[Intel-gfx] [RFC PATCH 1/3] drm: Extract amdgpu_sa.c as a generic suballocation helper
Christian König
christian.koenig at amd.com
Fri Feb 4 18:29:50 UTC 2022
Oh, that's on my TODO list for years!
Am 04.02.22 um 18:48 schrieb Maarten Lankhorst:
> Suballocating a buffer object is something that is not driver
> generic, and is useful for other drivers as well.
>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst at linux.intel.com>
> ---
> drivers/gpu/drm/Makefile | 4 +-
> drivers/gpu/drm/drm_suballoc.c | 424 +++++++++++++++++++++++++++++++++
> include/drm/drm_suballoc.h | 78 ++++++
> 3 files changed, 505 insertions(+), 1 deletion(-)
> create mode 100644 drivers/gpu/drm/drm_suballoc.c
> create mode 100644 include/drm/drm_suballoc.h
>
> diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile
> index 8675c2af7ae1..b848bcf8790c 100644
> --- a/drivers/gpu/drm/Makefile
> +++ b/drivers/gpu/drm/Makefile
> @@ -57,7 +57,9 @@ drm_kms_helper-y := drm_bridge_connector.o drm_crtc_helper.o \
> drm_scdc_helper.o drm_gem_atomic_helper.o \
> drm_gem_framebuffer_helper.o \
> drm_atomic_state_helper.o drm_damage_helper.o \
> - drm_format_helper.o drm_self_refresh_helper.o drm_rect.o
> + drm_format_helper.o drm_self_refresh_helper.o drm_rect.o \
> + drm_suballoc.o
> +
I think we should put that into a separate module like we now do with
other helpers as well.
> drm_kms_helper-$(CONFIG_DRM_PANEL_BRIDGE) += bridge/panel.o
> drm_kms_helper-$(CONFIG_DRM_FBDEV_EMULATION) += drm_fb_helper.o
>
> diff --git a/drivers/gpu/drm/drm_suballoc.c b/drivers/gpu/drm/drm_suballoc.c
> new file mode 100644
> index 000000000000..e0bb35367b71
> --- /dev/null
> +++ b/drivers/gpu/drm/drm_suballoc.c
> @@ -0,0 +1,424 @@
> +/*
> + * Copyright 2011 Red Hat Inc.
> + * All Rights Reserved.
> + *
> + * 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, sub license, 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 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 NON-INFRINGEMENT. IN NO EVENT SHALL
> + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS 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.
> + *
> + * The above copyright notice and this permission notice (including the
> + * next paragraph) shall be included in all copies or substantial portions
> + * of the Software.
> + *
> + */
> +/*
> + * Authors:
> + * Jerome Glisse <glisse at freedesktop.org>
> + */
That is hopelessly outdated. IIRC I completely rewrote that stuff in ~2012.
> +/* Algorithm:
> + *
> + * We store the last allocated bo in "hole", we always try to allocate
> + * after the last allocated bo. Principle is that in a linear GPU ring
> + * progression was is after last is the oldest bo we allocated and thus
> + * the first one that should no longer be in use by the GPU.
> + *
> + * If it's not the case we skip over the bo after last to the closest
> + * done bo if such one exist. If none exist and we are not asked to
> + * block we report failure to allocate.
> + *
> + * If we are asked to block we wait on all the oldest fence of all
> + * rings. We just wait for any of those fence to complete.
> + */
> +
> +#include <drm/drm_suballoc.h>
> +#include <drm/drm_print.h>
> +#include <linux/slab.h>
> +#include <linux/sched.h>
> +#include <linux/wait.h>
> +#include <linux/dma-fence.h>
> +
> +static void drm_suballoc_remove_locked(struct drm_suballoc *sa);
> +static void drm_suballoc_try_free(struct drm_suballoc_manager *sa_manager);
> +
> +/**
> + * drm_suballoc_manager_init - Initialise the drm_suballoc_manager
> + *
> + * @sa_manager: pointer to the sa_manager
> + * @size: number of bytes we want to suballocate
> + * @align: alignment for each suballocated chunk
> + *
> + * Prepares the suballocation manager for suballocations.
