[v4,1/4] drm: add unified vma offset manager

Commit Message

David Herrmann July 23, 2013, 12:47 p.m. UTC

If we want to map GPU memory into user-space, we need to linearize the
addresses to not confuse mm-core. Currently, GEM and TTM both implement
their own offset-managers to assign a pgoff to each object for user-space
CPU access. GEM uses a hash-table, TTM uses an rbtree.

This patch provides a unified implementation that can be used to replace
both. TTM allows partial mmaps with a given offset, so we cannot use
hashtables as the start address may not be known at mmap time. Hence, we
use the rbtree-implementation of TTM.

We could easily update drm_mm to use an rbtree instead of a linked list
for it's object list and thus drop the rbtree from the vma-manager.
However, this would slow down drm_mm object allocation for all other
use-cases (rbtree insertion) and add another 4-8 bytes to each mm node.
Hence, use the separate tree but allow for later migration.

This is a rewrite of the 2012-proposal by David Airlie <[e-mail]>

v2:
 - fix Docbook integration
 - drop drm_mm_node_linked() and use drm_mm_node_allocated()
 - remove unjustified likely/unlikely usage (but keep for rbtree paths)
 - remove BUG_ON() as drm_mm already does that
 - clarify page-based vs. byte-based addresses
 - use drm_vma_node_reset() for initialization, too
v4:
 - allow external locking via drm_vma_offset_un/lock_lookup()
 - add locked lookup helper drm_vma_offset_lookup_locked()

Signed-off-by: David Herrmann <[e-mail]>
---
 Documentation/DocBook/drm.tmpl    |   6 +
 drivers/gpu/drm/Makefile          |   2 +-
 drivers/gpu/drm/drm_vma_manager.c | 283 ++++++++++++++++++++++++++++++++++++++
 include/drm/drm_vma_manager.h     | 180 ++++++++++++++++++++++++
 4 files changed, 470 insertions(+), 1 deletion(-)
 create mode 100644 drivers/gpu/drm/drm_vma_manager.c
 create mode 100644 include/drm/drm_vma_manager.h

Comments

Daniel Vetter July 24, 2013, 3:35 p.m. UTC | #1

On Tue, Jul 23, 2013 at 02:47:13PM +0200, David Herrmann wrote:
> If we want to map GPU memory into user-space, we need to linearize the
> addresses to not confuse mm-core. Currently, GEM and TTM both implement
> their own offset-managers to assign a pgoff to each object for user-space
> CPU access. GEM uses a hash-table, TTM uses an rbtree.
> 
> This patch provides a unified implementation that can be used to replace
> both. TTM allows partial mmaps with a given offset, so we cannot use
> hashtables as the start address may not be known at mmap time. Hence, we
> use the rbtree-implementation of TTM.
> 
> We could easily update drm_mm to use an rbtree instead of a linked list
> for it's object list and thus drop the rbtree from the vma-manager.
> However, this would slow down drm_mm object allocation for all other
> use-cases (rbtree insertion) and add another 4-8 bytes to each mm node.
> Hence, use the separate tree but allow for later migration.
> 
> This is a rewrite of the 2012-proposal by David Airlie <[e-mail]>
> 
> v2:
>  - fix Docbook integration
>  - drop drm_mm_node_linked() and use drm_mm_node_allocated()
>  - remove unjustified likely/unlikely usage (but keep for rbtree paths)
>  - remove BUG_ON() as drm_mm already does that
>  - clarify page-based vs. byte-based addresses
>  - use drm_vma_node_reset() for initialization, too
> v4:
>  - allow external locking via drm_vma_offset_un/lock_lookup()
>  - add locked lookup helper drm_vma_offset_lookup_locked()
> 
> Signed-off-by: David Herrmann <[e-mail]>
> ---
>  Documentation/DocBook/drm.tmpl    |   6 +
>  drivers/gpu/drm/Makefile          |   2 +-
>  drivers/gpu/drm/drm_vma_manager.c | 283 ++++++++++++++++++++++++++++++++++++++
>  include/drm/drm_vma_manager.h     | 180 ++++++++++++++++++++++++
>  4 files changed, 470 insertions(+), 1 deletion(-)
>  create mode 100644 drivers/gpu/drm/drm_vma_manager.c
>  create mode 100644 include/drm/drm_vma_manager.h
> 
> diff --git a/Documentation/DocBook/drm.tmpl b/Documentation/DocBook/drm.tmpl
> index 7d1278e..87e22ec 100644
> --- a/Documentation/DocBook/drm.tmpl
> +++ b/Documentation/DocBook/drm.tmpl
> @@ -2212,6 +2212,12 @@ void intel_crt_init(struct drm_device *dev)
>  !Iinclude/drm/drm_rect.h
>  !Edrivers/gpu/drm/drm_rect.c
>      </sect2>
> +    <sect2>
> +      <title>VMA Offset Manager</title>
> +!Pdrivers/gpu/drm/drm_vma_manager.c vma offset manager
> +!Edrivers/gpu/drm/drm_vma_manager.c
> +!Iinclude/drm/drm_vma_manager.h
> +    </sect2>
>    </sect1>
>  
>    <!-- Internals: kms properties -->
> diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile
> index 801bcaf..d943b94 100644
> --- a/drivers/gpu/drm/Makefile
> +++ b/drivers/gpu/drm/Makefile
> @@ -13,7 +13,7 @@ drm-y       :=	drm_auth.o drm_buffer.o drm_bufs.o drm_cache.o \
>  		drm_crtc.o drm_modes.o drm_edid.o \
>  		drm_info.o drm_debugfs.o drm_encoder_slave.o \
>  		drm_trace_points.o drm_global.o drm_prime.o \
> -		drm_rect.o
> +		drm_rect.o drm_vma_manager.o
>  
>  drm-$(CONFIG_COMPAT) += drm_ioc32.o
>  drm-$(CONFIG_DRM_GEM_CMA_HELPER) += drm_gem_cma_helper.o
> diff --git a/drivers/gpu/drm/drm_vma_manager.c b/drivers/gpu/drm/drm_vma_manager.c
> new file mode 100644
> index 0000000..27760bd
> --- /dev/null
> +++ b/drivers/gpu/drm/drm_vma_manager.c
> @@ -0,0 +1,283 @@
> +/*
> + * Copyright (c) 2006-2009 VMware, Inc., Palo Alto, CA., USA
> + * Copyright (c) 2012 David Airlie <[e-mail]>
> + * Copyright (c) 2013 David Herrmann <[e-mail]>
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a
> + * copy of this software and associated documentation files (the "Software"),
> + * to deal in the Software without restriction, including without limitation
> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
> + * and/or sell copies of the Software, and to permit persons to whom the
> + * Software is furnished to do so, subject to the following conditions:
> + *
> + * The above copyright notice and this permission notice shall be included in
> + * all copies or substantial portions of the Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
> + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
> + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
> + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
> + * OTHER DEALINGS IN THE SOFTWARE.
> + */
> +
> +#include <drm/drmP.h>
> +#include <drm/drm_mm.h>
> +#include <drm/drm_vma_manager.h>
> +#include <linux/mm.h>
> +#include <linux/module.h>
> +#include <linux/rbtree.h>
> +#include <linux/slab.h>
> +#include <linux/spinlock.h>
> +#include <linux/types.h>
> +
> +/**
> + * DOC: vma offset manager
> + *
> + * The vma-manager is responsible to map arbitrary driver-dependent memory
> + * regions into the linear user address-space. It provides offsets to the
> + * caller which can then be used on the address_space of the drm-device. It
> + * takes care to not overlap regions, size them appropriately and to not
> + * confuse mm-core by inconsistent fake vm_pgoff fields.
> + * Drivers shouldn't use this for object placement in VMEM. This manager should
> + * only be used to manage mappings into linear user-space VMs.
> + *
> + * We use drm_mm as backend to manage object allocations. But it is highly
> + * optimized for alloc/free calls, not lookups. Hence, we use an rb-tree to
> + * speed up offset lookups.
> + *
> + * You must not use multiple offset managers on a single address_space.
> + * Otherwise, mm-core will be unable to tear down memory mappings as the VM will
> + * no longer be linear. Please use VM_NONLINEAR in that case and implement your
> + * own offset managers.
> + *
> + * This offset manager works on page-based addresses. That is, every argument
> + * and return code (with the exception of drm_vma_node_offset_addr()) is given
> + * in number of pages, not number of bytes. That means, object sizes and offsets
> + * must always be page-aligned (as usual).
> + * If you want to get a valid byte-based user-space address for a given offset,
> + * please see drm_vma_node_offset_addr().
> + */
> +
> +/**
> + * drm_vma_offset_manager_init - Initialize new offset-manager
> + * @mgr: Manager object
> + * @page_offset: Offset of available memory area (page-based)
> + * @size: Size of available memory area (page-based)

