Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
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
8/*
9 * Constructor for a conventional segment GDT (or LDT) entry.
10 * This is a macro so it can be used in initializers.
11 */
12#define GDT_ENTRY(flags, base, limit) \
13 ((((base) & _AC(0xff000000,ULL)) << (56-24)) | \
14 (((flags) & _AC(0x0000f0ff,ULL)) << 40) | \
15 (((limit) & _AC(0x000f0000,ULL)) << (48-16)) | \
16 (((base) & _AC(0x00ffffff,ULL)) << 16) | \
17 (((limit) & _AC(0x0000ffff,ULL))))
18
19/* Simple and small GDT entries for booting only: */
20
21#define GDT_ENTRY_BOOT_CS 2
22#define GDT_ENTRY_BOOT_DS 3
23#define GDT_ENTRY_BOOT_TSS 4
24#define __BOOT_CS (GDT_ENTRY_BOOT_CS*8)
25#define __BOOT_DS (GDT_ENTRY_BOOT_DS*8)
26#define __BOOT_TSS (GDT_ENTRY_BOOT_TSS*8)
27
28/*
29 * Bottom two bits of selector give the ring
30 * privilege level
31 */
32#define SEGMENT_RPL_MASK 0x3
33
34/* User mode is privilege level 3: */
35#define USER_RPL 0x3
36
37/* Bit 2 is Table Indicator (TI): selects between LDT or GDT */
38#define SEGMENT_TI_MASK 0x4
39/* LDT segment has TI set ... */
40#define SEGMENT_LDT 0x4
41/* ... GDT has it cleared */
42#define SEGMENT_GDT 0x0
43
44#define GDT_ENTRY_INVALID_SEG 0
45
46#ifdef CONFIG_X86_32
47/*
48 * The layout of the per-CPU GDT under Linux:
49 *
50 * 0 - null <=== cacheline #1
51 * 1 - reserved
52 * 2 - reserved
53 * 3 - reserved
54 *
55 * 4 - unused <=== cacheline #2
56 * 5 - unused
57 *
58 * ------- start of TLS (Thread-Local Storage) segments:
59 *
60 * 6 - TLS segment #1 [ glibc's TLS segment ]
61 * 7 - TLS segment #2 [ Wine's %fs Win32 segment ]
62 * 8 - TLS segment #3 <=== cacheline #3
63 * 9 - reserved
64 * 10 - reserved
65 * 11 - reserved
66 *
67 * ------- start of kernel segments:
68 *
69 * 12 - kernel code segment <=== cacheline #4
70 * 13 - kernel data segment
71 * 14 - default user CS
72 * 15 - default user DS
73 * 16 - TSS <=== cacheline #5
74 * 17 - LDT
75 * 18 - PNPBIOS support (16->32 gate)
76 * 19 - PNPBIOS support
77 * 20 - PNPBIOS support <=== cacheline #6
78 * 21 - PNPBIOS support
79 * 22 - PNPBIOS support
80 * 23 - APM BIOS support
81 * 24 - APM BIOS support <=== cacheline #7
82 * 25 - APM BIOS support
83 *
84 * 26 - ESPFIX small SS
85 * 27 - per-cpu [ offset to per-cpu data area ]
