Merged.

8 files changed, 84 insertions, 76 deletions

diff --git a/tools/examples/vif-nat b/tools/examples/vif-nat
index b2bfade528..d667fb89da 100644
--- a/tools/examples/vif-nat
+++ b/tools/examples/vif-nat
@@ -91,8 +91,6 @@ vif_int=$(( $(echo "((($vif_ip" | sed -e 's#\.#)\*256\+#g') ))
 netmask=$(dotted_quad $intmask)
 network=$(dotted_quad $(( $vif_int & $intmask )) )
 
-main_ip=$(dom0_ip)
-
 
 dhcp_remove_entry()
 {
@@ -140,7 +138,7 @@ case "$command" in
 
         do_or_die ip link set "$vif" up arp on
         do_or_die ip addr add "$router_ip" dev "$vif"
-        do_or_die ip route add "$vif_ip" dev "$vif" src "$main_ip"
+        do_or_die ip route add "$vif_ip" dev "$vif" src "$router_ip"
         echo 1 >/proc/sys/net/ipv4/conf/${vif}/proxy_arp
         [ "$dhcp" != 'no' ] && dhcp_up
         ;;
diff --git a/xen/arch/x86/Makefile b/xen/arch/x86/Makefile
index ecd0b64e68..952327cc16 100644
--- a/xen/arch/x86/Makefile
+++ b/xen/arch/x86/Makefile
@@ -37,7 +37,8 @@ endif
 default: $(TARGET)
 
 $(TARGET): $(TARGET)-syms boot/mkelf32
-	./boot/mkelf32 $(TARGET)-syms $(TARGET) 0x100000
+	./boot/mkelf32 $(TARGET)-syms $(TARGET) 0x100000 \
+	`nm $(TARGET)-syms | sort | tail -n 1 | sed -e 's/^\([^ ]*\).*/0x\1/'`
 
 $(CURDIR)/arch.o: $(OBJS)
 	$(LD) $(LDFLAGS) -r -o $@ $(OBJS)
diff --git a/xen/arch/x86/boot/mkelf32.c b/xen/arch/x86/boot/mkelf32.c
index 8fb665e068..e93dfdcf96 100644
--- a/xen/arch/x86/boot/mkelf32.c
+++ b/xen/arch/x86/boot/mkelf32.c
@@ -222,6 +222,7 @@ static void do_read(int fd, void *data, int len)
 
 int main(int argc, char **argv)
 {
+    u64        final_exec_addr;
     u32        loadbase, dat_siz, mem_siz;
     char      *inimage, *outimage;
     int        infd, outfd;
@@ -234,15 +235,17 @@ int main(int argc, char **argv)
     Elf64_Ehdr in64_ehdr;
     Elf64_Phdr in64_phdr;
 
-    if ( argc != 4 )
+    if ( argc != 5 )
     {
-        fprintf(stderr, "Usage: mkelf32 <in-image> <out-image> <load-base>\n");
+        fprintf(stderr, "Usage: mkelf32 <in-image> <out-image> "
+                "<load-base> <final-exec-addr>\n");
         return 1;
     }
 
     inimage  = argv[1];
     outimage = argv[2];
     loadbase = strtoul(argv[3], NULL, 16);
+    final_exec_addr = strtoul(argv[4], NULL, 16);
 
     infd = open(inimage, O_RDONLY);
     if ( infd == -1 )
@@ -286,7 +289,10 @@ int main(int argc, char **argv)
 
         (void)lseek(infd, in32_phdr.p_offset, SEEK_SET);
         dat_siz = (u32)in32_phdr.p_filesz;
-        mem_siz = (u32)in32_phdr.p_memsz;
+
+        /* Do not use p_memsz: it does not include BSS alignment padding. */
+        /*mem_siz = (u32)in32_phdr.p_memsz;*/
+        mem_siz = (u32)(final_exec_addr - in32_phdr.p_vaddr);
         break;
 
     case ELFCLASS64:
@@ -314,7 +320,10 @@ int main(int argc, char **argv)
 
