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/*
35 * When running on Xen PV, the actual privilege level of the kernel is 1,
36 * not 0. Testing the Requested Privilege Level in a segment selector to
37 * determine whether the context is user mode or kernel mode with
38 * SEGMENT_RPL_MASK is wrong because the PV kernel's privilege level
39 * matches the 0x3 mask.
40 *
41 * Testing with USER_SEGMENT_RPL_MASK is valid for both native and Xen PV
42 * kernels because privilege level 2 is never used.
43 */
44#define USER_SEGMENT_RPL_MASK 0x2
45
46/* User mode is privilege level 3: */
47#define USER_RPL 0x3
48
49/* Bit 2 is Table Indicator (TI): selects between LDT or GDT */
50#define SEGMENT_TI_MASK 0x4
51/* LDT segment has TI set ... */
52#define SEGMENT_LDT 0x4
53/* ... GDT has it cleared */
54#define SEGMENT_GDT 0x0
55
56#define GDT_ENTRY_INVALID_SEG 0
57
58#ifdef CONFIG_X86_32
59/*
60 * The layout of the per-CPU GDT under Linux:
61 *
62 * 0 - null <=== cacheline #1
63 * 1 - reserved
64 * 2 - reserved
65 * 3 - reserved
66 *
67 * 4 - unused <=== cacheline #2
68 * 5 - unused
69 *
70 * ------- start of TLS (Thread-Local Storage) segments:
71 *
72 * 6 - TLS segment #1 [ glibc's TLS segment ]
73 * 7 - TLS segment #2 [ Wine's %fs Win32 segment ]
74 * 8 - TLS segment #3 <=== cacheline #3
75 * 9 - reserved
76 * 10 - reserved
77 * 11 - reserved
78 *
79 * ------- start of kernel segments:
80 *
81 * 12 - kernel code segment <=== cacheline #4
82 * 13 - kernel data segment
83 * 14 - default user CS
84 * 15 - default user DS
85 * 16 - TSS <=== cacheline #5
86 * 17 - LDT
87 * 18 - PNPBIOS support (16->32 gate)
88 * 19 - PNPBIOS support
89 * 20 - PNPBIOS support <=== cacheline #6
90 * 21 - PNPBIOS support
91 * 22 - PNPBIOS support
92 * 23 - APM BIOS support
93 * 24 - APM BIOS support <=== cacheline #7
94 * 25 - APM BIOS support
95 *
96 * 26 - ESPFIX small SS
97 * 27 - per-cpu [ offset to per-cpu data area ]
98 * 28 - unused
99 * 29 - unused
100 * 30 - unused
101 * 31 - TSS for double fault handler
102 */
103#define GDT_ENTRY_TLS_MIN 6
104#define GDT_ENTRY_TLS_MAX (GDT_ENTRY_TLS_MIN + GDT_ENTRY_TLS_ENTRIES - 1)
105
106#define GDT_ENTRY_KERNEL_CS 12
107#define GDT_ENTRY_KERNEL_DS 13
108#define GDT_ENTRY_DEFAULT_USER_CS 14
109#define GDT_ENTRY_DEFAULT_USER_DS 15
110#define GDT_ENTRY_TSS 16
111#define GDT_ENTRY_LDT 17
112#define GDT_ENTRY_PNPBIOS_CS32 18
113#define GDT_ENTRY_PNPBIOS_CS16 19
114#define GDT_ENTRY_PNPBIOS_DS 20
115#define GDT_ENTRY_PNPBIOS_TS1 21
116#define GDT_ENTRY_PNPBIOS_TS2 22
117#define GDT_ENTRY_APMBIOS_BASE 23
118
119#define GDT_ENTRY_ESPFIX_SS 26
120#define GDT_ENTRY_PERCPU 27
121
122#define GDT_ENTRY_DOUBLEFAULT_TSS 31
123
124/*
125 * Number of entries in the GDT table:
126 */
127#define GDT_ENTRIES 32
128
129/*
130 * Segment selector values corresponding to the above entries:
131 */
132
133#define __KERNEL_CS (GDT_ENTRY_KERNEL_CS*8)
134#define __KERNEL_DS (GDT_ENTRY_KERNEL_DS*8)
135#define __USER_DS (GDT_ENTRY_DEFAULT_USER_DS*8 + 3)
136#define __USER_CS (GDT_ENTRY_DEFAULT_USER_CS*8 + 3)
137#define __ESPFIX_SS (GDT_ENTRY_ESPFIX_SS*8)
138
139/* segment for calling fn: */
140#define PNP_CS32 (GDT_ENTRY_PNPBIOS_CS32*8)
141/* code segment for BIOS: */
142#define PNP_CS16 (GDT_ENTRY_PNPBIOS_CS16*8)
143
144/* "Is this PNP code selector (PNP_CS32 or PNP_CS16)?" */
145#define SEGMENT_IS_PNP_CODE(x) (((x) & 0xf4) == PNP_CS32)
146
147/* data segment for BIOS: */
148#define PNP_DS (GDT_ENTRY_PNPBIOS_DS*8)
149/* transfer data segment: */
150#define PNP_TS1 (GDT_ENTRY_PNPBIOS_TS1*8)
151/* another data segment: */
152#define PNP_TS2 (GDT_ENTRY_PNPBIOS_TS2*8)
153
154#ifdef CONFIG_SMP
155# define __KERNEL_PERCPU (GDT_ENTRY_PERCPU*8)
156#else
157# define __KERNEL_PERCPU 0
158#endif
159
160#else /* 64-bit: */
161
162#include <asm/cache.h>
163
164#define GDT_ENTRY_KERNEL32_CS 1
165#define GDT_ENTRY_KERNEL_CS 2
166#define GDT_ENTRY_KERNEL_DS 3
167
168/*
169 * We cannot use the same code segment descriptor for user and kernel mode,
170 * not even in long flat mode, because of different DPL.
