bitkeeper revision 1.1577 (429831a31IZvEeEnalH3oKUTY7jV7A)

[PATCH] vmx-worldswitch-1-to-1.patch

Fix VMX world switch to use 1:1 page tables when the guest has paging
disabled. Also do a printk instead of VMX_DBG_LOG() anytime we crash
a domain.

Signed-off-by: Arun Sharma <arun.sharma@intel.com>

diff --git a/tools/libxc/xc_vmx_build.c b/tools/libxc/xc_vmx_build.c
index 360f3757448968d6b14233385ec5dc0dd2831652..5cbd9b85770eaaf9b134b42f90a9af42ebdaaca6 100644 (file)
--- a/tools/libxc/xc_vmx_build.c
+++ b/tools/libxc/xc_vmx_build.c
@@ -10,7 +10,7 @@
 #include <zlib.h>
 #include "linux_boot_params.h"
 
-#define L1_PROT (_PAGE_PRESENT|_PAGE_RW|_PAGE_ACCESSED)
+#define L1_PROT (_PAGE_PRESENT|_PAGE_RW|_PAGE_ACCESSED|_PAGE_USER)
 #define L2_PROT (_PAGE_PRESENT|_PAGE_RW|_PAGE_ACCESSED|_PAGE_DIRTY|_PAGE_USER)
 
 #define round_pgup(_p)    (((_p)+(PAGE_SIZE-1))&PAGE_MASK)

diff --git a/xen/arch/x86/domain.c b/xen/arch/x86/domain.c
index 30795b5831b0ac964ecc344ad88c71b01eabc157..1f42bd61ba657d369a4b0aceddeb0a1f43b78920 100644 (file)
--- a/xen/arch/x86/domain.c
+++ b/xen/arch/x86/domain.c
@@ -339,7 +339,6 @@ static int vmx_final_setup_guest(
     }
 
     ed->arch.schedule_tail = arch_vmx_do_launch;
-    clear_bit(VMX_CPU_STATE_PG_ENABLED, &ed->arch.arch_vmx.cpu_state);
 
 #if defined (__i386)
     ed->arch.arch_vmx.vmx_platform.real_mode_data = 

diff --git a/xen/arch/x86/vmx.c b/xen/arch/x86/vmx.c
index ac6dc6fde19a6b3c2c0445bd5d796aef0ab05392..3c649b1a7ac5335d1e68f70bb8ed887e648c63e5 100644 (file)
--- a/xen/arch/x86/vmx.c
+++ b/xen/arch/x86/vmx.c
@@ -122,7 +122,6 @@ static void inline __update_guest_eip(unsigned long inst_len)
 
 static int vmx_do_page_fault(unsigned long va, struct cpu_user_regs *regs) 
 {
-    struct exec_domain *ed = current;
     unsigned long eip;
     l1_pgentry_t gpte;
     unsigned long gpa; /* FIXME: PAE */
@@ -137,15 +136,8 @@ static int vmx_do_page_fault(unsigned long va, struct cpu_user_regs *regs)
     }
 #endif
 
-    /*
-     * If vpagetable is zero, then we are still emulating 1:1 page tables,
-     * and we should have never gotten here.
-     */
-    if ( !test_bit(VMX_CPU_STATE_PG_ENABLED, &ed->arch.arch_vmx.cpu_state) )
-    {
-        printk("vmx_do_page_fault while running on 1:1 page table\n");
-        return 0;
-    }
+    if (!vmx_paging_enabled(current))
+        handle_mmio(va, va);
 
     gpte = gva_to_gpte(va);
     if (!(l1e_get_flags(gpte) & _PAGE_PRESENT) )
@@ -399,7 +391,7 @@ static void vmx_io_instruction(struct cpu_user_regs *regs,
 