         (void)lseek(infd, in64_phdr.p_offset, SEEK_SET);
         dat_siz = (u32)in64_phdr.p_filesz;
-        mem_siz = (u32)in64_phdr.p_memsz;
+
+        /* Do not use p_memsz: it does not include BSS alignment padding. */
+        /*mem_siz = (u32)in64_phdr.p_memsz;*/
+        mem_siz = (u32)(final_exec_addr - in64_phdr.p_vaddr);
         break;
 
     default:
diff --git a/xen/arch/x86/boot/x86_32.S b/xen/arch/x86/boot/x86_32.S
index ca0c43555a..59e68ca7ef 100644
--- a/xen/arch/x86/boot/x86_32.S
+++ b/xen/arch/x86/boot/x86_32.S
@@ -74,10 +74,6 @@ __start:
         cmp     $0x2BADB002,%eax
         jne     not_multiboot
 
-        /* Save the Multiboot info structure for later use. */
-	add     $__PAGE_OFFSET,%ebx
-        push    %ebx
-
         /* Initialize BSS (no nasty surprises!) */
         mov     $__bss_start-__PAGE_OFFSET,%edi
         mov     $_end-__PAGE_OFFSET,%ecx
@@ -85,6 +81,10 @@ __start:
         xor     %eax,%eax
         rep     stosb
 
+        /* Save the Multiboot info structure for later use. */
+        add     $__PAGE_OFFSET,%ebx
+        push    %ebx
+
 #ifdef CONFIG_X86_PAE
         /* Initialize low and high mappings of all memory with 2MB pages */
         mov     $idle_pg_table_l2-__PAGE_OFFSET,%edi
@@ -238,27 +238,28 @@ ENTRY(gdt_table)
         .fill 2*NR_CPUS,8,0          /* space for TSS and LDT per CPU    */
 
         .org 0x2000
-/* Maximum STACK_ORDER for x86/32 is 1. We must therefore ensure that the */
-/* CPU0 stack is aligned on an even page boundary!                        */
-ENTRY(cpu0_stack)
-        .org 0x2000 + STACK_SIZE
 
 #ifdef CONFIG_X86_PAE
-
 ENTRY(idle_pg_table)
 ENTRY(idle_pg_table_l3)
-        .quad 0x100000 + 0x2000 + STACK_SIZE + 1*PAGE_SIZE + 0x01
-        .quad 0x100000 + 0x2000 + STACK_SIZE + 2*PAGE_SIZE + 0x01
-        .quad 0x100000 + 0x2000 + STACK_SIZE + 3*PAGE_SIZE + 0x01
-        .quad 0x100000 + 0x2000 + STACK_SIZE + 4*PAGE_SIZE + 0x01
-        .org 0x2000 + STACK_SIZE + 1*PAGE_SIZE
+        .long idle_pg_table_l2 + 0*PAGE_SIZE + 0x01 - __PAGE_OFFSET, 0
+        .long idle_pg_table_l2 + 1*PAGE_SIZE + 0x01 - __PAGE_OFFSET, 0
+        .long idle_pg_table_l2 + 2*PAGE_SIZE + 0x01 - __PAGE_OFFSET, 0
+        .long idle_pg_table_l2 + 3*PAGE_SIZE + 0x01 - __PAGE_OFFSET, 0
+.section ".bss.page_aligned","w"
 ENTRY(idle_pg_table_l2)
-        .org 0x2000 + STACK_SIZE + 5*PAGE_SIZE
-
-#else /* CONFIG_X86_PAE */
-
+        .fill 4*PAGE_SIZE,1,0
+#else
+.section ".bss.page_aligned","w"
 ENTRY(idle_pg_table)
-ENTRY(idle_pg_table_l2) # Initial page directory is 4kB
-        .org 0x2000 + STACK_SIZE + PAGE_SIZE
+ENTRY(idle_pg_table_l2)
+        .fill 1*PAGE_SIZE,1,0
+#endif
 
-#endif /* CONFIG_X86_PAE */
+#if (STACK_ORDER == 0)
+.section ".bss.page_aligned","w"
+#else
+.section ".bss.twopage_aligned","w"
+#endif
+ENTRY(cpu0_stack)
+        .fill STACK_SIZE,1,0
diff --git a/xen/arch/x86/boot/x86_64.S b/xen/arch/x86/boot/x86_64.S
index fbe5311124..126850a0f8 100644
--- a/xen/arch/x86/boot/x86_64.S
+++ b/xen/arch/x86/boot/x86_64.S
@@ -249,13 +249,8 @@ ENTRY(idle_pg_table_4)
 ENTRY(idle_pg_table_l3)
         .quad idle_pg_table_l2 - __PAGE_OFFSET + 7
 
