext *s, S390Opcode op,
594                             TCGReg r1, TCGReg r2)
595{
596    tcg_out32(s, (op << 16) | (r1 << 4) | r2);
597}
598
599/* RRF-a without the m4 field */
600static void tcg_out_insn_RRFa(TCGContext *s, S390Opcode op,
601                              TCGReg r1, TCGReg r2, TCGReg r3)
602{
603    tcg_out32(s, (op << 16) | (r3 << 12) | (r1 << 4) | r2);
604}
605
606/* RRF-a with the m4 field */
607static void tcg_out_insn_RRFam(TCGContext *s, S390Opcode op,
608                               TCGReg r1, TCGReg r2, TCGReg r3, int m4)
609{
610    tcg_out32(s, (op << 16) | (r3 << 12) | (m4 << 8) | (r1 << 4) | r2);
611}
612
613static void tcg_out_insn_RRFc(TCGContext *s, S390Opcode op,
614                              TCGReg r1, TCGReg r2, int m3)
615{
616    tcg_out32(s, (op << 16) | (m3 << 12) | (r1 << 4) | r2);
617}
618
619static void tcg_out_insn_RI(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
620{
621    tcg_out32(s, (op << 16) | (r1 << 20) | (i2 & 0xffff));
622}
623
624static void tcg_out_insn_RIEg(TCGContext *s, S390Opcode op, TCGReg r1,
625                             int i2, int m3)
626{
627    tcg_out16(s, (op & 0xff00) | (r1 << 4) | m3);
628    tcg_out32(s, (i2 << 16) | (op & 0xff));
629}
630
631static void tcg_out_insn_RIL(TCGContext *s, S390Opcode op, TCGReg r1, int i2)
632{
633    tcg_out16(s, op | (r1 << 4));
634    tcg_out32(s, i2);
635}
636
637static void tcg_out_insn_RS(TCGContext *s, S390Opcode op, TCGReg r1,
638                            TCGReg b2, TCGReg r3, int disp)
639{
640    tcg_out32(s, (op << 24) | (r1 << 20) | (r3 << 16) | (b2 << 12)
641              | (disp & 0xfff));
642}
643
644static void tcg_out_insn_RSY(TCGContext *s, S390Opcode op, TCGReg r1,
645                             TCGReg b2, TCGReg r3, int disp)
646{
647    tcg_out16(s, (op & 0xff00) | (r1 << 4) | r3);
648    tcg_out32(s, (op & 0xff) | (b2 << 28)
649              | ((disp & 0xfff) << 16) | ((disp & 0xff000) >> 4));
650}
651
652#define tcg_out_insn_RX   tcg_out_insn_RS
653#define tcg_out_insn_RXY  tcg_out_insn_RSY
654
655static int RXB(TCGReg v1, TCGReg v2, TCGReg v3, TCGReg v4)
656{
657    /*
658     * Shift bit 4 of each regno to its corresponding bit of RXB.
659     * RXB itself begins at bit 8 of the instruction so 8 - 4 = 4
660     * is the left-shift of the 4th operand.
661     */
662    return ((v1 & 0x10) << (4 + 3))
663         | ((v2 & 0x10) << (4 + 2))
664         | ((v3 & 0x10) << (4 + 1))
665         | ((v4 & 0x10) << (4 + 0));
666}
667
668static void tcg_out_insn_VRIa(TCGContext *s, S390Opcode op,
669                              TCGReg v1, uint16_t i2, int m3)
670{
671    tcg_debug_assert(is_vector_reg(v1));
672    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4));
673    tcg_out16(s, i2);
674    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m3 << 12));
675}
676
677static void tcg_out_insn_VRIb(TCGContext *s, S390Opcode op,
678                              TCGReg v1, uint8_t i2, uint8_t i3, int m4)
679{
680    tcg_debug_assert(is_vector_reg(v1));
681    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4));
682    tcg_out16(s, (i2 << 8) | (i3 & 0xff));
683    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m4 << 12));
684}
685
686static void tcg_out_insn_VRIc(TCGContext *s, S390Opcode op,
687                              TCGReg v1, uint16_t i2, TCGReg v3, int m4)
688{
689    tcg_debug_assert(is_vector_reg(v1));
690    tcg_debug_assert(is_vector_reg(v3));
691    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v3 & 0xf));
692    tcg_out16(s, i2);
693    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, v3, 0) | (m4 << 12));
694}
695
696static void tcg_out_insn_VRRa(TCGContext *s, S390Opcode op,
697                              TCGReg v1, TCGReg v2, int m3)
698{
699    tcg_debug_assert(is_vector_reg(v1));
700    tcg_debug_assert(is_vector_reg(v2));
701    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
702    tcg_out32(s, (op & 0x00ff) | RXB(v1, v2, 0, 0) | (m3 << 12));
703}
704
705static void tcg_out_insn_VRRc(TCGContext *s, S390Opcode op,
706                              TCGReg v1, TCGReg v2, TCGReg v3, int m4)
707{
708    tcg_debug_assert(is_vector_reg(v1));
709    tcg_debug_assert(is_vector_reg(v2));
710    tcg_debug_assert(is_vector_reg(v3));
711    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
712    tcg_out16(s, v3 << 12);
713    tcg_out16(s, (op & 0x00ff) | RXB(v1, v2, v3, 0) | (m4 << 12));
714}
715
716static void tcg_out_insn_VRRe(TCGContext *s, S390Opcode op,
717                              TCGReg v1, TCGReg v2, TCGReg v3, TCGReg v4)
718{
719    tcg_debug_assert(is_vector_reg(v1));
720    tcg_debug_assert(is_vector_reg(v2));
721    tcg_debug_assert(is_vector_reg(v3));
722    tcg_debug_assert(is_vector_reg(v4));
723    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v2 & 0xf));
724    tcg_out16(s, v3 << 12);
725    tcg_out16(s, (op & 0x00ff) | RXB(v1, v2, v3, v4) | (v4 << 12));
726}
727
728static void tcg_out_insn_VRRf(TCGContext *s, S390Opcode op,
729                              TCGReg v1, TCGReg r2, TCGReg r3)
730{
731    tcg_debug_assert(is_vector_reg(v1));
732    tcg_debug_assert(is_general_reg(r2));
733    tcg_debug_assert(is_general_reg(r3));
734    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | r2);
735    tcg_out16(s, r3 << 12);
736    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0));
737}
738
739static void tcg_out_insn_VRSa(TCGContext *s, S390Opcode op, TCGReg v1,
740                              intptr_t d2, TCGReg b2, TCGReg v3, int m4)
741{
742    tcg_debug_assert(is_vector_reg(v1));
743    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
744    tcg_debug_assert(is_general_reg(b2));
745    tcg_debug_assert(is_vector_reg(v3));
746    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | (v3 & 0xf));
747    tcg_out16(s, b2 << 12 | d2);
748    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, v3, 0) | (m4 << 12));
749}
750
751static void tcg_out_insn_VRSb(TCGContext *s, S390Opcode op, TCGReg v1,
752                              intptr_t d2, TCGReg b2, TCGReg r3, int m4)
753{
754    tcg_debug_assert(is_vector_reg(v1));
755    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
756    tcg_debug_assert(is_general_reg(b2));
757    tcg_debug_assert(is_general_reg(r3));
758    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | r3);
759    tcg_out16(s, b2 << 12 | d2);
760    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m4 << 12));
761}
762
763static void tcg_out_insn_VRSc(TCGContext *s, S390Opcode op, TCGReg r1,
764                              intptr_t d2, TCGReg b2, TCGReg v3, int m4)
765{
766    tcg_debug_assert(is_general_reg(r1));
767    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
768    tcg_debug_assert(is_general_reg(b2));
769    tcg_debug_assert(is_vector_reg(v3));
770    tcg_out16(s, (op & 0xff00) | (r1 << 4) | (v3 & 0xf));
771    tcg_out16(s, b2 << 12 | d2);
772    tcg_out16(s, (op & 0x00ff) | RXB(0, 0, v3, 0) | (m4 << 12));
773}
774
775static void tcg_out_insn_VRX(TCGContext *s, S390Opcode op, TCGReg v1,
776                             TCGReg b2, TCGReg x2, intptr_t d2, int m3)
777{
778    tcg_debug_assert(is_vector_reg(v1));
779    tcg_debug_assert(d2 >= 0 && d2 <= 0xfff);
780    tcg_debug_assert(is_general_reg(x2));
781    tcg_debug_assert(is_general_reg(b2));
782    tcg_out16(s, (op & 0xff00) | ((v1 & 0xf) << 4) | x2);
783    tcg_out16(s, (b2 << 12) | d2);
784    tcg_out16(s, (op & 0x00ff) | RXB(v1, 0, 0, 0) | (m3 << 12));
785}
786
787/* Emit an opcode with "type-checking" of the format.  */
788#define tcg_out_insn(S, FMT, OP, ...) \
789    glue(tcg_out_insn_,FMT)(S, glue(glue(FMT,_),OP), ## __VA_ARGS__)
790
791
792/* emit 64-bit shifts */
793static void tcg_out_sh64(TCGContext* s, S390Opcode op, TCGReg dest,
794                         TCGReg src, TCGReg sh_reg, int sh_imm)
795{
796    tcg_out_insn_RSY(s, op, dest, sh_reg, src, sh_imm);
797}
798
799/* emit 32-bit shifts */
800static void tcg_out_sh32(TCGContext* s, S390Opcode op, TCGReg dest,
801                         TCGReg sh_reg, int sh_imm)
802{
803    tcg_out_insn_RS(s, op, dest, sh_reg, 0, sh_imm);
804}
805
806static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg dst, TCGReg src)
807{
808    if (src == dst) {
809        return true;
810    }
811    switch (type) {
812    case TCG_TYPE_I32:
813        if (likely(is_general_reg(dst) && is_general_reg(src))) {
814            tcg_out_insn(s, RR, LR, dst, src);
815            break;
816        }
817        /* fallthru */
818
819    case TCG_TYPE_I64:
820        if (likely(is_general_reg(dst))) {
821            if (likely(is_general_reg(src))) {
822                tcg_out_insn(s, RRE, LGR, dst, src);
823            } else {
824                tcg_out_insn(s, VRSc, VLGV, dst, 0, 0, src, 3);
825            }
826            break;
827        } else if (is_general_reg(src)) {
828            tcg_out_insn(s, VRSb, VLVG, dst, 0, 0, src, 3);
829            break;
830        }
831        /* fallthru */
832
833    case TCG_TYPE_V64:
834    case TCG_TYPE_V128:
835        tcg_out_insn(s, VRRa, VLR, dst, src, 0);
836        break;
837
838    default:
839        g_assert_not_reached();
840    }
841    return true;
842}
843
844static const S390Opcode li_insns[4] = {
845    RI_LLILL, RI_LLILH, RI_LLIHL, RI_LLIHH
846};
847static const S390Opcode oi_insns[4] = {
848    RI_OILL, RI_OILH, RI_OIHL, RI_OIHH
849};
850static const S390Opcode lif_insns[2] = {
851    RIL_LLILF, RIL_LLIHF,
852};
853
854/* load a register with an immediate value */
855static void tcg_out_movi(TCGContext *s, TCGType type,
856                         TCGReg ret, tcg_target_long sval)
857{
858    tcg_target_ulong uval = sval;
859    ptrdiff_t pc_off;
860    int i;
861
862    if (type == TCG_TYPE_I32) {
863        uval = (uint32_t)sval;
864        sval = (int32_t)sval;
865    }
866
867    /* Try all 32-bit insns that can load it in one go.  */
868    if (sval >= -0x8000 && sval < 0x8000) {
869        tcg_out_insn(s, RI, LGHI, ret, sval);
870        return;
871    }
872
873    i = is_const_p16(uval);
874    if (i >= 0) {
875        tcg_out_insn_RI(s, li_insns[i], ret, uval >> (i * 16));
876        return;
877    }
878
879    /* Try all 48-bit insns that can load it in one go.  */
880    if (sval == (int32_t)sval) {
881        tcg_out_insn(s, RIL, LGFI, ret, sval);
882        return;
883    }
884
885    i = is_const_p32(uval);
886    if (i >= 0) {
887        tcg_out_insn_RIL(s, lif_insns[i], ret, uval >> (i * 32));
888        return;
889    }
890
891    /* Try for PC-relative address load.  For odd addresses, add one. */
892    pc_off = tcg_pcrel_diff(s, (void *)sval) >> 1;
893    if (pc_off == (int32_t)pc_off) {
894        tcg_out_insn(s, RIL, LARL, ret, pc_off);
895        if (sval & 1) {
896            tcg_out_insn(s, RI, AGHI, ret, 1);
897        }
898        return;
899    }
900
901    /* Otherwise, load it by parts. */
902    i = is_const_p16((uint32_t)uval);
903    if (i >= 0) {
904        tcg_out_insn_RI(s, li_insns[i], ret, uval >> (i * 16));
905    } else {
906        tcg_out_insn(s, RIL, LLILF, ret, uval);
907    }
908    uval >>= 32;
909    i = is_const_p16(uval);
910    if (i >= 0) {
911        tcg_out_insn_RI(s, oi_insns[i + 2], ret, uval >> (i * 16));
912    } else {
913        tcg_out_insn(s, RIL, OIHF, ret, uval);
914    }
915}
916
917/* Emit a load/store type instruction.  Inputs are:
918   DATA:     The register to be loaded or stored.
919   BASE+OFS: The effective address.
920   OPC_RX:   If the operation has an RX format opcode (e.g. STC), otherwise 0.
921   OPC_RXY:  The RXY format opcode for the operation (e.g. STCY).  */
922
923static void tcg_out_mem(TCGContext *s, S390Opcode opc_rx, S390Opcode opc_rxy,
924                        TCGReg data, TCGReg base, TCGReg index,
925                        tcg_target_long ofs)
926{
927    if (ofs < -0x80000 || ofs >= 0x80000) {
928        /* Combine the low 20 bits of the offset with the actual load insn;
929           the high 44 bits must come from an immediate load.  */
930        tcg_target_long low = ((ofs & 0xfffff) ^ 0x80000) - 0x80000;
931        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs - low);
932        ofs = low;
933
934        /* If we were already given an index register, add it in.  */
935        if (index != TCG_REG_NONE) {
936            tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
937        }
938        index = TCG_TMP0;
939    }
940
941    if (opc_rx && ofs >= 0 && ofs < 0x1000) {
942        tcg_out_insn_RX(s, opc_rx, data, base, index, ofs);
943    } else {
944        tcg_out_insn_RXY(s, opc_rxy, data, base, index, ofs);
945    }
946}
947
948static void tcg_out_vrx_mem(TCGContext *s, S390Opcode opc_vrx,
949                            TCGReg data, TCGReg base, TCGReg index,
950                            tcg_target_long ofs, int m3)
951{
952    if (ofs < 0 || ofs >= 0x1000) {
953        if (ofs >= -0x80000 && ofs < 0x80000) {
954            tcg_out_insn(s, RXY, LAY, TCG_TMP0, base, index, ofs);
955            base = TCG_TMP0;
956            index = TCG_REG_NONE;
957            ofs = 0;
958        } else {
959            tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, ofs);
960            if (index != TCG_REG_NONE) {
961                tcg_out_insn(s, RRE, AGR, TCG_TMP0, index);
962            }
963            index = TCG_TMP0;
964            ofs = 0;
965        }
966    }
967    tcg_out_insn_VRX(s, opc_vrx, data, base, index, ofs, m3);
968}
969
970/* load data without address translation or endianness conversion */
971static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg data,
972                       TCGReg base, intptr_t ofs)
973{
974    switch (type) {
975    case TCG_TYPE_I32:
976        if (likely(is_general_reg(data))) {
977            tcg_out_mem(s, RX_L, RXY_LY, data, base, TCG_REG_NONE, ofs);
978            break;
979        }
980        tcg_out_vrx_mem(s, VRX_VLLEZ, data, base, TCG_REG_NONE, ofs, MO_32);
981        break;
982
983    case TCG_TYPE_I64:
984        if (likely(is_general_reg(data))) {
985            tcg_out_mem(s, 0, RXY_LG, data, base, TCG_REG_NONE, ofs);
986            break;
987        }
988        /* fallthru */
989
990    case TCG_TYPE_V64:
991        tcg_out_vrx_mem(s, VRX_VLLEZ, data, base, TCG_REG_NONE, ofs, MO_64);
992        break;
993
994    case TCG_TYPE_V128:
995        /* Hint quadword aligned.  */
996        tcg_out_vrx_mem(s, VRX_VL, data, base, TCG_REG_NONE, ofs, 4);
997        break;
998
999    default:
1000        g_assert_not_reached();
1001    }
1002}
1003
1004static void tcg_out_st(TCGContext *s, TCGType type, TCGReg data,
1005                       TCGReg base, intptr_t ofs)
1006{
1007    switch (type) {
1008    case TCG_TYPE_I32:
1009        if (likely(is_general_reg(data))) {
1010            tcg_out_mem(s, RX_ST, RXY_STY, data, base, TCG_REG_NONE, ofs);
1011        } else {
1012            tcg_out_vrx_mem(s, VRX_VSTEF, data, base, TCG_REG_NONE, ofs, 1);
1013        }
1014        break;
1015
1016    case TCG_TYPE_I64:
1017        if (likely(is_general_reg(data))) {
1018            tcg_out_mem(s, 0, RXY_STG, data, base, TCG_REG_NONE, ofs);
1019            break;
1020        }
1021        /* fallthru */
1022
1023    case TCG_TYPE_V64:
1024        tcg_out_vrx_mem(s, VRX_VSTEG, data, base, TCG_REG_NONE, ofs, 0);
1025        break;
1026
1027    case TCG_TYPE_V128:
1028        /* Hint quadword aligned.  */
1029        tcg_out_vrx_mem(s, VRX_VST, data, base, TCG_REG_NONE, ofs, 4);
1030        break;
1031
1032    default:
1033        g_assert_not_reached();
1034    }
1035}
1036
1037static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
1038                               TCGReg base, intptr_t ofs)
1039{
1040    return false;
1041}
1042
1043static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
1044{
1045    return false;
1046}
1047
1048static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
1049                             tcg_target_long imm)
1050{
1051    /* This function is only used for passing structs by reference. */
1052    tcg_out_mem(s, RX_LA, RXY_LAY, rd, rs, TCG_REG_NONE, imm);
1053}
1054
1055static inline void tcg_out_risbg(TCGContext *s, TCGReg dest, TCGReg src,
1056                                 int msb, int lsb, int ofs, int z)
1057{
1058    /* Format RIE-f */
1059    tcg_out16(s, (RIEf_RISBG & 0xff00) | (dest << 4) | src);
1060    tcg_out16(s, (msb << 8) | (z << 7) | lsb);
1061    tcg_out16(s, (ofs << 8) | (RIEf_RISBG & 0xff));
1062}
1063
1064static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1065{
1066    tcg_out_insn(s, RRE, LGBR, dest, src);
1067}
1068
1069static void tcg_out_ext8u(TCGContext *s, TCGReg dest, TCGReg src)
1070{
1071    tcg_out_insn(s, RRE, LLGCR, dest, src);
1072}
1073
1074static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1075{
1076    tcg_out_insn(s, RRE, LGHR, dest, src);
1077}
1078
1079static void tcg_out_ext16u(TCGContext *s, TCGReg dest, TCGReg src)
1080{
1081    tcg_out_insn(s, RRE, LLGHR, dest, src);
1082}
1083
1084static void tcg_out_ext32s(TCGContext *s, TCGReg dest, TCGReg src)
1085{
1086    tcg_out_insn(s, RRE, LGFR, dest, src);
1087}
1088
1089static void tcg_out_ext32u(TCGContext *s, TCGReg dest, TCGReg src)
1090{
1091    tcg_out_insn(s, RRE, LLGFR, dest, src);
1092}
1093
1094static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg dest, TCGReg src)
1095{
1096    tcg_out_ext32s(s, dest, src);
1097}
1098
1099static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg dest, TCGReg src)
1100{
1101    tcg_out_ext32u(s, dest, src);
1102}
1103
1104static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg dest, TCGReg src)
1105{
1106    tcg_out_mov(s, TCG_TYPE_I32, dest, src);
1107}
1108
1109static void tgen_andi_risbg(TCGContext *s, TCGReg out, TCGReg in, uint64_t val)
1110{
1111    int msb, lsb;
1112    if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
1113        /* Achieve wraparound by swapping msb and lsb.  */
1114        msb = 64 - ctz64(~val);
1115        lsb = clz64(~val) - 1;
1116    } else {
1117        msb = clz64(val);
1118        lsb = 63 - ctz64(val);
1119    }
1120    tcg_out_risbg(s, out, in, msb, lsb, 0, 1);
1121}
1122
1123static void tgen_andi(TCGContext *s, TCGType type, TCGReg dest, uint64_t val)
1124{
1125    static const S390Opcode ni_insns[4] = {
1126        RI_NILL, RI_NILH, RI_NIHL, RI_NIHH
1127    };
1128    static const S390Opcode nif_insns[2] = {
1129        RIL_NILF, RIL_NIHF
1130    };
1131    uint64_t valid = (type == TCG_TYPE_I32 ? 0xffffffffull : -1ull);
1132    int i;
1133
1134    /* Look for the zero-extensions.  */
1135    if ((val & valid) == 0xffffffff) {
1136        tcg_out_ext32u(s, dest, dest);
1137        return;
1138    }
1139    if ((val & valid) == 0xff) {
1140        tcg_out_ext8u(s, dest, dest);
1141        return;
1142    }
1143    if ((val & valid) == 0xffff) {
1144        tcg_out_ext16u(s, dest, dest);
1145        return;
1146    }
1147
1148    i = is_const_p16(~val & valid);
1149    if (i >= 0) {
1150        tcg_out_insn_RI(s, ni_insns[i], dest, val >> (i * 16));
1151        return;
1152    }
1153
1154    i = is_const_p32(~val & valid);
1155    tcg_debug_assert(i == 0 || type != TCG_TYPE_I32);
1156    if (i >= 0) {
1157        tcg_out_insn_RIL(s, nif_insns[i], dest, val >> (i * 32));
1158        return;
1159    }
1160
1161    if (risbg_mask(val)) {
1162        tgen_andi_risbg(s, dest, dest, val);
1163        return;
1164    }
1165
1166    g_assert_not_reached();
1167}
1168
1169static void tgen_ori(TCGContext *s, TCGReg dest, uint64_t val)
1170{
1171    static const S390Opcode oif_insns[2] = {
1172        RIL_OILF, RIL_OIHF
1173    };
1174
1175    int i;
1176
1177    i = is_const_p16(val);
1178    if (i >= 0) {
1179        tcg_out_insn_RI(s, oi_insns[i], dest, val >> (i * 16));
1180        return;
1181    }
1182
1183    i = is_const_p32(val);
1184    if (i >= 0) {
1185        tcg_out_insn_RIL(s, oif_insns[i], dest, val >> (i * 32));
1186        return;
1187    }
1188
1189    g_assert_not_reached();
1190}
1191
1192static void tgen_xori(TCGContext *s, TCGReg dest, uint64_t val)
1193{
1194    switch (is_const_p32(val)) {
1195    case 0:
1196        tcg_out_insn(s, RIL, XILF, dest, val);
1197        break;
1198    case 1:
1199        tcg_out_insn(s, RIL, XIHF, dest, val >> 32);
1200        break;
1201    default:
1202        g_assert_not_reached();
1203    }
1204}
1205
1206static int tgen_cmp2(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1207                     TCGArg c2, bool c2const, bool need_carry, int *inv_cc)
1208{
1209    bool is_unsigned = is_unsigned_cond(c);
1210    TCGCond inv_c = tcg_invert_cond(c);
1211    S390Opcode op;
1212
1213    if (c2const) {
1214        if (c2 == 0) {
1215            if (!(is_unsigned && need_carry)) {
1216                if (type == TCG_TYPE_I32) {
1217                    tcg_out_insn(s, RR, LTR, r1, r1);
1218                } else {
1219                    tcg_out_insn(s, RRE, LTGR, r1, r1);
1220                }
1221                *inv_cc = tcg_cond_to_ltr_cond[inv_c];
1222                return tcg_cond_to_ltr_cond[c];
1223            }
1224        }
1225
1226        if (!is_unsigned && c2 == (int16_t)c2) {
1227            op = (type == TCG_TYPE_I32 ? RI_CHI : RI_CGHI);
1228            tcg_out_insn_RI(s, op, r1, c2);
1229            goto exit;
1230        }
1231
1232        if (type == TCG_TYPE_I32) {
1233            op = (is_unsigned ? RIL_CLFI : RIL_CFI);
1234            tcg_out_insn_RIL(s, op, r1, c2);
1235            goto exit;
1236        }
1237
1238        /*
1239         * Constraints are for a signed 33-bit operand, which is a
1240         * convenient superset of this signed/unsigned test.
1241         */
1242        if (c2 == (is_unsigned ? (TCGArg)(uint32_t)c2 : (TCGArg)(int32_t)c2)) {
1243            op = (is_unsigned ? RIL_CLGFI : RIL_CGFI);
1244            tcg_out_insn_RIL(s, op, r1, c2);
1245            goto exit;
1246        }
1247
1248        /* Load everything else into a register. */
1249        tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, c2);
1250        c2 = TCG_TMP0;
1251    }
1252
1253    if (type == TCG_TYPE_I32) {
1254        op = (is_unsigned ? RR_CLR : RR_CR);
1255        tcg_out_insn_RR(s, op, r1, c2);
1256    } else {
1257        op = (is_unsigned ? RRE_CLGR : RRE_CGR);
1258        tcg_out_insn_RRE(s, op, r1, c2);
1259    }
1260
1261 exit:
1262    *inv_cc = tcg_cond_to_s390_cond[inv_c];
1263    return tcg_cond_to_s390_cond[c];
1264}
1265
1266static int tgen_cmp(TCGContext *s, TCGType type, TCGCond c, TCGReg r1,
1267                    TCGArg c2, bool c2const, bool need_carry)
1268{
1269    int inv_cc;
1270    return tgen_cmp2(s, type, c, r1, c2, c2const, need_carry, &inv_cc);
1271}
1272
1273static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond,
1274                         TCGReg dest, TCGReg c1, TCGArg c2, int c2const)
1275{
1276    int cc;
1277
1278    /* With LOC2, we can always emit the minimum 3 insns.  */
1279    if (HAVE_FACILITY(LOAD_ON_COND2)) {
1280        /* Emit: d = 0, d = (cc ? 1 : d).  */
1281        cc = tgen_cmp(s, type, cond, c1, c2, c2const, false);
1282        tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1283        tcg_out_insn(s, RIEg, LOCGHI, dest, 1, cc);
1284        return;
1285    }
1286
1287 restart:
1288    switch (cond) {
1289    case TCG_COND_NE:
1290        /* X != 0 is X > 0.  */
1291        if (c2const && c2 == 0) {
1292            cond = TCG_COND_GTU;
1293        } else {
1294            break;
1295        }
1296        /* fallthru */
1297
1298    case TCG_COND_GTU:
1299    case TCG_COND_GT:
1300        /* The result of a compare has CC=2 for GT and CC=3 unused.
1301           ADD LOGICAL WITH CARRY considers (CC & 2) the carry bit.  */
1302        tgen_cmp(s, type, cond, c1, c2, c2const, true);
1303        tcg_out_movi(s, type, dest, 0);
1304        tcg_out_insn(s, RRE, ALCGR, dest, dest);
1305        return;
1306
1307    case TCG_COND_EQ:
1308        /* X == 0 is X <= 0.  */
1309        if (c2const && c2 == 0) {
1310            cond = TCG_COND_LEU;
1311        } else {
1312            break;
1313        }
1314        /* fallthru */
1315
1316    case TCG_COND_LEU:
1317    case TCG_COND_LE:
1318        /* As above, but we're looking for borrow, or !carry.
