xref: /openbmc/linux/arch/x86/include/asm/desc.h (revision 9dae47aba0a055f761176d9297371d5bb24289ec)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_DESC_H
3 #define _ASM_X86_DESC_H
4 
5 #include <asm/desc_defs.h>
6 #include <asm/ldt.h>
7 #include <asm/mmu.h>
8 #include <asm/fixmap.h>
9 #include <asm/irq_vectors.h>
10 #include <asm/cpu_entry_area.h>
11 
12 #include <linux/smp.h>
13 #include <linux/percpu.h>
14 
15 static inline void fill_ldt(struct desc_struct *desc, const struct user_desc *info)
16 {
17 	desc->limit0		= info->limit & 0x0ffff;
18 
19 	desc->base0		= (info->base_addr & 0x0000ffff);
20 	desc->base1		= (info->base_addr & 0x00ff0000) >> 16;
21 
22 	desc->type		= (info->read_exec_only ^ 1) << 1;
23 	desc->type	       |= info->contents << 2;
24 
25 	desc->s			= 1;
26 	desc->dpl		= 0x3;
27 	desc->p			= info->seg_not_present ^ 1;
28 	desc->limit1		= (info->limit & 0xf0000) >> 16;
29 	desc->avl		= info->useable;
30 	desc->d			= info->seg_32bit;
31 	desc->g			= info->limit_in_pages;
32 
33 	desc->base2		= (info->base_addr & 0xff000000) >> 24;
34 	/*
35 	 * Don't allow setting of the lm bit. It would confuse
36 	 * user_64bit_mode and would get overridden by sysret anyway.
37 	 */
38 	desc->l			= 0;
39 }
40 
41 extern struct desc_ptr idt_descr;
42 extern gate_desc idt_table[];
43 extern const struct desc_ptr debug_idt_descr;
44 extern gate_desc debug_idt_table[];
45 
46 struct gdt_page {
47 	struct desc_struct gdt[GDT_ENTRIES];
48 } __attribute__((aligned(PAGE_SIZE)));
49 
50 DECLARE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page);
51 
52 /* Provide the original GDT */
53 static inline struct desc_struct *get_cpu_gdt_rw(unsigned int cpu)
54 {
55 	return per_cpu(gdt_page, cpu).gdt;
56 }
57 
58 /* Provide the current original GDT */
59 static inline struct desc_struct *get_current_gdt_rw(void)
60 {
61 	return this_cpu_ptr(&gdt_page)->gdt;
62 }
63 
64 /* Provide the fixmap address of the remapped GDT */
65 static inline struct desc_struct *get_cpu_gdt_ro(int cpu)
66 {
67 	return (struct desc_struct *)&get_cpu_entry_area(cpu)->gdt;
68 }
69 
70 /* Provide the current read-only GDT */
71 static inline struct desc_struct *get_current_gdt_ro(void)
72 {
73 	return get_cpu_gdt_ro(smp_processor_id());
74 }
75 
76 /* Provide the physical address of the GDT page. */
77 static inline phys_addr_t get_cpu_gdt_paddr(unsigned int cpu)
78 {
79 	return per_cpu_ptr_to_phys(get_cpu_gdt_rw(cpu));
80 }
81 
82 static inline void pack_gate(gate_desc *gate, unsigned type, unsigned long func,
83 			     unsigned dpl, unsigned ist, unsigned seg)
84 {
85 	gate->offset_low	= (u16) func;
86 	gate->bits.p		= 1;
87 	gate->bits.dpl		= dpl;
88 	gate->bits.zero		= 0;
89 	gate->bits.type		= type;
90 	gate->offset_middle	= (u16) (func >> 16);
91 #ifdef CONFIG_X86_64
92 	gate->segment		= __KERNEL_CS;
93 	gate->bits.ist		= ist;
94 	gate->reserved		= 0;
95 	gate->offset_high	= (u32) (func >> 32);
96 #else
97 	gate->segment		= seg;
98 	gate->bits.ist		= 0;
99 #endif
100 }
101 
102 static inline int desc_empty(const void *ptr)
103 {
104 	const u32 *desc = ptr;
105 
106 	return !(desc[0] | desc[1]);
107 }
108 
109 #ifdef CONFIG_PARAVIRT
110 #include <asm/paravirt.h>
111 #else
112 #define load_TR_desc()				native_load_tr_desc()
113 #define load_gdt(dtr)				native_load_gdt(dtr)
114 #define load_idt(dtr)				native_load_idt(dtr)
115 #define load_tr(tr)				asm volatile("ltr %0"::"m" (tr))
116 #define load_ldt(ldt)				asm volatile("lldt %0"::"m" (ldt))
117 
118 #define store_gdt(dtr)				native_store_gdt(dtr)