-        .org 0x4000
-/* Maximum STACK_ORDER for x86/64 is 2. We must therefore ensure that the */
-/* CPU0 stack is aligned on a 4-page boundary.                            */
-ENTRY(cpu0_stack)
-
 /* Initial PDE -- level-2 page table. Maps first 64MB physical memory. */
-        .org 0x4000 + STACK_SIZE
+        .org 0x4000
 ENTRY(idle_pg_table_l2)
         .macro identmap from=0, count=32
         .if \count-1
@@ -265,7 +260,15 @@ ENTRY(idle_pg_table_l2)
         .quad 0x00000000000001e3 + \from
         .endif
         .endm
-        identmap /* Too orangey for crows :-) */
+        identmap
 
-        .org 0x4000 + STACK_SIZE + PAGE_SIZE
+        .org 0x4000 + PAGE_SIZE
         .code64
+
+#if (STACK_ORDER == 0)
+.section ".bss.page_aligned","w"
+#else
+.section ".bss.twopage_aligned","w"
+#endif
+ENTRY(cpu0_stack)
+        .fill STACK_SIZE,1,0
diff --git a/xen/arch/x86/mm.c b/xen/arch/x86/mm.c
index 31ec92fe42..b12eac5b91 100644
--- a/xen/arch/x86/mm.c
+++ b/xen/arch/x86/mm.c
@@ -128,8 +128,9 @@ static int mod_l1_entry(l1_pgentry_t *, l1_pgentry_t);
 
 /* Used to defer flushing of memory structures. */
 static struct {
-#define DOP_FLUSH_TLB   (1<<0) /* Flush the TLB.                 */
-#define DOP_RELOAD_LDT  (1<<1) /* Reload the LDT shadow mapping. */
+#define DOP_FLUSH_TLB      (1<<0) /* Flush the local TLB.                    */
+#define DOP_FLUSH_ALL_TLBS (1<<1) /* Flush TLBs of all VCPUs of current dom. */
+#define DOP_RELOAD_LDT     (1<<2) /* Reload the LDT shadow mapping.          */
     unsigned int   deferred_ops;
     /* If non-NULL, specifies a foreign subject domain for some operations. */
     struct domain *foreign;
@@ -1323,14 +1324,28 @@ void free_page_type(struct pfn_info *page, unsigned long type)
     struct domain *owner = page_get_owner(page);
     unsigned long gpfn;
 
-    if ( unlikely((owner != NULL) && shadow_mode_enabled(owner)) )
+    if ( likely(owner != NULL) )
     {
-        mark_dirty(owner, page_to_pfn(page));
-        if ( unlikely(shadow_mode_refcounts(owner)) )
-            return;
-        gpfn = __mfn_to_gpfn(owner, page_to_pfn(page));
-        ASSERT(VALID_M2P(gpfn));
-        remove_shadow(owner, gpfn, type & PGT_type_mask);
+        /*
+         * We have to flush before the next use of the linear mapping
+         * (e.g., update_va_mapping()) or we could end up modifying a page
+         * that is no longer a page table (and hence screw up ref counts).
+         */
+        percpu_info[smp_processor_id()].deferred_ops |= DOP_FLUSH_ALL_TLBS;
+
+        if ( unlikely(shadow_mode_enabled(owner)) )
+        {
+            /* Raw page tables are rewritten during save/restore. */
+            if ( !shadow_mode_translate(owner) )
+                mark_dirty(owner, page_to_pfn(page));
+
+            if ( shadow_mode_refcounts(owner) )
+                return;
+
+            gpfn = __mfn_to_gpfn(owner, page_to_pfn(page));
+            ASSERT(VALID_M2P(gpfn));
+            remove_shadow(owner, gpfn, type & PGT_type_mask);
+        }
     }
 
     switch ( type & PGT_type_mask )
@@ -1600,11 +1615,14 @@ static void process_deferred_ops(unsigned int cpu)
     deferred_ops = percpu_info[cpu].deferred_ops;
     percpu_info[cpu].deferred_ops = 0;
 
