1 //===-- PPCJITInfo.cpp - Implement the JIT interfaces for the PowerPC -----===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the JIT interfaces for the 32-bit PowerPC target.
12 //===----------------------------------------------------------------------===//
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/CodeGen/MachineCodeEmitter.h"
20 #include "llvm/Config/alloca.h"
21 #include "llvm/Support/Debug.h"
25 static TargetJITInfo::JITCompilerFn JITCompilerFunction;
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))
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)
49 static 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;
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
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
71 extern "C" void PPC32CompilationCallback();
72 extern "C" void PPC64CompilationCallback();
74 #if (defined(__POWERPC__) || defined (__ppc__) || defined(_POWER)) && \
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.
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
90 // PowerPC64 ABI linkage - 24 bytes
91 // parameters - 32 bytes
92 // 13 double registers - 104 bytes
93 // 8 int registers - 32 bytes
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"
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.
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"
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"
133 // Pop 3 frames off the stack and branch to target
140 #elif defined(__PPC__) && !defined(__ppc64__)
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.
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
157 // 8 double registers - 64 bytes
158 // 8 int registers - 32 bytes
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.
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"
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.
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"
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
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"
194 // Pop 3 frames off the stack and branch to target
201 void PPC32CompilationCallback() {
202 assert(0 && "This is not a power pc, you can't execute this!");
207 #if (defined(__POWERPC__) || defined (__ppc__) || defined(_POWER)) && \
212 ".globl _PPC64CompilationCallback\n"
213 "_PPC64CompilationCallback:\n"
214 // Make space for 8 ints r[3-10] and 13 doubles f[1-13] and the
215 // FIXME: need to save v[0-19] for altivec?
216 // Set up a proper stack frame
218 // PowerPC64 ABI linkage - 48 bytes
219 // parameters - 64 bytes
220 // 13 double registers - 104 bytes
221 // 8 int registers - 64 bytes
224 "stdu r1, -280(r1)\n"
225 // Save all int arg registers
226 "std r10, 272(r1)\n" "std r9, 264(r1)\n"
227 "std r8, 256(r1)\n" "std r7, 248(r1)\n"
228 "std r6, 240(r1)\n" "std r5, 232(r1)\n"
229 "std r4, 224(r1)\n" "std r3, 216(r1)\n"
230 // Save all call-clobbered FP regs.
231 "stfd f13, 208(r1)\n" "stfd f12, 200(r1)\n"
232 "stfd f11, 192(r1)\n" "stfd f10, 184(r1)\n"
233 "stfd f9, 176(r1)\n" "stfd f8, 168(r1)\n"
234 "stfd f7, 160(r1)\n" "stfd f6, 152(r1)\n"
235 "stfd f5, 144(r1)\n" "stfd f4, 136(r1)\n"
236 "stfd f3, 128(r1)\n" "stfd f2, 120(r1)\n"
238 // Arguments to Compilation Callback:
239 // r3 - our lr (address of the call instruction in stub plus 4)
240 // r4 - stub's lr (address of instruction that called the stub plus 4)
241 // r5 - is64Bit - always 1.
243 "ld r2, 280(r1)\n" // stub's frame
244 "ld r4, 16(r2)\n" // stub's lr
245 "li r5, 1\n" // 1 == 64 bit
246 "bl _PPCCompilationCallbackC\n"
248 // Restore all int arg registers
249 "ld r10, 272(r1)\n" "ld r9, 264(r1)\n"
250 "ld r8, 256(r1)\n" "ld r7, 248(r1)\n"
251 "ld r6, 240(r1)\n" "ld r5, 232(r1)\n"
252 "ld r4, 224(r1)\n" "ld r3, 216(r1)\n"
253 // Restore all FP arg registers
254 "lfd f13, 208(r1)\n" "lfd f12, 200(r1)\n"
255 "lfd f11, 192(r1)\n" "lfd f10, 184(r1)\n"
256 "lfd f9, 176(r1)\n" "lfd f8, 168(r1)\n"
257 "lfd f7, 160(r1)\n" "lfd f6, 152(r1)\n"
258 "lfd f5, 144(r1)\n" "lfd f4, 136(r1)\n"
259 "lfd f3, 128(r1)\n" "lfd f2, 120(r1)\n"
261 // Pop 3 frames off the stack and branch to target
268 void PPC64CompilationCallback() {
269 assert(0 && "This is not a power pc, you can't execute this!");
274 extern "C" void *PPCCompilationCallbackC(unsigned *StubCallAddrPlus4,
275 unsigned *OrigCallAddrPlus4,
277 // Adjust the pointer to the address of the call instruction in the stub
278 // emitted by emitFunctionStub, rather than the instruction after it.