119 #define store_tr(tr)				(tr = native_store_tr())
120 
121 #define load_TLS(t, cpu)			native_load_tls(t, cpu)
122 #define set_ldt					native_set_ldt
123 
124 #define write_ldt_entry(dt, entry, desc)	native_write_ldt_entry(dt, entry, desc)
125 #define write_gdt_entry(dt, entry, desc, type)	native_write_gdt_entry(dt, entry, desc, type)
126 #define write_idt_entry(dt, entry, g)		native_write_idt_entry(dt, entry, g)
127 
128 static inline void paravirt_alloc_ldt(struct desc_struct *ldt, unsigned entries)
129 {
130 }
131 
132 static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries)
133 {
134 }
135 #endif	/* CONFIG_PARAVIRT */
136 
137 #define store_ldt(ldt) asm("sldt %0" : "=m"(ldt))
138 
139 static inline void native_write_idt_entry(gate_desc *idt, int entry, const gate_desc *gate)
140 {
141 	memcpy(&idt[entry], gate, sizeof(*gate));
142 }
143 
144 static inline void native_write_ldt_entry(struct desc_struct *ldt, int entry, const void *desc)
145 {
146 	memcpy(&ldt[entry], desc, 8);
147 }
148 
149 static inline void
150 native_write_gdt_entry(struct desc_struct *gdt, int entry, const void *desc, int type)
151 {
152 	unsigned int size;
153 
154 	switch (type) {
155 	case DESC_TSS:	size = sizeof(tss_desc);	break;
156 	case DESC_LDT:	size = sizeof(ldt_desc);	break;
157 	default:	size = sizeof(*gdt);		break;
158 	}
159 
160 	memcpy(&gdt[entry], desc, size);
161 }
162 
163 static inline void set_tssldt_descriptor(void *d, unsigned long addr,
164 					 unsigned type, unsigned size)
165 {
166 	struct ldttss_desc *desc = d;
167 
168 	memset(desc, 0, sizeof(*desc));
169 
170 	desc->limit0		= (u16) size;
171 	desc->base0		= (u16) addr;
172 	desc->base1		= (addr >> 16) & 0xFF;
173 	desc->type		= type;
174 	desc->p			= 1;
175 	desc->limit1		= (size >> 16) & 0xF;
176 	desc->base2		= (addr >> 24) & 0xFF;
177 #ifdef CONFIG_X86_64
178 	desc->base3		= (u32) (addr >> 32);
179 #endif
180 }
181 
182 static inline void __set_tss_desc(unsigned cpu, unsigned int entry, struct x86_hw_tss *addr)
183 {
184 	struct desc_struct *d = get_cpu_gdt_rw(cpu);
185 	tss_desc tss;
186 
187 	set_tssldt_descriptor(&tss, (unsigned long)addr, DESC_TSS,
188 			      __KERNEL_TSS_LIMIT);
189 	write_gdt_entry(d, entry, &tss, DESC_TSS);
190 }
191 
192 #define set_tss_desc(cpu, addr) __set_tss_desc(cpu, GDT_ENTRY_TSS, addr)
193 
194 static inline void native_set_ldt(const void *addr, unsigned int entries)
195 {
196 	if (likely(entries == 0))
197 		asm volatile("lldt %w0"::"q" (0));
198 	else {
199 		unsigned cpu = smp_processor_id();
200 		ldt_desc ldt;
201 
202 		set_tssldt_descriptor(&ldt, (unsigned long)addr, DESC_LDT,
203 				      entries * LDT_ENTRY_SIZE - 1);
204 		write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_LDT,
205 				&ldt, DESC_LDT);
206 		asm volatile("lldt %w0"::"q" (GDT_ENTRY_LDT*8));
207 	}
208 }
209 
210 static inline void native_load_gdt(const struct desc_ptr *dtr)
211 {
212 	asm volatile("lgdt %0"::"m" (*dtr));
213 }
214 
215 static inline void native_load_idt(const struct desc_ptr *dtr)
216 {
217 	asm volatile("lidt %0"::"m" (*dtr));
218 }
219 
220 static inline void native_store_gdt(struct desc_ptr *dtr)
221 {
222 	asm volatile("sgdt %0":"=m" (*dtr));
223 }
224 
225 static inline void store_idt(struct desc_ptr *dtr)
226 {
227 	asm volatile("sidt %0":"=m" (*dtr));
228 }
229 
230 /*
231  * The LTR instruction marks the TSS GDT entry as busy. On 64-bit, the GDT is
232  * a read-only remapping. To prevent a page fault, the GDT is switched to the
233  * original writeable version when needed.
