more debugging of page table issues

xencall/src/domctl.rs

@@ -1,9 +1,10 @@
 use crate::sys::{
     AddressSize, ArchDomainConfig, CreateDomain, DomCtl, DomCtlValue, DomCtlVcpuContext,
-    GetDomainInfo, HypercallInit, MaxMem, MaxVcpus, VcpuGuestContext, VcpuGuestContextAny,
-    HYPERVISOR_DOMCTL, XEN_DOMCTL_CREATEDOMAIN, XEN_DOMCTL_DESTROYDOMAIN, XEN_DOMCTL_GETDOMAININFO,
-    XEN_DOMCTL_GETVCPUCONTEXT, XEN_DOMCTL_HYPERCALL_INIT, XEN_DOMCTL_INTERFACE_VERSION,
-    XEN_DOMCTL_MAX_MEM, XEN_DOMCTL_MAX_VCPUS, XEN_DOMCTL_PAUSEDOMAIN, XEN_DOMCTL_SETVCPUCONTEXT,
+    GetDomainInfo, GetPageFrameInfo3, HypercallInit, MaxMem, MaxVcpus, VcpuGuestContext,
+    VcpuGuestContextAny, HYPERVISOR_DOMCTL, XEN_DOMCTL_CREATEDOMAIN, XEN_DOMCTL_DESTROYDOMAIN,
+    XEN_DOMCTL_GETDOMAININFO, XEN_DOMCTL_GETPAGEFRAMEINFO3, XEN_DOMCTL_GETVCPUCONTEXT,
+    XEN_DOMCTL_HYPERCALL_INIT, XEN_DOMCTL_INTERFACE_VERSION, XEN_DOMCTL_MAX_MEM,
+    XEN_DOMCTL_MAX_VCPUS, XEN_DOMCTL_PAUSEDOMAIN, XEN_DOMCTL_SETVCPUCONTEXT,
     XEN_DOMCTL_SET_ADDRESS_SIZE,
 };
 use crate::{XenCall, XenCallError};
@@ -11,6 +12,7 @@ use log::trace;
 use std::ffi::c_ulong;
 use std::os::fd::AsRawFd;
 use std::ptr::addr_of_mut;
+use std::slice;
 
 pub struct DomainControl<'a> {
     call: &'a XenCall,
@@ -214,6 +216,34 @@ impl DomainControl<'_> {
         Ok(())
     }
 
+    pub fn get_page_frame_info(
+        &self,
+        domid: u32,
+        frames: &[u64],
+    ) -> Result<Vec<u64>, XenCallError> {
+        let mut buffer: Vec<u64> = frames.to_vec();
+        let mut domctl = DomCtl {
+            cmd: XEN_DOMCTL_GETPAGEFRAMEINFO3,
+            interface_version: XEN_DOMCTL_INTERFACE_VERSION,
+            domid,
+            value: DomCtlValue {
+                get_page_frame_info: GetPageFrameInfo3 {
+                    num: buffer.len() as u64,
+                    array: buffer.as_mut_ptr() as c_ulong,
+                },
+            },
+        };
+        self.call
+            .hypercall1(HYPERVISOR_DOMCTL, addr_of_mut!(domctl) as c_ulong)?;
+        let slice = unsafe {
+            slice::from_raw_parts_mut(
+                domctl.value.get_page_frame_info.array as *mut u64,
+                domctl.value.get_page_frame_info.num as usize,
+            )
+        };
+        Ok(slice.to_vec())
+    }
+
     pub fn hypercall_init(&self, domid: u32, gmfn: u64) -> Result<(), XenCallError> {
         trace!(
             "domctl fd={} hypercall_init domid={} gmfn={}",

xencall/src/memory.rs

@@ -1,10 +1,11 @@
 use crate::sys::{
-    MemoryReservation, MultiCallEntry, HYPERVISOR_MEMORY_OP, XEN_MEM_POPULATE_PHYSMAP,
+    MemoryMap, MemoryReservation, MmuExtOp, MultiCallEntry, HYPERVISOR_MEMORY_OP,
+    HYPERVISOR_MMUEXT_OP, XEN_MEM_MEMORY_MAP, XEN_MEM_POPULATE_PHYSMAP,
 };
 use crate::{XenCall, XenCallError};
 
 use log::trace;
-use std::ffi::c_ulong;
+use std::ffi::{c_uint, c_ulong};
 use std::os::fd::AsRawFd;
 use std::ptr::addr_of_mut;
 
