From: Mika Kuoppala <mika.kuoppala@linux.intel.com>
To: Chris Wilson <chris@chris-wilson.co.uk>, intel-gfx@lists.freedesktop.org
Subject: Re: [PATCH 18/23] drm/i915: Keep all partially allocated HWSP on a freelist
Date: Fri, 18 Jan 2019 14:12:58 +0200 [thread overview]
Message-ID: <87h8e6w4dx.fsf@gaia.fi.intel.com> (raw)
In-Reply-To: <20190117143519.16086-19-chris@chris-wilson.co.uk>
Chris Wilson <chris@chris-wilson.co.uk> writes:
> Keep track of partially allocated pages for use in allocating future
> timeline HWSP. This is still without migration, so it is possible for
timeline from HWSP?
> the system to end up with each timeline in its own page, but we ensure
> that no new allocation would needless allocate a fresh page!
s/needless/needlessly.
-Mika
>
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> Cc: John Harrison <John.C.Harrison@Intel.com>
> ---
> drivers/gpu/drm/i915/i915_drv.h | 3 +-
> drivers/gpu/drm/i915/i915_timeline.c | 81 +++++++++++++++++-----------
> 2 files changed, 50 insertions(+), 34 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index d59228dabb6e..0bebef428f1e 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -1981,8 +1981,7 @@ struct drm_i915_private {
>
> /* Pack multiple timelines' seqnos into the same page */
> spinlock_t hwsp_lock;
> - struct i915_vma *hwsp;
> - u64 bitmap;
> + struct list_head hwsp;
> } timelines;
>
> struct list_head active_rings;
> diff --git a/drivers/gpu/drm/i915/i915_timeline.c b/drivers/gpu/drm/i915/i915_timeline.c
> index e939a9e1a4ab..64bb1ce24318 100644
> --- a/drivers/gpu/drm/i915/i915_timeline.c
> +++ b/drivers/gpu/drm/i915/i915_timeline.c
> @@ -9,74 +9,94 @@
> #include "i915_timeline.h"
> #include "i915_syncmap.h"
>
> +struct i915_timeline_hwsp {
> + struct list_head link;
> + struct i915_vma *vma;
> + u64 bitmap;
> +};
> +
> static int hwsp_alloc(struct i915_timeline *timeline)
> {
> -#define NBITS BITS_PER_TYPE(typeof(gt->bitmap))
> struct drm_i915_private *i915 = timeline->i915;
> struct i915_gt_timelines *gt = &i915->gt.timelines;
> - struct i915_vma *vma;
> + struct i915_timeline_hwsp *hwsp;
> int offset;
>
> spin_lock(>->hwsp_lock);
>
> -restart:
> - offset = find_first_bit((unsigned long *)>->bitmap, NBITS);
> - if (offset == NBITS && gt->hwsp) {
> - i915_vma_put(gt->hwsp);
> - gt->hwsp = NULL;
> - }
> -
> - vma = gt->hwsp;
> - if (!vma) {
> + hwsp = list_first_entry_or_null(>->hwsp, typeof(*hwsp), link);
> + if (!hwsp) {
> struct drm_i915_gem_object *bo;
> + struct i915_vma *vma;
>
> spin_unlock(>->hwsp_lock);
>
> - BUILD_BUG_ON(NBITS * CACHELINE_BYTES > PAGE_SIZE);
> + hwsp = kmalloc(sizeof(*hwsp), GFP_KERNEL);
> + if (!hwsp)
> + return -ENOMEM;
> +
> + BUILD_BUG_ON(BITS_PER_TYPE(hwsp->bitmap) * CACHELINE_BYTES > PAGE_SIZE);
> bo = i915_gem_object_create_internal(i915, PAGE_SIZE);
> - if (IS_ERR(bo))
> + if (IS_ERR(bo)) {
> + kfree(hwsp);
> return PTR_ERR(bo);
> + }
>
> i915_gem_object_set_cache_level(bo, I915_CACHE_LLC);
>
> vma = i915_vma_instance(bo, &i915->ggtt.vm, NULL);
> if (IS_ERR(vma)) {