> + */
> +void drm_suballoc_manager_init(struct drm_suballoc_manager *sa_manager,
> + u32 size, u32 align)
> +{
> + u32 i;
> +
> + if (!align)
> + align = 1;
> +
> + /* alignment must be a power of 2 */
> + BUG_ON(align & (align - 1));
When we move that I think we should cleanup the code once more, e.g. use
is_power_of_2() function here for example.
There are also a bunch of places with extra {} and constructs like "if
(....) return true; else return false;" which could certainly be simplified.
Apart from that really great idea.
Regards,
Christian.
> +
> + init_waitqueue_head(&sa_manager->wq);
> + sa_manager->size = size;
> + sa_manager->align = align;
> + sa_manager->hole = &sa_manager->olist;
> + INIT_LIST_HEAD(&sa_manager->olist);
> + for (i = 0; i < DRM_SUBALLOC_MAX_QUEUES; ++i)
> + INIT_LIST_HEAD(&sa_manager->flist[i]);
> +}
> +EXPORT_SYMBOL(drm_suballoc_manager_init);
> +
> +/**
> + * drm_suballoc_manager_fini - Destroy the drm_suballoc_manager
> + *
> + * @sa_manager: pointer to the sa_manager
> + *
> + * Cleans up the suballocation manager after use. All fences added
> + * with drm_suballoc_free() must be signaled, or we cannot clean up
> + * the entire manager.
> + */
> +void drm_suballoc_manager_fini(struct drm_suballoc_manager *sa_manager)
> +{
> + struct drm_suballoc *sa, *tmp;
> +
> + if (!sa_manager->size)
> + return;
> +
> + if (!list_empty(&sa_manager->olist)) {
> + sa_manager->hole = &sa_manager->olist,
> + drm_suballoc_try_free(sa_manager);
> + if (!list_empty(&sa_manager->olist))
> + DRM_ERROR("sa_manager is not empty, clearing anyway\n");
> + }
> + list_for_each_entry_safe(sa, tmp, &sa_manager->olist, olist) {
> + drm_suballoc_remove_locked(sa);
> + }
> +
> + sa_manager->size = 0;
> +}
> +EXPORT_SYMBOL(drm_suballoc_manager_fini);
> +
> +static void drm_suballoc_remove_locked(struct drm_suballoc *sa)
> +{
> + struct drm_suballoc_manager *sa_manager = sa->manager;
> + if (sa_manager->hole == &sa->olist) {
> + sa_manager->hole = sa->olist.prev;
> + }
> + list_del_init(&sa->olist);
> + list_del_init(&sa->flist);
> + dma_fence_put(sa->fence);
> + kfree(sa);
> +}
> +
> +static void drm_suballoc_try_free(struct drm_suballoc_manager *sa_manager)
> +{
> + struct drm_suballoc *sa, *tmp;
> +
> + if (sa_manager->hole->next == &sa_manager->olist)
> + return;
> +
> + sa = list_entry(sa_manager->hole->next, struct drm_suballoc, olist);
> + list_for_each_entry_safe_from(sa, tmp, &sa_manager->olist, olist) {
> + if (sa->fence == NULL ||
> + !dma_fence_is_signaled(sa->fence)) {
> + return;
> + }
> + drm_suballoc_remove_locked(sa);
> + }
> +}
> +
> +static inline unsigned drm_suballoc_hole_soffset(struct drm_suballoc_manager *sa_manager)
> +{
> + struct list_head *hole = sa_manager->hole;
> +
> + if (hole != &sa_manager->olist) {
> + return list_entry(hole, struct drm_suballoc, olist)->eoffset;
> + }
> + return 0;
> +}
> +
> +static inline unsigned drm_suballoc_hole_eoffset(struct drm_suballoc_manager *sa_manager)
> +{
> + struct list_head *hole = sa_manager->hole;
> +
> + if (hole->next != &sa_manager->olist) {
> + return list_entry(hole->next, struct drm_suballoc, olist)->soffset;
> + }
> + return sa_manager->size;
> +}
> +
> +static bool drm_suballoc_try_alloc(struct drm_suballoc_manager *sa_manager,
> + struct drm_suballoc *sa,
