kernel - Implement spectre mitigations part 3 (stabilization)
[dragonfly.git] / sys / cpu / x86_64 / include / asmacros.h
blobfb84979b52cf2122528e06a0a47964471e339639
1 /*
2 * Copyright (c) 1993 The Regents of the University of California.
3 * Copyright (c) 2008 The DragonFly Project.
4 * All rights reserved.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. Neither the name of the University nor the names of its contributors
15 * may be used to endorse or promote products derived from this software
16 * without specific prior written permission.
18 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
30 * $FreeBSD: src/sys/amd64/include/asmacros.h,v 1.32 2006/10/28 06:04:29 bde Exp $
33 #ifndef _CPU_ASMACROS_H_
34 #define _CPU_ASMACROS_H_
36 #include <sys/cdefs.h>
37 #include <machine/specialreg.h>
39 /* XXX too much duplication in various asm*.h's. */
42 * CNAME is used to manage the relationship between symbol names in C
43 * and the equivalent assembly language names. CNAME is given a name as
44 * it would be used in a C program. It expands to the equivalent assembly
45 * language name.
47 #define CNAME(csym) csym
49 #define ALIGN_DATA .p2align 3 /* 8 byte alignment, zero filled */
50 #ifdef GPROF
51 #define ALIGN_TEXT .p2align 4,0x90 /* 16-byte alignment, nop filled */
52 #else
53 #define ALIGN_TEXT .p2align 4,0x90 /* 16-byte alignment, nop filled */
54 #endif
55 #define SUPERALIGN_TEXT .p2align 4,0x90 /* 16-byte alignment, nop filled */
57 #define GEN_ENTRY(name) ALIGN_TEXT; .globl CNAME(name); \
58 .type CNAME(name),@function; CNAME(name):
59 #define NON_GPROF_ENTRY(name) GEN_ENTRY(name)
60 #define NON_GPROF_RET .byte 0xc3 /* opcode for `ret' */
62 #define END(name) .size name, . - name
64 #ifdef GPROF
66 * __mcount is like [.]mcount except that doesn't require its caller to set
67 * up a frame pointer. It must be called before pushing anything onto the
68 * stack. gcc should eventually generate code to call __mcount in most
69 * cases. This would make -pg in combination with -fomit-frame-pointer
70 * useful. gcc has a configuration variable PROFILE_BEFORE_PROLOGUE to
71 * allow profiling before setting up the frame pointer, but this is
72 * inadequate for good handling of special cases, e.g., -fpic works best
73 * with profiling after the prologue.
75 * [.]mexitcount is a new function to support non-statistical profiling if an
76 * accurate clock is available. For C sources, calls to it are generated
77 * by the FreeBSD extension `-mprofiler-epilogue' to gcc. It is best to
78 * call [.]mexitcount at the end of a function like the MEXITCOUNT macro does,
79 * but gcc currently generates calls to it at the start of the epilogue to
80 * avoid problems with -fpic.
82 * [.]mcount and __mcount may clobber the call-used registers and %ef.
83 * [.]mexitcount may clobber %ecx and %ef.
85 * Cross-jumping makes non-statistical profiling timing more complicated.
86 * It is handled in many cases by calling [.]mexitcount before jumping. It
87 * is handled for conditional jumps using CROSSJUMP() and CROSSJUMP_LABEL().
88 * It is handled for some fault-handling jumps by not sharing the exit
89 * routine.
91 * ALTENTRY() must be before a corresponding ENTRY() so that it can jump to
92 * the main entry point. Note that alt entries are counted twice. They
93 * have to be counted as ordinary entries for gprof to get the call times
94 * right for the ordinary entries.
96 * High local labels are used in macros to avoid clashes with local labels
97 * in functions.
99 * Ordinary `ret' is used instead of a macro `RET' because there are a lot
100 * of `ret's. 0xc3 is the opcode for `ret' (`#define ret ... ret' can't
101 * be used because this file is sometimes preprocessed in traditional mode).
102 * `ret' clobbers eflags but this doesn't matter.