1319           The second insn computes d - d - borrow, or -1 for true
1320           and 0 for false.  So we must mask to 1 bit afterward.  */
1321        tgen_cmp(s, type, cond, c1, c2, c2const, true);
1322        tcg_out_insn(s, RRE, SLBGR, dest, dest);
1323        tgen_andi(s, type, dest, 1);
1324        return;
1325
1326    case TCG_COND_GEU:
1327    case TCG_COND_LTU:
1328    case TCG_COND_LT:
1329    case TCG_COND_GE:
1330        /* Swap operands so that we can use LEU/GTU/GT/LE.  */
1331        if (!c2const) {
1332            TCGReg t = c1;
1333            c1 = c2;
1334            c2 = t;
1335            cond = tcg_swap_cond(cond);
1336            goto restart;
1337        }
1338        break;
1339
1340    default:
1341        g_assert_not_reached();
1342    }
1343
1344    cc = tgen_cmp(s, type, cond, c1, c2, c2const, false);
1345    /* Emit: d = 0, t = 1, d = (cc ? t : d).  */
1346    tcg_out_movi(s, TCG_TYPE_I64, dest, 0);
1347    tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, 1);
1348    tcg_out_insn(s, RRFc, LOCGR, dest, TCG_TMP0, cc);
1349}
1350
1351static void tgen_movcond_int(TCGContext *s, TCGType type, TCGReg dest,
1352                             TCGArg v3, int v3const, TCGReg v4,
1353                             int cc, int inv_cc)
1354{
1355    TCGReg src;
1356
1357    if (v3const) {
1358        if (dest == v4) {
1359            if (HAVE_FACILITY(LOAD_ON_COND2)) {
1360                /* Emit: if (cc) dest = v3. */
1361                tcg_out_insn(s, RIEg, LOCGHI, dest, v3, cc);
1362                return;
1363            }
1364            tcg_out_insn(s, RI, LGHI, TCG_TMP0, v3);
1365            src = TCG_TMP0;
1366        } else {
1367            /* LGR+LOCGHI is larger than LGHI+LOCGR. */
1368            tcg_out_insn(s, RI, LGHI, dest, v3);
1369            cc = inv_cc;
1370            src = v4;
1371        }
1372    } else {
1373        if (HAVE_FACILITY(MISC_INSN_EXT3)) {
1374            /* Emit: dest = cc ? v3 : v4. */
1375            tcg_out_insn(s, RRFam, SELGR, dest, v3, v4, cc);
1376            return;
1377        }
1378        if (dest == v4) {
1379            src = v3;
1380        } else {
1381            tcg_out_mov(s, type, dest, v3);
1382            cc = inv_cc;
1383            src = v4;
1384        }
1385    }
1386
1387    /* Emit: if (cc) dest = src. */
1388    tcg_out_insn(s, RRFc, LOCGR, dest, src, cc);
1389}
1390
1391static void tgen_movcond(TCGContext *s, TCGType type, TCGCond c, TCGReg dest,
1392                         TCGReg c1, TCGArg c2, int c2const,
1393                         TCGArg v3, int v3const, TCGReg v4)
1394{
1395    int cc, inv_cc;
1396
1397    cc = tgen_cmp2(s, type, c, c1, c2, c2const, false, &inv_cc);
1398    tgen_movcond_int(s, type, dest, v3, v3const, v4, cc, inv_cc);
1399}
1400
1401static void tgen_clz(TCGContext *s, TCGReg dest, TCGReg a1,
1402                     TCGArg a2, int a2const)
1403{
1404    /* Since this sets both R and R+1, we have no choice but to store the
1405       result into R0, allowing R1 == TCG_TMP0 to be clobbered as well.  */
1406    QEMU_BUILD_BUG_ON(TCG_TMP0 != TCG_REG_R1);
1407    tcg_out_insn(s, RRE, FLOGR, TCG_REG_R0, a1);
1408
1409    if (a2const && a2 == 64) {
1410        tcg_out_mov(s, TCG_TYPE_I64, dest, TCG_REG_R0);
1411        return;
1412    }
1413
1414    /*
1415     * Conditions from FLOGR are:
1416     *   2 -> one bit found
1417     *   8 -> no one bit found
1418     */
1419    tgen_movcond_int(s, TCG_TYPE_I64, dest, a2, a2const, TCG_REG_R0, 8, 2);
1420}
1421
1422static void tgen_ctpop(TCGContext *s, TCGType type, TCGReg dest, TCGReg src)
1423{
1424    /* With MIE3, and bit 0 of m4 set, we get the complete result. */
1425    if (HAVE_FACILITY(MISC_INSN_EXT3)) {
1426        if (type == TCG_TYPE_I32) {
1427            tcg_out_ext32u(s, dest, src);
1428            src = dest;
1429        }
1430        tcg_out_insn(s, RRFc, POPCNT, dest, src, 8);
1431        return;
1432    }
1433
1434    /* Without MIE3, each byte gets the count of bits for the byte. */
1435    tcg_out_insn(s, RRFc, POPCNT, dest, src, 0);
1436
1437    /* Multiply to sum each byte at the top of the word. */
1438    if (type == TCG_TYPE_I32) {
1439        tcg_out_insn(s, RIL, MSFI, dest, 0x01010101);
1440        tcg_out_sh32(s, RS_SRL, dest, TCG_REG_NONE, 24);
1441    } else {
1442        tcg_out_movi(s, TCG_TYPE_I64, TCG_TMP0, 0x0101010101010101ull);
1443        tcg_out_insn(s, RRE, MSGR, dest, TCG_TMP0);
1444        tcg_out_sh64(s, RSY_SRLG, dest, dest, TCG_REG_NONE, 56);
1445    }
1446}
1447
1448static void tgen_deposit(TCGContext *s, TCGReg dest, TCGReg src,
1449                         int ofs, int len, int z)
1450{
1451    int lsb = (63 - ofs);
1452    int msb = lsb - (len - 1);
1453    tcg_out_risbg(s, dest, src, msb, lsb, ofs, z);
1454}
1455
1456static void tgen_extract(TCGContext *s, TCGReg dest, TCGReg src,
1457                         int ofs, int len)
1458{
1459    tcg_out_risbg(s, dest, src, 64 - len, 63, 64 - ofs, 1);
1460}
1461
1462static void tgen_gotoi(TCGContext *s, int cc, const tcg_insn_unit *dest)
1463{
1464    ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1;
1465    if (off == (int16_t)off) {
1466        tcg_out_insn(s, RI, BRC, cc, off);
1467    } else if (off == (int32_t)off) {
1468        tcg_out_insn(s, RIL, BRCL, cc, off);
1469    } else {
1470        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1471        tcg_out_insn(s, RR, BCR, cc, TCG_TMP0);
1472    }
1473}
1474
1475static void tgen_branch(TCGContext *s, int cc, TCGLabel *l)
1476{
1477    if (l->has_value) {
1478        tgen_gotoi(s, cc, l->u.value_ptr);
1479    } else {
1480        tcg_out16(s, RI_BRC | (cc << 4));
1481        tcg_out_reloc(s, s->code_ptr, R_390_PC16DBL, l, 2);
1482        s->code_ptr += 1;
1483    }
1484}
1485
1486static void tgen_compare_branch(TCGContext *s, S390Opcode opc, int cc,
1487                                TCGReg r1, TCGReg r2, TCGLabel *l)
1488{
1489    tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1490    /* Format RIE-b */
1491    tcg_out16(s, (opc & 0xff00) | (r1 << 4) | r2);
1492    tcg_out16(s, 0);
1493    tcg_out16(s, cc << 12 | (opc & 0xff));
1494}
1495
1496static void tgen_compare_imm_branch(TCGContext *s, S390Opcode opc, int cc,
1497                                    TCGReg r1, int i2, TCGLabel *l)
1498{
1499    tcg_out_reloc(s, s->code_ptr + 1, R_390_PC16DBL, l, 2);
1500    /* Format RIE-c */
1501    tcg_out16(s, (opc & 0xff00) | (r1 << 4) | cc);
1502    tcg_out16(s, 0);
1503    tcg_out16(s, (i2 << 8) | (opc & 0xff));
1504}
1505
1506static void tgen_brcond(TCGContext *s, TCGType type, TCGCond c,
1507                        TCGReg r1, TCGArg c2, int c2const, TCGLabel *l)
1508{
1509    int cc;
1510    bool is_unsigned = is_unsigned_cond(c);
1511    bool in_range;
1512    S390Opcode opc;
1513
1514    cc = tcg_cond_to_s390_cond[c];
1515
1516    if (!c2const) {
1517        opc = (type == TCG_TYPE_I32
1518               ? (is_unsigned ? RIEb_CLRJ : RIEb_CRJ)
1519               : (is_unsigned ? RIEb_CLGRJ : RIEb_CGRJ));
1520        tgen_compare_branch(s, opc, cc, r1, c2, l);
1521        return;
1522    }
1523
1524    /*
1525     * COMPARE IMMEDIATE AND BRANCH RELATIVE has an 8-bit immediate field.
1526     * If the immediate we've been given does not fit that range, we'll
1527     * fall back to separate compare and branch instructions using the
1528     * larger comparison range afforded by COMPARE IMMEDIATE.
1529     */
1530    if (type == TCG_TYPE_I32) {
1531        if (is_unsigned) {
1532            opc = RIEc_CLIJ;
1533            in_range = (uint32_t)c2 == (uint8_t)c2;
1534        } else {
1535            opc = RIEc_CIJ;
1536            in_range = (int32_t)c2 == (int8_t)c2;
1537        }
1538    } else {
1539        if (is_unsigned) {
1540            opc = RIEc_CLGIJ;
1541            in_range = (uint64_t)c2 == (uint8_t)c2;
1542        } else {
1543            opc = RIEc_CGIJ;
1544            in_range = (int64_t)c2 == (int8_t)c2;
1545        }
1546    }
1547    if (in_range) {
1548        tgen_compare_imm_branch(s, opc, cc, r1, c2, l);
1549        return;
1550    }
1551
1552    cc = tgen_cmp(s, type, c, r1, c2, c2const, false);
1553    tgen_branch(s, cc, l);
1554}
1555
1556static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *dest)
1557{
1558    ptrdiff_t off = tcg_pcrel_diff(s, dest) >> 1;
1559    if (off == (int32_t)off) {
1560        tcg_out_insn(s, RIL, BRASL, TCG_REG_R14, off);
1561    } else {
1562        tcg_out_movi(s, TCG_TYPE_PTR, TCG_TMP0, (uintptr_t)dest);
1563        tcg_out_insn(s, RR, BASR, TCG_REG_R14, TCG_TMP0);
1564    }
1565}
1566
1567static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest,
1568                         const TCGHelperInfo *info)
1569{
1570    tcg_out_call_int(s, dest);
1571}
1572
1573typedef struct {
1574    TCGReg base;
1575    TCGReg index;
1576    int disp;
1577    TCGAtomAlign aa;
1578} HostAddress;
1579
1580bool tcg_target_has_memory_bswap(MemOp memop)
1581{
1582    TCGAtomAlign aa;
1583
1584    if ((memop & MO_SIZE) <= MO_64) {
1585        return true;
1586    }
1587
1588    /*
1589     * Reject 16-byte memop with 16-byte atomicity,
1590     * but do allow a pair of 64-bit operations.
1591     */
1592    aa = atom_and_align_for_opc(tcg_ctx, memop, MO_ATOM_IFALIGN, true);
1593    return aa.atom <= MO_64;
1594}
1595
1596static void tcg_out_qemu_ld_direct(TCGContext *s, MemOp opc, TCGReg data,
1597                                   HostAddress h)
1598{
1599    switch (opc & (MO_SSIZE | MO_BSWAP)) {
1600    case MO_UB:
1601        tcg_out_insn(s, RXY, LLGC, data, h.base, h.index, h.disp);
1602        break;
1603    case MO_SB:
1604        tcg_out_insn(s, RXY, LGB, data, h.base, h.index, h.disp);
1605        break;
1606
1607    case MO_UW | MO_BSWAP:
1608        /* swapped unsigned halfword load with upper bits zeroed */
1609        tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp);
1610        tcg_out_ext16u(s, data, data);
1611        break;
1612    case MO_UW:
1613        tcg_out_insn(s, RXY, LLGH, data, h.base, h.index, h.disp);
1614        break;
1615
1616    case MO_SW | MO_BSWAP:
1617        /* swapped sign-extended halfword load */
1618        tcg_out_insn(s, RXY, LRVH, data, h.base, h.index, h.disp);
1619        tcg_out_ext16s(s, TCG_TYPE_REG, data, data);
1620        break;
1621    case MO_SW:
1622        tcg_out_insn(s, RXY, LGH, data, h.base, h.index, h.disp);
1623        break;
1624
1625    case MO_UL | MO_BSWAP:
1626        /* swapped unsigned int load with upper bits zeroed */
1627        tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp);
1628        tcg_out_ext32u(s, data, data);
1629        break;
1630    case MO_UL:
1631        tcg_out_insn(s, RXY, LLGF, data, h.base, h.index, h.disp);
1632        break;
1633
1634    case MO_SL | MO_BSWAP:
1635        /* swapped sign-extended int load */
1636        tcg_out_insn(s, RXY, LRV, data, h.base, h.index, h.disp);
1637        tcg_out_ext32s(s, data, data);
1638        break;
1639    case MO_SL:
1640        tcg_out_insn(s, RXY, LGF, data, h.base, h.index, h.disp);
1641        break;
1642
1643    case MO_UQ | MO_BSWAP:
1644        tcg_out_insn(s, RXY, LRVG, data, h.base, h.index, h.disp);
1645        break;
1646    case MO_UQ:
1647        tcg_out_insn(s, RXY, LG, data, h.base, h.index, h.disp);
1648        break;
1649
1650    default:
1651        g_assert_not_reached();
1652    }
1653}
1654
1655static void tcg_out_qemu_st_direct(TCGContext *s, MemOp opc, TCGReg data,
1656                                   HostAddress h)
1657{
1658    switch (opc & (MO_SIZE | MO_BSWAP)) {
1659    case MO_UB:
1660        if (h.disp >= 0 && h.disp < 0x1000) {
1661            tcg_out_insn(s, RX, STC, data, h.base, h.index, h.disp);
1662        } else {
1663            tcg_out_insn(s, RXY, STCY, data, h.base, h.index, h.disp);
1664        }
1665        break;
1666
1667    case MO_UW | MO_BSWAP:
1668        tcg_out_insn(s, RXY, STRVH, data, h.base, h.index, h.disp);
1669        break;
1670    case MO_UW:
1671        if (h.disp >= 0 && h.disp < 0x1000) {
1672            tcg_out_insn(s, RX, STH, data, h.base, h.index, h.disp);
1673        } else {
1674            tcg_out_insn(s, RXY, STHY, data, h.base, h.index, h.disp);
1675        }
1676        break;
1677
1678    case MO_UL | MO_BSWAP:
1679        tcg_out_insn(s, RXY, STRV, data, h.base, h.index, h.disp);
1680        break;
1681    case MO_UL:
1682        if (h.disp >= 0 && h.disp < 0x1000) {
1683            tcg_out_insn(s, RX, ST, data, h.base, h.index, h.disp);
1684        } else {
1685            tcg_out_insn(s, RXY, STY, data, h.base, h.index, h.disp);
1686        }
1687        break;
1688
1689    case MO_UQ | MO_BSWAP:
1690        tcg_out_insn(s, RXY, STRVG, data, h.base, h.index, h.disp);
1691        break;
1692    case MO_UQ:
1693        tcg_out_insn(s, RXY, STG, data, h.base, h.index, h.disp);
1694        break;
1695
1696    default:
1697        g_assert_not_reached();
1698    }
1699}
1700
1701static const TCGLdstHelperParam ldst_helper_param = {
1702    .ntmp = 1, .tmp = { TCG_TMP0 }
1703};
1704
1705static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1706{
1707    MemOp opc = get_memop(lb->oi);
1708
1709    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1710                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) {
1711        return false;
1712    }
1713
1714    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
1715    tcg_out_call_int(s, qemu_ld_helpers[opc & MO_SIZE]);
1716    tcg_out_ld_helper_ret(s, lb, false, &ldst_helper_param);
1717
1718    tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1719    return true;
1720}
1721
1722static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1723{
1724    MemOp opc = get_memop(lb->oi);
1725
1726    if (!patch_reloc(lb->label_ptr[0], R_390_PC16DBL,
1727                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 2)) {
1728        return false;
1729    }
1730
1731    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
1732    tcg_out_call_int(s, qemu_st_helpers[opc & MO_SIZE]);
1733
1734    tgen_gotoi(s, S390_CC_ALWAYS, lb->raddr);
1735    return true;
1736}
1737
1738/*
1739 * For softmmu, perform the TLB load and compare.