> + unsigned size)
> +{
> + unsigned soffset, eoffset;
> +
> + soffset = drm_suballoc_hole_soffset(sa_manager);
> + eoffset = drm_suballoc_hole_eoffset(sa_manager);
> +
> + if ((eoffset - soffset) >= size) {
> + sa->manager = sa_manager;
> + sa->soffset = soffset;
> + sa->eoffset = soffset + size;
> + list_add(&sa->olist, sa_manager->hole);
> + INIT_LIST_HEAD(&sa->flist);
> + sa_manager->hole = &sa->olist;
> + return true;
> + }
> + return false;
> +}
> +
> +/**
> + * drm_suballoc_event - Check if we can stop waiting
> + *
> + * @sa_manager: pointer to the sa_manager
> + * @size: number of bytes we want to allocate
> + * @align: alignment we need to match
> + *
> + * Check if either there is a fence we can wait for or
> + * enough free memory to satisfy the allocation directly
> + */
> +static bool drm_suballoc_event(struct drm_suballoc_manager *sa_manager,
> + u32 size)
> +{
> + unsigned soffset, eoffset, i;
> +
> + for (i = 0; i < DRM_SUBALLOC_MAX_QUEUES; ++i)
> + if (!list_empty(&sa_manager->flist[i]))
> + return true;
> +
> + soffset = drm_suballoc_hole_soffset(sa_manager);
> + eoffset = drm_suballoc_hole_eoffset(sa_manager);
> +
> + if ((eoffset - soffset) >= size) {
> + return true;
> + }
> +
> + return false;
> +}
> +
> +static bool drm_suballoc_next_hole(struct drm_suballoc_manager *sa_manager,
> + struct dma_fence **fences,
> + unsigned *tries)
> +{
> + struct drm_suballoc *best_bo = NULL;
> + unsigned i, best_idx, soffset, best, tmp;
> +
> + /* if hole points to the end of the buffer */
> + if (sa_manager->hole->next == &sa_manager->olist) {
> + /* try again with its beginning */
> + sa_manager->hole = &sa_manager->olist;
> + return true;
> + }
> +
> + soffset = drm_suballoc_hole_soffset(sa_manager);
> + /* to handle wrap around we add sa_manager->size */
> + best = sa_manager->size * 2;
> + /* go over all fence list and try to find the closest sa
> + * of the current last
> + */
> + for (i = 0; i < DRM_SUBALLOC_MAX_QUEUES; ++i) {
> + struct drm_suballoc *sa;
> +
> + fences[i] = NULL;
> +
> + if (list_empty(&sa_manager->flist[i]))
> + continue;
> +
> + sa = list_first_entry(&sa_manager->flist[i],
> + struct drm_suballoc, flist);
> +
> + if (!dma_fence_is_signaled(sa->fence)) {
> + fences[i] = sa->fence;
> + continue;
> + }
> +
> + /* limit the number of tries each freelist gets */
> + if (tries[i] > 2) {
> + continue;
> + }
> +
> + tmp = sa->soffset;
> + if (tmp < soffset) {
> + /* wrap around, pretend it's after */
> + tmp += sa_manager->size;
> + }
> + tmp -= soffset;
> + if (tmp < best) {
> + /* this sa bo is the closest one */
> + best = tmp;
> + best_idx = i;
> + best_bo = sa;
> + }
> + }
> +
> + if (best_bo) {
> + ++tries[best_idx];
> + sa_manager->hole = best_bo->olist.prev;
> +
> + /* we knew that this one is signaled,
> + so it's save to remote it */
> + drm_suballoc_remove_locked(best_bo);
> + return true;
> + }
> + return false;
> +}
> +
> +/**
> + * drm_suballoc_new - Make a suballocation.
> + *
> + * @sa_manager: pointer to the sa_manager
> + * @size: number of bytes we want to suballocate.
> + *
> + * Try to make a suballocation of size @size, which will be rounded
> + * up to the alignment specified in specified in drm_suballoc_manager_init().
> + *
> + * Returns a new suballocated bo, or an ERR_PTR.
