xref: /openbmc/linux/arch/x86/include/asm/desc.h (revision 0c380187)
1 #ifndef _ASM_X86_DESC_H
2 #define _ASM_X86_DESC_H
3 
4 #include <asm/desc_defs.h>
5 #include <asm/ldt.h>
6 #include <asm/mmu.h>
7 
8 #include <linux/smp.h>
9 #include <linux/percpu.h>
10 
11 static inline void fill_ldt(struct desc_struct *desc, const struct user_desc *info)
12 {
13 	desc->limit0		= info->limit & 0x0ffff;
14 
15 	desc->base0		= (info->base_addr & 0x0000ffff);
16 	desc->base1		= (info->base_addr & 0x00ff0000) >> 16;
17 
18 	desc->type		= (info->read_exec_only ^ 1) << 1;
19 	desc->type	       |= info->contents << 2;
20 
21 	desc->s			= 1;
22 	desc->dpl		= 0x3;
23 	desc->p			= info->seg_not_present ^ 1;
24 	desc->limit		= (info->limit & 0xf0000) >> 16;
25 	desc->avl		= info->useable;
26 	desc->d			= info->seg_32bit;
27 	desc->g			= info->limit_in_pages;
28 
29 	desc->base2		= (info->base_addr & 0xff000000) >> 24;
30 	/*
31 	 * Don't allow setting of the lm bit. It would confuse
32 	 * user_64bit_mode and would get overridden by sysret anyway.
33 	 */
34 	desc->l			= 0;
35 }
36 
37 extern struct desc_ptr idt_descr;
38 extern gate_desc idt_table[];
39 extern const struct desc_ptr debug_idt_descr;
40 extern gate_desc debug_idt_table[];
41 
42 struct gdt_page {
43 	struct desc_struct gdt[GDT_ENTRIES];
44 } __attribute__((aligned(PAGE_SIZE)));
45 
46 DECLARE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page);
47 
48 static inline struct desc_struct *get_cpu_gdt_table(unsigned int cpu)
49 {
50 	return per_cpu(gdt_page, cpu).gdt;
51 }
52 
53 #ifdef CONFIG_X86_64
54 
55 static inline void pack_gate(gate_desc *gate, unsigned type, unsigned long func,
56 			     unsigned dpl, unsigned ist, unsigned seg)
57 {
58 	gate->offset_low	= PTR_LOW(func);
59 	gate->segment		= __KERNEL_CS;
60 	gate->ist		= ist;
61 	gate->p			= 1;
62 	gate->dpl		= dpl;
63 	gate->zero0		= 0;
64 	gate->zero1		= 0;
65 	gate->type		= type;
66 	gate->offset_middle	= PTR_MIDDLE(func);
67 	gate->offset_high	= PTR_HIGH(func);
68 }
69 
70 #else
71 static inline void pack_gate(gate_desc *gate, unsigned char type,
72 			     unsigned long base, unsigned dpl, unsigned flags,
73 			     unsigned short seg)
74 {
75 	gate->a = (seg << 16) | (base & 0xffff);
76 	gate->b = (base & 0xffff0000) | (((0x80 | type | (dpl << 5)) & 0xff) << 8);
77 }
78 
79 #endif
80 
81 static inline int desc_empty(const void *ptr)
82 {
83 	const u32 *desc = ptr;
84 
85 	return !(desc[0] | desc[1]);
86 }
87 
88 #ifdef CONFIG_PARAVIRT
89 #include <asm/paravirt.h>
90 #else
91 #define load_TR_desc()				native_load_tr_desc()
92 #define load_gdt(dtr)				native_load_gdt(dtr)
93 #define load_idt(dtr)				native_load_idt(dtr)
94 #define load_tr(tr)				asm volatile("ltr %0"::"m" (tr))
95 #define load_ldt(ldt)				asm volatile("lldt %0"::"m" (ldt))
96 
97 #define store_gdt(dtr)				native_store_gdt(dtr)
98 #define store_idt(dtr)				native_store_idt(dtr)
99 #define store_tr(tr)				(tr = native_store_tr())
100 
101 #define load_TLS(t, cpu)			native_load_tls(t, cpu)
102 #define set_ldt					native_set_ldt
103 
104 #define write_ldt_entry(dt, entry, desc)	native_write_ldt_entry(dt, entry, desc)
105 #define write_gdt_entry(dt, entry, desc, type)	native_write_gdt_entry(dt, entry, desc, type)
106 #define write_idt_entry(dt, entry, g)		native_write_idt_entry(dt, entry, g)
107 
108 static inline void paravirt_alloc_ldt(struct desc_struct *ldt, unsigned entries)
