1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_SEGMENT_H
3 #define _ASM_X86_SEGMENT_H
4
5 #include <linux/const.h>
6 #include <asm/alternative.h>
7 #include <asm/ibt.h>
8
9 /*
10 * Constructor for a conventional segment GDT (or LDT) entry.
11 * This is a macro so it can be used in initializers.
12 */
13 #define GDT_ENTRY(flags, base, limit) \
14 ((((base) & _AC(0xff000000,ULL)) << (56-24)) | \
15 (((flags) & _AC(0x0000f0ff,ULL)) << 40) | \
16 (((limit) & _AC(0x000f0000,ULL)) << (48-16)) | \
17 (((base) & _AC(0x00ffffff,ULL)) << 16) | \
18 (((limit) & _AC(0x0000ffff,ULL))))
19
20 /* Simple and small GDT entries for booting only: */
21
22 #define GDT_ENTRY_BOOT_CS 2
23 #define GDT_ENTRY_BOOT_DS 3
24 #define GDT_ENTRY_BOOT_TSS 4
25 #define __BOOT_CS (GDT_ENTRY_BOOT_CS*8)
26 #define __BOOT_DS (GDT_ENTRY_BOOT_DS*8)
27 #define __BOOT_TSS (GDT_ENTRY_BOOT_TSS*8)
28
29 /*
30 * Bottom two bits of selector give the ring
31 * privilege level
32 */
33 #define SEGMENT_RPL_MASK 0x3
34
35 /*
36 * When running on Xen PV, the actual privilege level of the kernel is 1,
37 * not 0. Testing the Requested Privilege Level in a segment selector to
38 * determine whether the context is user mode or kernel mode with
39 * SEGMENT_RPL_MASK is wrong because the PV kernel's privilege level
40 * matches the 0x3 mask.
41 *
42 * Testing with USER_SEGMENT_RPL_MASK is valid for both native and Xen PV
43 * kernels because privilege level 2 is never used.
44 */
45 #define USER_SEGMENT_RPL_MASK 0x2
46
47 /* User mode is privilege level 3: */
48 #define USER_RPL 0x3
49
50 /* Bit 2 is Table Indicator (TI): selects between LDT or GDT */
51 #define SEGMENT_TI_MASK 0x4
52 /* LDT segment has TI set ... */
53 #define SEGMENT_LDT 0x4
54 /* ... GDT has it cleared */
55 #define SEGMENT_GDT 0x0
56
57 #define GDT_ENTRY_INVALID_SEG 0
58
59 #if defined(CONFIG_X86_32) && !defined(BUILD_VDSO32_64)
60 /*
61 * The layout of the per-CPU GDT under Linux:
62 *
63 * 0 - null <=== cacheline #1
64 * 1 - reserved
65 * 2 - reserved
66 * 3 - reserved
67 *
68 * 4 - unused <=== cacheline #2
69 * 5 - unused
70 *
71 * ------- start of TLS (Thread-Local Storage) segments:
72 *
73 * 6 - TLS segment #1 [ glibc's TLS segment ]
74 * 7 - TLS segment #2 [ Wine's %fs Win32 segment ]
75 * 8 - TLS segment #3 <=== cacheline #3
76 * 9 - reserved
77 * 10 - reserved
78 * 11 - reserved
79 *
80 * ------- start of kernel segments:
81 *
82 * 12 - kernel code segment <=== cacheline #4
83 * 13 - kernel data segment
84 * 14 - default user CS
85 * 15 - default user DS
86 * 16 - TSS <=== cacheline #5
87 * 17 - LDT
88 * 18 - PNPBIOS support (16->32 gate)
