paravirt_types.h

  1/* SPDX-License-Identifier: GPL-2.0 */
  2#ifndef _ASM_X86_PARAVIRT_TYPES_H
  3#define _ASM_X86_PARAVIRT_TYPES_H
  4
  5#ifdef CONFIG_PARAVIRT
  6
  7#ifndef __ASSEMBLER__
  8#include <linux/types.h>
  9
 10#include <asm/desc_defs.h>
 11#include <asm/pgtable_types.h>
 12#include <asm/nospec-branch.h>
 13
 14struct page;
 15struct thread_struct;
 16struct desc_ptr;
 17struct tss_struct;
 18struct mm_struct;
 19struct desc_struct;
 20struct task_struct;
 21struct cpumask;
 22struct flush_tlb_info;
 23struct mmu_gather;
 24struct vm_area_struct;
 25
 26/*
 27 * Wrapper type for pointers to code which uses the non-standard
 28 * calling convention.  See PV_CALL_SAVE_REGS_THUNK below.
 29 */
 30struct paravirt_callee_save {
 31	void *func;
 32};
 33
 34/* general info */
 35struct pv_info {
 36#ifdef CONFIG_PARAVIRT_XXL
 37	u16 extra_user_64bit_cs;  /* __USER_CS if none */
 38#endif
 39
 40	const char *name;
 41};
 42
 43#ifdef CONFIG_PARAVIRT_XXL
 44struct pv_lazy_ops {
 45	/* Set deferred update mode, used for batching operations. */
 46	void (*enter)(void);
 47	void (*leave)(void);
 48	void (*flush)(void);
 49} __no_randomize_layout;
 50#endif
 51
 52struct pv_cpu_ops {
 53	/* hooks for various privileged instructions */
 54	void (*io_delay)(void);
 55
 56#ifdef CONFIG_PARAVIRT_XXL
 57	unsigned long (*get_debugreg)(int regno);
 58	void (*set_debugreg)(int regno, unsigned long value);
 59
 60	unsigned long (*read_cr0)(void);
 61	void (*write_cr0)(unsigned long);
 62
 63	void (*write_cr4)(unsigned long);
 64
 65	/* Segment descriptor handling */
 66	void (*load_tr_desc)(void);
 67	void (*load_gdt)(const struct desc_ptr *);
 68	void (*load_idt)(const struct desc_ptr *);
 69	void (*set_ldt)(const void *desc, unsigned entries);
 70	unsigned long (*store_tr)(void);
 71	void (*load_tls)(struct thread_struct *t, unsigned int cpu);
 72	void (*load_gs_index)(unsigned int idx);
 73	void (*write_ldt_entry)(struct desc_struct *ldt, int entrynum,
 74				const void *desc);
 75	void (*write_gdt_entry)(struct desc_struct *,
 76				int entrynum, const void *desc, int size);
 77	void (*write_idt_entry)(gate_desc *,
 78				int entrynum, const gate_desc *gate);
 79	void (*alloc_ldt)(struct desc_struct *ldt, unsigned entries);
 80	void (*free_ldt)(struct desc_struct *ldt, unsigned entries);
 81
 82	void (*load_sp0)(unsigned long sp0);
 83
 84#ifdef CONFIG_X86_IOPL_IOPERM
 85	void (*invalidate_io_bitmap)(void);
 86	void (*update_io_bitmap)(void);
 87#endif
 88
 89	/* cpuid emulation, mostly so that caps bits can be disabled */
 90	void (*cpuid)(unsigned int *eax, unsigned int *ebx,
 91		      unsigned int *ecx, unsigned int *edx);
 92
 93	/* Unsafe MSR operations.  These will warn or panic on failure. */
 94	u64 (*read_msr)(u32 msr);
 95	void (*write_msr)(u32 msr, u64 val);
 96
 97	/*
 98	 * Safe MSR operations.
 99	 * Returns 0 or -EIO.
100	 */
101	int (*read_msr_safe)(u32 msr, u64 *val);
102	int (*write_msr_safe)(u32 msr, u64 val);
103
104	u64 (*read_pmc)(int counter);
105
106	void (*start_context_switch)(struct task_struct *prev);
107	void (*end_context_switch)(struct task_struct *next);
108#endif
109} __no_randomize_layout;
110
111struct pv_irq_ops {
112#ifdef CONFIG_PARAVIRT_XXL
113	/*
114	 * Get/set interrupt state.  save_fl is expected to use X86_EFLAGS_IF;
115	 * all other bits returned from save_fl are undefined.
116	 *
117	 * NOTE: These functions callers expect the callee to preserve
118	 * more registers than the standard C calling convention.
119	 */
120	struct paravirt_callee_save save_fl;
121	struct paravirt_callee_save irq_disable;
122	struct paravirt_callee_save irq_enable;
123#endif
124	void (*safe_halt)(void);
125	void (*halt)(void);
126} __no_randomize_layout;
127
128struct pv_mmu_ops {
129	/* TLB operations */
130	void (*flush_tlb_user)(void);
131	void (*flush_tlb_kernel)(void);
132	void (*flush_tlb_one_user)(unsigned long addr);
133	void (*flush_tlb_multi)(const struct cpumask *cpus,
134				const struct flush_tlb_info *info);
135
136	/* Hook for intercepting the destruction of an mm_struct. */
137	void (*exit_mmap)(struct mm_struct *mm);
138	void (*notify_page_enc_status_changed)(unsigned long pfn, int npages, bool enc);
139
140#ifdef CONFIG_PARAVIRT_XXL
141	struct paravirt_callee_save read_cr2;
142	void (*write_cr2)(unsigned long);
143
144	unsigned long (*read_cr3)(void);
145	void (*write_cr3)(unsigned long);
146
147	/* Hook for intercepting the creation/use of an mm_struct. */
148	void (*enter_mmap)(struct mm_struct *mm);
149
150	/* Hooks for allocating and freeing a pagetable top-level */
151	int  (*pgd_alloc)(struct mm_struct *mm);
152	void (*pgd_free)(struct mm_struct *mm, pgd_t *pgd);
153
154	/*
155	 * Hooks for allocating/releasing pagetable pages when they're
156	 * attached to a pagetable
157	 */
158	void (*alloc_pte)(struct mm_struct *mm, unsigned long pfn);
159	void (*alloc_pmd)(struct mm_struct *mm, unsigned long pfn);
160	void (*alloc_pud)(struct mm_struct *mm, unsigned long pfn);
161	void (*alloc_p4d)(struct mm_struct *mm, unsigned long pfn);
162	void (*release_pte)(unsigned long pfn);
163	void (*release_pmd)(unsigned long pfn);
164	void (*release_pud)(unsigned long pfn);
165	void (*release_p4d)(unsigned long pfn);
166
167	/* Pagetable manipulation functions */
168	void (*set_pte)(pte_t *ptep, pte_t pteval);
169	void (*set_pmd)(pmd_t *pmdp, pmd_t pmdval);
170
171	pte_t (*ptep_modify_prot_start)(struct vm_area_struct *vma, unsigned long addr,
172					pte_t *ptep);
173	void (*ptep_modify_prot_commit)(struct vm_area_struct *vma, unsigned long addr,
174					pte_t *ptep, pte_t pte);
175
176	struct paravirt_callee_save pte_val;
177	struct paravirt_callee_save make_pte;
178
179	struct paravirt_callee_save pgd_val;
180	struct paravirt_callee_save make_pgd;
181
182	void (*set_pud)(pud_t *pudp, pud_t pudval);
183
184	struct paravirt_callee_save pmd_val;
185	struct paravirt_callee_save make_pmd;
186
187	struct paravirt_callee_save pud_val;
188	struct paravirt_callee_save make_pud;
189
190	void (*set_p4d)(p4d_t *p4dp, p4d_t p4dval);
191
192	struct paravirt_callee_save p4d_val;
193	struct paravirt_callee_save make_p4d;
194
195	void (*set_pgd)(pgd_t *pgdp, pgd_t pgdval);
196
197	struct pv_lazy_ops lazy_mode;
198
199	/* dom0 ops */
200
201	/* Sometimes the physical address is a pfn, and sometimes its
202	   an mfn.  We can tell which is which from the index. */
203	void (*set_fixmap)(unsigned /* enum fixed_addresses */ idx,
204			   phys_addr_t phys, pgprot_t flags);
205#endif
206} __no_randomize_layout;
207
208struct arch_spinlock;
209#ifdef CONFIG_SMP
210#include <asm/spinlock_types.h>
211#endif
212
213struct qspinlock;
214
215struct pv_lock_ops {
216	void (*queued_spin_lock_slowpath)(struct qspinlock *lock, u32 val);
217	struct paravirt_callee_save queued_spin_unlock;
218
219	void (*wait)(u8 *ptr, u8 val);
220	void (*kick)(int cpu);
221
222	struct paravirt_callee_save vcpu_is_preempted;
223} __no_randomize_layout;
224
225/* This contains all the paravirt structures: we get a convenient
226 * number for each function using the offset which we use to indicate
227 * what to patch. */
228struct paravirt_patch_template {
229	struct pv_cpu_ops	cpu;
230	struct pv_irq_ops	irq;
231	struct pv_mmu_ops	mmu;
232	struct pv_lock_ops	lock;
233} __no_randomize_layout;
234
235extern struct pv_info pv_info;
236extern struct paravirt_patch_template pv_ops;
237
238#define paravirt_ptr(op)	[paravirt_opptr] "m" (pv_ops.op)
239
240/*
241 * This generates an indirect call based on the operation type number.