104 #define ALTENTRY(name) GEN_ENTRY(name) ; MCOUNT ; MEXITCOUNT ; jmp 9f
105 #define CROSSJUMP(jtrue, label, jfalse) \
106 jfalse 8f; MEXITCOUNT; jmp __CONCAT(to,label); 8:
107 #define CROSSJUMPTARGET(label) \
108 ALIGN_TEXT; __CONCAT(to,label): ; MCOUNT; jmp label
109 #define ENTRY(name) GEN_ENTRY(name) ; 9: ; MCOUNT
110 #define FAKE_MCOUNT(caller) pushq caller ; call __mcount ; popq %rcx
111 #define MCOUNT call __mcount
112 #define MCOUNT_LABEL(name) GEN_ENTRY(name) ; nop ; ALIGN_TEXT
113 #ifdef GUPROF
114 #define MEXITCOUNT call .mexitcount
115 #define ret MEXITCOUNT ; NON_GPROF_RET
116 #else
117 #define MEXITCOUNT
118 #endif
120 #else /* !GPROF */
122 * ALTENTRY() has to align because it is before a corresponding ENTRY().
123 * ENTRY() has to align to because there may be no ALTENTRY() before it.
124 * If there is a previous ALTENTRY() then the alignment code for ENTRY()
125 * is empty.
127 #define ALTENTRY(name) GEN_ENTRY(name)
128 #define CROSSJUMP(jtrue, label, jfalse) jtrue label
129 #define CROSSJUMPTARGET(label)
130 #define ENTRY(name) GEN_ENTRY(name)
131 #define FAKE_MCOUNT(caller)
132 #define MCOUNT
133 #define MCOUNT_LABEL(name)
134 #define MEXITCOUNT
135 #endif /* GPROF */
137 #ifdef LOCORE
139 * Convenience macro for declaring interrupt entry points.
141 #define IDTVEC(name) ALIGN_TEXT; .globl __CONCAT(X,name); \
142 .type __CONCAT(X,name),@function; __CONCAT(X,name):
145 * stack frame macro support - supports mmu isolation, swapgs, and
146 * stack frame pushing and popping.
150 * Kernel pmap isolation to work-around the massive Intel mmu bug
151 * that allows kernel memory to be sussed out due to speculative memory
152 * reads and instruction execution creating timing differences that can
153 * be detected by userland. e.g. force speculative read, speculatively
154 * execute a cmp/branch sequence, detect timing. Iterate cmp $values
155 * to suss-out content of speculatively read kernel memory.
157 * We do this by creating a trampoline area for all user->kernel and
158 * kernel->user transitions. The trampoline area allows us to limit
159 * the reach the kernel map in the isolated version of the user pmap
160 * to JUST the trampoline area (for all cpus), tss, and vector area.
162 * It is very important that these transitions not access any memory
163 * outside of the trampoline page while the isolated user process pmap
164 * is active in %cr3.
166 * The trampoline does not add much overhead when pmap isolation is
167 * disabled, so we just run with it regardless. Of course, when pmap
168 * isolation is enabled, the %cr3 loads add 150-250ns to every system
169 * call as well as (without PCID) smash the TLB.
171 * KMMUENTER - Executed by the trampoline when a user->kernel transition
172 * is detected. The stack pointer points into the pcpu
173 * trampoline space and is available for register save/restore.
174 * Other registers have not yet been saved. %gs points at
175 * the kernel pcpu structure.
177 * Caller has already determined that a transition is in
178 * progress and has already issued the swapgs. hwtf indicates
179 * how much hardware has already pushed.
181 * KMMUEXIT - Executed when a kernel->user transition is made. The stack
182 * pointer points into the pcpu trampoline space and we are
183 * almost ready to iretq. %gs still points at the kernel pcpu
184 * structure.
186 * Caller has already determined that a transition is in
187 * progress. hwtf indicates how much hardware has already
188 * pushed.
192 * KMMUENTER_CORE - Handles ISOMMU, IBRS, and IBPB. Caller has already
193 * saved %rcx and %rdx. We have to deal with %rax.
195 * XXX If IBPB is not supported, try to clear the
196 * call return hw cache w/ many x chained call sequence?
198 * IBRS2 note - We are leaving IBRS on full-time. However, Intel
199 * believes it is not safe unless the MSR is poked on each user->kernel
200 * transition, so poke the MSR for both IBRS1 and IBRS2.