Bikeshed on naming: s/memory area/address space range/. Feel free to
ignore.

> + *
> + * Initialize a new offset-manager. The offset and area size available for the
> + * manager are given as @page_offset and @size. Both are interpreted as
> + * page-numbers, not bytes.
> + *
> + * Adding/removing nodes from the manager is locked internally and protected
> + * against concurrent access. However, node allocation and destruction is left
> + * for the caller. While calling into the vma-manager, a given node must
> + * always be guaranteed to be referenced.
> + */
> +void drm_vma_offset_manager_init(struct drm_vma_offset_manager *mgr,
> +				 unsigned long page_offset, unsigned long size)
> +{
> +	rwlock_init(&mgr->vm_lock);
> +	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);
> +
> +/**
> + * drm_vma_offset_manager_destroy() - Destroy offset manager
> + * @mgr: Manager object
> + *
> + * Destroy an object manager which was previously created via
> + * drm_vma_offset_manager_init(). The caller must remove all allocated nodes
> + * before destroying the manager. Otherwise, drm_mm will refuse to free the
> + * requested resources.
> + *
> + * The manager must not be accessed after this function is called.
> + */
> +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);
> +
> +/**
> + * drm_vma_offset_lookup() - Find node in offset space
> + * @mgr: Manager object
> + * @start: Start address for object (page-based)
> + * @pages: Size of object (page-based)
> + *
> + * 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).
> + *
> + * RETURNS:
> + * Returns NULL if no suitable node can be found. Otherwise, the best match
> + * 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(struct drm_vma_offset_manager *mgr,
> +						  unsigned long start,
> +						  unsigned long pages)
> +{
> +	struct drm_vma_offset_node *node;
> +
> +	read_lock(&mgr->vm_lock);
> +	node = drm_vma_offset_lookup_locked(mgr, start, pages);
> +	read_unlock(&mgr->vm_lock);
> +
> +	return node;
> +}
> +EXPORT_SYMBOL(drm_vma_offset_lookup);
> +
> +/**
> + * drm_vma_offset_lookup_locked() - Find node in offset space
> + * @mgr: Manager object
> + * @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.
> + *
> + * RETURNS:
> + * Returns NULL if no suitable node can be found. Otherwise, the best match
> + * is returned.
> + */
> +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 rb_node *iter;
> +	unsigned long offset;
> +
> +	iter = mgr->vm_addr_space_rb.rb_node;
> +	best = NULL;
> +
> +	while (likely(iter)) {
> +		node = rb_entry(iter, struct drm_vma_offset_node, vm_rb);
> +		offset = node->vm_node.start;
> +		if (start >= offset) {
> +			iter = iter->rb_right;
> +			best = node;
> +			if (start == offset)
> +				break;
> +		} else {
> +			iter = iter->rb_left;
> +		}
> +	}
> +
> +	/* verify that the node spans the requested area */
> +	if (best) {
> +		offset = best->vm_node.start + best->vm_pages;
> +		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();
> +	}
> +
> +	rb_link_node(&node->vm_rb, parent, iter);
> +	rb_insert_color(&node->vm_rb, &mgr->vm_addr_space_rb);
> +}
> +
> +/**
> + * drm_vma_offset_add() - Add offset node to manager
> + * @mgr: Manager object
> + * @node: Node to be added
> + * @pages: Allocation size visible to user-space (in number of pages)
> + *
> + * Add a node to the offset-manager. If the node was already added, this does
> + * nothing and return 0. @pages is the size of the object given in number of
> + * pages.
> + * After this call succeeds, you can access the offset of the node until it
> + * is removed again.
> + *
> + * If this call fails, it is safe to retry the operation or call
> + * drm_vma_offset_remove(), anyway. However, no cleanup is required in that
> + * case.
> + *
> + * @pages is not required to be the same size as the underlying memory object
> + * that you want to map. It only limits the size that user-space can map into
> + * their address space.
> + *
> + * RETURNS:
> + * 0 on success, negative error code on failure.
> + */
> +int drm_vma_offset_add(struct drm_vma_offset_manager *mgr,
> +		       struct drm_vma_offset_node *node, unsigned long pages)
> +{
> +	int ret;
> +
> +	write_lock(&mgr->vm_lock);
> +
> +	if (drm_mm_node_allocated(&node->vm_node)) {
> +		ret = 0;
> +		goto out_unlock;
> +	}
> +
> +	ret = drm_mm_insert_node_generic(&mgr->vm_addr_space_mm,
> +					 &node->vm_node, pages, 0, 0);
> +	if (ret)
> +		goto out_unlock;
> +
> +	node->vm_pages = pages;

Why not just use node->vm_node.size and remove vm_pages?

> +	_drm_vma_offset_add_rb(mgr, node);
> +	goto out_unlock;

Unnecessary goto.

> +
> +out_unlock:
> +	write_unlock(&mgr->vm_lock);
> +	return ret;
> +}
> +EXPORT_SYMBOL(drm_vma_offset_add);
> +
> +/**
> + * drm_vma_offset_remove() - Remove offset node from manager
> + * @mgr: Manager object
> + * @node: Node to be removed
> + *
> + * Remove a node from the offset manager. If the node wasn't added before, this
> + * does nothing. After this call returns, the offset of the node must be not
> + * accessed, anymore.

