Index
> OS Construction > hypervisor |
| Author |
|
|
Feryno 01 Feb 2010, 10:34
I decided to start this thread thanks to:
- f0dder who pointed me to emulators capable to emulate virtualization http://board.flatassembler.net/topic.php?p=108769#108769 - booter who decided to work on hypervisors http://board.flatassembler.net/topic.php?t=11085 developping an hypervisor for intel is now available to everybody, you may developp it as driver for running OS or you may load it at boot stage development of a hypervisor is very close to constructing an OS so this thread is in OS construction I spent few hours to put everything together, so you need to do this: ----- compiling bochs to support intel vmx ----- download bochs 2.4.2 source from http://bochs.sourceforge.net and upack it update the file .conf.linux (or dedicated file for another platform) the necessities are these switches --enable-vmx (I hope there is no explanation needed why it is necessary) --enable-smp (if your hypervisor would run at more CPUs) --enable-sse=3 (or even higher than 3) --enable-sse-extension (vmx requires at least SSE3+SSE_extension enabled or SSE4 or higher) --enable-x86_64 (if you plane to work on 64 bit hypervisor) Code: #!/bin/sh # # .conf.linux # #which_config=normal which_config=plugins CC="gcc" CXX="c++" CFLAGS="-Wall -O3 -fomit-frame-pointer -pipe" # for speed #CFLAGS="-Wall -g -pipe" # for development CXXFLAGS="$CFLAGS" export CC export CXX export CFLAGS export CXXFLAGS case $which_config in normal) ####################################################################### # configuration 1 for release binary RPMs # Include a few standard options, speed optimizations, X11 only. ####################################################################### ./configure --enable-sb16 \ --enable-ne2000 \ --enable-all-optimizations \ --enable-cpu-level=6 \ --enable-x86-64 \ --enable-sse=3 \ --enable-sse-extension \ --enable-vmx \ --enable-smp \ --enable-pci \ --enable-acpi \ --enable-clgd54xx \ --enable-usb \ --enable-usb-ohci \ --enable-show-ips \ ${CONFIGURE_ARGS} ;; plugins) ####################################################################### # configuration 2 for release binary RPMs # Include plugins, every possible gui. ####################################################################### ./configure --enable-sb16 \ --enable-ne2000 \ --enable-all-optimizations \ --enable-cpu-level=6 \ --enable-x86-64 \ --enable-sse=3 \ --enable-sse-extension \ --enable-vmx \ --enable-smp \ --enable-pci \ --enable-acpi \ --enable-clgd54xx \ --enable-usb \ --enable-usb-ohci \ --enable-plugins \ --enable-show-ips \ --with-all-libs \ ${CONFIGURE_ARGS} ;; esac run these commands: ./.conf.linux make make install ----- creating disk image ----- prepare enough space on your disk (2 GB would be enough), then run this command: bximage when promted type these settings: hd flat 2000 hd0.img ----- installing win XP ----- I strongly suggest you to use 1 cpu for installation (simulating 2 CPUs in system drops the performace about twice, so you won't install 5 hours but you will 10 hours) edit the file bochsrc.bxrc, especially the path for XP installation source which is at the line beginning with ata1-slave Code: ############################################### # bochsrc.bxrc file for windows xp disk image # ############################################### # how much memory the emulated machine will have megs: 256 # this is the fastest emulation, very suitable for fast installation: cpu: vendor_string=GenuineIntel, msrs=msrs.def # Intel Core 2 Duo #cpu: count=1:2, vendor_string=GenuineIntel, msrs="msrs.def", brand_string="Intel(R) Core(TM)2 Duo CPU T8100 @2.10GHz" # Intel Core i7 (count=1, cores=4, threads_per_core=2), this is very slow emulation: #cpu: count=1:4:2, vendor_string=GenuineIntel, msrs="msrs.def", brand_string="Intel(R) Core i7" # hard disk ata0: enabled=1, ioaddr1=0x1f0, ioaddr2=0x3f0, irq=14 ata0-master: type=disk, path="hd0.img", mode=flat, cylinders=4063, heads=16, spt=63 ata1: enabled=1, ioaddr1=0x170, ioaddr2=0x370, irq=15 ata1-slave: type=cdrom, path=/mnt/sda4/Big_Fucking_Gun/044/windows_xp_professional_paneuro.iso, status=inserted # if the cdrom source is in drive, then you may use path=/dev/cdrom # choose the boot disk. #boot: c boot: cdrom, disk # where do we send log messages? log: bochsout.txt # enable the mouse mouse: enabled=1 some MSRs must be enabled and set properly also edit the msrs.def file: Code: ############################################### # msrs.def file for Intel Core 2 Duo CPU # ############################################### # # ---------------------------------- # Bochs CPU MSRs configuration # ---------------------------------- # # LEGEND: # ------ # # MSR ADDRESS - MSR address in hex (supplied in ECX register for RDMSR/WRMSR) # MSR TYPE - MSR type, see below # # The following fields have any meaning for MSRs with no type only: # # RESET_HI - reset value of the MSR (bits 63:32) # RESET_LO - reset value of the MSR (bits 31:00) # # NOTE: the