89 * 19 - PNPBIOS support
90 * 20 - PNPBIOS support <=== cacheline #6
91 * 21 - PNPBIOS support
92 * 22 - PNPBIOS support
93 * 23 - APM BIOS support
94 * 24 - APM BIOS support <=== cacheline #7
95 * 25 - APM BIOS support
96 *
97 * 26 - ESPFIX small SS
98 * 27 - per-cpu [ offset to per-cpu data area ]
99 * 28 - VDSO getcpu
100 * 29 - unused
101 * 30 - unused
102 * 31 - TSS for double fault handler
103 */
104 #define GDT_ENTRY_TLS_MIN 6
105 #define GDT_ENTRY_TLS_MAX (GDT_ENTRY_TLS_MIN + GDT_ENTRY_TLS_ENTRIES - 1)
106
107 #define GDT_ENTRY_KERNEL_CS 12
108 #define GDT_ENTRY_KERNEL_DS 13
109 #define GDT_ENTRY_DEFAULT_USER_CS 14
110 #define GDT_ENTRY_DEFAULT_USER_DS 15
111 #define GDT_ENTRY_TSS 16
112 #define GDT_ENTRY_LDT 17
113 #define GDT_ENTRY_PNPBIOS_CS32 18
114 #define GDT_ENTRY_PNPBIOS_CS16 19
115 #define GDT_ENTRY_PNPBIOS_DS 20
116 #define GDT_ENTRY_PNPBIOS_TS1 21
117 #define GDT_ENTRY_PNPBIOS_TS2 22
118 #define GDT_ENTRY_APMBIOS_BASE 23
119
120 #define GDT_ENTRY_ESPFIX_SS 26
121 #define GDT_ENTRY_PERCPU 27
122 #define GDT_ENTRY_CPUNODE 28
123
124 #define GDT_ENTRY_DOUBLEFAULT_TSS 31
125
126 /*
127 * Number of entries in the GDT table:
128 */
129 #define GDT_ENTRIES 32
130
131 /*
132 * Segment selector values corresponding to the above entries:
133 */
134
135 #define __KERNEL_CS (GDT_ENTRY_KERNEL_CS*8)
136 #define __KERNEL_DS (GDT_ENTRY_KERNEL_DS*8)
137 #define __USER_DS (GDT_ENTRY_DEFAULT_USER_DS*8 + 3)
138 #define __USER_CS (GDT_ENTRY_DEFAULT_USER_CS*8 + 3)
139 #define __USER32_CS __USER_CS
140 #define __ESPFIX_SS (GDT_ENTRY_ESPFIX_SS*8)
141
142 /* segment for calling fn: */
143 #define PNP_CS32 (GDT_ENTRY_PNPBIOS_CS32*8)
144 /* code segment for BIOS: */
145 #define PNP_CS16 (GDT_ENTRY_PNPBIOS_CS16*8)
146
147 /* "Is this PNP code selector (PNP_CS32 or PNP_CS16)?" */
148 #define SEGMENT_IS_PNP_CODE(x) (((x) & 0xf4) == PNP_CS32)
149
150 /* data segment for BIOS: */
151 #define PNP_DS (GDT_ENTRY_PNPBIOS_DS*8)
152 /* transfer data segment: */
153 #define PNP_TS1 (GDT_ENTRY_PNPBIOS_TS1*8)
154 /* another data segment: */
155 #define PNP_TS2 (GDT_ENTRY_PNPBIOS_TS2*8)
156
157 #ifdef CONFIG_SMP
158 # define __KERNEL_PERCPU (GDT_ENTRY_PERCPU*8)
159 #else
160 # define __KERNEL_PERCPU 0
161 #endif
162
163 #define __CPUNODE_SEG (GDT_ENTRY_CPUNODE*8 + 3)
164
165 #else /* 64-bit: */
166
167 #include <asm/cache.h>
168
169 #define GDT_ENTRY_KERNEL32_CS 1
170 #define GDT_ENTRY_KERNEL_CS 2
171 #define GDT_ENTRY_KERNEL_DS 3
172
173 /*
174 * We cannot use the same code segment descriptor for user and kernel mode,
175 * not even in long flat mode, because of different DPL.
