xserver-xorg-video-nouveau: Changes to 'debian-unstable'
ChangeLog | 309 ++++++++++++++++++++++++++++++++++++++
debian/README.Debian | 27 +--
debian/changelog | 10 +
debian/control | 2
man/nouveau.man | 15 +
src/drmmode_display.c | 107 ++++++++++++-
src/hwdefs/nv_object.xml.h | 248 +++++++++++++++---------------
src/nouveau_dri2.c | 274 +++++++++++++++++++++++++++++++--
src/nouveau_exa.c | 334 +++++++++++------------------------------
src/nouveau_local.h | 96 +++++++++++
src/nouveau_xv.c | 121 +++++----------
src/nv04_exa.c | 116 +++++++++-----
src/nv04_xv_blit.c | 22 +-
src/nv10_exa.c | 183 ++++++++++------------
src/nv30_exa.c | 126 +++++++--------
src/nv30_shaders.c | 25 +--
src/nv30_xv_tex.c | 35 +---
src/nv40_exa.c | 110 ++++++-------
src/nv40_xv_tex.c | 25 +--
src/nv50_accel.c | 124 +++++++--------
src/nv50_accel.h | 12 -
src/nv50_exa.c | 182 ++++++++++++++++------
src/nv50_xv.c | 51 ++----
src/nv_accel_common.c | 176 ++++++++++-----------
src/nv_const.h | 2
src/nv_driver.c | 43 +++++
src/nv_include.h | 3
src/nv_proto.h | 20 ++
src/nv_type.h | 4
src/nvc0_accel.c | 196 ++++++++++++------------
src/nvc0_accel.h | 17 --
src/nvc0_exa.c | 363 +++++++++++++++++----------------------------
src/nvc0_xv.c | 104 ++----------
33 files changed, 2038 insertions(+), 1444 deletions(-)
New commits:
commit 108dd83e6c5dc55c6e7df7089c46a8298cd1835a
Author: Sven Joachim <svenjoac@gmx.de>
Date: Wed Apr 11 18:32:51 2012 +0200
Bump Standards-Version to 3.9.3
diff --git a/debian/changelog b/debian/changelog
index 6300a23..c2aef8f 100644
--- a/debian/changelog
+++ b/debian/changelog
@@ -8,6 +8,7 @@ xserver-xorg-video-nouveau (1:0.0.16+git20120322+ab7291d-1) UNRELEASED; urgency=
* Update README.Debian:
- Nouveau moves out of staging in Linux 3.4.
- Generalize the firmware section.
+ * Bump Standards-Version to 3.9.3, no changes needed.
-- Julien Cristau <jcristau@debian.org> Mon, 09 Apr 2012 20:23:18 +0200
diff --git a/debian/control b/debian/control
index d8929dc..065d5cf 100644
--- a/debian/control
+++ b/debian/control
@@ -24,7 +24,7 @@ Build-Depends:
libtool,
xutils-dev,
quilt,
-Standards-Version: 3.9.2
+Standards-Version: 3.9.3
Homepage: http://nouveau.freedesktop.org/wiki/
Vcs-Git: git://git.debian.org/git/pkg-xorg/driver/xserver-xorg-video-nouveau
Vcs-Browser: http://git.debian.org/?p=pkg-xorg/driver/xserver-xorg-video-nouveau.git
commit efdea9e3434a1b5f796995231c9a4d2a3695acff
Author: Sven Joachim <svenjoac@gmx.de>
Date: Wed Apr 11 18:24:32 2012 +0200
README.Debian: Generalize the firmware section
There is little point in keeping the information there up to date,
especially since it is subject to change with new card generations and
with every new kernel version. So just state that new hardware
generally needs non-free firmware and will hopefully be fully
supported later.
diff --git a/debian/README.Debian b/debian/README.Debian
index aded10e..b3a807f 100644
--- a/debian/README.Debian
+++ b/debian/README.Debian
@@ -9,12 +9,12 @@ have to enable the module under Device Drivers -> Graphics support ->
Direct Rendering Manager (Linux 3.4 and newer) or Device Drivers ->
Staging drivers (Linux 3.3 and earlier).
-Some Fermi cards (GeForce 4xx/5xx series, NVC0 in Nouveau terminology)
-currently need non-free and not readily available firmware to provide
-any acceleration. Starting with Linux 3.1, the Linux kernel includes
-free firmware for some of these cards, but not all of them are
-supported yet. Please check the status at
-http://nouveau.freedesktop.org/wiki/InstallDRM#Firmware.
+Non-vintage Nvidia cards require that microcode (firmware) is uploaded
+by the driver to the GPU in order to provide any acceleration. The
+Nouveau project aims to provide the microcode in the kernel driver for
+all cards, but new hardware usually needs non-free and not readily
+available firmware for a transitional period. Please check the status
+at http://nouveau.freedesktop.org/wiki/InstallDRM#Firmware.
Incompatibilities with other drivers
diff --git a/debian/changelog b/debian/changelog
index a06cfc3..6300a23 100644
--- a/debian/changelog
+++ b/debian/changelog
@@ -7,6 +7,7 @@ xserver-xorg-video-nouveau (1:0.0.16+git20120322+ab7291d-1) UNRELEASED; urgency=
* New upstream snapshot.