I think we could clarify this a little "After this call returns the offset
node must not be used until a new offset is allocatd with
drm_vma_offset_add() again. Helper functions like
drm_vma_node_offset_addr() will return invalid data"

> + * It is safe to add the node via drm_vma_offset_add() again.
> + */
> +void drm_vma_offset_remove(struct drm_vma_offset_manager *mgr,
> +			   struct drm_vma_offset_node *node)
> +{
> +	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));
> +		node->vm_pages = 0;
> +	}
> +
> +	write_unlock(&mgr->vm_lock);
> +}
> +EXPORT_SYMBOL(drm_vma_offset_remove);
> diff --git a/include/drm/drm_vma_manager.h b/include/drm/drm_vma_manager.h
> new file mode 100644
> index 0000000..ace2925
> --- /dev/null
> +++ b/include/drm/drm_vma_manager.h
> @@ -0,0 +1,180 @@
> +#ifndef __DRM_VMA_MANAGER_H__
> +#define __DRM_VMA_MANAGER_H__
> +
> +/*
> + * Copyright (c) 2013 David Herrmann <[e-mail]>
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a
> + * copy of this software and associated documentation files (the "Software"),
> + * to deal in the Software without restriction, including without limitation
> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
> + * and/or sell copies of the Software, and to permit persons to whom the
> + * Software is furnished to do so, subject to the following conditions:
> + *
> + * The above copyright notice and this permission notice shall be included in
> + * all copies or substantial portions of the Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
> + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
> + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
> + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
> + * OTHER DEALINGS IN THE SOFTWARE.
> + */
> +
> +#include <drm/drm_mm.h>
> +#include <linux/module.h>
> +#include <linux/rbtree.h>
> +#include <linux/spinlock.h>
> +#include <linux/types.h>
> +
> +struct drm_vma_offset_node {
> +	struct drm_mm_node vm_node;
> +	struct rb_node vm_rb;
> +	unsigned long vm_pages;
> +};
> +
> +struct drm_vma_offset_manager {
> +	rwlock_t vm_lock;
> +	struct rb_root vm_addr_space_rb;
> +	struct drm_mm vm_addr_space_mm;
> +};
> +
> +void drm_vma_offset_manager_init(struct drm_vma_offset_manager *mgr,
> +				 unsigned long page_offset, unsigned long size);
> +void drm_vma_offset_manager_destroy(struct drm_vma_offset_manager *mgr);
> +
> +struct drm_vma_offset_node *drm_vma_offset_lookup(struct drm_vma_offset_manager *mgr,
> +						  unsigned long start,
> +						  unsigned long pages);
> +struct drm_vma_offset_node *drm_vma_offset_lookup_locked(struct drm_vma_offset_manager *mgr,
> +							   unsigned long start,
> +							   unsigned long pages);
> +int drm_vma_offset_add(struct drm_vma_offset_manager *mgr,
> +		       struct drm_vma_offset_node *node, unsigned long pages);
> +void drm_vma_offset_remove(struct drm_vma_offset_manager *mgr,
> +			   struct drm_vma_offset_node *node);
> +
> +/**
> + * drm_vma_offset_exact_lookup() - Look up node by exact address
> + * @mgr: Manager object
> + * @start: Start address (page-based, not byte-based)
> + *
> + * Same as drm_vma_offset_lookup() but does not allow any offset into the node.
> + * It only returns the exact object with the given start address.
> + *
> + * RETURNS:
> + * Node at exact start address @start.
> + */
> +static inline struct drm_vma_offset_node *
> +drm_vma_offset_exact_lookup(struct drm_vma_offset_manager *mgr,
> +			    unsigned long start)
> +{
> +	return drm_vma_offset_lookup(mgr, start, 1);
> +}
> +
> +/**
> + * drm_vma_offset_lock_lookup() - Lock lookup for extended private use
> + * @mgr: Manager object
> + *
> + * Lock VMA manager for extended lookups. Only *_locked() VMA function calls
> + * are allowed while holding this lock. All other contexts are blocked from VMA
> + * until the lock is released via drm_vma_offset_unlock_lookup().
> + *
> + * Use this if you need to take a reference to the objects returned by
> + * drm_vma_offset_lookup_locked() before releasing this lock again.
> + *
> + * This lock must not be used for anything else than extended lookups. You must
> + * not call any other VMA helpers while holding this lock.
> + *
> + * Note: You're in atomic-context while holding this lock!
> + *
> + * 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);
> + */
> +static inline void drm_vma_offset_lock_lookup(struct drm_vma_offset_manager *mgr)
> +{
> +	read_lock(&mgr->vm_lock);
> +}
> +
> +/**
> + * drm_vma_offset_unlock_lookup() - Unlock lookup for extended private use
> + * @mgr: Manager object
> + *
> + * Release lookup-lock. See drm_vma_offset_lock_lookup() for more information.
> + */
> +static inline void drm_vma_offset_unlock_lookup(struct drm_vma_offset_manager *mgr)
> +{
> +	read_unlock(&mgr->vm_lock);
> +}
> +
> +/**
> + * drm_vma_node_reset() - Initialize or reset node object
> + * @node: Node to initialize or reset
> + *
> + * Reset a node to its initial state. You must call this before using @node with
> + * any vma offset manager.

The must here feels a bit strong since this is just a memset and pretty
much everyone allocates with kzalloc. So calling this is redundant. Maybe

"This must be called if @node isn't already cleared (e.g. by allocating
with kzalloc) before it using it with any vma offset manager."

> + *
> + * This must not be called on an already allocated node, or you will leak
> + * memory.
> + */
> +static inline void drm_vma_node_reset(struct drm_vma_offset_node *node)
> +{
> +	memset(node, 0, sizeof(*node));
> +}
> +
> +/**
> + * drm_vma_node_start() - Return start address for page-based addressing
> + * @node: Node to inspect
> + *
> + * Return the start address of the given node. This can be used as offset into
> + * the linear VM space that is provided by the VMA offset manager. Note that
> + * this can only be used for page-based addressing. If you need a proper offset
> + * for user-space mappings, you must apply "<< PAGE_SHIFT" or use the
> + * drm_vma_node_offset_addr() helper instead.
> + *
> + * RETURNS:
> + * Start address of @node for page-based addressing. 0 if the node does not
> + * have an offset allocated.
> + */
> +static inline unsigned long drm_vma_node_start(struct drm_vma_offset_node *node)
> +{
> +	return node->vm_node.start;
> +}
> +
> +/**
> + * drm_vma_node_has_offset() - Check whether node is added to offset manager
> + * @node: Node to be checked
> + *
> + * RETURNS:
> + * true iff the node was previously allocated an offset and added to
> + * an vma offset manager.
> + */
> +static inline bool drm_vma_node_has_offset(struct drm_vma_offset_node *node)
> +{
> +	return drm_mm_node_allocated(&node->vm_node);
> +}
> +
> +/**
> + * drm_vma_node_offset_addr() - Return sanitized offset for user-space mmaps
> + * @node: Linked offset node
> + *
> + * Same as drm_vma_node_start() but returns the address as a valid offset that
> + * can be used for user-space mappings during mmap().
> + * This must not be called on unlinked nodes.
> + *
> + * RETURNS:
> + * Offset of @node for byte-based addressing. 0 if the node does not have an
> + * object allocated.
> + */
> +static inline __u64 drm_vma_node_offset_addr(struct drm_vma_offset_node *node)
> +{
> +	return ((__u64)node->vm_node.start) << PAGE_SHIFT;
> +}
> +
> +#endif /* __DRM_VMA_MANAGER_H__ */

With the above tiny bikesheds addressed this is:

Reviewed-by: Daniel Vetter <[e-mail]>

David Herrmann July 24, 2013, 4:10 p.m. UTC | #2

Hi

On Wed, Jul 24, 2013 at 5:35 PM, Daniel Vetter <[e-mail]> wrote:
> On Tue, Jul 23, 2013 at 02:47:13PM +0200, David Herrmann wrote:
>> If we want to map GPU memory into user-space, we need to linearize the
>> addresses to not confuse mm-core. Currently, GEM and TTM both implement
>> their own offset-managers to assign a pgoff to each object for user-space
>> CPU access. GEM uses a hash-table, TTM uses an rbtree.
>>
>> This patch provides a unified implementation that can be used to replace
>> both. TTM allows partial mmaps with a given offset, so we cannot use
>> hashtables as the start address may not be known at mmap time. Hence, we
>> use the rbtree-implementation of TTM.
>>
>> We could easily update drm_mm to use an rbtree instead of a linked list
>> for it's object list and thus drop the rbtree from the vma-manager.
>> However, this would slow down drm_mm object allocation for all other
>> use-cases (rbtree insertion) and add another 4-8 bytes to each mm node.
>> Hence, use the separate tree but allow for later migration.
>>
>> This is a rewrite of the 2012-proposal by David Airlie <[e-mail]>
>>
>> v2:
>>  - fix Docbook integration
>>  - drop drm_mm_node_linked() and use drm_mm_node_allocated()
>>  - remove unjustified likely/unlikely usage (but keep for rbtree paths)
>>  - remove BUG_ON() as drm_mm already does that
>>  - clarify page-based vs. byte-based addresses
>>  - use drm_vma_node_reset() for initialization, too
>> v4:
>>  - allow external locking via drm_vma_offset_un/lock_lookup()
>>  - add locked lookup helper drm_vma_offset_lookup_locked()
>>
>> Signed-off-by: David Herrmann <[e-mail]>
>> ---
>>  Documentation/DocBook/drm.tmpl    |   6 +
>>  drivers/gpu/drm/Makefile          |   2 +-
>>  drivers/gpu/drm/drm_vma_manager.c | 283 ++++++++++++++++++++++++++++++++++++++
>>  include/drm/drm_vma_manager.h     | 180 ++++++++++++++++++++++++
>>  4 files changed, 470 insertions(+), 1 deletion(-)
>>  create mode 100644 drivers/gpu/drm/drm_vma_manager.c
>>  create mode 100644 include/drm/drm_vma_manager.h
>>
>> diff --git a/Documentation/DocBook/drm.tmpl b/Documentation/DocBook/drm.tmpl
>> index 7d1278e..87e22ec 100644
>> --- a/Documentation/DocBook/drm.tmpl
>> +++ b/Documentation/DocBook/drm.tmpl
>> @@ -2212,6 +2212,12 @@ void intel_crt_init(struct drm_device *dev)
>>  !Iinclude/drm/drm_rect.h
>>  !Edrivers/gpu/drm/drm_rect.c
>>      </sect2>
>> +    <sect2>
>> +      <title>VMA Offset Manager</title>
>> +!Pdrivers/gpu/drm/drm_vma_manager.c vma offset manager
>> +!Edrivers/gpu/drm/drm_vma_manager.c
>> +!Iinclude/drm/drm_vma_manager.h
>> +    </sect2>
>>    </sect1>
>>
>>    <!-- Internals: kms properties -->
>> diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile
>> index 801bcaf..d943b94 100644
>> --- a/drivers/gpu/drm/Makefile
>> +++ b/drivers/gpu/drm/Makefile
>> @@ -13,7 +13,7 @@ drm-y       :=      drm_auth.o drm_buffer.o drm_bufs.o drm_cache.o \
>>               drm_crtc.o drm_modes.o drm_edid.o \
>>               drm_info.o drm_debugfs.o drm_encoder_slave.o \
>>               drm_trace_points.o drm_global.o drm_prime.o \
>> -             drm_rect.o
>> +             drm_rect.o drm_vma_manager.o
>>
>>  drm-$(CONFIG_COMPAT) += drm_ioc32.o
>>  drm-$(CONFIG_DRM_GEM_CMA_HELPER) += drm_gem_cma_helper.o
>> diff --git a/drivers/gpu/drm/drm_vma_manager.c b/drivers/gpu/drm/drm_vma_manager.c
>> new file mode 100644
>> index 0000000..27760bd
>> --- /dev/null
>> +++ b/drivers/gpu/drm/drm_vma_manager.c
>> @@ -0,0 +1,283 @@
>> +/*
>> + * Copyright (c) 2006-2009 VMware, Inc., Palo Alto, CA., USA
>> + * Copyright (c) 2012 David Airlie <[e-mail]>
>> + * Copyright (c) 2013 David Herrmann <[e-mail]>
>> + *
>> + * Permission is hereby granted, free of charge, to any person obtaining a
>> + * copy of this software and associated documentation files (the "Software"),
>> + * to deal in the Software without restriction, including without limitation
>> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
>> + * and/or sell copies of the Software, and to permit persons to whom the
>> + * Software is furnished to do so, subject to the following conditions:
>> + *
>> + * The above copyright notice and this permission notice shall be included in
>> + * all copies or substantial portions of the Software.
>> + *
>> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
>> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
>> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
>> + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
>> + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
>> + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
>> + * OTHER DEALINGS IN THE SOFTWARE.
>> + */
>> +
>> +#include <drm/drmP.h>
>> +#include <drm/drm_mm.h>
>> +#include <drm/drm_vma_manager.h>
>> +#include <linux/mm.h>
>> +#include <linux/module.h>
>> +#include <linux/rbtree.h>
>> +#include <linux/slab.h>
>> +#include <linux/spinlock.h>
>> +#include <linux/types.h>
>> +
>> +/**
>> + * DOC: vma offset manager
>> + *
>> + * The vma-manager is responsible to map arbitrary driver-dependent memory
>> + * regions into the linear user address-space. It provides offsets to the
>> + * caller which can then be used on the address_space of the drm-device. It
>> + * takes care to not overlap regions, size them appropriately and to not
>> + * confuse mm-core by inconsistent fake vm_pgoff fields.
>> + * Drivers shouldn't use this for object placement in VMEM. This manager should
>> + * only be used to manage mappings into linear user-space VMs.
>> + *
>> + * We use drm_mm as backend to manage object allocations. But it is highly
>> + * optimized for alloc/free calls, not lookups. Hence, we use an rb-tree to
>> + * speed up offset lookups.
>> + *
>> + * You must not use multiple offset managers on a single address_space.
>> + * Otherwise, mm-core will be unable to tear down memory mappings as the VM will
>> + * no longer be linear. Please use VM_NONLINEAR in that case and implement your
>> + * own offset managers.
>> + *
>> + * This offset manager works on page-based addresses. That is, every argument
>> + * and return code (with the exception of drm_vma_node_offset_addr()) is given
>> + * in number of pages, not number of bytes. That means, object sizes and offsets
>> + * must always be page-aligned (as usual).
>> + * If you want to get a valid byte-based user-space address for a given offset,
>> + * please see drm_vma_node_offset_addr().
>> + */
>> +
>> +/**
>> + * drm_vma_offset_manager_init - Initialize new offset-manager
>> + * @mgr: Manager object
>> + * @page_offset: Offset of available memory area (page-based)
>> + * @size: Size of available memory area (page-based)
>
> Bikeshed on naming: s/memory area/address space range/. Feel free to
> ignore.

Makes sense, yepp.