@@ -17,6 +18,26 @@ impl MemoryControl<'_> {
         MemoryControl { call }
     }
 
+    pub fn get_memory_map(&self, size_of_entry: usize) -> Result<Vec<u8>, XenCallError> {
+        let mut memory_map = MemoryMap {
+            count: 0,
+            buffer: 0,
+        };
+        self.call.hypercall2(
+            HYPERVISOR_MEMORY_OP,
+            XEN_MEM_MEMORY_MAP as c_ulong,
+            addr_of_mut!(memory_map) as c_ulong,
+        )?;
+        let mut buffer = vec![0u8; memory_map.count as usize * size_of_entry];
+        memory_map.buffer = buffer.as_mut_ptr() as c_ulong;
+        self.call.hypercall2(
+            HYPERVISOR_MEMORY_OP,
+            XEN_MEM_MEMORY_MAP as c_ulong,
+            addr_of_mut!(memory_map) as c_ulong,
+        )?;
+        Ok(buffer)
+    }
+
     pub fn populate_physmap(
         &self,
         domid: u32,
@@ -62,4 +83,24 @@ impl MemoryControl<'_> {
         let extents = extent_starts[0..code as usize].to_vec();
         Ok(extents)
     }
+
+    pub fn mmuext(
+        &self,
+        domid: u32,
+        cmd: c_uint,
+        arg1: u64,
+        arg2: u64,
+    ) -> Result<(), XenCallError> {
+        let mut ops = MmuExtOp { cmd, arg1, arg2 };
+
+        self.call
+            .hypercall4(
+                HYPERVISOR_MMUEXT_OP,
+                addr_of_mut!(ops) as c_ulong,
+                1,
+                0,
+                domid as c_ulong,
+            )
+            .map(|_| ())
+    }
 }

xencall/src/sys.rs

@@ -223,6 +223,7 @@ pub union DomCtlValue {
     pub hypercall_init: HypercallInit,
     pub vcpu_context: DomCtlVcpuContext,
     pub address_size: AddressSize,
+    pub get_page_frame_info: GetPageFrameInfo3,
     pub pad: [u8; 128],
 }
 
@@ -287,6 +288,13 @@ pub struct GetDomainInfo {
     pub arch: ArchDomainConfig,
 }
 
+#[repr(C)]
+#[derive(Copy, Clone, Debug)]
+pub struct GetPageFrameInfo3 {
+    pub num: u64,
+    pub array: c_ulong,
+}
+
 #[repr(C)]
 #[derive(Copy, Clone, Debug)]
 pub struct ArchDomainConfig {
@@ -342,6 +350,14 @@ pub struct MultiCallEntry {
 }
 
 pub const XEN_MEM_POPULATE_PHYSMAP: u32 = 6;
+pub const XEN_MEM_MEMORY_MAP: u32 = 9;
+
+#[repr(C)]
+#[derive(Copy, Clone, Debug)]
+pub struct MemoryMap {
+    pub count: c_uint,
+    pub buffer: c_ulong,
+}
 
 #[repr(C)]
 #[derive(Copy, Clone, Debug)]
@@ -456,3 +472,13 @@ impl Default for VcpuGuestContext {
 pub union VcpuGuestContextAny {
     pub value: VcpuGuestContext,
 }
+
+#[repr(C)]
+#[derive(Debug, Copy, Clone)]
+pub struct MmuExtOp {
+    pub cmd: c_uint,
+    pub arg1: c_ulong,
+    pub arg2: c_ulong,
+}
+
+pub const MMUEXT_PIN_L4_TABLE: u32 = 3;

xenclient/examples/boot.rs

@@ -19,7 +19,6 @@ fn main() -> Result<(), XenClientError> {
     let call = XenCall::open()?;
     let domctl = DomainControl::new(&call);
     let domid = domctl.create_domain(CreateDomain::default())?;
-    domctl.pause_domain(domid)?;
     domctl.set_max_vcpus(domid, 1)?;
     let result = boot(domid, kernel_image_path.as_str(), &call, &domctl);
     domctl.destroy_domain(domid)?;

