Boot with virtual == physical to get closer to native Linux.

1) This allows us to get alot closer to booting bzImages.

2) It means we don't have to know page_offset.

3) The Guest needs to modify the boot pagetables to create the
   PAGE_OFFSET mapping before jumping to C code.

4) guest_pa() walks the page tables rather than using page_offset.

5) We don't use page_offset to figure out whether to emulate: it was
   always kinda quesationable, and won't work for instructions done
   before remapping (bzImage unpacking in particular).

6) We still want the kernel address for tlb flushing: have the initial
   hypercall give us that, too.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

6 files changed, 62 insertions, 32 deletions

diff --git a/drivers/lguest/hypercalls.c b/drivers/lguest/hypercalls.c
index 02d0ae26826..13b5f2f813d 100644
--- a/drivers/lguest/hypercalls.c
+++ b/drivers/lguest/hypercalls.c
@@ -181,15 +181,15 @@ static void initialize(struct lguest *lg)
         /* The Guest tells us where we're not to deliver interrupts by putting
          * the range of addresses into "struct lguest_data". */
         if (get_user(lg->noirq_start, &lg->lguest_data->noirq_start)
-            || get_user(lg->noirq_end, &lg->lguest_data->noirq_end)
-            /* We tell the Guest that it can't use the top 4MB of virtual
-             * addresses used by the Switcher. */
-            || put_user(4U*1024*1024, &lg->lguest_data->reserve_mem))
+            || get_user(lg->noirq_end, &lg->lguest_data->noirq_end))
                 kill_guest(lg, "bad guest page %p", lg->lguest_data);
 
         /* We write the current time into the Guest's data page once now. */
         write_timestamp(lg);
 
+        /* page_tables.c will also do some setup. */
+        page_table_guest_data_init(lg);
+
         /* This is the one case where the above accesses might have been the
          * first write to a Guest page.  This may have caused a copy-on-write
          * fault, but the Guest might be referring to the old (read-only)
diff --git a/drivers/lguest/interrupts_and_traps.c b/drivers/lguest/interrupts_and_traps.c
index a57d757eab6..3271c0031a1 100644
--- a/drivers/lguest/interrupts_and_traps.c
+++ b/drivers/lguest/interrupts_and_traps.c
@@ -62,8 +62,9 @@ static void push_guest_stack(struct lguest *lg, unsigned long *gstack, u32 val)
  * it). */
 static void set_guest_interrupt(struct lguest *lg, u32 lo, u32 hi, int has_err)
 {
-        unsigned long gstack;
+        unsigned long gstack, origstack;
         u32 eflags, ss, irq_enable;
+        unsigned long virtstack;
 
         /* There are two cases for interrupts: one where the Guest is already
          * in the kernel, and a more complex one where the Guest is in
@@ -71,8 +72,10 @@ static void set_guest_interrupt(struct lguest *lg, u32 lo, u32 hi, int has_err)
         if ((lg->regs->ss&0x3) != GUEST_PL) {
                 /* The Guest told us their kernel stack with the SET_STACK
                  * hypercall: both the virtual address and the segment */
-                gstack = guest_pa(lg, lg->esp1);
+                virtstack = lg->esp1;
                 ss = lg->ss1;
+
+                origstack = gstack = guest_pa(lg, virtstack);
                 /* We push the old stack segment and pointer onto the new
                  * stack: when the Guest does an "iret" back from the interrupt
                  * handler the CPU will notice they're dropping privilege
@@ -81,8 +84,10 @@ static void set_guest_interrupt(struct lguest *lg, u32 lo, u32 hi, int has_err)
                 push_guest_stack(lg, &gstack, lg->regs->esp);
         } else {
                 /* We're staying on the same Guest (kernel) stack. */
-                gstack = guest_pa(lg, lg->regs->esp);
+                virtstack = lg->regs->esp;
                 ss = lg->regs->ss;
+
+                origstack = gstack = guest_pa(lg, virtstack);
         }
 
         /* Remember that we never let the Guest actually disable interrupts, so
@@ -108,7 +113,7 @@ static void set_guest_interrupt(struct lguest *lg, u32 lo, u32 hi, int has_err)
         /* Now we've pushed all the old state, we change the stack, the code
          * segment and the address to execute. */
         lg->regs->ss = ss;
-        lg->regs->esp = gstack + lg->page_offset;
+        lg->regs->esp = virtstack + (gstack - origstack);
         lg->regs->cs = (__KERNEL_CS|GUEST_PL);
         lg->regs->eip = idt_address(lo, hi);
 
diff --git a/drivers/lguest/lg.h b/drivers/lguest/lg.h
index 7408cebe995..e4845d7f068 100644
--- a/drivers/lguest/lg.h
+++ b/drivers/lguest/lg.h
@@ -63,7 +63,7 @@ struct lguest
         /* This provides the offset to the base of guest-physical
          * memory in the Launcher. */
         void __user *mem_base;
-        u32 page_offset;
+        unsigned long kernel_address;
         u32 cr2;
         int halted;
         int ts;
@@ -165,6 +165,8 @@ void guest_set_pte(struct lguest *lg, unsigned long gpgdir,
 void map_switcher_in_guest(struct lguest *lg, struct lguest_pages *pages);
 int demand_page(struct lguest *info, unsigned long cr2, int errcode);
 void pin_page(struct lguest *lg, unsigned long vaddr);
+unsigned long guest_pa(struct lguest *lg, unsigned long vaddr);
+void page_table_guest_data_init(struct lguest *lg);
 
 /* <arch>/core.c: */
 void lguest_arch_host_init(void);
@@ -229,9 +231,5 @@ do {								\
 } while(0)
 /* (End of aside) :*/
 
-static inline unsigned long guest_pa(struct lguest *lg, unsigned long vaddr)
-{
-        return vaddr - lg->page_offset;
-}
 #endif  /* __ASSEMBLY__ */
 #endif  /* _LGUEST_H */
diff --git a/drivers/lguest/lguest_user.c b/drivers/lguest/lguest_user.c
index b184652e45d..61b177e1e64 100644
--- a/drivers/lguest/lguest_user.c
+++ b/drivers/lguest/lguest_user.c
@@ -111,7 +111,7 @@ static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
         return run_guest(lg, (unsigned long __user *)user);
 }
 
-/*L:020 The initialization write supplies 5 pointer sized (32 or 64 bit)
+/*L:020 The initialization write supplies 4 pointer sized (32 or 64 bit)
  * values (in addition to the LHREQ_INITIALIZE value).  These are:
  *
  * base: The start of the Guest-physical memory inside the Launcher memory.
@@ -124,12 +124,6 @@ static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
  * pagetables (which are set up by the Launcher).
  *
  * start: The first instruction to execute ("eip" in x86-speak).
- *
- * page_offset: The PAGE_OFFSET constant in the Guest kernel.  We should
- * probably wean the code off this, but it's a very useful constant!  Any
- * address above this is within the Guest kernel, and any kernel address can
- * quickly converted from physical to virtual by adding PAGE_OFFSET.  It's
- * 0xC0000000 (3G) by default, but it's configurable at kernel build time.
  */
 static int initialize(struct file *file, const unsigned long __user *input)
 {
@@ -137,7 +131,7 @@ static int initialize(struct file *file, const unsigned long __user *input)
          * Guest. */
         struct lguest *lg;
         int err;
-        unsigned long args[5];
+        unsigned long args[4];
 
         /* We grab the Big Lguest lock, which protects against multiple
          * simultaneous initializations. */
@@ -162,7 +156,6 @@ static int initialize(struct file *file, const unsigned long __user *input)
         /* Populate the easy fields of our "struct lguest" */
         lg->mem_base = (void __user *)(long)args[0];
         lg->pfn_limit = args[1];
-        lg->page_offset = args[4];
 
         /* We need a complete page for the Guest registers: they are accessible
          * to the Guest and we can only grant it access to whole pages. */
diff --git a/drivers/lguest/page_tables.c b/drivers/lguest/page_tables.c
index bfe3650b28d..fe3c7575647 100644
--- a/drivers/lguest/page_tables.c
+++ b/drivers/lguest/page_tables.c
@@ -13,6 +13,7 @@
 #include <linux/random.h>
 #include <linux/percpu.h>
 #include <asm/tlbflush.h>
+#include <asm/uaccess.h>
 #include "lg.h"
 
 /*M:008 We hold reference to pages, which prevents them from being swapped.
@@ -345,7 +346,7 @@ static void flush_user_mappings(struct lguest *lg, int idx)
 {
         unsigned int i;
         /* Release every pgd entry up to the kernel's address. */
-        for (i = 0; i < pgd_index(lg->page_offset); i++)
+        for (i = 0; i < pgd_index(lg->kernel_address); i++)
                 release_pgd(lg, lg->pgdirs[idx].pgdir + i);
 }
 
@@ -358,6 +359,25 @@ void guest_pagetable_flush_user(struct lguest *lg)
 }
 /*:*/
 
+/* We walk down the guest page tables to get a guest-physical address */
+unsigned long guest_pa(struct lguest *lg, unsigned long vaddr)
+{
+        pgd_t gpgd;
+        pte_t gpte;
+
+        /* First step: get the top-level Guest page table entry. */
+        gpgd = __pgd(lgread_u32(lg, gpgd_addr(lg, vaddr)));
+        /* Toplevel not present?  We can't map it in. */
+        if (!(pgd_flags(gpgd) & _PAGE_PRESENT))
+                kill_guest(lg, "Bad address %#lx", vaddr);
+
+        gpte = __pte(lgread_u32(lg, gpte_addr(lg, gpgd, vaddr)));
+        if (!(pte_flags(gpte) & _PAGE_PRESENT))
+                kill_guest(lg, "Bad address %#lx", vaddr);
+
+        return pte_pfn(gpte) * PAGE_SIZE | (vaddr & ~PAGE_MASK);
+}
+
 /* We keep several page tables.  This is a simple routine to find the page
  * table (if any) corresponding to this top-level address the Guest has given
  * us. */
@@ -500,7 +520,7 @@ void guest_set_pte(struct lguest *lg,
 {
         /* Kernel mappings must be changed on all top levels.  Slow, but
          * doesn't happen often. */
-        if (vaddr >= lg->page_offset) {
+        if (vaddr >= lg->kernel_address) {
                 unsigned int i;
                 for (i = 0; i < ARRAY_SIZE(lg->pgdirs); i++)
                         if (lg->pgdirs[i].pgdir)
@@ -550,11 +570,6 @@ void guest_set_pmd(struct lguest *lg, unsigned long gpgdir, u32 idx)
  * its first page table is.  We set some things up here: */
 int init_guest_pagetable(struct lguest *lg, unsigned long pgtable)
 {
-        /* In flush_user_mappings() we loop from 0 to
-         * "pgd_index(lg->page_offset)".  This assumes it won't hit
-         * the Switcher mappings, so check that now. */
-        if (pgd_index(lg->page_offset) >= SWITCHER_PGD_INDEX)
-                return -EINVAL;
         /* We start on the first shadow page table, and give it a blank PGD
          * page. */
         lg->pgdidx = 0;
@@ -565,6 +580,24 @@ int init_guest_pagetable(struct lguest *lg, unsigned long pgtable)
         return 0;
 }
 
+/* When the Guest calls LHCALL_LGUEST_INIT we do more setup. */
+void page_table_guest_data_init(struct lguest *lg)
+{
+        /* We get the kernel address: above this is all kernel memory. */
+        if (get_user(lg->kernel_address, &lg->lguest_data->kernel_address)
+            /* We tell the Guest that it can't use the top 4MB of virtual
+             * addresses used by the Switcher. */
+            || put_user(4U*1024*1024, &lg->lguest_data->reserve_mem)
+            || put_user(lg->pgdirs[lg->pgdidx].gpgdir,&lg->lguest_data->pgdir))
+                kill_guest(lg, "bad guest page %p", lg->lguest_data);
+
+        /* In flush_user_mappings() we loop from 0 to
+         * "pgd_index(lg->kernel_address)".  This assumes it won't hit the
+         * Switcher mappings, so check that now. */
+        if (pgd_index(lg->kernel_address) >= SWITCHER_PGD_INDEX)
+                kill_guest(lg, "bad kernel address %#lx", lg->kernel_address);
+}
+
 /* When a Guest dies, our cleanup is fairly simple. */
 void free_guest_pagetable(struct lguest *lg)
 {
diff --git a/drivers/lguest/x86/core.c b/drivers/lguest/x86/core.c
index a125109446d..39f64c95de1 100644
--- a/drivers/lguest/x86/core.c
+++ b/drivers/lguest/x86/core.c
@@ -216,9 +216,10 @@ static int emulate_insn(struct lguest *lg)
          * guest_pa just subtracts the Guest's page_offset. */
         unsigned long physaddr = guest_pa(lg, lg->regs->eip);
 
-        /* The guest_pa() function only works for Guest kernel addresses, but
-         * that's all we're trying to do anyway. */
-        if (lg->regs->eip < lg->page_offset)
+        /* This must be the Guest kernel trying to do something, not userspace!
+         * The bottom two bits of the CS segment register are the privilege
+         * level. */
+        if ((lg->regs->cs & 3) != GUEST_PL)
                 return 0;
 
         /* Decoding x86 instructions is icky. */