>> + *
>> + * Initialize a new offset-manager. The offset and area size available for the
>> + * manager are given as @page_offset and @size. Both are interpreted as
>> + * page-numbers, not bytes.
>> + *
>> + * Adding/removing nodes from the manager is locked internally and protected
>> + * against concurrent access. However, node allocation and destruction is left
>> + * for the caller. While calling into the vma-manager, a given node must
>> + * always be guaranteed to be referenced.
>> + */
>> +void drm_vma_offset_manager_init(struct drm_vma_offset_manager *mgr,
>> +                              unsigned long page_offset, unsigned long size)
>> +{
>> +     rwlock_init(&mgr->vm_lock);
>> +     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);
>> +
>> +/**
>> + * drm_vma_offset_manager_destroy() - Destroy offset manager
>> + * @mgr: Manager object
>> + *
>> + * Destroy an object manager which was previously created via
>> + * drm_vma_offset_manager_init(). The caller must remove all allocated nodes
>> + * before destroying the manager. Otherwise, drm_mm will refuse to free the
>> + * requested resources.
>> + *
>> + * The manager must not be accessed after this function is called.
>> + */
>> +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);
>> +
>> +/**
>> + * drm_vma_offset_lookup() - Find node in offset space
>> + * @mgr: Manager object
>> + * @start: Start address for object (page-based)
>> + * @pages: Size of object (page-based)
>> + *
>> + * 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).
>> + *
>> + * RETURNS:
>> + * Returns NULL if no suitable node can be found. Otherwise, the best match
>> + * 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(struct drm_vma_offset_manager *mgr,
>> +                                               unsigned long start,
>> +                                               unsigned long pages)
>> +{
>> +     struct drm_vma_offset_node *node;
>> +
>> +     read_lock(&mgr->vm_lock);
>> +     node = drm_vma_offset_lookup_locked(mgr, start, pages);
>> +     read_unlock(&mgr->vm_lock);
>> +
>> +     return node;
>> +}
>> +EXPORT_SYMBOL(drm_vma_offset_lookup);
>> +
>> +/**
>> + * drm_vma_offset_lookup_locked() - Find node in offset space
>> + * @mgr: Manager object
>> + * @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.
>> + *
>> + * RETURNS:
>> + * Returns NULL if no suitable node can be found. Otherwise, the best match
>> + * is returned.
>> + */
>> +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 rb_node *iter;
>> +     unsigned long offset;
>> +
>> +     iter = mgr->vm_addr_space_rb.rb_node;
>> +     best = NULL;
>> +
>> +     while (likely(iter)) {
>> +             node = rb_entry(iter, struct drm_vma_offset_node, vm_rb);
>> +             offset = node->vm_node.start;
>> +             if (start >= offset) {
>> +                     iter = iter->rb_right;
>> +                     best = node;
>> +                     if (start == offset)
>> +                             break;
>> +             } else {
>> +                     iter = iter->rb_left;
>> +             }
>> +     }
>> +
>> +     /* verify that the node spans the requested area */
>> +     if (best) {
>> +             offset = best->vm_node.start + best->vm_pages;
>> +             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();
>> +     }
>> +
>> +     rb_link_node(&node->vm_rb, parent, iter);
>> +     rb_insert_color(&node->vm_rb, &mgr->vm_addr_space_rb);
>> +}
>> +
>> +/**
>> + * drm_vma_offset_add() - Add offset node to manager
>> + * @mgr: Manager object
>> + * @node: Node to be added
>> + * @pages: Allocation size visible to user-space (in number of pages)
>> + *
>> + * Add a node to the offset-manager. If the node was already added, this does
>> + * nothing and return 0. @pages is the size of the object given in number of
>> + * pages.
>> + * After this call succeeds, you can access the offset of the node until it
>> + * is removed again.
>> + *
>> + * If this call fails, it is safe to retry the operation or call
>> + * drm_vma_offset_remove(), anyway. However, no cleanup is required in that
>> + * case.
>> + *
>> + * @pages is not required to be the same size as the underlying memory object
>> + * that you want to map. It only limits the size that user-space can map into
>> + * their address space.
>> + *
>> + * RETURNS:
>> + * 0 on success, negative error code on failure.
>> + */
>> +int drm_vma_offset_add(struct drm_vma_offset_manager *mgr,
>> +                    struct drm_vma_offset_node *node, unsigned long pages)
>> +{
>> +     int ret;
>> +
>> +     write_lock(&mgr->vm_lock);
>> +
>> +     if (drm_mm_node_allocated(&node->vm_node)) {
>> +             ret = 0;
>> +             goto out_unlock;
>> +     }
>> +
>> +     ret = drm_mm_insert_node_generic(&mgr->vm_addr_space_mm,
>> +                                      &node->vm_node, pages, 0, 0);
>> +     if (ret)
>> +             goto out_unlock;
>> +
>> +     node->vm_pages = pages;
>
> Why not just use node->vm_node.size and remove vm_pages?

In the first draft I initialized the size directly. I added it to
drm_vma_offset_add() later, so that's a left-over. I will remove the
duplicate and use node->vm_node.size directly.

>> +     _drm_vma_offset_add_rb(mgr, node);
>> +     goto out_unlock;
>
> Unnecessary goto.

Fixed.

>> +
>> +out_unlock:
>> +     write_unlock(&mgr->vm_lock);
>> +     return ret;
>> +}
>> +EXPORT_SYMBOL(drm_vma_offset_add);
>> +
>> +/**
>> + * drm_vma_offset_remove() - Remove offset node from manager
>> + * @mgr: Manager object
>> + * @node: Node to be removed
>> + *
>> + * Remove a node from the offset manager. If the node wasn't added before, this
>> + * does nothing. After this call returns, the offset of the node must be not
>> + * accessed, anymore.
>
> I think we could clarify this a little "After this call returns the offset
> node must not be used until a new offset is allocatd with
> drm_vma_offset_add() again. Helper functions like
> drm_vma_node_offset_addr() will return invalid data"

They guarantee to return 0. I use that in the debugfs paths so we
don't have to check drm_vma_node_has_offset() all the time. So I guess
I will change it to "return 0 (invalid)" or alike instead of just
"invalid".

