On Tue, Apr 13, 2010 at 04:34:51AM -0700, Marcin Slusarz wrote: > On Tue, Apr 13, 2010 at 11:15:28AM +1000, Christopher James Halse Rogers > wrote: > > On Mon, 2010-04-12 at 20:04 +0300, Pekka Paalanen wrote: > > > On Mon, 12 Apr 2010 12:50:13 +0200 > > > Marcin Slusarz <[email protected]> wrote: > > > > > > > I'm not aware of any trick to make both nv and nouveau working. > > > > Maybe there should be a way to prevent nv (and vesa, but not > > > > fbdev) from loading when KMS is in use? > > > > > > Yeah, there was a brief discussion about that recently: > > > http://people.freedesktop.org/~cbrill/dri-log/index.php?date=2010-03-21&channel=nouveau&show_html=true&highlight_names=&update=Update&date=2010-03-21 > > > > > > I haven't heard anything about it since. > > > > > > > I wrote a patch for vesa on bugzilla[1] to do exactly that; a similar > > one for nv wouldn't be hard. > > > > [1] https://bugs.freedesktop.org/show_bug.cgi?id=26878 > > Yay, thanks. > I adapted it to xf86-video-nv and it works! > Now we need to trick Aaron to apply it. > > Aaron: please, please, please apply. > :)
Sorry for the slow reply. It sucks that this is necessary, but I guess it makes sense. Comments below. > --- > From 112a8c9aa9de6ea30b4c3c98e5d7cbce312aaaf8 Mon Sep 17 00:00:00 2001 > From: Marcin Slusarz <[email protected]> > Date: Tue, 13 Apr 2010 13:17:56 +0200 > Subject: [PATCH] Refuse to bind to a device which has kernel modesetting > active. > > Touching directly the hardware while kernel modesetting is active > breaks console output. > > Patch almost entrirely based on VESA patch by > Christopher James Halse Rogers <[email protected]> > > Signed-off-by: Marcin Slusarz <[email protected]> > --- > configure.ac | 11 +++++++++++ > src/Makefile.am | 2 +- > src/nv_driver.c | 33 +++++++++++++++++++++++++++++++++ > 3 files changed, 45 insertions(+), 1 deletions(-) > > diff --git a/configure.ac b/configure.ac > index d717805..3028005 100644 > --- a/configure.ac > +++ b/configure.ac > @@ -134,6 +134,17 @@ AC_SUBST([modes_dir]) > AC_SUBST([XORG_CFLAGS]) > AC_SUBST([moduledir]) > > +if test "x$XSERVER_LIBPCIACCESS" = xyes; then > + PKG_CHECK_MODULES(LIBDRM, [libdrm > 2.4.3 xf86driproto], HAVE_KMS="yes", > HAVE_KMS="no") > + if test "x$HAVE_KMS" = xyes; then > + AC_DEFINE(HAVE_KMS, 1, [Have kernel modesetting]) > + else > + AC_MSG_WARN(Support for detecting kernel modesetting drivers is not > available.) > + AC_MSG_WARN(This driver can cause display problems in the presence > of kernel modesetting.) > + AC_MSG_WARN(Please install libdrm > 2.4.3 and xf86driproto to enable > kms detection.) s/kms/KMS/ > + fi > +fi > + > DRIVER_NAME=nv > AC_SUBST([DRIVER_NAME]) > > diff --git a/src/Makefile.am b/src/Makefile.am > index 2677d9c..5d57010 100644 > --- a/src/Makefile.am > +++ b/src/Makefile.am > @@ -23,7 +23,7 @@ > # -avoid-version prevents gratuitous .0.0.0 version numbers on the end > # _ladir passes a dummy rpath to libtool so the thing will actually link > # TODO: -nostdlib/-Bstatic/-lgcc platform magic, not installing the .a, etc. > -AM_CFLAGS = @XMODES_CFLAGS@ @XORG_CFLAGS@ > +AM_CFLAGS = @XMODES_CFLAGS@ @XORG_CFLAGS@ $(LIBDRM_CFLAGS) > nv_drv_la_LTLIBRARIES = nv_drv.la > nv_drv_la_LDFLAGS = -module -avoid-version > nv_drv_ladir = @moduledir@/drivers > diff --git a/src/nv_driver.c b/src/nv_driver.c > index 0d017f9..e117626 100644 > --- a/src/nv_driver.c > +++ b/src/nv_driver.c > @@ -33,6 +33,11 @@ > #include "xf86int10.h" > #include "vbeModes.h" > > +#ifdef HAVE_KMS > +#include <xf86drmMode.h> > +#include <dri.h> > +#endif > + > const OptionInfoRec * RivaAvailableOptions(int chipid, int busid); > Bool RivaGetScrnInfoRec(PciChipsets *chips, int chip); > Bool G80GetScrnInfoRec(PciChipsets *chips, int chip); > @@ -830,6 +835,26 @@ NVIsSupported(CARD32 id) > return FALSE; > } > > +#ifdef HAVE_KMS > +static Bool NVKernelModesettingEnabled (struct pci_device *device) > +{ > + char *busIdString; > + int ret; > + > + if (!xf86LoaderCheckSymbol("DRICreatePCIBusID")) > + return FALSE; I know the rest of the code uses tabs, but please use spaces for new code. > + > + busIdString = DRICreatePCIBusID(device); Eww. Why doesn't drmCheckModesettingSupported take a pci_device*? Oh well, whatever. > + ret = drmCheckModesettingSupported(busIdString); There's no chance of false positives here, right? E.g. if you set nouveau.modeset=0 on the kernel command line, does that make this return nonzero? > + xfree(busIdString); > + > + return (ret == 0); > +} > +#else > +static inline Bool NVKernelModesettingEnabled (struct pci_device *device) { > return FALSE; } > +#endif //HAVE_KMS > + > /* Mandatory */ > #if XSERVER_LIBPCIACCESS > static Bool > @@ -864,6 +889,14 @@ NVPciProbe(DriverPtr drv, int entity, struct pci_device > *dev, intptr_t data) > NV_NAME ": Found NVIDIA %s at %2...@%2.2x:%2.2x:%1.1x\n", > name, dev->bus, dev->domain, dev->dev, dev->func); > > + /* Trying to bring up a NV mode while kernel modesetting is enabled > + results in badness */ > + if (NVKernelModesettingEnabled (dev)) { > + xf86Msg (X_ERROR, > + "NV: Kernel modesetting driver in use, refusing to load\n"); Please use NV_NAME instead of hard-coding the "NV:" prefix. > + return FALSE; > + } > + > if(NVIsG80(id)) > return G80GetScrnInfoRec(NULL, entity); > else if(dev->vendor_id == PCI_VENDOR_NVIDIA_SGS) > -- > 1.7.0.4 _______________________________________________ Nouveau mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/nouveau
