ia64/xen-unstable

changeset 10562:dad1f890aca5

[IA64] update VTI Readme file

Update VTI how-to readme file with physical disk

Signed-off-by: Fred yang <Fred.Yang@intel.com>
author awilliam@xenbuild.aw
date Mon Jul 03 08:15:43 2006 -0600 (2006-07-03)
parents 550786d7d352
children aa8257def6dc
files xen/arch/ia64/tools/README.RunVT
line diff
     1.1 --- a/xen/arch/ia64/tools/README.RunVT	Mon Jul 03 08:12:16 2006 -0600
     1.2 +++ b/xen/arch/ia64/tools/README.RunVT	Mon Jul 03 08:15:43 2006 -0600
     1.3 @@ -1,59 +1,46 @@
     1.4  INSTRUCTIONS FOR Running IPF/Xen with VT-enabled Tiger4 pltform
     1.5  
     1.6 -Note: the Domain0 must be an unmodified Linux
     1.7 -
     1.8 -1) Perform operations in README.xenia64 to get a flattened Xen IPF source tree
     1.9 -
    1.10 -2) Build an unmodified Linux 2.6 kernel
    1.11 -	a) tar xvfz  linux-2.6.11.tar.gz
    1.12 -        b) cp arch/ia64/configs/tiger_defconfig .config
    1.13 -	c) Build linux.
    1.14 -   		1) yes "" | make oldconfig
    1.15 -   		2) make
    1.16 -
    1.17 -3) Build IPF VT-enabled Xen image
    1.18 -	edit xen/arch/ia64/Rules.mk for
    1.19 -		CONFIG_VTI	?= y 	to enable VT-enable build
    1.20 -4) Setup ELILO.CONF
    1.21 -	image=xen
    1.22 -        	label=xen
    1.23 -        	initrd=vmlinux2.6.11		// unmodified Linux kernel image
    1.24 -        	read-only
    1.25 -        	append="nomca root=/dev/sda3"
    1.26 -
    1.27 -STATUS as 4/28/05 - Features implemented for Domain0
    1.28 -
    1.29 -0. Runs unmodified Linux kernel as Domain0
    1.30 -    Validated with Linux 2.6.11 to run Xwindow and NIC on UP logical processor
    1.31 +1. Install a Linux Disk, VT_Disk, to be used by VT
    1.32 +2. Setup the target VT_Disk
    1.33 +	1. Boot VT_Disk
    1.34 +	2. modify following files of VT_Disk
    1.35 +		/boot/efi/efi/redhat/elilo.conf -
    1.36 +			modify "append=" line to have "root=/dev/hda3"
    1.37 +			** note /dev/hda3 must reflect VT_Disk /root partition
    1.38  
    1.39 -1. Take advantage of VT-enabled processor
    1.40 -   a. Processor intercepts guest privileged instruction and deliver Opcode/Cause to Hypervisor
    1.41 -   b. One VPD (Virtual Processor Descriptor) per Virtual Processor
    1.42 -   c. Domains are in a different virtual address space from hypervisor. Domains have one less VA bit than hypervisor, where hypervisor runs in 0xF00000... address protected by the processor from Domains.
    1.43 -
    1.44 -2. vTLB and guest_VHPT
    1.45 -   a. vTLB extending machine TLB entries through hypervisor internal data structure
    1.46 -      vTLB caches Domains installed TR's and TC's, and then installs TC's for Domains instead.
    1.47 -      vTLB implements collision chains
    1.48 -   b. Processor walks hypervisor internal VHPT, not the domain VHPT.  On TLB miss, vTLB is consulted first to put hypervisor cached entry into VHPT without inject TLB miss to domain.
    1.49 -
    1.50 -3. Region ID fix-partitioning
    1.51 -   a. currently hard partition 24bits of RIDs into 16 partitions by using top 4bit.
    1.52 -   b. Hypervisor uses the very last partition RIDs, i.e., 0xFxxxxx RIDs
    1.53 -   c. Effectively supports Domain0 and 14 other DomainN
    1.54 -
    1.55 -4. HyperVisor is mapped with 2 sets of RIDs during runtime, its own RIDs and the active Domain RIDs
    1.56 -   a. Domain RIDs are used by processor to access guest_VHPT during Domain runtime
    1.57 -   b. Hypervisor RIDs are used when Hypervisor is running
    1.58 -   c. Implies there are some Region registers transition on entering/exiting hypervisor
    1.59 -
    1.60 -5. Linux styled pt_regs with minor modification for VT and instruction emulation
    1.61 -   a. Part of Domain registers are saved/restored from VPD
    1.62 -   b. Extended pt_regs to include r4~r7 and Domain's iipa & isr for possible instruction emulation, so no need to save a complete set of switch_stack on IVT entry
    1.63 -
    1.64 -6. Linux styled per virtual processor memory/RSE stacks, which is the same as non-VT domain0
    1.65 -
    1.66 -7. Handles splitted I/DCache design
    1.67 -   Newer IPF processors has split I/Dcaches.  The design takes this into consideration when Xen recopy Domain0 to target address for execution
    1.68 +		/etc/fstab -
    1.69 +			LABEL=/     /     ext3    DEFAULTS  1   1
    1.70 +		  to
    1.71 +			/dev/hda3   /     ext3    DEFAULTS  1   1
    1.72 +                  and other entries accordingly
    1.73 +3. Install Xen and boot XenLinux on your standard Linux disk
    1.74 +        1. modify /boot/efi/efi/redhat/elilo.conf -
    1.75 +			"append=" entry to have "root=/dev/sda3"
    1.76 +	2. modify /etc/fstab -
    1.77 +                        LABEL=/     /     ext3    DEFAULTS  1   1
    1.78 +                  to
    1.79 +                        /dev/sda3   /     ext3    DEFAULTS  1   1
    1.80 +                  and other entries accordingly
    1.81 +4. Reboot XenLinux with VT_Disk in /dev/sdb slot
    1.82 +	1. copy Guest_Firmware.bin into /usr/lib/xen/boot/guest_firmware.bin
    1.83 +	2. modify /etc/xen/xmexample.vti
    1.84 +		disk = [ 'phy:/dev/sdb,ioemu:hda,w' ]
    1.85 +	   and make sure
    1.86 +		kernel=/usr/lib/xen/boot/guest_firmware.bin
    1.87 +5. Make sure XenLinux has SDL installed by
    1.88 +	> rpm -q -a | grep SDL
    1.89 +		SDL-1.2.7-8 SDL-devel-1.2.7-8 
    1.90 +6. Start vncserver from XenLinux
    1.91 +	1. ifconfig  to get XenLinux IP address
    1.92 +	2. vncserver
    1.93 +7. Start VT Domain
    1.94 +	1. From a remote system connect to XenLinux through vnc viewer
    1.95 +	2. On vnc windows
    1.96 +		> xend start
    1.97 +		> xm create /etc/xen/xmexample.vti
    1.98 +	   an EFI shell will popup
    1.99 +		> fs0:
   1.100 +		fs0:> cd efi\redhat
   1.101 +		fs0:> elilo linux
   1.102  
   1.103