86 * 28 - stack_canary-20 [ for stack protector ] <=== cacheline #8
87 * 29 - unused
88 * 30 - unused
89 * 31 - TSS for double fault handler
90 */
91#define GDT_ENTRY_TLS_MIN 6
92#define GDT_ENTRY_TLS_MAX (GDT_ENTRY_TLS_MIN + GDT_ENTRY_TLS_ENTRIES - 1)
93
94#define GDT_ENTRY_KERNEL_CS 12
95#define GDT_ENTRY_KERNEL_DS 13
96#define GDT_ENTRY_DEFAULT_USER_CS 14
97#define GDT_ENTRY_DEFAULT_USER_DS 15
98#define GDT_ENTRY_TSS 16
99#define GDT_ENTRY_LDT 17
100#define GDT_ENTRY_PNPBIOS_CS32 18
101#define GDT_ENTRY_PNPBIOS_CS16 19
102#define GDT_ENTRY_PNPBIOS_DS 20
103#define GDT_ENTRY_PNPBIOS_TS1 21
104#define GDT_ENTRY_PNPBIOS_TS2 22
105#define GDT_ENTRY_APMBIOS_BASE 23
106
107#define GDT_ENTRY_ESPFIX_SS 26
108#define GDT_ENTRY_PERCPU 27
109#define GDT_ENTRY_STACK_CANARY 28
110
111#define GDT_ENTRY_DOUBLEFAULT_TSS 31
112
113/*
114 * Number of entries in the GDT table:
115 */
116#define GDT_ENTRIES 32
117
118/*
119 * Segment selector values corresponding to the above entries:
120 */
121
122#define __KERNEL_CS (GDT_ENTRY_KERNEL_CS*8)
123#define __KERNEL_DS (GDT_ENTRY_KERNEL_DS*8)
124#define __USER_DS (GDT_ENTRY_DEFAULT_USER_DS*8 + 3)
125#define __USER_CS (GDT_ENTRY_DEFAULT_USER_CS*8 + 3)
126#define __ESPFIX_SS (GDT_ENTRY_ESPFIX_SS*8)
127
128/* segment for calling fn: */
129#define PNP_CS32 (GDT_ENTRY_PNPBIOS_CS32*8)
130/* code segment for BIOS: */
131#define PNP_CS16 (GDT_ENTRY_PNPBIOS_CS16*8)
132
133/* "Is this PNP code selector (PNP_CS32 or PNP_CS16)?" */
134#define SEGMENT_IS_PNP_CODE(x) (((x) & 0xf4) == PNP_CS32)
135
136/* data segment for BIOS: */
137#define PNP_DS (GDT_ENTRY_PNPBIOS_DS*8)
138/* transfer data segment: */
139#define PNP_TS1 (GDT_ENTRY_PNPBIOS_TS1*8)
140/* another data segment: */
141#define PNP_TS2 (GDT_ENTRY_PNPBIOS_TS2*8)
142
143#ifdef CONFIG_SMP
144# define __KERNEL_PERCPU (GDT_ENTRY_PERCPU*8)
145#else
146# define __KERNEL_PERCPU 0
147#endif
148
149#ifdef CONFIG_STACKPROTECTOR
150# define __KERNEL_STACK_CANARY (GDT_ENTRY_STACK_CANARY*8)
151#else
152# define __KERNEL_STACK_CANARY 0
153#endif
154
155#else /* 64-bit: */
156
157#include <asm/cache.h>
158
159#define GDT_ENTRY_KERNEL32_CS 1
160#define GDT_ENTRY_KERNEL_CS 2
161#define GDT_ENTRY_KERNEL_DS 3
162
163/*
164 * We cannot use the same code segment descriptor for user and kernel mode,
165 * not even in long flat mode, because of different DPL.
