x86/boot: copy/clear sections more efficiently

Both the trampoline copy and BSS initialise can be performed more
efficiently by using 4-byte variants of the string operations.

On Intel systems with ERMSB (efficient rep movsb), this is no practical
difference.  On all other systems, this is 4 times more efficient.

Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>

xen/arch/x86/boot/head.S

@@ -128,7 +128,8 @@ __start:
         mov     $sym_phys(__bss_end),%ecx
         sub     %edi,%ecx
         xor     %eax,%eax
-        rep     stosb
+        shr     $2,%ecx
+        rep stosl
 
         /* Interrogate CPU extended features via CPUID. */
         mov     $0x80000000,%eax
@@ -192,8 +193,8 @@ __start:
 
         /* Copy bootstrap trampoline to low memory, below 1MB. */
         mov     $sym_phys(trampoline_start),%esi
-        mov     $trampoline_end - trampoline_start,%ecx
-        rep     movsb
+        mov     $((trampoline_end - trampoline_start) / 4),%ecx
+        rep movsl
 
         /* Jump into the relocated trampoline. */
         lret
@@ -205,6 +206,6 @@ reloc:
 
 ENTRY(trampoline_start)
 #include "trampoline.S"
-GLOBAL(trampoline_end)
+ENTRY(trampoline_end)
 
 #include "x86_64.S"

xen/arch/x86/xen.lds.S

@@ -314,3 +314,8 @@ ASSERT(IS_ALIGNED(cpu0_stack, STACK_SIZE), "cpu0_stack misaligned")
 
 ASSERT(IS_ALIGNED(__init_begin, PAGE_SIZE), "__init_begin misaligned")
 ASSERT(IS_ALIGNED(__init_end,   PAGE_SIZE), "__init_end misaligned")
+
+ASSERT(IS_ALIGNED(trampoline_start, 4), "trampoline_start misaligned")
+ASSERT(IS_ALIGNED(trampoline_end,   4), "trampoline_end misaligned")
+ASSERT(IS_ALIGNED(__bss_start,      4), "__bss_start misaligned")
+ASSERT(IS_ALIGNED(__bss_end,        4), "__bss_end misaligned")