* Update README.Debian:
- Nouveau moves out of staging in Linux 3.4.
+ - Generalize the firmware section.
-- Julien Cristau <jcristau@debian.org> Mon, 09 Apr 2012 20:23:18 +0200
commit 6cc7edb20bdb8230f8049cda16852ed851521c8a
Author: Sven Joachim <svenjoac@gmx.de>
Date: Wed Apr 11 18:23:36 2012 +0200
README.Debian: Nouveau moves out of staging in Linux 3.4
diff --git a/debian/README.Debian b/debian/README.Debian
index 870e6c6..aded10e 100644
--- a/debian/README.Debian
+++ b/debian/README.Debian
@@ -2,13 +2,12 @@ Requirements for using Nouveau
==============================
You need a Linux kernel that contains a nouveau.ko module which works
-with the libdrm-nouveau package in Debian. Unfortunately,
-libdrm-nouveau and the kernel module are coupled very tightly, and you
-may not be able to upgrade one without the other. As of this writing,
-supported kernels are 2.6.34-rc1 and newer, earlier versions lack the
-nouveau module altogether or contain an incompatible version. If you
-build your own kernel, you have to enable the module under Drivers ->
-Staging drivers.
+with the libdrm-nouveau package in Debian. Supported kernels are
+2.6.34 and newer, earlier versions lack the nouveau module altogether
+or contain an incompatible version. If you build your own kernel, you
+have to enable the module under Device Drivers -> Graphics support ->
+Direct Rendering Manager (Linux 3.4 and newer) or Device Drivers ->
+Staging drivers (Linux 3.3 and earlier).
Some Fermi cards (GeForce 4xx/5xx series, NVC0 in Nouveau terminology)
currently need non-free and not readily available firmware to provide
@@ -61,4 +60,4 @@ http://nouveau.freedesktop.org/wiki/FAQ
http://nouveau.freedesktop.org/wiki/TroubleShooting
http://nouveau.freedesktop.org/wiki/KernelModeSetting
- -- Sven Joachim <svenjoac@gmx.de>, Thu, 1 Dec 2011 16:17:41 +0100
+ -- Sven Joachim <svenjoac@gmx.de>, Wed, 11 Apr 2012 17:08:14 +0200
diff --git a/debian/changelog b/debian/changelog
index a5015ed..a06cfc3 100644
--- a/debian/changelog
+++ b/debian/changelog
@@ -5,6 +5,8 @@ xserver-xorg-video-nouveau (1:0.0.16+git20120322+ab7291d-1) UNRELEASED; urgency=
[ Sven Joachim ]
* New upstream snapshot.
+ * Update README.Debian:
+ - Nouveau moves out of staging in Linux 3.4.
-- Julien Cristau <jcristau@debian.org> Mon, 09 Apr 2012 20:23:18 +0200
commit e9680e1894cc3c011b8ea85eac8c4704500fdb95
Author: Sven Joachim <svenjoac@gmx.de>
Date: Wed Apr 11 15:36:40 2012 +0200
New upstream snapshot
diff --git a/ChangeLog b/ChangeLog
index 2a5e32e..d1d022b 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,312 @@
+commit ab7291d368ddc66ad21c5ad5caa0faeee42ccaf1
+Author: Ben Skeggs <bskeggs@redhat.com>
+Date: Wed Mar 21 14:56:55 2012 +1000
+
+ xv: don't try to init textured video without 3d engine object
+
+ Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
+
+commit fe7e0425a4a43b2a7d4743a0a90a86724c3b6775
+Author: Ben Skeggs <bskeggs@redhat.com>
+Date: Mon Mar 12 11:59:29 2012 +1000
+
+ recognise kepler
+
+ Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
+
+commit f5d1cd2cb6808838ae1a188cef888eaa9582c76d
+Author: Ben Skeggs <bskeggs@redhat.com>
+Date: Tue Mar 6 08:16:51 2012 +1000
+
+ remove use of xf86PciInfo.h, deprecated now apparently
+
+ Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
+
+commit 248de8cdbd6d0bc062633b49896fa4791148cd3b
+Author: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+Date: Thu Feb 16 00:45:24 2012 +0100
+
+ dri2: Fix corner case crash for swaplimit > 1
+
+ If a swaplimit > 1 is set on a server which
+ supports the swaplimit api (XOrg 1.12.0+),
+ the following can happen:
+
+ 1. Client calls glXSwapBuffersMscOML() with a
+ swap target > 1 vblank in the future, or a
+ client calls glXSwapbuffers() while the swap
+ interval is set to > 1 (unusual but possible).
+
+ 2. nouveau_dri2_finish_swap() is therefore called
+ only at the target vblank, instead of immediately.
+
+ 3. Because of the deferred execution of
+ nouveu_dri2_finish_swap(), the OpenGL client
+ can call x-servers DRI2GetBuffersWithFormat()
+ before nouveau_dri2_finish_swap() executes and
+ it deletes pixmaps that would be needed by
+ nouveau_dri2_finish_swap() --> Segfault --> Crash.