> + */
> +struct drm_suballoc *
> +drm_suballoc_new(struct drm_suballoc_manager *sa_manager, u32 size)
> +{
> + struct dma_fence *fences[DRM_SUBALLOC_MAX_QUEUES];
> + unsigned tries[DRM_SUBALLOC_MAX_QUEUES];
> + unsigned count;
> + int i, r;
> + struct drm_suballoc *sa;
> +
> + size = ALIGN(size, sa_manager->align);
> + if (WARN_ON_ONCE(size > sa_manager->size))
> + return ERR_PTR(-EINVAL);
> +
> + sa = kmalloc(sizeof(struct drm_suballoc), GFP_KERNEL);
> + if (!sa)
> + return ERR_PTR(-ENOMEM);
> + sa->manager = sa_manager;
> + sa->fence = NULL;
> + INIT_LIST_HEAD(&sa->olist);
> + INIT_LIST_HEAD(&sa->flist);
> +
> + spin_lock(&sa_manager->wq.lock);
> + do {
> + for (i = 0; i < DRM_SUBALLOC_MAX_QUEUES; ++i)
> + tries[i] = 0;
> +
> + do {
> + drm_suballoc_try_free(sa_manager);
> +
> + if (drm_suballoc_try_alloc(sa_manager, sa,
> + size)) {
> + spin_unlock(&sa_manager->wq.lock);
> + return sa;
> + }
> +
> + /* see if we can skip over some allocations */
> + } while (drm_suballoc_next_hole(sa_manager, fences, tries));
> +
> + for (i = 0, count = 0; i < DRM_SUBALLOC_MAX_QUEUES; ++i)
> + if (fences[i])
> + fences[count++] = dma_fence_get(fences[i]);
> +
> + if (count) {
> + long t;
> +
> + spin_unlock(&sa_manager->wq.lock);
> + t = dma_fence_wait_any_timeout(fences, count, true,
> + MAX_SCHEDULE_TIMEOUT,
> + NULL);
> + for (i = 0; i < count; ++i)
> + dma_fence_put(fences[i]);
> +
> + r = (t > 0) ? 0 : t;
> + spin_lock(&sa_manager->wq.lock);
> + } else {
> + /* if we have nothing to wait for block */
> + r = wait_event_interruptible_locked(
> + sa_manager->wq,
> + drm_suballoc_event(sa_manager, size)
> + );
> + }
> +
> + } while (!r);
> +
> + spin_unlock(&sa_manager->wq.lock);
> + kfree(sa);
> + return ERR_PTR(r);
> +}
> +EXPORT_SYMBOL(drm_suballoc_new);
> +
> +/**
> + * drm_suballoc_free - Free a suballocation
> + *
> + * @suballoc: pointer to the suballocation
> + * @fence: fence that signals when suballocation is idle
> + * @queue: the index to which queue the suballocation will be placed on the free list.
> + *
> + * Free the suballocation. The suballocation can be re-used after @fence signals.
> + * @queue is used to allow waiting on multiple fence contexts in parallel in
> + * drm_suballoc_new().