xenclient/src/boot.rs

@@ -4,19 +4,20 @@ use crate::sys::{
     XEN_PAGE_SHIFT,
 };
 use crate::x86::{
-    PageTable, PageTableMapping, StartInfo, MAX_GUEST_CMDLINE, X86_GUEST_MAGIC, X86_PAGE_SHIFT,
-    X86_PAGE_SIZE, X86_PAGE_TABLE_MAX_MAPPINGS, X86_PGTABLE_LEVELS, X86_PGTABLE_LEVEL_SHIFT,
-    X86_VIRT_MASK,
+    PageTable, PageTableMapping, SharedInfo, StartInfo, MAX_GUEST_CMDLINE, X86_GUEST_MAGIC,
+    X86_PAGE_SHIFT, X86_PAGE_SIZE, X86_PAGE_TABLE_MAX_MAPPINGS, X86_PGTABLE_LEVELS,
+    X86_PGTABLE_LEVEL_SHIFT, X86_VIRT_MASK,
 };
 use crate::XenClientError;
 use libc::c_char;
 use log::{debug, trace};
 use slice_copy::copy;
 use std::cmp::{max, min};
+use std::mem::size_of;
 use std::slice;
 use xencall::domctl::DomainControl;
 use xencall::memory::MemoryControl;
-use xencall::sys::VcpuGuestContext;
+use xencall::sys::{VcpuGuestContext, MMUEXT_PIN_L4_TABLE};
 use xencall::XenCall;
 
 pub trait BootImageLoader {
@@ -198,7 +199,12 @@ impl BootSetup<'_> {
                         .as_str(),
                     ));
                 }
-                copy(p2m.as_mut_slice(), result.as_slice());
+
+                for (i, item) in result.iter().enumerate() {
+                    let p = (pfn_base + j + i as u64) as usize;
+                    let m = *item;
+                    p2m[p] = m;
+                }
                 j += allocsz;
             }
         }
@@ -263,6 +269,22 @@ impl BootSetup<'_> {
         self.setup_start_info(state, cmdline)?;
         self.setup_hypercall_page(&state.image_info)?;
 
+        self.phys.unmap(state.page_table_segment.pfn)?;
+        self.phys.unmap(state.p2m_segment.pfn)?;
+        let pg_pfn = state.page_table_segment.pfn;
+        let pg_mfn = self.phys.p2m[pg_pfn as usize];
+        debug!(
+            "domain info: {:?}",
+            self.domctl.get_domain_info(self.domid)?
+        );
+        let page_frame_info = self.domctl.get_page_frame_info(self.domid, &[pg_pfn])?;
+        debug!("pgtable page frame info: {:#x}", page_frame_info[0]);
+        debug!("pinning l4 table: mfn={:#x}", pg_mfn);
+        self.memctl
+            .mmuext(self.domid, MMUEXT_PIN_L4_TABLE, pg_mfn, 0)?;
+        debug!("pinned l4 table: {:#x}", state.page_table_segment.pfn);
+        self.setup_shared_info()?;
+
         let mut vcpu = VcpuGuestContext::default();
         vcpu.user_regs.rip = state.image_info.virt_entry;
         vcpu.user_regs.rsp =
@@ -273,7 +295,7 @@ impl BootSetup<'_> {
         vcpu.debugreg[6] = 0xffff0ff0;
         vcpu.debugreg[7] = 0x00000400;
         vcpu.flags = VGCF_IN_KERNEL | VGCF_ONLINE;
-        let cr3_pfn = self.phys.p2m[state.page_table_segment.pfn as usize];
+        let cr3_pfn = pg_mfn;
         debug!(
             "cr3: pfn {:#x} mfn {:#x}",
             state.page_table_segment.pfn, cr3_pfn
@@ -358,9 +380,9 @@ impl BootSetup<'_> {
 
     const PAGE_PRESENT: u64 = 0x001;
     const PAGE_RW: u64 = 0x002;
+    const PAGE_USER: u64 = 0x004;
     const PAGE_ACCESSED: u64 = 0x020;
     const PAGE_DIRTY: u64 = 0x040;
-    const PAGE_USER: u64 = 0x004;
     fn get_pg_prot(&mut self, l: usize, pfn: u64, table: &PageTable) -> u64 {
         let prot = [
             BootSetup::PAGE_PRESENT | BootSetup::PAGE_RW | BootSetup::PAGE_ACCESSED,
@@ -425,6 +447,21 @@ impl BootSetup<'_> {
         Ok(())
     }
 
