xref: /titanic_51/usr/src/uts/i86pc/vm/hat_i86.h (revision bd93c05dbd9b8f1e8d2edf48c777bc881f927608)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 /*
26  * Copyright (c) 2014 by Delphix. All rights reserved.
27  */
28 
29 #ifndef	_VM_HAT_I86_H
30 #define	_VM_HAT_I86_H
31 
32 
33 #ifdef	__cplusplus
34 extern "C" {
35 #endif
36 
37 /*
38  * VM - Hardware Address Translation management.
39  *
40  * This file describes the contents of the x86_64 HAT data structures.
41  */
42 #include <sys/types.h>
43 #include <sys/t_lock.h>
44 #include <sys/cpuvar.h>
45 #include <sys/x_call.h>
46 #include <vm/seg.h>
47 #include <vm/page.h>
48 #include <sys/vmparam.h>
49 #include <sys/vm_machparam.h>
50 #include <sys/promif.h>
51 #include <vm/hat_pte.h>
52 #include <vm/htable.h>
53 #include <vm/hment.h>
54 
55 /*
56  * The essential data types involved:
57  *
58  * htable_t	- There is one of these for each page table and it is used
59  *		by the HAT to manage the page table.
60  *
61  * hment_t	- Links together multiple PTEs to a single page.
62  */
63 
64 /*
65  * VLP processes have a 32 bit address range, so their top level is 2 and
66  * with only 4 PTEs in that table.
67  */
68 #define	VLP_LEVEL	(2)
69 #define	VLP_NUM_PTES	(4)
70 #define	VLP_SIZE	(VLP_NUM_PTES * sizeof (x86pte_t))
71 #define	TOP_LEVEL(h)	(((h)->hat_flags & HAT_VLP) ? VLP_LEVEL : mmu.max_level)
72 #define	VLP_COPY(fromptep, toptep) { \
73 	toptep[0] = fromptep[0]; \
74 	toptep[1] = fromptep[1]; \
75 	toptep[2] = fromptep[2]; \
76 	toptep[3] = fromptep[3]; \
77 }
78 
79 /*
80  * The hat struct exists for each address space.
81  */
82 struct hat {
83 	kmutex_t	hat_mutex;
84 	struct as	*hat_as;
85 	uint_t		hat_stats;
86 	pgcnt_t		hat_pages_mapped[MAX_PAGE_LEVEL + 1];
87 	pgcnt_t		hat_ism_pgcnt;
88 	cpuset_t	hat_cpus;
89 	uint16_t	hat_flags;
90 	htable_t	*hat_htable;	/* top level htable */
91 	struct hat	*hat_next;
92 	struct hat	*hat_prev;
93 	uint_t		hat_num_hash;	/* number of htable hash buckets */
94 	htable_t	**hat_ht_hash;	/* htable hash buckets */
95 	htable_t	*hat_ht_cached;	/* cached free htables */
96 	x86pte_t	hat_vlp_ptes[VLP_NUM_PTES];
97 #if defined(__amd64) && defined(__xpv)
98 	pfn_t		hat_user_ptable; /* alt top ptable for user mode */
99 #endif
100 };
101 typedef struct hat hat_t;
102 
103 #define	PGCNT_INC(hat, level)	\
104 	atomic_inc_ulong(&(hat)->hat_pages_mapped[level]);
105 #define	PGCNT_DEC(hat, level)	\
106 	atomic_dec_ulong(&(hat)->hat_pages_mapped[level]);
107 
108 /*
109  * Flags for the hat_flags field
110  *
111  * HAT_FREEING - set when HAT is being destroyed - mostly used to detect that
112  *	demap()s can be avoided.
113  *
114  * HAT_VLP - indicates a 32 bit process has a virtual address range less than
115  *	the hardware's physical address range. (VLP->Virtual Less-than Physical)
116  *	Note - never used on the hypervisor.
117  *
118  * HAT_VICTIM - This is set while a hat is being examined for page table
119  *	stealing and prevents it from being freed.
120  *
121  * HAT_SHARED - The hat has exported it's page tables via hat_share()
122  *
123  * HAT_PINNED - On the hypervisor, indicates the top page table has been pinned.
124  */
125 #define	HAT_FREEING	(0x0001)
126 #define	HAT_VLP		(0x0002)
127 #define	HAT_VICTIM	(0x0004)
128 #define	HAT_SHARED	(0x0008)
129 #define	HAT_PINNED	(0x0010)
130 
131 /*
132  * Additional platform attribute for hat_devload() to force no caching.
133  */
134 #define	HAT_PLAT_NOCACHE	(0x100000)
135 
136 /*
137  * Simple statistics for the HAT. These are just counters that are
138  * atomically incremented. They can be reset directly from the kernel
139  * debugger.