+
+ Prevent this: When a swap is scheduled into the
+ future, we temporarily reduce the swaplimit to 1
+ until nouveau_dri2_finish_swap() is done, then
+ restore it to its original value. This throttles
+ the client inside the server in DRI2ThrottleClient()
+ before it can call the evil DRI2GetbuffersWithFormat().
+
+ The client will still be released one video refresh
+ interval before swap completion, so there is still
+ some potential win.
+
+ This doesn't affect the common case of swapping at
+ the next vblank, where this throttling is not needed
+ or done.
+
+ Signed-off-by: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+ Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
+
+commit 247465ec5e21ce7151ef1e6a3e7644bc74d48a5b
+Author: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+Date: Thu Feb 16 00:45:23 2012 +0100
+
+ dri2: Reimplement hack for triple-buffering on old X-Servers.
+
+ X-Servers before 1.12.0 don't have the DRI2SwapLimit()
+ API. On these, default to a swaplimit of 1 - double-buffering.
+
+ This patch implements support for swap limit of 2,
+ triple-buffering, on old x-servers via Francisco Jerez
+ previous hack:
+
+ Return DRI2SwapComplete() before the swap has completed,
+ so clients don't get blocked on the pending swap. This
+ allows for a "triple-buffering look-alike" behaviour, but
+ breaks the swap scheduling and timestamping defined
+ in the OML_sync_control spec, so applications which
+ rely on conformant behaviour will break with a swap
+ limit of 2 on pre 1.12.0 x-servers.
+
+ Signed-off-by: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+ Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
+
+commit a1902ffaab1f6807fdaac0fce49f784e7208d1a5
+Author: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+Date: Thu Feb 16 00:45:22 2012 +0100
+
+ dri2: Allow vblank controlled swaps for redirected windows. Part II
+
+ This part implements proper throttling for clients. For
+ vblank synchronized blits, it defers DRI2SwapComplete()
+ until 1 vblank after the framebuffer blit is submitted to
+ the gpu.
+
+ Rationale:
+
+ For unredirected windows, this is the earliest time the
+ "blit swap" can complete, as blits are submitted one vblank
+ before the target vblank and synchronized with vblank in the
+ gpu. This makes swap completion timestamps at least reasonable.
+
+ For redirected windows, the compositor will probably pick
+ up the "blit swapped" frontbuffer pixmap of the window quickly,
+ but defer its own recomposition to the next vblank, at least
+ if sync to vblank for the compositor is on.
+
+ Signed-off-by: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+ Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
+
+commit c8572d329014afc0c79b5765468dcdfb65be0868
+Author: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+Date: Thu Feb 16 00:45:21 2012 +0100
+
+ dri2: Allow vblank controlled swaps for redirected windows. Part I
+
+ Make sure that swaps for redirected windows under a
+ compositor are still scheduled via vblank events, to
+ avoid effects like 2900 fps swaps under a compositor.
+
+ See discussion with Francisco Jerez at:
+
+ http://lists.freedesktop.org/archives/nouveau/2011-September/009278.html
+ http://lists.freedesktop.org/archives/nouveau/2011-September/009292.html
+
+ This is part I of the agreed upon band-aid, in a separate patch.
+
+ It allows to use vblank related functions on redirected
+ windows and thereby fixes functions from sgi_sync_control
+ and oml_sync_control extension, e.g., glXWaitForMscOML(),
+ glXGetSyncValuesOML(), glXWaitVideoSyncSGI, ...
+
+ Signed-off-by: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+ Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
+
+commit 919643888045b102f3e10dfdeb6210ab14b6ecab
+Author: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+Date: Thu Feb 16 00:45:20 2012 +0100
+
+ dri2: Fixes to swap scheduling.
+
+ Fix some small off-by-one errors and a mismatch
+ between 32 bit kernel interfaces for vblank count
+ and 64 bit dri2 interfaces for target_msc et al.
+
+ Return corrected target_msc to swap scheduling in
+ x-server.
+
+ A revised version of the patch discussed here:
+ http://lists.freedesktop.org/archives/nouveau/2011-September/009143.html
+
+ Signed-off-by: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+ Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
+
+commit 5ede4538366a352aa9726560f8d42e85df4aec82
+Author: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+Date: Thu Feb 16 00:45:19 2012 +0100
+
+ dri2: Update front buffer pixmap and name before exchanging buffers
+
+ Buffer exchange assumes that the front buffer pixmap and name
+ information is accurate. That may not be the case eg. if the window
+ has been (un)redirected since the buffer was created.
+
+ This is a translation to nouveau of a fix that was originally developed
+ by Ville Syrjala <syrjala@sci.fi> for the ati/radeon ddx to fix the
+ same bug there.
+
+ See thread at:
+
+ http://lists.x.org/archives/xorg-devel/2011-May/021908.html
+
+ Fixes FDO bug #35452.
+
+ Signed-off-by: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+ Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
+
+commit 6fdf60b7288d49f889c80705aabc1db3bc327ba3
+Author: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+Date: Thu Feb 16 00:45:18 2012 +0100
+
+ dri2: Add support for DRI2SwapLimit() API.
+
+ Uses the new DRI2SwapLimit() API of X-Server 1.12+
+ to allow to change the maximum number of pending
+ swaps on a drawable before the OpenGL client is
+ throttled by the server.