     vio = (vcpu_iodata_t *) d->arch.arch_vmx.vmx_platform.shared_page_va;
     if (vio == 0) {
-        VMX_DBG_LOG(DBG_LEVEL_1, "bad shared page: %lx", (unsigned long) vio);
+        printk("bad shared page: %lx", (unsigned long) vio);
         domain_crash_synchronous(); 
     }
     p = &vio->vp_ioreq;
@@ -423,7 +415,10 @@ static void vmx_io_instruction(struct cpu_user_regs *regs,
             laddr = (p->dir == IOREQ_WRITE) ? regs->esi : regs->edi;
         }
         p->pdata_valid = 1;
-        p->u.pdata = (void *) gva_to_gpa(laddr);
+
+        p->u.data = laddr;
+        if (vmx_paging_enabled(d))
+                p->u.pdata = (void *) gva_to_gpa(p->u.data);
         p->df = (eflags & X86_EFLAGS_DF) ? 1 : 0;
 
         if (test_bit(5, &exit_qualification)) /* "rep" prefix */
@@ -481,7 +476,7 @@ vmx_copy(void *buf, unsigned long laddr, int size, int dir)
        return 0;
     }
 
-    mfn = phys_to_machine_mapping(l1e_get_pfn(gva_to_gpte(laddr)));
+    mfn = phys_to_machine_mapping(laddr >> PAGE_SHIFT);
     addr = map_domain_mem((mfn << PAGE_SHIFT) | (laddr & ~PAGE_MASK));
 
     if (dir == COPY_IN)
@@ -570,6 +565,12 @@ vmx_world_restore(struct exec_domain *d, struct vmx_assist_context *c)
 
     error |= __vmwrite(CR0_READ_SHADOW, c->cr0);
 
+    if (!vmx_paging_enabled(d)) {
+       VMX_DBG_LOG(DBG_LEVEL_VMMU, "switching to vmxassist. use phys table");
+       __vmwrite(GUEST_CR3, pagetable_val(d->domain->arch.phys_table));
+        goto skip_cr3;
+    }
+
     if (c->cr3 == d->arch.arch_vmx.cpu_cr3) {
        /* 
         * This is simple TLB flush, implying the guest has 
@@ -578,7 +579,7 @@ vmx_world_restore(struct exec_domain *d, struct vmx_assist_context *c)
         */
        mfn = phys_to_machine_mapping(c->cr3 >> PAGE_SHIFT);
        if ((mfn << PAGE_SHIFT) != pagetable_val(d->arch.guest_table)) {
-           VMX_DBG_LOG(DBG_LEVEL_VMMU, "Invalid CR3 value=%lx", c->cr3);
+           printk("Invalid CR3 value=%lx", c->cr3);
            domain_crash_synchronous();
            return 0;
        }
@@ -590,7 +591,7 @@ vmx_world_restore(struct exec_domain *d, struct vmx_assist_context *c)
         */
        VMX_DBG_LOG(DBG_LEVEL_VMMU, "CR3 c->cr3 = %lx", c->cr3);
        if ((c->cr3 >> PAGE_SHIFT) > d->domain->max_pages) {
-           VMX_DBG_LOG(DBG_LEVEL_VMMU, "Invalid CR3 value=%lx", c->cr3);
+           printk("Invalid CR3 value=%lx", c->cr3);
            domain_crash_synchronous(); 
            return 0;
        }
@@ -605,6 +606,8 @@ vmx_world_restore(struct exec_domain *d, struct vmx_assist_context *c)
        __vmwrite(GUEST_CR3, pagetable_val(d->arch.shadow_table));
     }
 
+skip_cr3:
+
     error |= __vmread(CR4_READ_SHADOW, &old_cr4);
     error |= __vmwrite(GUEST_CR4, (c->cr4 | X86_CR4_VMXE));
     error |= __vmwrite(CR4_READ_SHADOW, c->cr4);
@@ -731,18 +734,18 @@ static int vmx_set_cr0(unsigned long value)
     struct exec_domain *d = current;
     unsigned long old_base_mfn, mfn;
     unsigned long eip;
+    int paging_enabled;
 
     /* 
      * CR0: We don't want to lose PE and PG.
      */
+    paging_enabled = vmx_paging_enabled(d);
     __vmwrite(GUEST_CR0, (value | X86_CR0_PE | X86_CR0_PG));
+    __vmwrite(CR0_READ_SHADOW, value);
 
