2 * Copyright (c) 2002 Apple Computer, Inc. All rights reserved.
4 * @APPLE_LICENSE_HEADER_START@
6 * Copyright (c) 1999-2003 Apple Computer, Inc. All Rights Reserved.
8 * This file contains Original Code and/or Modifications of Original Code
9 * as defined in and that are subject to the Apple Public Source License
10 * Version 2.0 (the 'License'). You may not use this file except in
11 * compliance with the License. Please obtain a copy of the License at
12 * http://www.opensource.apple.com/apsl/ and read it before using this
15 * The Original Code and all software distributed under the License are
16 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
17 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
18 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
20 * Please see the License for the specific language governing rights and
21 * limitations under the License.
23 * @APPLE_LICENSE_HEADER_END@
28 * Software emulation of instructions not handled in hw, on 64-bit machines.
31 #include <sys/appleapiopts.h>
34 #include <ppc/proc_reg.h>
35 #include <ppc/exception.h>
36 #include <mach/machine/vm_param.h>
37 #include <ppc/cpu_capabilities.h>
40 // CR bit set if the instruction is an "update" form (LFDU, STWU, etc):
43 // CR bit set if interrupt occured in trace mode (ie, MSR_SE_BIT):
46 // CR bit set if notification on alignment interrupts is requested (notifyUnalignbit in spcFlags):
49 // CR bit distinguishes between alignment and program exceptions:
54 // *************************************
55 // * P R O G R A M I N T E R R U P T *
56 // *************************************
58 // These are floating pt exceptions, illegal instructions, privileged mode violations,
59 // and traps. All we're interested in at this low level is illegal instructions.
60 // The ones we "emulate" are:
61 // DCBA, which is not implemented in the IBM 970. The emulation is to ignore it,
62 // as it is just a hint.
63 // MCRXR, which is not implemented on the IBM 970, but is in the PPC ISA.
65 // Additionally, to facilitate debugging the alignment handler, we recognize a special
66 // diagnostic mode that is used to simulate alignment exceptions. When in this mode,
67 // if the instruction has opcode==0 and the extended opcode is one of the X-form
68 // instructions that can take an alignment interrupt, then we change the opcode to
69 // 31 and pretend it got an alignment interrupt. This exercises paths that
70 // are hard to drive or perhaps never driven on this particular CPU.
76 crclr kAlignment // not an alignment exception
77 b a64AlignAssistJoin // join alignment handler
80 // Return from alignment handler with all the regs loaded for opcode emulation.
83 rlwinm. r0,r29,0,SRR1_PRG_ILL_INS_BIT,SRR1_PRG_ILL_INS_BIT // illegal opcode?
84 beq a64PassAlong // No, must have been trap or priv violation etc
85 rlwinm r3,r20,6,26,31 // right justify opcode field (bits 0-5)
86 rlwinm r4,r20,31,22,31 // right justify extended opcode field (bits 21-30)
87 cmpwi cr0,r3,31 // X-form?
88 cmpwi cr1,r4,758 // DCBA?
89 cmpwi cr4,r4,512 // MCRXR?
90 crand cr1_eq,cr0_eq,cr1_eq // merge the two tests for DCBA
91 crand cr4_eq,cr0_eq,cr4_eq // and for MCRXR
92 beq++ cr1_eq,a64ExitEm // was DCBA, so ignore
93 bne-- cr4_eq,a64NotEmulated // skip if not MCRXR
95 // Was MCRXR, so emulate.
97 ld r3,savexer(r13) // get the XER
98 lwz r4,savecr(r13) // and the CR
99 rlwinm r5,r20,11,27,29 // get (CR# * 4) from instruction
100 rlwinm r6,r3,0,4,31 // zero XER[32-35] (also XER[0-31])
101 sld r4,r4,r5 // move target CR field to bits 32-35
102 rlwimi r4,r3,0,0,3 // move XER[32-35] into CR field
103 stw r6,savexer+4(r13) // update XER
104 srd r4,r4,r5 // re-position CR
105 stw r4,savecr(r13) // update CR
108 // Not an opcode we normally emulate. If in special diagnostic mode and opcode=0,
109 // emulate as an alignment exception. This special case is for test software.
112 lwz r30,dgFlags(0) // Get the flags
113 rlwinm. r0,r30,0,enaDiagEMb,enaDiagEMb // Do we want to try to emulate something?
114 beq++ a64PassAlong // No emulation allowed
115 cmpwi r3,0 // opcode==0 ?
116 bne a64PassAlong // not the special case
117 oris r20,r20,0x7C00 // change opcode to 31
118 crset kAlignment // say we took alignment exception
119 rlwinm r5,r4,0,26+1,26-1 // mask Update bit (32) out of extended opcode
120 rlwinm r5,r5,0,0,31 // Clean out leftover junk from rlwinm
122 cmpwi r4,1014 // dcbz/dcbz128 ?
124 cmpwi r5,21 // ldx/ldux ?
125 cror cr1_eq,cr0_eq,cr1_eq
126 cmpwi r5,599 // lfdx/lfdux ?
127 cror cr1_eq,cr0_eq,cr1_eq
128 cmpwi r5,535 // lfsx/lfsux ?
