xref: /titanic_51/usr/src/uts/i86pc/vm/hat_i86.h (revision 71269a2275bf5a143dad6461eee2710a344e7261)
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 #ifndef	_VM_HAT_I86_H
27 #define	_VM_HAT_I86_H
28 
29 
30 #ifdef	__cplusplus
31 extern "C" {
32 #endif
33 
34 /*
35  * VM - Hardware Address Translation management.
36  *
37  * This file describes the contents of the x86_64 HAT data structures.
38  */
39 #include <sys/types.h>
40 #include <sys/t_lock.h>
41 #include <sys/cpuvar.h>
42 #include <sys/x_call.h>
43 #include <vm/seg.h>
44 #include <vm/page.h>
45 #include <sys/vmparam.h>
46 #include <sys/vm_machparam.h>
47 #include <sys/promif.h>
48 #include <vm/hat_pte.h>
49 #include <vm/htable.h>
50 #include <vm/hment.h>
51 
52 /*
53  * The essential data types involved:
54  *
55  * htable_t	- There is one of these for each page table and it is used
56  *		by the HAT to manage the page table.
57  *
58  * hment_t	- Links together multiple PTEs to a single page.
59  */
60 
61 /*
62  * VLP processes have a 32 bit address range, so their top level is 2 and
63  * with only 4 PTEs in that table.
64  */
65 #define	VLP_LEVEL	(2)
66 #define	VLP_NUM_PTES	(4)
67 #define	VLP_SIZE	(VLP_NUM_PTES * sizeof (x86pte_t))
68 #define	TOP_LEVEL(h)	(((h)->hat_flags & HAT_VLP) ? VLP_LEVEL : mmu.max_level)
69 #define	VLP_COPY(fromptep, toptep) { \
70 	toptep[0] = fromptep[0]; \
71 	toptep[1] = fromptep[1]; \
72 	toptep[2] = fromptep[2]; \
73 	toptep[3] = fromptep[3]; \
74 }
75 
76 /*
77  * The hat struct exists for each address space.
78  */
79 struct hat {
80 	kmutex_t	hat_mutex;
81 	struct as	*hat_as;
82 	uint_t		hat_stats;
83 	pgcnt_t		hat_pages_mapped[MAX_PAGE_LEVEL + 1];
84 	pgcnt_t		hat_ism_pgcnt;
85 	cpuset_t	hat_cpus;
86 	uint16_t	hat_flags;
87 	htable_t	*hat_htable;	/* top level htable */
88 	struct hat	*hat_next;
89 	struct hat	*hat_prev;
90 	uint_t		hat_num_hash;	/* number of htable hash buckets */
91 	htable_t	**hat_ht_hash;	/* htable hash buckets */
92 	htable_t	*hat_ht_cached;	/* cached free htables */
93 	x86pte_t	hat_vlp_ptes[VLP_NUM_PTES];
94 #if defined(__amd64) && defined(__xpv)
95 	pfn_t		hat_user_ptable; /* alt top ptable for user mode */
96 #endif
97 };
98 typedef struct hat hat_t;
99 
100 #define	PGCNT_INC(hat, level)	\
101 	atomic_add_long(&(hat)->hat_pages_mapped[level], 1);
102 #define	PGCNT_DEC(hat, level)	\
103 	atomic_add_long(&(hat)->hat_pages_mapped[level], -1);
104 
105 /*
106  * Flags for the hat_flags field
107  *
108  * HAT_FREEING - set when HAT is being destroyed - mostly used to detect that
109  *	demap()s can be avoided.
110  *
111  * HAT_VLP - indicates a 32 bit process has a virtual address range less than
112  *	the hardware's physical address range. (VLP->Virtual Less-than Physical)
113  *	Note - never used on the hypervisor.
114  *
115  * HAT_VICTIM - This is set while a hat is being examined for page table
116  *	stealing and prevents it from being freed.
117  *
118  * HAT_SHARED - The hat has exported it's page tables via hat_share()
119  *
120  * HAT_PINNED - On the hypervisor, indicates the top page table has been pinned.
121  */
122 #define	HAT_FREEING	(0x0001)
123 #define	HAT_VLP		(0x0002)
124 #define	HAT_VICTIM	(0x0004)
125 #define	HAT_SHARED	(0x0008)
126 #define	HAT_PINNED	(0x0010)
127 
128 /*
129  * Additional platform attribute for hat_devload() to force no caching.
130  */
131 #define	HAT_PLAT_NOCACHE	(0x100000)
132 
133 /*
134  * Simple statistics for the HAT. These are just counters that are
135  * atomically incremented. They can be reset directly from the kernel
136  * debugger.