234  */
235 #ifdef CONFIG_X86_64
236 static inline void native_load_tr_desc(void)
237 {
238 	struct desc_ptr gdt;
239 	int cpu = raw_smp_processor_id();
240 	bool restore = 0;
241 	struct desc_struct *fixmap_gdt;
242 
243 	native_store_gdt(&gdt);
244 	fixmap_gdt = get_cpu_gdt_ro(cpu);
245 
246 	/*
247 	 * If the current GDT is the read-only fixmap, swap to the original
248 	 * writeable version. Swap back at the end.
249 	 */
250 	if (gdt.address == (unsigned long)fixmap_gdt) {
251 		load_direct_gdt(cpu);
252 		restore = 1;
253 	}
254 	asm volatile("ltr %w0"::"q" (GDT_ENTRY_TSS*8));
255 	if (restore)
256 		load_fixmap_gdt(cpu);
257 }
258 #else
259 static inline void native_load_tr_desc(void)
260 {
261 	asm volatile("ltr %w0"::"q" (GDT_ENTRY_TSS*8));
262 }
263 #endif
264 
265 static inline unsigned long native_store_tr(void)
266 {
267 	unsigned long tr;
268 
269 	asm volatile("str %0":"=r" (tr));
270 
271 	return tr;
272 }
273 
274 static inline void native_load_tls(struct thread_struct *t, unsigned int cpu)
275 {
276 	struct desc_struct *gdt = get_cpu_gdt_rw(cpu);
277 	unsigned int i;
278 
279 	for (i = 0; i < GDT_ENTRY_TLS_ENTRIES; i++)
280 		gdt[GDT_ENTRY_TLS_MIN + i] = t->tls_array[i];
281 }
282 
283 DECLARE_PER_CPU(bool, __tss_limit_invalid);
284 
285 static inline void force_reload_TR(void)
286 {
287 	struct desc_struct *d = get_current_gdt_rw();
288 	tss_desc tss;
289 
290 	memcpy(&tss, &d[GDT_ENTRY_TSS], sizeof(tss_desc));
291 
292 	/*
293 	 * LTR requires an available TSS, and the TSS is currently
294 	 * busy.  Make it be available so that LTR will work.
295 	 */
296 	tss.type = DESC_TSS;
297 	write_gdt_entry(d, GDT_ENTRY_TSS, &tss, DESC_TSS);
298 
299 	load_TR_desc();
300 	this_cpu_write(__tss_limit_invalid, false);
301 }
302 
303 /*
304  * Call this if you need the TSS limit to be correct, which should be the case
305  * if and only if you have TIF_IO_BITMAP set or you're switching to a task
306  * with TIF_IO_BITMAP set.
307  */
308 static inline void refresh_tss_limit(void)
309 {
310 	DEBUG_LOCKS_WARN_ON(preemptible());
311 
312 	if (unlikely(this_cpu_read(__tss_limit_invalid)))
313 		force_reload_TR();
314 }
315 
316 /*
317  * If you do something evil that corrupts the cached TSS limit (I'm looking
318  * at you, VMX exits), call this function.
319  *
320  * The optimization here is that the TSS limit only matters for Linux if the
321  * IO bitmap is in use.  If the TSS limit gets forced to its minimum value,
322  * everything works except that IO bitmap will be ignored and all CPL 3 IO
323  * instructions will #GP, which is exactly what we want for normal tasks.
