On Thu, Oct 08, 2020 at 03:29:49PM +0200, [email protected] wrote: > >Synopsis: Page fault when watching 2k videos > >Category: kernel > >Environment: > System : OpenBSD 6.8 > Details : OpenBSD 6.8-current (GENERIC.MP) #99: Wed Oct 7 22:24:13 > MDT 2020 > > [email protected]:/usr/src/sys/arch/amd64/compile/GENERIC.MP > > Architecture: OpenBSD.amd64 > Machine : amd64 > >Description: > The whole system crashes when I'm using the GPU too much (e.g. watching 2k > videos on YouTube). While they run smooth at the start, graphical artifacts > appear a few minutes later ending with a complete crash to the kernel > console. The logs are full of no-retry page-faults (see below). > I'm using an AMD Ryzen 2400g (APU) with a AB350N-Gaming WIFI Motherboard and > the latest OpenBSD snapshot, connected via HDMI to my 2K 27" monitor. > > By the way: The amdgpu driver didn't work at all for me in 6.7 and it crashed > with "/bsd: [drm] *ERROR* construct: Invalid Connector ObjectID from Adapter > Service for connector index:1! type 0 expected 3", so the 6.8 update is still > a huge improvement as it works somewhat at least - thanks for your work! :) > >How-To-Repeat: > Watching high quality videos, generally doing GPU-extensive tasks. > >Fix: > General GPU usage works great and smooth, so not doing GPU-extensive tasks is > a fix.
The most direct way I am aware of to reproduce the no-retry page fault situation is with piglit from ports piglit run quick -t "[email protected]@execution@texelfetch fs sampler2d 1x281-501x281" out With the following diff it no longer occurs in testing on vega10 / Vega 56. diff --git sys/dev/pci/drm/drm_mm.c sys/dev/pci/drm/drm_mm.c index 58a1afe504b..94c837ae655 100644 --- sys/dev/pci/drm/drm_mm.c +++ sys/dev/pci/drm/drm_mm.c @@ -93,19 +93,11 @@ * some basic allocator dumpers for debugging. * * Note that this range allocator is not thread-safe, drivers need to protect - * modifications with their on locking. The idea behind this is that for a full + * modifications with their own locking. The idea behind this is that for a full * memory manager additional data needs to be protected anyway, hence internal * locking would be fully redundant. */ -static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_mm *mm, - u64 size, - u64 alignment, - unsigned long color, - u64 start, - u64 end, - enum drm_mm_search_flags flags); - #ifdef CONFIG_DRM_DEBUG_MM #include <linux/stackdepot.h> @@ -115,25 +107,19 @@ static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_ static noinline void save_stack(struct drm_mm_node *node) { unsigned long entries[STACKDEPTH]; - struct stack_trace trace = { - .entries = entries, - .max_entries = STACKDEPTH, - .skip = 1 - }; + unsigned int n; - save_stack_trace(&trace); - if (trace.nr_entries != 0 && - trace.entries[trace.nr_entries-1] == ULONG_MAX) - trace.nr_entries--; + n = stack_trace_save(entries, ARRAY_SIZE(entries), 1); /* May be called under spinlock, so avoid sleeping */ - node->stack = depot_save_stack(&trace, GFP_NOWAIT); + node->stack = stack_depot_save(entries, n, GFP_NOWAIT); } static void show_leaks(struct drm_mm *mm) { struct drm_mm_node *node; - unsigned long entries[STACKDEPTH]; + unsigned long *entries; + unsigned int nr_entries; char *buf; buf = kmalloc(BUFSZ, GFP_KERNEL); @@ -141,19 +127,14 @@ static void show_leaks(struct drm_mm *mm) return; list_for_each_entry(node, drm_mm_nodes(mm), node_list) { - struct stack_trace trace = { - .entries = entries, - .max_entries = STACKDEPTH - }; - if (!node->stack) { DRM_ERROR("node [%08llx + %08llx]: unknown owner\n", node->start, node->size); continue; } - depot_fetch_stack(node->stack, &trace); - snprint_stack_trace(buf, BUFSZ, &trace, 0); + nr_entries = stack_depot_fetch(node->stack, &entries); + stack_trace_snprint(buf, BUFSZ, entries, nr_entries, 0); DRM_ERROR("node [%08llx + %08llx]: inserted at\n%s", node->start, node->size, buf); } @@ -176,39 +157,85 @@ INTERVAL_TREE_DEFINE(struct drm_mm_node, rb, u64, __subtree_last, START, LAST, static inline, drm_mm_interval_tree) #else -struct drm_mm_node * -drm_mm_interval_tree_iter_first(struct rb_root *rb, u64 start, u64 last) +static struct drm_mm_node * +drm_mm_interval_tree_iter_first(struct rb_root_cached *root, uint64_t start, + uint64_t last) { - struct drm_mm *mm = container_of(rb, typeof(*mm), interval_tree); struct drm_mm_node *node; + struct rb_node *rb; + + for (rb = rb_first_cached(root); rb; rb = rb_next(rb)) { + node = rb_entry(rb, typeof(*node), rb); + if (LAST(node) >= start && START(node) <= last) + return node; + } + return NULL; +} - drm_mm_for_each_node(node, mm) { +static struct drm_mm_node * +drm_mm_interval_tree_iter_next(struct drm_mm_node *node, uint64_t start, + uint64_t last) +{ + STUB(); + struct rb_node *rb = &node->rb; + + for (rb = rb_next(rb); rb; rb = rb_next(rb)) { + node = rb_entry(rb, typeof(*node), rb); if (LAST(node) >= start && START(node) <= last) return node; } return NULL; } + +static void +drm_mm_interval_tree_remove(struct drm_mm_node *node, + struct rb_root_cached *root) +{ + rb_erase_cached(&node->rb, root); +} + +static void +drm_mm_interval_tree_insert(struct drm_mm_node *node, + struct rb_root_cached *root) +{ + struct rb_node **iter = &root->rb_root.rb_node; + struct rb_node *parent = NULL; + struct drm_mm_node *iter_node; + + while (*iter) { + parent = *iter; + iter_node = rb_entry(*iter, struct drm_mm_node, rb); + + if (node->start < iter_node->start) + iter = &(*iter)->rb_left; + else + iter = &(*iter)->rb_right; + } + + rb_link_node(&node->rb, parent, iter); + rb_insert_color_cached(&node->rb, root, false); +} #endif struct drm_mm_node * __drm_mm_interval_first(const struct drm_mm *mm, u64 start, u64 last) { - return drm_mm_interval_tree_iter_first((struct rb_root *)&mm->interval_tree, - start, last); + return drm_mm_interval_tree_iter_first((struct rb_root_cached *)&mm->interval_tree, + start, last) ?: (struct drm_mm_node *)&mm->head_node; } EXPORT_SYMBOL(__drm_mm_interval_first); -#ifdef __linux__ static void drm_mm_interval_tree_add_node(struct drm_mm_node *hole_node, struct drm_mm_node *node) { struct drm_mm *mm = hole_node->mm; struct rb_node **link, *rb; struct drm_mm_node *parent; + bool leftmost; node->__subtree_last = LAST(node); - if (hole_node->allocated) { + if (drm_mm_node_allocated(hole_node)) { rb = &hole_node->rb; while (rb) { parent = rb_entry(rb, struct drm_mm_node, rb); @@ -221,9 +248,11 @@ static void drm_mm_interval_tree_add_node(struct drm_mm_node *hole_node, rb = &hole_node->rb; link = &hole_node->rb.rb_right; + leftmost = false; } else { rb = NULL; - link = &mm->interval_tree.rb_node; + link = &mm->interval_tree.rb_root.rb_node; + leftmost = true; } while (*link) { @@ -231,84 +260,192 @@ static void drm_mm_interval_tree_add_node(struct drm_mm_node *hole_node, parent = rb_entry(rb, struct drm_mm_node, rb); if (parent->__subtree_last < node->__subtree_last) parent->__subtree_last = node->__subtree_last; - if (node->start < parent->start) + if (node->start < parent->start) { link = &parent->rb.rb_left; - else + } else { link = &parent->rb.rb_right; + leftmost = false; + } } rb_link_node(&node->rb, rb, link); - rb_insert_augmented(&node->rb, - &mm->interval_tree, - &drm_mm_interval_tree_augment); -} +#ifdef notyet + rb_insert_augmented_cached(&node->rb, &mm->interval_tree, leftmost, + &drm_mm_interval_tree_augment); +#else + rb_insert_color_cached(&node->rb, &mm->interval_tree, leftmost); #endif +} -static void drm_mm_insert_helper(struct drm_mm_node *hole_node, - struct drm_mm_node *node, - u64 size, u64 alignment, - unsigned long color, - u64 range_start, u64 range_end, - enum drm_mm_allocator_flags flags) +#define DRM_RB_INSERT(root, member, expr) do { \ + struct rb_node **link = &root.rb_node, *rb = NULL; \ + u64 x = expr(node); \ + while (*link) { \ + rb = *link; \ + if (x < expr(rb_entry(rb, struct drm_mm_node, member))) \ + link = &rb->rb_left; \ + else \ + link = &rb->rb_right; \ + } \ + rb_link_node(&node->member, rb, link); \ + rb_insert_color(&node->member, &root); \ +} while (0) + +#define HOLE_SIZE(NODE) ((NODE)->hole_size) +#define HOLE_ADDR(NODE) (__drm_mm_hole_node_start(NODE)) + +static u64 rb_to_hole_size(struct rb_node *rb) { - struct drm_mm *mm = hole_node->mm; - u64 hole_start = drm_mm_hole_node_start(hole_node); - u64 hole_end = drm_mm_hole_node_end(hole_node); - u64 adj_start = hole_start; - u64 adj_end = hole_end; + return rb_entry(rb, struct drm_mm_node, rb_hole_size)->hole_size; +} - DRM_MM_BUG_ON(!drm_mm_hole_follows(hole_node) || node->allocated); +static void insert_hole_size(struct rb_root_cached *root, + struct drm_mm_node *node) +{ + struct rb_node **link = &root->rb_root.rb_node, *rb = NULL; + u64 x = node->hole_size; + bool first = true; - if (mm->color_adjust) - mm->color_adjust(hole_node, color, &adj_start, &adj_end); + while (*link) { + rb = *link; + if (x > rb_to_hole_size(rb)) { + link = &rb->rb_left; + } else { + link = &rb->rb_right; + first = false; + } + } - adj_start = max(adj_start, range_start); - adj_end = min(adj_end, range_end); + rb_link_node(&node->rb_hole_size, rb, link); + rb_insert_color_cached(&node->rb_hole_size, root, first); +} - if (flags & DRM_MM_CREATE_TOP) - adj_start = adj_end - size; +static void add_hole(struct drm_mm_node *node) +{ + struct drm_mm *mm = node->mm; - if (alignment) { - u64 rem; + node->hole_size = + __drm_mm_hole_node_end(node) - __drm_mm_hole_node_start(node); + DRM_MM_BUG_ON(!drm_mm_hole_follows(node)); - div64_u64_rem(adj_start, alignment, &rem); - if (rem) { - if (flags & DRM_MM_CREATE_TOP) - adj_start -= rem; - else - adj_start += alignment - rem; + insert_hole_size(&mm->holes_size, node); + DRM_RB_INSERT(mm->holes_addr, rb_hole_addr, HOLE_ADDR); + + list_add(&node->hole_stack, &mm->hole_stack); +} + +static void rm_hole(struct drm_mm_node *node) +{ + DRM_MM_BUG_ON(!drm_mm_hole_follows(node)); + + list_del(&node->hole_stack); + rb_erase_cached(&node->rb_hole_size, &node->mm->holes_size); + rb_erase(&node->rb_hole_addr, &node->mm->holes_addr); + node->hole_size = 0; + + DRM_MM_BUG_ON(drm_mm_hole_follows(node)); +} + +static inline struct drm_mm_node *rb_hole_size_to_node(struct rb_node *rb) +{ + return rb_entry_safe(rb, struct drm_mm_node, rb_hole_size); +} + +static inline struct drm_mm_node *rb_hole_addr_to_node(struct rb_node *rb) +{ + return rb_entry_safe(rb, struct drm_mm_node, rb_hole_addr); +} + +static inline u64 rb_hole_size(struct rb_node *rb) +{ + return rb_entry(rb, struct drm_mm_node, rb_hole_size)->hole_size; +} + +static struct drm_mm_node *best_hole(struct drm_mm *mm, u64 size) +{ + struct rb_node *rb = mm->holes_size.rb_root.rb_node; + struct drm_mm_node *best = NULL; + + do { + struct drm_mm_node *node = + rb_entry(rb, struct drm_mm_node, rb_hole_size); + + if (size <= node->hole_size) { + best = node; + rb = rb->rb_right; + } else { + rb = rb->rb_left; } - } + } while (rb); + + return best; +} + +static struct drm_mm_node *find_hole(struct drm_mm *mm, u64 addr) +{ + struct rb_node *rb = mm->holes_addr.rb_node; + struct drm_mm_node *node = NULL; + + while (rb) { + u64 hole_start; - if (adj_start == hole_start) { - hole_node->hole_follows = 0; - list_del(&hole_node->hole_stack); + node = rb_hole_addr_to_node(rb); + hole_start = __drm_mm_hole_node_start(node); + + if (addr < hole_start) + rb = node->rb_hole_addr.rb_left; + else if (addr > hole_start + node->hole_size) + rb = node->rb_hole_addr.rb_right; + else + break; } - node->start = adj_start; - node->size = size; - node->mm = mm; - node->color = color; - node->allocated = 1; + return node; +} - list_add(&node->node_list, &hole_node->node_list); +static struct drm_mm_node * +first_hole(struct drm_mm *mm, + u64 start, u64 end, u64 size, + enum drm_mm_insert_mode mode) +{ + switch (mode) { + default: + case DRM_MM_INSERT_BEST: + return best_hole(mm, size); -#ifdef __linux__ - drm_mm_interval_tree_add_node(hole_node, node); -#endif + case DRM_MM_INSERT_LOW: + return find_hole(mm, start); - DRM_MM_BUG_ON(node->start < range_start); - DRM_MM_BUG_ON(node->start < adj_start); - DRM_MM_BUG_ON(node->start + node->size > adj_end); - DRM_MM_BUG_ON(node->start + node->size > range_end); + case DRM_MM_INSERT_HIGH: + return find_hole(mm, end); - node->hole_follows = 0; - if (__drm_mm_hole_node_start(node) < hole_end) { - list_add(&node->hole_stack, &mm->hole_stack); - node->hole_follows = 1; + case DRM_MM_INSERT_EVICT: + return list_first_entry_or_null(&mm->hole_stack, + struct drm_mm_node, + hole_stack); } +} - save_stack(node); +static struct drm_mm_node * +next_hole(struct drm_mm *mm, + struct drm_mm_node *node, + enum drm_mm_insert_mode mode) +{ + switch (mode) { + default: + case DRM_MM_INSERT_BEST: + return rb_hole_size_to_node(rb_prev(&node->rb_hole_size)); + + case DRM_MM_INSERT_LOW: + return rb_hole_addr_to_node(rb_next(&node->rb_hole_addr)); + + case DRM_MM_INSERT_HIGH: + return rb_hole_addr_to_node(rb_prev(&node->rb_hole_addr)); + + case DRM_MM_INSERT_EVICT: + node = list_next_entry(node, hole_stack); + return &node->hole_stack == &mm->hole_stack ? NULL : node; + } } /** @@ -327,31 +464,22 @@ static void drm_mm_insert_helper(struct drm_mm_node *hole_node, */ int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node) { - u64 end = node->start + node->size; struct drm_mm_node *hole; u64 