129 cror cr1_eq,cr0_eq,cr1_eq
130 cmpwi r5,343 // lhax/lhaux ?
131 cror cr1_eq,cr0_eq,cr1_eq
132 cmpwi r4,790 // lhbrx ?
133 cror cr1_eq,cr0_eq,cr1_eq
134 cmpwi r5,279 // lhzx/lhzux ?
135 cror cr1_eq,cr0_eq,cr1_eq
136 cmpwi r4,597 // lswi ?
137 cror cr1_eq,cr0_eq,cr1_eq
138 cmpwi r4,533 // lswx ?
139 cror cr1_eq,cr0_eq,cr1_eq
140 cmpwi r5,341 // lwax/lwaux ?
141 cror cr1_eq,cr0_eq,cr1_eq
142 cmpwi r4,534 // lwbrx ?
143 cror cr1_eq,cr0_eq,cr1_eq
144 cmpwi r5,23 // lwz/lwzx ?
145 cror cr1_eq,cr0_eq,cr1_eq
146 cmpwi r5,149 // stdx/stdux ?
147 cror cr1_eq,cr0_eq,cr1_eq
148 cmpwi r5,727 // stfdx/stfdux ?
149 cror cr1_eq,cr0_eq,cr1_eq
150 cmpwi r4,983 // stfiwx ?
151 cror cr1_eq,cr0_eq,cr1_eq
152 cmpwi r5,663 // stfsx/stfsux ?
153 cror cr1_eq,cr0_eq,cr1_eq
154 cmpwi r4,918 // sthbrx ?
155 cror cr1_eq,cr0_eq,cr1_eq
156 cmpwi r5,407 // sthx/sthux ?
157 cror cr1_eq,cr0_eq,cr1_eq
158 cmpwi r4,725 // stswi ?
159 cror cr1_eq,cr0_eq,cr1_eq
160 cmpwi r4,661 // stswx ?
161 cror cr1_eq,cr0_eq,cr1_eq
162 cmpwi r4,662 // stwbrx ?
163 cror cr1_eq,cr0_eq,cr1_eq
164 cmpwi r5,151 // stwx/stwux ?
165 cror cr1_eq,cr0_eq,cr1_eq
167 beq++ cr1,a64GotInstruction // it was one of the X-forms we handle
168 crclr kAlignment // revert to program interrupt
169 b a64PassAlong // not recognized extended opcode
172 // *****************************************
173 // * A L I G N M E N T I N T E R R U P T *
174 // *****************************************
176 // We get here in exception context, ie with interrupts disabled, translation off, and
177 // in 64-bit mode, with:
178 // r13 = save-area pointer, with general context already saved in it
179 // cr6 = feature flags
180 // We preserve r13 and cr6. Other GPRs and CRs, the LR and CTR are used.
182 // Current 64-bit processors (GPUL) handle almost all misaligned operations in hardware,
183 // so this routine usually isn't called very often. Only floating pt ops that cross a page
184 // boundary and are not word aligned, and LMW/STMW can take exceptions to cacheable memory.
185 // However, in contrast to G3 and G4, any misaligned load/store will get an alignment
186 // interrupt on uncached memory.
188 // We always emulate scalar ops with a series of byte load/stores. Doing so is no slower
189 // than LWZ/STW in cases where a scalar op gets an alignment exception.
191 // This routine supports all legal permutations of alignment interrupts occuring in user or
192 // supervisor mode, 32 or 64-bit addressing, and translation on or off. We do not emulate
193 // instructions that go past the end of an address space, such as "LHZ -1(0)"; we just pass
194 // along the alignment exception rather than wrap around to byte 0. (Treatment of address
195 // space wrap is a moot point in Mac OS X, since we do not map either the last page or
198 // First, check for a few special cases such as virtual machines, etc.
200 .globl EXT(AlignAssist64)
203 crset kAlignment // mark as alignment interrupt
205 a64AlignAssistJoin: // join here from program interrupt handler
206 mfsprg r31,0 // get the per_proc data ptr
207 mcrf cr3,cr6 // save feature flags here...
208 lwz r21,spcFlags(r31) // grab the special flags
209 ld r29,savesrr1(r13) // get the MSR etc at the fault
210 ld r28,savesrr0(r13) // get the EA of faulting instruction
211 mfmsr r26 // save MSR at entry
212 rlwinm. r0,r21,0,runningVMbit,runningVMbit // Are we running a VM?
213 lwz r19,dgFlags(0) // Get the diagnostics flags
214 bne-- a64PassAlong // yes, let the virtual machine monitor handle
217 // Set up the MSR shadow regs. We turn on FP in this routine, and usually set DR and RI
218 // when accessing user space (the SLB is still set up with all the user space translations.)
219 // However, if the interrupt occured in the kernel with DR off, we keep it off while
220 // accessing the "target" address space. If we set DR to access the target space, we also
221 // set RI. The RI bit tells the exception handlers to clear cr0 beq and return if we get an
222 // exception accessing the user address space. We are careful to test cr0 beq after every such
223 // access. We keep the following "shadows" of the MSR in global regs across this code:
224 // r25 = MSR at entry, plus FP and probably DR and RI (used to access target space)
225 // r26 = MSR at entry
227 // r29 = SRR1 (ie, MSR at interrupt)
228 // Note that EE and IR are always off, and SF is always on in this code.