-    if (value & (X86_CR0_PE | X86_CR0_PG) &&
-        !test_bit(VMX_CPU_STATE_PG_ENABLED, &d->arch.arch_vmx.cpu_state)) {
-        /*
-         * Enable paging
-         */
-        set_bit(VMX_CPU_STATE_PG_ENABLED, &d->arch.arch_vmx.cpu_state);
+    VMX_DBG_LOG(DBG_LEVEL_VMMU, "Update CR0 value = %lx\n", value);
+    if ((value & X86_CR0_PE) && (value & X86_CR0_PG) 
+        && !paging_enabled) {
         /*
          * The guest CR3 must be pointing to the guest physical.
          */
@@ -750,8 +753,7 @@ static int vmx_set_cr0(unsigned long value)
                             d->arch.arch_vmx.cpu_cr3 >> PAGE_SHIFT)) ||
              !get_page(pfn_to_page(mfn), d->domain) )
         {
-            VMX_DBG_LOG(DBG_LEVEL_VMMU, "Invalid CR3 value = %lx",
-                        d->arch.arch_vmx.cpu_cr3);
+            printk("Invalid CR3 value = %lx", d->arch.arch_vmx.cpu_cr3);
             domain_crash_synchronous(); /* need to take a clean path */
         }
         old_base_mfn = pagetable_get_pfn(d->arch.guest_table);
@@ -776,8 +778,7 @@ static int vmx_set_cr0(unsigned long value)
     } else {
         if ((value & X86_CR0_PE) == 0) {
             __vmread(GUEST_EIP, &eip);
-            VMX_DBG_LOG(DBG_LEVEL_1,
-               "Disabling CR0.PE at %%eip 0x%lx", eip);
+            VMX_DBG_LOG(DBG_LEVEL_1, "Disabling CR0.PE at %%eip 0x%lx\n", eip);
            if (vmx_assist(d, VMX_ASSIST_INVOKE)) {
                set_bit(VMX_CPU_STATE_ASSIST_ENABLED,
                                        &d->arch.arch_vmx.cpu_state);
@@ -838,7 +839,6 @@ static int mov_to_cr(int gp, int cr, struct cpu_user_regs *regs)
     switch(cr) {
     case 0: 
     {
-       __vmwrite(CR0_READ_SHADOW, value);
        return vmx_set_cr0(value);
     }
     case 3: 
@@ -848,7 +848,7 @@ static int mov_to_cr(int gp, int cr, struct cpu_user_regs *regs)
         /*
          * If paging is not enabled yet, simply copy the value to CR3.
          */
-        if (!test_bit(VMX_CPU_STATE_PG_ENABLED, &d->arch.arch_vmx.cpu_state)) {
+        if (!vmx_paging_enabled(d)) {
             d->arch.arch_vmx.cpu_cr3 = value;
             break;
         }
@@ -876,8 +876,7 @@ static int mov_to_cr(int gp, int cr, struct cpu_user_regs *regs)
                  !VALID_MFN(mfn = phys_to_machine_mapping(value >> PAGE_SHIFT)) ||
                  !get_page(pfn_to_page(mfn), d->domain) )
             {
-                VMX_DBG_LOG(DBG_LEVEL_VMMU, 
-                        "Invalid CR3 value=%lx", value);
+                printk("Invalid CR3 value=%lx", value);
                 domain_crash_synchronous(); /* need to take a clean path */
             }
             old_base_mfn = pagetable_get_pfn(d->arch.guest_table);
@@ -1133,6 +1132,7 @@ asmlinkage void vmx_vmexit_handler(struct cpu_user_regs regs)
         VMX_DBG_LOG(DBG_LEVEL_0, "exit reason = %x", exit_reason);
 
     if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
+        printk("Failed vm entry\n");
         domain_crash_synchronous();         
         return;
     }

diff --git a/xen/arch/x86/vmx_io.c b/xen/arch/x86/vmx_io.c
index 8fc73da632fff0095277c4c6c8ab0a5530ae0ef0..dc18839d2d9fabbee0ba651e0f5b76018e868705 100644 (file)
--- a/xen/arch/x86/vmx_io.c
+++ b/xen/arch/x86/vmx_io.c
@@ -465,7 +465,7 @@ void vmx_intr_assist(struct exec_domain *d)
 void vmx_do_resume(struct exec_domain *d) 
 {
     vmx_stts();
-    if ( test_bit(VMX_CPU_STATE_PG_ENABLED, &d->arch.arch_vmx.cpu_state) )
+    if ( vmx_paging_enabled(d) )
         __vmwrite(GUEST_CR3, pagetable_val(d->arch.shadow_table));
     else
         // paging is not enabled in the guest

diff --git a/xen/arch/x86/vmx_platform.c b/xen/arch/x86/vmx_platform.c
index d3460890512f2d0696efa9e5b571912c213816d9..0dfe6d855a0d58fba918c8c28e978a5140557179 100644 (file)
--- a/xen/arch/x86/vmx_platform.c
+++ b/xen/arch/x86/vmx_platform.c
@@ -418,8 +418,12 @@ int inst_copy_from_guest(unsigned char *buf, unsigned long guest_eip, int inst_l
     }
 
     if ((guest_eip & PAGE_MASK) == ((guest_eip + inst_len) & PAGE_MASK)) {
-        gpte = gva_to_gpte(guest_eip);
-        mfn = phys_to_machine_mapping(l1e_get_pfn(gpte));
+        if (vmx_paging_enabled(current)) {
+                gpte = gva_to_gpte(guest_eip);
+                mfn = phys_to_machine_mapping(l1e_get_pfn(gpte));
+        } else {
+                mfn = phys_to_machine_mapping(guest_eip >> PAGE_SHIFT);
+        }
         ma = (mfn << PAGE_SHIFT) | (guest_eip & (PAGE_SIZE - 1));
         inst_start = (unsigned char *)map_domain_mem(ma);
                 
@@ -508,7 +512,7 @@ static void send_mmio_req(unsigned long gpa,
     } else
         p->count = 1;
 
-    if (pvalid)
+    if ((pvalid) && vmx_paging_enabled(current))
         p->u.pdata = (void *) gva_to_gpa(p->u.data);
 
 #if 0

diff --git a/xen/arch/x86/vmx_vmcs.c b/xen/arch/x86/vmx_vmcs.c
index e840c4c3d0de4a4c06aa5e85d75f933c6a8bb26b..8f1768602fdbd8e86000f1cdc3610aa9d3e0d2ad 100644 (file)
--- a/xen/arch/x86/vmx_vmcs.c
+++ b/xen/arch/x86/vmx_vmcs.c
@@ -291,7 +291,7 @@ construct_init_vmcs_guest(struct cpu_user_regs *regs,
 
     /* Initally PG, PE are not set*/
     shadow_cr = host_env->cr0;
-    shadow_cr &= ~(X86_CR0_PE | X86_CR0_PG);
+    shadow_cr &= ~X86_CR0_PG;
     error |= __vmwrite(CR0_READ_SHADOW, shadow_cr);
     /* CR3 is set in vmx_final_setup_guest */
     error |= __vmwrite(GUEST_CR4, host_env->cr4);

diff --git a/xen/arch/x86/x86_32/traps.c b/xen/arch/x86/x86_32/traps.c
index 2070c8a49d295454b42b25639eaa10e1d7f09295..f545efcba4c3a73d469c8ea8cc25a6a028073978 100644 (file)
--- a/xen/arch/x86/x86_32/traps.c
+++ b/xen/arch/x86/x86_32/traps.c
@@ -20,8 +20,9 @@ void show_registers(struct cpu_user_regs *regs)
     unsigned long ss, ds, es, fs, gs, cs;
     unsigned long eip, esp, eflags;
     const char *context;
-
 #ifdef CONFIG_VMX
+    unsigned long cr0, cr3;
+
     if ( VMX_DOMAIN(current) && (regs->eflags == 0) )
     {
         __vmread(GUEST_EIP, &eip);
@@ -33,6 +34,8 @@ void show_registers(struct cpu_user_regs *regs)
         __vmread(GUEST_FS_SELECTOR, &fs);
         __vmread(GUEST_GS_SELECTOR, &gs);
         __vmread(GUEST_CS_SELECTOR, &cs);
+        __vmread(CR0_READ_SHADOW, &cr0);
+        __vmread(GUEST_CR3, &cr3);
         context = "vmx guest";
     }
     else
@@ -77,6 +80,9 @@ void show_registers(struct cpu_user_regs *regs)
     printk("ds: %04lx   es: %04lx   fs: %04lx   gs: %04lx   "
            "ss: %04lx   cs: %04lx\n",
            ds, es, fs, gs, ss, cs);
+#ifdef CONFIG_VMX
+    printk("cr0: %08lx   cr3: %08lx\n", cr0, cr3);
+#endif
 
     if ( GUEST_MODE(regs) )
         show_guest_stack();

diff --git a/xen/include/asm-x86/shadow.h b/xen/include/asm-x86/shadow.h
index 5b9eeaf17f9371131b0521644067fcda563cb4fc..42a9c1e6f1e303db468deb5770dbb26c9c191b2e 100644 (file)
--- a/xen/include/asm-x86/shadow.h
+++ b/xen/include/asm-x86/shadow.h
@@ -31,6 +31,9 @@
 #include <asm/processor.h>
 #include <asm/domain_page.h>
 #include <public/dom0_ops.h>
+#ifdef CONFIG_VMX
+#include <asm/vmx.h>
+#endif
 
 /* Shadow PT operation mode : shadow-mode variable in arch_domain. */
 
@@ -1672,8 +1675,8 @@ static inline void update_pagetables(struct exec_domain *ed)
 
 #ifdef CONFIG_VMX
     if ( VMX_DOMAIN(ed) )
-        paging_enabled =
-            test_bit(VMX_CPU_STATE_PG_ENABLED, &ed->arch.arch_vmx.cpu_state);
+        paging_enabled = vmx_paging_enabled(ed);
+            
     else
 #endif
         // HACK ALERT: there's currently no easy way to figure out if a domU

diff --git a/xen/include/asm-x86/vmx.h b/xen/include/asm-x86/vmx.h
index d9bed6621b4e6eaf0836558d20c0c9bb299660f8..a66ebb5f3f39e1909f8d7c450a3870bce4351ec1 100644 (file)
--- a/xen/include/asm-x86/vmx.h
+++ b/xen/include/asm-x86/vmx.h
@@ -294,5 +294,14 @@ static inline void vmx_stts()
     if (!(cr0 & X86_CR0_TS))
        __vm_set_bit(EXCEPTION_BITMAP, EXCEPTION_BITMAP_NM);
 }
- 
+
+/* Works only for ed == current */
+static inline int vmx_paging_enabled(struct exec_domain *ed)
+{
+    unsigned long cr0;
+
+    __vmread(CR0_READ_SHADOW, &cr0);
+    return (cr0 & X86_CR0_PE) && (cr0 & X86_CR0_PG);
+}
+
 #endif /* __ASM_X86_VMX_H__ */

diff --git a/xen/include/asm-x86/vmx_vmcs.h b/xen/include/asm-x86/vmx_vmcs.h
index 70a9c34594d0893d121572e2ff1d23407a25ff26..7ccfac0b6bf8d2ef0fb3892e8e290069dab0cb50 100644 (file)
--- a/xen/include/asm-x86/vmx_vmcs.h
+++ b/xen/include/asm-x86/vmx_vmcs.h
@@ -29,7 +29,6 @@ extern void stop_vmx(void);
 
 void vmx_enter_scheduler(void);
 
-#define VMX_CPU_STATE_PG_ENABLED        0       
 #define        VMX_CPU_STATE_ASSIST_ENABLED    1
 
 struct vmcs_struct {