1//===-- PPCJITInfo.cpp - Implement the JIT interfaces for the PowerPC -----===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This file implements the JIT interfaces for the 32-bit PowerPC target.
11//
12//===----------------------------------------------------------------------===//
13
14#define DEBUG_TYPE "jit"
15#include "PPCJITInfo.h"
16#include "PPCRelocations.h"
17#include "PPCTargetMachine.h"
18#include "llvm/Function.h"
19#include "llvm/Support/Memory.h"
20#include "llvm/Support/Debug.h"
21#include "llvm/Support/ErrorHandling.h"
22#include "llvm/Support/raw_ostream.h"
23using namespace llvm;
24
25static TargetJITInfo::JITCompilerFn JITCompilerFunction;
26
27#define BUILD_ADDIS(RD,RS,IMM16) \
28  ((15 << 26) | ((RD) << 21) | ((RS) << 16) | ((IMM16) & 65535))
29#define BUILD_ORI(RD,RS,UIMM16) \
30  ((24 << 26) | ((RS) << 21) | ((RD) << 16) | ((UIMM16) & 65535))
31#define BUILD_ORIS(RD,RS,UIMM16) \
32  ((25 << 26) | ((RS) << 21) | ((RD) << 16) | ((UIMM16) & 65535))
33#define BUILD_RLDICR(RD,RS,SH,ME) \
34  ((30 << 26) | ((RS) << 21) | ((RD) << 16) | (((SH) & 31) << 11) | \
35   (((ME) & 63) << 6) | (1 << 2) | ((((SH) >> 5) & 1) << 1))
36#define BUILD_MTSPR(RS,SPR)      \
37  ((31 << 26) | ((RS) << 21) | ((SPR) << 16) | (467 << 1))
38#define BUILD_BCCTRx(BO,BI,LINK) \
39  ((19 << 26) | ((BO) << 21) | ((BI) << 16) | (528 << 1) | ((LINK) & 1))
40#define BUILD_B(TARGET, LINK) \
41  ((18 << 26) | (((TARGET) & 0x00FFFFFF) << 2) | ((LINK) & 1))
42
43// Pseudo-ops
44#define BUILD_LIS(RD,IMM16)    BUILD_ADDIS(RD,0,IMM16)
45#define BUILD_SLDI(RD,RS,IMM6) BUILD_RLDICR(RD,RS,IMM6,63-IMM6)
46#define BUILD_MTCTR(RS)        BUILD_MTSPR(RS,9)
47#define BUILD_BCTR(LINK)       BUILD_BCCTRx(20,0,LINK)
48
49static void EmitBranchToAt(uint64_t At, uint64_t To, bool isCall, bool is64Bit){
50  intptr_t Offset = ((intptr_t)To - (intptr_t)At) >> 2;
51  unsigned *AtI = (unsigned*)(intptr_t)At;
52
53  if (Offset >= -(1 << 23) && Offset < (1 << 23)) {   // In range?
54    AtI[0] = BUILD_B(Offset, isCall);     // b/bl target
55  } else if (!is64Bit) {
56    AtI[0] = BUILD_LIS(12, To >> 16);     // lis r12, hi16(address)
57    AtI[1] = BUILD_ORI(12, 12, To);       // ori r12, r12, lo16(address)
58    AtI[2] = BUILD_MTCTR(12);             // mtctr r12
59    AtI[3] = BUILD_BCTR(isCall);          // bctr/bctrl
60  } else {
61    AtI[0] = BUILD_LIS(12, To >> 48);      // lis r12, hi16(address)
62    AtI[1] = BUILD_ORI(12, 12, To >> 32);  // ori r12, r12, lo16(address)
63    AtI[2] = BUILD_SLDI(12, 12, 32);       // sldi r12, r12, 32
64    AtI[3] = BUILD_ORIS(12, 12, To >> 16); // oris r12, r12, hi16(address)
65    AtI[4] = BUILD_ORI(12, 12, To);        // ori r12, r12, lo16(address)
66    AtI[5] = BUILD_MTCTR(12);              // mtctr r12
67    AtI[6] = BUILD_BCTR(isCall);           // bctr/bctrl
68  }
69}
70
71extern "C" void PPC32CompilationCallback();
72extern "C" void PPC64CompilationCallback();
73
74#if (defined(__POWERPC__) || defined (__ppc__) || defined(_POWER)) && \
75    !(defined(__ppc64__) || defined(__FreeBSD__))
76// CompilationCallback stub - We can't use a C function with inline assembly in
77// it, because we the prolog/epilog inserted by GCC won't work for us.  Instead,
78// write our own wrapper, which does things our way, so we have complete control
79// over register saving and restoring.