230 rlwinm r3,r29,0,MSR_DR_BIT,MSR_DR_BIT // was translation on at fault?
231 rlwimi r3,r3,32-MSR_RI_BIT+MSR_DR_BIT,MSR_RI_BIT,MSR_RI_BIT // if DR was set, set RI too
232 or r25,r26,r3 // assemble MSR to use accessing target space
235 // Because the DSISR and DAR are either not set or are not to be trusted on some 64-bit
236 // processors on an alignment interrupt, we must fetch the faulting instruction ourselves,
237 // then decode/hash the opcode and reconstruct the EA manually.
239 mtmsr r25 // turn on FP and (if it was on at fault) DR and RI
240 isync // wait for it to happen
241 cmpw r0,r0 // turn on beq so we can check for DSIs
242 lwz r20,0(r28) // fetch faulting instruction, probably with DR on
243 bne-- a64RedriveAsISI // got a DSI trying to fetch it, pretend it was an ISI
244 mtmsr r26 // turn DR back off
245 isync // wait for it to happen
248 // Set a few flags while we wait for the faulting instruction to arrive from cache.
250 rlwinm. r0,r29,0,MSR_SE_BIT,MSR_SE_BIT // Were we single stepping?
251 stw r20,savemisc2(r13) // Save the instruction image in case we notify
253 rlwinm. r0,r19,0,enaNotifyEMb,enaNotifyEMb // Should we notify?
257 // Hash the intruction into a 5-bit value "AAAAB" used to index the branch table, and a
258 // 1-bit kUpdate flag, as follows:
259 // ¥ for X-form instructions (with primary opcode 31):
260 // the "AAAA" bits are bits 21-24 of the instruction
261 // the "B" bit is the XOR of bits 29 and 30
262 // the update bit is instruction bit 25
263 // ¥ for D and DS-form instructions (actually, any primary opcode except 31):
264 // the "AAAA" bits are bits 1-4 of the instruction
266 // the update bit is instruction bit 5
268 // Just for fun (and perhaps a little speed on deep-pipe machines), we compute the hash,
269 // update flag, and EA without branches and with ipc >= 2.
271 // When we "bctr" to the opcode-specific reoutine, the following are all set up:
272 // MSR = EE and IR off, SF and FP on
273 // r13 = save-area pointer (physical)
274 // r14 = ptr to saver0 in save-area (ie, to base of GPRs)
275 // r15 = 0x00000000FFFFFFFF if 32-bit mode fault, 0xFFFFFFFFFFFFFFFF if 64
276 // r16 = RA * 8 (ie, reg# not reg value)
278 // r18 = (RA|0) (reg value)
279 // r19 = -1 if X-form, 0 if D-form
280 // r20 = faulting instruction
281 // r21 = RT * 8 (ie, reg# not reg value)
282 // r22 = addr(aaFPopTable)+(RT*32), ie ptr to floating pt table for target register
283 // r25 = MSR at entrance, probably with DR and RI set (for access to target space)
284 // r26 = MSR at entrance
286 // r28 = SRR0 (ie, EA of faulting instruction)
287 // r29 = SRR1 (ie, MSR at fault)
288 // r30 = scratch, usually user data
289 // r31 = per-proc pointer
290 // cr2 = kTrace, kNotify, and kAlignment flags
291 // cr3 = saved copy of feature flags used in lowmem vector code
292 // cr6 = bits 24-27 of CR are bits 24-27 of opcode if X-form, or bits 4-5 and 00 if D-form
293 // bit 25 is the kUpdate flag, set for update form instructions
294 // cr7 = bits 28-31 of CR are bits 28-31 of opcode if X-form, or 0 if D-form
296 a64GotInstruction: // here from program interrupt with instruction in r20
297 rlwinm r21,r20,6+6,20,25 // move the primary opcode (bits 0-6) to bits 20-25
298 la r14,saver0(r13) // r14 <- base address of GPR registers
299 xori r19,r21,0x07C0 // iff primary opcode is 31, set r19 to 0
300 rlwinm r16,r20,16+3,24,28 // r16 <- RA*8
301 subi r19,r19,1 // set bit 0 iff X-form (ie, if primary opcode is 31)
302 rlwinm r17,r20,21+3,24,28 // r17 <- RB*8 (if X-form)
303 sradi r19,r19,63 // r19 <- -1 if X-form, 0 if D-form
304 extsh r22,r20 // r22 <- displacement (if D-form)
306 ldx r23,r14,r17 // get (RB), if any
307 and r15,r20,r19 // instruction if X, 0 if D
308 andc r17,r21,r19 // primary opcode in bits 20-25 if D, 0 if X
309 ldx r18,r14,r16 // get (RA)
310 subi r24,r16,1 // set bit 0 iff RA==0
311 or r21,r15,r17 // r21 <- instruction if X, or bits 0-5 in bits 20-25 if D
312 sradi r24,r24,63 // r24 <- -1 if RA==0, 0 otherwise
313 rlwinm r17,r21,32-4,25,28 // shift opcode bits 21-24 to 25-28 (hash "AAAA" bits)
314 lis r10,ha16(a64BranchTable) // start to build up branch table address
315 rlwimi r17,r21,0,29,29 // move opcode bit 29 into hash as start of "B" bit
