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 - stack_canary-20 [ for stack protector ] <=== cacheline #8
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#define GDT_ENTRY_STACK_CANARY 28
122
123#define GDT_ENTRY_DOUBLEFAULT_TSS 31
124
125/*
126 * Number of entries in the GDT table:
127 */
128#define GDT_ENTRIES 32
129
130/*
131 * Segment selector values corresponding to the above entries:
132 */
133
134#define __KERNEL_CS (GDT_ENTRY_KERNEL_CS*8)
135#define __KERNEL_DS (GDT_ENTRY_KERNEL_DS*8)
136#define __USER_DS (GDT_ENTRY_DEFAULT_USER_DS*8 + 3)
137#define __USER_CS (GDT_ENTRY_DEFAULT_USER_CS*8 + 3)
138#define __ESPFIX_SS (GDT_ENTRY_ESPFIX_SS*8)
139
140/* segment for calling fn: */
141#define PNP_CS32 (GDT_ENTRY_PNPBIOS_CS32*8)
142/* code segment for BIOS: */
143#define PNP_CS16 (GDT_ENTRY_PNPBIOS_CS16*8)
144
145/* "Is this PNP code selector (PNP_CS32 or PNP_CS16)?" */
146#define SEGMENT_IS_PNP_CODE(x) (((x) & 0xf4) == PNP_CS32)
147
148/* data segment for BIOS: */
149#define PNP_DS (GDT_ENTRY_PNPBIOS_DS*8)
150/* transfer data segment: */
151#define PNP_TS1 (GDT_ENTRY_PNPBIOS_TS1*8)
152/* another data segment: */
153#define PNP_TS2 (GDT_ENTRY_PNPBIOS_TS2*8)
154
155#ifdef CONFIG_SMP
156# define __KERNEL_PERCPU (GDT_ENTRY_PERCPU*8)
157#else
158# define __KERNEL_PERCPU 0
159#endif
160
161#ifdef CONFIG_STACKPROTECTOR
162# define __KERNEL_STACK_CANARY (GDT_ENTRY_STACK_CANARY*8)
163#else
164# define __KERNEL_STACK_CANARY 0
165#endif
166
167#else /* 64-bit: */
168
169#include <asm/cache.h>
170
171#define GDT_ENTRY_KERNEL32_CS 1
172#define GDT_ENTRY_KERNEL_CS 2
173#define GDT_ENTRY_KERNEL_DS 3
174
175/*
176 * We cannot use the same code segment descriptor for user and kernel mode,
177 * not even in long flat mode, because of different DPL.
178 *
179 * GDT layout to get 64-bit SYSCALL/SYSRET support right. SYSRET hardcodes
180 * selectors:
181 *
182 * if returning to 32-bit userspace: cs = STAR.SYSRET_CS,
183 * if returning to 64-bit userspace: cs = STAR.SYSRET_CS+16,
184 *
185 * ss = STAR.SYSRET_CS+8 (in either case)
186 *
187 * thus USER_DS should be between 32-bit and 64-bit code selectors:
188 */
189#define GDT_ENTRY_DEFAULT_USER32_CS 4
190#define GDT_ENTRY_DEFAULT_USER_DS 5
191#define GDT_ENTRY_DEFAULT_USER_CS 6
192
193/* Needs two entries */
194#define GDT_ENTRY_TSS 8
195/* Needs two entries */
196#define GDT_ENTRY_LDT 10
197
198#define GDT_ENTRY_TLS_MIN 12
199#define GDT_ENTRY_TLS_MAX 14
200
201#define GDT_ENTRY_CPUNODE 15
202
203/*
204 * Number of entries in the GDT table:
205 */
206#define GDT_ENTRIES 16
207
208/*
209 * Segment selector values corresponding to the above entries:
210 *
211 * Note, selectors also need to have a correct RPL,
212 * expressed with the +3 value for user-space selectors:
213 */
214#define __KERNEL32_CS (GDT_ENTRY_KERNEL32_CS*8)
215#define __KERNEL_CS (GDT_ENTRY_KERNEL_CS*8)
216#define __KERNEL_DS (GDT_ENTRY_KERNEL_DS*8)
217#define __USER32_CS (GDT_ENTRY_DEFAULT_USER32_CS*8 + 3)
218#define __USER_DS (GDT_ENTRY_DEFAULT_USER_DS*8 + 3)
219#define __USER32_DS __USER_DS
220#define __USER_CS (GDT_ENTRY_DEFAULT_USER_CS*8 + 3)
221#define __CPUNODE_SEG (GDT_ENTRY_CPUNODE*8 + 3)
222
223#endif
224
225#ifndef CONFIG_PARAVIRT_XXL
226# define get_kernel_rpl() 0
227#endif
228
229#define IDT_ENTRIES 256
230#define NUM_EXCEPTION_VECTORS 32
231
232/* Bitmask of exception vectors which push an error code on the stack: */
233#define EXCEPTION_ERRCODE_MASK 0x00027d00
234
235#define GDT_SIZE (GDT_ENTRIES*8)
236#define GDT_ENTRY_TLS_ENTRIES 3
237#define TLS_SIZE (GDT_ENTRY_TLS_ENTRIES* 8)
238
239#ifdef CONFIG_X86_64
240
241/* Bit size and mask of CPU number stored in the per CPU data (and TSC_AUX) */
242#define VDSO_CPUNODE_BITS 12
243#define VDSO_CPUNODE_MASK 0xfff
244
245#ifndef __ASSEMBLY__
246
247/* Helper functions to store/load CPU and node numbers */
248
249static inline unsigned long vdso_encode_cpunode(int cpu, unsigned long node)
250{
251 return (node << VDSO_CPUNODE_BITS) | cpu;
252}
253
254static inline void vdso_read_cpunode(unsigned *cpu, unsigned *node)
255{
256 unsigned int p;
257
258 /*
259 * Load CPU and node number from the GDT. LSL is faster than RDTSCP
260 * and works on all CPUs. This is volatile so that it orders
261 * correctly with respect to barrier() and to keep GCC from cleverly
262 * hoisting it out of the calling function.