171 *
172 * GDT layout to get 64-bit SYSCALL/SYSRET support right. SYSRET hardcodes
173 * selectors:
174 *
175 * if returning to 32-bit userspace: cs = STAR.SYSRET_CS,
176 * if returning to 64-bit userspace: cs = STAR.SYSRET_CS+16,
177 *
178 * ss = STAR.SYSRET_CS+8 (in either case)
179 *
180 * thus USER_DS should be between 32-bit and 64-bit code selectors:
181 */
182#define GDT_ENTRY_DEFAULT_USER32_CS 4
183#define GDT_ENTRY_DEFAULT_USER_DS 5
184#define GDT_ENTRY_DEFAULT_USER_CS 6
185
186/* Needs two entries */
187#define GDT_ENTRY_TSS 8
188/* Needs two entries */
189#define GDT_ENTRY_LDT 10
190
191#define GDT_ENTRY_TLS_MIN 12
192#define GDT_ENTRY_TLS_MAX 14
193
194#define GDT_ENTRY_CPUNODE 15
195
196/*
197 * Number of entries in the GDT table:
198 */
199#define GDT_ENTRIES 16
200
201/*
202 * Segment selector values corresponding to the above entries:
203 *
204 * Note, selectors also need to have a correct RPL,
205 * expressed with the +3 value for user-space selectors:
206 */
207#define __KERNEL32_CS (GDT_ENTRY_KERNEL32_CS*8)
208#define __KERNEL_CS (GDT_ENTRY_KERNEL_CS*8)
209#define __KERNEL_DS (GDT_ENTRY_KERNEL_DS*8)
210#define __USER32_CS (GDT_ENTRY_DEFAULT_USER32_CS*8 + 3)
211#define __USER_DS (GDT_ENTRY_DEFAULT_USER_DS*8 + 3)
212#define __USER32_DS __USER_DS
213#define __USER_CS (GDT_ENTRY_DEFAULT_USER_CS*8 + 3)
214#define __CPUNODE_SEG (GDT_ENTRY_CPUNODE*8 + 3)
215
216#endif
217
218#define IDT_ENTRIES 256
219#define NUM_EXCEPTION_VECTORS 32
220
221/* Bitmask of exception vectors which push an error code on the stack: */
222#define EXCEPTION_ERRCODE_MASK 0x20027d00
223
224#define GDT_SIZE (GDT_ENTRIES*8)
225#define GDT_ENTRY_TLS_ENTRIES 3
226#define TLS_SIZE (GDT_ENTRY_TLS_ENTRIES* 8)
227
228#ifdef CONFIG_X86_64
229
230/* Bit size and mask of CPU number stored in the per CPU data (and TSC_AUX) */
231#define VDSO_CPUNODE_BITS 12
232#define VDSO_CPUNODE_MASK 0xfff
233
234#ifndef __ASSEMBLY__
235
236/* Helper functions to store/load CPU and node numbers */
237
238static inline unsigned long vdso_encode_cpunode(int cpu, unsigned long node)
239{
240 return (node << VDSO_CPUNODE_BITS) | cpu;
241}
242
243static inline void vdso_read_cpunode(unsigned *cpu, unsigned *node)
244{
245 unsigned int p;
246
247 /*
248 * Load CPU and node number from the GDT. LSL is faster than RDTSCP
249 * and works on all CPUs. This is volatile so that it orders
250 * correctly with respect to barrier() and to keep GCC from cleverly
251 * hoisting it out of the calling function.
252 *
253 * If RDPID is available, use it.
254 */
255 alternative_io ("lsl %[seg],%[p]",
256 ".byte 0xf3,0x0f,0xc7,0xf8", /* RDPID %eax/rax */
257 X86_FEATURE_RDPID,
258 [p] "=a" (p), [seg] "r" (__CPUNODE_SEG));
259
260 if (cpu)
261 *cpu = (p & VDSO_CPUNODE_MASK);
262 if (node)
263 *node = (p >> VDSO_CPUNODE_BITS);
264}
265
266#endif /* !__ASSEMBLY__ */
267#endif /* CONFIG_X86_64 */
268
269#ifdef __KERNEL__
270
271/*
272 * early_idt_handler_array is an array of entry points referenced in the
273 * early IDT. For simplicity, it's a real array with one entry point
274 * every nine bytes. That leaves room for an optional 'push $0' if the
275 * vector has no error code (two bytes), a 'push $vector_number' (two
276 * bytes), and a jump to the common entry code (up to five bytes).
277 */
278#define EARLY_IDT_HANDLER_SIZE 9
279
280/*
281 * xen_early_idt_handler_array is for Xen pv guests: for each entry in
282 * early_idt_handler_array it contains a prequel in the form of
283 * pop %rcx; pop %r11; jmp early_idt_handler_array[i]; summing up to
284 * max 8 bytes.
285 */
286#define XEN_EARLY_IDT_HANDLER_SIZE 8
287
288#ifndef __ASSEMBLY__
289
290extern const char early_idt_handler_array[NUM_EXCEPTION_VECTORS][EARLY_IDT_HANDLER_SIZE];
291extern void early_ignore_irq(void);
292
293#ifdef CONFIG_XEN_PV
294extern const char xen_early_idt_handler_array[NUM_EXCEPTION_VECTORS][XEN_EARLY_IDT_HANDLER_SIZE];
295#endif
296
297/*
298 * Load a segment. Fall back on loading the zero segment if something goes
299 * wrong. This variant assumes that loading zero fully clears the segment.
300 * This is always the case on Intel CPUs and, even on 64-bit AMD CPUs, any
301 * failure to fully clear the cached descriptor is only observable for
302 * FS and GS.
303 */
304#define __loadsegment_simple(seg, value) \
305do { \
306 unsigned short __val = (value); \
307 \
308 asm volatile(" \n" \
309 "1: movl %k0,%%" #seg " \n" \
310 _ASM_EXTABLE_TYPE_REG(1b, 1b, EX_TYPE_ZERO_REG, %k0)\
311 : "+r" (__val) : : "memory"); \
312} while (0)
313
314#define __loadsegment_ss(value) __loadsegment_simple(ss, (value))
315#define __loadsegment_ds(value) __loadsegment_simple(ds, (value))
316#define __loadsegment_es(value) __loadsegment_simple(es, (value))
317
318#ifdef CONFIG_X86_32
319
320/*
321 * On 32-bit systems, the hidden parts of FS and GS are unobservable if
322 * the selector is NULL, so there's no funny business here.
323 */
324#define __loadsegment_fs(value) __loadsegment_simple(fs, (value))
325#define __loadsegment_gs(value) __loadsegment_simple(gs, (value))
326
327#else
328
329static inline void __loadsegment_fs(unsigned short value)
330{
331 asm volatile(" \n"
332 "1: movw %0, %%fs \n"
333 "2: \n"
334
335 _ASM_EXTABLE_TYPE(1b, 2b, EX_TYPE_CLEAR_FS)
336
337 : : "rm" (value) : "memory");
338}
339
340/* __loadsegment_gs is intentionally undefined. Use load_gs_index instead. */
341
342#endif
343
344#define loadsegment(seg, value) __loadsegment_ ## seg (value)
345
346/*
347 * Save a segment register away:
348 */
349#define savesegment(seg, value) \
350 asm("mov %%" #seg ",%0":"=r" (value) : : "memory")
351
352/*
353 * x86-32 user GS accessors. This is ugly and could do with some cleaning up.
354 */
355#ifdef CONFIG_X86_32
356# define get_user_gs(regs) (u16)({ unsigned long v; savesegment(gs, v); v; })
357# define set_user_gs(regs, v) loadsegment(gs, (unsigned long)(v))
358# define task_user_gs(tsk) ((tsk)->thread.gs)
359# define lazy_save_gs(v) savesegment(gs, (v))
360# define lazy_load_gs(v) loadsegment(gs, (v))
361# define load_gs_index(v) loadsegment(gs, (v))
362#endif /* X86_32 */
363
364#endif /* !__ASSEMBLY__ */
365#endif /* __KERNEL__ */
366
367#endif /* _ASM_X86_SEGMENT_H */