316 rlwinm r30,r21,1,29,29 // position opcode bit 30 in position 29
317 and r12,r23,r19 // RB if X-form, 0 if D-form
318 andc r11,r22,r19 // 0 if X-form, sign extended displacement if D-form
319 xor r17,r17,r30 // bit 29 ("B") of hash is xor(bit29,bit30)
320 addi r10,r10,lo16(a64BranchTable)
321 or r12,r12,r11 // r12 <- (RB) or displacement, as appropriate
322 lwzx r30,r10,r17 // get address from branch table
323 mtcrf 0x01,r21 // move opcode bits 28-31 to CR7
324 sradi r15,r29,32 // propogate SF bit from SRR1 (MSR_SF, which is bit 0)
325 andc r18,r18,r24 // r18 <- (RA|0)
326 mtcrf 0x02,r21 // move opcode bits 24-27 to CR6 (kUpdate is bit 25)
327 add r17,r18,r12 // r17 <- EA, which might need to be clamped to 32 bits
328 mtctr r30 // set up branch address
330 oris r15,r15,0xFFFF // start to fill low word of r15 with 1s
331 rlwinm r21,r20,11+3,24,28 // r21 <- RT * 8
332 lis r22,ha16(EXT(aaFPopTable)) // start to compute address of floating pt table
333 ori r15,r15,0xFFFF // now bits 32-63 of r15 are 1s
334 addi r22,r22,lo16(EXT(aaFPopTable))
335 and r17,r17,r15 // clamp EA to 32 bits if necessary
336 rlwimi r22,r21,2,22,26 // move RT into aaFPopTable address (which is 1KB aligned)
338 bf-- kAlignment,a64HandleProgramInt // return to Program Interrupt handler
339 bctr // if alignment interrupt, jump to opcode-specific routine
342 // Floating-pt load single (lfs[u], lfsx[u])
345 bl a64Load4Bytes // get data in r30
346 mtctr r22 // set up address of "lfs fRT,emfp0(r31)"
347 stw r30,emfp0(r31) // put word here for aaFPopTable routine
349 b a64UpdateCheck // update RA if necessary and exit
352 // Floating-pt store single (stfs[u], stfsx[u])
355 ori r22,r22,8 // set dir==1 (ie, single store) in aaFPopTable
356 mtctr r22 // set up address of "stfs fRT,emfp0(r31)"
357 bctrl // execute the store into emfp0
358 lwz r30,emfp0(r31) // get the word
359 bl a64Store4Bytes // store r30 into user space
360 b a64UpdateCheck // update RA if necessary and exit
363 // Floating-pt store as integer word (stfiwx)
366 ori r22,r22,16+8 // set size=1, dir==1 (ie, double store) in aaFPopTable
367 mtctr r22 // set up FP register table address
368 bctrl // double precision store into emfp0
369 lwz r30,emfp0+4(r31) // get the low-order word
370 bl a64Store4Bytes // store r30 into user space
371 b a64Exit // successfully emulated
374 // Floating-pt load double (lfd[u], lfdx[u])
377 ori r22,r22,16 // set Double bit in aaFPopTable address
378 bl a64Load8Bytes // get data in r30
379 mtctr r22 // set up address of "lfd fRT,emfp0(r31)"
380 std r30,emfp0(r31) // put doubleword here for aaFPopTable routine
381 bctrl // execute the load
382 b a64UpdateCheck // update RA if necessary and exit
385 // Floating-pt store double (stfd[u], stfdx[u])
388 ori r22,r22,16+8 // set size=1, dir==1 (ie, double store) in aaFPopTable address
389 mtctr r22 // address of routine to stfd RT
390 bctrl // store into emfp0
391 ld r30,emfp0(r31) // get the doubleword
392 bl a64Store8Bytes // store r30 into user space
393 b a64UpdateCheck // update RA if necessary and exit
396 // Load halfword w 0-fill (lhz[u], lhzx[u])
399 bl a64Load2Bytes // load into r30 from user space (w 0-fill)
400 stdx r30,r14,r21 // store into RT slot in register file
401 b a64UpdateCheck // update RA if necessary and exit
404 // Load halfword w sign fill (lha[u], lhax[u])
407 bl a64Load2Bytes // load into r30 from user space (w 0-fill)
408 extsh r30,r30 // sign-extend
409 stdx r30,r14,r21 // store into RT slot in register file
410 b a64UpdateCheck // update RA if necessary and exit
413 // Load halfword byte reversed (lhbrx)
416 bl a64Load2Bytes // load into r30 from user space (w 0-fill)
417 rlwinm r3,r30,8,16,23 // reverse bytes into r3
418 rlwimi r3,r30,24,24,31
419 stdx r3,r14,r21 // store into RT slot in register file
420 b a64Exit // successfully emulated
423 // Store halfword (sth[u], sthx[u])
426 ldx r30,r14,r21 // get RT
427 bl a64Store2Bytes // store r30 into user space
428 b a64UpdateCheck // update RA if necessary and exit
431 // Store halfword byte reversed (sthbrx)
434 addi r21,r21,6 // point to low two bytes of RT
435 lhbrx r30,r14,r21 // load and reverse
436 bl a64Store2Bytes // store r30 into user space
437 b a64Exit // successfully emulated
440 // Load word w 0-fill (lwz[u], lwzx[u]), also lwarx.
