[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] Details on p2m_mem_access_resume()


I've seen that, once the hypervisor code decides that it should send a mem_event to userspace in p2m_mem_access_check() (file: xen/arch/x86/mm/p2m.c), it first pauses the VCPU that tried to write to a page that doesn't allow writes. Then, control is handed over to userspace, and then back to the hypervisor via p2m_mem_access_resume(), which looks like this:

void p2m_mem_access_resume(struct domain *d)
    mem_event_response_t rsp;

    /* Pull all responses off the ring */
    while( mem_event_get_response(d, &d->mem_event->access, &rsp) )
        if ( rsp.flags & MEM_EVENT_FLAG_DUMMY )
        /* Unpause domain */
        if ( rsp.flags & MEM_EVENT_FLAG_VCPU_PAUSED )

What I want to do is: based on the response received from userspace (i.e. if ( rsp.flags & MY_FLAG ) ...), allow the write instruction (without unprotecting the page) and move on. This means: I want to be able to receive a page fault mem_event caused by a write only once, _without_ lifting the write restrictions.

I've tried to achieve this by using hvm_emulate_one(), but in order to be able to use it, I need a valid struct cpu_user_regs *parameter. As I've written to xen-devel in a previous thread, I've initially tried to obtain that via guest_cpu_user_regs(), which was not the way to go. I've recently tried &(d->vcpu[rsp.vcpu_id]->arch.user_regs) instead, with similar success. I've even passed a valid struct cpu_user_regs *parameter to p2m_mem_access_check(), from ept_handle_violation() down, and kept that pointer to use later on from p2m_mem_access_resume(). Nothing has worked so far.

Is it possible, in theory, to achieve what I'm after? Is there another way I might be able to achieve it, if emulating the write instruction is not possible there?

How does p2m_mem_access_resume() fit into the overall code flow, and what can (and can't) I do from it's scope, related to the specific vcpu I'm interested in?

Thank you in advance for your replies,
Razvan Cojocaru

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.