> + */
> +void drm_suballoc_free(struct drm_suballoc *suballoc,
> + struct dma_fence *fence,
> + u32 queue)
> +{
> + struct drm_suballoc_manager *sa_manager;
> +
> + if (!suballoc)
> + return;
> +
> + sa_manager = suballoc->manager;
> + BUG_ON(queue >= DRM_SUBALLOC_MAX_QUEUES);
> +
> + spin_lock(&sa_manager->wq.lock);
> + if (fence && !dma_fence_is_signaled(fence)) {
> + suballoc->fence = dma_fence_get(fence);
> + list_add_tail(&suballoc->flist, &sa_manager->flist[queue]);
> + } else {
> + drm_suballoc_remove_locked(suballoc);
> + }
> + wake_up_all_locked(&sa_manager->wq);
> + spin_unlock(&sa_manager->wq.lock);
> +}
> +EXPORT_SYMBOL(drm_suballoc_free);
> +
> +#ifdef CONFIG_DEBUG_FS
> +void drm_suballoc_dump_debug_info(struct drm_suballoc_manager *sa_manager,
> + struct seq_file *m, u64 suballoc_base)
> +{
> + struct drm_suballoc *i;
> +
> + spin_lock(&sa_manager->wq.lock);
> + list_for_each_entry(i, &sa_manager->olist, olist) {
> + uint64_t soffset = i->soffset;
> + uint64_t eoffset = i->eoffset;
> + if (&i->olist == sa_manager->hole) {
> + seq_printf(m, ">");
> + } else {
> + seq_printf(m, " ");
> + }
> + seq_printf(m, "[0x%010llx 0x%010llx] size %8lld",
> + suballoc_base + soffset, suballoc_base + eoffset, eoffset - soffset);
> +
> + if (i->fence)
> + seq_printf(m, " protected by 0x%016llx on context %llu",
> + i->fence->seqno, i->fence->context);
> +
> + seq_printf(m, "\n");
> + }
> + spin_unlock(&sa_manager->wq.lock);
> +}
> +EXPORT_SYMBOL(drm_suballoc_dump_debug_info);
> +#endif
> diff --git a/include/drm/drm_suballoc.h b/include/drm/drm_suballoc.h
> new file mode 100644
> index 000000000000..846c4a792fac
> --- /dev/null
> +++ b/include/drm/drm_suballoc.h
> @@ -0,0 +1,78 @@
> +/*
> + * SPDX-License-Identifier: MIT
> + *
> + * Copyright © 2022 Intel Corporation
> + */
> +#ifndef _DRM_SUBALLOC_H_
> +#define _DRM_SUBALLOC_H_
> +
> +#include <linux/types.h>
> +#include <linux/list.h>
> +#include <linux/wait.h>
> +
> +struct dma_fence;
> +struct seq_file;
> +
> +/* sub-allocation manager, it has to be protected by another lock.
> + * By conception this is an helper for other part of the driver
> + * like the indirect buffer or semaphore, which both have their
> + * locking.
> + *
> + * Principe is simple, we keep a list of sub allocation in offset
> + * order (first entry has offset == 0, last entry has the highest
> + * offset).
> + *
> + * When allocating new object we first check if there is room at
> + * the end total_size - (last_object_offset + last_object_size) >=
> + * alloc_size. If so we allocate new object there.
> + *
> + * When there is not enough room at the end, we start waiting for
> + * each sub object until we reach object_offset+object_size >=
> + * alloc_size, this object then become the sub object we return.
> + *
> + * Alignment can't be bigger than page size.
> + *
> + * Hole are not considered for allocation to keep things simple.
> + * Assumption is that there won't be hole (all object on same
> + * alignment).
> + *
> + * The actual buffer object handling depends on the driver,
> + * and is not part of the helper implementation.
> + */
> +#define DRM_SUBALLOC_MAX_QUEUES 32
> +
> +struct drm_suballoc_manager {
> + wait_queue_head_t wq;
> + struct list_head *hole, olist, flist[DRM_SUBALLOC_MAX_QUEUES];
> + u32 size, align;
> +};
> +
> +/* sub-allocation buffer */
> +struct drm_suballoc {
> + struct list_head olist, flist;
> + struct drm_suballoc_manager *manager;
> + u32 soffset, eoffset;
> + struct dma_fence *fence;
> +};
> +
> +void drm_suballoc_manager_init(struct drm_suballoc_manager *sa_manager,
> + u32 size, u32 align);
> +void drm_suballoc_manager_fini(struct drm_suballoc_manager *sa_manager);
> +struct drm_suballoc *drm_suballoc_new(struct drm_suballoc_manager *sa_manager,
> + u32 size);
> +void drm_suballoc_free(struct drm_suballoc *sa_bo,
> + struct dma_fence *fence,
> + u32 queue);
> +
> +#ifdef CONFIG_DEBUG_FS
> +void drm_suballoc_dump_debug_info(struct drm_suballoc_manager *sa_manager,
> + struct seq_file *m, u64 suballoc_base);
> +#else
> +static inline void
> +drm_suballoc_dump_debug_info(struct drm_suballoc_manager *sa_manager,
> + struct seq_file *m, u64 suballoc_base)
> +{ }
> +
> +#endif
> +
> +#endif /* _DRM_SUBALLOC_H_ */
More information about the Intel-gfx
mailing list