443 andc r3,r19,r20 // light bit 30 of r3 iff lwarx
444 andi. r0,r3,2 // is it lwarx?
445 bne-- a64PassAlong // yes, never try to emulate a lwarx
446 bl a64Load4Bytes // load 4 bytes from user space into r30 (0-filled)
447 stdx r30,r14,r21 // update register file
448 b a64UpdateCheck // update RA if necessary and exit
451 // Load word w sign fill (lwa, lwax[u])
454 crclr kUpdate // no update form of lwa (its a reserved encoding)
456 bl a64Load4Bytes // load 4 bytes from user space into r30 (0-filled)
457 extsw r30,r30 // sign extend
458 stdx r30,r14,r21 // update register file
459 b a64UpdateCheck // update RA if necessary and exit
462 // Load word byte reversed (lwbrx)
465 bl a64Load4Bytes // load 4 bytes from user space into r30 (0-filled)
466 rlwinm r3,r30,24,0,31 // flip bytes 1234 to 4123
467 rlwimi r3,r30,8,8,15 // r3 is now 4323
468 rlwimi r3,r30,8,24,31 // r3 is now 4321
469 stdx r3,r14,r21 // update register file
470 b a64Exit // successfully emulated
473 // Store word (stw[u], stwx[u])
476 ldx r30,r14,r21 // get RT
477 bl a64Store4Bytes // store r30 into user space
478 b a64UpdateCheck // update RA if necessary and exit
481 // Store word byte reversed (stwbrx)
484 addi r21,r21,4 // point to low word of RT
485 lwbrx r30,r14,r21 // load and reverse
486 bl a64Store4Bytes // store r30 into user space
487 b a64Exit // successfully emulated
490 // Load doubleword (ld[u], ldx[u]), also lwa.
492 a64LdLwa: // these are DS form: ld=0, ldu=1, and lwa=2
493 andi. r0,r20,2 // ld[u] or lwa? (test bit 30 of DS field)
494 rlwinm r3,r20,0,30,31 // must adjust EA by subtracting DS field
496 and r17,r17,r15 // re-clamp to 32 bits if necessary
497 bne a64Lwa // handle lwa
499 bl a64Load8Bytes // load 8 bytes from user space into r30
500 stdx r30,r14,r21 // update register file
501 b a64UpdateCheck // update RA if necessary and exit
504 // Store doubleword (stdx[u], std[u])
507 bf-- 30,a64PassAlong // stwcx, so pass along alignment exception
508 b a64Stdx // was stdx
510 bt 30,a64Stfiwx // handle stfiwx
511 rlwinm. r3,r20,0,30,31 // must adjust EA by subtracting DS field
513 and r17,r17,r15 // re-clamp to 32 bits if necessary
515 ldx r30,r14,r21 // get RT
516 bl a64Store8Bytes // store RT into user space
517 b a64UpdateCheck // update RA if necessary and exit
520 // Dcbz and Dcbz128 (bit 10 distinguishes the two forms)
523 andis. r0,r20,0x0020 // bit 10 set?
524 li r3,0 // get a 0 to store
525 li r0,4 // assume 32-bit version, store 8 bytes 4x
526 li r4,_COMM_PAGE_BASE_ADDRESS
527 rldicr r17,r17,0,63-5 // 32-byte align EA
528 beq a64DcbzSetup // it was the 32-byte version
529 rldicr r17,r17,0,63-7 // zero low 7 bits of EA
530 li r0,16 // store 8 bytes 16x
532 xor r4,r4,r28 // was dcbz in the commpage(s)?
533 and r4,r4,r15 // mask off high-order bits if 32-bit mode
534 srdi. r4,r4,12 // check SRR0
535 bne a64NotCommpage // not in commpage
536 rlwinm. r4,r29,0,MSR_PR_BIT,MSR_PR_BIT // did fault occur in user mode?
537 beq-- a64NotCommpage // do not zero cr7 if kernel got alignment exception
538 lwz r4,savecr(r13) // if we take a dcbz{128} in the commpage...
539 rlwinm r4,r4,0,0,27 // ...clear user's cr7...
540 stw r4,savecr(r13) // ...as a flag for _COMM_PAGE_BIGCOPY
543 cmpw r0,r0 // turn cr0 beq on so we can check for DSIs
544 mtmsr r25 // turn on DR and RI so we can address user space
545 isync // wait for it to happen
547 std r3,0(r17) // store into user space
548 bne-- a64RedriveAsDSI
552 mtmsr r26 // restore MSR
553 isync // wait for it to happen
557 // Load and store multiple (lmw, stmw), distinguished by bit 25
560 subfic r22,r21,32*8 // how many regs to load or store?