+
+ The new optional xorg.conf parameter "SwapLimit"
+ allows to select a new swap limit >= 1. The default
+ swap limit is 2 for triple-buffering on XOrg 1.12+,
+ 1 for double-buffering on older servers, as we can't
+ change the swap limit there.
+
+ Signed-off-by: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+ Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
+
+commit 820916e722db2997a53128f2f370d14a02d4401c
+Author: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+Date: Thu Feb 16 00:45:17 2012 +0100
+
+ dri2: Implement handling of pageflip completion events.
+
+ Requests pageflip completion events from the kernel.
+ Implements pageflip completion handler to finalize
+ and timestamp swaps.
+
+ Completion handler includes a consistency check, and
+ disambiguation if multiple crtc's are involved in a
+ pageflip (e.g., clone mode, extendend desktop). Only
+ the timestamp of the crtc whose vblank event initially
+ triggered the swap is used, but handler waits for flip
+ completion on all involved crtc's before completing the
+ swap and releasing the old framebuffer.
+
+ This code is almost identical to the code used in the
+ ati/radeon ddx and intel ddx.
+
+ Signed-off-by: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+ Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
+
+commit 4da68cedbefef4f72c00bb088e179071490882c8
+Author: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+Date: Thu Mar 1 19:11:14 2012 +0100
+
+ dri2: Fix can_exchange() to allow page-flipping on more mesa versions.
+
+ can_exchange() returns false and thereby prevents page
+ flipping on some drawables where page flipping would
+ work fine. This due to non-matching drawable depths values
+ between front buffer pixmap and back buffer pixmap, because
+ front buffer pixmaps inherit the depth of the screen, typically
+ 24 bits, whereas the depth value of back buffer pixmaps for a
+ given RGB8 or RGBA8 visual depends on the mesa version in use,
+ either 24 bits or 32 bits.
+
+ Use bitsPerPixel instead of depth to decide if drawable
+ is flippable. This will still catch really incompatible
+ formats like 32 bpp vs. 16 bpp buffers.
+
+ Tested for screen DefaultDepth 24 and also 30 bits
+ (for RGB10 framebuffers) on NV-50.
+
+ The problem was fixed in the same way in the ati & intel ddx.
+
+ Signed-off-by: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
+ Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
+
+commit b0d7f4ddd94154491d04dcb9698d63d61e9240dc
+Author: Christoph Bumiller <e0425955@student.tuwien.ac.at>
+Date: Mon Jan 9 15:17:27 2012 +0100
+
+ dri2: CreateBuffer format may be 0, use depth in that case
+
+commit 3d9f6b32e0ad72dc8c7db07d039d87e83d8fdb1d
+Author: Ben Skeggs <bskeggs@redhat.com>
+Date: Fri Dec 9 17:45:31 2011 +1000
+
+ nv50/xv: bump max dims to 8k
+
+ Untested, but I don't see why this wouldn't work...
+
+commit 57c62228e2e67370feeeb421c369fbf940d30ca4
+Author: Ben Skeggs <bskeggs@redhat.com>
+Date: Fri Dec 9 17:44:07 2011 +1000
+
+ xv: stop using custom m2mf code
+
+commit ad402333157c437ccd2c639f5b3f3bcb81e78d6c
+Author: Christoph Bumiller <e0425955@student.tuwien.ac.at>
+Date: Sat Dec 10 13:32:24 2011 +0100
+
+ dri2: fix allocation of Z16 depth attachments
+
+commit 3d2a752ca8aafee5e1e94dabfd7deec439890e95
+Author: Ben Skeggs <bskeggs@redhat.com>
+Date: Thu Dec 1 11:31:49 2011 +1000
+
+ move away from libdrm's BEGIN_RING (and variants) macros
+
+ subchan auto-assignment is now gone.
+
+commit b6cfdc01aeb5c2ab401ee81fff66bd0560abdfcc
+Author: Ben Skeggs <bskeggs@redhat.com>
+Date: Thu Dec 1 11:25:10 2011 +1000
+
+ Append _CLASS to object class id definitions
+
+ I want to be able to use ring macros of the same name
+
+commit 1dadb5826363d0d8ed1a72209be18277550f8345
+Author: Ben Skeggs <bskeggs@redhat.com>
+Date: Thu Dec 1 09:06:24 2011 +1000
+
+ exa: create general m2mf rect copy for each chipset, use for UTS/DFS
+
+ No point having duplicated UTS/DFS logic in each implementation. This also
+ splits nv50 out into its own code, in preparation for it being made to not
+ use relocs at all.
+
+ Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
+
commit b5534a15a5f5e1c3bd6c92e3ed8391ff628c09bb
Author: Marcin Slusarz <marcin.slusarz@gmail.com>
Date: Wed Nov 30 18:56:54 2011 +0100
diff --git a/debian/changelog b/debian/changelog
index 3712479..a5015ed 100644
--- a/debian/changelog
+++ b/debian/changelog
@@ -1,7 +1,11 @@
-xserver-xorg-video-nouveau (1:0.0.16+git20111201+b5534a1-2) UNRELEASED; urgency=low
+xserver-xorg-video-nouveau (1:0.0.16+git20120322+ab7291d-1) UNRELEASED; urgency=low
+ [ Julien Cristau ]
* Add Dm-Upload-Allowed control field to allow Sven to upload this package.