-    if ( deferred_ops & DOP_FLUSH_TLB )
+    if ( deferred_ops & (DOP_FLUSH_ALL_TLBS|DOP_FLUSH_TLB) )
     {
         if ( shadow_mode_enabled(d) )
             shadow_sync_all(d);
-        local_flush_tlb();
+        if ( deferred_ops & DOP_FLUSH_ALL_TLBS )
+            flush_tlb_mask(d->cpumask);
+        else
+            local_flush_tlb();
     }
         
     if ( deferred_ops & DOP_RELOAD_LDT )
diff --git a/xen/arch/x86/x86_32/xen.lds b/xen/arch/x86/x86_32/xen.lds
index fec62dbf24..f3c168f662 100644
--- a/xen/arch/x86/x86_32/xen.lds
+++ b/xen/arch/x86/x86_32/xen.lds
@@ -23,7 +23,6 @@ SECTIONS
   _etext = .;			/* End of text section */
 
   .rodata : { *(.rodata) *(.rodata.*) } :text
-  .kstrtab : { *(.kstrtab) } :text
 
   . = ALIGN(32);		/* Exception table */
   __start___ex_table = .;
@@ -35,24 +34,11 @@ SECTIONS
   __pre_ex_table : { *(__pre_ex_table) } :text
   __stop___pre_ex_table = .;
 
-  __start___ksymtab = .;	/* Kernel symbol table */
-  __ksymtab : { *(__ksymtab) } :text
-  __stop___ksymtab = .;
-
-  __start___kallsyms = .;	/* All kernel symbols */
-  __kallsyms : { *(__kallsyms) } :text
-  __stop___kallsyms = .;
-
   .data : {			/* Data */
 	*(.data)
 	CONSTRUCTORS
 	} :text
 
-  _edata = .;			/* End of data section */
-
-  . = ALIGN(8192);		/* init_task */
-  .data.init_task : { *(.data.init_task) } :text
-
   . = ALIGN(4096);		/* Init code and data */
   __init_begin = .;
   .text.init : { *(.text.init) } :text
@@ -64,10 +50,13 @@ SECTIONS
   __initcall_start = .;
   .initcall.init : { *(.initcall.init) } :text
   __initcall_end = .;
+  . = ALIGN(8192);
   __init_end = .;
 
   __bss_start = .;		/* BSS */
   .bss : {
+	*(.bss.twopage_aligned)
+	*(.bss.page_aligned)
 	*(.bss)
 	} :text
   _end = . ;
diff --git a/xen/arch/x86/x86_64/xen.lds b/xen/arch/x86/x86_64/xen.lds
index 30a2b0ca67..837d335f0d 100644
--- a/xen/arch/x86/x86_64/xen.lds
+++ b/xen/arch/x86/x86_64/xen.lds
@@ -21,7 +21,6 @@ SECTIONS
   _etext = .;			/* End of text section */
 
   .rodata : { *(.rodata) *(.rodata.*) } :text
-  .kstrtab : { *(.kstrtab) } :text
 
   . = ALIGN(32);		/* Exception table */
   __start___ex_table = .;
@@ -33,24 +32,11 @@ SECTIONS
   __pre_ex_table : { *(__pre_ex_table) } :text
   __stop___pre_ex_table = .;
 
-  __start___ksymtab = .;	/* Kernel symbol table */
-  __ksymtab : { *(__ksymtab) } :text
-  __stop___ksymtab = .;
-
-  __start___kallsyms = .;	/* All kernel symbols */
-  __kallsyms : { *(__kallsyms) } :text
-  __stop___kallsyms = .;
-
   .data : {			/* Data */
 	*(.data)
 	CONSTRUCTORS
 	} :text
 
-  _edata = .;			/* End of data section */
-
-  . = ALIGN(8192);		/* init_task */
-  .data.init_task : { *(.data.init_task) } :text
-
   . = ALIGN(4096);		/* Init code and data */
   __init_begin = .;
   .text.init : { *(.text.init) } :text
@@ -62,10 +48,13 @@ SECTIONS
   __initcall_start = .;
   .initcall.init : { *(.initcall.init) } :text
   __initcall_end = .;
+  . = ALIGN(8192);
   __init_end = .;
 
   __bss_start = .;		/* BSS */
   .bss : {
+	*(.bss.twopage_aligned)
+	*(.bss.page_aligned)
 	*(.bss)
 	} :text
   _end = . ;