hole_start, hole_end; u64 adj_start, adj_end; + u64 end; end = node->start + node->size; if (unlikely(end <= node->start)) return -ENOSPC; /* Find the relevant hole to add our node to */ - hole = drm_mm_interval_tree_iter_first(&mm->interval_tree, - node->start, ~(u64)0); - if (hole) { - if (hole->start < end) - return -ENOSPC; - } else { - hole = list_entry(drm_mm_nodes(mm), typeof(*hole), node_list); - } - - hole = list_last_entry(&hole->node_list, typeof(*hole), node_list); - if (!drm_mm_hole_follows(hole)) + hole = find_hole(mm, node->start); + if (!hole) return -ENOSPC; adj_start = hole_start = __drm_mm_hole_node_start(hole); - adj_end = hole_end = __drm_mm_hole_node_end(hole); + adj_end = hole_end = hole_start + hole->hole_size; if (mm->color_adjust) mm->color_adjust(hole, node->color, &adj_start, &adj_end); @@ -360,72 +488,148 @@ int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node) return -ENOSPC; node->mm = mm; - node->allocated = 1; + __set_bit(DRM_MM_NODE_ALLOCATED_BIT, &node->flags); list_add(&node->node_list, &hole->node_list); - -#ifdef __linux__ drm_mm_interval_tree_add_node(hole, node); -#endif + node->hole_size = 0; - if (node->start == hole_start) { - hole->hole_follows = 0; - list_del(&hole->hole_stack); - } - - node->hole_follows = 0; - if (end != hole_end) { - list_add(&node->hole_stack, &mm->hole_stack); - node->hole_follows = 1; - } + rm_hole(hole); + if (node->start > hole_start) + add_hole(hole); + if (end < hole_end) + add_hole(node); save_stack(node); - return 0; } EXPORT_SYMBOL(drm_mm_reserve_node); +static u64 rb_to_hole_size_or_zero(struct rb_node *rb) +{ + return rb ? rb_to_hole_size(rb) : 0; +} + /** - * drm_mm_insert_node_in_range_generic - ranged search for space and insert @node + * drm_mm_insert_node_in_range - ranged search for space and insert @node * @mm: drm_mm to allocate from * @node: preallocate node to insert * @size: size of the allocation * @alignment: alignment of the allocation * @color: opaque tag value to use for this node - * @start: start of the allowed range for this node - * @end: end of the allowed range for this node - * @sflags: flags to fine-tune the allocation search - * @aflags: flags to fine-tune the allocation behavior + * @range_start: start of the allowed range for this node + * @range_end: end of the allowed range for this node + * @mode: fine-tune the allocation search and placement * * The preallocated @node must be cleared to 0. * * Returns: * 0 on success, -ENOSPC if there's no suitable hole. */ -int drm_mm_insert_node_in_range_generic(struct drm_mm *mm, struct drm_mm_node *node, - u64 size, u64 alignment, - unsigned long color, - u64 start, u64 end, - enum drm_mm_search_flags sflags, - enum drm_mm_allocator_flags aflags) -{ - struct drm_mm_node *hole_node; - - if (WARN_ON(size == 0)) - return -EINVAL; - - hole_node = drm_mm_search_free_in_range_generic(mm, - size, alignment, color, - start, end, sflags); - if (!hole_node) +int drm_mm_insert_node_in_range(struct drm_mm * const mm, + struct drm_mm_node * const node, + u64 size, u64 alignment, + unsigned long color, + u64 range_start, u64 range_end, + enum drm_mm_insert_mode mode) +{ + struct drm_mm_node *hole; + u64 remainder_mask; + bool once; + + DRM_MM_BUG_ON(range_start > range_end); + + if (unlikely(size == 0 || range_end - range_start < size)) return -ENOSPC; - drm_mm_insert_helper(hole_node, node, - size, alignment, color, - start, end, aflags); - return 0; + if (rb_to_hole_size_or_zero(rb_first_cached(&mm->holes_size)) < size) + return -ENOSPC; + + if (alignment <= 1) + alignment = 0; + + once = mode & DRM_MM_INSERT_ONCE; + mode &= ~DRM_MM_INSERT_ONCE; + + remainder_mask = is_power_of_2(alignment) ? alignment - 1 : 0; + for (hole = first_hole(mm, range_start, range_end, size, mode); + hole; + hole = once ? NULL : next_hole(mm, hole, mode)) { + u64 hole_start = __drm_mm_hole_node_start(hole); + u64 hole_end = hole_start + hole->hole_size; + u64 adj_start, adj_end; + u64 col_start, col_end; + + if (mode == DRM_MM_INSERT_LOW && hole_start >= range_end) + break; + + if (mode == DRM_MM_INSERT_HIGH && hole_end <= range_start) + break; + + col_start = hole_start; + col_end = hole_end; + if (mm->color_adjust) + mm->color_adjust(hole, color, &col_start, &col_end); + + adj_start = max(col_start, range_start); + adj_end = min(col_end, range_end); + + if (adj_end <= adj_start || adj_end - adj_start < size) + continue; + + if (mode == DRM_MM_INSERT_HIGH) + adj_start = adj_end - size; + + if (alignment) { + u64 rem; + + if (likely(remainder_mask)) + rem = adj_start & remainder_mask; + else + div64_u64_rem(adj_start, alignment, &rem); + if (rem) { + adj_start -= rem; + if (mode != DRM_MM_INSERT_HIGH) + adj_start += alignment; + + if (adj_start < max(col_start, range_start) || + min(col_end, range_end) - adj_start < size) + continue; + + if (adj_end <= adj_start || + adj_end - adj_start < size) + continue; + } + } + + node->mm = mm; + node->size = size; + node->start = adj_start; + node->color = color; + node->hole_size = 0; + + __set_bit(DRM_MM_NODE_ALLOCATED_BIT, &node->flags); + list_add(&node->node_list, &hole->node_list); + drm_mm_interval_tree_add_node(hole, node); + + rm_hole(hole); + if (adj_start > hole_start) + add_hole(hole); + if (adj_start + size < hole_end) + add_hole(node); + + save_stack(node); + return 0; + } + + return -ENOSPC; +} +EXPORT_SYMBOL(drm_mm_insert_node_in_range); + +static inline bool drm_mm_node_scanned_block(const struct drm_mm_node *node) +{ + return test_bit(DRM_MM_NODE_SCANNED_BIT, &node->flags); } -EXPORT_SYMBOL(drm_mm_insert_node_in_range_generic); /** * drm_mm_remove_node - Remove a memory node from the allocator. @@ -440,97 +644,24 @@ void drm_mm_remove_node(struct drm_mm_node *node) struct drm_mm *mm = node->mm; struct drm_mm_node *prev_node; - DRM_MM_BUG_ON(!node->allocated); - DRM_MM_BUG_ON(node->scanned_block); - - prev_node = - list_entry(node->node_list.prev, struct drm_mm_node, node_list); + DRM_MM_BUG_ON(!drm_mm_node_allocated(node)); + DRM_MM_BUG_ON(drm_mm_node_scanned_block(node)); - if (drm_mm_hole_follows(node)) { - DRM_MM_BUG_ON(__drm_mm_hole_node_start(node) == - __drm_mm_hole_node_end(node)); - list_del(&node->hole_stack); - } else { - DRM_MM_BUG_ON(__drm_mm_hole_node_start(node) != - __drm_mm_hole_node_end(node)); - } + prev_node = list_prev_entry(node, node_list); - if (!drm_mm_hole_follows(prev_node)) { - prev_node->hole_follows = 1; - list_add(&prev_node->hole_stack, &mm->hole_stack); - } else - list_move(&prev_node->hole_stack, &mm->hole_stack); + if (drm_mm_hole_follows(node)) + rm_hole(node); -#ifdef __linux__ drm_mm_interval_tree_remove(node, &mm->interval_tree); -#endif