1740 * For useronly, perform any required alignment tests.
1741 * In both cases, return a TCGLabelQemuLdst structure if the slow path
1742 * is required and fill in @h with the host address for the fast path.
1743 */
1744static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
1745                                           TCGReg addr_reg, MemOpIdx oi,
1746                                           bool is_ld)
1747{
1748    TCGLabelQemuLdst *ldst = NULL;
1749    MemOp opc = get_memop(oi);
1750    MemOp s_bits = opc & MO_SIZE;
1751    unsigned a_mask;
1752
1753    h->aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, s_bits == MO_128);
1754    a_mask = (1 << h->aa.align) - 1;
1755
1756#ifdef CONFIG_SOFTMMU
1757    unsigned s_mask = (1 << s_bits) - 1;
1758    int mem_index = get_mmuidx(oi);
1759    int fast_off = TLB_MASK_TABLE_OFS(mem_index);
1760    int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1761    int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1762    int ofs, a_off;
1763    uint64_t tlb_mask;
1764
1765    ldst = new_ldst_label(s);
1766    ldst->is_ld = is_ld;
1767    ldst->oi = oi;
1768    ldst->addrlo_reg = addr_reg;
1769
1770    tcg_out_sh64(s, RSY_SRLG, TCG_TMP0, addr_reg, TCG_REG_NONE,
1771                 s->page_bits - CPU_TLB_ENTRY_BITS);
1772
1773    QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0);
1774    QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -(1 << 19));
1775    tcg_out_insn(s, RXY, NG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, mask_off);
1776    tcg_out_insn(s, RXY, AG, TCG_TMP0, TCG_AREG0, TCG_REG_NONE, table_off);
1777
1778    /*
1779     * For aligned accesses, we check the first byte and include the alignment
1780     * bits within the address.  For unaligned access, we check that we don't
1781     * cross pages using the address of the last byte of the access.
1782     */
1783    a_off = (a_mask >= s_mask ? 0 : s_mask - a_mask);
1784    tlb_mask = (uint64_t)s->page_mask | a_mask;
1785    if (a_off == 0) {
1786        tgen_andi_risbg(s, TCG_REG_R0, addr_reg, tlb_mask);
1787    } else {
1788        tcg_out_insn(s, RX, LA, TCG_REG_R0, addr_reg, TCG_REG_NONE, a_off);
1789        tgen_andi(s, TCG_TYPE_TL, TCG_REG_R0, tlb_mask);
1790    }
1791
1792    if (is_ld) {
1793        ofs = offsetof(CPUTLBEntry, addr_read);
1794    } else {
1795        ofs = offsetof(CPUTLBEntry, addr_write);
1796    }
1797    if (TARGET_LONG_BITS == 32) {
1798        tcg_out_insn(s, RX, C, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs);
1799    } else {
1800        tcg_out_insn(s, RXY, CG, TCG_REG_R0, TCG_TMP0, TCG_REG_NONE, ofs);
1801    }
1802
1803    tcg_out16(s, RI_BRC | (S390_CC_NE << 4));
1804    ldst->label_ptr[0] = s->code_ptr++;
1805
1806    h->index = TCG_TMP0;
1807    tcg_out_insn(s, RXY, LG, h->index, TCG_TMP0, TCG_REG_NONE,
1808                 offsetof(CPUTLBEntry, addend));
1809
1810    if (TARGET_LONG_BITS == 32) {
1811        tcg_out_insn(s, RRE, ALGFR, h->index, addr_reg);
1812        h->base = TCG_REG_NONE;
1813    } else {
1814        h->base = addr_reg;
1815    }
1816    h->disp = 0;
1817#else
1818    if (a_mask) {
1819        ldst = new_ldst_label(s);
1820        ldst->is_ld = is_ld;
1821        ldst->oi = oi;
1822        ldst->addrlo_reg = addr_reg;
1823
1824        /* We are expecting a_bits to max out at 7, much lower than TMLL. */
1825        tcg_debug_assert(a_mask <= 0xffff);
1826        tcg_out_insn(s, RI, TMLL, addr_reg, a_mask);
1827
1828        tcg_out16(s, RI_BRC | (7 << 4)); /* CC in {1,2,3} */
1829        ldst->label_ptr[0] = s->code_ptr++;
1830    }
1831
1832    h->base = addr_reg;
1833    if (TARGET_LONG_BITS == 32) {
1834        tcg_out_ext32u(s, TCG_TMP0, addr_reg);
1835        h->base = TCG_TMP0;
1836    }
1837    if (guest_base < 0x80000) {
1838        h->index = TCG_REG_NONE;
1839        h->disp = guest_base;
1840    } else {
1841        h->index = TCG_GUEST_BASE_REG;
1842        h->disp = 0;
1843    }
1844#endif
1845
1846    return ldst;
1847}
1848
1849static void tcg_out_qemu_ld(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
1850                            MemOpIdx oi, TCGType data_type)
1851{
1852    TCGLabelQemuLdst *ldst;
1853    HostAddress h;
1854
1855    ldst = prepare_host_addr(s, &h, addr_reg, oi, true);
1856    tcg_out_qemu_ld_direct(s, get_memop(oi), data_reg, h);
1857
1858    if (ldst) {
1859        ldst->type = data_type;
1860        ldst->datalo_reg = data_reg;
1861        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1862    }
1863}
1864
1865static void tcg_out_qemu_st(TCGContext* s, TCGReg data_reg, TCGReg addr_reg,
1866                            MemOpIdx oi, TCGType data_type)
1867{
1868    TCGLabelQemuLdst *ldst;
1869    HostAddress h;
1870
1871    ldst = prepare_host_addr(s, &h, addr_reg, oi, false);
1872    tcg_out_qemu_st_direct(s, get_memop(oi), data_reg, h);
1873
1874    if (ldst) {
1875        ldst->type = data_type;
1876        ldst->datalo_reg = data_reg;
1877        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1878    }
1879}
1880
1881static void tcg_out_qemu_ldst_i128(TCGContext *s, TCGReg datalo, TCGReg datahi,
1882                                   TCGReg addr_reg, MemOpIdx oi, bool is_ld)
1883{
1884    TCGLabel *l1 = NULL, *l2 = NULL;
1885    TCGLabelQemuLdst *ldst;
1886    HostAddress h;
1887    bool need_bswap;
1888    bool use_pair;
1889    S390Opcode insn;
1890
1891    ldst = prepare_host_addr(s, &h, addr_reg, oi, is_ld);
1892
1893    use_pair = h.aa.atom < MO_128;
1894    need_bswap = get_memop(oi) & MO_BSWAP;
1895
1896    if (!use_pair) {
1897        /*
1898         * Atomicity requires we use LPQ.  If we've already checked for
1899         * 16-byte alignment, that's all we need.  If we arrive with
1900         * lesser alignment, we have determined that less than 16-byte
1901         * alignment can be satisfied with two 8-byte loads.
1902         */
1903        if (h.aa.align < MO_128) {
1904            use_pair = true;
1905            l1 = gen_new_label();
1906            l2 = gen_new_label();
1907
1908            tcg_out_insn(s, RI, TMLL, addr_reg, 15);
1909            tgen_branch(s, 7, l1); /* CC in {1,2,3} */
1910        }
1911
1912        tcg_debug_assert(!need_bswap);
1913        tcg_debug_assert(datalo & 1);
1914        tcg_debug_assert(datahi == datalo - 1);
1915        insn = is_ld ? RXY_LPQ : RXY_STPQ;
1916        tcg_out_insn_RXY(s, insn, datahi, h.base, h.index, h.disp);
1917
1918        if (use_pair) {
1919            tgen_branch(s, S390_CC_ALWAYS, l2);
1920            tcg_out_label(s, l1);
1921        }
1922    }
1923    if (use_pair) {
1924        TCGReg d1, d2;
1925
1926        if (need_bswap) {
1927            d1 = datalo, d2 = datahi;
1928            insn = is_ld ? RXY_LRVG : RXY_STRVG;
1929        } else {
1930            d1 = datahi, d2 = datalo;
1931            insn = is_ld ? RXY_LG : RXY_STG;
1932        }
1933
1934        if (h.base == d1 || h.index == d1) {
1935            tcg_out_insn(s, RXY, LAY, TCG_TMP0, h.base, h.index, h.disp);
1936            h.base = TCG_TMP0;
1937            h.index = TCG_REG_NONE;
1938            h.disp = 0;
1939        }
1940        tcg_out_insn_RXY(s, insn, d1, h.base, h.index, h.disp);
1941        tcg_out_insn_RXY(s, insn, d2, h.base, h.index, h.disp + 8);
1942    }
1943    if (l2) {
1944        tcg_out_label(s, l2);
1945    }
1946
1947    if (ldst) {
1948        ldst->type = TCG_TYPE_I128;
1949        ldst->datalo_reg = datalo;
1950        ldst->datahi_reg = datahi;
1951        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1952    }
1953}
1954
1955static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
1956{
1957    /* Reuse the zeroing that exists for goto_ptr.  */
1958    if (a0 == 0) {
1959        tgen_gotoi(s, S390_CC_ALWAYS, tcg_code_gen_epilogue);
1960    } else {
1961        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, a0);
1962        tgen_gotoi(s, S390_CC_ALWAYS, tb_ret_addr);
1963    }
1964}
1965
1966static void tcg_out_goto_tb(TCGContext *s, int which)
1967{
1968    /*
1969     * Branch displacement must be aligned for atomic patching;
1970     * see if we need to add extra nop before branch
1971     */
1972    if (!QEMU_PTR_IS_ALIGNED(s->code_ptr + 1, 4)) {
1973        tcg_out16(s, NOP);
1974    }
1975    tcg_out16(s, RIL_BRCL | (S390_CC_ALWAYS << 4));
1976    set_jmp_insn_offset(s, which);
1977    s->code_ptr += 2;
1978    set_jmp_reset_offset(s, which);
1979}
1980
1981void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
1982                              uintptr_t jmp_rx, uintptr_t jmp_rw)
1983{
1984    if (!HAVE_FACILITY(GEN_INST_EXT)) {
1985        return;
1986    }
1987    /* patch the branch destination */
1988    uintptr_t addr = tb->jmp_target_addr[n];
1989    intptr_t disp = addr - (jmp_rx - 2);
1990    qatomic_set((int32_t *)jmp_rw, disp / 2);
1991    /* no need to flush icache explicitly */
1992}
1993
1994# define OP_32_64(x) \
1995        case glue(glue(INDEX_op_,x),_i32): \
1996        case glue(glue(INDEX_op_,x),_i64)
1997
1998static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
1999                              const TCGArg args[TCG_MAX_OP_ARGS],
2000                              const int const_args[TCG_MAX_OP_ARGS])
2001{
2002    S390Opcode op, op2;
2003    TCGArg a0, a1, a2;
2004
2005    switch (opc) {
2006    case INDEX_op_goto_ptr:
2007        a0 = args[0];
2008        tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, a0);
2009        break;
2010
2011    OP_32_64(ld8u):
2012        /* ??? LLC (RXY format) is only present with the extended-immediate
2013           facility, whereas LLGC is always present.  */
2014        tcg_out_mem(s, 0, RXY_LLGC, args[0], args[1], TCG_REG_NONE, args[2]);
2015        break;
2016
2017    OP_32_64(ld8s):
2018        /* ??? LB is no smaller than LGB, so no point to using it.  */
2019        tcg_out_mem(s, 0, RXY_LGB, args[0], args[1], TCG_REG_NONE, args[2]);
2020        break;
2021
2022    OP_32_64(ld16u):
2023        /* ??? LLH (RXY format) is only present with the extended-immediate
2024           facility, whereas LLGH is always present.  */
2025        tcg_out_mem(s, 0, RXY_LLGH, args[0], args[1], TCG_REG_NONE, args[2]);
2026        break;
2027
2028    case INDEX_op_ld16s_i32:
2029        tcg_out_mem(s, RX_LH, RXY_LHY, args[0], args[1], TCG_REG_NONE, args[2]);
2030        break;
2031
2032    case INDEX_op_ld_i32:
2033        tcg_out_ld(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2034        break;
2035
2036    OP_32_64(st8):
2037        tcg_out_mem(s, RX_STC, RXY_STCY, args[0], args[1],
2038                    TCG_REG_NONE, args[2]);
2039        break;
2040
2041    OP_32_64(st16):
2042        tcg_out_mem(s, RX_STH, RXY_STHY, args[0], args[1],
2043                    TCG_REG_NONE, args[2]);
2044        break;
2045
2046    case INDEX_op_st_i32:
2047        tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2048        break;
2049
2050    case INDEX_op_add_i32:
2051        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2052        if (const_args[2]) {
2053        do_addi_32:
2054            if (a0 == a1) {
2055                if (a2 == (int16_t)a2) {
2056                    tcg_out_insn(s, RI, AHI, a0, a2);
2057                    break;
2058                }
2059                tcg_out_insn(s, RIL, AFI, a0, a2);
2060                break;
2061            }
2062            tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
2063        } else if (a0 == a1) {
2064            tcg_out_insn(s, RR, AR, a0, a2);
2065        } else {
2066            tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
2067        }
2068        break;
2069    case INDEX_op_sub_i32:
2070        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2071        if (const_args[2]) {
2072            a2 = -a2;
2073            goto do_addi_32;
2074        } else if (a0 == a1) {
2075            tcg_out_insn(s, RR, SR, a0, a2);
2076        } else {
2077            tcg_out_insn(s, RRFa, SRK, a0, a1, a2);
2078        }
2079        break;
2080
2081    case INDEX_op_and_i32:
2082        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2083        if (const_args[2]) {
2084            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2085            tgen_andi(s, TCG_TYPE_I32, a0, a2);
2086        } else if (a0 == a1) {
2087            tcg_out_insn(s, RR, NR, a0, a2);
2088        } else {
2089            tcg_out_insn(s, RRFa, NRK, a0, a1, a2);
2090        }
2091        break;
2092    case INDEX_op_or_i32:
2093        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2094        if (const_args[2]) {
2095            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2096            tgen_ori(s, a0, a2);
2097        } else if (a0 == a1) {
2098            tcg_out_insn(s, RR, OR, a0, a2);
2099        } else {
2100            tcg_out_insn(s, RRFa, ORK, a0, a1, a2);
2101        }
2102        break;
2103    case INDEX_op_xor_i32:
2104        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2105        if (const_args[2]) {
2106            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2107            tcg_out_insn(s, RIL, XILF, a0, a2);
2108        } else if (a0 == a1) {
2109            tcg_out_insn(s, RR, XR, args[0], args[2]);
2110        } else {
2111            tcg_out_insn(s, RRFa, XRK, a0, a1, a2);
2112        }
2113        break;
2114
2115    case INDEX_op_andc_i32:
2116        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2117        if (const_args[2]) {