109 {
110 }
111 
112 static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries)
113 {
114 }
115 #endif	/* CONFIG_PARAVIRT */
116 
117 #define store_ldt(ldt) asm("sldt %0" : "=m"(ldt))
118 
119 static inline void native_write_idt_entry(gate_desc *idt, int entry, const gate_desc *gate)
120 {
121 	memcpy(&idt[entry], gate, sizeof(*gate));
122 }
123 
124 static inline void native_write_ldt_entry(struct desc_struct *ldt, int entry, const void *desc)
125 {
126 	memcpy(&ldt[entry], desc, 8);
127 }
128 
129 static inline void
130 native_write_gdt_entry(struct desc_struct *gdt, int entry, const void *desc, int type)
131 {
132 	unsigned int size;
133 
134 	switch (type) {
135 	case DESC_TSS:	size = sizeof(tss_desc);	break;
136 	case DESC_LDT:	size = sizeof(ldt_desc);	break;
137 	default:	size = sizeof(*gdt);		break;
138 	}
139 
140 	memcpy(&gdt[entry], desc, size);
141 }
142 
143 static inline void pack_descriptor(struct desc_struct *desc, unsigned long base,
144 				   unsigned long limit, unsigned char type,
145 				   unsigned char flags)
146 {
147 	desc->a = ((base & 0xffff) << 16) | (limit & 0xffff);
148 	desc->b = (base & 0xff000000) | ((base & 0xff0000) >> 16) |
149 		(limit & 0x000f0000) | ((type & 0xff) << 8) |
150 		((flags & 0xf) << 20);
151 	desc->p = 1;
152 }
153 
154 
155 static inline void set_tssldt_descriptor(void *d, unsigned long addr, unsigned type, unsigned size)
156 {
157 #ifdef CONFIG_X86_64
158 	struct ldttss_desc64 *desc = d;
159 
160 	memset(desc, 0, sizeof(*desc));
161 
162 	desc->limit0		= size & 0xFFFF;
163 	desc->base0		= PTR_LOW(addr);
164 	desc->base1		= PTR_MIDDLE(addr) & 0xFF;
165 	desc->type		= type;
166 	desc->p			= 1;
167 	desc->limit1		= (size >> 16) & 0xF;
168 	desc->base2		= (PTR_MIDDLE(addr) >> 8) & 0xFF;
169 	desc->base3		= PTR_HIGH(addr);
170 #else
171 	pack_descriptor((struct desc_struct *)d, addr, size, 0x80 | type, 0);
172 #endif
173 }
174 
175 static inline void __set_tss_desc(unsigned cpu, unsigned int entry, void *addr)
176 {
177 	struct desc_struct *d = get_cpu_gdt_table(cpu);
178 	tss_desc tss;
179 
180 	set_tssldt_descriptor(&tss, (unsigned long)addr, DESC_TSS,
181 			      __KERNEL_TSS_LIMIT);
182 	write_gdt_entry(d, entry, &tss, DESC_TSS);
183 }
184 
185 #define set_tss_desc(cpu, addr) __set_tss_desc(cpu, GDT_ENTRY_TSS, addr)
186 
187 static inline void native_set_ldt(const void *addr, unsigned int entries)
188 {
189 	if (likely(entries == 0))
190 		asm volatile("lldt %w0"::"q" (0));
191 	else {
192 		unsigned cpu = smp_processor_id();
193 		ldt_desc ldt;
194 
195 		set_tssldt_descriptor(&ldt, (unsigned long)addr, DESC_LDT,
196 				      entries * LDT_ENTRY_SIZE - 1);
197 		write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_LDT,
198 				&ldt, DESC_LDT);
199 		asm volatile("lldt %w0"::"q" (GDT_ENTRY_LDT*8));
200 	}
201 }
202 
203 static inline void native_load_tr_desc(void)
204 {
205 	asm volatile("ltr %w0"::"q" (GDT_ENTRY_TSS*8));
206 }
207 
208 DECLARE_PER_CPU(bool, __tss_limit_invalid);
209 
210 static inline void force_reload_TR(void)
211 {
212 	struct desc_struct *d = get_cpu_gdt_table(smp_processor_id());
213 	tss_desc tss;
214 
215 	memcpy(&tss, &d[GDT_ENTRY_TSS], sizeof(tss_desc));
216 
217 	/*
218 	 * LTR requires an available TSS, and the TSS is currently
219 	 * busy.  Make it be available so that LTR will work.
220 	 */
221 	tss.type = DESC_TSS;
222 	write_gdt_entry(d, GDT_ENTRY_TSS, &tss, DESC_TSS);
223 
224 	load_TR_desc();
225 	this_cpu_write(__tss_limit_invalid, false);
226 }
227 
228 /*
229  * Call this if you need the TSS limit to be correct, which should be the case
230  * if and only if you have TIF_IO_BITMAP set or you're switching to a task
231  * with TIF_IO_BITMAP set.
232  */
233 static inline void refresh_tss_limit(void)
234 {
235 	DEBUG_LOCKS_WARN_ON(preemptible());
236 
237 	if (unlikely(this_cpu_read(__tss_limit_invalid)))
238 		force_reload_TR();
239 }
240 
241 /*
242  * If you do something evil that corrupts the cached TSS limit (I'm looking
243  * at you, VMX exits), call this function.
244  *
245  * The optimization here is that the TSS limit only matters for Linux if the
246  * IO bitmap is in use.  If the TSS limit gets forced to its minimum value,
247  * everything works except that IO bitmap will be ignored and all CPL 3 IO
248  * instructions will #GP, which is exactly what we want for normal tasks.
249  */
250 static inline void invalidate_tss_limit(void)
251 {
252 	DEBUG_LOCKS_WARN_ON(preemptible());
253 
254 	if (unlikely(test_thread_flag(TIF_IO_BITMAP)))
255 		force_reload_TR();
256 	else
257 		this_cpu_write(__tss_limit_invalid, true);
258 }
259 
260 static inline void native_load_gdt(const struct desc_ptr *dtr)
261 {
262 	asm volatile("lgdt %0"::"m" (*dtr));
263 }
264 
265 static inline void native_load_idt(const struct desc_ptr *dtr)
266 {
267 	asm volatile("lidt %0"::"m" (*dtr));
268 }
269 
270 static inline void native_store_gdt(struct desc_ptr *dtr)
271 {
272 	asm volatile("sgdt %0":"=m" (*dtr));
273 }
274 
275 static inline void native_store_idt(struct desc_ptr *dtr)
276 {
277 	asm volatile("sidt %0":"=m" (*dtr));
278 }
279 
280 static inline unsigned long native_store_tr(void)
281 {
282 	unsigned long tr;
283 
284 	asm volatile("str %0":"=r" (tr));
285 
286 	return tr;
287 }
288 
289 static inline void native_load_tls(struct thread_struct *t, unsigned int cpu)
290 {
291 	struct desc_struct *gdt = get_cpu_gdt_table(cpu);
292 	unsigned int i;
293 
294 	for (i = 0; i < GDT_ENTRY_TLS_ENTRIES; i++)
295 		gdt[GDT_ENTRY_TLS_MIN + i] = t->tls_array[i];
296 }
297 
298 /* This intentionally ignores lm, since 32-bit apps don't have that field. */
299 #define LDT_empty(info)					\
300 	((info)->base_addr		== 0	&&	\
301 	 (info)->limit			== 0	&&	\
302 	 (info)->contents		== 0	&&	\
303 	 (info)->read_exec_only		== 1	&&	\
304 	 (info)->seg_32bit		== 0	&&	\
305 	 (info)->limit_in_pages		== 0	&&	\
306 	 (info)->seg_not_present	== 1	&&	\
307 	 (info)->useable		== 0)
308 
309 /* Lots of programs expect an all-zero user_desc to mean "no segment at all". */
310 static inline bool LDT_zero(const struct user_desc *info)
311 {
312 	return (info->base_addr		== 0 &&
313 		info->limit		== 0 &&
314 		info->contents		== 0 &&
315 		info->read_exec_only	== 0 &&
316 		info->seg_32bit		== 0 &&
317 		info->limit_in_pages	== 0 &&
318 		info->seg_not_present	== 0 &&
319 		info->useable		== 0);
320 }
321 
322 static inline void clear_LDT(void)
323 {
324 	set_ldt(NULL, 0);
325 }
326 
327 static inline unsigned long get_desc_base(const struct desc_struct *desc)
328 {
329 	return (unsigned)(desc->base0 | ((desc->base1) << 16) | ((desc->base2) << 24));
330 }
331 
332 static inline void set_desc_base(struct desc_struct *desc, unsigned long base)
333 {
334 	desc->base0 = base & 0xffff;
335 	desc->base1 = (base >> 16) & 0xff;
336 	desc->base2 = (base >> 24) & 0xff;
337 }
338 
339 static inline unsigned long get_desc_limit(const struct desc_struct *desc)
340 {
341 	return desc->limit0 | (desc->limit << 16);
342 }
343 
344 static inline void set_desc_limit(struct desc_struct *desc, unsigned long limit)
345 {
346 	desc->limit0 = limit & 0xffff;
347 	desc->limit = (limit >> 16) & 0xf;
348 }
349 
350 #ifdef CONFIG_X86_64
351 static inline void set_nmi_gate(int gate, void *addr)
352 {
353 	gate_desc s;
354 
355 	pack_gate(&s, GATE_INTERRUPT, (unsigned long)addr, 0, 0, __KERNEL_CS);
356 	write_idt_entry(debug_idt_table, gate, &s);
357 }
358 #endif
359 
360 #ifdef CONFIG_TRACING
361 extern struct desc_ptr trace_idt_descr;
362 extern gate_desc trace_idt_table[];
363 static inline void write_trace_idt_entry(int entry, const gate_desc *gate)
364 {
365 	write_idt_entry(trace_idt_table, entry, gate);
366 }
367 
368 static inline void _trace_set_gate(int gate, unsigned type, void *addr,
369 				   unsigned dpl, unsigned ist, unsigned seg)
370 {
371 	gate_desc s;
372 
373 	pack_gate(&s, type, (unsigned long)addr, dpl, ist, seg);
374 	/*
375 	 * does not need to be atomic because it is only done once at
376 	 * setup time
377 	 */
378 	write_trace_idt_entry(gate, &s);
379 }
380 #else
381 static inline void write_trace_idt_entry(int entry, const gate_desc *gate)
382 {
383 }
384 
385 #define _trace_set_gate(gate, type, addr, dpl, ist, seg)
386 #endif
387 
388 static inline void _set_gate(int gate, unsigned type, void *addr,
389 			     unsigned dpl, unsigned ist, unsigned seg)
390 {
391 	gate_desc s;
392 
393 	pack_gate(&s, type, (unsigned long)addr, dpl, ist, seg);
394 	/*
395 	 * does not need to be atomic because it is only done once at
396 	 * setup time
397 	 */
398 	write_idt_entry(idt_table, gate, &s);
399 	write_trace_idt_entry(gate, &s);
400 }
401 
402 /*
403  * This needs to use 'idt_table' rather than 'idt', and
404  * thus use the _nonmapped_ version of the IDT, as the
405  * Pentium F0 0F bugfix can have resulted in the mapped
406  * IDT being write-protected.
407  */
408 #define set_intr_gate_notrace(n, addr)					\
409 	do {								\
410 		BUG_ON((unsigned)n > 0xFF);				\
411 		_set_gate(n, GATE_INTERRUPT, (void *)addr, 0, 0,	\
412 			  __KERNEL_CS);					\
413 	} while (0)
414 
415 #define set_intr_gate(n, addr)						\
416 	do {								\
417 		set_intr_gate_notrace(n, addr);				\
418 		_trace_set_gate(n, GATE_INTERRUPT, (void *)trace_##addr,\
419 				0, 0, __KERNEL_CS);			\
420 	} while (0)
421 
422 extern int first_system_vector;
423 /* used_vectors is BITMAP for irq is not managed by percpu vector_irq */
424 extern unsigned long used_vectors[];
425 
426 static inline void alloc_system_vector(int vector)
427 {
428 	if (!test_bit(vector, used_vectors)) {
429 		set_bit(vector, used_vectors);
430 		if (first_system_vector > vector)
431 			first_system_vector = vector;
432 	} else {
433 		BUG();
434 	}
435 }
436 
437 #define alloc_intr_gate(n, addr)				\
438 	do {							\
439 		alloc_system_vector(n);				\
440 		set_intr_gate(n, addr);				\
441 	} while (0)
442 
443 /*
444  * This routine sets up an interrupt gate at directory privilege level 3.
