xref: /illumos-gate/usr/src/uts/intel/amd64/sys/privregs.h (revision b69c34dad3717624ff6b4f32b71014ee05b6a678)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * Copyright 2019 Joyent, Inc.
29  */
30 
31 #ifndef	_AMD64_SYS_PRIVREGS_H
32 #define	_AMD64_SYS_PRIVREGS_H
33 
34 #include <sys/ccompile.h>
35 
36 #ifdef __cplusplus
37 extern "C" {
38 #endif
39 
40 /*
41  * This file describes the cpu's privileged register set, and
42  * how the machine state is saved on the stack when a trap occurs.
43  */
44 
45 #if !defined(__amd64)
46 #error	"non-amd64 code depends on amd64 privileged header!"
47 #endif
48 
49 #ifndef _ASM
50 
51 /*
52  * This is NOT the structure to use for general purpose debugging;
53  * see /proc for that.  This is NOT the structure to use to decode
54  * the ucontext or grovel about in a core file; see <sys/regset.h>.
55  */
56 
57 struct regs {
58 	/*
59 	 * Extra frame for mdb to follow through high level interrupts and
60 	 * system traps.  Set them to 0 to terminate stacktrace.
61 	 */
62 	greg_t	r_savfp;	/* a copy of %rbp */
63 	greg_t	r_savpc;	/* a copy of %rip */
64 
65 	greg_t	r_rdi;		/* 1st arg to function */
66 	greg_t	r_rsi;		/* 2nd arg to function */
67 	greg_t	r_rdx;		/* 3rd arg to function, 2nd return register */
68 	greg_t	r_rcx;		/* 4th arg to function */
69 
70 	greg_t	r_r8;		/* 5th arg to function */
71 	greg_t	r_r9;		/* 6th arg to function */
72 	greg_t	r_rax;		/* 1st return register, # SSE registers */
73 	greg_t	r_rbx;		/* callee-saved, optional base pointer */
74 
75 	greg_t	r_rbp;		/* callee-saved, optional frame pointer */
76 	greg_t	r_r10;		/* temporary register, static chain pointer */
77 	greg_t	r_r11;		/* temporary register */
78 	greg_t	r_r12;		/* callee-saved */
79 
80 	greg_t	r_r13;		/* callee-saved */
81 	greg_t	r_r14;		/* callee-saved */
82 	greg_t	r_r15;		/* callee-saved */
83 
84 	/*
85 	 * fsbase and gsbase are sampled on every exception in DEBUG kernels
86 	 * only.  They remain in the non-DEBUG kernel to avoid any flag days.
87 	 */
88 	greg_t	__r_fsbase;	/* no longer used in non-DEBUG builds */
89 	greg_t	__r_gsbase;	/* no longer used in non-DEBUG builds */
90 	greg_t	r_ds;
91 	greg_t	r_es;
92 	greg_t	r_fs;		/* %fs is *never* used by the kernel */
93 	greg_t	r_gs;
94 
95 	greg_t	r_trapno;
96 
97 	/*
98 	 * (the rest of these are defined by the hardware)
99 	 */
100 	greg_t	r_err;
101 	greg_t	r_rip;
102 	greg_t	r_cs;
103 	greg_t	r_rfl;
104 	greg_t	r_rsp;
105 	greg_t	r_ss;
106 };
107 
108 #define	r_r0	r_rax	/* r0 for portability */
109 #define	r_r1	r_rdx	/* r1 for portability */
110 #define	r_fp	r_rbp	/* kernel frame pointer */
111 #define	r_sp	r_rsp	/* user stack pointer */
112 #define	r_pc	r_rip	/* user's instruction pointer */
113 #define	r_ps	r_rfl	/* user's RFLAGS */
114 
115 #ifdef _KERNEL
116 #define	lwptoregs(lwp)	((struct regs *)((lwp)->lwp_regs))
117 #endif /* _KERNEL */
118 
119 #else	/* !_ASM */
120 
121 #if defined(_MACHDEP)
122 
123 #include <sys/machprivregs.h>
124 #include <sys/pcb.h>
125 
126 /*
127  * We can not safely sample {fs,gs}base on the hypervisor. The rdmsr
128  * instruction triggers a #gp fault which is emulated in the hypervisor
129  * on behalf of the guest. This is normally ok but if the guest is in
130  * the special failsafe handler it must not fault again or the hypervisor
131  * will kill the domain. We could use something different than INTR_PUSH
132  * in xen_failsafe_callback but for now we will not sample them.