166 *
167 * GDT layout to get 64-bit SYSCALL/SYSRET support right. SYSRET hardcodes
168 * selectors:
169 *
170 * if returning to 32-bit userspace: cs = STAR.SYSRET_CS,
171 * if returning to 64-bit userspace: cs = STAR.SYSRET_CS+16,
172 *
173 * ss = STAR.SYSRET_CS+8 (in either case)
174 *
175 * thus USER_DS should be between 32-bit and 64-bit code selectors:
176 */
177#define GDT_ENTRY_DEFAULT_USER32_CS 4
178#define GDT_ENTRY_DEFAULT_USER_DS 5
179#define GDT_ENTRY_DEFAULT_USER_CS 6
180
181/* Needs two entries */
182#define GDT_ENTRY_TSS 8
183/* Needs two entries */
184#define GDT_ENTRY_LDT 10
185
186#define GDT_ENTRY_TLS_MIN 12
187#define GDT_ENTRY_TLS_MAX 14
188
189#define GDT_ENTRY_CPUNODE 15
190
191/*
192 * Number of entries in the GDT table:
193 */
194#define GDT_ENTRIES 16
195
196/*
197 * Segment selector values corresponding to the above entries:
198 *
199 * Note, selectors also need to have a correct RPL,
200 * expressed with the +3 value for user-space selectors:
201 */
202#define __KERNEL32_CS (GDT_ENTRY_KERNEL32_CS*8)
203#define __KERNEL_CS (GDT_ENTRY_KERNEL_CS*8)
204#define __KERNEL_DS (GDT_ENTRY_KERNEL_DS*8)
205#define __USER32_CS (GDT_ENTRY_DEFAULT_USER32_CS*8 + 3)
206#define __USER_DS (GDT_ENTRY_DEFAULT_USER_DS*8 + 3)
207#define __USER32_DS __USER_DS
208#define __USER_CS (GDT_ENTRY_DEFAULT_USER_CS*8 + 3)
209#define __CPUNODE_SEG (GDT_ENTRY_CPUNODE*8 + 3)
210
211#endif
212
213#ifndef CONFIG_PARAVIRT_XXL
214# define get_kernel_rpl() 0
215#endif
216
217#define IDT_ENTRIES 256
218#define NUM_EXCEPTION_VECTORS 32
219
220/* Bitmask of exception vectors which push an error code on the stack: */
221#define EXCEPTION_ERRCODE_MASK 0x00027d00
222
223#define GDT_SIZE (GDT_ENTRIES*8)
224#define GDT_ENTRY_TLS_ENTRIES 3
225#define TLS_SIZE (GDT_ENTRY_TLS_ENTRIES* 8)
226
227#ifdef CONFIG_X86_64
228
229/* Bit size and mask of CPU number stored in the per CPU data (and TSC_AUX) */
230#define VDSO_CPUNODE_BITS 12
231#define VDSO_CPUNODE_MASK 0xfff
232
233#ifndef __ASSEMBLY__
234
235/* Helper functions to store/load CPU and node numbers */
236
237static inline unsigned long vdso_encode_cpunode(int cpu, unsigned long node)
238{
239 return (node << VDSO_CPUNODE_BITS) | cpu;
240}
241
242static inline void vdso_read_cpunode(unsigned *cpu, unsigned *node)
243{
244 unsigned int p;
245
246 /*
247 * Load CPU and node number from the GDT. LSL is faster than RDTSCP
248 * and works on all CPUs. This is volatile so that it orders
249 * correctly with respect to barrier() and to keep GCC from cleverly
250 * hoisting it out of the calling function.
251 *
252 * If RDPID is available, use it.
253 */
254 alternative_io ("lsl %[seg],%[p]",
255 ".byte 0xf3,0x0f,0xc7,0xf8", /* RDPID %eax/rax */
256 X86_FEATURE_RDPID,
257 [p] "=a" (p), [seg] "r" (__CPUNODE_SEG));
258
259 if (cpu)
260 *cpu = (p & VDSO_CPUNODE_MASK);
261 if (node)
262 *node = (p >> VDSO_CPUNODE_BITS);
263}
264
265#endif /* !__ASSEMBLY__ */
266#endif /* CONFIG_X86_64 */
267
268#ifdef __KERNEL__
269
270/*
271 * early_idt_handler_array is an array of entry points referenced in the
272 * early IDT. For simplicity, it's a real array with one entry point
273 * every nine bytes. That leaves room for an optional 'push $0' if the
274 * vector has no error code (two bytes), a 'push $vector_number' (two
275 * bytes), and a jump to the common entry code (up to five bytes).
276 */
277#define EARLY_IDT_HANDLER_SIZE 9
278
279/*
280 * xen_early_idt_handler_array is for Xen pv guests: for each entry in
281 * early_idt_handler_array it contains a prequel in the form of
282 * pop %rcx; pop %r11; jmp early_idt_handler_array[i]; summing up to
283 * max 8 bytes.