279 unsigned *StubCallAddr = StubCallAddrPlus4 - 1;
280 unsigned *OrigCallAddr = OrigCallAddrPlus4 - 1;
282 void *Target = JITCompilerFunction(StubCallAddr);
284 // Check to see if *OrigCallAddr is a 'bl' instruction, and if we can rewrite
285 // it to branch directly to the destination. If so, rewrite it so it does not
286 // need to go through the stub anymore.
287 unsigned OrigCallInst = *OrigCallAddr;
288 if ((OrigCallInst >> 26) == 18) { // Direct call.
289 intptr_t Offset = ((intptr_t)Target - (intptr_t)OrigCallAddr) >> 2;
291 if (Offset >= -(1 << 23) && Offset < (1 << 23)) { // In range?
292 // Clear the original target out.
293 OrigCallInst &= (63 << 26) | 3;
294 // Fill in the new target.
295 OrigCallInst |= (Offset & ((1 << 24)-1)) << 2;
297 *OrigCallAddr = OrigCallInst;
301 // Assert that we are coming from a stub that was created with our
303 if ((*StubCallAddr >> 26) == 18)
306 assert((*StubCallAddr >> 26) == 19 && "Call in stub is not indirect!");
307 StubCallAddr -= is64Bit ? 9 : 6;
310 // Rewrite the stub with an unconditional branch to the target, for any users
311 // who took the address of the stub.
312 EmitBranchToAt((intptr_t)StubCallAddr, (intptr_t)Target, false, is64Bit);
314 // Put the address of the target function to call and the address to return to
315 // after calling the target function in a place that is easy to get on the
316 // stack after we restore all regs.
322 TargetJITInfo::LazyResolverFn
323 PPCJITInfo::getLazyResolverFunction(JITCompilerFn Fn) {
324 JITCompilerFunction = Fn;
325 return is64Bit ? PPC64CompilationCallback : PPC32CompilationCallback;
328 #if (defined(__POWERPC__) || defined (__ppc__) || defined(_POWER)) && \
330 extern "C" void sys_icache_invalidate(const void *Addr, size_t len);
333 /// SyncICache - On PPC, the JIT emitted code must be explicitly refetched to
334 /// ensure correct execution.
335 static void SyncICache(const void *Addr, size_t len) {
336 #if (defined(__POWERPC__) || defined (__ppc__) || defined(_POWER)) && \
338 sys_icache_invalidate(Addr, len);
342 void *PPCJITInfo::emitFunctionStub(const Function* F, void *Fn,
343 MachineCodeEmitter &MCE) {
344 // If this is just a call to an external function, emit a branch instead of a
345 // call. The code is the same except for one bit of the last instruction.