+ [ Sven Joachim ]
+ * New upstream snapshot.
+
-- Julien Cristau <jcristau@debian.org> Mon, 09 Apr 2012 20:23:18 +0200
xserver-xorg-video-nouveau (1:0.0.16+git20111201+b5534a1-1) unstable; urgency=low
commit ab7291d368ddc66ad21c5ad5caa0faeee42ccaf1
Author: Ben Skeggs <bskeggs@redhat.com>
Date: Wed Mar 21 14:56:55 2012 +1000
xv: don't try to init textured video without 3d engine object
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
diff --git a/src/nouveau_xv.c b/src/nouveau_xv.c
index ad203a0..54ae3be 100644
--- a/src/nouveau_xv.c
+++ b/src/nouveau_xv.c
@@ -2022,6 +2022,28 @@ NV50SetupTexturedVideo (ScreenPtr pScreen)
return adapt;
}
+void
+NVSetupTexturedVideo (ScreenPtr pScreen, XF86VideoAdaptorPtr *textureAdaptor)
+{
+ ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
+ NVPtr pNv = NVPTR(pScrn);
+
+ if (!pNv->Nv3D)
+ return;
+
+ if (pNv->Architecture == NV_ARCH_30) {
+ textureAdaptor[0] = NV30SetupTexturedVideo(pScreen, FALSE);
+ textureAdaptor[1] = NV30SetupTexturedVideo(pScreen, TRUE);
+ } else
+ if (pNv->Architecture == NV_ARCH_40) {
+ textureAdaptor[0] = NV40SetupTexturedVideo(pScreen, FALSE);
+ textureAdaptor[1] = NV40SetupTexturedVideo(pScreen, TRUE);
+ } else
+ if (pNv->Architecture >= NV_ARCH_50) {
+ textureAdaptor[0] = NV50SetupTexturedVideo(pScreen);
+ }
+}
+
/**
* NVInitVideo
* tries to initialize the various supported adapters
@@ -2056,17 +2078,7 @@ NVInitVideo(ScreenPtr pScreen)
blitAdaptor = NVSetupBlitVideo(pScreen);
}
- if (pNv->Architecture == NV_ARCH_30) {
- textureAdaptor[0] = NV30SetupTexturedVideo(pScreen, FALSE);
- textureAdaptor[1] = NV30SetupTexturedVideo(pScreen, TRUE);
- } else
- if (pNv->Architecture == NV_ARCH_40) {
- textureAdaptor[0] = NV40SetupTexturedVideo(pScreen, FALSE);
- textureAdaptor[1] = NV40SetupTexturedVideo(pScreen, TRUE);
- } else
- if (pNv->Architecture >= NV_ARCH_50) {
- textureAdaptor[0] = NV50SetupTexturedVideo(pScreen);
- }
+ NVSetupTexturedVideo(pScreen, textureAdaptor);
}
num_adaptors = xf86XVListGenericAdaptors(pScrn, &adaptors);
commit fe7e0425a4a43b2a7d4743a0a90a86724c3b6775
Author: Ben Skeggs <bskeggs@redhat.com>
Date: Mon Mar 12 11:59:29 2012 +1000
recognise kepler
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
diff --git a/src/nv_driver.c b/src/nv_driver.c
index 667a0a7..56865e2 100644
--- a/src/nv_driver.c
+++ b/src/nv_driver.c
@@ -265,6 +265,7 @@ NVPciProbe(DriverPtr drv, int entity_num, struct pci_device *pci_dev,
case 0xa0:
case 0xc0:
case 0xd0:
+ case 0xe0:
break;
default:
xf86DrvMsg(-1, X_ERROR, "Unknown chipset: NV%02x\n", chipset);
@@ -694,6 +695,9 @@ NVPreInit(ScrnInfoPtr pScrn, int flags)
case 0xd0:
pNv->Architecture = NV_ARCH_C0;
break;
+ case 0xe0:
+ pNv->Architecture = NV_ARCH_E0;
+ break;
default:
return FALSE;
}
diff --git a/src/nv_type.h b/src/nv_type.h
index 73328fc..c8f27ac 100644
--- a/src/nv_type.h
+++ b/src/nv_type.h
@@ -25,6 +25,7 @@
#define NV_ARCH_40 0x40
#define NV_ARCH_50 0x50
#define NV_ARCH_C0 0xc0
+#define NV_ARCH_E0 0xe0
/* NV50 */
typedef struct _NVRec *NVPtr;
commit f5d1cd2cb6808838ae1a188cef888eaa9582c76d
Author: Ben Skeggs <bskeggs@redhat.com>
Date: Tue Mar 6 08:16:51 2012 +1000
remove use of xf86PciInfo.h, deprecated now apparently
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
diff --git a/src/nv_driver.c b/src/nv_driver.c
index 7512464..667a0a7 100644
--- a/src/nv_driver.c
+++ b/src/nv_driver.c
@@ -65,8 +65,8 @@ static Bool NVUnmapMem(ScrnInfoPtr pScrn);
0x00030000, 0x00ffffff, 0 }
static const struct pci_id_match nouveau_device_match[] = {
- NOUVEAU_PCI_DEVICE(PCI_VENDOR_NVIDIA, PCI_MATCH_ANY),
- NOUVEAU_PCI_DEVICE(PCI_VENDOR_NVIDIA_SGS, PCI_MATCH_ANY),
+ NOUVEAU_PCI_DEVICE(0x12d2, PCI_MATCH_ANY),
+ NOUVEAU_PCI_DEVICE(0x10de, PCI_MATCH_ANY),
{ 0, 0, 0 },
};
diff --git a/src/nv_include.h b/src/nv_include.h
index 496eefe..407b7c9 100644