list_del(&node->node_list); - node->allocated = 0; -} -EXPORT_SYMBOL(drm_mm_remove_node); - -static int check_free_hole(u64 start, u64 end, u64 size, u64 alignment) -{ - if (end - start < size) - return 0; - - if (alignment) { - u64 rem; - - div64_u64_rem(start, alignment, &rem); - if (rem) - start += alignment - rem; - } - - return end >= start + size; -} - -static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_mm *mm, - u64 size, - u64 alignment, - unsigned long color, - u64 start, - u64 end, - enum drm_mm_search_flags flags) -{ - struct drm_mm_node *entry; - struct drm_mm_node *best; - u64 adj_start; - u64 adj_end; - u64 best_size; - DRM_MM_BUG_ON(mm->scan_active); - - best = NULL; - best_size = ~0UL; - - __drm_mm_for_each_hole(entry, mm, adj_start, adj_end, - flags & DRM_MM_SEARCH_BELOW) { - u64 hole_size = adj_end - adj_start; - - if (mm->color_adjust) { - mm->color_adjust(entry, color, &adj_start, &adj_end); - if (adj_end <= adj_start) - continue; - } - - adj_start = max(adj_start, start); - adj_end = min(adj_end, end); - - if (!check_free_hole(adj_start, adj_end, size, alignment)) - continue; - - if (!(flags & DRM_MM_SEARCH_BEST)) - return entry; - - if (hole_size < best_size) { - best = entry; - best_size = hole_size; - } - } + if (drm_mm_hole_follows(prev_node)) + rm_hole(prev_node); + add_hole(prev_node); - return best; + clear_bit_unlock(DRM_MM_NODE_ALLOCATED_BIT, &node->flags); } +EXPORT_SYMBOL(drm_mm_remove_node); /** * drm_mm_replace_node - move an allocation from @old to @new @@ -543,22 +674,27 @@ static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_ */ void drm_mm_replace_node(struct drm_mm_node *old, struct drm_mm_node *new) { - DRM_MM_BUG_ON(!old->allocated); + struct drm_mm *mm = old->mm; + DRM_MM_BUG_ON(!drm_mm_node_allocated(old)); + + *new = *old; + + __set_bit(DRM_MM_NODE_ALLOCATED_BIT, &new->flags); list_replace(&old->node_list, &new->node_list); - list_replace(&old->hole_stack, &new->hole_stack); -#ifdef __linux__ - rb_replace_node(&old->rb, &new->rb, &old->mm->interval_tree); -#endif - new->hole_follows = old->hole_follows; - new->mm = old->mm; - new->start = old->start; - new->size = old->size; - new->color = old->color; - new->__subtree_last = old->__subtree_last; + rb_replace_node_cached(&old->rb, &new->rb, &mm->interval_tree); + + if (drm_mm_hole_follows(old)) { + list_replace(&old->hole_stack, &new->hole_stack); + rb_replace_node_cached(&old->rb_hole_size, + &new->rb_hole_size, + &mm->holes_size); + rb_replace_node(&old->rb_hole_addr, + &new->rb_hole_addr, + &mm->holes_addr); + } - old->allocated = 0; - new->allocated = 1; + clear_bit_unlock(DRM_MM_NODE_ALLOCATED_BIT, &old->flags); } EXPORT_SYMBOL(drm_mm_replace_node); @@ -603,7 +739,7 @@ EXPORT_SYMBOL(drm_mm_replace_node); * @color: opaque tag value to use for the allocation * @start: start of the allowed range for the allocation * @end: end of the allowed range for the allocation - * @flags: flags to specify how the allocation will be performed afterwards + * @mode: fine-tune the allocation search and placement * * This simply sets up the scanning routines with the parameters for the desired * hole. @@ -619,7 +755,7 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan, unsigned long color, u64 start, u64 end, - unsigned int flags) + enum drm_mm_insert_mode mode) { DRM_MM_BUG_ON(start >= end); DRM_MM_BUG_ON(!size || size > end - start); @@ -634,7 +770,7 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan, scan->alignment = alignment; scan->remainder_mask = is_power_of_2(alignment) ? alignment - 1 : 0; scan->size = size; - scan->flags = flags; + scan->mode = mode; DRM_MM_BUG_ON(end <= start); scan->range_start = start; @@ -666,9 +802,9 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan, u64 adj_start, adj_end; DRM_MM_BUG_ON(node->mm != mm); - DRM_MM_BUG_ON(!node->allocated); - DRM_MM_BUG_ON(node->scanned_block); - node->scanned_block = true; + DRM_MM_BUG_ON(!drm_mm_node_allocated(node)); + DRM_MM_BUG_ON(drm_mm_node_scanned_block(node)); + __set_bit(DRM_MM_NODE_SCANNED_BIT, &node->flags); mm->scan_active++; /* Remove this block from the node_list so that we enlarge the hole @@ -693,7 +829,7 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan, if (adj_end <= adj_start || adj_end - adj_start < scan->size) return false; - if (scan->flags == DRM_MM_CREATE_TOP) + if (scan->mode == DRM_MM_INSERT_HIGH) adj_start = adj_end - scan->size; if (scan->alignment) { @@ -705,7 +841,7 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan, div64_u64_rem(adj_start, scan->alignment, &rem); if (rem) { adj_start -= rem; - if (scan->flags != DRM_MM_CREATE_TOP) + if (scan->mode != DRM_MM_INSERT_HIGH) adj_start += scan->alignment; if (adj_start < max(col_start, scan->range_start) || min(col_end, scan->range_end) - adj_start < scan->size) @@ -741,7 +877,7 @@ EXPORT_SYMBOL(drm_mm_scan_add_block); * When the scan list is empty, the selected memory nodes can be freed. An * immediately following drm_mm_insert_node_in_range_generic() or one of the * simpler versions of that function with !DRM_MM_SEARCH_BEST will then return - * the just freed block (because its at the top of the free_stack list). + * the just freed block (because it's at the top of the free_stack list). * * Returns: * True if this block should be evicted, false otherwise. Will always @@ -753,8 +889,8 @@ bool drm_mm_scan_remove_block(struct drm_mm_scan *scan, struct drm_mm_node *prev_node; DRM_MM_BUG_ON(node->mm != scan->mm); - DRM_MM_BUG_ON(!node->scanned_block); - node->scanned_block = false; + DRM_MM_BUG_ON(!drm_mm_node_scanned_block(node)); + __clear_bit(DRM_MM_NODE_SCANNED_BIT, &node->flags); DRM_MM_BUG_ON(!node->mm->scan_active); node->mm->scan_active--; @@ -799,9 +935,24 @@ struct drm_mm_node *drm_mm_scan_color_evict(struct drm_mm_scan *scan) if (!mm->color_adjust) return NULL; - hole = list_first_entry(&mm->hole_stack, typeof(*hole), hole_stack); - hole_start = __drm_mm_hole_node_start(hole); - hole_end = __drm_mm_hole_node_end(hole); + /* + * The hole found during scanning should ideally be the first element + * in the hole_stack list, but due to side-effects in the driver it + * may not be. + */ + list_for_each_entry(hole, &mm->hole_stack, hole_stack) { + hole_start = __drm_mm_hole_node_start(hole); + hole_end = hole_start + hole->hole_size; + + if (hole_start <= scan->hit_start && + hole_end >= scan->hit_end) + break; + } + + /* We should only be called after we found the hole previously */ + DRM_MM_BUG_ON(&hole->hole_stack == &mm->hole_stack); + if (unlikely(&hole->hole_stack == &mm->hole_stack)) + return NULL; DRM_MM_BUG_ON(hole_start > scan->hit_start); DRM_MM_BUG_ON(hole_end < scan->hit_end); @@ -828,21 +979,22 @@ void drm_mm_init(struct drm_mm *mm, u64 start, u64 size) { DRM_MM_BUG_ON(start + size <= start); + mm->color_adjust = NULL; + INIT_LIST_HEAD(&mm->hole_stack); - mm->scan_active = 0; + mm->interval_tree = RB_ROOT_CACHED; + mm->holes_size = RB_ROOT_CACHED; + mm->holes_addr = RB_ROOT; /* Clever trick to avoid a special case in the free hole tracking. */ INIT_LIST_HEAD(&mm->head_node.node_list); - mm->head_node.allocated = 0; - mm->head_node.hole_follows = 1; + mm->head_node.flags = 0; mm->head_node.mm = mm; mm->head_node.start = start + size; - mm->head_node.size = start - mm->head_node.start; - list_add_tail(&mm->head_node.hole_stack, &mm->hole_stack); + mm->head_node.size = -size; + add_hole(&mm->head_node); - mm->interval_tree = RB_ROOT; - - mm->color_adjust = NULL; + mm->scan_active = 0; } EXPORT_SYMBOL(drm_mm_init); @@ -863,20 +1015,17 @@ EXPORT_SYMBOL(drm_mm_takedown); static u64 drm_mm_dump_hole(struct drm_printer *p, const struct drm_mm_node *entry) { - u64 hole_start, hole_end, hole_size; + u64 start, size; - if (entry->hole_follows) { - hole_start = drm_mm_hole_node_start(entry); - hole_end = drm_mm_hole_node_end(entry); - hole_size = hole_end - hole_start; - drm_printf(p, "%#018llx-%#018llx: %llu: free\n", hole_start, - hole_end, hole_size); - return hole_size; + size = entry->hole_size; + if (size) { + start = drm_mm_hole_node_start(entry); + drm_printf(p, "%#018llx-%#018llx: %llu: free\n", + start, start + size, size); } - return 0; + return size; } - /** * drm_mm_print - print allocator state * @mm: drm_mm allocator to print diff --git sys/dev/pci/drm/drm_vma_manager.c sys/dev/pci/drm/drm_vma_manager.c index 7ef7ce59c94..9d80892528d 100644 --- sys/dev/pci/drm/drm_vma_manager.c +++ sys/dev/pci/drm/drm_vma_manager.c @@ -1,4 +1,4 @@ -/* $OpenBSD: drm_vma_manager.c,v 1.5 2020/06/08 04:47:58 jsg Exp $ */ +// SPDX-License-Identifier: GPL-2.0 OR MIT /* * Copyright (c) 2006-2009 VMware, Inc., Palo Alto, CA., USA * Copyright (c) 2012 David Airlie <[email protected]> @@ -86,7 +86,6 @@ void drm_vma_offset_manager_init(struct drm_vma_offset_manager *mgr, unsigned long page_offset, unsigned long size) { rw_init(&mgr->vm_lock, "drmvmo"); - mgr->vm_addr_space_rb = RB_ROOT; drm_mm_init(&mgr->vm_addr_space_mm, page_offset, size); } EXPORT_SYMBOL(drm_vma_offset_manager_init); @@ -104,10 +103,7 @@ EXPORT_SYMBOL(drm_vma_offset_manager_init); */ void drm_vma_offset_manager_destroy(struct drm_vma_offset_manager *mgr) { - /* take the lock to protect against buggy drivers */ - write_lock(&mgr->vm_lock); drm_mm_takedown(&mgr->vm_addr_space_mm); - write_unlock(&mgr->vm_lock); } EXPORT_SYMBOL(drm_vma_offset_manager_destroy); @@ -117,27 +113,44 @@ EXPORT_SYMBOL(drm_vma_offset_manager_destroy); * @start: Start address for object (page-based) * @pages: Size of object (page-based) * - * Same as drm_vma_offset_lookup() but requires the caller to lock offset lookup - * manually. See drm_vma_offset_lock_lookup() for an example. + * Find a node given a start address and object size. This returns the _best_ + * match for the given node. That is, @start may point somewhere into a valid + * region and the given node will be returned, as long as the node spans the + * whole requested area (given the size in number of pages as @pages). + * + * Note that before lookup the vma offset manager lookup lock must be acquired + * with drm_vma_offset_lock_lookup(). See there for an example. This can then be + * used to implement weakly referenced lookups using kref_get_unless_zero(). + * + * Example: + * + * :: + * + * drm_vma_offset_lock_lookup(mgr); + * node = drm_vma_offset_lookup_locked(mgr); + * if (node) + * kref_get_unless_zero(container_of(node, sth, entr)); + * drm_vma_offset_unlock_lookup(mgr); * * RETURNS: * Returns NULL if no suitable node can be found. Otherwise, the best match - * is returned. + * is returned. It's the caller's responsibility to make sure the node doesn't + * get destroyed before the caller can access it. */ struct drm_vma_offset_node *drm_vma_offset_lookup_locked(struct drm_vma_offset_manager *mgr, unsigned long start, unsigned long pages) { - struct drm_vma_offset_node *node, *best; + struct drm_mm_node *node, *best; struct rb_node *iter; unsigned long offset; - iter = mgr->vm_addr_space_rb.rb_node; + iter = mgr->vm_addr_space_mm.interval_tree.rb_root.rb_node; best = NULL; while (likely(iter)) { - node = rb_entry(iter, struct drm_vma_offset_node, vm_rb); - offset = node->vm_node.start; + node = rb_entry(iter, struct drm_mm_node, rb); + offset = node->start; if (start >= offset) { iter = iter->rb_right; best = node; @@ -150,38 +163,17 @@ struct drm_vma_offset_node *drm_vma_offset_lookup_locked(struct drm_vma_offset_m /* verify that the node spans the requested area */ if (best) { - offset = best->vm_node.start + best->vm_node.size; + offset = best->start + best->size; if (offset < start + pages) best = NULL; } - return best; -} -EXPORT_SYMBOL(drm_vma_offset_lookup_locked); - -/* internal helper to link @node into the rb-tree */ -static void _drm_vma_offset_add_rb(struct drm_vma_offset_manager *mgr, - struct drm_vma_offset_node *node) -{ - struct rb_node **iter = &mgr->vm_addr_space_rb.rb_node; - struct rb_node *parent = NULL; - struct drm_vma_offset_node *iter_node; - - while (likely(*iter)) { - parent = *iter; - iter_node = rb_entry(*iter, struct drm_vma_offset_node, vm_rb); - - if (node->vm_node.start < iter_node->vm_node.start) - iter = &(*iter)->rb_left; - else if (node->vm_node.start > iter_node->vm_node.start) - iter = &(*iter)->rb_right; - else - BUG(); - } + if (!best) + return NULL; - rb_link_node(&node->vm_rb, parent, iter); - rb_insert_color(&node->vm_rb, &mgr->vm_addr_space_rb); + return container_of(best, struct drm_vma_offset_node, vm_node); } +EXPORT_SYMBOL(drm_vma_offset_lookup_locked); /** * drm_vma_offset_add() - Add offset node to manager @@ -209,24 +201,16 @@ static void _drm_vma_offset_add_rb(struct drm_vma_offset_manager *mgr, int drm_vma_offset_add(struct drm_vma_offset_manager *mgr, struct drm_vma_offset_node *node, unsigned long pages) { - int ret; + int ret = 0; write_lock(&mgr->vm_lock); - if (drm_mm_node_allocated(&node->vm_node)) { - ret = 0; - goto out_unlock; - } - - ret = drm_mm_insert_node(&mgr->vm_addr_space_mm, &node->vm_node, - pages, 0, DRM_MM_SEARCH_DEFAULT); - if (ret) - goto out_unlock; - - _drm_vma_offset_add_rb(mgr, node); + if (!drm_mm_node_allocated(&node->vm_node)) + ret = drm_mm_insert_node(&mgr->vm_addr_space_mm, + &node->vm_node, pages); -out_unlock: write_unlock(&mgr->vm_lock); + return ret; } EXPORT_SYMBOL(drm_vma_offset_add); @@ -248,7 +232,6 @@ void drm_vma_offset_remove(struct drm_vma_offset_manager *mgr, write_lock(&mgr->vm_lock); if (drm_mm_node_allocated(&node->vm_node)) { - rb_erase(&node->vm_rb, &mgr->vm_addr_space_rb); drm_mm_remove_node(&node->vm_node); memset(&node->vm_node, 0, sizeof(node->vm_node)); } @@ -260,9 +243,9 @@ EXPORT_SYMBOL(drm_vma_offset_remove); /** * drm_vma_node_allow - Add open-file to list of allowed users * @node: Node to modify - * @filp: Open file to add + * @tag: Tag of file to remove * - * Add @filp to the list of allowed open-files for this node. If @filp is + * Add @tag to the list of allowed open-files for this node. If @tag is * already on this list, the ref-count is incremented. * * The list of allowed-users is preserved across drm_vma_offset_add() and @@ -277,7 +260,7 @@ EXPORT_SYMBOL(drm_vma_offset_remove); * RETURNS: * 0 on success, negative error code on internal failure (out-of-mem) */ -int drm_vma_node_allow(struct drm_vma_offset_node *node, struct file *filp) +int drm_vma_node_allow(struct drm_vma_offset_node *node, struct file *tag) { struct rb_node **iter; struct rb_node *parent = NULL; @@ -298,10 +281,10 @@ int drm_vma_node_allow(struct drm_vma_offset_node *node, struct file *filp) parent = *iter; entry = rb_entry(*iter, struct drm_vma_offset_file, vm_rb); - if (filp == entry->vm_filp) { + if (tag == entry->vm_tag) { entry->vm_count++; goto unlock; - } else if (filp > entry->vm_filp) { + } else if (tag > entry->vm_tag) { iter = &(*iter)->rb_right; } else { iter = &(*iter)->rb_left; @@ -313,7 +296,7 @@ int drm_vma_node_allow(struct drm_vma_offset_node *node, struct file *filp) goto unlock; } - new->vm_filp = filp; + new->vm_tag = tag; new->vm_count = 1; rb_link_node(&new->vm_rb, parent, iter); rb_insert_color(&new->vm_rb, &node->vm_files); @@ -329,17 +312,18 @@ EXPORT_SYMBOL(drm_vma_node_allow); /** * drm_vma_node_revoke - Remove open-file from list of allowed users * @node: Node to modify - * @filp: Open file to remove + * @tag: Tag of file to remove * - * Decrement the ref-count of @filp in the list of allowed open-files on @node. - * If the ref-count drops to zero, remove @filp from the list. You must call - * this once for every drm_vma_node_allow() on @filp. + * Decrement the ref-count of @tag in the list of allowed open-files on @node. + * If the ref-count drops to zero, remove @tag from the list. You must call + * this once for every drm_vma_node_allow() on @tag. * * This is locked against concurrent access internally. * - * If @filp is not on the list, nothing is done. + * If @tag is not on the list, nothing is done. */ -void drm_vma_node_revoke(struct drm_vma_offset_node *node, struct file *filp) +void drm_vma_node_revoke(struct drm_vma_offset_node *node, + struct file *tag) { struct drm_vma_offset_file *entry; struct rb_node *iter; @@ -349,13 +333,13 @@ void drm_vma_node_revoke(struct drm_vma_offset_node *node, struct file *filp) iter = node->vm_files.rb_node; while (likely(iter)) { entry = rb_entry(iter, struct drm_vma_offset_file, vm_rb); - if (filp == entry->vm_filp) { + if (tag == entry->vm_tag) { if (!--entry->vm_count) { rb_erase(&entry->vm_rb, &node->vm_files); kfree(entry); } break; - } else if (filp > entry->vm_filp) { + } else if (tag > entry->vm_tag) { iter = iter->rb_right; } else { iter = iter->rb_left; @@ -369,9 +353,9 @@ EXPORT_SYMBOL(drm_vma_node_revoke); /** * drm_vma_node_is_allowed - Check whether an open-file is granted access * @node: Node to check - * @filp: Open-file to check for + * @tag: Tag of file to remove * - * Search the list in @node whether @filp is currently on the list of allowed + * Search the list in @node whether @tag is currently on the list of allowed * open-files (see drm_vma_node_allow()). * * This is locked against concurrent access internally. @@ -380,7 +364,7 @@ EXPORT_SYMBOL(drm_vma_node_revoke); * true iff @filp is on the list */ bool drm_vma_node_is_allowed(struct drm_vma_offset_node *node, - struct file *filp) + struct file *tag) { struct drm_vma_offset_file *entry; struct rb_node *iter; @@ -390,9 +374,9 @@ bool drm_vma_node_is_allowed(struct drm_vma_offset_node *node, iter = node->vm_files.rb_node; while (likely(iter)) { entry = rb_entry(iter, struct drm_vma_offset_file, vm_rb); - if (filp == entry->vm_filp) + if (tag == entry->vm_tag) break; - else if (filp > entry->vm_filp) + else if (tag > entry->vm_tag) iter = iter->rb_right; else iter = iter->rb_left; diff --git sys/dev/pci/drm/i915/gem/i915_gem_execbuffer.c sys/dev/pci/drm/i915/gem/i915_gem_execbuffer.c index b6bf81f647b..971ed84f371 100644 --- sys/dev/pci/drm/i915/gem/i915_gem_execbuffer.c +++ sys/dev/pci/drm/i915/gem/i915_gem_execbuffer.c @@ -877,13 +877,7 @@ static void reloc_cache_init(struct reloc_cache *cache, cache->use_64bit_reloc = HAS_64BIT_RELOC(i915); cache->has_fence = cache->gen < 4; cache->needs_unfenced = INTEL_INFO(i915)->unfenced_needs_alignment; -#ifdef notyet cache->node.flags = 0; -#else - cache->node.hole_follows = 0; - cache->node.allocated = 0; - cache->node.scanned_block = 0; -#endif cache->rq = NULL; cache->rq_size = 0; diff --git sys/dev/pci/drm/i915/i915_gem.c sys/dev/pci/drm/i915/i915_gem.c index 92ed27c93ea..5857612aa3c 100644 --- sys/dev/pci/drm/i915/i915_gem.c +++ sys/dev/pci/drm/i915/i915_gem.c @@ -438,13 +438,7 @@ i915_gem_gtt_pread(struct drm_i915_gem_object *obj, PIN_NOEVICT); if (!IS_ERR(vma)) { node.start = i915_ggtt_offset(vma); -#ifdef notyet node.flags = 0; -#else - node.hole_follows = 0; - node.allocated = 0; - node.scanned_block = 0; -#endif } else { ret = insert_mappable_node(ggtt, &node, PAGE_SIZE); if (ret) @@ -670,13 +664,7 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_gem_object *obj, PIN_NOEVICT); if (!IS_ERR(vma)) { node.start = i915_ggtt_offset(vma); -#ifdef notyet node.flags = 0; -#else - node.hole_follows = 0; - node.allocated = 0; - node.scanned_block = 0; -#endif } else { ret = insert_mappable_node(ggtt, &node, PAGE_SIZE); if (ret) diff --git sys/dev/pci/drm/include/drm/drm_mm.h sys/dev/pci/drm/include/drm/drm_mm.h index 63a94d20f41..ee8b0e80ca9 100644 --- sys/dev/pci/drm/include/drm/drm_mm.h +++ sys/dev/pci/drm/include/drm/drm_mm.h @@ -40,6 +40,7 @@ #include <linux/bug.h> #include <linux/rbtree.h> #include <linux/kernel.h> +#include <linux/mm_types.h> #include <linux/list.h> #include <linux/spinlock.h> #ifdef CONFIG_DRM_DEBUG_MM @@ -53,20 +54,6 @@ #define DRM_MM_BUG_ON(expr) BUILD_BUG_ON_INVALID(expr) #endif -enum drm_mm_search_flags { - DRM_MM_SEARCH_DEFAULT = 0, - DRM_MM_SEARCH_BEST = 1 << 0, - DRM_MM_SEARCH_BELOW = 1 << 1, -}; - -enum drm_mm_allocator_flags { - DRM_MM_CREATE_DEFAULT = 0, - DRM_MM_CREATE_TOP = 1 << 