445  */
446 static inline void set_system_intr_gate(unsigned int n, void *addr)
447 {
448 	BUG_ON((unsigned)n > 0xFF);
449 	_set_gate(n, GATE_INTERRUPT, addr, 0x3, 0, __KERNEL_CS);
450 }
451 
452 static inline void set_system_trap_gate(unsigned int n, void *addr)
453 {
454 	BUG_ON((unsigned)n > 0xFF);
455 	_set_gate(n, GATE_TRAP, addr, 0x3, 0, __KERNEL_CS);
456 }
457 
458 static inline void set_trap_gate(unsigned int n, void *addr)
459 {
460 	BUG_ON((unsigned)n > 0xFF);
461 	_set_gate(n, GATE_TRAP, addr, 0, 0, __KERNEL_CS);
462 }
463 
464 static inline void set_task_gate(unsigned int n, unsigned int gdt_entry)
465 {
466 	BUG_ON((unsigned)n > 0xFF);
467 	_set_gate(n, GATE_TASK, (void *)0, 0, 0, (gdt_entry<<3));
468 }
469 
470 static inline void set_intr_gate_ist(int n, void *addr, unsigned ist)
471 {
472 	BUG_ON((unsigned)n > 0xFF);
473 	_set_gate(n, GATE_INTERRUPT, addr, 0, ist, __KERNEL_CS);
474 }
475 
476 static inline void set_system_intr_gate_ist(int n, void *addr, unsigned ist)
477 {
478 	BUG_ON((unsigned)n > 0xFF);
479 	_set_gate(n, GATE_INTERRUPT, addr, 0x3, ist, __KERNEL_CS);
480 }
481 
482 #ifdef CONFIG_X86_64
483 DECLARE_PER_CPU(u32, debug_idt_ctr);
484 static inline bool is_debug_idt_enabled(void)
485 {
486 	if (this_cpu_read(debug_idt_ctr))
487 		return true;
488 
489 	return false;
490 }
491 
492 static inline void load_debug_idt(void)
493 {
494 	load_idt((const struct desc_ptr *)&debug_idt_descr);
495 }
496 #else
497 static inline bool is_debug_idt_enabled(void)
498 {
499 	return false;
500 }
501 
502 static inline void load_debug_idt(void)
503 {
504 }
505 #endif
506 
507 #ifdef CONFIG_TRACING
508 extern atomic_t trace_idt_ctr;
509 static inline bool is_trace_idt_enabled(void)
510 {
511 	if (atomic_read(&trace_idt_ctr))
512 		return true;
513 
514 	return false;
515 }
516 
517 static inline void load_trace_idt(void)
518 {
519 	load_idt((const struct desc_ptr *)&trace_idt_descr);
520 }
521 #else
522 static inline bool is_trace_idt_enabled(void)
523 {
524 	return false;
525 }
526 
527 static inline void load_trace_idt(void)
528 {
529 }
530 #endif
531 
532 /*
533  * The load_current_idt() must be called with interrupts disabled
534  * to avoid races. That way the IDT will always be set back to the expected
535  * descriptor. It's also called when a CPU is being initialized, and
536  * that doesn't need to disable interrupts, as nothing should be
537  * bothering the CPU then.
538  */
539 static inline void load_current_idt(void)
540 {
541 	if (is_debug_idt_enabled())
542 		load_debug_idt();
543 	else if (is_trace_idt_enabled())
544 		load_trace_idt();
545 	else
546 		load_idt((const struct desc_ptr *)&idt_descr);
547 }
548 #endif /* _ASM_X86_DESC_H */
549