202 #define KMMUENTER_CORE \
203 testq $PCB_ISOMMU,PCPU(trampoline)+TR_PCB_FLAGS ; \
204 je 40f ; \
205 movq PCPU(trampoline)+TR_PCB_CR3,%rcx ; \
206 movq %rcx,%cr3 ; \
207 40: testq $PCB_IBRS1|PCB_IBRS2|PCB_IBPB,PCPU(trampoline)+TR_PCB_GFLAGS ;\
208 je 43f ; \
209 movq %rax, PCPU(trampoline)+TR_RAX ; \
210 testq $PCB_IBRS1|PCB_IBRS2,PCPU(trampoline)+TR_PCB_GFLAGS ; \
211 je 41f ; \
212 movl $MSR_SPEC_CTRL,%ecx ; \
213 movl $MSR_IBRS_ENABLE,%eax ; \
214 xorl %edx,%edx ; \
215 wrmsr ; \
216 41: testq $PCB_IBPB,PCPU(trampoline)+TR_PCB_GFLAGS ; \
217 je 42f ; \
218 movl $MSR_PRED_CMD,%ecx ; \
219 movl $MSR_IBPB_BARRIER,%eax ; \
220 xorl %edx,%edx ; \
221 wrmsr ; \
222 42: movq PCPU(trampoline)+TR_RAX, %rax ; \
223 43: \
227 * Enter with trampoline, hardware pushed up to %rip
229 #define KMMUENTER_TFRIP \
230 subq $TR_RIP, %rsp ; \
231 movq %rcx, TR_RCX(%rsp) ; \
232 movq %rdx, TR_RDX(%rsp) ; \
233 KMMUENTER_CORE ; \
234 movq %rsp, %rcx ; /* trampoline rsp */ \
235 movq PCPU(trampoline)+TR_PCB_RSP,%rsp ; /* kstack rsp */ \
236 movq TR_SS(%rcx), %rdx ; \
237 pushq %rdx ; \
238 movq TR_RSP(%rcx), %rdx ; \
239 pushq %rdx ; \
240 movq TR_RFLAGS(%rcx), %rdx ; \
241 pushq %rdx ; \
242 movq TR_CS(%rcx), %rdx ; \
243 pushq %rdx ; \
244 movq TR_RIP(%rcx), %rdx ; \
245 pushq %rdx ; \
246 movq TR_RDX(%rcx), %rdx ; \
247 movq TR_RCX(%rcx), %rcx \
250 * Enter with trampoline, hardware pushed up to ERR
252 #define KMMUENTER_TFERR \
253 subq $TR_ERR, %rsp ; \
254 movq %rcx, TR_RCX(%rsp) ; \
255 movq %rdx, TR_RDX(%rsp) ; \
256 KMMUENTER_CORE ; \
257 movq %rsp, %rcx ; /* trampoline rsp */ \
258 movq PCPU(trampoline)+TR_PCB_RSP,%rsp ; /* kstack rsp */ \
259 movq TR_SS(%rcx), %rdx ; \
260 pushq %rdx ; \
261 movq TR_RSP(%rcx), %rdx ; \
262 pushq %rdx ; \
263 movq TR_RFLAGS(%rcx), %rdx ; \
264 pushq %rdx ; \
265 movq TR_CS(%rcx), %rdx ; \
266 pushq %rdx ; \
267 movq TR_RIP(%rcx), %rdx ; \
268 pushq %rdx ; \
269 movq TR_ERR(%rcx), %rdx ; \
270 pushq %rdx ; \
271 movq TR_RDX(%rcx), %rdx ; \
272 movq TR_RCX(%rcx), %rcx \
275 * Enter with trampoline, hardware pushed up to ERR and
276 * we need to save %cr2 early (before potentially reloading %cr3).
278 #define KMMUENTER_TFERR_SAVECR2 \
279 subq $TR_ERR, %rsp ; \
280 movq %rcx, TR_RCX(%rsp) ; \
281 movq %rdx, TR_RDX(%rsp) ; \
282 movq %cr2, %rcx ; \
283 movq %rcx, PCPU(trampoline)+TR_CR2 ; \
284 KMMUENTER_CORE ; \
285 movq %rsp, %rcx ; /* trampoline rsp */ \
286 movq PCPU(trampoline)+TR_PCB_RSP,%rsp ; /* kstack rsp */ \
287 movq TR_SS(%rcx), %rdx ; \
288 pushq %rdx ; \
289 movq TR_RSP(%rcx), %rdx ; \
290 pushq %rdx ; \
291 movq TR_RFLAGS(%rcx), %rdx ; \
292 pushq %rdx ; \
293 movq TR_CS(%rcx), %rdx ; \
294 pushq %rdx ; \
295 movq TR_RIP(%rcx), %rdx ; \
296 pushq %rdx ; \
297 movq TR_ERR(%rcx), %rdx ; \
298 pushq %rdx ; \
299 movq TR_RDX(%rcx), %rdx ; \
300 movq TR_RCX(%rcx), %rcx \
303 * Set %cr3 if necessary on syscall entry. No registers may be
304 * disturbed.