--- a/src/nv_include.h
+++ b/src/nv_include.h
@@ -17,9 +17,6 @@
#define PPC_MMIO_IS_BE
#include "compiler.h"
-/* Drivers for PCI hardware need this */
-#include "xf86PciInfo.h"
-
/* Drivers that need to access the PCI config space directly need this */
#include "xf86Pci.h"
commit 248de8cdbd6d0bc062633b49896fa4791148cd3b
Author: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
Date: Thu Feb 16 00:45:24 2012 +0100
dri2: Fix corner case crash for swaplimit > 1
If a swaplimit > 1 is set on a server which
supports the swaplimit api (XOrg 1.12.0+),
the following can happen:
1. Client calls glXSwapBuffersMscOML() with a
swap target > 1 vblank in the future, or a
client calls glXSwapbuffers() while the swap
interval is set to > 1 (unusual but possible).
2. nouveau_dri2_finish_swap() is therefore called
only at the target vblank, instead of immediately.
3. Because of the deferred execution of
nouveu_dri2_finish_swap(), the OpenGL client
can call x-servers DRI2GetBuffersWithFormat()
before nouveau_dri2_finish_swap() executes and
it deletes pixmaps that would be needed by
nouveau_dri2_finish_swap() --> Segfault --> Crash.
Prevent this: When a swap is scheduled into the
future, we temporarily reduce the swaplimit to 1
until nouveau_dri2_finish_swap() is done, then
restore it to its original value. This throttles
the client inside the server in DRI2ThrottleClient()
before it can call the evil DRI2GetbuffersWithFormat().
The client will still be released one video refresh
interval before swap completion, so there is still
some potential win.
This doesn't affect the common case of swapping at
the next vblank, where this throttling is not needed
or done.
Signed-off-by: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
diff --git a/src/nouveau_dri2.c b/src/nouveau_dri2.c
index f0c7fec..7878a5a 100644
--- a/src/nouveau_dri2.c
+++ b/src/nouveau_dri2.c
@@ -445,6 +445,26 @@ nouveau_dri2_schedule_swap(ClientPtr client, DrawablePtr draw,
if (*target_msc == 0)
*target_msc = 1;
+#if DRI2INFOREC_VERSION >= 6
+ /* Is this a swap in the future, ie. the vblank event will
+ * not be immediately dispatched, but only at a future vblank?
+ * If so, we need to temporarily lower the swaplimit to 1, so
+ * that DRI2GetBuffersWithFormat() requests from the client get
+ * deferred in the x-server until the vblank event has been
+ * dispatched to us and nouveau_dri2_finish_swap() is done. If
+ * we wouldn't do this, DRI2GetBuffersWithFormat() would operate
+ * on wrong (pre-swap) buffers, and cause a segfault later on in
+ * nouveau_dri2_finish_swap(). Our vblank event handler restores
+ * the old swaplimit immediately after nouveau_dri2_finish_swap()
+ * is done, so we still get 1 video refresh cycle worth of
+ * triple-buffering. For a swap at next vblank, dispatch of the
+ * vblank event happens immediately, so there isn't any need
+ * for this lowered swaplimit.
+ */
+ if (current_msc < *target_msc - 1)
+ DRI2SwapLimit(draw, 1);
+#endif
+
/* Request a vblank event one frame before the target */
ret = nouveau_wait_vblank(draw, DRM_VBLANK_ABSOLUTE |
DRM_VBLANK_EVENT,
@@ -557,6 +577,12 @@ nouveau_dri2_vblank_handler(int fd, unsigned int frame,
switch (s->action) {
case SWAP:
nouveau_dri2_finish_swap(draw, frame, tv_sec, tv_usec, s);
+#if DRI2INFOREC_VERSION >= 6
+ /* Restore real swap limit on drawable, now that it is safe. */
+ ScrnInfoPtr scrn = xf86Screens[draw->pScreen->myNum];
+ DRI2SwapLimit(draw, NVPTR(scrn)->swap_limit);
+#endif
+
break;
case WAIT:
commit 247465ec5e21ce7151ef1e6a3e7644bc74d48a5b
Author: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
Date: Thu Feb 16 00:45:23 2012 +0100
dri2: Reimplement hack for triple-buffering on old X-Servers.
X-Servers before 1.12.0 don't have the DRI2SwapLimit()
API. On these, default to a swaplimit of 1 - double-buffering.
This patch implements support for swap limit of 2,
triple-buffering, on old x-servers via Francisco Jerez
previous hack:
Return DRI2SwapComplete() before the swap has completed,
so clients don't get blocked on the pending swap. This
allows for a "triple-buffering look-alike" behaviour, but
breaks the swap scheduling and timestamping defined
in the OML_sync_control spec, so applications which
rely on conformant behaviour will break with a swap
limit of 2 on pre 1.12.0 x-servers.
Signed-off-by: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
diff --git a/man/nouveau.man b/man/nouveau.man
index 59f6c1a..7c72907 100644
--- a/man/nouveau.man
+++ b/man/nouveau.man
@@ -101,7 +101,11 @@ a drawable before a client is blocked.
A value of 1 corresponds to double-buffering. A value of 2 corresponds
to triple-buffering. Higher values may allow higher framerate, but also
increase lag for interactive applications, e.g., games. Nouveau currently
-supports a maximum value of 2 on XOrg 1.12+ and a maximum of 1 on older servers.
+reliably supports a maximum value of 2 on XOrg 1.12+. A maximum setting of 2
+on older x-servers is allowed, but it will break conformance with the
+OpenML OML_sync_control specification and will cause failure of software
+that relies on correct presentation timing behaviour as defined in that
+specification.
.br
Default: 2 for XOrg 1.12+, 1 for older servers.
.SH "SEE ALSO"
diff --git a/src/nouveau_dri2.c b/src/nouveau_dri2.c
index fdc5148..f0c7fec 100644
--- a/src/nouveau_dri2.c
+++ b/src/nouveau_dri2.c
@@ -255,6 +255,18 @@ nouveau_dri2_swap_limit_validate(DrawablePtr draw, int swap_limit)
}
#endif
+/* Shall we intentionally violate the OML_sync_control spec to
+ * get some sort of triple-buffering behaviour on a pre 1.12.0
+ * x-server?
+ */
+static Bool violate_oml(DrawablePtr draw)
+{
+ ScrnInfoPtr scrn = xf86Screens[draw->pScreen->myNum];
+ NVPtr pNv = NVPTR(scrn);
+
+ return (DRI2INFOREC_VERSION < 6) && (pNv->swap_limit > 1);
+}
+
static void
nouveau_dri2_finish_swap(DrawablePtr draw, unsigned int frame,
unsigned int tv_sec, unsigned int tv_usec,
@@ -319,7 +331,9 @@ nouveau_dri2_finish_swap(DrawablePtr draw, unsigned int frame,
if (DRI2CanFlip(draw)) {
type = DRI2_FLIP_COMPLETE;
- ret = drmmode_page_flip(draw, src_pix, s, ref_crtc_hw_id);
+ ret = drmmode_page_flip(draw, src_pix,
+ violate_oml(draw) ? NULL : s,
+ ref_crtc_hw_id);
if (!ret)
goto out;
}
@@ -330,7 +344,7 @@ nouveau_dri2_finish_swap(DrawablePtr draw, unsigned int frame,
DamageRegionProcessPending(draw);
/* If it is a page flip, finish it in the flip event handler. */
- if (type == DRI2_FLIP_COMPLETE)
+ if ((type == DRI2_FLIP_COMPLETE) && !violate_oml(draw))
return;
} else {
type = DRI2_BLIT_COMPLETE;
@@ -344,7 +358,7 @@ nouveau_dri2_finish_swap(DrawablePtr draw, unsigned int frame,
REGION_TRANSLATE(0, ®, -draw->x, -draw->y);
nouveau_dri2_copy_region(draw, ®, s->dst, s->src);
- if (can_sync_to_vblank(draw)) {
+ if (can_sync_to_vblank(draw) && !violate_oml(draw)) {
/* Request a vblank event one vblank from now, the most
* likely (optimistic?) time a direct framebuffer blit
* will complete or a desktop compositor will update its
@@ -361,6 +375,14 @@ nouveau_dri2_finish_swap(DrawablePtr draw, unsigned int frame,
}
}
+ /* Special triple-buffering hack for old pre 1.12.0 x-servers used? */
+ if (violate_oml(draw)) {
+ /* Signal to client that swap completion timestamps and counts
+ * are invalid - they violate the specification.
+ */
+ frame = tv_sec = tv_usec = 0;
+ }
+
/*
* Tell the X server buffers are already swapped even if they're
* not, to prevent it from blocking the client on the next
@@ -371,6 +393,10 @@ nouveau_dri2_finish_swap(DrawablePtr draw, unsigned int frame,
* It is still needed as a fallback for some copy swaps as
* we lack a method to detect true swap completion for
* DRI2_BLIT_COMPLETE.
+ *
+ * It is also used if triple-buffering is requested on
+ * old x-servers which don't support the DRI2SwapLimit()
+ * function.
*/
DRI2SwapComplete(s->client, draw, frame, tv_sec, tv_usec,
type, s->func, s->data);
diff --git a/src/nv_driver.c b/src/nv_driver.c
index 5def531..7512464 100644
--- a/src/nv_driver.c
+++ b/src/nv_driver.c
@@ -863,11 +863,12 @@ NVPreInit(ScrnInfoPtr pScrn, int flags)
reason = "";
from = X_CONFIG;
- if (DRI2INFOREC_VERSION < 6) {
- /* No swap limit api in server. Stick to server default of 1. */
- pNv->swap_limit = 1;
- from = X_DEFAULT;
- reason = ": This X-Server only supports a swap limit of 1.";
+ if ((DRI2INFOREC_VERSION < 6) && (pNv->swap_limit > 1)) {
+ /* No swap limit api in server. A value > 1 requires use
+ * of problematic hacks.
