mellownight / aethervisor Goto Github PK

as my title say, amd cpu user is less than intell, how about add intel souport?

as title says, my pc is completely freezing and then restarting after I load the aethervisor driver, no bsods or anything else. I have latest Windows 10 installed, my cpu is ryzen 5 5600x and AMD SVM enabled in bios. Here is a screenshot of what I could catch with dbgview

when trying to map the driver with kdmapper it gives a PAGE_FAULT_IN_NONPAGED_AREA bsod

After enabling the hypervisor, I attempted to use ExAllocatePoolZero to allocate memory within a vmcall, but it failed. Therefore, I concluded that memory must be allocated before enabling the hypervisor. So, I proceeded to do it that way.

I attempted to allocate a block of memory using the ExAllocatePoolZero function within the driver entry function, which is the system process 4, before enabling the hypervisor. However, I discovered that this block of memory is not accessible in the vmcall. Any attempts to read or write to it result in a blue screen. Why is this happening? I can see the allocated memory and its virtual address using windbg, so there is no doubt that the virtual address exists. Why is it not possible to operate on it?

In other words, is it incorrect to use vmcall to pass data between the host and guest machines? Is it not possible to use a buffer to transfer data ?

This is the commit which contains code for getting the small page from a large page using the guest CR3.

When I am working with nested page tables in the vmexit handler, I have to make sure that I am using the correct CR3 context for guest physical address translations. Instead of using MmGetPhysicalAddress which uses the VM root CR3 (usually 0x1AD000), I have to use GetPte with the guest CR3 value to get the physical address of the small page inside of a large page. If I want to read/write from this physical address, I will have to set the PFN of a page mapping window to the physical address of the small page.

I can look at minivisor, pagewalkr, or physmem-scanner, to quickly set up code for mapping physical pages.

I tried using kdmapper to map the driver, but I get a SYSTEM THREAD EXCEPTION NOT HANDLED bsod, but it doesn't seem to do it every time. When it works, I try to run the EXE, but it doesn't give me any output.
I don't know English very well, so I used Google Translate.

I was looking through the source, and wanted to make a recommendation for checking the guest privilege level.

Reference:

In the VMCB, there is a field Cpl which you could use. Something akin to guest_context->state_save_area.cpl. If that's not available for some reason, or you'd prefer an alternative then checking the DPL of the SS register would be ideal. Having a wrapper like below:

segment_descriptor_access_rights ar =
{
	static_cast< unsigned int >(guest_ss_access_rights))
};

const auto cpl = ar.fields.dpl;

I saw the snippet I linked above spread around in various functions, and having a utility function for returning the guest CPL in the VMCB, or from SS might help clean things up; and just be more generally correct.

Glad to see hypervisor development is still an interest to many. It's great fun. Keep on goin'!

VMRUN instruction should raise a #UD exception when EFER.SVME is 0 instead of a #GP(0).

hello i was trying to mapping aethervisor by kdmapper ,
but BSOD was occured - in AllocateNewTable Function, by KeBugCheckEx(MANUALLY_INITIATED_CRASH, 0, 1, 2, 3)
why last_reserved_count value keep increasing? i saw its over 60000 and keep overflowing,,
i recheck my system AMD-V was enabled, clean format again, but same issue was happen. here is my system status

how to solving this and mapping successfully?
should i modify some codes or change my system setting?

======
05.12 add
im using 32GB memory now. is it can make probelm?
i know GetPhysicalMemoryRanges() function obtain memory ranges but AllocateNewTable function's max reserve is 60000.
i didnt know exactly whole codes, can size of physical memory has an effect?

title says it all, I cannot find a reference to any xsetbv handling code, which could easily crash a system

is this undetec on eAC sp i cam use for hack fortnite?

Hello everyone!

I am develop a hypervisor that virtualizes an existing system, and after that I need to trace the guest VM (in particular the application) according to one instruction

I initially thought that amd has something like intel, where I vmwrite in CPU_BASED_VM_EXEC_CONTROL(0x00004002) value CPU_BASED_MONITOR_TRAP_FLAG(0x08000000), but amd-v does not have this, and the tf flag does not work. I also found information that the rf flag can also affect this somehow, but I didn't understand how, since when I switch the rf flag to 1, nothing is intercepted in SvmExitHandler either.

Thus the main question: Can someone tell me how can I trace instructions in the case of amd-v? i.e., for example, initially I virtualize the entire system, after that, for example, I compiled a large application through the clang compiler with my optimization phases, then I need to trace it like:

1. vmrun
2. instruction from a guest
3. vmexit
4. my SvmExitHandler function with Exitcode dispatcher
5. some actions in Exitcode dispatcher
6. back to the first stage

Where can I find any trace rudiments in the source code?