242 *
243 * Since alternatives run after enabling CET/IBT -- the latter setting/clearing
244 * capabilities and the former requiring all capabilities being finalized --
245 * these indirect calls are subject to IBT and the paravirt stubs should have
246 * ENDBR on.
247 *
248 * OTOH since this is effectively a __nocfi indirect call, the paravirt stubs
249 * don't need to bother with CFI prefixes.
250 */
251#define PARAVIRT_CALL					\
252	ANNOTATE_RETPOLINE_SAFE "\n\t"			\
253	"call *%[paravirt_opptr];"
254
255/*
256 * These macros are intended to wrap calls through one of the paravirt
257 * ops structs, so that they can be later identified and patched at
258 * runtime.
259 *
260 * Normally, a call to a pv_op function is a simple indirect call:
261 * (pv_op_struct.operations)(args...).
262 *
263 * Unfortunately, this is a relatively slow operation for modern CPUs,
264 * because it cannot necessarily determine what the destination
265 * address is.  In this case, the address is a runtime constant, so at
266 * the very least we can patch the call to a simple direct call, or,
267 * ideally, patch an inline implementation into the callsite.  (Direct
268 * calls are essentially free, because the call and return addresses
269 * are completely predictable.)
270 *
271 * For i386, these macros rely on the standard gcc "regparm(3)" calling
272 * convention, in which the first three arguments are placed in %eax,
273 * %edx, %ecx (in that order), and the remaining arguments are placed
274 * on the stack.  All caller-save registers (eax,edx,ecx) are expected
275 * to be modified (either clobbered or used for return values).
276 * X86_64, on the other hand, already specifies a register-based calling
277 * conventions, returning at %rax, with parameters going in %rdi, %rsi,
278 * %rdx, and %rcx. Note that for this reason, x86_64 does not need any
279 * special handling for dealing with 4 arguments, unlike i386.
280 * However, x86_64 also has to clobber all caller saved registers, which
281 * unfortunately, are quite a bit (r8 - r11)
282 *
283 * Unfortunately there's no way to get gcc to generate the args setup
284 * for the call, and then allow the call itself to be generated by an
285 * inline asm.  Because of this, we must do the complete arg setup and
286 * return value handling from within these macros.  This is fairly
287 * cumbersome.
288 *
289 * There are 5 sets of PVOP_* macros for dealing with 0-4 arguments.
290 * It could be extended to more arguments, but there would be little
291 * to be gained from that.  For each number of arguments, there are
292 * two VCALL and CALL variants for void and non-void functions.
293 *
294 * When there is a return value, the invoker of the macro must specify
295 * the return type.  The macro then uses sizeof() on that type to
296 * determine whether it's a 32 or 64 bit value and places the return
297 * in the right register(s) (just %eax for 32-bit, and %edx:%eax for
298 * 64-bit). For x86_64 machines, it just returns in %rax regardless of
299 * the return value size.
300 *
301 * 64-bit arguments are passed as a pair of adjacent 32-bit arguments;
302 * i386 also passes 64-bit arguments as a pair of adjacent 32-bit arguments
303 * in low,high order
304 *
305 * Small structures are passed and returned in registers.  The macro
306 * calling convention can't directly deal with this, so the wrapper
307 * functions must do it.
308 *
309 * These PVOP_* macros are only defined within this header.  This
310 * means that all uses must be wrapped in inline functions.  This also
311 * makes sure the incoming and outgoing types are always correct.
312 */
313#ifdef CONFIG_X86_32
314#define PVOP_CALL_ARGS							\
315	unsigned long __eax = __eax, __edx = __edx, __ecx = __ecx;
316
317#define PVOP_CALL_ARG1(x)		"a" ((unsigned long)(x))
318#define PVOP_CALL_ARG2(x)		"d" ((unsigned long)(x))
319#define PVOP_CALL_ARG3(x)		"c" ((unsigned long)(x))
320
321#define PVOP_VCALL_CLOBBERS		"=a" (__eax), "=d" (__edx),	\
322					"=c" (__ecx)
323#define PVOP_CALL_CLOBBERS		PVOP_VCALL_CLOBBERS
324
325#define PVOP_VCALLEE_CLOBBERS		"=a" (__eax), "=d" (__edx)
326#define PVOP_CALLEE_CLOBBERS		PVOP_VCALLEE_CLOBBERS
327
328#define EXTRA_CLOBBERS
329#define VEXTRA_CLOBBERS
330#else  /* CONFIG_X86_64 */
331/* [re]ax isn't an arg, but the return val */
332#define PVOP_CALL_ARGS						\
333	unsigned long __edi = __edi, __esi = __esi,		\
334		__edx = __edx, __ecx = __ecx, __eax = __eax;
335
336#define PVOP_CALL_ARG1(x)		"D" ((unsigned long)(x))
337#define PVOP_CALL_ARG2(x)		"S" ((unsigned long)(x))
338#define PVOP_CALL_ARG3(x)		"d" ((unsigned long)(x))
339#define PVOP_CALL_ARG4(x)		"c" ((unsigned long)(x))
340
341#define PVOP_VCALL_CLOBBERS	"=D" (__edi),				\
342				"=S" (__esi), "=d" (__edx),		\
343				"=c" (__ecx)
344#define PVOP_CALL_CLOBBERS	PVOP_VCALL_CLOBBERS, "=a" (__eax)
345
346/*
347 * void functions are still allowed [re]ax for scratch.
348 *
349 * The ZERO_CALL_USED REGS feature may end up zeroing out callee-saved
350 * registers. Make sure we model this with the appropriate clobbers.
351 */
352#ifdef CONFIG_ZERO_CALL_USED_REGS
353#define PVOP_VCALLEE_CLOBBERS	"=a" (__eax), PVOP_VCALL_CLOBBERS
354#else
355#define PVOP_VCALLEE_CLOBBERS	"=a" (__eax)
356#endif
357#define PVOP_CALLEE_CLOBBERS	PVOP_VCALLEE_CLOBBERS
358
359#define EXTRA_CLOBBERS	 , "r8", "r9", "r10", "r11"
360#define VEXTRA_CLOBBERS	 , "rax", "r8", "r9", "r10", "r11"
361#endif	/* CONFIG_X86_32 */
362
363#ifdef CONFIG_PARAVIRT_DEBUG
364#define PVOP_TEST_NULL(op)	BUG_ON(pv_ops.op == NULL)
365#else
366#define PVOP_TEST_NULL(op)	((void)pv_ops.op)
367#endif
368
369#define PVOP_RETVAL(rettype)						\
370	({	unsigned long __mask = ~0UL;				\
371		BUILD_BUG_ON(sizeof(rettype) > sizeof(unsigned long));	\
372		switch (sizeof(rettype)) {				\
373		case 1: __mask =       0xffUL; break;			\
374		case 2: __mask =     0xffffUL; break;			\
375		case 4: __mask = 0xffffffffUL; break;			\
376		default: break;						\
377		}							\
378		__mask & __eax;						\
379	})
380
381/*
382 * Use alternative patching for paravirt calls:
383 * - For replacing an indirect call with a direct one, use the "normal"
384 *   ALTERNATIVE() macro with the indirect call as the initial code sequence,
385 *   which will be replaced with the related direct call by using the
386 *   ALT_FLAG_DIRECT_CALL special case and the "always on" feature.