133  */
134 #if defined(DEBUG) && !defined(__xpv)
135 #define	__SAVE_BASES				\
136 	movl    $MSR_AMD_FSBASE, %ecx;          \
137 	rdmsr;                                  \
138 	movl    %eax, REGOFF_FSBASE(%rsp);      \
139 	movl    %edx, REGOFF_FSBASE+4(%rsp);    \
140 	movl    $MSR_AMD_GSBASE, %ecx;          \
141 	rdmsr;                                  \
142 	movl    %eax, REGOFF_GSBASE(%rsp);      \
143 	movl    %edx, REGOFF_GSBASE+4(%rsp)
144 #else
145 #define	__SAVE_BASES
146 #endif
147 
148 /*
149  * Create a struct regs on the stack suitable for an
150  * interrupt trap.
151  *
152  * Assumes that the trap handler has already pushed an
153  * appropriate r_err and r_trapno
154  */
155 #define	__SAVE_REGS				\
156 	movq	%r15, REGOFF_R15(%rsp);		\
157 	movq	%r14, REGOFF_R14(%rsp);		\
158 	movq	%r13, REGOFF_R13(%rsp);		\
159 	movq	%r12, REGOFF_R12(%rsp);		\
160 	movq	%r11, REGOFF_R11(%rsp);		\
161 	movq	%r10, REGOFF_R10(%rsp);		\
162 	movq	%rbp, REGOFF_RBP(%rsp);		\
163 	movq	%rbx, REGOFF_RBX(%rsp);		\
164 	movq	%rax, REGOFF_RAX(%rsp);		\
165 	movq	%r9, REGOFF_R9(%rsp);		\
166 	movq	%r8, REGOFF_R8(%rsp);		\
167 	movq	%rcx, REGOFF_RCX(%rsp);		\
168 	movq	%rdx, REGOFF_RDX(%rsp);		\
169 	movq	%rsi, REGOFF_RSI(%rsp);		\
170 	movq	%rdi, REGOFF_RDI(%rsp);		\
171 	movq	%rbp, REGOFF_SAVFP(%rsp);	\
172 	movq	REGOFF_RIP(%rsp), %rcx;		\
173 	movq	%rcx, REGOFF_SAVPC(%rsp);	\
174 	xorl	%ecx, %ecx;			\
175 	movw	%gs, %cx;			\
176 	movq	%rcx, REGOFF_GS(%rsp);		\
177 	movw	%fs, %cx;			\
178 	movq	%rcx, REGOFF_FS(%rsp);		\
179 	movw	%es, %cx;			\
180 	movq	%rcx, REGOFF_ES(%rsp);		\
181 	movw	%ds, %cx;			\
182 	movq	%rcx, REGOFF_DS(%rsp);		\
183 	__SAVE_BASES
184 
185 #define	__RESTORE_REGS				\
186 	movq	REGOFF_RDI(%rsp),	%rdi;	\
187 	movq	REGOFF_RSI(%rsp),	%rsi;	\
188 	movq	REGOFF_RDX(%rsp),	%rdx;	\
189 	movq	REGOFF_RCX(%rsp),	%rcx;	\
190 	movq	REGOFF_R8(%rsp),	%r8;	\
191 	movq	REGOFF_R9(%rsp),	%r9;	\
192 	movq	REGOFF_RAX(%rsp),	%rax;	\
193 	movq	REGOFF_RBX(%rsp),	%rbx;	\
194 	movq	REGOFF_RBP(%rsp),	%rbp;	\
195 	movq	REGOFF_R10(%rsp),	%r10;	\
196 	movq	REGOFF_R11(%rsp),	%r11;	\
197 	movq	REGOFF_R12(%rsp),	%r12;	\
198 	movq	REGOFF_R13(%rsp),	%r13;	\
199 	movq	REGOFF_R14(%rsp),	%r14;	\
200 	movq	REGOFF_R15(%rsp),	%r15
201 
202 /*
203  * Push register state onto the stack. If we've
204  * interrupted userland, do a swapgs as well.
205  */
206 #define	INTR_PUSH				\
207 	subq	$REGOFF_TRAPNO, %rsp;		\
208 	__SAVE_REGS;				\
209 	cmpw	$KCS_SEL, REGOFF_CS(%rsp);	\
210 	je	6f;				\
211 	movq	$0, REGOFF_SAVFP(%rsp);		\
212 	SWAPGS;					\
213 6:	lfence; /* swapgs mitigation */		\
214 	CLEAN_CS
215 
216 #define	INTR_POP			\
217 	leaq	sys_lcall32(%rip), %r11;\
218 	cmpq	%r11, REGOFF_RIP(%rsp);	\
219 	__RESTORE_REGS;			\
220 	je	5f;			\
221 	cmpw	$KCS_SEL, REGOFF_CS(%rsp);\
222 	je	8f;			\
223 5:	SWAPGS;				\
224 8:	lfence; /* swapgs mitigation */	\
225 	addq	$REGOFF_RIP, %rsp
226 
227 /*
228  * No need for swapgs mitigation: it's unconditional, and we're heading
229  * back to userspace.