176 *
177 * GDT layout to get 64-bit SYSCALL/SYSRET support right. SYSRET hardcodes
178 * selectors:
179 *
180 * if returning to 32-bit userspace: cs = STAR.SYSRET_CS,
181 * if returning to 64-bit userspace: cs = STAR.SYSRET_CS+16,
182 *
183 * ss = STAR.SYSRET_CS+8 (in either case)
184 *
185 * thus USER_DS should be between 32-bit and 64-bit code selectors:
186 */
187 #define GDT_ENTRY_DEFAULT_USER32_CS 4
188 #define GDT_ENTRY_DEFAULT_USER_DS 5
189 #define GDT_ENTRY_DEFAULT_USER_CS 6
190
191 /* Needs two entries */
192 #define GDT_ENTRY_TSS 8
193 /* Needs two entries */
194 #define GDT_ENTRY_LDT 10
195
196 #define GDT_ENTRY_TLS_MIN 12
197 #define GDT_ENTRY_TLS_MAX 14
198
199 #define GDT_ENTRY_CPUNODE 15
200
201 /*
202 * Number of entries in the GDT table:
203 */
204 #define GDT_ENTRIES 16
205
206 /*
207 * Segment selector values corresponding to the above entries:
208 *
209 * Note, selectors also need to have a correct RPL,
210 * expressed with the +3 value for user-space selectors:
211 */
212 #define __KERNEL32_CS (GDT_ENTRY_KERNEL32_CS*8)
213 #define __KERNEL_CS (GDT_ENTRY_KERNEL_CS*8)
214 #define __KERNEL_DS (GDT_ENTRY_KERNEL_DS*8)
215 #define __USER32_CS (GDT_ENTRY_DEFAULT_USER32_CS*8 + 3)
216 #define __USER_DS (GDT_ENTRY_DEFAULT_USER_DS*8 + 3)
217 #define __USER_CS (GDT_ENTRY_DEFAULT_USER_CS*8 + 3)
218 #define __CPUNODE_SEG (GDT_ENTRY_CPUNODE*8 + 3)
219
220 #endif
221
222 #define IDT_ENTRIES 256
223 #define NUM_EXCEPTION_VECTORS 32
224
225 /* Bitmask of exception vectors which push an error code on the stack: */
226 #define EXCEPTION_ERRCODE_MASK 0x20027d00
227
228 #define GDT_SIZE (GDT_ENTRIES*8)
229 #define GDT_ENTRY_TLS_ENTRIES 3
230 #define TLS_SIZE (GDT_ENTRY_TLS_ENTRIES* 8)
231
232 /* Bit size and mask of CPU number stored in the per CPU data (and TSC_AUX) */
233 #define VDSO_CPUNODE_BITS 12
234 #define VDSO_CPUNODE_MASK 0xfff
235
236 #ifndef __ASSEMBLY__
237
238 /* Helper functions to store/load CPU and node numbers */
239
240 static inline unsigned long vdso_encode_cpunode(int cpu, unsigned long node)
241 {
242 return (node << VDSO_CPUNODE_BITS) | cpu;
243 }
244
245 static inline void vdso_read_cpunode(unsigned *cpu, unsigned *node)
246 {
247 unsigned int p;
248
249 /*
250 * Load CPU and node number from the GDT. LSL is faster than RDTSCP
251 * and works on all CPUs. This is volatile so that it orders
252 * correctly with respect to barrier() and to keep GCC from cleverly
253 * hoisting it out of the calling function.
254 *
255 * If RDPID is available, use it.