2118            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2119            tgen_andi(s, TCG_TYPE_I32, a0, (uint32_t)~a2);
2120	} else {
2121            tcg_out_insn(s, RRFa, NCRK, a0, a1, a2);
2122	}
2123        break;
2124    case INDEX_op_orc_i32:
2125        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2126        if (const_args[2]) {
2127            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2128            tgen_ori(s, a0, (uint32_t)~a2);
2129        } else {
2130            tcg_out_insn(s, RRFa, OCRK, a0, a1, a2);
2131        }
2132        break;
2133    case INDEX_op_eqv_i32:
2134        a0 = args[0], a1 = args[1], a2 = (uint32_t)args[2];
2135        if (const_args[2]) {
2136            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2137            tcg_out_insn(s, RIL, XILF, a0, ~a2);
2138        } else {
2139            tcg_out_insn(s, RRFa, NXRK, a0, a1, a2);
2140        }
2141        break;
2142    case INDEX_op_nand_i32:
2143        tcg_out_insn(s, RRFa, NNRK, args[0], args[1], args[2]);
2144        break;
2145    case INDEX_op_nor_i32:
2146        tcg_out_insn(s, RRFa, NORK, args[0], args[1], args[2]);
2147        break;
2148
2149    case INDEX_op_neg_i32:
2150        tcg_out_insn(s, RR, LCR, args[0], args[1]);
2151        break;
2152    case INDEX_op_not_i32:
2153        tcg_out_insn(s, RRFa, NORK, args[0], args[1], args[1]);
2154        break;
2155
2156    case INDEX_op_mul_i32:
2157        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2158        if (const_args[2]) {
2159            tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
2160            if (a2 == (int16_t)a2) {
2161                tcg_out_insn(s, RI, MHI, a0, a2);
2162            } else {
2163                tcg_out_insn(s, RIL, MSFI, a0, a2);
2164            }
2165        } else if (a0 == a1) {
2166            tcg_out_insn(s, RRE, MSR, a0, a2);
2167        } else {
2168            tcg_out_insn(s, RRFa, MSRKC, a0, a1, a2);
2169        }
2170        break;
2171
2172    case INDEX_op_div2_i32:
2173        tcg_debug_assert(args[0] == args[2]);
2174        tcg_debug_assert(args[1] == args[3]);
2175        tcg_debug_assert((args[1] & 1) == 0);
2176        tcg_debug_assert(args[0] == args[1] + 1);
2177        tcg_out_insn(s, RR, DR, args[1], args[4]);
2178        break;
2179    case INDEX_op_divu2_i32:
2180        tcg_debug_assert(args[0] == args[2]);
2181        tcg_debug_assert(args[1] == args[3]);
2182        tcg_debug_assert((args[1] & 1) == 0);
2183        tcg_debug_assert(args[0] == args[1] + 1);
2184        tcg_out_insn(s, RRE, DLR, args[1], args[4]);
2185        break;
2186
2187    case INDEX_op_shl_i32:
2188        op = RS_SLL;
2189        op2 = RSY_SLLK;
2190    do_shift32:
2191        a0 = args[0], a1 = args[1], a2 = (int32_t)args[2];
2192        if (a0 == a1) {
2193            if (const_args[2]) {
2194                tcg_out_sh32(s, op, a0, TCG_REG_NONE, a2);
2195            } else {
2196                tcg_out_sh32(s, op, a0, a2, 0);
2197            }
2198        } else {
2199            /* Using tcg_out_sh64 here for the format; it is a 32-bit shift.  */
2200            if (const_args[2]) {
2201                tcg_out_sh64(s, op2, a0, a1, TCG_REG_NONE, a2);
2202            } else {
2203                tcg_out_sh64(s, op2, a0, a1, a2, 0);
2204            }
2205        }
2206        break;
2207    case INDEX_op_shr_i32:
2208        op = RS_SRL;
2209        op2 = RSY_SRLK;
2210        goto do_shift32;
2211    case INDEX_op_sar_i32:
2212        op = RS_SRA;
2213        op2 = RSY_SRAK;
2214        goto do_shift32;
2215
2216    case INDEX_op_rotl_i32:
2217        /* ??? Using tcg_out_sh64 here for the format; it is a 32-bit rol.  */
2218        if (const_args[2]) {
2219            tcg_out_sh64(s, RSY_RLL, args[0], args[1], TCG_REG_NONE, args[2]);
2220        } else {
2221            tcg_out_sh64(s, RSY_RLL, args[0], args[1], args[2], 0);
2222        }
2223        break;
2224    case INDEX_op_rotr_i32:
2225        if (const_args[2]) {
2226            tcg_out_sh64(s, RSY_RLL, args[0], args[1],
2227                         TCG_REG_NONE, (32 - args[2]) & 31);
2228        } else {
2229            tcg_out_insn(s, RR, LCR, TCG_TMP0, args[2]);
2230            tcg_out_sh64(s, RSY_RLL, args[0], args[1], TCG_TMP0, 0);
2231        }
2232        break;
2233
2234    case INDEX_op_bswap16_i32:
2235        a0 = args[0], a1 = args[1], a2 = args[2];
2236        tcg_out_insn(s, RRE, LRVR, a0, a1);
2237        if (a2 & TCG_BSWAP_OS) {
2238            tcg_out_sh32(s, RS_SRA, a0, TCG_REG_NONE, 16);
2239        } else {
2240            tcg_out_sh32(s, RS_SRL, a0, TCG_REG_NONE, 16);
2241        }
2242        break;
2243    case INDEX_op_bswap16_i64:
2244        a0 = args[0], a1 = args[1], a2 = args[2];
2245        tcg_out_insn(s, RRE, LRVGR, a0, a1);
2246        if (a2 & TCG_BSWAP_OS) {
2247            tcg_out_sh64(s, RSY_SRAG, a0, a0, TCG_REG_NONE, 48);
2248        } else {
2249            tcg_out_sh64(s, RSY_SRLG, a0, a0, TCG_REG_NONE, 48);
2250        }
2251        break;
2252
2253    case INDEX_op_bswap32_i32:
2254        tcg_out_insn(s, RRE, LRVR, args[0], args[1]);
2255        break;
2256    case INDEX_op_bswap32_i64:
2257        a0 = args[0], a1 = args[1], a2 = args[2];
2258        tcg_out_insn(s, RRE, LRVR, a0, a1);
2259        if (a2 & TCG_BSWAP_OS) {
2260            tcg_out_ext32s(s, a0, a0);
2261        } else if ((a2 & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
2262            tcg_out_ext32u(s, a0, a0);
2263        }
2264        break;
2265
2266    case INDEX_op_add2_i32:
2267        if (const_args[4]) {
2268            tcg_out_insn(s, RIL, ALFI, args[0], args[4]);
2269        } else {
2270            tcg_out_insn(s, RR, ALR, args[0], args[4]);
2271        }
2272        tcg_out_insn(s, RRE, ALCR, args[1], args[5]);
2273        break;
2274    case INDEX_op_sub2_i32:
2275        if (const_args[4]) {
2276            tcg_out_insn(s, RIL, SLFI, args[0], args[4]);
2277        } else {
2278            tcg_out_insn(s, RR, SLR, args[0], args[4]);
2279        }
2280        tcg_out_insn(s, RRE, SLBR, args[1], args[5]);
2281        break;
2282
2283    case INDEX_op_br:
2284        tgen_branch(s, S390_CC_ALWAYS, arg_label(args[0]));
2285        break;
2286
2287    case INDEX_op_brcond_i32:
2288        tgen_brcond(s, TCG_TYPE_I32, args[2], args[0],
2289                    args[1], const_args[1], arg_label(args[3]));
2290        break;
2291    case INDEX_op_setcond_i32:
2292        tgen_setcond(s, TCG_TYPE_I32, args[3], args[0], args[1],
2293                     args[2], const_args[2]);
2294        break;
2295    case INDEX_op_movcond_i32:
2296        tgen_movcond(s, TCG_TYPE_I32, args[5], args[0], args[1],
2297                     args[2], const_args[2], args[3], const_args[3], args[4]);
2298        break;
2299
2300    case INDEX_op_qemu_ld_a32_i32:
2301    case INDEX_op_qemu_ld_a64_i32:
2302        tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I32);
2303        break;
2304    case INDEX_op_qemu_ld_a32_i64:
2305    case INDEX_op_qemu_ld_a64_i64:
2306        tcg_out_qemu_ld(s, args[0], args[1], args[2], TCG_TYPE_I64);
2307        break;
2308    case INDEX_op_qemu_st_a32_i32:
2309    case INDEX_op_qemu_st_a64_i32:
2310        tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I32);
2311        break;
2312    case INDEX_op_qemu_st_a32_i64:
2313    case INDEX_op_qemu_st_a64_i64:
2314        tcg_out_qemu_st(s, args[0], args[1], args[2], TCG_TYPE_I64);
2315        break;
2316    case INDEX_op_qemu_ld_a32_i128:
2317    case INDEX_op_qemu_ld_a64_i128:
2318        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], true);
2319        break;
2320    case INDEX_op_qemu_st_a32_i128:
2321    case INDEX_op_qemu_st_a64_i128:
2322        tcg_out_qemu_ldst_i128(s, args[0], args[1], args[2], args[3], false);
2323        break;
2324
2325    case INDEX_op_ld16s_i64:
2326        tcg_out_mem(s, 0, RXY_LGH, args[0], args[1], TCG_REG_NONE, args[2]);
2327        break;
2328    case INDEX_op_ld32u_i64:
2329        tcg_out_mem(s, 0, RXY_LLGF, args[0], args[1], TCG_REG_NONE, args[2]);
2330        break;
2331    case INDEX_op_ld32s_i64:
2332        tcg_out_mem(s, 0, RXY_LGF, args[0], args[1], TCG_REG_NONE, args[2]);
2333        break;
2334    case INDEX_op_ld_i64:
2335        tcg_out_ld(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2336        break;
2337
2338    case INDEX_op_st32_i64:
2339        tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
2340        break;
2341    case INDEX_op_st_i64:
2342        tcg_out_st(s, TCG_TYPE_I64, args[0], args[1], args[2]);
2343        break;
2344
2345    case INDEX_op_add_i64:
2346        a0 = args[0], a1 = args[1], a2 = args[2];
2347        if (const_args[2]) {
2348        do_addi_64:
2349            if (a0 == a1) {
2350                if (a2 == (int16_t)a2) {
2351                    tcg_out_insn(s, RI, AGHI, a0, a2);
2352                    break;
2353                }
2354                if (a2 == (int32_t)a2) {
2355                    tcg_out_insn(s, RIL, AGFI, a0, a2);
2356                    break;
2357                }
2358                if (a2 == (uint32_t)a2) {
2359                    tcg_out_insn(s, RIL, ALGFI, a0, a2);
2360                    break;
2361                }
2362                if (-a2 == (uint32_t)-a2) {
2363                    tcg_out_insn(s, RIL, SLGFI, a0, -a2);
2364                    break;
2365                }
2366            }
2367            tcg_out_mem(s, RX_LA, RXY_LAY, a0, a1, TCG_REG_NONE, a2);
2368        } else if (a0 == a1) {
2369            tcg_out_insn(s, RRE, AGR, a0, a2);
2370        } else {
2371            tcg_out_insn(s, RX, LA, a0, a1, a2, 0);
2372        }
2373        break;
2374    case INDEX_op_sub_i64:
2375        a0 = args[0], a1 = args[1], a2 = args[2];
2376        if (const_args[2]) {
2377            a2 = -a2;
2378            goto do_addi_64;
2379        } else {
2380            tcg_out_insn(s, RRFa, SGRK, a0, a1, a2);
2381        }
2382        break;
2383
2384    case INDEX_op_and_i64:
2385        a0 = args[0], a1 = args[1], a2 = args[2];
2386        if (const_args[2]) {
2387            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2388            tgen_andi(s, TCG_TYPE_I64, args[0], args[2]);
2389        } else {
2390            tcg_out_insn(s, RRFa, NGRK, a0, a1, a2);
2391        }
2392        break;
2393    case INDEX_op_or_i64:
2394        a0 = args[0], a1 = args[1], a2 = args[2];
2395        if (const_args[2]) {
2396            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2397            tgen_ori(s, a0, a2);
2398        } else {
2399            tcg_out_insn(s, RRFa, OGRK, a0, a1, a2);
2400        }
2401        break;
2402    case INDEX_op_xor_i64:
2403        a0 = args[0], a1 = args[1], a2 = args[2];
2404        if (const_args[2]) {
2405            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2406            tgen_xori(s, a0, a2);
2407        } else {
2408            tcg_out_insn(s, RRFa, XGRK, a0, a1, a2);
2409        }
2410        break;
2411
2412    case INDEX_op_andc_i64:
2413        a0 = args[0], a1 = args[1], a2 = args[2];
2414        if (const_args[2]) {
2415            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2416            tgen_andi(s, TCG_TYPE_I64, a0, ~a2);
2417        } else {
2418            tcg_out_insn(s, RRFa, NCGRK, a0, a1, a2);
2419        }
2420        break;
2421    case INDEX_op_orc_i64:
2422        a0 = args[0], a1 = args[1], a2 = args[2];
2423        if (const_args[2]) {
2424            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2425            tgen_ori(s, a0, ~a2);
2426        } else {
2427            tcg_out_insn(s, RRFa, OCGRK, a0, a1, a2);
2428        }
2429        break;
2430    case INDEX_op_eqv_i64:
2431        a0 = args[0], a1 = args[1], a2 = args[2];
2432        if (const_args[2]) {
2433            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2434            tgen_xori(s, a0, ~a2);
2435        } else {
2436            tcg_out_insn(s, RRFa, NXGRK, a0, a1, a2);
2437        }
2438        break;
2439    case INDEX_op_nand_i64:
2440        tcg_out_insn(s, RRFa, NNGRK, args[0], args[1], args[2]);
2441        break;
2442    case INDEX_op_nor_i64:
2443        tcg_out_insn(s, RRFa, NOGRK, args[0], args[1], args[2]);
2444        break;
2445
2446    case INDEX_op_neg_i64:
2447        tcg_out_insn(s, RRE, LCGR, args[0], args[1]);
2448        break;
2449    case INDEX_op_not_i64:
2450        tcg_out_insn(s, RRFa, NOGRK, args[0], args[1], args[1]);
2451        break;
2452    case INDEX_op_bswap64_i64:
2453        tcg_out_insn(s, RRE, LRVGR, args[0], args[1]);
2454        break;
2455
2456    case INDEX_op_mul_i64:
2457        a0 = args[0], a1 = args[1], a2 = args[2];
2458        if (const_args[2]) {
2459            tcg_out_mov(s, TCG_TYPE_I64, a0, a1);
2460            if (a2 == (int16_t)a2) {
2461                tcg_out_insn(s, RI, MGHI, a0, a2);
2462            } else {
2463                tcg_out_insn(s, RIL, MSGFI, a0, a2);
2464            }
2465        } else if (a0 == a1) {
2466            tcg_out_insn(s, RRE, MSGR, a0, a2);
2467        } else {
2468            tcg_out_insn(s, RRFa, MSGRKC, a0, a1, a2);
2469        }
2470        break;
2471
2472    case INDEX_op_div2_i64:
2473        /*
2474         * ??? We get an unnecessary sign-extension of the dividend
2475         * into op0 with this definition, but as we do in fact always
2476         * produce both quotient and remainder using INDEX_op_div_i64
2477         * instead requires jumping through even more hoops.
2478         */
2479        tcg_debug_assert(args[0] == args[2]);
2480        tcg_debug_assert(args[1] == args[3]);
2481        tcg_debug_assert((args[1] & 1) == 0);
2482        tcg_debug_assert(args[0] == args[1] + 1);
2483        tcg_out_insn(s, RRE, DSGR, args[1], args[4]);
2484        break;
2485    case INDEX_op_divu2_i64:
2486        tcg_debug_assert(args[0] == args[2]);
2487        tcg_debug_assert(args[1] == args[3]);
2488        tcg_debug_assert((args[1] & 1) == 0);
2489        tcg_debug_assert(args[0] == args[1] + 1);
2490        tcg_out_insn(s, RRE, DLGR, args[1], args[4]);
2491        break;
2492    case INDEX_op_mulu2_i64:
2493        tcg_debug_assert(args[0] == args[2]);
2494        tcg_debug_assert((args[1] & 1) == 0);
2495        tcg_debug_assert(args[0] == args[1] + 1);
2496        tcg_out_insn(s, RRE, MLGR, args[1], args[3]);
2497        break;
2498    case INDEX_op_muls2_i64:
2499        tcg_debug_assert((args[1] & 1) == 0);
2500        tcg_debug_assert(args[0] == args[1] + 1);
2501        tcg_out_insn(s, RRFa, MGRK, args[1], args[2], args[3]);
2502        break;
2503
2504    case INDEX_op_shl_i64:
2505        op = RSY_SLLG;
2506    do_shift64:
2507        if (const_args[2]) {
2508            tcg_out_sh64(s, op, args[0], args[1], TCG_REG_NONE, args[2]);
2509        } else {
2510            tcg_out_sh64(s, op, args[0], args[1], args[2], 0);
2511        }
2512        break;
2513    case INDEX_op_shr_i64:
2514        op = RSY_SRLG;
2515        goto do_shift64;
2516    case INDEX_op_sar_i64:
2517        op = RSY_SRAG;
2518        goto do_shift64;
2519
2520    case INDEX_op_rotl_i64:
2521        if (const_args[2]) {
2522            tcg_out_sh64(s, RSY_RLLG, args[0], args[1],
2523                         TCG_REG_NONE, args[2]);
2524        } else {
2525            tcg_out_sh64(s, RSY_RLLG, args[0], args[1], args[2], 0);
2526        }
2527        break;
2528    case INDEX_op_rotr_i64:
2529        if (const_args[2]) {
2530            tcg_out_sh64(s, RSY_RLLG, args[0], args[1],
2531                         TCG_REG_NONE, (64 - args[2]) & 63);
2532        } else {
2533            /* We can use the smaller 32-bit negate because only the
2534               low 6 bits are examined for the rotate.  */
2535            tcg_out_insn(s, RR, LCR, TCG_TMP0, args[2]);
2536            tcg_out_sh64(s, RSY_RLLG, args[0], args[1], TCG_TMP0, 0);
2537        }
2538        break;
2539
2540    case INDEX_op_add2_i64:
2541        if (const_args[4]) {
2542            if ((int64_t)args[4] >= 0) {
2543                tcg_out_insn(s, RIL, ALGFI, args[0], args[4]);
2544            } else {
2545                tcg_out_insn(s, RIL, SLGFI, args[0], -args[4]);
2546            }
2547        } else {
2548            tcg_out_insn(s, RRE, ALGR, args[0], args[4]);
2549        }
2550        tcg_out_insn(s, RRE, ALCGR, args[1], args[5]);
2551        break;
2552    case INDEX_op_sub2_i64:
2553        if (const_args[4]) {
2554            if ((int64_t)args[4] >= 0) {
2555                tcg_out_insn(s, RIL, SLGFI, args[0], args[4]);
2556            } else {
2557                tcg_out_insn(s, RIL, ALGFI, args[0], -args[4]);
2558            }
2559        } else {
2560            tcg_out_insn(s, RRE, SLGR, args[0], args[4]);
2561        }
2562        tcg_out_insn(s, RRE, SLBGR, args[1], args[5]);
2563        break;
2564
2565    case INDEX_op_brcond_i64:
2566        tgen_brcond(s, TCG_TYPE_I64, args[2], args[0],
2567                    args[1], const_args[1], arg_label(args[3]));
2568        break;
2569    case INDEX_op_setcond_i64:
2570        tgen_setcond(s, TCG_TYPE_I64, args[3], args[0], args[1],
2571                     args[2], const_args[2]);
2572        break;
2573    case INDEX_op_movcond_i64:
2574        tgen_movcond(s, TCG_TYPE_I64, args[5], args[0], args[1],
2575                     args[2], const_args[2], args[3], const_args[3], args[4]);
2576        break;
2577
2578    OP_32_64(deposit):
2579        a0 = args[0], a1 = args[1], a2 = args[2];
2580        if (const_args[1]) {
2581            tgen_deposit(s, a0, a2, args[3], args[4], 1);
2582        } else {
2583            /* Since we can't support "0Z" as a constraint, we allow a1 in
2584               any register.  Fix things up as if a matching constraint.  */
2585            if (a0 != a1) {
2586                TCGType type = (opc == INDEX_op_deposit_i64);
2587                if (a0 == a2) {
2588                    tcg_out_mov(s, type, TCG_TMP0, a2);
2589                    a2 = TCG_TMP0;
2590                }
2591                tcg_out_mov(s, type, a0, a1);
2592            }
2593            tgen_deposit(s, a0, a2, args[3], args[4], 0);
2594        }
2595        break;
2596
2597    OP_32_64(extract):
2598        tgen_extract(s, args[0], args[1], args[2], args[3]);
2599        break;
2600
2601    case INDEX_op_clz_i64:
2602        tgen_clz(s, args[0], args[1], args[2], const_args[2]);
2603        break;
2604
2605    case INDEX_op_ctpop_i32:
2606        tgen_ctpop(s, TCG_TYPE_I32, args[0], args[1]);
2607        break;
2608    case INDEX_op_ctpop_i64:
2609        tgen_ctpop(s, TCG_TYPE_I64, args[0], args[1]);
2610        break;
2611
2612    case INDEX_op_mb:
2613        /* The host memory model is quite strong, we simply need to
2614           serialize the instruction stream.  */
2615        if (args[0] & TCG_MO_ST_LD) {
2616            /* fast-bcr-serialization facility (45) is present */
2617            tcg_out_insn(s, RR, BCR, 14, 0);
2618        }
2619        break;
2620
2621    case INDEX_op_mov_i32:  /* Always emitted via tcg_out_mov.  */
2622    case INDEX_op_mov_i64:
2623    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
2624    case INDEX_op_exit_tb:  /* Always emitted via tcg_out_exit_tb.  */
2625    case INDEX_op_goto_tb:  /* Always emitted via tcg_out_goto_tb.  */
2626    case INDEX_op_ext8s_i32:  /* Always emitted via tcg_reg_alloc_op.  */
2627    case INDEX_op_ext8s_i64:
2628    case INDEX_op_ext8u_i32:
2629    case INDEX_op_ext8u_i64:
2630    case INDEX_op_ext16s_i32:
2631    case INDEX_op_ext16s_i64:
2632    case INDEX_op_ext16u_i32:
2633    case INDEX_op_ext16u_i64:
2634    case INDEX_op_ext32s_i64:
2635    case INDEX_op_ext32u_i64:
2636    case INDEX_op_ext_i32_i64:
2637    case INDEX_op_extu_i32_i64:
2638    case INDEX_op_extrl_i64_i32:
2639    default:
2640        g_assert_not_reached();
2641    }
2642}
2643
2644static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
2645                            TCGReg dst, TCGReg src)
2646{
2647    if (is_general_reg(src)) {
2648        /* Replicate general register into two MO_64. */
2649        tcg_out_insn(s, VRRf, VLVGP, dst, src, src);
2650        if (vece == MO_64) {
2651            return true;
2652        }
2653        src = dst;
2654    }
2655
2656    /*
2657     * Recall that the "standard" integer, within a vector, is the
2658     * rightmost element of the leftmost doubleword, a-la VLLEZ.
2659     */
2660    tcg_out_insn(s, VRIc, VREP, dst, (8 >> vece) - 1, src, vece);
2661    return true;
2662}
2663
2664static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
2665                             TCGReg dst, TCGReg base, intptr_t offset)
2666{
2667    tcg_out_vrx_mem(s, VRX_VLREP, dst, base, TCG_REG_NONE, offset, vece);
2668    return true;
2669}
2670
2671static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
2672                             TCGReg dst, int64_t val)
2673{
2674    int i, mask, msb, lsb;
2675
2676    /* Look for int16_t elements.  */
2677    if (vece <= MO_16 ||
2678        (vece == MO_32 ? (int32_t)val : val) == (int16_t)val) {
2679        tcg_out_insn(s, VRIa, VREPI, dst, val, vece);
2680        return;
2681    }
2682
2683    /* Look for bit masks.  */
2684    if (vece == MO_32) {
2685        if (risbg_mask((int32_t)val)) {
2686            /* Handle wraparound by swapping msb and lsb.  */
2687            if ((val & 0x80000001u) == 0x80000001u) {
2688                msb = 32 - ctz32(~val);
2689                lsb = clz32(~val) - 1;
2690            } else {
2691                msb = clz32(val);
2692                lsb = 31 - ctz32(val);
2693            }
2694            tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_32);
2695            return;
2696        }
2697    } else {
2698        if (risbg_mask(val)) {
2699            /* Handle wraparound by swapping msb and lsb.  */
2700            if ((val & 0x8000000000000001ull) == 0x8000000000000001ull) {
2701                /* Handle wraparound by swapping msb and lsb.  */
2702                msb = 64 - ctz64(~val);
2703                lsb = clz64(~val) - 1;
2704            } else {
2705                msb = clz64(val);
2706                lsb = 63 - ctz64(val);
2707            }
2708            tcg_out_insn(s, VRIb, VGM, dst, msb, lsb, MO_64);
2709            return;
2710        }
2711    }
2712
2713    /* Look for all bytes 0x00 or 0xff.  */
2714    for (i = mask = 0; i < 8; i++) {
2715        uint8_t byte = val >> (i * 8);
2716        if (byte == 0xff) {
2717            mask |= 1 << i;
2718        } else if (byte != 0) {
2719            break;
2720        }
2721    }
2722    if (i == 8) {
2723        tcg_out_insn(s, VRIa, VGBM, dst, mask * 0x0101, 0);
2724        return;
2725    }
2726
2727    /* Otherwise, stuff it in the constant pool.  */
2728    tcg_out_insn(s, RIL, LARL, TCG_TMP0, 0);
2729    new_pool_label(s, val, R_390_PC32DBL, s->code_ptr - 2, 2);
2730    tcg_out_insn(s, VRX, VLREP, dst, TCG_TMP0, TCG_REG_NONE, 0, MO_64);
2731}
2732
2733static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
2734                           unsigned vecl, unsigned vece,
2735                           const TCGArg args[TCG_MAX_OP_ARGS],
2736                           const int const_args[TCG_MAX_OP_ARGS])
2737{
2738    TCGType type = vecl + TCG_TYPE_V64;