>> + * It is safe to add the node via drm_vma_offset_add() again.
>> + */
>> +void drm_vma_offset_remove(struct drm_vma_offset_manager *mgr,
>> +                        struct drm_vma_offset_node *node)
>> +{
>> +     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));
>> +             node->vm_pages = 0;
>> +     }
>> +
>> +     write_unlock(&mgr->vm_lock);
>> +}
>> +EXPORT_SYMBOL(drm_vma_offset_remove);
>> diff --git a/include/drm/drm_vma_manager.h b/include/drm/drm_vma_manager.h
>> new file mode 100644
>> index 0000000..ace2925
>> --- /dev/null
>> +++ b/include/drm/drm_vma_manager.h
>> @@ -0,0 +1,180 @@
>> +#ifndef __DRM_VMA_MANAGER_H__
>> +#define __DRM_VMA_MANAGER_H__
>> +
>> +/*
>> + * Copyright (c) 2013 David Herrmann <[e-mail]>
>> + *
>> + * Permission is hereby granted, free of charge, to any person obtaining a
>> + * copy of this software and associated documentation files (the "Software"),
>> + * to deal in the Software without restriction, including without limitation
>> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
>> + * and/or sell copies of the Software, and to permit persons to whom the
>> + * Software is furnished to do so, subject to the following conditions:
>> + *
>> + * The above copyright notice and this permission notice shall be included in
>> + * all copies or substantial portions of the Software.
>> + *
>> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
>> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
>> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
>> + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
>> + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
>> + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
>> + * OTHER DEALINGS IN THE SOFTWARE.
>> + */
>> +
>> +#include <drm/drm_mm.h>
>> +#include <linux/module.h>
>> +#include <linux/rbtree.h>
>> +#include <linux/spinlock.h>
>> +#include <linux/types.h>
>> +
>> +struct drm_vma_offset_node {
>> +     struct drm_mm_node vm_node;
>> +     struct rb_node vm_rb;
>> +     unsigned long vm_pages;
>> +};
>> +
>> +struct drm_vma_offset_manager {
>> +     rwlock_t vm_lock;
>> +     struct rb_root vm_addr_space_rb;
>> +     struct drm_mm vm_addr_space_mm;
>> +};
>> +
>> +void drm_vma_offset_manager_init(struct drm_vma_offset_manager *mgr,
>> +                              unsigned long page_offset, unsigned long size);
>> +void drm_vma_offset_manager_destroy(struct drm_vma_offset_manager *mgr);
>> +
>> +struct drm_vma_offset_node *drm_vma_offset_lookup(struct drm_vma_offset_manager *mgr,
>> +                                               unsigned long start,
>> +                                               unsigned long pages);
>> +struct drm_vma_offset_node *drm_vma_offset_lookup_locked(struct drm_vma_offset_manager *mgr,
>> +                                                        unsigned long start,
>> +                                                        unsigned long pages);
>> +int drm_vma_offset_add(struct drm_vma_offset_manager *mgr,
>> +                    struct drm_vma_offset_node *node, unsigned long pages);
>> +void drm_vma_offset_remove(struct drm_vma_offset_manager *mgr,
>> +                        struct drm_vma_offset_node *node);
>> +
>> +/**
>> + * drm_vma_offset_exact_lookup() - Look up node by exact address
>> + * @mgr: Manager object
>> + * @start: Start address (page-based, not byte-based)
>> + *
>> + * Same as drm_vma_offset_lookup() but does not allow any offset into the node.
>> + * It only returns the exact object with the given start address.
>> + *
>> + * RETURNS:
>> + * Node at exact start address @start.
>> + */
>> +static inline struct drm_vma_offset_node *
>> +drm_vma_offset_exact_lookup(struct drm_vma_offset_manager *mgr,
>> +                         unsigned long start)
>> +{
>> +     return drm_vma_offset_lookup(mgr, start, 1);
>> +}
>> +
>> +/**
>> + * drm_vma_offset_lock_lookup() - Lock lookup for extended private use
>> + * @mgr: Manager object
>> + *
>> + * Lock VMA manager for extended lookups. Only *_locked() VMA function calls
>> + * are allowed while holding this lock. All other contexts are blocked from VMA
>> + * until the lock is released via drm_vma_offset_unlock_lookup().
>> + *
>> + * Use this if you need to take a reference to the objects returned by
>> + * drm_vma_offset_lookup_locked() before releasing this lock again.
>> + *
>> + * This lock must not be used for anything else than extended lookups. You must
>> + * not call any other VMA helpers while holding this lock.
>> + *
>> + * Note: You're in atomic-context while holding this lock!
>> + *
>> + * 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);
>> + */
>> +static inline void drm_vma_offset_lock_lookup(struct drm_vma_offset_manager *mgr)
>> +{
>> +     read_lock(&mgr->vm_lock);
>> +}
>> +
>> +/**
>> + * drm_vma_offset_unlock_lookup() - Unlock lookup for extended private use
>> + * @mgr: Manager object
>> + *
>> + * Release lookup-lock. See drm_vma_offset_lock_lookup() for more information.
>> + */
>> +static inline void drm_vma_offset_unlock_lookup(struct drm_vma_offset_manager *mgr)
>> +{
>> +     read_unlock(&mgr->vm_lock);
>> +}
>> +
>> +/**
>> + * drm_vma_node_reset() - Initialize or reset node object
>> + * @node: Node to initialize or reset
>> + *
>> + * Reset a node to its initial state. You must call this before using @node with
>> + * any vma offset manager.
>
> The must here feels a bit strong since this is just a memset and pretty
> much everyone allocates with kzalloc. So calling this is redundant. Maybe
>
> "This must be called if @node isn't already cleared (e.g. by allocating
> with kzalloc) before it using it with any vma offset manager."

Makes sense. drm_mm has the same semantics, anyway.

>> + *
>> + * This must not be called on an already allocated node, or you will leak
>> + * memory.
>> + */
>> +static inline void drm_vma_node_reset(struct drm_vma_offset_node *node)
>> +{
>> +     memset(node, 0, sizeof(*node));
>> +}
>> +
>> +/**
>> + * drm_vma_node_start() - Return start address for page-based addressing
>> + * @node: Node to inspect
>> + *
>> + * Return the start address of the given node. This can be used as offset into
>> + * the linear VM space that is provided by the VMA offset manager. Note that
>> + * this can only be used for page-based addressing. If you need a proper offset
>> + * for user-space mappings, you must apply "<< PAGE_SHIFT" or use the
>> + * drm_vma_node_offset_addr() helper instead.
>> + *
>> + * RETURNS:
>> + * Start address of @node for page-based addressing. 0 if the node does not
>> + * have an offset allocated.
>> + */
>> +static inline unsigned long drm_vma_node_start(struct drm_vma_offset_node *node)
>> +{
>> +     return node->vm_node.start;
>> +}
>> +
>> +/**
>> + * drm_vma_node_has_offset() - Check whether node is added to offset manager
>> + * @node: Node to be checked
>> + *
>> + * RETURNS:
>> + * true iff the node was previously allocated an offset and added to
>> + * an vma offset manager.
>> + */
>> +static inline bool drm_vma_node_has_offset(struct drm_vma_offset_node *node)
>> +{
>> +     return drm_mm_node_allocated(&node->vm_node);
>> +}
>> +
>> +/**
>> + * drm_vma_node_offset_addr() - Return sanitized offset for user-space mmaps
>> + * @node: Linked offset node
>> + *
>> + * Same as drm_vma_node_start() but returns the address as a valid offset that
>> + * can be used for user-space mappings during mmap().
>> + * This must not be called on unlinked nodes.
>> + *
>> + * RETURNS:
>> + * Offset of @node for byte-based addressing. 0 if the node does not have an
>> + * object allocated.
>> + */
>> +static inline __u64 drm_vma_node_offset_addr(struct drm_vma_offset_node *node)
>> +{
>> +     return ((__u64)node->vm_node.start) << PAGE_SHIFT;
>> +}
>> +
>> +#endif /* __DRM_VMA_MANAGER_H__ */
>
> With the above tiny bikesheds addressed this is:
>
> Reviewed-by: Daniel Vetter <[e-mail]>

Thanks
David

Patch

diff --git a/Documentation/DocBook/drm.tmpl b/Documentation/DocBook/drm.tmpl
index 7d1278e..87e22ec 100644
--- a/Documentation/DocBook/drm.tmpl
+++ b/Documentation/DocBook/drm.tmpl
@@ -2212,6 +2212,12 @@  void intel_crt_init(struct drm_device *dev)
 !Iinclude/drm/drm_rect.h
 !Edrivers/gpu/drm/drm_rect.c
     </sect2>
+    <sect2>
+      <title>VMA Offset Manager</title>
+!Pdrivers/gpu/drm/drm_vma_manager.c vma offset manager
+!Edrivers/gpu/drm/drm_vma_manager.c
+!Iinclude/drm/drm_vma_manager.h
+    </sect2>
   </sect1>
 