0, -}; - -#define DRM_MM_BOTTOMUP DRM_MM_SEARCH_DEFAULT, DRM_MM_CREATE_DEFAULT -#define DRM_MM_TOPDOWN DRM_MM_SEARCH_BELOW, DRM_MM_CREATE_TOP - /** * enum drm_mm_insert_mode - control search and allocation behaviour * @@ -173,14 +160,17 @@ struct drm_mm_node { /** @size: Size of the allocated block. */ u64 size; /* private: */ + struct drm_mm *mm; struct list_head node_list; struct list_head hole_stack; struct rb_node rb; - unsigned hole_follows : 1; - unsigned allocated : 1; - bool scanned_block : 1; + struct rb_node rb_hole_size; + struct rb_node rb_hole_addr; u64 __subtree_last; - struct drm_mm *mm; + u64 hole_size; + unsigned long flags; +#define DRM_MM_NODE_ALLOCATED_BIT 0 +#define DRM_MM_NODE_SCANNED_BIT 1 #ifdef CONFIG_DRM_DEBUG_MM depot_stack_handle_t stack; #endif @@ -215,7 +205,9 @@ struct drm_mm { * according to the (increasing) start address of the memory node. */ struct drm_mm_node head_node; /* Keep an interval_tree for fast lookup of drm_mm_nodes by address. */ - struct rb_root interval_tree; + struct rb_root_cached interval_tree; + struct rb_root_cached holes_size; + struct rb_root holes_addr; unsigned long scan_active; }; @@ -244,7 +236,7 @@ struct drm_mm_scan { u64 hit_end; unsigned long color; - unsigned int flags; + enum drm_mm_insert_mode mode; }; /** @@ -262,7 +254,7 @@ struct drm_mm_scan { */ static inline bool drm_mm_node_allocated(const struct drm_mm_node *node) { - return node->allocated; + return test_bit(DRM_MM_NODE_ALLOCATED_BIT, &node->flags); } /** @@ -280,7 +272,7 @@ static inline bool drm_mm_node_allocated(const struct drm_mm_node *node) */ static inline bool drm_mm_initialized(const struct drm_mm *mm) { - return mm->hole_stack.next; + return READ_ONCE(mm->hole_stack.next); } /** @@ -297,7 +289,7 @@ static inline bool drm_mm_initialized(const struct drm_mm *mm) */ static inline bool drm_mm_hole_follows(const struct drm_mm_node *node) { - return node->hole_follows; + return node->hole_size; } static inline u64 __drm_mm_hole_node_start(const struct drm_mm_node *hole_node) @@ -380,17 +372,9 @@ static inline u64 drm_mm_hole_node_end(const struct drm_mm_node *hole_node) #define drm_mm_for_each_node_safe(entry, next, mm) \ list_for_each_entry_safe(entry, next, drm_mm_nodes(mm), node_list) -#define __drm_mm_for_each_hole(entry, mm, hole_start, hole_end, backwards) \ - for (entry = list_entry((backwards) ? (mm)->hole_stack.prev : (mm)->hole_stack.next, struct drm_mm_node, hole_stack); \ - &entry->hole_stack != &(mm)->hole_stack ? \ - hole_start = drm_mm_hole_node_start(entry), \ - hole_end = drm_mm_hole_node_end(entry), \ - 1 : 0; \ - entry = list_entry((backwards) ? entry->hole_stack.prev : entry->hole_stack.next, struct drm_mm_node, hole_stack)) - /** * drm_mm_for_each_hole - iterator to walk over all holes - * @entry: &drm_mm_node used internally to track progress + * @pos: &drm_mm_node used internally to track progress * @mm: &drm_mm allocator to walk * @hole_start: ulong variable to assign the hole start to on each iteration * @hole_end: ulong variable to assign the hole end to on each iteration @@ -403,79 +387,28 @@ static inline u64 drm_mm_hole_node_end(const struct drm_mm_node *hole_node) * Implementation Note: * We need to inline list_for_each_entry in order to be able to set hole_start * and hole_end on each iteration while keeping the macro sane. - * - * The __drm_mm_for_each_hole version is similar, but with added support for - * going backwards. */ -#define drm_mm_for_each_hole(entry, mm, hole_start, hole_end) \ - __drm_mm_for_each_hole(entry, mm, hole_start, hole_end, 0) +#define drm_mm_for_each_hole(pos, mm, hole_start, hole_end) \ + for (pos = list_first_entry(&(mm)->hole_stack, \ + typeof(*pos), hole_stack); \ + &pos->hole_stack != &(mm)->hole_stack ? \ + hole_start = drm_mm_hole_node_start(pos), \ + hole_end = hole_start + pos->hole_size, \ + 1 : 0; \ + pos = list_next_entry(pos, hole_stack)) /* * Basic range manager support (drm_mm.c) */ int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node); -int drm_mm_insert_node_in_range_generic(struct drm_mm *mm, - struct drm_mm_node *node, - u64 size, - u64 alignment, - unsigned long color, - u64 start, - u64 end, - enum drm_mm_search_flags sflags, - enum drm_mm_allocator_flags aflags); - -/** - * drm_mm_insert_node_in_range - ranged search for space and insert @node - * @mm: drm_mm to allocate from - * @node: preallocate node to insert - * @size: size of the allocation - * @alignment: alignment of the allocation - * @start: start of the allowed range for this node - * @end: end of the allowed range for this node - * @flags: flags to fine-tune the allocation - * - * This is a simplified version of drm_mm_insert_node_in_range_generic() with - * @color set to 0. - * - * The preallocated node must be cleared to 0. - * - * Returns: - * 0 on success, -ENOSPC if there's no suitable hole. - */ -static inline int drm_mm_insert_node_in_range(struct drm_mm *mm, - struct drm_mm_node *node, - u64 size, - u64 alignment, - unsigned long color, - u64 start, - u64 end, - enum drm_mm_insert_mode mode) -{ - enum drm_mm_search_flags sflags; - enum drm_mm_allocator_flags aflags; - switch (mode) { - case DRM_MM_INSERT_HIGHEST: - sflags = DRM_MM_SEARCH_BELOW; - aflags = DRM_MM_CREATE_TOP; - break; - case DRM_MM_INSERT_BEST: - sflags = DRM_MM_SEARCH_BEST; - aflags = DRM_MM_CREATE_DEFAULT; - break; - case DRM_MM_INSERT_LOW: - case DRM_MM_INSERT_HIGH: - case DRM_MM_INSERT_EVICT: - case DRM_MM_INSERT_ONCE: - case DRM_MM_INSERT_LOWEST: - default: - sflags = DRM_MM_SEARCH_DEFAULT; - aflags = DRM_MM_CREATE_DEFAULT; - break; - } - return drm_mm_insert_node_in_range_generic(mm, node, size, alignment, - color, start, end, - sflags, aflags); -} +int drm_mm_insert_node_in_range(struct drm_mm *mm, + struct drm_mm_node *node, + u64 size, + u64 alignment, + unsigned long color, + u64 start, + u64 end, + enum drm_mm_insert_mode mode); /** * drm_mm_insert_node_generic - search for space and insert @node @@ -484,10 +417,9 @@ static inline int drm_mm_insert_node_in_range(struct drm_mm *mm, * @size: size of the allocation * @alignment: alignment of the allocation * @color: opaque tag value to use for this node - * @sflags: flags to fine-tune the allocation search - * @aflags: flags to fine-tune the allocation behavior + * @mode: fine-tune the allocation search and placement * - * This is a simplified version of drm_mm_insert_node_in_range_generic() with no + * This is a simplified version of drm_mm_insert_node_in_range() with no * range restrictions applied. * * The preallocated node must be cleared to 0. @@ -499,13 +431,11 @@ static inline int drm_mm_insert_node_generic(struct