2739    TCGArg a0 = args[0], a1 = args[1], a2 = args[2];
2740
2741    switch (opc) {
2742    case INDEX_op_ld_vec:
2743        tcg_out_ld(s, type, a0, a1, a2);
2744        break;
2745    case INDEX_op_st_vec:
2746        tcg_out_st(s, type, a0, a1, a2);
2747        break;
2748    case INDEX_op_dupm_vec:
2749        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
2750        break;
2751
2752    case INDEX_op_abs_vec:
2753        tcg_out_insn(s, VRRa, VLP, a0, a1, vece);
2754        break;
2755    case INDEX_op_neg_vec:
2756        tcg_out_insn(s, VRRa, VLC, a0, a1, vece);
2757        break;
2758    case INDEX_op_not_vec:
2759        tcg_out_insn(s, VRRc, VNO, a0, a1, a1, 0);
2760        break;
2761
2762    case INDEX_op_add_vec:
2763        tcg_out_insn(s, VRRc, VA, a0, a1, a2, vece);
2764        break;
2765    case INDEX_op_sub_vec:
2766        tcg_out_insn(s, VRRc, VS, a0, a1, a2, vece);
2767        break;
2768    case INDEX_op_and_vec:
2769        tcg_out_insn(s, VRRc, VN, a0, a1, a2, 0);
2770        break;
2771    case INDEX_op_andc_vec:
2772        tcg_out_insn(s, VRRc, VNC, a0, a1, a2, 0);
2773        break;
2774    case INDEX_op_mul_vec:
2775        tcg_out_insn(s, VRRc, VML, a0, a1, a2, vece);
2776        break;
2777    case INDEX_op_or_vec:
2778        tcg_out_insn(s, VRRc, VO, a0, a1, a2, 0);
2779        break;
2780    case INDEX_op_orc_vec:
2781        tcg_out_insn(s, VRRc, VOC, a0, a1, a2, 0);
2782        break;
2783    case INDEX_op_xor_vec:
2784        tcg_out_insn(s, VRRc, VX, a0, a1, a2, 0);
2785        break;
2786    case INDEX_op_nand_vec:
2787        tcg_out_insn(s, VRRc, VNN, a0, a1, a2, 0);
2788        break;
2789    case INDEX_op_nor_vec:
2790        tcg_out_insn(s, VRRc, VNO, a0, a1, a2, 0);
2791        break;
2792    case INDEX_op_eqv_vec:
2793        tcg_out_insn(s, VRRc, VNX, a0, a1, a2, 0);
2794        break;
2795
2796    case INDEX_op_shli_vec:
2797        tcg_out_insn(s, VRSa, VESL, a0, a2, TCG_REG_NONE, a1, vece);
2798        break;
2799    case INDEX_op_shri_vec:
2800        tcg_out_insn(s, VRSa, VESRL, a0, a2, TCG_REG_NONE, a1, vece);
2801        break;
2802    case INDEX_op_sari_vec:
2803        tcg_out_insn(s, VRSa, VESRA, a0, a2, TCG_REG_NONE, a1, vece);
2804        break;
2805    case INDEX_op_rotli_vec:
2806        tcg_out_insn(s, VRSa, VERLL, a0, a2, TCG_REG_NONE, a1, vece);
2807        break;
2808    case INDEX_op_shls_vec:
2809        tcg_out_insn(s, VRSa, VESL, a0, 0, a2, a1, vece);
2810        break;
2811    case INDEX_op_shrs_vec:
2812        tcg_out_insn(s, VRSa, VESRL, a0, 0, a2, a1, vece);
2813        break;
2814    case INDEX_op_sars_vec:
2815        tcg_out_insn(s, VRSa, VESRA, a0, 0, a2, a1, vece);
2816        break;
2817    case INDEX_op_rotls_vec:
2818        tcg_out_insn(s, VRSa, VERLL, a0, 0, a2, a1, vece);
2819        break;
2820    case INDEX_op_shlv_vec:
2821        tcg_out_insn(s, VRRc, VESLV, a0, a1, a2, vece);
2822        break;
2823    case INDEX_op_shrv_vec:
2824        tcg_out_insn(s, VRRc, VESRLV, a0, a1, a2, vece);
2825        break;
2826    case INDEX_op_sarv_vec:
2827        tcg_out_insn(s, VRRc, VESRAV, a0, a1, a2, vece);
2828        break;
2829    case INDEX_op_rotlv_vec:
2830        tcg_out_insn(s, VRRc, VERLLV, a0, a1, a2, vece);
2831        break;
2832
2833    case INDEX_op_smin_vec:
2834        tcg_out_insn(s, VRRc, VMN, a0, a1, a2, vece);
2835        break;
2836    case INDEX_op_smax_vec:
2837        tcg_out_insn(s, VRRc, VMX, a0, a1, a2, vece);
2838        break;
2839    case INDEX_op_umin_vec:
2840        tcg_out_insn(s, VRRc, VMNL, a0, a1, a2, vece);
2841        break;
2842    case INDEX_op_umax_vec:
2843        tcg_out_insn(s, VRRc, VMXL, a0, a1, a2, vece);
2844        break;
2845
2846    case INDEX_op_bitsel_vec:
2847        tcg_out_insn(s, VRRe, VSEL, a0, a2, args[3], a1);
2848        break;
2849
2850    case INDEX_op_cmp_vec:
2851        switch ((TCGCond)args[3]) {
2852        case TCG_COND_EQ:
2853            tcg_out_insn(s, VRRc, VCEQ, a0, a1, a2, vece);
2854            break;
2855        case TCG_COND_GT:
2856            tcg_out_insn(s, VRRc, VCH, a0, a1, a2, vece);
2857            break;
2858        case TCG_COND_GTU:
2859            tcg_out_insn(s, VRRc, VCHL, a0, a1, a2, vece);
2860            break;
2861        default:
2862            g_assert_not_reached();
2863        }
2864        break;
2865
2866    case INDEX_op_s390_vuph_vec:
2867        tcg_out_insn(s, VRRa, VUPH, a0, a1, vece);
2868        break;
2869    case INDEX_op_s390_vupl_vec:
2870        tcg_out_insn(s, VRRa, VUPL, a0, a1, vece);
2871        break;
2872    case INDEX_op_s390_vpks_vec:
2873        tcg_out_insn(s, VRRc, VPKS, a0, a1, a2, vece);
2874        break;
2875
2876    case INDEX_op_mov_vec:   /* Always emitted via tcg_out_mov.  */
2877    case INDEX_op_dup_vec:   /* Always emitted via tcg_out_dup_vec.  */
2878    default:
2879        g_assert_not_reached();
2880    }
2881}
2882
2883int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
2884{
2885    switch (opc) {
2886    case INDEX_op_abs_vec:
2887    case INDEX_op_add_vec:
2888    case INDEX_op_and_vec:
2889    case INDEX_op_andc_vec:
2890    case INDEX_op_bitsel_vec:
2891    case INDEX_op_eqv_vec:
2892    case INDEX_op_nand_vec:
2893    case INDEX_op_neg_vec:
2894    case INDEX_op_nor_vec:
2895    case INDEX_op_not_vec:
2896    case INDEX_op_or_vec:
2897    case INDEX_op_orc_vec:
2898    case INDEX_op_rotli_vec:
2899    case INDEX_op_rotls_vec:
2900    case INDEX_op_rotlv_vec:
2901    case INDEX_op_sari_vec:
2902    case INDEX_op_sars_vec:
2903    case INDEX_op_sarv_vec:
2904    case INDEX_op_shli_vec:
2905    case INDEX_op_shls_vec:
2906    case INDEX_op_shlv_vec:
2907    case INDEX_op_shri_vec:
2908    case INDEX_op_shrs_vec:
2909    case INDEX_op_shrv_vec:
2910    case INDEX_op_smax_vec:
2911    case INDEX_op_smin_vec:
2912    case INDEX_op_sub_vec:
2913    case INDEX_op_umax_vec:
2914    case INDEX_op_umin_vec:
2915    case INDEX_op_xor_vec:
2916        return 1;
2917    case INDEX_op_cmp_vec:
2918    case INDEX_op_cmpsel_vec:
2919    case INDEX_op_rotrv_vec:
2920        return -1;
2921    case INDEX_op_mul_vec:
2922        return vece < MO_64;
2923    case INDEX_op_ssadd_vec:
2924    case INDEX_op_sssub_vec:
2925        return vece < MO_64 ? -1 : 0;
2926    default:
2927        return 0;
2928    }
2929}
2930
2931static bool expand_vec_cmp_noinv(TCGType type, unsigned vece, TCGv_vec v0,
2932                                 TCGv_vec v1, TCGv_vec v2, TCGCond cond)
2933{
2934    bool need_swap = false, need_inv = false;
2935
2936    switch (cond) {
2937    case TCG_COND_EQ:
2938    case TCG_COND_GT:
2939    case TCG_COND_GTU:
2940        break;
2941    case TCG_COND_NE:
2942    case TCG_COND_LE:
2943    case TCG_COND_LEU:
2944        need_inv = true;
2945        break;
2946    case TCG_COND_LT:
2947    case TCG_COND_LTU:
2948        need_swap = true;
2949        break;
2950    case TCG_COND_GE:
2951    case TCG_COND_GEU:
2952        need_swap = need_inv = true;
2953        break;
2954    default:
2955        g_assert_not_reached();
2956    }
2957
2958    if (need_inv) {
2959        cond = tcg_invert_cond(cond);
2960    }
2961    if (need_swap) {
2962        TCGv_vec t1;
2963        t1 = v1, v1 = v2, v2 = t1;
2964        cond = tcg_swap_cond(cond);
2965    }
2966
2967    vec_gen_4(INDEX_op_cmp_vec, type, vece, tcgv_vec_arg(v0),
2968              tcgv_vec_arg(v1), tcgv_vec_arg(v2), cond);
2969
2970    return need_inv;
2971}
2972
2973static void expand_vec_cmp(TCGType type, unsigned vece, TCGv_vec v0,
2974                           TCGv_vec v1, TCGv_vec v2, TCGCond cond)
2975{
2976    if (expand_vec_cmp_noinv(type, vece, v0, v1, v2, cond)) {
2977        tcg_gen_not_vec(vece, v0, v0);
2978    }
2979}
2980
2981static void expand_vec_cmpsel(TCGType type, unsigned vece, TCGv_vec v0,
2982                              TCGv_vec c1, TCGv_vec c2,
2983                              TCGv_vec v3, TCGv_vec v4, TCGCond cond)
2984{
2985    TCGv_vec t = tcg_temp_new_vec(type);
2986
2987    if (expand_vec_cmp_noinv(type, vece, t, c1, c2, cond)) {
2988        /* Invert the sense of the compare by swapping arguments.  */
2989        tcg_gen_bitsel_vec(vece, v0, t, v4, v3);
2990    } else {
2991        tcg_gen_bitsel_vec(vece, v0, t, v3, v4);
2992    }
2993    tcg_temp_free_vec(t);
2994}
2995
2996static void expand_vec_sat(TCGType type, unsigned vece, TCGv_vec v0,
2997                           TCGv_vec v1, TCGv_vec v2, TCGOpcode add_sub_opc)
2998{
2999    TCGv_vec h1 = tcg_temp_new_vec(type);
3000    TCGv_vec h2 = tcg_temp_new_vec(type);
3001    TCGv_vec l1 = tcg_temp_new_vec(type);
3002    TCGv_vec l2 = tcg_temp_new_vec(type);
3003
3004    tcg_debug_assert (vece < MO_64);
3005
3006    /* Unpack with sign-extension. */
3007    vec_gen_2(INDEX_op_s390_vuph_vec, type, vece,
3008              tcgv_vec_arg(h1), tcgv_vec_arg(v1));
3009    vec_gen_2(INDEX_op_s390_vuph_vec, type, vece,
3010              tcgv_vec_arg(h2), tcgv_vec_arg(v2));
3011
3012    vec_gen_2(INDEX_op_s390_vupl_vec, type, vece,
3013              tcgv_vec_arg(l1), tcgv_vec_arg(v1));
3014    vec_gen_2(INDEX_op_s390_vupl_vec, type, vece,
3015              tcgv_vec_arg(l2), tcgv_vec_arg(v2));
3016
3017    /* Arithmetic on a wider element size. */
3018    vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(h1),
3019              tcgv_vec_arg(h1), tcgv_vec_arg(h2));
3020    vec_gen_3(add_sub_opc, type, vece + 1, tcgv_vec_arg(l1),
3021              tcgv_vec_arg(l1), tcgv_vec_arg(l2));
3022
3023    /* Pack with saturation. */
3024    vec_gen_3(INDEX_op_s390_vpks_vec, type, vece + 1,
3025              tcgv_vec_arg(v0), tcgv_vec_arg(h1), tcgv_vec_arg(l1));
3026
3027    tcg_temp_free_vec(h1);
3028    tcg_temp_free_vec(h2);
3029    tcg_temp_free_vec(l1);
3030    tcg_temp_free_vec(l2);
3031}
3032
3033void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
3034                       TCGArg a0, ...)
3035{
3036    va_list va;
3037    TCGv_vec v0, v1, v2, v3, v4, t0;
3038
3039    va_start(va, a0);
3040    v0 = temp_tcgv_vec(arg_temp(a0));
3041    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3042    v2 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3043
3044    switch (opc) {
3045    case INDEX_op_cmp_vec:
3046        expand_vec_cmp(type, vece, v0, v1, v2, va_arg(va, TCGArg));
3047        break;
3048
3049    case INDEX_op_cmpsel_vec:
3050        v3 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3051        v4 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3052        expand_vec_cmpsel(type, vece, v0, v1, v2, v3, v4, va_arg(va, TCGArg));
3053        break;
3054
3055    case INDEX_op_rotrv_vec:
3056        t0 = tcg_temp_new_vec(type);
3057        tcg_gen_neg_vec(vece, t0, v2);
3058        tcg_gen_rotlv_vec(vece, v0, v1, t0);
3059        tcg_temp_free_vec(t0);
3060        break;
3061
3062    case INDEX_op_ssadd_vec:
3063        expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_add_vec);
3064        break;
3065    case INDEX_op_sssub_vec:
3066        expand_vec_sat(type, vece, v0, v1, v2, INDEX_op_sub_vec);
3067        break;
3068
3069    default:
3070        g_assert_not_reached();
3071    }
3072    va_end(va);
3073}
3074
3075static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
3076{
3077    switch (op) {
3078    case INDEX_op_goto_ptr:
3079        return C_O0_I1(r);
3080
3081    case INDEX_op_ld8u_i32:
3082    case INDEX_op_ld8u_i64:
3083    case INDEX_op_ld8s_i32:
3084    case INDEX_op_ld8s_i64:
3085    case INDEX_op_ld16u_i32:
3086    case INDEX_op_ld16u_i64:
3087    case INDEX_op_ld16s_i32:
3088    case INDEX_op_ld16s_i64:
3089    case INDEX_op_ld_i32:
3090    case INDEX_op_ld32u_i64:
3091    case INDEX_op_ld32s_i64:
3092    case INDEX_op_ld_i64:
3093        return C_O1_I1(r, r);
3094
3095    case INDEX_op_st8_i32:
3096    case INDEX_op_st8_i64:
3097    case INDEX_op_st16_i32:
3098    case INDEX_op_st16_i64:
3099    case INDEX_op_st_i32:
3100    case INDEX_op_st32_i64:
3101    case INDEX_op_st_i64:
3102        return C_O0_I2(r, r);
3103
3104    case INDEX_op_add_i32:
3105    case INDEX_op_add_i64:
3106    case INDEX_op_shl_i64:
3107    case INDEX_op_shr_i64:
3108    case INDEX_op_sar_i64:
3109    case INDEX_op_rotl_i32:
3110    case INDEX_op_rotl_i64:
3111    case INDEX_op_rotr_i32:
3112    case INDEX_op_rotr_i64:
3113    case INDEX_op_setcond_i32:
3114        return C_O1_I2(r, r, ri);
3115    case INDEX_op_setcond_i64:
3116        return C_O1_I2(r, r, rA);
3117
3118    case INDEX_op_clz_i64:
3119        return C_O1_I2(r, r, rI);
3120
3121    case INDEX_op_sub_i32:
3122    case INDEX_op_sub_i64:
3123    case INDEX_op_and_i32:
3124    case INDEX_op_or_i32:
3125    case INDEX_op_xor_i32:
3126        return C_O1_I2(r, r, ri);
3127    case INDEX_op_and_i64:
3128        return C_O1_I2(r, r, rNKR);
3129    case INDEX_op_or_i64:
3130    case INDEX_op_xor_i64:
3131        return C_O1_I2(r, r, rK);
3132
3133    case INDEX_op_andc_i32:
3134    case INDEX_op_orc_i32:
3135    case INDEX_op_eqv_i32:
3136        return C_O1_I2(r, r, ri);
3137    case INDEX_op_andc_i64:
3138        return C_O1_I2(r, r, rKR);
3139    case INDEX_op_orc_i64:
3140    case INDEX_op_eqv_i64:
3141        return C_O1_I2(r, r, rNK);
3142
3143    case INDEX_op_nand_i32:
3144    case INDEX_op_nand_i64:
3145    case INDEX_op_nor_i32:
3146    case INDEX_op_nor_i64:
3147        return C_O1_I2(r, r, r);
3148
3149    case INDEX_op_mul_i32:
3150        return (HAVE_FACILITY(MISC_INSN_EXT2)
3151                ? C_O1_I2(r, r, ri)
3152                : C_O1_I2(r, 0, ri));
3153    case INDEX_op_mul_i64:
3154        return (HAVE_FACILITY(MISC_INSN_EXT2)
3155                ? C_O1_I2(r, r, rJ)
3156                : C_O1_I2(r, 0, rJ));
3157
3158    case INDEX_op_shl_i32:
3159    case INDEX_op_shr_i32:
3160    case INDEX_op_sar_i32:
3161        return C_O1_I2(r, r, ri);
3162
3163    case INDEX_op_brcond_i32:
3164        return C_O0_I2(r, ri);
3165    case INDEX_op_brcond_i64:
3166        return C_O0_I2(r, rA);
3167
3168    case INDEX_op_bswap16_i32:
3169    case INDEX_op_bswap16_i64:
3170    case INDEX_op_bswap32_i32:
3171    case INDEX_op_bswap32_i64:
3172    case INDEX_op_bswap64_i64:
3173    case INDEX_op_neg_i32:
3174    case INDEX_op_neg_i64:
3175    case INDEX_op_not_i32:
3176    case INDEX_op_not_i64:
3177    case INDEX_op_ext8s_i32:
3178    case INDEX_op_ext8s_i64:
3179    case INDEX_op_ext8u_i32:
3180    case INDEX_op_ext8u_i64:
3181    case INDEX_op_ext16s_i32:
3182    case INDEX_op_ext16s_i64:
3183    case INDEX_op_ext16u_i32:
3184    case INDEX_op_ext16u_i64:
3185    case INDEX_op_ext32s_i64:
3186    case INDEX_op_ext32u_i64:
3187    case INDEX_op_ext_i32_i64:
3188    case INDEX_op_extu_i32_i64:
3189    case INDEX_op_extract_i32:
3190    case INDEX_op_extract_i64:
3191    case INDEX_op_ctpop_i32:
3192    case INDEX_op_ctpop_i64:
3193        return C_O1_I1(r, r);
3194
3195    case INDEX_op_qemu_ld_a32_i32:
3196    case INDEX_op_qemu_ld_a64_i32:
3197    case INDEX_op_qemu_ld_a32_i64:
3198    case INDEX_op_qemu_ld_a64_i64:
3199        return C_O1_I1(r, r);
3200    case INDEX_op_qemu_st_a32_i64:
3201    case INDEX_op_qemu_st_a64_i64:
3202    case INDEX_op_qemu_st_a32_i32:
3203    case INDEX_op_qemu_st_a64_i32:
3204        return C_O0_I2(r, r);
3205    case INDEX_op_qemu_ld_a32_i128:
3206    case INDEX_op_qemu_ld_a64_i128:
3207        return C_O2_I1(o, m, r);
3208    case INDEX_op_qemu_st_a32_i128:
3209    case INDEX_op_qemu_st_a64_i128:
3210        return C_O0_I3(o, m, r);
3211
3212    case INDEX_op_deposit_i32:
3213    case INDEX_op_deposit_i64:
3214        return C_O1_I2(r, rZ, r);
3215
3216    case INDEX_op_movcond_i32:
3217        return C_O1_I4(r, r, ri, rI, r);
3218    case INDEX_op_movcond_i64:
3219        return C_O1_I4(r, r, rA, rI, r);
3220
3221    case INDEX_op_div2_i32:
3222    case INDEX_op_div2_i64:
3223    case INDEX_op_divu2_i32:
3224    case INDEX_op_divu2_i64:
3225        return C_O2_I3(o, m, 0, 1, r);
3226
3227    case INDEX_op_mulu2_i64:
3228        return C_O2_I2(o, m, 0, r);
3229    case INDEX_op_muls2_i64:
3230        return C_O2_I2(o, m, r, r);
3231
3232    case INDEX_op_add2_i32:
3233    case INDEX_op_sub2_i32:
3234        return C_O2_I4(r, r, 0, 1, ri, r);
3235
3236    case INDEX_op_add2_i64:
3237    case INDEX_op_sub2_i64:
3238        return C_O2_I4(r, r, 0, 1, rA, r);
3239
3240    case INDEX_op_st_vec:
3241        return C_O0_I2(v, r);
3242    case INDEX_op_ld_vec:
3243    case INDEX_op_dupm_vec:
3244        return C_O1_I1(v, r);
3245    case INDEX_op_dup_vec:
3246        return C_O1_I1(v, vr);
3247    case INDEX_op_abs_vec:
3248    case INDEX_op_neg_vec:
3249    case INDEX_op_not_vec:
3250    case INDEX_op_rotli_vec:
3251    case INDEX_op_sari_vec:
3252    case INDEX_op_shli_vec:
3253    case INDEX_op_shri_vec:
3254    case INDEX_op_s390_vuph_vec:
3255    case INDEX_op_s390_vupl_vec:
3256        return C_O1_I1(v, v);
3257    case INDEX_op_add_vec:
3258    case INDEX_op_sub_vec:
3259    case INDEX_op_and_vec:
3260    case INDEX_op_andc_vec:
3261    case INDEX_op_or_vec:
3262    case INDEX_op_orc_vec:
3263    case INDEX_op_xor_vec:
3264    case INDEX_op_nand_vec:
3265    case INDEX_op_nor_vec:
3266    case INDEX_op_eqv_vec:
3267    case INDEX_op_cmp_vec:
3268    case INDEX_op_mul_vec:
3269    case INDEX_op_rotlv_vec:
3270    case INDEX_op_rotrv_vec:
3271    case INDEX_op_shlv_vec:
3272    case INDEX_op_shrv_vec:
3273    case INDEX_op_sarv_vec:
3274    case INDEX_op_smax_vec:
3275    case INDEX_op_smin_vec:
3276    case INDEX_op_umax_vec:
3277    case INDEX_op_umin_vec:
3278    case INDEX_op_s390_vpks_vec:
3279        return C_O1_I2(v, v, v);
3280    case INDEX_op_rotls_vec:
3281    case INDEX_op_shls_vec:
3282    case INDEX_op_shrs_vec:
3283    case INDEX_op_sars_vec:
3284        return C_O1_I2(v, v, r);
3285    case INDEX_op_bitsel_vec:
3286        return C_O1_I3(v, v, v, v);
3287
3288    default:
3289        g_assert_not_reached();
3290    }
3291}
3292
3293/*
3294 * Mainline glibc added HWCAP_S390_VX before it was kernel abi.
3295 * Some distros have fixed this up locally, others have not.
3296 */
3297#ifndef HWCAP_S390_VXRS
3298#define HWCAP_S390_VXRS 2048
3299#endif
3300
3301static void query_s390_facilities(void)
3302{
3303    unsigned long hwcap = qemu_getauxval(AT_HWCAP);
3304    const char *which;
3305
3306    /* Is STORE FACILITY LIST EXTENDED available?  Honestly, I believe this
3307       is present on all 64-bit systems, but let's check for it anyway.  */
3308    if (hwcap & HWCAP_S390_STFLE) {
3309        register int r0 __asm__("0") = ARRAY_SIZE(s390_facilities) - 1;
3310        register void *r1 __asm__("1") = s390_facilities;
3311
3312        /* stfle 0(%r1) */
3313        asm volatile(".word 0xb2b0,0x1000"
3314                     : "=r"(r0) : "r"(r0), "r"(r1) : "memory", "cc");
3315    }
3316
3317    /*
3318     * Use of vector registers requires os support beyond the facility bit.
3319     * If the kernel does not advertise support, disable the facility bits.
3320     * There is nothing else we currently care about in the 3rd word, so
3321     * disable VECTOR with one store.
3322     */
3323    if (!(hwcap & HWCAP_S390_VXRS)) {
3324        s390_facilities[2] = 0;
3325    }
3326
3327    /*
3328     * Minimum supported cpu revision is z196.
3329     * Check for all required facilities.
3330     * ZARCH_ACTIVE is done via preprocessor check for 64-bit.
3331     */
3332    if (!HAVE_FACILITY(LONG_DISP)) {
3333        which = "long-displacement";
3334        goto fail;
3335    }
3336    if (!HAVE_FACILITY(EXT_IMM)) {
3337        which = "extended-immediate";
3338        goto fail;
3339    }
3340    if (!HAVE_FACILITY(GEN_INST_EXT)) {
3341        which = "general-instructions-extension";
3342        goto fail;
3343    }
3344    /*
3345     * Facility 45 is a big bin that contains: distinct-operands,
3346     * fast-BCR-serialization, high-word, population-count,
3347     * interlocked-access-1, and load/store-on-condition-1
3348     */
3349    if (!HAVE_FACILITY(45)) {
3350        which = "45";
3351        goto fail;
3352    }
3353    return;
3354
3355 fail:
3356    error_report("%s: missing required facility %s", __func__, which);
3357    exit(EXIT_FAILURE);
3358}
3359
3360static void tcg_target_init(TCGContext *s)
3361{
3362    query_s390_facilities();
3363
3364    tcg_target_available_regs[TCG_TYPE_I32] = 0xffff;
3365    tcg_target_available_regs[TCG_TYPE_I64] = 0xffff;
3366    if (HAVE_FACILITY(VECTOR)) {
3367        tcg_target_available_regs[TCG_TYPE_V64] = 0xffffffff00000000ull;
3368        tcg_target_available_regs[TCG_TYPE_V128] = 0xffffffff00000000ull;
3369    }
3370
3371    tcg_target_call_clobber_regs = 0;
3372    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
3373    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1);
3374    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
3375    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
3376    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R4);
3377    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R5);
3378    /* The r6 register is technically call-saved, but it's also a parameter
3379       register, so it can get killed by setup for the qemu_st helper.  */
3380    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R6);
3381    /* The return register can be considered call-clobbered.  */
3382    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14);
3383
3384    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V0);
3385    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V1);
3386    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V2);
3387    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V3);
3388    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V4);
3389    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V5);
3390    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V6);
3391    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V7);
3392    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V16);
3393    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V17);
3394    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V18);
3395    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V19);
3396    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V20);
3397    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V21);
3398    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V22);
3399    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V23);
3400    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V24);
3401    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V25);
3402    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V26);
3403    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V27);
3404    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V28);
3405    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V29);
3406    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V30);
3407    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_V31);
3408
3409    s->reserved_regs = 0;
3410    tcg_regset_set_reg(s->reserved_regs, TCG_TMP0);
3411    /* XXX many insns can't be used with R0, so we better avoid it for now */
3412    tcg_regset_set_reg(s->reserved_regs, TCG_REG_R0);
3413    tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
3414}
3415
3416#define FRAME_SIZE  ((int)(TCG_TARGET_CALL_STACK_OFFSET          \
3417                           + TCG_STATIC_CALL_ARGS_SIZE           \
3418                           + CPU_TEMP_BUF_NLONGS * sizeof(long)))
3419
3420static void tcg_target_qemu_prologue(TCGContext *s)
3421{
3422    /* stmg %r6,%r15,48(%r15) (save registers) */
3423    tcg_out_insn(s, RXY, STMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15, 48);
3424
3425    /* aghi %r15,-frame_size */
3426    tcg_out_insn(s, RI, AGHI, TCG_REG_R15, -FRAME_SIZE);
3427
3428    tcg_set_frame(s, TCG_REG_CALL_STACK,
3429                  TCG_STATIC_CALL_ARGS_SIZE + TCG_TARGET_CALL_STACK_OFFSET,
3430                  CPU_TEMP_BUF_NLONGS * sizeof(long));
3431
3432#ifndef CONFIG_SOFTMMU
3433    if (guest_base >= 0x80000) {
3434        tcg_out_movi(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base);
3435        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
3436    }
3437#endif
3438
3439    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
3440
3441    /* br %r3 (go to TB) */
3442    tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, tcg_target_call_iarg_regs[1]);
3443
3444    /*
3445     * Return path for goto_ptr. Set return value to 0, a-la exit_tb,
3446     * and fall through to the rest of the epilogue.
3447     */
3448    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
3449    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R2, 0);
3450
3451    /* TB epilogue */
3452    tb_ret_addr = tcg_splitwx_to_rx(s->code_ptr);
3453
3454    /* lmg %r6,%r15,fs+48(%r15) (restore registers) */
3455    tcg_out_insn(s, RXY, LMG, TCG_REG_R6, TCG_REG_R15, TCG_REG_R15,
3456                 FRAME_SIZE + 48);
3457
3458    /* br %r14 (return) */
3459    tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_R14);
3460}
3461
3462static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
3463{
3464    memset(p, 0x07, count * sizeof(tcg_insn_unit));
3465}
3466
3467typedef struct {
3468    DebugFrameHeader h;
3469    uint8_t fde_def_cfa[4];
3470    uint8_t fde_reg_ofs[18];
3471} DebugFrame;
3472
3473/* We're expecting a 2 byte uleb128 encoded value.  */
3474QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3475
3476#define ELF_HOST_MACHINE  EM_S390
3477
3478static const DebugFrame debug_frame = {
3479    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3480    .h.cie.id = -1,
3481    .h.cie.version = 1,
3482    .h.cie.code_align = 1,
3483    .h.cie.data_align = 8,                /* sleb128 8 */
3484    .h.cie.return_column = TCG_REG_R14,
3485
3486    /* Total FDE size does not include the "len" member.  */
3487    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
3488
3489    .fde_def_cfa = {
3490        12, TCG_REG_CALL_STACK,         /* DW_CFA_def_cfa %r15, ... */
3491        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3492        (FRAME_SIZE >> 7)
3493    },
3494    .fde_reg_ofs = {
3495        0x86, 6,                        /* DW_CFA_offset, %r6, 48 */
3496        0x87, 7,                        /* DW_CFA_offset, %r7, 56 */
3497        0x88, 8,                        /* DW_CFA_offset, %r8, 64 */
3498        0x89, 9,                        /* DW_CFA_offset, %r92, 72 */
3499        0x8a, 10,                       /* DW_CFA_offset, %r10, 80 */
3500        0x8b, 11,                       /* DW_CFA_offset, %r11, 88 */
3501        0x8c, 12,                       /* DW_CFA_offset, %r12, 96 */
3502        0x8d, 13,                       /* DW_CFA_offset, %r13, 104 */
3503        0x8e, 14,                       /* DW_CFA_offset, %r14, 112 */
3504    }
3505};
3506
3507void tcg_register_jit(const void *buf, size_t buf_size)
3508{
3509    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3510}
3511