80asm(
81    ".text\n"
82    ".align 2\n"
83    ".globl _PPC32CompilationCallback\n"
84"_PPC32CompilationCallback:\n"
85    // Make space for 8 ints r[3-10] and 13 doubles f[1-13] and the
86    // FIXME: need to save v[0-19] for altivec?
87    // FIXME: could shrink frame
88    // Set up a proper stack frame
89    // FIXME Layout
90    //   PowerPC32 ABI linkage    -  24 bytes
91    //                 parameters -  32 bytes
92    //   13 double registers      - 104 bytes
93    //   8 int registers          -  32 bytes
94    "mflr r0\n"
95    "stw r0,  8(r1)\n"
96    "stwu r1, -208(r1)\n"
97    // Save all int arg registers
98    "stw r10, 204(r1)\n"    "stw r9,  200(r1)\n"
99    "stw r8,  196(r1)\n"    "stw r7,  192(r1)\n"
100    "stw r6,  188(r1)\n"    "stw r5,  184(r1)\n"
101    "stw r4,  180(r1)\n"    "stw r3,  176(r1)\n"
102    // Save all call-clobbered FP regs.
103    "stfd f13, 168(r1)\n"   "stfd f12, 160(r1)\n"
104    "stfd f11, 152(r1)\n"   "stfd f10, 144(r1)\n"
105    "stfd f9,  136(r1)\n"   "stfd f8,  128(r1)\n"
106    "stfd f7,  120(r1)\n"   "stfd f6,  112(r1)\n"
107    "stfd f5,  104(r1)\n"   "stfd f4,   96(r1)\n"
108    "stfd f3,   88(r1)\n"   "stfd f2,   80(r1)\n"
109    "stfd f1,   72(r1)\n"
110    // Arguments to Compilation Callback:
111    // r3 - our lr (address of the call instruction in stub plus 4)
112    // r4 - stub's lr (address of instruction that called the stub plus 4)
113    // r5 - is64Bit - always 0.
114    "mr   r3, r0\n"
115    "lwz  r2, 208(r1)\n" // stub's frame
116    "lwz  r4, 8(r2)\n" // stub's lr
117    "li   r5, 0\n"       // 0 == 32 bit
118    "bl _PPCCompilationCallbackC\n"
119    "mtctr r3\n"
120    // Restore all int arg registers
121    "lwz r10, 204(r1)\n"    "lwz r9,  200(r1)\n"
122    "lwz r8,  196(r1)\n"    "lwz r7,  192(r1)\n"
123    "lwz r6,  188(r1)\n"    "lwz r5,  184(r1)\n"
124    "lwz r4,  180(r1)\n"    "lwz r3,  176(r1)\n"
125    // Restore all FP arg registers
126    "lfd f13, 168(r1)\n"    "lfd f12, 160(r1)\n"
127    "lfd f11, 152(r1)\n"    "lfd f10, 144(r1)\n"
128    "lfd f9,  136(r1)\n"    "lfd f8,  128(r1)\n"
129    "lfd f7,  120(r1)\n"    "lfd f6,  112(r1)\n"
130    "lfd f5,  104(r1)\n"    "lfd f4,   96(r1)\n"
131    "lfd f3,   88(r1)\n"    "lfd f2,   80(r1)\n"
132    "lfd f1,   72(r1)\n"
133    // Pop 3 frames off the stack and branch to target
134    "lwz  r1, 208(r1)\n"
135    "lwz  r2, 8(r1)\n"
136    "mtlr r2\n"
137    "bctr\n"
138    );
139
140#elif defined(__PPC__) && !defined(__ppc64__)
141// Linux & FreeBSD / PPC 32 support
142
143// CompilationCallback stub - We can't use a C function with inline assembly in
144// it, because we the prolog/epilog inserted by GCC won't work for us.  Instead,
145// write our own wrapper, which does things our way, so we have complete control
146// over register saving and restoring.
147asm(
148    ".text\n"
149    ".align 2\n"
150    ".globl PPC32CompilationCallback\n"
151"PPC32CompilationCallback:\n"
152    // Make space for 8 ints r[3-10] and 8 doubles f[1-8] and the
153    // FIXME: need to save v[0-19] for altivec?
154    // FIXME: could shrink frame
155    // Set up a proper stack frame
156    // FIXME Layout
157    //   8 double registers       -  64 bytes
158    //   8 int registers          -  32 bytes
159    "mflr 0\n"
160    "stw 0,  4(1)\n"
161    "stwu 1, -104(1)\n"
162    // Save all int arg registers
163    "stw 10, 100(1)\n"   "stw 9,  96(1)\n"
164    "stw 8,  92(1)\n"    "stw 7,  88(1)\n"
165    "stw 6,  84(1)\n"    "stw 5,  80(1)\n"
166    "stw 4,  76(1)\n"    "stw 3,  72(1)\n"
167    // Save all call-clobbered FP regs.
168    "stfd 8,  64(1)\n"
169    "stfd 7,  56(1)\n"   "stfd 6,  48(1)\n"
170    "stfd 5,  40(1)\n"   "stfd 4,  32(1)\n"
171    "stfd 3,  24(1)\n"   "stfd 2,  16(1)\n"
172    "stfd 1,  8(1)\n"
173    // Arguments to Compilation Callback:
174    // r3 - our lr (address of the call instruction in stub plus 4)
175    // r4 - stub's lr (address of instruction that called the stub plus 4)
176    // r5 - is64Bit - always 0.
177    "mr   3, 0\n"
178    "lwz  5, 104(1)\n" // stub's frame
179    "lwz  4, 4(5)\n" // stub's lr
180    "li   5, 0\n"       // 0 == 32 bit
181    "bl PPCCompilationCallbackC\n"
182    "mtctr 3\n"
183    // Restore all int arg registers
184    "lwz 10, 100(1)\n"   "lwz 9,  96(1)\n"
185    "lwz 8,  92(1)\n"    "lwz 7,  88(1)\n"
186    "lwz 6,  84(1)\n"    "lwz 5,  80(1)\n"
187    "lwz 4,  76(1)\n"    "lwz 3,  72(1)\n"
188    // Restore all FP arg registers
189    "lfd 8,  64(1)\n"
190    "lfd 7,  56(1)\n"    "lfd 6,  48(1)\n"
191    "lfd 5,  40(1)\n"    "lfd 4,  32(1)\n"
192    "lfd 3,  24(1)\n"    "lfd 2,  16(1)\n"
193    "lfd 1,  8(1)\n"
194    // Pop 3 frames off the stack and branch to target
195    "lwz  1, 104(1)\n"
196    "lwz  0, 4(1)\n"
197    "mtlr 0\n"
198    "bctr\n"
199    );
200#else
201void PPC32CompilationCallback() {
202  llvm_unreachable("This is not a power pc, you can't execute this!");
203}
204#endif
205
206#if (defined(__POWERPC__) || defined (__ppc__) || defined(_POWER)) && \
207    defined(__ppc64__)
208#ifdef __ELF__
209asm(
210    ".text\n"
211    ".align 2\n"
212    ".globl PPC64CompilationCallback\n"
213    ".section \".opd\",\"aw\",@progbits\n"
214    ".align 3\n"
215"PPC64CompilationCallback:\n"
216    ".quad .L.PPC64CompilationCallback,.TOC.@tocbase,0\n"
217    ".size PPC64CompilationCallback,24\n"
218    ".previous\n"
219    ".align 4\n"
220    ".type PPC64CompilationCallback,@function\n"
221".L.PPC64CompilationCallback:\n"
222#else
223asm(
224    ".text\n"
225    ".align 2\n"
226    ".globl _PPC64CompilationCallback\n"
227"_PPC64CompilationCallback:\n"
228#endif
229    // Make space for 8 ints r[3-10] and 13 doubles f[1-13] and the
230    // FIXME: need to save v[0-19] for altivec?
231    // Set up a proper stack frame
232    // Layout
233    //   PowerPC64 ABI linkage    -  48 bytes
234    //                 parameters -  64 bytes
235    //   13 double registers      - 104 bytes
236    //   8 int registers          -  64 bytes
237    "mflr 0\n"
238    "std  0,  16(1)\n"
239    "stdu 1, -280(1)\n"
240    // Save all int arg registers
241    "std 10, 272(1)\n"    "std 9,  264(1)\n"
242    "std 8,  256(1)\n"    "std 7,  248(1)\n"
243    "std 6,  240(1)\n"    "std 5,  232(1)\n"
244    "std 4,  224(1)\n"    "std 3,  216(1)\n"
245    // Save all call-clobbered FP regs.
246    "stfd 13, 208(1)\n"    "stfd 12, 200(1)\n"
247    "stfd 11, 192(1)\n"    "stfd 10, 184(1)\n"
248    "stfd 9,  176(1)\n"    "stfd 8,  168(1)\n"
249    "stfd 7,  160(1)\n"    "stfd 6,  152(1)\n"
250    "stfd 5,  144(1)\n"    "stfd 4,  136(1)\n"
251    "stfd 3,  128(1)\n"    "stfd 2,  120(1)\n"
252    "stfd 1,  112(1)\n"
253    // Arguments to Compilation Callback:
254    // r3 - our lr (address of the call instruction in stub plus 4)
255    // r4 - stub's lr (address of instruction that called the stub plus 4)
256    // r5 - is64Bit - always 1.
257    "mr   3, 0\n"      // return address (still in r0)
258    "ld   5, 280(1)\n" // stub's frame
259    "ld   4, 16(5)\n"  // stub's lr
260    "li   5, 1\n"      // 1 == 64 bit
261#ifdef __ELF__
262    "bl PPCCompilationCallbackC\n"
263    "nop\n"
264#else
265    "bl _PPCCompilationCallbackC\n"
266#endif
267    "mtctr 3\n"
268    // Restore all int arg registers
269    "ld 10, 272(1)\n"    "ld 9,  264(1)\n"
270    "ld 8,  256(1)\n"    "ld 7,  248(1)\n"
271    "ld 6,  240(1)\n"    "ld 5,  232(1)\n"
272    "ld 4,  224(1)\n"    "ld 3,  216(1)\n"
273    // Restore all FP arg registers
274    "lfd 13, 208(1)\n"    "lfd 12, 200(1)\n"
275    "lfd 11, 192(1)\n"    "lfd 10, 184(1)\n"
276    "lfd 9,  176(1)\n"    "lfd 8,  168(1)\n"
277    "lfd 7,  160(1)\n"    "lfd 6,  152(1)\n"
278    "lfd 5,  144(1)\n"    "lfd 4,  136(1)\n"
279    "lfd 3,  128(1)\n"    "lfd 2,  120(1)\n"
280    "lfd 1,  112(1)\n"
281    // Pop 3 frames off the stack and branch to target
282    "ld  1, 280(1)\n"
283    "ld  0, 16(1)\n"
284    "mtlr 0\n"
285    // XXX: any special TOC handling in the ELF case for JIT?
286    "bctr\n"
287    );
288#else
289void PPC64CompilationCallback() {
290  llvm_unreachable("This is not a power pc, you can't execute this!");
291}
292#endif
293
294extern "C" {
295static void* LLVM_ATTRIBUTE_USED PPCCompilationCallbackC(unsigned *StubCallAddrPlus4,
296                                                         unsigned *OrigCallAddrPlus4,
297                                                         bool is64Bit) {
298  // Adjust the pointer to the address of the call instruction in the stub
299  // emitted by emitFunctionStub, rather than the instruction after it.
300  unsigned *StubCallAddr = StubCallAddrPlus4 - 1;
301  unsigned *OrigCallAddr = OrigCallAddrPlus4 - 1;
302
303  void *Target = JITCompilerFunction(StubCallAddr);
304
305  // Check to see if *OrigCallAddr is a 'bl' instruction, and if we can rewrite
306  // it to branch directly to the destination.  If so, rewrite it so it does not
307  // need to go through the stub anymore.