140  */
141 struct hatstats {
142 	ulong_t	hs_reap_attempts;
143 	ulong_t	hs_reaped;
144 	ulong_t	hs_steals;
145 	ulong_t	hs_ptable_allocs;
146 	ulong_t	hs_ptable_frees;
147 	ulong_t	hs_htable_rgets;	/* allocs from reserve */
148 	ulong_t	hs_htable_rputs;	/* putbacks to reserve */
149 	ulong_t	hs_htable_shared;	/* number of htables shared */
150 	ulong_t	hs_htable_unshared;	/* number of htables unshared */
151 	ulong_t	hs_hm_alloc;
152 	ulong_t	hs_hm_free;
153 	ulong_t	hs_hm_put_reserve;
154 	ulong_t	hs_hm_get_reserve;
155 	ulong_t	hs_hm_steals;
156 	ulong_t	hs_hm_steal_exam;
157 	ulong_t hs_tlb_inval_delayed;
158 };
159 extern struct hatstats hatstat;
160 #ifdef DEBUG
161 #define	HATSTAT_INC(x)	(++hatstat.x)
162 #else
163 #define	HATSTAT_INC(x)	(0)
164 #endif
165 
166 #if defined(_KERNEL)
167 
168 /*
169  * Useful macro to align hat_XXX() address arguments to a page boundary
170  */
171 #define	ALIGN2PAGE(a)		((uintptr_t)(a) & MMU_PAGEMASK)
172 #define	IS_PAGEALIGNED(a)	(((uintptr_t)(a) & MMU_PAGEOFFSET) == 0)
173 
174 extern uint_t	khat_running;	/* set at end of hat_kern_setup() */
175 extern cpuset_t khat_cpuset;	/* cpuset for kernal address demap Xcalls */
176 extern kmutex_t hat_list_lock;
177 extern kcondvar_t hat_list_cv;
178 
179 
180 
181 /*
182  * Interfaces to setup a cpu private mapping (ie. preemption disabled).
183  * The attr and flags arguments are the same as for hat_devload().
184  * setup() must be called once, then any number of calls to remap(),
185  * followed by a final call to release()
186  *
187  * Used by ppcopy(), page_zero(), the memscrubber, and the kernel debugger.
188  */
189 typedef paddr_t hat_mempte_t;				/* phys addr of PTE */
190 extern hat_mempte_t hat_mempte_setup(caddr_t addr);
191 extern void hat_mempte_remap(pfn_t, caddr_t, hat_mempte_t,
192 	uint_t attr, uint_t flags);
193 extern void hat_mempte_release(caddr_t addr, hat_mempte_t);
194 
195 /*
196  * Interfaces to manage which thread has access to htable and hment reserves.
197  * The USE_HAT_RESERVES macro should always be recomputed in full. Its value
198  * (due to curthread) can change after any call into kmem/vmem.
199  */
200 extern uint_t can_steal_post_boot;
201 extern uint_t use_boot_reserve;
202 #define	USE_HAT_RESERVES()					\
203 	(use_boot_reserve || curthread->t_hatdepth > 1 ||	\
204 	panicstr != NULL || vmem_is_populator())
205 
206 /*
207  * initialization stuff needed by by startup, mp_startup...
208  */
209 extern void hat_cpu_online(struct cpu *);
210 extern void hat_cpu_offline(struct cpu *);
211 extern void setup_vaddr_for_ppcopy(struct cpu *);
212 extern void teardown_vaddr_for_ppcopy(struct cpu *);
213 extern void clear_boot_mappings(uintptr_t, uintptr_t);
214 
215 /*
216  * magic value to indicate that all TLB entries should be demapped.
217  */
218 #define	DEMAP_ALL_ADDR	(~(uintptr_t)0)
219 
220 /*
221  * not in any include file???
222  */
223 extern void halt(char *fmt);
224 
225 /*
226  * x86 specific routines for use online in setup or i86pc/vm files
227  */
228 extern void hat_kern_alloc(caddr_t segmap_base, size_t segmap_size,
229 	caddr_t ekernelheap);
230 extern void hat_kern_setup(void);
231 extern void hat_tlb_inval(struct hat *hat, uintptr_t va);
232 extern void hat_pte_unmap(htable_t *ht, uint_t entry, uint_t flags,
233 	x86pte_t old_pte, void *pte_ptr, boolean_t tlb);
234 extern void hat_init_finish(void);
235 extern caddr_t hat_kpm_pfn2va(pfn_t pfn);
236 extern pfn_t hat_kpm_va2pfn(caddr_t);
237 extern page_t *hat_kpm_vaddr2page(caddr_t);
238 extern uintptr_t hat_kernelbase(uintptr_t);
239 extern void hat_kmap_init(uintptr_t base, size_t len);
240 
241 extern hment_t *hati_page_unmap(page_t *pp, htable_t *ht, uint_t entry);
242 
243 #if !defined(__xpv)
244 /*
245  * routines to deal with delayed TLB invalidations for idle CPUs
246  */
247 extern void tlb_going_idle(void);
248 extern void tlb_service(void);
249 #endif
250 
251 /*
252  * Hat switch function invoked to load a new context into %cr3
253  */
254 extern void hat_switch(struct hat *hat);
255 
256 #ifdef __xpv
257 /*
258  * Interfaces to use around code that maps/unmaps grant table references.
259  */
260 extern void hat_prepare_mapping(hat_t *, caddr_t, uint64_t *);
261 extern void hat_release_mapping(hat_t *, caddr_t);
262 
263 #define	XPV_DISALLOW_MIGRATE()	xen_block_migrate()
264 #define	XPV_ALLOW_MIGRATE()	xen_allow_migrate()
265 
266 #else
267 
268 #define	XPV_DISALLOW_MIGRATE()	/* nothing */
269 #define	XPV_ALLOW_MIGRATE()	/* nothing */
270 
271 #define	pfn_is_foreign(pfn)	__lintzero
272 
273 #endif
274 
275 
276 #endif	/* _KERNEL */
277 
278 #ifdef	__cplusplus
279 }
280 #endif
281 
282 #endif	/* _VM_HAT_I86_H */
283