561 srwi r22,r22,1 // get bytes to load/store
562 bf 25,a64LoadMultiple // handle lmw
563 b a64StoreMultiple // it was stmw
566 // Load string word immediate (lswi)
569 rlwinm r22,r20,21,27,31 // get #bytes in r22
570 and r17,r18,r15 // recompute EA as (RA|0), and clamp
571 subi r3,r22,1 // r22==0?
572 rlwimi r22,r3,6,26,26 // map count of 0 to 32
576 // Store string word immediate (stswi)
579 rlwinm r22,r20,21,27,31 // get #bytes in r22
580 and r17,r18,r15 // recompute EA as (RA|0), and clamp
581 subi r3,r22,1 // r22==0?
582 rlwimi r22,r3,6,26,26 // map count of 0 to 32
586 // Load string word indexed (lswx), also lwbrx
589 bf 30,a64Lwbrx // was lwbrx
590 ld r22,savexer(r13) // get the xer
591 rlwinm r22,r22,0,25,31 // isolate the byte count
592 b a64LoadMultiple // join common code
595 // Store string word indexed (stswx), also stwbrx
598 bf 30,a64Stwbrx // was stwbrx
599 ld r22,savexer(r13) // get the xer
600 rlwinm r22,r22,0,25,31 // isolate the byte count
601 b a64StoreMultiple // join common code
604 // Load multiple words. This handles lmw, lswi, and lswx.
606 a64LoadMultiple: // r22 = byte count, may be 0
607 subic. r3,r22,1 // get (#bytes-1)
608 blt a64Exit // done if 0
609 add r4,r17,r3 // get EA of last operand byte
610 and r4,r4,r15 // clamp
611 cmpld r4,r17 // address space wrap?
612 blt-- a64PassAlong // pass along exception if so
613 srwi. r4,r22,2 // get # full words to load
614 rlwinm r22,r22,0,30,31 // r22 <- leftover byte count
615 cmpwi cr1,r22,0 // leftover bytes?
616 beq a64Lm3 // no words
617 mtctr r4 // set up word count
618 cmpw r0,r0 // set beq for DSI test
620 mtmsr r25 // turn on DR and RI
621 isync // wait for it to happen
623 bne-- a64RedriveAsDSI // got a DSI
625 bne-- a64RedriveAsDSI // got a DSI
627 bne-- a64RedriveAsDSI // got a DSI
629 bne-- a64RedriveAsDSI // got a DSI
630 rlwinm r30,r3,24,0,7 // pack bytes into r30
634 mtmsr r26 // turn DR back off so we can store into register file
636 addi r17,r17,4 // bump EA
637 stdx r30,r14,r21 // pack into register file
638 addi r21,r21,8 // bump register file offset
639 rlwinm r21,r21,0,24,28 // wrap around to 0
641 a64Lm3: // cr1/r22 = leftover bytes (0-3), cr0 beq set
642 beq cr1,a64Exit // no leftover bytes
644 mtmsr r25 // turn on DR so we can access user space
646 lbz r3,0(r17) // get 1st leftover byte
647 bne-- a64RedriveAsDSI // got a DSI
648 rlwinm r30,r3,24,0,7 // position in byte 4 of r30 (and clear rest of r30)
649 bdz a64Lm4 // only 1 byte leftover
650 lbz r3,1(r17) // get 2nd byte
651 bne-- a64RedriveAsDSI // got a DSI
652 rldimi r30,r3,16,40 // insert into byte 5 of r30
653 bdz a64Lm4 // only 2 bytes leftover
654 lbz r3,2(r17) // get 3rd byte
655 bne-- a64RedriveAsDSI // got a DSI
656 rldimi r30,r3,8,48 // insert into byte 6
658 mtmsr r26 // turn DR back off so we can store into register file
660 stdx r30,r14,r21 // pack partially-filled word into register file
664 // Store multiple words. This handles stmw, stswi, and stswx.
666 a64StoreMultiple: // r22 = byte count, may be 0
667 subic. r3,r22,1 // get (#bytes-1)
668 blt a64Exit // done if 0
669 add r4,r17,r3 // get EA of last operand byte
670 and r4,r4,r15 // clamp
671 cmpld r4,r17 // address space wrap?
672 blt-- a64PassAlong // pass along exception if so
673 srwi. r4,r22,2 // get # full words to load
674 rlwinm r22,r22,0,30,31 // r22 <- leftover byte count
675 cmpwi cr1,r22,0 // leftover bytes?
676 beq a64Sm3 // no words
677 mtctr r4 // set up word count
678 cmpw r0,r0 // turn on beq so we can check for DSIs
680 ldx r30,r14,r21 // get next register
681 addi r21,r21,8 // bump register file offset
682 rlwinm r21,r21,0,24,28 // wrap around to 0
683 srwi r3,r30,24 // shift the four bytes into position
686 mtmsr r25 // turn on DR so we can access user space
687 isync // wait for it to happen
689 bne-- a64RedriveAsDSI // got a DSI
691 bne-- a64RedriveAsDSI // got a DSI
693 bne-- a64RedriveAsDSI // got a DSI
695 bne-- a64RedriveAsDSI // got a DSI
696 mtmsr r26 // turn DR back off
698 addi r17,r17,4 // bump EA
700 a64Sm3: // r22 = 0-3, cr1 set on r22, cr0 beq set
701 beq cr1,a64Exit // no leftover bytes
702 ldx r30,r14,r21 // get last register
704 mtmsr r25 // turn on DR so we can access user space
705 isync // wait for it to happen
707 rlwinm r30,r30,8,0,31 // position next byte
708 stb r30,0(r17) // pack into user space
709 addi r17,r17,1 // bump user space ptr
710 bne-- a64RedriveAsDSI // got a DSI
712 mtmsr r26 // turn DR back off
717 // Subroutines to load bytes from user space.
