similar to: [Patches][nouveau/ddx]: Improvements to bufferswap implementation and timestamping v2

Hi, here a set of patches against the nouveau-ddx. This is an extended and revised set, based on Francisco Jerez feedback from autumn last year. [1/9] Makes pageflipping work again on X-Server 1.12rc. It apparently stopped working somewhere around Xorg 1.11+. [2/9] Implements handling of pageflip completion events from the kernel. Francisco Jerez argument against including it was that the

Port of commit ae45d7e6d8e6844cd4586c9ee97c21b257fa788f in xf86-video-ati. Fixes https://bugs.freedesktop.org/show_bug.cgi?id=49351 (Additionally, don't try to pageflip if user disabled it in xorg.conf. Currently this change is a no-op, because can_exchange returns true only when page flipping is enabled, but commit 169512fbe91f0671a90dfee5e280357f0a4ef701 - which changed can_exchange

A short series of compiler visibility warning fixes that I prepared whilst trialing improvements to xf86-video-nouveau's use of the core xorg-server utility macros. Rhys Kidd (4): wfb: Remove declaration for undefined function nouveau_wfb_init() dri2: Mark local create/destroy buffer and copy region functions as static xv: Mark local NVSetupTexturedVideo function as static

Hi all, the following patch fixes a race-condition in the nouveau ddx which caused frequent x-server crashes for at least 1 user under some loads when OpenGL triple-buffering is enabled, which it is by default on XOrg 1.12 and later. As a side effect, it provides a small optimization for the common case of bufferswap at next vblank. The other way to avoid the race + possible crash is to set

Based on the original patch by Christoph Bumiller, but since it depends on kernel support patched I cannot push it yet. The changes are that I enable vblank by default, and offset takes OFFSET_HIGH/LOW instead of something relative to notifier_bo. Signed-off-by: Maarten Lankhorst <maarten.lankhorst at canonical.com> --- diff --git a/man/nouveau.man b/man/nouveau.man index 7c72907..8765569

With this patch the DDX almost works with GK20A, the missing piece is adding COHERENT mappings to the right places. ;-) If you specify NOUVEAU_BO_APER the kernel will truncate valid_domains to the domains specified at creation time. This means that as long as we only specify the correct domain in nouveau_allocate_surface the effect is still the same. Signed-off-by: Maarten Lankhorst <dev at

It turns out there's a difference in X between a CRTC being "disabled" and simply having it's DPMS turned off. This is problematic though, because if DPMS is turned off you can't really use the CRTC as a normal CRTC anyway since page flipping and vblanks will be non-functional. As a result, we've been considering DPMS-on CRTCs as enabled and attempt to perform pageflips,

--- src/nouveau_dri2.c | 15 ++++++++++----- 1 file changed, 10 insertions(+), 5 deletions(-) diff --git a/src/nouveau_dri2.c b/src/nouveau_dri2.c index f22e319..4398559 100644 --- a/src/nouveau_dri2.c +++ b/src/nouveau_dri2.c @@ -142,6 +142,7 @@ nouveau_dri2_copy_region2(ScreenPtr pScreen, DrawablePtr pDraw, RegionPtr pRegio NVPtr pNv = NVPTR(xf86ScreenToScrn(pScreen)); RegionPtr pCopyClip;

Well, I don't pretend to know anything about X, but this is the commit that added the code in question: commit 297fd0d0755bda698be1d0b30cc60a41d7673c0b Author: Dave Airlie <airlied at redhat.com> Date: Tue Oct 16 16:15:16 2012 +1000 nouveau/dri2: fix pixmap/window offset calcs. This should fix prime rendering under kwin, and not break it under the others.

Had those patches living in my local tree for a while now. Here is a respin on top of latest master Patch 1: regression fix, preventing the ddx from loading if kernel module is not loaded Patches 2-5: Completely nuke dri1, make dri2 hard dependency Patches 6-7: Assist people with first-time build of nouveau Git complains about whitespace errors in patch 7, which for the sake of me I cannot

https://bugs.freedesktop.org/show_bug.cgi?id=33999 Summary: 2.6.37 - NV11 crashes X if glxgears maximised Product: xorg Version: unspecified Platform: x86 (IA32) OS/Version: Linux (All) Status: NEW Severity: normal Priority: medium Component: Driver/nouveau AssignedTo: nouveau at

Signed-off-by: Christopher James Halse Rogers <christopher.halse.rogers at canonical.com> --- This is mostly just for testing, although it should be safe to apply to nouveau trunk. src/nouveau_dri2.c | 34 +++++++++++ src/nv_driver.c | 163 +++++++++++++++++++++++++++++++++++++++++++++++----- src/nv_type.h | 7 +++ 3 files changed, 191 insertions(+), 13 deletions(-) diff

First of all, we were not forcing the pixmap to be moved in before attempting to present it. Secondly, in certain configurations, we avoid ever backing pixmaps with vram (in GPUs with 32MB of VRAM or less). This fixes the observed cases where we ended up assuming that a bo was backing a pixmap without doing the explicit move-in. Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu> ---

For now pixmaps will only be tiled if driver pixmaps are being used and we're told to with the NOUVEAU_CREATE_PIXMAP_TILED usage hint. Signed-off-by: Francisco Jerez <currojerez at riseup.net> --- src/nouveau_exa.c | 31 ++++++++++++++++++++----------- src/nv50_exa.c | 6 +++--- src/nv50_xv.c | 2 +- src/nv_proto.h | 2 +- src/nv_type.h | 1 + 5 files

From: Christopher James Halse Rogers <raof at ubuntu.com> Barring some (admittedly significant) missing optimisations? this is reasonably complete, but can't be applied until the Xserver patch has landed, and that needs more work. This demonstrates the approach, however. There's probably some code to be shared with XWayland support, around the output handling (or lack thereof) and

https://bugs.freedesktop.org/show_bug.cgi?id=70511 Priority: medium Bug ID: 70511 Assignee: nouveau at lists.freedesktop.org Summary: nouveau_bo_name_get segmentation fault while running root tutorials/gl/glbox.C QA Contact: xorg-team at lists.x.org Severity: normal Classification: Unclassified

Not clear what the depth % 8 was trying to protect against, but it was breaking 30bpp visuals with DRI3. Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu> --- src/nouveau_dri2.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/nouveau_dri2.c b/src/nouveau_dri2.c index cbb7b2a..07b6022 100644 --- a/src/nouveau_dri2.c +++ b/src/nouveau_dri2.c @@ -1076,7 +1076,7 @@

Make it a Bool consistently, as declared in header. Reported-by: Ilia Mirkin <imirkin at alum.mit.edu> Signed-off-by: Mario Kleiner <mario.kleiner.de at gmail.com> --- src/nouveau_present.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/src/nouveau_present.c b/src/nouveau_present.c index 4de1e6e..699a58d 100644 --- a/src/nouveau_present.c +++

X-Server versions older than 1.16.3 have bugs in their DRI3/Present implementation which impair nouveau, so it is better to stick to good old DRI2 by default on such servers. E.g., page flipping doesn't work at all under DRI3/Present with older servers, and use of extensions like OML_sync_control, SGI_video_sync or INTEL_swap_events also causes failure of Present. nouveau's glamor accel

That would be great, as long as it does only that and does not go into the "drivername" territory. As the said driver ;-) "A driver name to use can be provided instead of simple boolean value, which will be passed to the GL implementation for it to load the appropriate backend." -Emil On 4 July 2015 at 18:17, Ilia Mirkin <imirkin at alum.mit.edu> wrote: > IMO