137  */
138 struct hatstats {
139 	ulong_t	hs_reap_attempts;
140 	ulong_t	hs_reaped;
141 	ulong_t	hs_steals;
142 	ulong_t	hs_ptable_allocs;
143 	ulong_t	hs_ptable_frees;
144 	ulong_t	hs_htable_rgets;	/* allocs from reserve */
145 	ulong_t	hs_htable_rputs;	/* putbacks to reserve */
146 	ulong_t	hs_htable_shared;	/* number of htables shared */
147 	ulong_t	hs_htable_unshared;	/* number of htables unshared */
148 	ulong_t	hs_hm_alloc;
149 	ulong_t	hs_hm_free;
150 	ulong_t	hs_hm_put_reserve;
151 	ulong_t	hs_hm_get_reserve;
152 	ulong_t	hs_hm_steals;
153 	ulong_t	hs_hm_steal_exam;
154 	ulong_t hs_tlb_inval_delayed;
155 };
156 extern struct hatstats hatstat;
157 #ifdef DEBUG
158 #define	HATSTAT_INC(x)	(++hatstat.x)
159 #else
160 #define	HATSTAT_INC(x)	(0)
161 #endif
162 
163 #if defined(_KERNEL)
164 
165 /*
166  * Useful macro to align hat_XXX() address arguments to a page boundary
167  */
168 #define	ALIGN2PAGE(a)		((uintptr_t)(a) & MMU_PAGEMASK)
169 #define	IS_PAGEALIGNED(a)	(((uintptr_t)(a) & MMU_PAGEOFFSET) == 0)
170 
171 extern uint_t	khat_running;	/* set at end of hat_kern_setup() */
172 extern cpuset_t khat_cpuset;	/* cpuset for kernal address demap Xcalls */
173 extern kmutex_t hat_list_lock;
174 extern kcondvar_t hat_list_cv;
175 
176 
177 
178 /*
179  * Interfaces to setup a cpu private mapping (ie. preemption disabled).
180  * The attr and flags arguments are the same as for hat_devload().
181  * setup() must be called once, then any number of calls to remap(),
182  * followed by a final call to release()
183  *
184  * Used by ppcopy(), page_zero(), the memscrubber, and the kernel debugger.
185  */
186 typedef paddr_t hat_mempte_t;				/* phys addr of PTE */
187 extern hat_mempte_t hat_mempte_setup(caddr_t addr);
188 extern void hat_mempte_remap(pfn_t, caddr_t, hat_mempte_t,
189 	uint_t attr, uint_t flags);
190 extern void hat_mempte_release(caddr_t addr, hat_mempte_t);
191 
192 /*
193  * Interfaces to manage which thread has access to htable and hment reserves.
194  * The USE_HAT_RESERVES macro should always be recomputed in full. Its value
195  * (due to curthread) can change after any call into kmem/vmem.
196  */
197 extern uint_t can_steal_post_boot;
198 extern uint_t use_boot_reserve;
199 #define	USE_HAT_RESERVES()					\
200 	(use_boot_reserve || curthread->t_hatdepth > 1 ||	\
201 	panicstr != NULL || vmem_is_populator())
202 
203 /*
204  * initialization stuff needed by by startup, mp_startup...
205  */
206 extern void hat_cpu_online(struct cpu *);
207 extern void hat_cpu_offline(struct cpu *);
208 extern void setup_vaddr_for_ppcopy(struct cpu *);
209 extern void teardown_vaddr_for_ppcopy(struct cpu *);
210 extern void clear_boot_mappings(uintptr_t, uintptr_t);
211 
212 /*
213  * magic value to indicate that all TLB entries should be demapped.
214  */
215 #define	DEMAP_ALL_ADDR	(~(uintptr_t)0)
216 
217 /*
218  * not in any include file???
219  */
220 extern void halt(char *fmt);
221 
222 /*
223  * x86 specific routines for use online in setup or i86pc/vm files
224  */
225 extern void hat_kern_alloc(caddr_t segmap_base, size_t segmap_size,
226 	caddr_t ekernelheap);
227 extern void hat_kern_setup(void);
228 extern void hat_tlb_inval(struct hat *hat, uintptr_t va);
229 extern void hat_pte_unmap(htable_t *ht, uint_t entry, uint_t flags,
230 	x86pte_t old_pte, void *pte_ptr);
231 extern void hat_init_finish(void);
232 extern caddr_t hat_kpm_pfn2va(pfn_t pfn);
233 extern pfn_t hat_kpm_va2pfn(caddr_t);
234 extern page_t *hat_kpm_vaddr2page(caddr_t);
235 extern uintptr_t hat_kernelbase(uintptr_t);
236 extern void hat_kmap_init(uintptr_t base, size_t len);
237 
238 extern hment_t *hati_page_unmap(page_t *pp, htable_t *ht, uint_t entry);
239 
240 #if !defined(__xpv)
241 /*
242  * routines to deal with delayed TLB invalidations for idle CPUs
243  */
244 extern void tlb_going_idle(void);
245 extern void tlb_service(void);
246 #endif
247 
248 /*
249  * Hat switch function invoked to load a new context into %cr3
250  */
251 extern void hat_switch(struct hat *hat);
252 
253 #ifdef __xpv
254 /*
255  * Interfaces to use around code that maps/unmaps grant table references.
256  */
257 extern void hat_prepare_mapping(hat_t *, caddr_t, uint64_t *);
258 extern void hat_release_mapping(hat_t *, caddr_t);
259 
260 #define	XPV_DISALLOW_MIGRATE()	xen_block_migrate()
261 #define	XPV_ALLOW_MIGRATE()	xen_allow_migrate()
262 
263 #else
264 
265 #define	XPV_DISALLOW_MIGRATE()	/* nothing */
266 #define	XPV_ALLOW_MIGRATE()	/* nothing */
267 
268 #define	pfn_is_foreign(pfn)	__lintzero
269 
270 #endif
271 
272 
273 #endif	/* _KERNEL */
274 
275 #ifdef	__cplusplus
276 }
277 #endif
278 
279 #endif	/* _VM_HAT_I86_H */
280