+ */
+ from = X_WARNING;
+ reason = ": Caution: Use of this swap limit > 1 violates OML_sync_control spec on this X-Server!\n";
}
} else {
/* Driver default: Double buffering on old servers, triple-buffering
commit a1902ffaab1f6807fdaac0fce49f784e7208d1a5
Author: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
Date: Thu Feb 16 00:45:22 2012 +0100
dri2: Allow vblank controlled swaps for redirected windows. Part II
This part implements proper throttling for clients. For
vblank synchronized blits, it defers DRI2SwapComplete()
until 1 vblank after the framebuffer blit is submitted to
the gpu.
Rationale:
For unredirected windows, this is the earliest time the
"blit swap" can complete, as blits are submitted one vblank
before the target vblank and synchronized with vblank in the
gpu. This makes swap completion timestamps at least reasonable.
For redirected windows, the compositor will probably pick
up the "blit swapped" frontbuffer pixmap of the window quickly,
but defer its own recomposition to the next vblank, at least
if sync to vblank for the compositor is on.
Signed-off-by: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
diff --git a/src/nouveau_dri2.c b/src/nouveau_dri2.c
index 6a0800c..fdc5148 100644
--- a/src/nouveau_dri2.c
+++ b/src/nouveau_dri2.c
@@ -135,6 +135,7 @@ nouveau_dri2_copy_region(DrawablePtr pDraw, RegionPtr pRegion,
struct nouveau_dri2_vblank_state {
enum {
SWAP,
+ BLIT,
WAIT
} action;
@@ -342,6 +343,22 @@ nouveau_dri2_finish_swap(DrawablePtr draw, unsigned int frame,
REGION_TRANSLATE(0, ®, -draw->x, -draw->y);
nouveau_dri2_copy_region(draw, ®, s->dst, s->src);
+
+ if (can_sync_to_vblank(draw)) {
+ /* Request a vblank event one vblank from now, the most
+ * likely (optimistic?) time a direct framebuffer blit
+ * will complete or a desktop compositor will update its
+ * screen. This defers DRI2SwapComplete() to the earliest
+ * likely time of real swap completion.
+ */
+ s->action = BLIT;
+ ret = nouveau_wait_vblank(draw, DRM_VBLANK_EVENT |
+ DRM_VBLANK_RELATIVE, 1,
+ NULL, NULL, s);
+ /* Done, if success. Otherwise use fallback below. */
+ if (!ret)
+ return;
+ }
}
/*
@@ -351,8 +368,9 @@ nouveau_dri2_finish_swap(DrawablePtr draw, unsigned int frame,
*
* XXX - The DRI2SwapLimit() API allowed us to move this to
* the flip handler with no FPS hit for page flipped swaps.
- * It is still needed for copy swaps as we lack a method
- * to detect true swap completion for DRI2_BLIT_COMPLETE.
+ * It is still needed as a fallback for some copy swaps as
+ * we lack a method to detect true swap completion for
+ * DRI2_BLIT_COMPLETE.
*/
DRI2SwapComplete(s->client, draw, frame, tv_sec, tv_usec,
type, s->func, s->data);
@@ -505,8 +523,10 @@ nouveau_dri2_vblank_handler(int fd, unsigned int frame,
ret = dixLookupDrawable(&draw, s->draw, serverClient,
M_ANY, DixWriteAccess);
- if (ret)
+ if (ret) {
+ free(s);
return;
+ }
switch (s->action) {
case SWAP:
@@ -517,6 +537,12 @@ nouveau_dri2_vblank_handler(int fd, unsigned int frame,
DRI2WaitMSCComplete(s->client, draw, frame, tv_sec, tv_usec);
free(s);
break;
+
+ case BLIT:
+ DRI2SwapComplete(s->client, draw, frame, tv_sec, tv_usec,
+ DRI2_BLIT_COMPLETE, s->func, s->data);
+ free(s);
+ break;
}
}
commit c8572d329014afc0c79b5765468dcdfb65be0868
Author: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
Date: Thu Feb 16 00:45:21 2012 +0100
dri2: Allow vblank controlled swaps for redirected windows. Part I
Make sure that swaps for redirected windows under a
compositor are still scheduled via vblank events, to
avoid effects like 2900 fps swaps under a compositor.
See discussion with Francisco Jerez at:
http://lists.freedesktop.org/archives/nouveau/2011-September/009278.html
http://lists.freedesktop.org/archives/nouveau/2011-September/009292.html
This is part I of the agreed upon band-aid, in a separate patch.
It allows to use vblank related functions on redirected
windows and thereby fixes functions from sgi_sync_control
and oml_sync_control extension, e.g., glXWaitForMscOML(),
glXGetSyncValuesOML(), glXWaitVideoSyncSGI, ...
Reply to: