Difference between revisions of "Lenovo IdeaPad Y580"

From ArchWiki
Jump to: navigation, search
(Installation)
(DISPLAY: : original author (https://bbs.archlinux.org/viewtopic.php?pid=1127892#p1127892) didn't care to explain any of this laptop's "compatibility issues" except those CUDA-related)
(19 intermediate revisions by 8 users not shown)
Line 1: Line 1:
[[Category:Lenovo]]
+
==DISPLAY==
The Lenovo IdeaPad Y580 started shipping in June 2012. It is a fairly powerful machine, but it has its own compatibility issues. The purpose of this article is to help with setting up Arch Linux on this machine.
+
As of the arrival of the 3.7.X series kernel, this machine boots into black screen. You will have to add acpi_backlight=vendor to kernel command line to boot into a visible display. You can alternatively increase the screen brightness during every boot also, since the screen brightness is set to 0 wrongly by the kernel acpi by default.
  
===Installation===
+
As of September 2013, this doesn't seem to be an issue anymore, at least not on kernel 3.10.
  
Install Arch on the machine if you have not done it already. Read [[Installation Guide]] for more information.
+
===Intel Card===
  
If you cannot get wireless working, you can use an ethernet connection with the help of the driver alx. This driver is not yet part of the Linux kernel, and that is why your ethernet card is not recognized at first. The links from QCA upstream [https://www.kernel.org/pub/linux/kernel/projects/backports/2013/03/04/compat-drivers-2013-03-04-u.tar.bz2 alx source files]works for current  3.8.x kernels.
+
This machine has two video cards, explained in detail below
This sources is found to work for both AR8161 and AR8162 ethernet devices.
+
{{bc|<nowiki>
Copy the tarball to a flash drive, boot the live cd, mount the flash drive, cd to the directory with the tarball and use:
+
$ lspci | grep VGA
 +
00:02.0 VGA compatible controller: Intel Corporation 3rd Gen Core processor Graphics Controller (rev 09)
 +
01:00.0 VGA compatible controller: NVIDIA Corporation GK107M [GeForce GTX 660M] (rev a1)
 +
</nowiki>}}
  
tar -xjvf compat*
+
In order to get the machine working for basic purposes, simply install xf86-video-intel and use the intel driver. You can leave the nvidia card disabled (and powered down?) just by not installing nvidia-304xx.
cd compat*
+
./scripts/driver-select alx
+
make
+
sudo make install 
+
  
After this, load the module (modprobe alx) and you should be able to get ethernet working easily.
 
 
The machine comes with Windows 7 and some Lenovo partitions that may be important if you need to recover the Windows install.
 
 
In case you don't need to restore the machine to their original state, you can just delete the existing partition. Other way, backup the partions before erasing them
 
 
Also, this laptop can use UEFI. If you want to use it, you need to a GPT partition (see below).
 
 
===UEFI===
 
 
Even thought its easier to switch from UEFI to legacy mode BIOS in this laptop and install Arch Linux on mSATA SSD and without disturbing factory installed windows 8 located on the HDD, this section will guide you to install Arch Linux with UEFI. UEFI is complicated and so proper understanding is advised before you start.
 
 
Read these pages: [http://www.rodsbooks.com/bios2uefi/index.html Rod smith's UEFI guide] [[Unified Extensible Firmware Interface]] and [[GRUB2]].
 
 
Among other things, you need a gpt partition and grub2.
 
 
To install Arch on the SSD drive, you need at least two partitions: one small (100 MB) boot partition, and another partition for /. To partition the drives, you can use cgdisk, which you can get by installing the package gpttools.
 
 
You should also create a third partition (with about 1GB) for EFI. This partition needs to be of EFI system type (code ef00 on gdisk) and it should be formatted as FAT32. If the partition is /dev/sda2, use:
 
 
mkfs.vfat -F32 /dev/sda3 
 
 
After you have your base system up and running, install the grub 2 firmware:
 
 
pacman -S grub2-efi-x86_64
 
 
Mount the system partition at /boot/efi:
 
 
mkdir /boot/efi
 
mount -t vfat /dev/sda3 /boot/efi
 
 
Install grub2 efi app (grubx64.efi) to /boot/efi/EFI/arch_grub, and its modules to /boot/efi/EFI/grub/x86_64-efi:
 
 
grub-install --directory=/usr/lib/grub/x86_64-efi --target=x86_64-efi --efi-directory=/boot/efi --bootloader-id=arch_grub --boot-directory=/boot/efi/EFI --recheck --debug
 
mkdir -p /boot/efi/EFI/grub/locale
 
cp /usr/share/locale/en@quot/LC_MESSAGES/grub.mo /boot/efi/EFI/grub/locale/en.mo
 
 
The grub2 wiki page says to copy the app to other places. This is probably not necessary, but you may want to use:
 
 
mkdir /boot/efi/EFI/tools
 
cp /boot/efi/EFI/arch_grub/grubx64.efi /boot/efi/shellx64.efi
 
cp /boot/efi/EFI/arch_grub/grubx64.efi /boot/efi/EFI/shellx64.efi
 
cp /boot/efi/EFI/arch_grub/grubx64.efi /boot/efi/EFI/tools/shellx64.efi 
 
 
Now comes the part where the grub2 page is not very clear. You need to add the system to the UEFI menu (the menu that shows up when you press F12 at boot). To do this, you need an UEFI shell. The Y580 does not come with a shell built in, but you can put one in a flash drive and boot from it. To do this, get a bootable flash drive, create a partition (1GB is enough) and format it as FAT32. Assuming that the partition is /dev/sdc1, type the following:
 
 
mount /dev/sdc1 /media
 
mkdir -p /media/efi/boot
 
cd /media/efi/boot
 
wget https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2/ShellBinPkg/UefiShell/X64/Shell.efi
 
mv Shell.efi bootx64.efi
 
cd /
 
umount /media 
 
 
The code creates a directory /efi/boot in the flash drive, downloads the shell, copies it to /efi/boot and renames it as bootx64.efi. The shell is downloaded from the link given here: https://wiki.archlinux.org/index.php/Unified_Extensible_Firmware_Interface#UEFI_Shell. Note that you need a 2.0 shell, otherwise you will not be able to add an entry to the menu. 
 
 
Note that it is also possible to simply place the UEFI shell in your UEFI system partition if you do not have a flash drive at hand:
 
wget https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2/ShellBinPkg/UefiShell/X64/Shell.efi
 
mkdir -p /boot/efi/EFI/boot
 
mv Shell.efi /boot/efi/EFI/boot/bootx64.efi
 
 
This will cause the ideapad UEFI firmware to automatically add a new boot option named "EFI HDD Device" which will by default come first.
 
 
Now reboot, go to the BIOS (press F2), enable UEFI and exit, then press F12 and the flash drive should show up in the menu (you want to choose the UEFI entry). Now you should be in the UEFI shell. There is quite a lot that you can do, but be careful because a mistake can seriously compromise the machine. This guide may be worth looking at:  http://software.intel.com/en-us/articles/uefi-shell/
 
 
For now, you just need the command bcfg. To add Arch to the first entry of the menu, use:
 
 
bcfg boot add 0 fs1:\EFI\arch_grub\grubx64.efi "Arch Linux" 
 
 
The command assumes that the system partition is installed on the first drive. This partition has the loader (grubx64.efi) and this little program is what loads grub2. If you add Arch to the first entry, you can boot to it without pressing F12. 
 
 
 
To see the menu entries, use:
 
 
bcfg boot dump -v
 
 
To delete, say the 3rd entry:
 
 
bcfg boot rm 3
 
 
Once you are happy with the menu entries, reboot and you should be able to boot into Arch.
 
 
===Dual-Boot With Windows 8===
 
 
To avoid problems with the Arch install, put Windows on the second drive (this means that the second drive should also have a GUID partition table, since Windows 8 only works with one).
 
 
Windows 8 uses UEFI, so you can press F12 to choose between Arch and Windows, or just adjust this at the BIOS. Another option is to use the Windows program EasyBCD and add Arch to the Windows boot loader. In this example, Windows is going to be added to grub2.
 
 
To do this, from Arch, mount the Windows system partition and find its UUID:
 
 
mount /dev/sdb1 /mnt
 
grub-probe --target=fs_uuid /media/EFI/Microsoft/Boot/bootmgfw.efi 
 
 
Take note of the output (something like 1ce5-7f28). Now copy the output of:
 
 
grub-probe --target=hints_string /media/EFI/Microsoft/Boot/bootmgfw.efi
 
 
Then, add something like this to /etc/grub.d/0_custom:
 
 
menuentry "Microsoft Windows 8 x86_64 UEFI-GPT" {
 
    insmod part_gpt
 
    insmod fat
 
    insmod search_fs_uuid
 
    insmod chain
 
    search --fs-uuid --no-floppy --set=root --hint-bios=hd0,gpt1 --hint-efi=hd0,gpt1 --hint-baremetal=ahci0,gpt1 1ce5-7f28
 
    chainloader /efi/Microsoft/Boot/bootmgfw.efi
 
}
 
 
Finally generate the grub2 configuration file (grub.cfg):
 
 
grub-mkconfig -o /boot/efi/EFI/grub/grub.cfg
 
 
Now you should be able to see an entry for Windows 8 on grub2.
 
 
The Y580 has a huge 1TB second drive, making it easy to install other OS. With grub2, you can install other distros, and then run grub-mkconfig to add the new entry. To make this easier, install os-prober so that grub2 can find other OS automatically (it does not work for Windows 8):
 
 
pacman -S os-prober
 
==DISPLAY==
 
Since the arrival of 3.7.X series kernel, this machine boots into black screen. You will have to add acpi_backlight=vendor to kernel command line to boot into a visible display. You can alternatively increase the screen brightness during every boot also, since the screen brightness is set to 0 wrongly by the kernel acpi by default.
 
 
===NVIDIA Card===
 
===NVIDIA Card===
  
The Y580 uses NVIDIA's Optimus technology, which is not officially supported on Linux. A possible solution is to install Bumblebee (https://wiki.archlinux.org/index.php/Bumblebee) and to access the card with optirun. It seems this currently does not work with the Y580. However, you can still use CUDA, which is good if you use apps like Blender or if you develop CUDA C programs.
+
The Y580 uses NVIDIA's Optimus technology, which is not officially supported on Linux yet. A possible solution is to install Bumblebee (https://wiki.archlinux.org/index.php/Bumblebee) and to access the card with optirun. However, you can still use CUDA, which is good if you use apps like Blender or if you develop CUDA C programs.
 
+
Lenovo machines (Y470, Y570, Y580) require a kernel patch for the correct acpi handling if you are using 3.7.X kernel. The modified PKGBUILD with sources are located [http://dl.dropbox.com/u/106654446/linux.tar.gz here ].
+
Kernel 3.8 will need a [https://patchwork.kernel.org/patch/1917651/ different patch] than what is linked above. There is proposal from kernel developer Rafael J. Wysocki for kernel 3.9 for a proper fix in the kernel tree. Until then patching the kernel is necessary.
+
 
+
Relevant external links:
+
 
+
1.[https://github.com/Bumblebee-Project/bbswitch/issues/2#issuecomment-11711656 Bumblebee-issue]
+
 
+
2.[https://bugzilla.kernel.org/show_bug.cgi?id=42696 kernel bugzilla]
+
 
+
3.[https://patchwork.kernel.org/patch/1917651/ patch for 3.8 kernel]
+
  
 
===Nvidia Bumblebee support===
 
===Nvidia Bumblebee support===
In Linux currently [https://wiki.archlinux.org/index.php/Bumblebee bumblebee ] is the easiest option for a optimus like solution. For this laptop you will need a patched kernel as given in NVIDIA Card section, [https://aur.archlinux.org/packages/dkms-bbswitch/ dkms-bbswitch], [https://aur.archlinux.org/packages/nvidia-bumblebee/ bumblee-nvidia], [https://aur.archlinux.org/packages/nvidia-utils-bumblebee/ bumblebee-nvidia-utils] package all available from AUR. After installation of these packages you will need to add your user name to bumblebee group.You have to start the bumblebeed service manually. A short guide given here [https://wiki.archlinux.org/index.php/Bumblebee#Start_Bumblebee to start bumblebee]. After this setup you will have to edit the {{ic|/etc/bumblebee/xorg.conf.nvidia}} Device option "ConnectedMonitor". You will have to switch from "FDP" to "CRT-0".
+
In Linux currently [https://wiki.archlinux.org/index.php/Bumblebee bumblebee ] is the easiest option for a optimus like solution. For this laptop you will need bbswitch, Bumblebee, nvidia from official Arch repos. After installation of these packages you will need to add your user name to bumblebee group.You have to start the bumblebeed service manually. A short guide given here [https://wiki.archlinux.org/index.php/Bumblebee#Bumblebee:_Optimus_for_Linux bumblebee].
Make sure nouvea is blacklisted.
+
  
 
===Driver===
 
===Driver===
  
 
To compile and run CUDA programs, you need a NVIDIA driver and the cuda-toolkit. Any driver with version 295.59 or higher will work.
 
To compile and run CUDA programs, you need a NVIDIA driver and the cuda-toolkit. Any driver with version 295.59 or higher will work.
{{Note: The only exception being the 302.17}}
+
{{Note|The only exception being the 302.17}}
  
 
As of 10/24/2012, the newest driver is 304.60. To install it, you need a modified version of the package nvidia-utils from extra. The easiest way to do this is to install the package nvidia-utils-custom from the AUR: https://aur.archlinux.org/packages.php?ID=60991.  
 
As of 10/24/2012, the newest driver is 304.60. To install it, you need a modified version of the package nvidia-utils from extra. The easiest way to do this is to install the package nvidia-utils-custom from the AUR: https://aur.archlinux.org/packages.php?ID=60991.  
Line 165: Line 34:
  
 
===nvidia-utils PKGBUILD===
 
===nvidia-utils PKGBUILD===
<pre>
+
{{bc|<nowiki>
 
# $Id$
 
# $Id$
 
# Maintainer: Thomas Baechler <thomas@archlinux.org>
 
# Maintainer: Thomas Baechler <thomas@archlinux.org>
Line 231: Line 100:
 
     # GLX extension module for X
 
     # GLX extension module for X
 
     #install -D -m755 "libglx.so.${pkgver}" "${pkgdir}/usr/lib/xorg/modules/extensions/libglx.so.${pkgver}"
 
     #install -D -m755 "libglx.so.${pkgver}" "${pkgdir}/usr/lib/xorg/modules/extensions/libglx.so.${pkgver}"
     #ln -s "libglx.so.${pkgver}" "${pkgdir}/usr/lib/xorg/modules/extensions/libglx.so" # X doesn't find glx otherwise
+
     #ln -s "libglx.so.${pkgver}" "${pkgdir}/usr/lib/xorg/modules/extensions/libglx.so" # X does not find glx otherwise
 
     # OpenGL library
 
     # OpenGL library
 
     #install -D -m755 "libGL.so.${pkgver}" "${pkgdir}/usr/lib/libGL.so.${pkgver}"
 
     #install -D -m755 "libGL.so.${pkgver}" "${pkgdir}/usr/lib/libGL.so.${pkgver}"
Line 264: Line 133:
 
     install -D -m755 nvidia-smi "${pkgdir}/usr/bin/nvidia-smi"
 
     install -D -m755 nvidia-smi "${pkgdir}/usr/bin/nvidia-smi"
 
     install -D -m644 nvidia-smi.1.gz "${pkgdir}/usr/share/man/man1/nvidia-smi.1.gz"
 
     install -D -m644 nvidia-smi.1.gz "${pkgdir}/usr/share/man/man1/nvidia-smi.1.gz"
 
  
 
     install -D -m644 LICENSE "${pkgdir}/usr/share/licenses/nvidia/LICENSE"
 
     install -D -m644 LICENSE "${pkgdir}/usr/share/licenses/nvidia/LICENSE"
Line 275: Line 143:
 
}
 
}
  
</pre>
+
</nowiki>}}
  
 
It is not necessary to patch other packages. If you want, you can also install nvidia-custom (https://aur.archlinux.org/packages.php?ID=60981) and opencl-nvidia-custom (https://aur.archlinux.org/packages.php?ID=61443) from the AUR, but they are not much different from the official ones (nvidia and opencl-nvidia from extra).
 
It is not necessary to patch other packages. If you want, you can also install nvidia-custom (https://aur.archlinux.org/packages.php?ID=60981) and opencl-nvidia-custom (https://aur.archlinux.org/packages.php?ID=61443) from the AUR, but they are not much different from the official ones (nvidia and opencl-nvidia from extra).
Line 287: Line 155:
 
===Configurations===
 
===Configurations===
  
You need to load the acpi-handle-hack module first, then the nvidia module. Depending on your system, this may be enough, but it may be necessary to create devices for CUDA. One way to accomplish this is to add the following to your /etc/rc.local:
+
Since Linux 3.9, the acpi-handle-hack module is no longer necessary for the Nvidia video adapter to get recognized. Depending on your system, just loading the nvidia module may be enough, but it may be necessary to create devices for CUDA. One way to accomplish this is to add the following to your /etc/rc.local:
  
<pre>
+
{{bc|<nowiki>
/sbin/modprobe acpi-handle-hack
+
 
/sbin/modprobe nvidia
 
/sbin/modprobe nvidia
  
Line 310: Line 177:
 
else
 
else
 
exit 1
 
exit 1
</pre>
+
fi
 +
</nowiki>}}
  
 
If you are using systemd, you can get the /etc/rc.local loaded at boot by adding a new service. Create the following file:
 
If you are using systemd, you can get the /etc/rc.local loaded at boot by adding a new service. Create the following file:
  
<pre>
+
{{bc|1=
 
#/etc/systemd/system/rc-local.service
 
#/etc/systemd/system/rc-local.service
  
Line 329: Line 197:
 
[Install]
 
[Install]
 
WantedBy=multi-user.target
 
WantedBy=multi-user.target
</pre>
+
}}
  
 
Then make this service load at boot with:
 
Then make this service load at boot with:
Line 351: Line 219:
  
 
If this works without errors, you are all set!
 
If this works without errors, you are all set!
 
===Other Distributions===
 
 
1.The above setup not only works only with Arch Linux and it may be even easier with other distros. For example, with Ubuntu 12.04 or Linux Mint 13, install the acpi-handle-hack module and then get the official nvidia-current (no need to patch it) package: 
 
 
apt-get install nvidia-current
 
 
2. For Debian wheezy you shall need the lenovo-hack as described above and additional packages from [http://suwako.nomanga.net/ suwako repos] which has dkms-bbswtich bumblebee and bumblebee-nvidia.The easiest option is available from Debian. No need to blacklist anything.
 
 
3. For Fedora, RHEL 6.X series and its clones the external repos from  [http://techies.ncsu.edu/wiki/bumblebee ncsu.edu] with installation of non-free nvidia from their non-free repos at [http://techies.ncsu.edu/wiki/bumblebee-nvidia bumblebee-nvidia] is necessary. Installing nvidia from rpmfusion shall break your X. You will have to edit the connected monitor section to "CRT-0" from "FDP" in Fedora 18. You will have to blacklist nouvea drivers at kernel command line to (xdriver=vesa nouveau.modeset=0 rd.driver.blacklist=nouveau).
 
 
4. Starting from 3.7.X kernel you will need to patch the kernel for proper acpi handling for any distro. For 3.9 linux kernel it may not be necessary.
 

Revision as of 19:06, 16 September 2013

DISPLAY

As of the arrival of the 3.7.X series kernel, this machine boots into black screen. You will have to add acpi_backlight=vendor to kernel command line to boot into a visible display. You can alternatively increase the screen brightness during every boot also, since the screen brightness is set to 0 wrongly by the kernel acpi by default.

As of September 2013, this doesn't seem to be an issue anymore, at least not on kernel 3.10.

Intel Card

This machine has two video cards, explained in detail below

$ lspci | grep VGA
00:02.0 VGA compatible controller: Intel Corporation 3rd Gen Core processor Graphics Controller (rev 09)
01:00.0 VGA compatible controller: NVIDIA Corporation GK107M [GeForce GTX 660M] (rev a1)

In order to get the machine working for basic purposes, simply install xf86-video-intel and use the intel driver. You can leave the nvidia card disabled (and powered down?) just by not installing nvidia-304xx.

NVIDIA Card

The Y580 uses NVIDIA's Optimus technology, which is not officially supported on Linux yet. A possible solution is to install Bumblebee (https://wiki.archlinux.org/index.php/Bumblebee) and to access the card with optirun. However, you can still use CUDA, which is good if you use apps like Blender or if you develop CUDA C programs.

Nvidia Bumblebee support

In Linux currently bumblebee is the easiest option for a optimus like solution. For this laptop you will need bbswitch, Bumblebee, nvidia from official Arch repos. After installation of these packages you will need to add your user name to bumblebee group.You have to start the bumblebeed service manually. A short guide given here bumblebee.

Driver

To compile and run CUDA programs, you need a NVIDIA driver and the cuda-toolkit. Any driver with version 295.59 or higher will work.

Note: The only exception being the 302.17

As of 10/24/2012, the newest driver is 304.60. To install it, you need a modified version of the package nvidia-utils from extra. The easiest way to do this is to install the package nvidia-utils-custom from the AUR: https://aur.archlinux.org/packages.php?ID=60991.

Alternatively, you can use ABS and patch the package yourself. If you are not familiar with the process, read the wiki: https://wiki.archlinux.org/index.php/Arch_Build_System. The exact way to rebuild a package depends on your own preferences, one way is to download the source code (in this case, the driver, obtained here: ftp://download.nvidia.com/XFree86/Linux-x86_64/304.60/NVIDIA-Linux-x86_64-304.60-no-compat32.run), copy it to the build directory, edit the PKGBUILD, and then run makepkg.

Edit the PKGBUILD of nvidia-utils. You may need to change the pkgver, the source item, and the md5sum. The most important thing is that nvidia-utils conflicts with libgl, but if you uninstall libgl, Gnome only starts in fallback mode (not sure about other DE). Because of this, you need to modify the PKGBUILD, so that it either does not install certain libraries (libglx.so, libGL.so) or that it installs them in another location. With the PKGBUILD below, those libraries are not installed (note the commented lines for the GLX extension module, and the empty 'conflicts' line).

nvidia-utils PKGBUILD

# $Id$
# Maintainer: Thomas Baechler <thomas@archlinux.org>
# Contributor: James Rayner <iphitus@gmail.com>
pkgbase=nvidia-utils
pkgname=('nvidia-utils' 'opencl-nvidia')
pkgver=304.60
pkgrel=1
arch=('i686' 'x86_64')
url="http://www.nvidia.com/"
license=('custom')
options=('!strip')

if [ "$CARCH" = "i686" ]; then
    _arch='x86'
    _pkg="NVIDIA-Linux-${_arch}-${pkgver}"
    source=("ftp://download.nvidia.com/XFree86/Linux-${_arch}/${pkgver}/${_pkg}.run")
    md5sums=('42b9887076b2ebcf1af5ee13bc332ccb')
elif [ "$CARCH" = "x86_64" ]; then
    _arch='x86_64'
    _pkg="NVIDIA-Linux-${_arch}-${pkgver}-no-compat32"
    source=("ftp://download.nvidia.com/XFree86/Linux-${_arch}/${pkgver}/${_pkg}.run")
    md5sums=('7248399a125808e3bbc9c66da99a098d')
fi

create_links() {
    # create soname links
    while read -d '' _lib; do
        _soname="$(dirname "${_lib}")/$(readelf -d "${_lib}" | sed -nr 's/.*Library soname: \[(.*)\].*/\1/p')"
        [[ -e "${_soname}" ]] || ln -s "$(basename "${_lib}")" "${_soname}"
        [[ -e "${_soname/.[0-9]*/}" ]] || ln -s "$(basename "${_soname}")" "${_soname/.[0-9]*/}"
    done < <(find "${pkgdir}" -type f -name '*.so*' -print0)
}

build() {
    cd "${srcdir}"
    sh "${_pkg}.run" --extract-only
}

package_opencl-nvidia() {
    pkgdesc="OpenCL implemention for NVIDIA"
    depends=('libcl' 'zlib')
    optdepends=('opencl-headers: headers necessary for OpenCL development')
    cd "${srcdir}/${_pkg}"

    # OpenCL
    install -D -m644 nvidia.icd "${pkgdir}/etc/OpenCL/vendors/nvidia.icd"
    install -D -m755 "libnvidia-compiler.so.${pkgver}" "${pkgdir}/usr/lib/libnvidia-compiler.so.${pkgver}"
    install -D -m755 "libnvidia-opencl.so.${pkgver}" "${pkgdir}/usr/lib/libnvidia-opencl.so.${pkgver}"

    create_links
}

package_nvidia-utils() {
    pkgdesc="NVIDIA drivers utilities and libraries."
    depends=('xorg-server' 'libxvmc')
    optdepends=('gtk2: nvidia-settings' 'pkg-config: nvidia-xconfig'
                'opencl-nvidia: OpenCL support')
    conflicts=('')
    provides=('libgl')
    cd "${srcdir}/${_pkg}"

    # X driver
    install -D -m755 nvidia_drv.so "${pkgdir}/usr/lib/xorg/modules/drivers/nvidia_drv.so"
    # GLX extension module for X
    #install -D -m755 "libglx.so.${pkgver}" "${pkgdir}/usr/lib/xorg/modules/extensions/libglx.so.${pkgver}"
    #ln -s "libglx.so.${pkgver}" "${pkgdir}/usr/lib/xorg/modules/extensions/libglx.so"	# X does not find glx otherwise
    # OpenGL library
    #install -D -m755 "libGL.so.${pkgver}" "${pkgdir}/usr/lib/libGL.so.${pkgver}"
    # OpenGL core library
    install -D -m755 "libnvidia-glcore.so.${pkgver}" "${pkgdir}/usr/lib/libnvidia-glcore.so.${pkgver}"
    # XvMC
    install -D -m644 libXvMCNVIDIA.a "${pkgdir}/usr/lib/libXvMCNVIDIA.a"
    install -D -m755 "libXvMCNVIDIA.so.${pkgver}" "${pkgdir}/usr/lib/libXvMCNVIDIA.so.${pkgver}"
    # VDPAU
    install -D -m755 "libvdpau_nvidia.so.${pkgver}" "${pkgdir}/usr/lib/vdpau/libvdpau_nvidia.so.${pkgver}"
    # nvidia-tls library
    install -D -m755 "tls/libnvidia-tls.so.${pkgver}" "${pkgdir}/usr/lib/libnvidia-tls.so.${pkgver}"
    install -D -m755 "libnvidia-cfg.so.${pkgver}" "${pkgdir}/usr/lib/libnvidia-cfg.so.${pkgver}"

    install -D -m755 "libnvidia-ml.so.${pkgver}" "${pkgdir}/usr/lib/libnvidia-ml.so.${pkgver}"
    # CUDA
    install -D -m755 "libcuda.so.${pkgver}" "${pkgdir}/usr/lib/libcuda.so.${pkgver}"
    install -D -m755 "libnvcuvid.so.${pkgver}" "${pkgdir}/usr/lib/libnvcuvid.so.${pkgver}"

    # nvidia-xconfig
    install -D -m755 nvidia-xconfig "${pkgdir}/usr/bin/nvidia-xconfig"
    install -D -m644 nvidia-xconfig.1.gz "${pkgdir}/usr/share/man/man1/nvidia-xconfig.1.gz"
    # nvidia-settings
    install -D -m755 nvidia-settings "${pkgdir}/usr/bin/nvidia-settings"
    install -D -m644 nvidia-settings.1.gz "${pkgdir}/usr/share/man/man1/nvidia-settings.1.gz"
    install -D -m644 nvidia-settings.desktop "${pkgdir}/usr/share/applications/nvidia-settings.desktop"
    install -D -m644 nvidia-settings.png "${pkgdir}/usr/share/pixmaps/nvidia-settings.png"
    sed -e 's:__UTILS_PATH__:/usr/bin:' -e 's:__PIXMAP_PATH__:/usr/share/pixmaps:' -i "${pkgdir}/usr/share/applications/nvidia-settings.desktop"
    # nvidia-bug-report
    install -D -m755 nvidia-bug-report.sh "${pkgdir}/usr/bin/nvidia-bug-report.sh"
    # nvidia-smi
    install -D -m755 nvidia-smi "${pkgdir}/usr/bin/nvidia-smi"
    install -D -m644 nvidia-smi.1.gz "${pkgdir}/usr/share/man/man1/nvidia-smi.1.gz"

    install -D -m644 LICENSE "${pkgdir}/usr/share/licenses/nvidia/LICENSE"
    ln -s nvidia "${pkgdir}/usr/share/licenses/nvidia-utils"
    install -D -m644 README.txt "${pkgdir}/usr/share/doc/nvidia/README"
    install -D -m644 NVIDIA_Changelog "${pkgdir}/usr/share/doc/nvidia/NVIDIA_Changelog"
    ln -s nvidia "${pkgdir}/usr/share/doc/nvidia-utils"

    create_links
}

It is not necessary to patch other packages. If you want, you can also install nvidia-custom (https://aur.archlinux.org/packages.php?ID=60981) and opencl-nvidia-custom (https://aur.archlinux.org/packages.php?ID=61443) from the AUR, but they are not much different from the official ones (nvidia and opencl-nvidia from extra).

If you rebuild nvidia-utils yourself, you may want to add it to the IgnorePkg line of your /etc/pacman.conf, so that the next system update does not break your system.

CUDA Toolkit

Install the package from community (https://www.archlinux.org/packages/community/x86_64/cuda/).

Configurations

Since Linux 3.9, the acpi-handle-hack module is no longer necessary for the Nvidia video adapter to get recognized. Depending on your system, just loading the nvidia module may be enough, but it may be necessary to create devices for CUDA. One way to accomplish this is to add the following to your /etc/rc.local:

/sbin/modprobe nvidia

if [ "$?" -eq 0 ]; then

# Count the number of NVIDIA controllers found.

N3D=`lspci | grep -i NVIDIA | grep "3D controller" | wc -l`

NVGA=`lspci | grep -i NVIDIA | grep "VGA compatible controller" | wc -l`

N=`expr $N3D + $NVGA - 1`

for i in `seq 0 $N`; do
mknod -m 666 /dev/nvidia$i c 195 $i
done

mknod -m 666 /dev/nvidiactl c 195 255
else
exit 1
fi

If you are using systemd, you can get the /etc/rc.local loaded at boot by adding a new service. Create the following file:

#/etc/systemd/system/rc-local.service

[Unit]
Description=/etc/rc.local Compatibility

[Service]
Type=oneshot
ExecStart=/etc/rc.local
TimeoutSec=0
StandardInput=tty
RemainAfterExit=yes

[Install]
WantedBy=multi-user.target

Then make this service load at boot with:

systemctl enable rc-local.service

Testing it

The cuda package includes both the cuda-toolkit and sdk. Before testing it, reboot and it should be working. To compile and run deviceQuery from the sdk:

cd /opt/cuda/samples/1_Utilities/deviceQuery
make
./deviceQuery

Alternatively, you can compile your own code and run it. To compile, say hello.cu, use:

nvcc hello.cu

Now you can run the executable:

./a.out

If this works without errors, you are all set!