263 *
264 * If RDPID is available, use it.
265 */
266 alternative_io ("lsl %[seg],%[p]",
267 ".byte 0xf3,0x0f,0xc7,0xf8", /* RDPID %eax/rax */
268 X86_FEATURE_RDPID,
269 [p] "=a" (p), [seg] "r" (__CPUNODE_SEG));
270
271 if (cpu)
272 *cpu = (p & VDSO_CPUNODE_MASK);
273 if (node)
274 *node = (p >> VDSO_CPUNODE_BITS);
275}
276
277#endif /* !__ASSEMBLY__ */
278#endif /* CONFIG_X86_64 */
279
280#ifdef __KERNEL__
281
282/*
283 * early_idt_handler_array is an array of entry points referenced in the
284 * early IDT. For simplicity, it's a real array with one entry point
285 * every nine bytes. That leaves room for an optional 'push $0' if the
286 * vector has no error code (two bytes), a 'push $vector_number' (two
287 * bytes), and a jump to the common entry code (up to five bytes).
288 */
289#define EARLY_IDT_HANDLER_SIZE 9
290
291/*
292 * xen_early_idt_handler_array is for Xen pv guests: for each entry in
293 * early_idt_handler_array it contains a prequel in the form of
294 * pop %rcx; pop %r11; jmp early_idt_handler_array[i]; summing up to
295 * max 8 bytes.
296 */
297#define XEN_EARLY_IDT_HANDLER_SIZE 8
298
299#ifndef __ASSEMBLY__
300
301extern const char early_idt_handler_array[NUM_EXCEPTION_VECTORS][EARLY_IDT_HANDLER_SIZE];
302extern void early_ignore_irq(void);
303
304#ifdef CONFIG_XEN_PV
305extern const char xen_early_idt_handler_array[NUM_EXCEPTION_VECTORS][XEN_EARLY_IDT_HANDLER_SIZE];
306#endif
307
308/*
309 * Load a segment. Fall back on loading the zero segment if something goes
310 * wrong. This variant assumes that loading zero fully clears the segment.
311 * This is always the case on Intel CPUs and, even on 64-bit AMD CPUs, any
312 * failure to fully clear the cached descriptor is only observable for
313 * FS and GS.
314 */
315#define __loadsegment_simple(seg, value) \
316do { \
317 unsigned short __val = (value); \
318 \
319 asm volatile(" \n" \
320 "1: movl %k0,%%" #seg " \n" \
321 \
322 ".section .fixup,\"ax\" \n" \
323 "2: xorl %k0,%k0 \n" \
324 " jmp 1b \n" \
325 ".previous \n" \
326 \
327 _ASM_EXTABLE(1b, 2b) \
328 \
329 : "+r" (__val) : : "memory"); \
330} while (0)
331
332#define __loadsegment_ss(value) __loadsegment_simple(ss, (value))
333#define __loadsegment_ds(value) __loadsegment_simple(ds, (value))
334#define __loadsegment_es(value) __loadsegment_simple(es, (value))
335
336#ifdef CONFIG_X86_32
337
338/*
339 * On 32-bit systems, the hidden parts of FS and GS are unobservable if
340 * the selector is NULL, so there's no funny business here.
341 */
342#define __loadsegment_fs(value) __loadsegment_simple(fs, (value))
343#define __loadsegment_gs(value) __loadsegment_simple(gs, (value))
344
345#else
346
347static inline void __loadsegment_fs(unsigned short value)
348{
349 asm volatile(" \n"
350 "1: movw %0, %%fs \n"
351 "2: \n"
352
353 _ASM_EXTABLE_HANDLE(1b, 2b, ex_handler_clear_fs)
354
355 : : "rm" (value) : "memory");
356}
357
358/* __loadsegment_gs is intentionally undefined. Use load_gs_index instead. */
359
360#endif
361
362#define loadsegment(seg, value) __loadsegment_ ## seg (value)
363
364/*
365 * Save a segment register away:
366 */
367#define savesegment(seg, value) \
368 asm("mov %%" #seg ",%0":"=r" (value) : : "memory")
369
370/*
371 * x86-32 user GS accessors:
372 */
373#ifdef CONFIG_X86_32
374# ifdef CONFIG_X86_32_LAZY_GS
375# define get_user_gs(regs) (u16)({ unsigned long v; savesegment(gs, v); v; })
376# define set_user_gs(regs, v) loadsegment(gs, (unsigned long)(v))
377# define task_user_gs(tsk) ((tsk)->thread.gs)
378# define lazy_save_gs(v) savesegment(gs, (v))
379# define lazy_load_gs(v) loadsegment(gs, (v))
380# else /* X86_32_LAZY_GS */
381# define get_user_gs(regs) (u16)((regs)->gs)
382# define set_user_gs(regs, v) do { (regs)->gs = (v); } while (0)
383# define task_user_gs(tsk) (task_pt_regs(tsk)->gs)
384# define lazy_save_gs(v) do { } while (0)
385# define lazy_load_gs(v) do { } while (0)
386# endif /* X86_32_LAZY_GS */
387#endif /* X86_32 */
388
389#endif /* !__ASSEMBLY__ */
390#endif /* __KERNEL__ */
391
392#endif /* _ASM_X86_SEGMENT_H */