   <!-- Internals: kms properties -->
diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile
index 801bcaf..d943b94 100644
--- a/drivers/gpu/drm/Makefile
+++ b/drivers/gpu/drm/Makefile
@@ -13,7 +13,7 @@  drm-y       :=	drm_auth.o drm_buffer.o drm_bufs.o drm_cache.o \
 		drm_crtc.o drm_modes.o drm_edid.o \
 		drm_info.o drm_debugfs.o drm_encoder_slave.o \
 		drm_trace_points.o drm_global.o drm_prime.o \
-		drm_rect.o
+		drm_rect.o drm_vma_manager.o
 
 drm-$(CONFIG_COMPAT) += drm_ioc32.o
 drm-$(CONFIG_DRM_GEM_CMA_HELPER) += drm_gem_cma_helper.o
diff --git a/drivers/gpu/drm/drm_vma_manager.c b/drivers/gpu/drm/drm_vma_manager.c
new file mode 100644
index 0000000..27760bd
--- /dev/null
+++ b/drivers/gpu/drm/drm_vma_manager.c
@@ -0,0 +1,283 @@ 
+/*
+ * Copyright (c) 2006-2009 VMware, Inc., Palo Alto, CA., USA
+ * Copyright (c) 2012 David Airlie <[e-mail]>
+ * Copyright (c) 2013 David Herrmann <[e-mail]>
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include <drm/drmP.h>
+#include <drm/drm_mm.h>
+#include <drm/drm_vma_manager.h>
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/rbtree.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/types.h>
+
+/**
+ * DOC: vma offset manager
+ *
+ * The vma-manager is responsible to map arbitrary driver-dependent memory
+ * regions into the linear user address-space. It provides offsets to the
+ * caller which can then be used on the address_space of the drm-device. It
+ * takes care to not overlap regions, size them appropriately and to not
+ * confuse mm-core by inconsistent fake vm_pgoff fields.
+ * Drivers shouldn't use this for object placement in VMEM. This manager should
+ * only be used to manage mappings into linear user-space VMs.
+ *
+ * We use drm_mm as backend to manage object allocations. But it is highly
+ * optimized for alloc/free calls, not lookups. Hence, we use an rb-tree to
+ * speed up offset lookups.
+ *
+ * You must not use multiple offset managers on a single address_space.
+ * Otherwise, mm-core will be unable to tear down memory mappings as the VM will
+ * no longer be linear. Please use VM_NONLINEAR in that case and implement your
+ * own offset managers.
+ *
+ * This offset manager works on page-based addresses. That is, every argument
+ * and return code (with the exception of drm_vma_node_offset_addr()) is given
+ * in number of pages, not number of bytes. That means, object sizes and offsets
+ * must always be page-aligned (as usual).
+ * If you want to get a valid byte-based user-space address for a given offset,
+ * please see drm_vma_node_offset_addr().
+ */
+
+/**
+ * drm_vma_offset_manager_init - Initialize new offset-manager
+ * @mgr: Manager object
+ * @page_offset: Offset of available memory area (page-based)
+ * @size: Size of available memory area (page-based)
+ *
+ * Initialize a new offset-manager. The offset and area size available for the
+ * manager are given as @page_offset and @size. Both are interpreted as
+ * page-numbers, not bytes.
+ *
+ * Adding/removing nodes from the manager is locked internally and protected
+ * against concurrent access. However, node allocation and destruction is left
+ * for the caller. While calling into the vma-manager, a given node must
+ * always be guaranteed to be referenced.
+ */
+void drm_vma_offset_manager_init(struct drm_vma_offset_manager *mgr,
+				 unsigned long page_offset, unsigned long size)
+{
+	rwlock_init(&mgr->vm_lock);
+	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);
+
+/**
+ * drm_vma_offset_manager_destroy() - Destroy offset manager
+ * @mgr: Manager object
+ *
+ * Destroy an object manager which was previously created via
+ * drm_vma_offset_manager_init(). The caller must remove all allocated nodes
+ * before destroying the manager. Otherwise, drm_mm will refuse to free the
+ * requested resources.
+ *
+ * The manager must not be accessed after this function is called.
+ */
+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);
+
+/**
+ * drm_vma_offset_lookup() - Find node in offset space
+ * @mgr: Manager object
+ * @start: Start address for object (page-based)
+ * @pages: Size of object (page-based)
+ *
+ * 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).
+ *
+ * RETURNS:
+ * Returns NULL if no suitable node can be found. Otherwise, the best match
+ * 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(struct drm_vma_offset_manager *mgr,
+						  unsigned long start,
+						  unsigned long pages)
+{
+	struct drm_vma_offset_node *node;
+
+	read_lock(&mgr->vm_lock);
+	node = drm_vma_offset_lookup_locked(mgr, start, pages);
+	read_unlock(&mgr->vm_lock);
+
+	return node;
+}
+EXPORT_SYMBOL(drm_vma_offset_lookup);
+
+/**
+ * drm_vma_offset_lookup_locked() - Find node in offset space
+ * @mgr: Manager object
+ * @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.
+ *
+ * RETURNS:
+ * Returns NULL if no suitable node can be found. Otherwise, the best match
+ * is returned.
+ */
+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 rb_node *iter;
+	unsigned long offset;
+
+	iter = mgr->vm_addr_space_rb.rb_node;
+	best = NULL;
+
+	while (likely(iter)) {
+		node = rb_entry(iter, struct drm_vma_offset_node, vm_rb);
+		offset = node->vm_node.start;
+		if (start >= offset) {
+			iter = iter->rb_right;
+			best = node;
+			if (start == offset)
+				break;
+		} else {
+			iter = iter->rb_left;
+		}
+	}
+
+	/* verify that the node spans the requested area */
+	if (best) {
+		offset = best->vm_node.start + best->vm_pages;
+		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();
+	}
+
+	rb_link_node(&node->vm_rb, parent, iter);
+	rb_insert_color(&node->vm_rb, &mgr->vm_addr_space_rb);
+}
+
+/**
+ * drm_vma_offset_add() - Add offset node to manager
+ * @mgr: Manager object
+ * @node: Node to be added
+ * @pages: Allocation size visible to user-space (in number of pages)
+ *
+ * Add a node to the offset-manager. If the node was already added, this does
+ * nothing and return 0. @pages is the size of the object given in number of
+ * pages.
+ * After this call succeeds, you can access the offset of the node until it
+ * is removed again.
+ *
+ * If this call fails, it is safe to retry the operation or call
+ * drm_vma_offset_remove(), anyway. However, no cleanup is required in that
+ * case.
+ *
+ * @pages is not required to be the same size as the underlying memory object
+ * that you want to map. It only limits the size that user-space can map into
+ * their address space.
+ *
+ * RETURNS:
+ * 0 on success, negative error code on failure.
+ */