value of the MSR doesn't change on INIT (software reset). # # RSRVD_HI - mask of reserved bits (bits 63:32) # RSRVD_LO - mask of reserved bits (bits 31:00) # # NOTE: #GP fault will be generated when trying to modify any of MSR # reserved bits. # # IGNRD_HI - mask of ignored bits (bits 63:32) # IGNRD_LO - mask of ignored bits (bits 31:00) # # NOTE: Ignored bits will keep their reset value, all writes to these # bits are ignored. # # MSR TYPES: # --------- # # 0 - No type. # 1 - MSR contains linear address, # #GP if writing non-canonical address in 64-bit mode. # 2 - MSR contains physical address, # #GP if writing a value which exceeds emulated physical address size. # # ADDRESS TYPE RESET_HI RESET_LO RSRVD_HI RSRVD_LO IGNRD_HI IGNRD_LO # --------------------------------------------------------------------------------- # 0x02c 0 00000000 00000000 00000000 00000000 00000000 00000000 0x03a 0 00000000 00000005 00000000 00000000 00000000 00000000 0x08b 0 0000060C 00000000 00000000 00000000 00000000 00000000 0x485 0 00000000 000403C0 00000000 00000000 00000000 00000000 run this command: bochs -q -f bochsrc.bxrc copying files last about half an hour and the second stage about 5 hours at an 6 years old AMD64 Athlon 3200+ running at 2 GHz with 1 GB RAM under Fedora Core 3 after you finish installation and everything is set up correctly, you may increase the count of CPUs in system (this will slow donw the performace very much, just edit the line for cpu in bochsrc.bxrc and reboot, then XP discovers more cpus, it installs drivers and kernel for SMP don't try to perform the XP installation at SMP (more CPUs than 1) because it is very slow, increase the count of CPUs only after you finish installation) later I'll post some screenshots about sucessful test of 32 bit hypervisor skeleton, I'm writing this at hurry, in case I missed something, don't hesitate to post here If you read my presentation about hypervisor shutdown http://board.flatassembler.net/topic.php?p=100097#100097 under bochs the shutdown required to insert one instruction of writing into stack, e.g. the code sequence Code: mov ecx,cr3 ; this is in my presentation mov cr3,ecx ; this is in my presentation mov [esp],eax ; this is required by bochs |
|||
01 Feb 2010, 10:34 |
|
- I assume there's no way to utilize multicore systems for emulating in bochs... an important distinction between bochs and other Virtual Machines: others try to run code natively as much as possible, whereas bochs is 100% emulation... which is both it's strength and it's weakness. Mostly strength |
Feryno 01 Feb 2010, 13:31
bochs requires SSE3 + SSE3_extension (or SSE4 or even higher) to enable bit 9 of ECX of CPUID_EAX=1 (I found that reading source code of bochs)
cpuid.cc Code: #if (BX_SUPPORT_SSE >= 4) || (BX_SUPPORT_SSE >= 3 && BX_SUPPORT_SSE_EXTENSION > 0) features |= (1<<9); // support SSE3E #endif without enabling that - if you strongly follow manuals then you discover that CPU support virtualization but doesn't support VMX-Capabality MSR so you will be unable to set correctly some VMCS fields (you are even unable to execute the VMXON instruction because you don't know VMCS revision identifier got by reading MSR_IA32_VMX_BASIC = 480h), you won't be either able to construct vm exit handler for reading/writing CR0/CR4 (you need some bit masks from VMX-capability MSRs to protect some bits of control registers against writing) this is fragment from my code checking CPU: Code: align 4 check_CPU: ; returns 0 = CPU OK, other = CPU disliked by this hypervisor design push ebx ; CPUID destroys EBX register so we must preserve it ; discovering GenuineIntel CPU xor eax,eax cpuid sub ebx,'Genu' sub edx,'ineI' sub ecx,'ntel' or ebx,edx or ebx,ecx jnz check_CPU_wrong ; DISCOVERING SUPPORT FOR VMX ; Before system software enters into VMX operation, it must discover the presence of VMX support in the processor. System software can determine whether a processor supports VMX operation using CPUID. If CPUID.1:ECX.VMX[bit 5] = 1, then VMX operation is supported. ; mov eax,1 ; this saves 2 bytes of code but is a bit slowlier (doesn't matter, this procedure is executed jut once...) ; push 1 ; pop rax ; this is fast and small, RBX=0 now lea eax,[ebx+1] cpuid test cl,1 shl 5 jz check_CPU_wrong ; the VMX-Capabality MSRs (480h-48Bh) may be read only if CPUID.01H:ECX.[bit 9] is set to 1 test ch,1 shl (9-8) jz check_CPU_wrong check_CPU_OK: xor eax,eax pop ebx ret check_CPU_wrong: ; mov eax,1 ; this is smaller: mov al,1 pop ebx ret |
|||
|
01 Feb 2010, 13:31 |
|
|
Feryno 01 Feb 2010, 14:41
promissed screenshots
funny how OS underestimated the CPU speed at 3 MHz... 
I measure the count of cycles for executing the CPUID instruction to validate that hypervisor is running, because the CPUID causes unconditional vm exit at Intel VMX - in bochs the change is from 5 CPU cycles to 35, in bare CPU the cycles increase about from 300 (no vmx) to 2000 (under vmx) |
|||
|
01 Feb 2010, 14:41 |
|
< Last Thread | Next Thread > |
Forum Rules:
|