308  unsigned OrigCallInst = *OrigCallAddr;
309  if ((OrigCallInst >> 26) == 18) {     // Direct call.
310    intptr_t Offset = ((intptr_t)Target - (intptr_t)OrigCallAddr) >> 2;
311
312    if (Offset >= -(1 << 23) && Offset < (1 << 23)) {   // In range?
313      // Clear the original target out.
314      OrigCallInst &= (63 << 26) | 3;
315      // Fill in the new target.
316      OrigCallInst |= (Offset & ((1 << 24)-1)) << 2;
317      // Replace the call.
318      *OrigCallAddr = OrigCallInst;
319    }
320  }
321
322  // Assert that we are coming from a stub that was created with our
323  // emitFunctionStub.
324  if ((*StubCallAddr >> 26) == 18)
325    StubCallAddr -= 3;
326  else {
327  assert((*StubCallAddr >> 26) == 19 && "Call in stub is not indirect!");
328    StubCallAddr -= is64Bit ? 9 : 6;
329  }
330
331  // Rewrite the stub with an unconditional branch to the target, for any users
332  // who took the address of the stub.
333  EmitBranchToAt((intptr_t)StubCallAddr, (intptr_t)Target, false, is64Bit);
334  sys::Memory::InvalidateInstructionCache(StubCallAddr, 7*4);
335
336  // Put the address of the target function to call and the address to return to
337  // after calling the target function in a place that is easy to get on the
338  // stack after we restore all regs.
339  return Target;
340}
341}
342
343
344
345TargetJITInfo::LazyResolverFn
346PPCJITInfo::getLazyResolverFunction(JITCompilerFn Fn) {
347  JITCompilerFunction = Fn;
348  return is64Bit ? PPC64CompilationCallback : PPC32CompilationCallback;
349}
350
351TargetJITInfo::StubLayout PPCJITInfo::getStubLayout() {
352  // The stub contains up to 10 4-byte instructions, aligned at 4 bytes: 3
353  // instructions to save the caller's address if this is a lazy-compilation
354  // stub, plus a 1-, 4-, or 7-instruction sequence to load an arbitrary address
355  // into a register and jump through it.
356  StubLayout Result = {10*4, 4};
357  return Result;
358}
359
360#if (defined(__POWERPC__) || defined (__ppc__) || defined(_POWER)) && \
361defined(__APPLE__)
362extern "C" void sys_icache_invalidate(const void *Addr, size_t len);
363#endif
364
365void *PPCJITInfo::emitFunctionStub(const Function* F, void *Fn,
366                                   JITCodeEmitter &JCE) {
367  // If this is just a call to an external function, emit a branch instead of a
368  // call.  The code is the same except for one bit of the last instruction.
369  if (Fn != (void*)(intptr_t)PPC32CompilationCallback &&
370      Fn != (void*)(intptr_t)PPC64CompilationCallback) {
371    void *Addr = (void*)JCE.getCurrentPCValue();
372    JCE.emitWordBE(0);
373    JCE.emitWordBE(0);
374    JCE.emitWordBE(0);
375    JCE.emitWordBE(0);
376    JCE.emitWordBE(0);
377    JCE.emitWordBE(0);
378    JCE.emitWordBE(0);
379    EmitBranchToAt((intptr_t)Addr, (intptr_t)Fn, false, is64Bit);
380    sys::Memory::InvalidateInstructionCache(Addr, 7*4);
381    return Addr;
382  }
383
384  void *Addr = (void*)JCE.getCurrentPCValue();
385  if (is64Bit) {
386    JCE.emitWordBE(0xf821ffb1);     // stdu r1,-80(r1)
387    JCE.emitWordBE(0x7d6802a6);     // mflr r11
388    JCE.emitWordBE(0xf9610060);     // std r11, 96(r1)
389  } else if (TM.getSubtargetImpl()->isDarwinABI()){
390    JCE.emitWordBE(0x9421ffe0);     // stwu r1,-32(r1)
391    JCE.emitWordBE(0x7d6802a6);     // mflr r11
392    JCE.emitWordBE(0x91610028);     // stw r11, 40(r1)
