1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Copyright © 2022 Intel Corporation
  4 */
  5#include "xe_pt_walk.h"
  6
  7/**
  8 * DOC: GPU page-table tree walking.
  9 * The utilities in this file are similar to the CPU page-table walk
 10 * utilities in mm/pagewalk.c. The main difference is that we distinguish
 11 * the various levels of a page-table tree with an unsigned integer rather
 12 * than by name. 0 is the lowest level, and page-tables with level 0 can
 13 * not be directories pointing to lower levels, whereas all other levels
 14 * can. The user of the utilities determines the highest level.
 15 *
 16 * Nomenclature:
 17 * Each struct xe_ptw, regardless of level is referred to as a page table, and
 18 * multiple page tables typically form a page table tree with page tables at
 19 * intermediate levels being page directories pointing at page tables at lower
 20 * levels. A shared page table for a given address range is a page-table which
 21 * is neither fully within nor fully outside the address range and that can
 22 * thus be shared by two or more address ranges.
 23 *
 24 * Please keep this code generic so that it can used as a drm-wide page-
 25 * table walker should other drivers find use for it.
 26 */
 27static u64 xe_pt_addr_end(u64 addr, u64 end, unsigned int level,
 28			  const struct xe_pt_walk *walk)
 29{
 30	u64 size = 1ull << walk->shifts[level];
 31	u64 tmp = round_up(addr + 1, size);
 32
 33	return min_t(u64, tmp, end);
 34}
 35
 36static bool xe_pt_next(pgoff_t *offset, u64 *addr, u64 next, u64 end,
 37		       unsigned int level, const struct xe_pt_walk *walk)
 38{
 39	pgoff_t step = 1;
 40
 41	/* Shared pt walk skips to the last pagetable */
 42	if (unlikely(walk->shared_pt_mode)) {
 43		unsigned int shift = walk->shifts[level];
 44		u64 skip_to = round_down(end, 1ull << shift);
 45
 46		if (skip_to > next) {
 47			step += (skip_to - next) >> shift;
 48			next = skip_to;
 49		}
 50	}
 51
 52	*addr = next;
 53	*offset += step;
 54
 55	return next != end;
 56}
 57
 58/**
 59 * xe_pt_walk_range() - Walk a range of a gpu page table tree with callbacks
 60 * for each page-table entry in all levels.
 61 * @parent: The root page table for walk start.
 62 * @level: The root page table level.
 63 * @addr: Virtual address start.
 64 * @end: Virtual address end + 1.
 65 * @walk: Walk info.
 66 *
 67 * Similar to the CPU page-table walker, this is a helper to walk
 68 * a gpu page table and call a provided callback function for each entry.
 69 *
 70 * Return: 0 on success, negative error code on error. The error is
 71 * propagated from the callback and on error the walk is terminated.
 72 */
 73int xe_pt_walk_range(struct xe_ptw *parent, unsigned int level,
 74		     u64 addr, u64 end, struct xe_pt_walk *walk)
 75{
 76	pgoff_t offset = xe_pt_offset(addr, level, walk);
 77	struct xe_ptw **entries = walk->staging ? (parent->staging ?: NULL) :
 78		(parent->children ?: NULL);
 79	const struct xe_pt_walk_ops *ops = walk->ops;
 80	enum page_walk_action action;
 81	struct xe_ptw *child;
 82	int err = 0;
 83	u64 next;
 84
 85	do {
 86		next = xe_pt_addr_end(addr, end, level, walk);
 87		if (walk->shared_pt_mode && xe_pt_covers(addr, next, level,
 88							 walk))
 89			continue;
 90again:
 91		action = ACTION_SUBTREE;
 92		child = entries ? entries[offset] : NULL;
 93		err = ops->pt_entry(parent, offset, level, addr, next,
 94				    &child, &action, walk);
 95		if (err)
 96			break;
 97
 98		/* Probably not needed yet for gpu pagetable walk. */
 99		if (unlikely(action == ACTION_AGAIN))
100			goto again;
101
102		if (likely(!level || !child || action == ACTION_CONTINUE))
103			continue;
104
105		err = xe_pt_walk_range(child, level - 1, addr, next, walk);
106
107		if (!err && ops->pt_post_descend)
108			err = ops->pt_post_descend(parent, offset, level, addr,
109						   next, &child, &action, walk);
110		if (err)
111			break;
112
113	} while (xe_pt_next(&offset, &addr, next, end, level, walk));
114
115	return err;
116}
117
118/**
119 * xe_pt_walk_shared() - Walk shared page tables of a page-table tree.
120 * @parent: Root page table directory.
121 * @level: Level of the root.
122 * @addr: Start address.
123 * @end: Last address + 1.
124 * @walk: Walk info.
125 *
126 * This function is similar to xe_pt_walk_range() but it skips page tables
127 * that are private to the range. Since the root (or @parent) page table is
128 * typically also a shared page table this function is different in that it
129 * calls the pt_entry callback and the post_descend callback also for the
130 * root. The root can be detected in the callbacks by checking whether
131 * parent == *child.
132 * Walking only the shared page tables is common for unbind-type operations
133 * where the page-table entries for an address range are cleared or detached
134 * from the main page-table tree.
135 *
136 * Return: 0 on success, negative error code on error: If a callback
137 * returns an error, the walk will be terminated and the error returned by
138 * this function.
139 */
140int xe_pt_walk_shared(struct xe_ptw *parent, unsigned int level,
141		      u64 addr, u64 end, struct xe_pt_walk *walk)
142{
143	const struct xe_pt_walk_ops *ops = walk->ops;
144	enum page_walk_action action = ACTION_SUBTREE;
145	struct xe_ptw *child = parent;
146	int err;
147
148	walk->shared_pt_mode = true;
149	err = walk->ops->pt_entry(parent, 0, level + 1, addr, end,
150				  &child, &action, walk);
151
152	if (err || action != ACTION_SUBTREE)
153		return err;
154
155	err = xe_pt_walk_range(parent, level, addr, end, walk);
156	if (!err && ops->pt_post_descend) {
157		err = ops->pt_post_descend(parent, 0, level + 1, addr, end,
158					   &child, &action, walk);
159	}
160	return err;
161}