324  */
325 static inline void invalidate_tss_limit(void)
326 {
327 	DEBUG_LOCKS_WARN_ON(preemptible());
328 
329 	if (unlikely(test_thread_flag(TIF_IO_BITMAP)))
330 		force_reload_TR();
331 	else
332 		this_cpu_write(__tss_limit_invalid, true);
333 }
334 
335 /* This intentionally ignores lm, since 32-bit apps don't have that field. */
336 #define LDT_empty(info)					\
337 	((info)->base_addr		== 0	&&	\
338 	 (info)->limit			== 0	&&	\
339 	 (info)->contents		== 0	&&	\
340 	 (info)->read_exec_only		== 1	&&	\
341 	 (info)->seg_32bit		== 0	&&	\
342 	 (info)->limit_in_pages		== 0	&&	\
343 	 (info)->seg_not_present	== 1	&&	\
344 	 (info)->useable		== 0)
345 
346 /* Lots of programs expect an all-zero user_desc to mean "no segment at all". */
347 static inline bool LDT_zero(const struct user_desc *info)
348 {
349 	return (info->base_addr		== 0 &&
350 		info->limit		== 0 &&
351 		info->contents		== 0 &&
352 		info->read_exec_only	== 0 &&
353 		info->seg_32bit		== 0 &&
354 		info->limit_in_pages	== 0 &&
355 		info->seg_not_present	== 0 &&
356 		info->useable		== 0);
357 }
358 
359 static inline void clear_LDT(void)
360 {
361 	set_ldt(NULL, 0);
362 }
363 
364 static inline unsigned long get_desc_base(const struct desc_struct *desc)
365 {
366 	return (unsigned)(desc->base0 | ((desc->base1) << 16) | ((desc->base2) << 24));
367 }
368 
369 static inline void set_desc_base(struct desc_struct *desc, unsigned long base)
370 {
371 	desc->base0 = base & 0xffff;
372 	desc->base1 = (base >> 16) & 0xff;
373 	desc->base2 = (base >> 24) & 0xff;
374 }
375 
376 static inline unsigned long get_desc_limit(const struct desc_struct *desc)
377 {
378 	return desc->limit0 | (desc->limit1 << 16);
379 }
380 
381 static inline void set_desc_limit(struct desc_struct *desc, unsigned long limit)
382 {
383 	desc->limit0 = limit & 0xffff;
384 	desc->limit1 = (limit >> 16) & 0xf;
385 }
386 
387 void update_intr_gate(unsigned int n, const void *addr);
388 void alloc_intr_gate(unsigned int n, const void *addr);
389 
390 extern unsigned long system_vectors[];
391 
392 #ifdef CONFIG_X86_64
393 DECLARE_PER_CPU(u32, debug_idt_ctr);
394 static inline bool is_debug_idt_enabled(void)
395 {
396 	if (this_cpu_read(debug_idt_ctr))
397 		return true;
398 
399 	return false;
400 }
401 
402 static inline void load_debug_idt(void)
403 {
404 	load_idt((const struct desc_ptr *)&debug_idt_descr);
405 }
406 #else
407 static inline bool is_debug_idt_enabled(void)
408 {
409 	return false;
410 }
411 
412 static inline void load_debug_idt(void)
413 {
414 }
415 #endif
416 
417 /*
418  * The load_current_idt() must be called with interrupts disabled
419  * to avoid races. That way the IDT will always be set back to the expected
420  * descriptor. It's also called when a CPU is being initialized, and
421  * that doesn't need to disable interrupts, as nothing should be
422  * bothering the CPU then.
423  */
424 static inline void load_current_idt(void)
425 {
426 	if (is_debug_idt_enabled())
427 		load_debug_idt();
428 	else
429 		load_idt((const struct desc_ptr *)&idt_descr);
430 }
431 
432 extern void idt_setup_early_handler(void);
433 extern void idt_setup_early_traps(void);
434 extern void idt_setup_traps(void);
435 extern void idt_setup_apic_and_irq_gates(void);
436 
437 #ifdef CONFIG_X86_64
438 extern void idt_setup_early_pf(void);
439 extern void idt_setup_ist_traps(void);
440 extern void idt_setup_debugidt_traps(void);
441 #else
442 static inline void idt_setup_early_pf(void) { }
443 static inline void idt_setup_ist_traps(void) { }
444 static inline void idt_setup_debugidt_traps(void) { }
445 #endif
446 
447 extern void idt_invalidate(void *addr);
448 
449 #endif /* _ASM_X86_DESC_H */
450