+int drm_vma_offset_add(struct drm_vma_offset_manager *mgr,
+		       struct drm_vma_offset_node *node, unsigned long pages)
+{
+	int ret;
+
+	write_lock(&mgr->vm_lock);
+
+	if (drm_mm_node_allocated(&node->vm_node)) {
+		ret = 0;
+		goto out_unlock;
+	}
+
+	ret = drm_mm_insert_node_generic(&mgr->vm_addr_space_mm,
+					 &node->vm_node, pages, 0, 0);
+	if (ret)
+		goto out_unlock;
+
+	node->vm_pages = pages;
+	_drm_vma_offset_add_rb(mgr, node);
+	goto out_unlock;
+
+out_unlock:
+	write_unlock(&mgr->vm_lock);
+	return ret;
+}
+EXPORT_SYMBOL(drm_vma_offset_add);
+
+/**
+ * drm_vma_offset_remove() - Remove offset node from manager
+ * @mgr: Manager object
+ * @node: Node to be removed
+ *
+ * Remove a node from the offset manager. If the node wasn't added before, this
+ * does nothing. After this call returns, the offset of the node must be not
+ * accessed, anymore.
+ * It is safe to add the node via drm_vma_offset_add() again.
+ */
+void drm_vma_offset_remove(struct drm_vma_offset_manager *mgr,
+			   struct drm_vma_offset_node *node)
+{
+	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));
+		node->vm_pages = 0;
+	}
+
+	write_unlock(&mgr->vm_lock);
+}
+EXPORT_SYMBOL(drm_vma_offset_remove);
diff --git a/include/drm/drm_vma_manager.h b/include/drm/drm_vma_manager.h
new file mode 100644
index 0000000..ace2925
--- /dev/null
+++ b/include/drm/drm_vma_manager.h
@@ -0,0 +1,180 @@ 
+#ifndef __DRM_VMA_MANAGER_H__
+#define __DRM_VMA_MANAGER_H__
+
+/*
+ * Copyright (c) 2013 David Herrmann <[e-mail]>
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include <drm/drm_mm.h>
+#include <linux/module.h>
+#include <linux/rbtree.h>
+#include <linux/spinlock.h>
+#include <linux/types.h>
+
+struct drm_vma_offset_node {
+	struct drm_mm_node vm_node;
+	struct rb_node vm_rb;
+	unsigned long vm_pages;
+};
+
+struct drm_vma_offset_manager {
+	rwlock_t vm_lock;
+	struct rb_root vm_addr_space_rb;
+	struct drm_mm vm_addr_space_mm;
+};
+
+void drm_vma_offset_manager_init(struct drm_vma_offset_manager *mgr,
+				 unsigned long page_offset, unsigned long size);
+void drm_vma_offset_manager_destroy(struct drm_vma_offset_manager *mgr);
+
+struct drm_vma_offset_node *drm_vma_offset_lookup(struct drm_vma_offset_manager *mgr,
+						  unsigned long start,
+						  unsigned long pages);
+struct drm_vma_offset_node *drm_vma_offset_lookup_locked(struct drm_vma_offset_manager *mgr,
+							   unsigned long start,
+							   unsigned long pages);
+int drm_vma_offset_add(struct drm_vma_offset_manager *mgr,
+		       struct drm_vma_offset_node *node, unsigned long pages);
+void drm_vma_offset_remove(struct drm_vma_offset_manager *mgr,
+			   struct drm_vma_offset_node *node);
+
+/**
+ * drm_vma_offset_exact_lookup() - Look up node by exact address
+ * @mgr: Manager object
+ * @start: Start address (page-based, not byte-based)
+ *
+ * Same as drm_vma_offset_lookup() but does not allow any offset into the node.
+ * It only returns the exact object with the given start address.
+ *
+ * RETURNS:
+ * Node at exact start address @start.
+ */
+static inline struct drm_vma_offset_node *
+drm_vma_offset_exact_lookup(struct drm_vma_offset_manager *mgr,
+			    unsigned long start)
+{
+	return drm_vma_offset_lookup(mgr, start, 1);
+}
+
+/**
+ * drm_vma_offset_lock_lookup() - Lock lookup for extended private use
+ * @mgr: Manager object
+ *
+ * Lock VMA manager for extended lookups. Only *_locked() VMA function calls
+ * are allowed while holding this lock. All other contexts are blocked from VMA
+ * until the lock is released via drm_vma_offset_unlock_lookup().
+ *
+ * Use this if you need to take a reference to the objects returned by
+ * drm_vma_offset_lookup_locked() before releasing this lock again.
+ *
+ * This lock must not be used for anything else than extended lookups. You must
+ * not call any other VMA helpers while holding this lock.
+ *
+ * Note: You're in atomic-context while holding this lock!
+ *
+ * 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);
+ */
+static inline void drm_vma_offset_lock_lookup(struct drm_vma_offset_manager *mgr)
+{
+	read_lock(&mgr->vm_lock);
+}
+
+/**
+ * drm_vma_offset_unlock_lookup() - Unlock lookup for extended private use
+ * @mgr: Manager object
+ *
+ * Release lookup-lock. See drm_vma_offset_lock_lookup() for more information.
+ */
+static inline void drm_vma_offset_unlock_lookup(struct drm_vma_offset_manager *mgr)
+{
+	read_unlock(&mgr->vm_lock);
+}
+
+/**
+ * drm_vma_node_reset() - Initialize or reset node object
+ * @node: Node to initialize or reset
+ *
+ * Reset a node to its initial state. You must call this before using @node with
+ * any vma offset manager.
+ *
+ * This must not be called on an already allocated node, or you will leak
+ * memory.
+ */
+static inline void drm_vma_node_reset(struct drm_vma_offset_node *node)
+{
+	memset(node, 0, sizeof(*node));
+}
+
+/**
+ * drm_vma_node_start() - Return start address for page-based addressing
+ * @node: Node to inspect
+ *
+ * Return the start address of the given node. This can be used as offset into
+ * the linear VM space that is provided by the VMA offset manager. Note that
+ * this can only be used for page-based addressing. If you need a proper offset
+ * for user-space mappings, you must apply "<< PAGE_SHIFT" or use the
+ * drm_vma_node_offset_addr() helper instead.
+ *
+ * RETURNS:
+ * Start address of @node for page-based addressing. 0 if the node does not
+ * have an offset allocated.
+ */
+static inline unsigned long drm_vma_node_start(struct drm_vma_offset_node *node)
+{
+	return node->vm_node.start;
+}
+
+/**
+ * drm_vma_node_has_offset() - Check whether node is added to offset manager
+ * @node: Node to be checked
+ *
+ * RETURNS:
+ * true iff the node was previously allocated an offset and added to
+ * an vma offset manager.
+ */
+static inline bool drm_vma_node_has_offset(struct drm_vma_offset_node *node)
+{
+	return drm_mm_node_allocated(&node->vm_node);
+}
+
+/**
+ * drm_vma_node_offset_addr() - Return sanitized offset for user-space mmaps
+ * @node: Linked offset node
+ *
+ * Same as drm_vma_node_start() but returns the address as a valid offset that
+ * can be used for user-space mappings during mmap().
+ * This must not be called on unlinked nodes.
+ *
+ * RETURNS:
+ * Offset of @node for byte-based addressing. 0 if the node does not have an
+ * object allocated.
+ */
+static inline __u64 drm_vma_node_offset_addr(struct drm_vma_offset_node *node)
+{
+	return ((__u64)node->vm_node.start) << PAGE_SHIFT;
+}
+
+#endif /* __DRM_VMA_MANAGER_H__ */