306 * NOTE: TR_CR2 is used by the caller to save %rsp, we cannot use it here.
308 #define KMMUENTER_SYSCALL \
309 movq %rcx, PCPU(trampoline)+TR_RCX ; \
310 movq %rdx, PCPU(trampoline)+TR_RDX ; \
311 KMMUENTER_CORE ; \
312 movq PCPU(trampoline)+TR_RDX, %rdx ; \
313 movq PCPU(trampoline)+TR_RCX, %rcx \
316 * KMMUEXIT_CORE handles IBRS and IBPB, but not ISOMMU
318 * We don't re-execute the IBPB barrier on exit atm.
320 #define KMMUEXIT_CORE \
321 testq $PCB_IBRS1,PCPU(trampoline)+TR_PCB_GFLAGS ; \
322 je 41f ; \
323 movq %rax, PCPU(trampoline)+TR_RAX ; \
324 movq %rcx, PCPU(trampoline)+TR_RCX ; \
325 movq %rdx, PCPU(trampoline)+TR_RDX ; \
326 movl $MSR_SPEC_CTRL,%ecx ; \
327 movl $MSR_IBRS_DISABLE,%eax ; \
328 xorl %edx,%edx ; \
329 wrmsr ; \
330 movq PCPU(trampoline)+TR_RDX, %rdx ; \
331 movq PCPU(trampoline)+TR_RCX, %rcx ; \
332 movq PCPU(trampoline)+TR_RAX, %rax ; \
336 * We are positioned at the base of the trapframe. Advance the trapframe
337 * and handle MMU isolation. MMU isolation requires us to copy the
338 * hardware frame to the trampoline area before setting %cr3 to the
339 * isolated map. We then set the %rsp for iretq to TR_RIP in the
340 * trampoline area (after restoring the register we saved in TR_ERR).
342 #define KMMUEXIT \
343 addq $TF_RIP,%rsp ; \
344 KMMUEXIT_CORE ; \
345 testq $PCB_ISOMMU,PCPU(trampoline)+TR_PCB_FLAGS ; \
346 je 40f ; \
347 movq %rcx, PCPU(trampoline)+TR_ERR ; /* save in TR_ERR */ \
348 popq %rcx ; /* copy %rip */ \
349 movq %rcx, PCPU(trampoline)+TR_RIP ; \
350 popq %rcx ; /* copy %cs */ \
351 movq %rcx, PCPU(trampoline)+TR_CS ; \
352 popq %rcx ; /* copy %rflags */ \
353 movq %rcx, PCPU(trampoline)+TR_RFLAGS ; \
354 popq %rcx ; /* copy %rsp */ \
355 movq %rcx, PCPU(trampoline)+TR_RSP ; \
356 popq %rcx ; /* copy %ss */ \
357 movq %rcx, PCPU(trampoline)+TR_SS ; \
358 movq %gs:0,%rcx ; \
359 addq $GD_TRAMPOLINE+TR_ERR,%rcx ; \
360 movq %rcx,%rsp ; \
361 movq PCPU(trampoline)+TR_PCB_CR3_ISO,%rcx ; \
362 movq %rcx,%cr3 ; \
363 popq %rcx ; /* positioned at TR_RIP after this */ \
364 40: \
367 * Warning: user stack pointer already loaded into %rsp at this
368 * point. We still have the kernel %gs.
370 * Caller will sysexit, we do not have to copy anything to the
371 * trampoline area.
373 #define KMMUEXIT_SYSCALL \
374 KMMUEXIT_CORE ; \
375 testq $PCB_ISOMMU,PCPU(trampoline)+TR_PCB_FLAGS ; \
376 je 40f ; \
377 movq %rcx, PCPU(trampoline)+TR_RCX ; \
378 movq PCPU(trampoline)+TR_PCB_CR3_ISO,%rcx ; \
379 movq %rcx,%cr3 ; \
380 movq PCPU(trampoline)+TR_RCX, %rcx ; \
381 40: \
384 * Macros to create and destroy a trap frame. rsp has already been shifted
385 * to the base of the trapframe in the thread structure.