387 * - In case the replacement is either a direct call or a short code sequence
388 *   depending on a feature bit, the ALTERNATIVE_2() macro is being used.
389 *   The indirect call is the initial code sequence again, while the special
390 *   code sequence is selected with the specified feature bit. In case the
391 *   feature is not active, the direct call is used as above via the
392 *   ALT_FLAG_DIRECT_CALL special case and the "always on" feature.
393 */
394#define ____PVOP_CALL(ret, op, call_clbr, extra_clbr, ...)	\
395	({								\
396		PVOP_CALL_ARGS;						\
397		PVOP_TEST_NULL(op);					\
398		asm volatile(ALTERNATIVE(PARAVIRT_CALL, ALT_CALL_INSTR,	\
399				ALT_CALL_ALWAYS)			\
400			     : call_clbr, ASM_CALL_CONSTRAINT		\
401			     : paravirt_ptr(op),			\
402			       ##__VA_ARGS__				\
403			     : "memory", "cc" extra_clbr);		\
404		ret;							\
405	})
406
407#define ____PVOP_ALT_CALL(ret, op, alt, cond, call_clbr,		\
408			  extra_clbr, ...)				\
409	({								\
410		PVOP_CALL_ARGS;						\
411		PVOP_TEST_NULL(op);					\
412		asm volatile(ALTERNATIVE_2(PARAVIRT_CALL,		\
413				 ALT_CALL_INSTR, ALT_CALL_ALWAYS,	\
414				 alt, cond)				\
415			     : call_clbr, ASM_CALL_CONSTRAINT		\
416			     : paravirt_ptr(op),			\
417			       ##__VA_ARGS__				\
418			     : "memory", "cc" extra_clbr);		\
419		ret;							\
420	})
421
422#define __PVOP_CALL(rettype, op, ...)					\
423	____PVOP_CALL(PVOP_RETVAL(rettype), op,				\
424		      PVOP_CALL_CLOBBERS, EXTRA_CLOBBERS, ##__VA_ARGS__)
425
426#define __PVOP_ALT_CALL(rettype, op, alt, cond, ...)			\
427	____PVOP_ALT_CALL(PVOP_RETVAL(rettype), op, alt, cond,		\
428			  PVOP_CALL_CLOBBERS, EXTRA_CLOBBERS,		\
429			  ##__VA_ARGS__)
430
431#define __PVOP_CALLEESAVE(rettype, op, ...)				\
432	____PVOP_CALL(PVOP_RETVAL(rettype), op.func,			\
433		      PVOP_CALLEE_CLOBBERS, , ##__VA_ARGS__)
434
435#define __PVOP_ALT_CALLEESAVE(rettype, op, alt, cond, ...)		\
436	____PVOP_ALT_CALL(PVOP_RETVAL(rettype), op.func, alt, cond,	\
437			  PVOP_CALLEE_CLOBBERS, , ##__VA_ARGS__)
438
439
440#define __PVOP_VCALL(op, ...)						\
441	(void)____PVOP_CALL(, op, PVOP_VCALL_CLOBBERS,			\
442		       VEXTRA_CLOBBERS, ##__VA_ARGS__)
443
444#define __PVOP_ALT_VCALL(op, alt, cond, ...)				\
445	(void)____PVOP_ALT_CALL(, op, alt, cond,			\
446				PVOP_VCALL_CLOBBERS, VEXTRA_CLOBBERS,	\
447				##__VA_ARGS__)
448
449#define __PVOP_VCALLEESAVE(op, ...)					\
450	(void)____PVOP_CALL(, op.func,					\
451			    PVOP_VCALLEE_CLOBBERS, , ##__VA_ARGS__)
452
453#define __PVOP_ALT_VCALLEESAVE(op, alt, cond, ...)			\
454	(void)____PVOP_ALT_CALL(, op.func, alt, cond,			\
455				PVOP_VCALLEE_CLOBBERS, , ##__VA_ARGS__)
456
457
458#define PVOP_CALL0(rettype, op)						\
459	__PVOP_CALL(rettype, op)
460#define PVOP_VCALL0(op)							\
461	__PVOP_VCALL(op)
462#define PVOP_ALT_CALL0(rettype, op, alt, cond)				\
463	__PVOP_ALT_CALL(rettype, op, alt, cond)
464#define PVOP_ALT_VCALL0(op, alt, cond)					\
465	__PVOP_ALT_VCALL(op, alt, cond)
466
467#define PVOP_CALLEE0(rettype, op)					\
468	__PVOP_CALLEESAVE(rettype, op)
469#define PVOP_VCALLEE0(op)						\
470	__PVOP_VCALLEESAVE(op)
471#define PVOP_ALT_CALLEE0(rettype, op, alt, cond)			\
472	__PVOP_ALT_CALLEESAVE(rettype, op, alt, cond)
473#define PVOP_ALT_VCALLEE0(op, alt, cond)				\
474	__PVOP_ALT_VCALLEESAVE(op, alt, cond)
475
476
477#define PVOP_CALL1(rettype, op, arg1)					\
478	__PVOP_CALL(rettype, op, PVOP_CALL_ARG1(arg1))
479#define PVOP_VCALL1(op, arg1)						\
480	__PVOP_VCALL(op, PVOP_CALL_ARG1(arg1))
481#define PVOP_ALT_VCALL1(op, arg1, alt, cond)				\
482	__PVOP_ALT_VCALL(op, alt, cond, PVOP_CALL_ARG1(arg1))
483
484#define PVOP_CALLEE1(rettype, op, arg1)					\
485	__PVOP_CALLEESAVE(rettype, op, PVOP_CALL_ARG1(arg1))
486#define PVOP_VCALLEE1(op, arg1)						\
487	__PVOP_VCALLEESAVE(op, PVOP_CALL_ARG1(arg1))
488#define PVOP_ALT_CALLEE1(rettype, op, arg1, alt, cond)			\
489	__PVOP_ALT_CALLEESAVE(rettype, op, alt, cond, PVOP_CALL_ARG1(arg1))
490#define PVOP_ALT_VCALLEE1(op, arg1, alt, cond)				\
491	__PVOP_ALT_VCALLEESAVE(op, alt, cond, PVOP_CALL_ARG1(arg1))
492
493
494#define PVOP_CALL2(rettype, op, arg1, arg2)				\
495	__PVOP_CALL(rettype, op, PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2))
496#define PVOP_VCALL2(op, arg1, arg2)					\
497	__PVOP_VCALL(op, PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2))
498
499#define PVOP_CALL3(rettype, op, arg1, arg2, arg3)			\
500	__PVOP_CALL(rettype, op, PVOP_CALL_ARG1(arg1),			\
501		    PVOP_CALL_ARG2(arg2), PVOP_CALL_ARG3(arg3))
502#define PVOP_VCALL3(op, arg1, arg2, arg3)				\
503	__PVOP_VCALL(op, PVOP_CALL_ARG1(arg1),				\
504		     PVOP_CALL_ARG2(arg2), PVOP_CALL_ARG3(arg3))
505
506#define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4)			\
507	__PVOP_CALL(rettype, op,					\
508		    PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2),		\
509		    PVOP_CALL_ARG3(arg3), PVOP_CALL_ARG4(arg4))
510#define PVOP_VCALL4(op, arg1, arg2, arg3, arg4)				\
511	__PVOP_VCALL(op, PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2),	\
512		     PVOP_CALL_ARG3(arg3), PVOP_CALL_ARG4(arg4))
513
514unsigned long paravirt_ret0(void);
515#ifdef CONFIG_PARAVIRT_XXL
516u64 _paravirt_ident_64(u64);
517unsigned long pv_native_save_fl(void);
518void pv_native_irq_disable(void);
519void pv_native_irq_enable(void);
520unsigned long pv_native_read_cr2(void);
521#endif
522
523#define paravirt_nop	((void *)nop_func)
524
525#endif	/* __ASSEMBLER__ */
526
527#define ALT_NOT_XEN	ALT_NOT(X86_FEATURE_XENPV)
528
529#endif  /* CONFIG_PARAVIRT */
530#endif	/* _ASM_X86_PARAVIRT_TYPES_H */