230  */
231 #define	USER_POP			\
232 	__RESTORE_REGS;			\
233 	SWAPGS;				\
234 	addq	$REGOFF_RIP, %rsp	/* Adjust %rsp to prepare for iretq */
235 
236 #define	USER32_POP			\
237 	movl	REGOFF_RDI(%rsp), %edi;	\
238 	movl	REGOFF_RSI(%rsp), %esi;	\
239 	movl	REGOFF_RDX(%rsp), %edx;	\
240 	movl	REGOFF_RCX(%rsp), %ecx;	\
241 	movl	REGOFF_RAX(%rsp), %eax;	\
242 	movl	REGOFF_RBX(%rsp), %ebx;	\
243 	movl	REGOFF_RBP(%rsp), %ebp;	\
244 	SWAPGS;				\
245 	addq	$REGOFF_RIP, %rsp	/* Adjust %rsp to prepare for iretq */
246 
247 #define	DFTRAP_PUSH				\
248 	subq	$REGOFF_TRAPNO, %rsp;		\
249 	__SAVE_REGS
250 
251 #endif	/* _MACHDEP */
252 
253 /*
254  * Used to set rflags to known values at the head of an
255  * interrupt gate handler, i.e. interrupts are -already- disabled.
256  */
257 #define	INTGATE_INIT_KERNEL_FLAGS	\
258 	pushq	$F_OFF;			\
259 	popfq
260 
261 #endif	/* !_ASM */
262 
263 #include <sys/controlregs.h>
264 
265 #if defined(_KERNEL) && !defined(_ASM)
266 #if !defined(__lint) && defined(__GNUC__)
267 
268 extern __GNU_INLINE ulong_t
269 getcr8(void)
270 {
271 	uint64_t value;
272 
273 	__asm__ __volatile__(
274 	    "movq %%cr8, %0"
275 	    : "=r" (value));
276 	return (value);
277 }
278 
279 extern __GNU_INLINE void
280 setcr8(ulong_t value)
281 {
282 	__asm__ __volatile__(
283 	    "movq %0, %%cr8"
284 	    : /* no output */
285 	    : "r" (value));
286 }
287 
288 #else
289 
290 extern ulong_t getcr8(void);
291 extern void setcr8(ulong_t);
292 
293 #endif	/* !defined(__lint) && defined(__GNUC__) */
294 #endif	/* _KERNEL && !_ASM */
295 
296 /* Control register layout for panic dump */
297 
298 #define	CREGSZ		0x68
299 #define	CREG_GDT	0
300 #define	CREG_IDT	0x10
301 #define	CREG_LDT	0x20
302 #define	CREG_TASKR	0x28
303 #define	CREG_CR0	0x30
304 #define	CREG_CR2	0x38
305 #define	CREG_CR3	0x40
306 #define	CREG_CR4	0x48
307 #define	CREG_CR8	0x50
308 #define	CREG_KGSBASE	0x58
309 #define	CREG_EFER	0x60
310 
311 #if !defined(_ASM) && defined(_INT64_TYPE)
312 
313 typedef	uint64_t	creg64_t;
314 typedef	upad128_t	creg128_t;
315 
316 struct cregs {
317 	creg128_t	cr_gdt;
318 	creg128_t	cr_idt;
319 	creg64_t	cr_ldt;
320 	creg64_t	cr_task;
321 	creg64_t	cr_cr0;
322 	creg64_t	cr_cr2;
323 	creg64_t	cr_cr3;
324 	creg64_t	cr_cr4;
325 	creg64_t	cr_cr8;
326 	creg64_t	cr_kgsbase;
327 	creg64_t	cr_efer;
328 };
329 
330 #if defined(_KERNEL)
331 extern void getcregs(struct cregs *);
332 #endif	/* _KERNEL */
333 
334 #endif	/* !_ASM && _INT64_TYPE */
335 
336 #ifdef __cplusplus
337 }
338 #endif
339 
340 #endif	/* !_AMD64_SYS_PRIVREGS_H */
341