284 */
285#define XEN_EARLY_IDT_HANDLER_SIZE 8
286
287#ifndef __ASSEMBLY__
288
289extern const char early_idt_handler_array[NUM_EXCEPTION_VECTORS][EARLY_IDT_HANDLER_SIZE];
290extern void early_ignore_irq(void);
291
292#if defined(CONFIG_X86_64) && defined(CONFIG_XEN_PV)
293extern const char xen_early_idt_handler_array[NUM_EXCEPTION_VECTORS][XEN_EARLY_IDT_HANDLER_SIZE];
294#endif
295
296/*
297 * Load a segment. Fall back on loading the zero segment if something goes
298 * wrong. This variant assumes that loading zero fully clears the segment.
299 * This is always the case on Intel CPUs and, even on 64-bit AMD CPUs, any
300 * failure to fully clear the cached descriptor is only observable for
301 * FS and GS.
302 */
303#define __loadsegment_simple(seg, value) \
304do { \
305 unsigned short __val = (value); \
306 \
307 asm volatile(" \n" \
308 "1: movl %k0,%%" #seg " \n" \
309 \
310 ".section .fixup,\"ax\" \n" \
311 "2: xorl %k0,%k0 \n" \
312 " jmp 1b \n" \
313 ".previous \n" \
314 \
315 _ASM_EXTABLE(1b, 2b) \
316 \
317 : "+r" (__val) : : "memory"); \
318} while (0)
319
320#define __loadsegment_ss(value) __loadsegment_simple(ss, (value))
321#define __loadsegment_ds(value) __loadsegment_simple(ds, (value))
322#define __loadsegment_es(value) __loadsegment_simple(es, (value))
323
324#ifdef CONFIG_X86_32
325
326/*
327 * On 32-bit systems, the hidden parts of FS and GS are unobservable if
328 * the selector is NULL, so there's no funny business here.
329 */
330#define __loadsegment_fs(value) __loadsegment_simple(fs, (value))
331#define __loadsegment_gs(value) __loadsegment_simple(gs, (value))
332
333#else
334
335static inline void __loadsegment_fs(unsigned short value)
336{
337 asm volatile(" \n"
338 "1: movw %0, %%fs \n"
339 "2: \n"
340
341 _ASM_EXTABLE_HANDLE(1b, 2b, ex_handler_clear_fs)
342
343 : : "rm" (value) : "memory");
344}
345
346/* __loadsegment_gs is intentionally undefined. Use load_gs_index instead. */
347
348#endif
349
350#define loadsegment(seg, value) __loadsegment_ ## seg (value)
351
352/*
353 * Save a segment register away:
354 */
355#define savesegment(seg, value) \
356 asm("mov %%" #seg ",%0":"=r" (value) : : "memory")
357
358/*
359 * x86-32 user GS accessors:
360 */
361#ifdef CONFIG_X86_32
362# ifdef CONFIG_X86_32_LAZY_GS
363# define get_user_gs(regs) (u16)({ unsigned long v; savesegment(gs, v); v; })
364# define set_user_gs(regs, v) loadsegment(gs, (unsigned long)(v))
365# define task_user_gs(tsk) ((tsk)->thread.gs)
366# define lazy_save_gs(v) savesegment(gs, (v))
367# define lazy_load_gs(v) loadsegment(gs, (v))
368# else /* X86_32_LAZY_GS */
369# define get_user_gs(regs) (u16)((regs)->gs)
370# define set_user_gs(regs, v) do { (regs)->gs = (v); } while (0)
371# define task_user_gs(tsk) (task_pt_regs(tsk)->gs)
372# define lazy_save_gs(v) do { } while (0)
373# define lazy_load_gs(v) do { } while (0)
374# endif /* X86_32_LAZY_GS */
375#endif /* X86_32 */
376
377#endif /* !__ASSEMBLY__ */
378#endif /* __KERNEL__ */
379
380#endif /* _ASM_X86_SEGMENT_H */