346 if (Fn != (void*)(intptr_t)PPC32CompilationCallback &&
347 Fn != (void*)(intptr_t)PPC64CompilationCallback) {
348 MCE.startFunctionStub(F, 7*4);
349 intptr_t Addr = (intptr_t)MCE.getCurrentPCValue();
357 EmitBranchToAt(Addr, (intptr_t)Fn, false, is64Bit);
358 SyncICache((void*)Addr, 7*4);
359 return MCE.finishFunctionStub(F);
362 MCE.startFunctionStub(F, 10*4);
363 intptr_t Addr = (intptr_t)MCE.getCurrentPCValue();
365 MCE.emitWordBE(0xf821ffb1); // stdu r1,-80(r1)
366 MCE.emitWordBE(0x7d6802a6); // mflr r11
367 MCE.emitWordBE(0xf9610060); // std r11, 96(r1)
368 } else if (TM.getSubtargetImpl()->isMachoABI()){
369 MCE.emitWordBE(0x9421ffe0); // stwu r1,-32(r1)
370 MCE.emitWordBE(0x7d6802a6); // mflr r11
371 MCE.emitWordBE(0x91610028); // stw r11, 40(r1)
373 MCE.emitWordBE(0x9421ffe0); // stwu r1,-32(r1)
374 MCE.emitWordBE(0x7d6802a6); // mflr r11
375 MCE.emitWordBE(0x91610024); // stw r11, 36(r1)
377 intptr_t BranchAddr = (intptr_t)MCE.getCurrentPCValue();
385 EmitBranchToAt(BranchAddr, (intptr_t)Fn, true, is64Bit);
386 SyncICache((void*)Addr, 10*4);
387 return MCE.finishFunctionStub(F);
391 void PPCJITInfo::relocate(void *Function, MachineRelocation *MR,
392 unsigned NumRelocs, unsigned char* GOTBase) {
393 for (unsigned i = 0; i != NumRelocs; ++i, ++MR) {
394 unsigned *RelocPos = (unsigned*)Function + MR->getMachineCodeOffset()/4;
395 intptr_t ResultPtr = (intptr_t)MR->getResultPointer();
396 switch ((PPC::RelocationType)MR->getRelocationType()) {
397 default: assert(0 && "Unknown relocation type!");
398 case PPC::reloc_pcrel_bx:
399 // PC-relative relocation for b and bl instructions.
400 ResultPtr = (ResultPtr-(intptr_t)RelocPos) >> 2;
401 assert(ResultPtr >= -(1 << 23) && ResultPtr < (1 << 23) &&
402 "Relocation out of range!");
403 *RelocPos |= (ResultPtr & ((1 << 24)-1)) << 2;
405 case PPC::reloc_pcrel_bcx:
406 // PC-relative relocation for BLT,BLE,BEQ,BGE,BGT,BNE, or other
408 ResultPtr = (ResultPtr-(intptr_t)RelocPos) >> 2;
409 assert(ResultPtr >= -(1 << 13) && ResultPtr < (1 << 13) &&
410 "Relocation out of range!");
411 *RelocPos |= (ResultPtr & ((1 << 14)-1)) << 2;
413 case PPC::reloc_absolute_high: // high bits of ref -> low 16 of instr
414 case PPC::reloc_absolute_low: { // low bits of ref -> low 16 of instr
415 ResultPtr += MR->getConstantVal();
417 // If this is a high-part access, get the high-part.
418 if (MR->getRelocationType() == PPC::reloc_absolute_high) {
419 // If the low part will have a carry (really a borrow) from the low
420 // 16-bits into the high 16, add a bit to borrow from.
421 if (((int)ResultPtr << 16) < 0)
422 ResultPtr += 1 << 16;
426 // Do the addition then mask, so the addition does not overflow the 16-bit
427 // immediate section of the instruction.
428 unsigned LowBits = (*RelocPos + ResultPtr) & 65535;
429 unsigned HighBits = *RelocPos & ~65535;
430 *RelocPos = LowBits | HighBits; // Slam into low 16-bits
433 case PPC::reloc_absolute_low_ix: { // low bits of ref -> low 14 of instr
434 ResultPtr += MR->getConstantVal();
435 // Do the addition then mask, so the addition does not overflow the 16-bit
436 // immediate section of the instruction.
437 unsigned LowBits = (*RelocPos + ResultPtr) & 0xFFFC;
438 unsigned HighBits = *RelocPos & 0xFFFF0003;
439 *RelocPos = LowBits | HighBits; // Slam into low 14-bits.
446 void PPCJITInfo::replaceMachineCodeForFunction(void *Old, void *New) {
447 EmitBranchToAt((intptr_t)Old, (intptr_t)New, false, is64Bit);