> i915_gem_object_put(bo);
> + kfree(hwsp);
> return PTR_ERR(vma);
> }
>
> - spin_lock(>->hwsp_lock);
> - if (gt->hwsp) {
> - i915_gem_object_put(bo);
> - goto restart;
> - }
> + vma->private = hwsp;
> + hwsp->vma = vma;
> + hwsp->bitmap = ~0ull;
>
> - gt->hwsp = vma;
> - gt->bitmap = ~0ull;
> - offset = 0;
> + spin_lock(>->hwsp_lock);
> + list_add(&hwsp->link, >->hwsp);
> }
>
> - gt->bitmap &= ~BIT_ULL(offset);
> + GEM_BUG_ON(!hwsp->bitmap);
> + offset = __ffs64(hwsp->bitmap);
> + hwsp->bitmap &= ~BIT_ULL(offset);
> + if (!hwsp->bitmap)
> + list_del(&hwsp->link);
>
> spin_unlock(>->hwsp_lock);
>
> - timeline->hwsp_ggtt = i915_vma_get(vma);
> + timeline->hwsp_ggtt = i915_vma_get(hwsp->vma);
> timeline->hwsp_offset = offset * CACHELINE_BYTES;
>
> + GEM_BUG_ON(timeline->hwsp_ggtt->private != hwsp);
> +
> return 0;
> -#undef NBITS
> }
>
> static void hwsp_free(struct i915_timeline *timeline)
> {
> struct i915_gt_timelines *gt = &timeline->i915->gt.timelines;
> + struct i915_timeline_hwsp *hwsp;
>
> - if (timeline->hwsp_ggtt != gt->hwsp)
> + hwsp = timeline->hwsp_ggtt->private;
> + if (!hwsp)
> return;
>
> spin_lock(>->hwsp_lock);
> - if (timeline->hwsp_ggtt == gt->hwsp)
> - gt->bitmap |= BIT_ULL(timeline->hwsp_offset / CACHELINE_BYTES);
> +
> + if (!hwsp->bitmap)
> + list_add_tail(&hwsp->link, >->hwsp);
> +
> + hwsp->bitmap |= BIT_ULL(timeline->hwsp_offset / CACHELINE_BYTES);
> +
> + if (hwsp->bitmap == ~0ull) {
> + i915_vma_put(hwsp->vma);
> + list_del(&hwsp->link);
> + kfree(hwsp);
> + }
> +
> spin_unlock(>->hwsp_lock);
> }
>
> @@ -148,6 +168,7 @@ void i915_timelines_init(struct drm_i915_private *i915)
> INIT_LIST_HEAD(>->list);
>
> spin_lock_init(>->hwsp_lock);
> + INIT_LIST_HEAD(>->hwsp);
>
> /* via i915_gem_wait_for_idle() */
> i915_gem_shrinker_taints_mutex(i915, >->mutex);
> @@ -264,13 +285,9 @@ void __i915_timeline_free(struct kref *kref)
> void i915_timelines_fini(struct drm_i915_private *i915)
> {
> struct i915_gt_timelines *gt = &i915->gt.timelines;
> - struct i915_vma *vma;
>
> GEM_BUG_ON(!list_empty(>->list));
> -
> - vma = fetch_and_zero(&i915->gt.timelines.hwsp);
> - if (vma)
> - i915_vma_put(vma);
> + GEM_BUG_ON(!list_empty(>->hwsp));
>
> mutex_destroy(>->mutex);
> }
> --
> 2.20.1
>
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
next prev parent reply other threads:[~2019-01-18 12:14 UTC|newest]
Thread overview: 50+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-01-17 14:34 Swapping a single global interrupt handler for a herd Chris Wilson
2019-01-17 14:34 ` [PATCH 01/23] drm/i915: Make all GPU resets atomic Chris Wilson
2019-01-17 14:34 ` [PATCH 02/23] drm/i915/guc: Disable global reset Chris Wilson
2019-01-17 14:34 ` [PATCH 03/23] drm/i915: Remove GPU reset dependence on struct_mutex Chris Wilson
2019-01-17 14:34 ` [PATCH 04/23] drm/i915/selftests: Trim struct_mutex duration for set-wedged selftest Chris Wilson
2019-01-17 14:34 ` [PATCH 05/23] drm/i915: Issue engine resets onto idle engines Chris Wilson
2019-01-18 12:06 ` Mika Kuoppala