393  } else {
394    JCE.emitWordBE(0x9421ffe0);     // stwu r1,-32(r1)
395    JCE.emitWordBE(0x7d6802a6);     // mflr r11
396    JCE.emitWordBE(0x91610024);     // stw r11, 36(r1)
397  }
398  intptr_t BranchAddr = (intptr_t)JCE.getCurrentPCValue();
399  JCE.emitWordBE(0);
400  JCE.emitWordBE(0);
401  JCE.emitWordBE(0);
402  JCE.emitWordBE(0);
403  JCE.emitWordBE(0);
404  JCE.emitWordBE(0);
405  JCE.emitWordBE(0);
406  EmitBranchToAt(BranchAddr, (intptr_t)Fn, true, is64Bit);
407  sys::Memory::InvalidateInstructionCache(Addr, 10*4);
408  return Addr;
409}
410
411
412void PPCJITInfo::relocate(void *Function, MachineRelocation *MR,
413                          unsigned NumRelocs, unsigned char* GOTBase) {
414  for (unsigned i = 0; i != NumRelocs; ++i, ++MR) {
415    unsigned *RelocPos = (unsigned*)Function + MR->getMachineCodeOffset()/4;
416    intptr_t ResultPtr = (intptr_t)MR->getResultPointer();
417    switch ((PPC::RelocationType)MR->getRelocationType()) {
418    default: llvm_unreachable("Unknown relocation type!");
419    case PPC::reloc_pcrel_bx:
420      // PC-relative relocation for b and bl instructions.
421      ResultPtr = (ResultPtr-(intptr_t)RelocPos) >> 2;
422      assert(ResultPtr >= -(1 << 23) && ResultPtr < (1 << 23) &&
423             "Relocation out of range!");
424      *RelocPos |= (ResultPtr & ((1 << 24)-1))  << 2;
425      break;
426    case PPC::reloc_pcrel_bcx:
427      // PC-relative relocation for BLT,BLE,BEQ,BGE,BGT,BNE, or other
428      // bcx instructions.
429      ResultPtr = (ResultPtr-(intptr_t)RelocPos) >> 2;
430      assert(ResultPtr >= -(1 << 13) && ResultPtr < (1 << 13) &&
431             "Relocation out of range!");
432      *RelocPos |= (ResultPtr & ((1 << 14)-1))  << 2;
433      break;
434    case PPC::reloc_absolute_high:     // high bits of ref -> low 16 of instr
435    case PPC::reloc_absolute_low: {    // low bits of ref  -> low 16 of instr
436      ResultPtr += MR->getConstantVal();
437
438      // If this is a high-part access, get the high-part.
439      if (MR->getRelocationType() == PPC::reloc_absolute_high) {
440        // If the low part will have a carry (really a borrow) from the low
441        // 16-bits into the high 16, add a bit to borrow from.
442        if (((int)ResultPtr << 16) < 0)
443          ResultPtr += 1 << 16;
444        ResultPtr >>= 16;
445      }
446
447      // Do the addition then mask, so the addition does not overflow the 16-bit
448      // immediate section of the instruction.
449      unsigned LowBits  = (*RelocPos + ResultPtr) & 65535;
450      unsigned HighBits = *RelocPos & ~65535;
451      *RelocPos = LowBits | HighBits;  // Slam into low 16-bits
452      break;
453    }
454    case PPC::reloc_absolute_low_ix: {  // low bits of ref  -> low 14 of instr
455      ResultPtr += MR->getConstantVal();
456      // Do the addition then mask, so the addition does not overflow the 16-bit
457      // immediate section of the instruction.
458      unsigned LowBits  = (*RelocPos + ResultPtr) & 0xFFFC;
459      unsigned HighBits = *RelocPos & 0xFFFF0003;
460      *RelocPos = LowBits | HighBits;  // Slam into low 14-bits.
461      break;
462    }
463    }
464  }
465}
466
467void PPCJITInfo::replaceMachineCodeForFunction(void *Old, void *New) {
468  EmitBranchToAt((intptr_t)Old, (intptr_t)New, false, is64Bit);
469  sys::Memory::InvalidateInstructionCache(Old, 7*4);
470}
471