+    fn setup_shared_info(&mut self) -> Result<(), XenClientError> {
+        let domain_info = self.domctl.get_domain_info(self.domid)?;
+        let info = self.phys.pfn_to_ptr(domain_info.shared_info_frame, 1)? as *mut SharedInfo;
+        unsafe {
+            let size = size_of::<SharedInfo>();
+            let info_as_buff = slice::from_raw_parts_mut(info as *mut u8, size);
+            info_as_buff.fill(0);
+            for i in 0..32 {
+                (*info).vcpu_info[i].evtchn_upcall_mask = 1;
+            }
+            trace!("BootSetup setup_shared_info shared_info={:?}", *info);
+        }
+        Ok(())
+    }
+
     fn load_kernel_segment(
         &mut self,
         image_loader: &dyn BootImageLoader,

xenclient/src/mem.rs

@@ -1,5 +1,7 @@
 use crate::sys::XEN_PAGE_SHIFT;
 use crate::XenClientError;
+use libc::munmap;
+use std::ffi::c_void;
 
 use crate::x86::X86_PAGE_SHIFT;
 use xencall::sys::MmapEntry;
@@ -104,4 +106,28 @@ impl PhysicalPages<'_> {
         self.pages.push(page);
         Ok(addr)
     }
+
+    pub fn unmap(&mut self, pfn: u64) -> Result<(), XenClientError> {
+        let mut page: Option<&PhysicalPage> = None;
+        for item in &self.pages {
+            if pfn >= item.pfn && pfn < (item.pfn + item.count) {
+                break;
+            }
+            page = Some(item);
+        }
+        if page.is_none() {
+            return Err(XenClientError::new("failed to unmap pfn"));
+        }
+        let page = page.unwrap();
+        unsafe {
+            let err = munmap(
+                page.ptr as *mut c_void,
+                (page.count << X86_PAGE_SHIFT) as usize,
+            );
+            if err != 0 {
+                return Err(XenClientError::new("failed to munmap pfn"));
+            }
+        }
+        Ok(())
+    }
 }

xenclient/src/x86.rs

@@ -62,3 +62,53 @@ pub struct StartInfo {
 }
 
 pub const X86_GUEST_MAGIC: &str = "xen-3.0-x86_64";
+
+#[repr(C)]
+#[derive(Debug)]
+pub struct ArchVcpuInfo {
+    pub cr2: u64,
+    pub pad: u64,
+}
+
+#[repr(C)]
+#[derive(Debug)]
+pub struct VcpuInfoTime {
+    pub version: u32,
+    pub pad0: u32,
+    pub tsc_timestamp: u64,
+    pub system_time: u64,
+    pub tsc_to_system_mul: u32,
+    pub tsc_shift: i8,
+    pub flags: u8,
+    pub pad1: [u8; 2],
+}
+
+#[repr(C)]
+#[derive(Debug)]
+pub struct VcpuInfo {
+    pub evtchn_upcall_pending: u8,
+    pub evtchn_upcall_mask: u8,
+    pub evtchn_pending_sel: u64,
+    pub arch_vcpu_info: ArchVcpuInfo,
+    pub vcpu_info_time: VcpuInfoTime,
+}
+
+#[repr(C)]
+#[derive(Debug)]
+pub struct SharedInfo {
+    pub vcpu_info: [VcpuInfo; 32],
+    pub evtchn_pending: [u64; u64::BITS as usize],
+    pub evtchn_mask: [u64; u64::BITS as usize],
+    pub wc_version: u32,
+    pub wc_sec: u32,
+    pub wc_nsec: u32,
+    pub wc_sec_hi: u32,
+
+    // arch shared info
+    pub max_pfn: u64,
+    pub pfn_to_mfn_frame_list_list: u64,
+    pub nmi_reason: u64,
+    pub p2m_cr3: u64,
+    pub p2m_vaddr: u64,
+    pub p2m_generation: u64,
+}