2019-01-17 14:34 ` [PATCH 06/23] drm/i915: Stop tracking MRU activity on VMA Chris Wilson
2019-01-17 14:34 ` [PATCH 07/23] drm/i915: Pull VM lists under the VM mutex Chris Wilson
2019-01-18 10:07 ` Tvrtko Ursulin
2019-01-17 14:34 ` [PATCH 08/23] drm/i915: Move vma lookup to its own lock Chris Wilson
2019-01-17 16:27 ` Tvrtko Ursulin
2019-01-17 16:31 ` Chris Wilson
2019-01-17 16:36 ` Chris Wilson
2019-01-17 16:51 ` Tvrtko Ursulin
2019-01-17 16:44 ` Chris Wilson
2019-01-17 14:34 ` [PATCH 09/23] drm/i915: Use b->irq_enable() as predicate for mock engine Chris Wilson
2019-01-17 16:44 ` Tvrtko Ursulin
2019-01-17 16:52 ` Chris Wilson
2019-01-17 18:00 ` Tvrtko Ursulin
2019-01-17 14:34 ` [PATCH 10/23] drm/i915/selftests: Allocate mock ring/timeline per context Chris Wilson
2019-01-17 14:34 ` [PATCH 11/23] drm/i915/selftests: Make evict tolerant of foreign objects Chris Wilson
2019-01-17 17:29 ` Tvrtko Ursulin
2019-01-18 11:23 ` Chris Wilson
2019-01-17 14:34 ` [PATCH 12/23] drm/i915: Always allocate an object/vma for the HWSP Chris Wilson
2019-01-17 14:34 ` [PATCH 13/23] drm/i915: Move list of timelines under its own lock Chris Wilson
2019-01-17 17:54 ` Tvrtko Ursulin
2019-01-18 11:31 ` Chris Wilson
2019-01-17 14:34 ` [PATCH 14/23] drm/i915: Introduce concept of per-timeline (context) HWSP Chris Wilson
2019-01-18 10:18 ` Tvrtko Ursulin
2019-01-17 14:34 ` [PATCH 15/23] drm/i915: Enlarge vma->pin_count Chris Wilson
2019-01-17 14:34 ` [PATCH 16/23] drm/i915: Allocate a status page for each timeline Chris Wilson
2019-01-18 11:19 ` Tvrtko Ursulin
2019-01-17 14:34 ` [PATCH 17/23] drm/i915: Share per-timeline HWSP using a slab suballocator Chris Wilson
2019-01-18 12:08 ` Tvrtko Ursulin
2019-01-17 14:34 ` [PATCH 18/23] drm/i915: Keep all partially allocated HWSP on a freelist Chris Wilson
2019-01-18 12:12 ` Mika Kuoppala [this message]
2019-01-18 12:25 ` Tvrtko Ursulin
2019-01-17 14:35 ` [PATCH 19/23] drm/i915: Track the context's seqno in its own timeline HWSP Chris Wilson
2019-01-18 14:10 ` Tvrtko Ursulin
2019-01-17 14:35 ` [PATCH 20/23] drm/i915: Identify active requests Chris Wilson
2019-01-17 14:35 ` [PATCH 21/23] drm/i915: Remove the intel_engine_notify tracepoint Chris Wilson
2019-01-17 14:35 ` [PATCH 22/23] drm/i915: Replace global breadcrumbs with per-context interrupt tracking Chris Wilson
2019-01-17 14:35 ` [PATCH 23/23] drm/i915: Drop fake breadcrumb irq Chris Wilson
2019-01-17 14:37 ` Swapping a single global interrupt handler for a herd Chris Wilson
2019-01-17 14:56 ` ✗ Fi.CI.CHECKPATCH: warning for series starting with [01/23] drm/i915: Make all GPU resets atomic Patchwork
2019-01-17 15:05 ` ✗ Fi.CI.SPARSE: " Patchwork
2019-01-17 15:38 ` ✓ Fi.CI.BAT: success " Patchwork
2019-01-17 23:36 ` ✗ Fi.CI.IGT: failure " Patchwork
2019-01-17 23:41 ` Chris Wilson
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=87h8e6w4dx.fsf@gaia.fi.intel.com \
--to=mika.kuoppala@linux.intel.com \
--cc=chris@chris-wilson.co.uk \
--cc=intel-gfx@lists.freedesktop.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox