1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2020 Google LLC
4 * Author: Quentin Perret <qperret@google.com>
7 #include <asm/kvm_hyp.h>
13 * Index the hyp_vmemmap to find a potential buddy page, but make no assumption
14 * about its current state.
16 * Example buddy-tree for a 4-pages physically contiguous pool:
27 * Example of requests on this pool:
28 * __find_buddy_nocheck(pool, page 0, order 0) => page 1
29 * __find_buddy_nocheck(pool, page 0, order 1) => page 2
30 * __find_buddy_nocheck(pool, page 1, order 0) => page 0
31 * __find_buddy_nocheck(pool, page 2, order 0) => page 3
33 static struct hyp_page *__find_buddy_nocheck(struct hyp_pool *pool,
37 phys_addr_t addr = hyp_page_to_phys(p);
39 addr ^= (PAGE_SIZE << order);
42 * Don't return a page outside the pool range -- it belongs to
43 * something else and may not be mapped in hyp_vmemmap.
45 if (addr < pool->range_start || addr >= pool->range_end)
48 return hyp_phys_to_page(addr);
51 /* Find a buddy page currently available for allocation */
52 static struct hyp_page *__find_buddy_avail(struct hyp_pool *pool,
56 struct hyp_page *buddy = __find_buddy_nocheck(pool, p, order);
58 if (!buddy || buddy->order != order || buddy->refcount)
66 * Pages that are available for allocation are tracked in free-lists, so we use
67 * the pages themselves to store the list nodes to avoid wasting space. As the
68 * allocator always returns zeroed pages (which are zeroed on the hyp_put_page()
69 * path to optimize allocation speed), we also need to clean-up the list node in
70 * each page when we take it out of the list.
72 static inline void page_remove_from_list(struct hyp_page *p)
74 struct list_head *node = hyp_page_to_virt(p);
76 __list_del_entry(node);
77 memset(node, 0, sizeof(*node));
80 static inline void page_add_to_list(struct hyp_page *p, struct list_head *head)
82 struct list_head *node = hyp_page_to_virt(p);
85 list_add_tail(node, head);
88 static inline struct hyp_page *node_to_page(struct list_head *node)
90 return hyp_virt_to_page(node);
93 static void __hyp_attach_page(struct hyp_pool *pool,
96 unsigned short order = p->order;
97 struct hyp_page *buddy;
99 memset(hyp_page_to_virt(p), 0, PAGE_SIZE << p->order);
102 * Only the first struct hyp_page of a high-order page (otherwise known
103 * as the 'head') should have p->order set. The non-head pages should
104 * have p->order = HYP_NO_ORDER. Here @p may no longer be the head
105 * after coallescing, so make sure to mark it HYP_NO_ORDER proactively.
107 p->order = HYP_NO_ORDER;
108 for (; (order + 1) < pool->max_order; order++) {
109 buddy = __find_buddy_avail(pool, p, order);
113 /* Take the buddy out of its list, and coallesce with @p */
114 page_remove_from_list(buddy);
115 buddy->order = HYP_NO_ORDER;
119 /* Mark the new head, and insert it */
121 page_add_to_list(p, &pool->free_area[order]);
124 static struct hyp_page *__hyp_extract_page(struct hyp_pool *pool,
126 unsigned short order)
128 struct hyp_page *buddy;
130 page_remove_from_list(p);
131 while (p->order > order) {
133 * The buddy of order n - 1 currently has HYP_NO_ORDER as it
134 * is covered by a higher-level page (whose head is @p). Use
135 * __find_buddy_nocheck() to find it and inject it in the
136 * free_list[n - 1], effectively splitting @p in half.
139 buddy = __find_buddy_nocheck(pool, p, p->order);
140 buddy->order = p->order;
141 page_add_to_list(buddy, &pool->free_area[buddy->order]);
147 static inline void hyp_page_ref_inc(struct hyp_page *p)
149 BUG_ON(p->refcount == USHRT_MAX);
153 static inline int hyp_page_ref_dec_and_test(struct hyp_page *p)
156 return (p->refcount == 0);
159 static inline void hyp_set_page_refcounted(struct hyp_page *p)
165 static void __hyp_put_page(struct hyp_pool *pool, struct hyp_page *p)
167 if (hyp_page_ref_dec_and_test(p))
168 __hyp_attach_page(pool, p);
172 * Changes to the buddy tree and page refcounts must be done with the hyp_pool
173 * lock held. If a refcount change requires an update to the buddy tree (e.g.
174 * hyp_put_page()), both operations must be done within the same critical
175 * section to guarantee transient states (e.g. a page with null refcount but
176 * not yet attached to a free list) can't be observed by well-behaved readers.
178 void hyp_put_page(struct hyp_pool *pool, void *addr)
180 struct hyp_page *p = hyp_virt_to_page(addr);
182 hyp_spin_lock(&pool->lock);
183 __hyp_put_page(pool, p);
184 hyp_spin_unlock(&pool->lock);
187 void hyp_get_page(struct hyp_pool *pool, void *addr)
189 struct hyp_page *p = hyp_virt_to_page(addr);
191 hyp_spin_lock(&pool->lock);
193 hyp_spin_unlock(&pool->lock);
196 void *hyp_alloc_pages(struct hyp_pool *pool, unsigned short order)
198 unsigned short i = order;
201 hyp_spin_lock(&pool->lock);
203 /* Look for a high-enough-order page */
204 while (i < pool->max_order && list_empty(&pool->free_area[i]))
206 if (i >= pool->max_order) {
207 hyp_spin_unlock(&pool->lock);
211 /* Extract it from the tree at the right order */
212 p = node_to_page(pool->free_area[i].next);
213 p = __hyp_extract_page(pool, p, order);
215 hyp_set_page_refcounted(p);
216 hyp_spin_unlock(&pool->lock);
218 return hyp_page_to_virt(p);
221 int hyp_pool_init(struct hyp_pool *pool, u64 pfn, unsigned int nr_pages,
222 unsigned int reserved_pages)
224 phys_addr_t phys = hyp_pfn_to_phys(pfn);
228 hyp_spin_lock_init(&pool->lock);
229 pool->max_order = min(MAX_ORDER, get_order(nr_pages << PAGE_SHIFT));
230 for (i = 0; i < pool->max_order; i++)
231 INIT_LIST_HEAD(&pool->free_area[i]);
232 pool->range_start = phys;
233 pool->range_end = phys + (nr_pages << PAGE_SHIFT);
235 /* Init the vmemmap portion */
236 p = hyp_phys_to_page(phys);
237 for (i = 0; i < nr_pages; i++) {
239 hyp_set_page_refcounted(&p[i]);
242 /* Attach the unused pages to the buddy tree */
243 for (i = reserved_pages; i < nr_pages; i++)
244 __hyp_put_page(pool, &p[i]);