719 a64Load2Bytes: // load 2 bytes right-justified into r30
720 addi r7,r17,1 // get EA of last byte
721 and r7,r7,r15 // clamp
722 cmpld r7,r17 // address wrap?
723 blt-- a64PassAlong // yes
724 mtmsr r25 // turn on DR so we can access user space
725 isync // wait for it to happen
726 sub. r30,r30,r30 // 0-fill dest and set beq
727 b a64Load2 // jump into routine
728 a64Load4Bytes: // load 4 bytes right-justified into r30 (ie, low order word)
729 addi r7,r17,3 // get EA of last byte
730 and r7,r7,r15 // clamp
731 cmpld r7,r17 // address wrap?
732 blt-- a64PassAlong // yes
733 mtmsr r25 // turn on DR so we can access user space
734 isync // wait for it to happen
735 sub. r30,r30,r30 // 0-fill dest and set beq
736 b a64Load4 // jump into routine
737 a64Load8Bytes: // load 8 bytes into r30
738 addi r7,r17,7 // get EA of last byte
739 and r7,r7,r15 // clamp
740 cmpld r7,r17 // address wrap?
741 blt-- a64PassAlong // yes
742 mtmsr r25 // turn on DR so we can access user space
743 isync // wait for it to happen
744 sub. r30,r30,r30 // 0-fill dest and set beq
745 lbz r3,-7(r7) // get byte 0
746 bne-- a64RedriveAsDSI // got a DSI
747 lbz r4,-6(r7) // and byte 1, etc
748 bne-- a64RedriveAsDSI // got a DSI
750 bne-- a64RedriveAsDSI // got a DSI
752 bne-- a64RedriveAsDSI // got a DSI
753 rldimi r30,r3,56,0 // position bytes in upper word
759 bne-- a64RedriveAsDSI // got a DSI
761 bne-- a64RedriveAsDSI // got a DSI
762 rldimi r30,r3,24,32 // insert bytes 4 and 5 into r30
766 bne-- a64RedriveAsDSI // got a DSI
768 bne-- a64RedriveAsDSI // got a DSI
769 mtmsr r26 // turn DR back off
771 rldimi r30,r3,8,48 // insert bytes 6 and 7 into r30
776 // Subroutines to store bytes into user space.
778 a64Store2Bytes: // store bytes 6 and 7 of r30
779 addi r7,r17,1 // get EA of last byte
780 and r7,r7,r15 // clamp
781 cmpld r7,r17 // address wrap?
782 blt-- a64PassAlong // yes
783 mtmsr r25 // turn on DR so we can access user space
784 isync // wait for it to happen
785 cmpw r0,r0 // set beq so we can check for DSI
786 b a64Store2 // jump into routine
787 a64Store4Bytes: // store bytes 4-7 of r30 (ie, low order word)
788 addi r7,r17,3 // get EA of last byte
789 and r7,r7,r15 // clamp
790 cmpld r7,r17 // address wrap?
791 blt-- a64PassAlong // yes
792 mtmsr r25 // turn on DR so we can access user space
793 isync // wait for it to happen
794 cmpw r0,r0 // set beq so we can check for DSI
795 b a64Store4 // jump into routine
796 a64Store8Bytes: // r30 = bytes
797 addi r7,r17,7 // get EA of last byte
798 and r7,r7,r15 // clamp
799 cmpld r7,r17 // address wrap?
800 blt-- a64PassAlong // yes
801 mtmsr r25 // turn on DR so we can access user space
802 isync // wait for it to happen
803 cmpw r0,r0 // set beq so we can check for DSI
804 rotldi r3,r30,8 // shift byte 0 into position
805 rotldi r4,r30,16 // and byte 1
806 rotldi r5,r30,24 // and byte 2
807 rotldi r6,r30,32 // and byte 3
808 stb r3,-7(r7) // store byte 0
809 bne-- a64RedriveAsDSI // got a DSI
810 stb r4,-6(r7) // and byte 1 etc...
811 bne-- a64RedriveAsDSI // got a DSI
813 bne-- a64RedriveAsDSI // got a DSI
815 bne-- a64RedriveAsDSI // got a DSI
817 rotldi r3,r30,40 // shift byte 4 into position
818 rotldi r4,r30,48 // and byte 5
820 bne-- a64RedriveAsDSI // got a DSI
822 bne-- a64RedriveAsDSI // got a DSI
824 rotldi r3,r30,56 // shift byte 6 into position
825 stb r3,-1(r7) // store byte 6
826 bne-- a64RedriveAsDSI // got a DSI
827 stb r30,0(r7) // store byte 7, which is already positioned
828 bne-- a64RedriveAsDSI // got a DSI
829 mtmsr r26 // turn off DR
837 li r30,T_EMULATE // Change exception code to emulate
838 stw r30,saveexception(r13) // Save it
839 b a64Exit // Join standard exit routine...