drm_mm *mm, struct drm_mm_node *node, u64 size, u64 alignment, unsigned long color, - enum drm_mm_search_flags sflags, - enum drm_mm_allocator_flags aflags) + enum drm_mm_insert_mode mode) { - return drm_mm_insert_node_in_range_generic(mm, node, - size, alignment, 0, - 0, U64_MAX, - sflags, aflags); + return drm_mm_insert_node_in_range(mm, node, + size, alignment, color, + 0, U64_MAX, mode); } /** @@ -513,8 +443,6 @@ drm_mm_insert_node_generic(struct drm_mm *mm, struct drm_mm_node *node, * @mm: drm_mm to allocate from * @node: preallocate node to insert * @size: size of the allocation - * @alignment: alignment of the allocation - * @flags: flags to fine-tune the allocation * * This is a simplified version of drm_mm_insert_node_generic() with @color set * to 0. @@ -526,13 +454,9 @@ drm_mm_insert_node_generic(struct drm_mm *mm, struct drm_mm_node *node, */ static inline int drm_mm_insert_node(struct drm_mm *mm, struct drm_mm_node *node, - u64 size, - u64 alignment, - enum drm_mm_search_flags flags) + u64 size) { - return drm_mm_insert_node_generic(mm, node, - size, alignment, 0, - flags, DRM_MM_CREATE_DEFAULT); + return drm_mm_insert_node_generic(mm, node, size, 0, 0, 0); } void drm_mm_remove_node(struct drm_mm_node *node); @@ -569,17 +493,20 @@ __drm_mm_interval_first(const struct drm_mm *mm, u64 start, u64 last); * but using the internal interval tree to accelerate the search for the * starting node, and so not safe against removal of elements. It assumes * that @end is within (or is the upper limit of) the drm_mm allocator. + * If [@start, @end] are beyond the range of the drm_mm, the iterator may walk + * over the special _unallocated_ &drm_mm.head_node, and may even continue + * indefinitely. */ #define drm_mm_for_each_node_in_range(node__, mm__, start__, end__) \ for (node__ = __drm_mm_interval_first((mm__), (start__), (end__)-1); \ - node__ && node__->start < (end__); \ + node__->start < (end__); \ node__ = list_next_entry(node__, node_list)) void drm_mm_scan_init_with_range(struct drm_mm_scan *scan, struct drm_mm *mm, u64 size, u64 alignment, unsigned long color, u64 start, u64 end, - unsigned int flags); + enum drm_mm_insert_mode mode); /** * drm_mm_scan_init - initialize lru scanning @@ -588,7 +515,7 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan, * @size: size of the allocation * @alignment: alignment of the allocation * @color: opaque tag value to use for the allocation - * @flags: flags to specify how the allocation will be performed afterwards + * @mode: fine-tune the allocation search and placement * * This is a simplified version of drm_mm_scan_init_with_range() with no range * restrictions applied. @@ -605,12 +532,11 @@ static inline void drm_mm_scan_init(struct drm_mm_scan *scan, u64 size, u64 alignment, unsigned long color, - unsigned int flags) + enum drm_mm_insert_mode mode) { drm_mm_scan_init_with_range(scan, mm, size, alignment, color, - 0, U64_MAX, - flags); + 0, U64_MAX, mode); } bool drm_mm_scan_add_block(struct drm_mm_scan *scan, diff --git sys/dev/pci/drm/include/drm/drm_vma_manager.h sys/dev/pci/drm/include/drm/drm_vma_manager.h index 0a1812412b4..c0b83478bdc 100644 --- sys/dev/pci/drm/include/drm/drm_vma_manager.h +++ sys/dev/pci/drm/include/drm/drm_vma_manager.h @@ -1,4 +1,3 @@ -/* $OpenBSD: drm_vma_manager.h,v 1.3 2020/06/08 04:48:14 jsg Exp $ */ #ifndef __DRM_VMA_MANAGER_H__ #define __DRM_VMA_MANAGER_H__ @@ -46,21 +45,19 @@ struct drm_file; struct drm_vma_offset_file { struct rb_node vm_rb; - struct file *vm_filp; + struct file *vm_tag; unsigned long vm_count; }; struct drm_vma_offset_node { rwlock_t vm_lock; struct drm_mm_node vm_node; - struct rb_node vm_rb; struct rb_root vm_files; bool readonly:1; }; struct drm_vma_offset_manager { rwlock_t vm_lock; - struct rb_root vm_addr_space_rb; struct drm_mm vm_addr_space_mm; }; @@ -76,10 +73,11 @@ int drm_vma_offset_add(struct drm_vma_offset_manager *mgr, void drm_vma_offset_remove(struct drm_vma_offset_manager *mgr, struct drm_vma_offset_node *node); -int drm_vma_node_allow(struct drm_vma_offset_node *node, struct file *filp); -void drm_vma_node_revoke(struct drm_vma_offset_node *node, struct file *filp); +int drm_vma_node_allow(struct drm_vma_offset_node *node, struct file *tag); +void drm_vma_node_revoke(struct drm_vma_offset_node *node, + struct file *tag); bool drm_vma_node_is_allowed(struct drm_vma_offset_node *node, - struct file *filp); + struct file *tag); /** * drm_vma_offset_exact_lookup_locked() - Look up node by exact address @@ -206,7 +204,6 @@ static inline __u64 drm_vma_node_offset_addr(struct drm_vma_offset_node *node) return ((__u64)node->vm_node.start) << PAGE_SHIFT; } -#ifdef __linux__ /** * drm_vma_node_unmap() - Unmap offset node * @node: Offset node @@ -219,6 +216,7 @@ static inline __u64 drm_vma_node_offset_addr(struct drm_vma_offset_node *node) * This call is unlocked. The caller must guarantee that drm_vma_offset_remove() * is not called on this node concurrently. */ +#ifdef __linux__ static inline void drm_vma_node_unmap(struct drm_vma_offset_node *node, struct address_space *file_mapping) { @@ -232,9 +230,9 @@ static inline void drm_vma_node_unmap(struct drm_vma_offset_node *node, /** * drm_vma_node_verify_access() - Access verification helper for TTM * @node: Offset node - * @filp: Open-file + * @tag: Tag of file to check * - * This checks whether @filp is granted access to @node. It is the same as + * This checks whether @tag is granted access to @node. It is the same as * drm_vma_node_is_allowed() but suitable as drop-in helper for TTM * verify_access() callbacks. * @@ -242,9 +240,9 @@ static inline void drm_vma_node_unmap(struct drm_vma_offset_node *node, * 0 if access is granted, -EACCES otherwise. */ static inline int drm_vma_node_verify_access(struct drm_vma_offset_node *node, - struct file *filp) + struct file *tag) { - return drm_vma_node_is_allowed(node, filp) ? 0 : -EACCES; + return drm_vma_node_is_allowed(node, tag) ? 0 : -EACCES; } #endif /* __DRM_VMA_MANAGER_H__ */ diff --git sys/dev/pci/drm/include/linux/rbtree.h sys/dev/pci/drm/include/linux/rbtree.h index fe45a4eeeb1..1e66b0cca8d 100644 --- sys/dev/pci/drm/include/linux/rbtree.h +++ sys/dev/pci/drm/include/linux/rbtree.h @@ -89,6 +89,8 @@ RB_PROTOTYPE(linux_root, rb_node, __entry, panic_cmp); #define rb_erase_cached(node, root) \ linux_root_RB_REMOVE((struct linux_root *)(&(root)->rb_root), (node)) #define rb_first_cached(root) RB_MIN(linux_root, (struct linux_root *)(&(root)->rb_root)) +#define rb_replace_node_cached(old, new, root) \ + rb_replace_node(old, new, &(root)->rb_root) static inline struct rb_node * __rb_deepest_left(struct rb_node *node)