387 #define PUSH_FRAME_REGS \
388 movq %rdi,TF_RDI(%rsp) ; \
389 movq %rsi,TF_RSI(%rsp) ; \
390 movq %rdx,TF_RDX(%rsp) ; \
391 movq %rcx,TF_RCX(%rsp) ; \
392 movq %r8,TF_R8(%rsp) ; \
393 movq %r9,TF_R9(%rsp) ; \
394 movq %rax,TF_RAX(%rsp) ; \
395 movq %rbx,TF_RBX(%rsp) ; \
396 movq %rbp,TF_RBP(%rsp) ; \
397 movq %r10,TF_R10(%rsp) ; \
398 movq %r11,TF_R11(%rsp) ; \
399 movq %r12,TF_R12(%rsp) ; \
400 movq %r13,TF_R13(%rsp) ; \
401 movq %r14,TF_R14(%rsp) ; \
402 movq %r15,TF_R15(%rsp)
405 * PUSH_FRAME is the first thing executed upon interrupt entry. We are
406 * responsible for swapgs execution and the KMMUENTER dispatch.
408 #define PUSH_FRAME_TFRIP \
409 testb $SEL_RPL_MASK,TF_CS-TF_RIP(%rsp) ; /* from userland? */ \
410 jz 1f ; \
411 swapgs ; /* from userland */ \
412 KMMUENTER_TFRIP ; /* from userland */ \
413 1: \
414 subq $TF_RIP,%rsp ; \
415 PUSH_FRAME_REGS \
417 #define PUSH_FRAME_TFERR \
418 testb $SEL_RPL_MASK,TF_CS-TF_ERR(%rsp) ; /* from userland? */ \
419 jz 1f ; \
420 swapgs ; /* from userland */ \
421 KMMUENTER_TFERR ; /* from userland */ \
422 1: \
423 subq $TF_ERR,%rsp ; \
424 PUSH_FRAME_REGS \
426 #define PUSH_FRAME_TFERR_SAVECR2 \
427 testb $SEL_RPL_MASK,TF_CS-TF_ERR(%rsp) ; \
428 jz 1f ; \
429 swapgs ; /* from userland */ \
430 KMMUENTER_TFERR_SAVECR2 ;/* from userland */ \
431 subq $TF_ERR,%rsp ; \
432 PUSH_FRAME_REGS ; \
433 movq PCPU(trampoline)+TR_CR2, %r10 ; \
434 jmp 2f ; \
435 1: \
436 subq $TF_ERR,%rsp ; \
437 PUSH_FRAME_REGS ; \
438 movq %cr2, %r10 ; \
439 2: \
440 movq %r10, TF_ADDR(%rsp)
443 * POP_FRAME is issued just prior to the iretq, or just prior to a
444 * jmp doreti_iret. These must be passed in to the macro.
446 #define POP_FRAME(lastinsn) \
447 movq TF_RDI(%rsp),%rdi ; \
448 movq TF_RSI(%rsp),%rsi ; \
449 movq TF_RDX(%rsp),%rdx ; \
450 movq TF_RCX(%rsp),%rcx ; \
451 movq TF_R8(%rsp),%r8 ; \
452 movq TF_R9(%rsp),%r9 ; \
453 movq TF_RAX(%rsp),%rax ; \
454 movq TF_RBX(%rsp),%rbx ; \
455 movq TF_RBP(%rsp),%rbp ; \
456 movq TF_R10(%rsp),%r10 ; \
457 movq TF_R11(%rsp),%r11 ; \
458 movq TF_R12(%rsp),%r12 ; \
459 movq TF_R13(%rsp),%r13 ; \
460 movq TF_R14(%rsp),%r14 ; \
461 movq TF_R15(%rsp),%r15 ; \
462 cli ; \
463 testb $SEL_RPL_MASK,TF_CS(%rsp) ; /* return to user? */ \
464 jz 1f ; \
465 KMMUEXIT ; /* return to user */ \
466 swapgs ; /* return to user */ \
467 jmp 2f ; \
468 1: \
469 addq $TF_RIP,%rsp ; /* setup for iretq */ \
470 2: \
471 lastinsn
474 * Access per-CPU data.
476 #define PCPU(member) %gs:gd_ ## member
477 #define PCPU_E8(member,idx) %gs:gd_ ## member(,idx,8)
478 #define PCPU_ADDR(member, reg) \
479 movq %gs:PC_PRVSPACE, reg ; \
480 addq $PC_ ## member, reg
482 #endif /* LOCORE */
484 #endif /* !_CPU_ASMACROS_H_ */