841 a64PassAlong: // unhandled exception, just pass it along
842 crset kNotify // return T_ALIGNMENT or T_PROGRAM
843 crclr kTrace // not a trace interrupt
845 a64UpdateCheck: // successfully emulated, may be update form
846 bf kUpdate,a64Exit // update?
847 stdx r17,r14,r16 // yes, store EA into RA
848 a64Exit: // instruction successfully emulated
849 addi r28,r28,4 // bump SRR0 past the emulated instruction
850 li r30,T_IN_VAIN // eat the interrupt since we emulated it
851 and r28,r28,r15 // clamp to address space size (32 vs 64)
852 std r28,savesrr0(r13) // save, so we return to next instruction
854 bt-- kTrace,a64Trace // were we in single-step at fault?
855 bt-- kNotify,a64Notify // should we say T_ALIGNMENT anyway?
857 mcrf cr6,cr3 // restore feature flags
858 mr r11,r30 // pass back exception code (T_IN_VAIN etc) in r11
859 b EXT(EmulExit) // return to exception processing
862 // Notification requested: pass exception upstairs even though it might have been emulated.
865 li r30,T_ALIGNMENT // somebody wants to know about it (but don't redrive)
866 bt kAlignment,a64Exit2 // was an alignment exception
867 li r30,T_PROGRAM // was an emulated instruction
871 // Emulate a trace interrupt after handling alignment interrupt.
874 lwz r9,SAVflags(r13) // get the save-area flags
876 oris r9,r9,hi16(SAVredrive) // Set the redrive bit
877 stw r30,saveexception(r13) // Set the exception code
878 stw r9,SAVflags(r13) // Set the flags
879 b a64Exit2 // Exit and do trace interrupt...
882 // Got a DSI accessing user space. Redrive. One way this can happen is if another
883 // processor removes a mapping while we are emulating.
885 a64RedriveAsISI: // this DSI happened fetching the opcode (r1==DSISR r4==DAR)
886 mtmsr r26 // turn DR back off
887 isync // wait for it to happen
888 li r30,T_INSTRUCTION_ACCESS
889 rlwimi r29,r1,0,0,4 // insert the fault type from DSI's DSISR
890 std r29,savesrr1(r13) // update SRR1 to look like an ISI
893 a64RedriveAsDSI: // r0==DAR r1==DSISR
894 mtmsr r26 // turn DR back off
895 isync // wait for it to happen
896 stw r1,savedsisr(r13) // Set the DSISR of failed access
897 std r0,savedar(r13) // Set the address of the failed access
898 li r30,T_DATA_ACCESS // Set failing data access code
900 lwz r9,SAVflags(r13) // Pick up the flags
901 stw r30,saveexception(r13) // Set the replacement code
902 oris r9,r9,hi16(SAVredrive) // Set the redrive bit
903 stw r9,SAVflags(r13) // Set redrive request
904 crclr kTrace // don't take a trace interrupt
905 crclr kNotify // don't pass alignment exception
909 // This is the branch table, indexed by the "AAAAB" opcode hash.
912 .long a64LwzLwzxLwarx // 00000 lwz[u], lwzx[u], lwarx
913 .long a64Ldx // 00001 ldx[u]
914 .long a64PassAlong // 00010 ldarx (never emulate these)
915 .long a64PassAlong // 00011
916 .long a64StwStwx // 00100 stw[u], stwx[u]
917 .long a64StdxStwcx // 00101 stdx[u], stwcx
918 .long a64PassAlong // 00110
919 .long a64PassAlong // 00111 stdcx (never emulate these)
920 .long a64LhzLhzx // 01000 lhz[u], lhzx[u]
921 .long a64PassAlong // 01001
922 .long a64LhaLhax // 01010 lha[u], lhax[u]
923 .long a64Lwax // 01011 lwax[u]
924 .long a64SthSthx // 01100 sth[u], sthx[u]
925 .long a64PassAlong // 01101
926 .long a64LmwStmw // 01110 lmw, stmw
927 .long a64PassAlong // 01111
928 .long a64LfsLfsx // 10000 lfs[u], lfsx[u]
929 .long a64LswxLwbrx // 10001 lswx, lwbrx
930 .long a64LfdLfdx // 10010 lfd[u], lfdx[u]
931 .long a64Lswi // 10011 lswi
932 .long a64StfsStfsx // 10100 stfs[u], stfsx[u]
933 .long a64StswxStwbrx // 10101 stswx, stwbrx
934 .long a64StfdStfdx // 10110 stfd[u], stfdx[u]
935 .long a64Stswi // 10111 stswi
936 .long a64PassAlong // 11000
937 .long a64Lhbrx // 11001 lhbrx
938 .long a64LdLwa // 11010 ld[u], lwa
939 .long a64PassAlong // 11011
940 .long a64PassAlong // 11100
941 .long a64Sthbrx // 11101 sthbrx
942 .long a64StdStfiwx // 11110 std[u], stfiwx
943 .long a64DcbzDcbz128 // 11111 dcbz, dcbz128