256 */
257 alternative_io ("lsl %[seg],%[p]",
258 ".byte 0xf3,0x0f,0xc7,0xf8", /* RDPID %eax/rax */
259 X86_FEATURE_RDPID,
260 [p] "=a" (p), [seg] "r" (__CPUNODE_SEG));
261
262 if (cpu)
263 *cpu = (p & VDSO_CPUNODE_MASK);
264 if (node)
265 *node = (p >> VDSO_CPUNODE_BITS);
266 }
267
268 #endif /* !__ASSEMBLY__ */
269
270 #ifdef __KERNEL__
271
272 /*
273 * early_idt_handler_array is an array of entry points referenced in the
274 * early IDT. For simplicity, it's a real array with one entry point
275 * every nine bytes. That leaves room for an optional 'push $0' if the
276 * vector has no error code (two bytes), a 'push $vector_number' (two
277 * bytes), and a jump to the common entry code (up to five bytes).
278 */
279 #define EARLY_IDT_HANDLER_SIZE (9 + ENDBR_INSN_SIZE)
280
281 /*
282 * xen_early_idt_handler_array is for Xen pv guests: for each entry in
283 * early_idt_handler_array it contains a prequel in the form of
284 * pop %rcx; pop %r11; jmp early_idt_handler_array[i]; summing up to
285 * max 8 bytes.
286 */
287 #define XEN_EARLY_IDT_HANDLER_SIZE (8 + ENDBR_INSN_SIZE)
288
289 #ifndef __ASSEMBLY__
290
291 extern const char early_idt_handler_array[NUM_EXCEPTION_VECTORS][EARLY_IDT_HANDLER_SIZE];
292 extern void early_ignore_irq(void);
293
294 #ifdef CONFIG_XEN_PV
295 extern const char xen_early_idt_handler_array[NUM_EXCEPTION_VECTORS][XEN_EARLY_IDT_HANDLER_SIZE];
296 #endif
297
298 /*
299 * Load a segment. Fall back on loading the zero segment if something goes
300 * wrong. This variant assumes that loading zero fully clears the segment.
301 * This is always the case on Intel CPUs and, even on 64-bit AMD CPUs, any
302 * failure to fully clear the cached descriptor is only observable for
303 * FS and GS.
304 */
305 #define __loadsegment_simple(seg, value) \
306 do { \
307 unsigned short __val = (value); \
308 \
309 asm volatile(" \n" \
310 "1: movl %k0,%%" #seg " \n" \
311 _ASM_EXTABLE_TYPE_REG(1b, 1b, EX_TYPE_ZERO_REG, %k0)\
312 : "+r" (__val) : : "memory"); \
313 } while (0)
314
315 #define __loadsegment_ss(value) __loadsegment_simple(ss, (value))
316 #define __loadsegment_ds(value) __loadsegment_simple(ds, (value))
317 #define __loadsegment_es(value) __loadsegment_simple(es, (value))
318
319 #ifdef CONFIG_X86_32
320
321 /*
322 * On 32-bit systems, the hidden parts of FS and GS are unobservable if
323 * the selector is NULL, so there's no funny business here.
324 */
325 #define __loadsegment_fs(value) __loadsegment_simple(fs, (value))
326 #define __loadsegment_gs(value) __loadsegment_simple(gs, (value))
327
328 #else
329
330 static inline void __loadsegment_fs(unsigned short value)
331 {
332 asm volatile(" \n"
333 "1: movw %0, %%fs \n"
334 "2: \n"
335
336 _ASM_EXTABLE_TYPE(1b, 2b, EX_TYPE_CLEAR_FS)
337
338 : : "rm" (value) : "memory");
339 }
340
341 /* __loadsegment_gs is intentionally undefined. Use load_gs_index instead. */
342
343 #endif
344
345 #define loadsegment(seg, value) __loadsegment_ ## seg (value)
346
347 /*
348 * Save a segment register away:
349 */
350 #define savesegment(seg, value) \
351 asm("mov %%" #seg ",%0":"=r" (value) : : "memory")
352
353 #endif /* !__ASSEMBLY__ */
354 #endif /* __KERNEL__ */
355
356 #endif /* _ASM_X86_SEGMENT_H */
357
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.