From: Thomas Zimmermann <tzimmermann@suse.de>
To: daniel@ffwll.ch, noralf@tronnes.org, airlied@linux.ie,
rong.a.chen@intel.com, feng.tang@intel.com, ying.huang@intel.com,
sean@poorly.run, maxime.ripard@bootlin.com,
maarten.lankhorst@linux.intel.com, dave@stgolabs.net,
kraxel@redhat.com
Cc: Thomas Zimmermann <tzimmermann@suse.de>, dri-devel@lists.freedesktop.org
Subject: [PATCH v4 4/4] drm/vram: Implement lazy unmapping for GEM VRAM buffers
Date: Fri, 6 Sep 2019 14:20:56 +0200 [thread overview]
Message-ID: <20190906122056.32018-5-tzimmermann@suse.de> (raw)
In-Reply-To: <20190906122056.32018-1-tzimmermann@suse.de>
Frequent mapping and unmapping a buffer object adds overhead for
modifying the page table and creates debug output. Unmapping a buffer
is only required when the memory manager evicts the buffer from its
current location.
v4:
* WARN_ON if buffer is still mapped during BO cleanup
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
---
drivers/gpu/drm/drm_gem_vram_helper.c | 49 ++++++++++++++++++++++-----
include/drm/drm_gem_vram_helper.h | 4 +++
2 files changed, 45 insertions(+), 8 deletions(-)
diff --git a/drivers/gpu/drm/drm_gem_vram_helper.c b/drivers/gpu/drm/drm_gem_vram_helper.c
index 1e17f11cc7b9..0990f0e03213 100644
--- a/drivers/gpu/drm/drm_gem_vram_helper.c
+++ b/drivers/gpu/drm/drm_gem_vram_helper.c
@@ -28,6 +28,7 @@ static void drm_gem_vram_cleanup(struct drm_gem_vram_object *gbo)
*/
WARN_ON(gbo->kmap_use_count);
+ WARN_ON(gbo->kmap.virtual);
drm_gem_object_release(&gbo->bo.base);
}
@@ -356,18 +357,17 @@ EXPORT_SYMBOL(drm_gem_vram_kmap);
static void drm_gem_vram_kunmap_locked(struct drm_gem_vram_object *gbo)
{
- struct ttm_bo_kmap_obj *kmap = &gbo->kmap;
-
if (WARN_ON_ONCE(!gbo->kmap_use_count))
return;
if (--gbo->kmap_use_count > 0)
return;
- if (!kmap->virtual)
- return;
-
- ttm_bo_kunmap(kmap);
- kmap->virtual = NULL;
+ /*
+ * Permanently mapping and unmapping buffers adds overhead from
+ * updating the page tables and creates debugging output. Therefore,
+ * we delay the actual unmap operation until the BO gets evicted
+ * from memory. See drm_gem_vram_bo_driver_move_notify().
+ */
}
/**
@@ -497,6 +497,38 @@ int drm_gem_vram_bo_driver_verify_access(struct ttm_buffer_object *bo,
}
EXPORT_SYMBOL(drm_gem_vram_bo_driver_verify_access);
+/**
+ * drm_gem_vram_bo_driver_move_notify() -
+ * Implements &struct ttm_bo_driver.move_notify
+ * @bo: TTM buffer object. Refers to &struct drm_gem_vram_object.bo
+ * @evict: True, if the BO is being evicted from graphics memory;
+ * false otherwise.
+ * @new_mem: New memory region, or NULL on destruction
+ */
+void drm_gem_vram_bo_driver_move_notify(struct ttm_buffer_object *bo,
+ bool evict,
+ struct ttm_mem_reg *new_mem)
+{
+ struct drm_gem_vram_object *gbo;
+ struct ttm_bo_kmap_obj *kmap;
+
+ /* TTM may pass BOs that are not GEM VRAM BOs. */
+ if (!drm_is_gem_vram(bo))
+ return;
+
+ gbo = drm_gem_vram_of_bo(bo);
+ kmap = &gbo->kmap;
+
+ if (WARN_ON_ONCE(gbo->kmap_use_count))
+ return;
+
+ if (!kmap->virtual)
+ return;
+ ttm_bo_kunmap(kmap);
+ kmap->virtual = NULL;
+}
+EXPORT_SYMBOL(drm_gem_vram_bo_driver_move_notify);
+
/*
* drm_gem_vram_mm_funcs - Functions for &struct drm_vram_mm
*
@@ -506,7 +538,8 @@ EXPORT_SYMBOL(drm_gem_vram_bo_driver_verify_access);
*/
const struct drm_vram_mm_funcs drm_gem_vram_mm_funcs = {
.evict_flags = drm_gem_vram_bo_driver_evict_flags,
- .verify_access = drm_gem_vram_bo_driver_verify_access
+ .verify_access = drm_gem_vram_bo_driver_verify_access,
+ .move_notify = drm_gem_vram_bo_driver_move_notify,
};
EXPORT_SYMBOL(drm_gem_vram_mm_funcs);
diff --git a/include/drm/drm_gem_vram_helper.h b/include/drm/drm_gem_vram_helper.h
index 4f0e207ee097..b47c46516466 100644
--- a/include/drm/drm_gem_vram_helper.h
+++ b/include/drm/drm_gem_vram_helper.h
@@ -112,6 +112,10 @@ int drm_gem_vram_fill_create_dumb(struct drm_file *file,
void drm_gem_vram_bo_driver_evict_flags(struct ttm_buffer_object *bo,
struct ttm_placement *pl);
+void drm_gem_vram_bo_driver_move_notify(struct ttm_buffer_object *bo,
+ bool evict,
+ struct ttm_mem_reg *new_mem);
+
int drm_gem_vram_bo_driver_verify_access(struct ttm_buffer_object *bo,
struct file *filp);
--
2.23.0
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
next prev parent reply other threads:[~2019-09-06 12:21 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-09-06 12:20 [PATCH v4 0/4] Implement lazy unmapping for GEM VRAM buffers Thomas Zimmermann
2019-09-06 12:20 ` [PATCH v4 1/4] drm/vram: Add kmap ref-counting to GEM VRAM objects Thomas Zimmermann
2019-09-06 12:20 ` [PATCH v4 2/4] drm/vram: Acquire lock only once per call to vmap()/vunmap() Thomas Zimmermann
2019-09-06 12:20 ` [PATCH v4 3/4] drm/vram: Add infrastructure for move_notify() Thomas Zimmermann
2019-09-06 12:20 ` Thomas Zimmermann [this message]
2019-09-06 18:33 ` [PATCH v4 0/4] Implement lazy unmapping for GEM VRAM buffers Daniel Vetter
2019-09-06 18:39 ` Davidlohr Bueso
2019-09-09 9:47 ` Gerd Hoffmann
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=20190906122056.32018-5-tzimmermann@suse.de \
--to=tzimmermann@suse.de \
--cc=airlied@linux.ie \
--cc=daniel@ffwll.ch \
--cc=dave@stgolabs.net \
--cc=dri-devel@lists.freedesktop.org \
--cc=feng.tang@intel.com \
--cc=kraxel@redhat.com \
--cc=maarten.lankhorst@linux.intel.com \
--cc=maxime.ripard@bootlin.com \
--cc=noralf@tronnes.org \
--cc=rong.a.chen@intel.com \
--cc=sean@poorly.run \
--cc=ying.huang@intel.com \
/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