hv: drop the temperory stack for AP startup

By switching AP wakeup from broadcast to one by one, we could
set correct stack for each AP and drop the temp stack used during
AP boot.

Tracked-On: #2034
Signed-off-by: Yin Fengwei <fengwei.yin@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>

hypervisor/arch/x86/boot/trampoline.S

@@ -33,7 +33,7 @@
  * the macros involved are changed.
  */
 
-    .extern     cpu_secondary_init
+    .extern     init_secondary_cpu
 
     .section     .trampoline_reset,"ax"
 
@@ -149,20 +149,7 @@ trampoline_start64:
     mov     %eax, %fs
     mov     %eax, %gs
 
-    /* Obtain CPU spin-lock to serialize trampoline for different APs */
-    movq    trampoline_spinlock_ptr(%rip), %rdi
-    spinlock_obtain(%rdi)
-
-    /* Initialize temporary stack pointer
-       NOTE:  Using the PML4 memory (PDPT address is top of memory
-              for the PML4 page) for the temporary stack
-              as we are only using the very first entry in
-              this page and the stack is growing down from
-              the top of this page.  This stack is only
-              used for a VERY short period of time, so
-              this reuse of PML4 memory should be acceptable. */
-
-    lea     trampoline_pdpt_addr(%rip), %rsp
+    movq    secondary_cpu_stack(%rip), %rsp
 
     /* Jump to C entry */
     movq    main_entry(%rip), %rax
@@ -173,11 +160,11 @@ trampoline_start64:
     .align  8
     .global main_entry
 main_entry:
-    .quad   cpu_secondary_init /* default entry is AP start entry */
+    .quad   init_secondary_cpu /* default entry is AP start entry */
 
-    .global trampoline_spinlock_ptr
-trampoline_spinlock_ptr:
-    .quad   trampoline_spinlock
+    .global secondary_cpu_stack
+secondary_cpu_stack:
+    .quad   0
 
 /* GDT table */
     .align  4