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>

3 files changed, 39 insertions, 10 deletions

diff --git a/arch/x86/kernel/asm-offsets_32.c b/arch/x86/kernel/asm-offsets_32.c
index f8764716b0c0..0e45981b2dd7 100644
--- a/arch/x86/kernel/asm-offsets_32.c
+++ b/arch/x86/kernel/asm-offsets_32.c
@@ -136,6 +136,7 @@ void foo(void)
 #ifdef CONFIG_LGUEST_GUEST
         BLANK();
         OFFSET(LGUEST_DATA_irq_enabled, lguest_data, irq_enabled);
+        OFFSET(LGUEST_DATA_pgdir, lguest_data, pgdir);
         OFFSET(LGUEST_PAGES_host_gdt_desc, lguest_pages, state.host_gdt_desc);
         OFFSET(LGUEST_PAGES_host_idt_desc, lguest_pages, state.host_idt_desc);
         OFFSET(LGUEST_PAGES_host_cr3, lguest_pages, state.host_cr3);
diff --git a/arch/x86/lguest/boot.c b/arch/x86/lguest/boot.c
index 3a06b51c98ad..090f30cbf24c 100644
--- a/arch/x86/lguest/boot.c
+++ b/arch/x86/lguest/boot.c
@@ -86,6 +86,7 @@ struct lguest_data lguest_data = {
         .hcall_status = { [0 ... LHCALL_RING_SIZE-1] = 0xFF },
         .noirq_start = (u32)lguest_noirq_start,
         .noirq_end = (u32)lguest_noirq_end,
+        .kernel_address = PAGE_OFFSET,
         .blocked_interrupts = { 1 }, /* Block timer interrupts */
         .syscall_vec = SYSCALL_VECTOR,
 };
@@ -1033,11 +1034,7 @@ __init void lguest_init(void *boot)
 
         /*G:070 Now we've seen all the paravirt_ops, we return to
          * lguest_init() where the rest of the fairly chaotic boot setup
-         * occurs.
-         *
-         * The Host expects our first hypercall to tell it where our "struct
-         * lguest_data" is, so we do that first. */
-        hcall(LHCALL_LGUEST_INIT, __pa(&lguest_data), 0, 0);
+         * occurs. */
 
         /* The native boot code sets up initial page tables immediately after
          * the kernel itself, and sets init_pg_tables_end so they're not
diff --git a/arch/x86/lguest/i386_head.S b/arch/x86/lguest/i386_head.S
index 6d7a74f07c41..ba4282eba5bf 100644
--- a/arch/x86/lguest/i386_head.S
+++ b/arch/x86/lguest/i386_head.S
@@ -1,5 +1,6 @@
 #include <linux/linkage.h>
 #include <linux/lguest.h>
+#include <asm/lguest_hcall.h>
 #include <asm/asm-offsets.h>
 #include <asm/thread_info.h>
 #include <asm/processor-flags.h>
@@ -8,18 +9,48 @@
  * looks for.  The plan is that the Linux boot protocol will be extended with a
  * "platform type" field which will guide us here from the normal entry point,
  * but for the moment this suffices.  The normal boot code uses %esi for the
- * boot header, so we do too.  We convert it to a virtual address by adding
- * PAGE_OFFSET, and hand it to lguest_init() as its argument (ie. %eax).
+ * boot header, so we do too.
+ *
+ * WARNING: be very careful here!  We're running at addresses equal to physical
+ * addesses (around 0), not above PAGE_OFFSET as most code expectes
+ * (eg. 0xC0000000).  Jumps are relative, so they're OK, but we can't touch any
+ * data.
  *
  * The .section line puts this code in .init.text so it will be discarded after
  * boot. */
 .section .init.text, "ax", @progbits
 .ascii "GenuineLguest"
-        /* Set up initial stack. */
-        movl $(init_thread_union+THREAD_SIZE),%esp
+        /* Make initial hypercall now, so we can set up the pagetables. */
+        movl $LHCALL_LGUEST_INIT, %eax
+        movl $lguest_data - __PAGE_OFFSET, %edx
+        int $LGUEST_TRAP_ENTRY
+
+        /* Set up boot information pointer to hand to lguest_init(): it wants
+         * a virtual address. */
         movl %esi, %eax
         addl $__PAGE_OFFSET, %eax
-        jmp lguest_init
+
+        /* The Host put the toplevel pagetable in lguest_data.pgdir.  The movsl
+         * instruction uses %esi, so we needed to save it above. */
+        movl lguest_data - __PAGE_OFFSET + LGUEST_DATA_pgdir, %esi
+
+        /* Copy first 32 entries of page directory to __PAGE_OFFSET entries.
+         * This means the first 128M of kernel memory will be mapped at
+         * PAGE_OFFSET where the kernel expects to run.  This will get it far
+         * enough through boot to switch to its own pagetables. */
+        movl $32, %ecx
+        movl %esi, %edi
+        addl $((__PAGE_OFFSET >> 22) * 4), %edi
+        rep
+        movsl
+
+        /* Set up the initial stack so we can run C code. */
+        movl $(init_thread_union+THREAD_SIZE),%esp
+
+
+        /* Jumps are relative, and we're running __PAGE_OFFSET too low at the
+         * moment. */
+        jmp lguest_init+__PAGE_OFFSET
 
 /*G:055 We create a macro which puts the assembler code between lgstart_ and
  * lgend_ markers.  These templates are put in the .text section: they can't be