x86/realmode: Make stack lock work in trampoline_compat()

The stack locking and stack assignment macro LOAD_REALMODE_ESP fails to
work when invoked from the 64bit trampoline entry point:

trampoline_start64
  trampoline_compat
    LOAD_REALMODE_ESP <- lock

Accessing tr_lock is only possible from 16bit mode. For the compat entry
point this needs to be pa_tr_lock so that the required relocation entry is
generated. Otherwise it locks the non-relocated address which is
aside of being wrong never cleared in secondary_startup_64() causing all
but the first CPU to get stuck on the lock.

Make the macro take an argument lock_pa which defaults to 0 and rename it
to LOCK_AND_LOAD_REALMODE_ESP to make it clear what this is about.

Fixes: f6f1ae9 ("x86/smpboot: Implement a bit spinlock to protect the realmode stack")
Reported-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Link: https://lore.kernel.org/r/87h6rujdvl.ffs@tglx

Author: KAGA-KOKO

arch/x86/realmode/rm/trampoline_64.S

@@ -37,12 +37,16 @@
 	.text
 	.code16
 
-.macro LOAD_REALMODE_ESP
+.macro LOCK_AND_LOAD_REALMODE_ESP lock_pa=0
 	/*
 	 * Make sure only one CPU fiddles with the realmode stack
 	 */
 .Llock_rm\@:
+	.if \lock_pa
+        lock btsl       $0, pa_tr_lock
+	.else
         lock btsl       $0, tr_lock
+	.endif
         jnc             2f
         pause
         jmp             .Llock_rm\@
@@ -63,7 +67,7 @@ SYM_CODE_START(trampoline_start)
 	mov	%ax, %es
 	mov	%ax, %ss
 
-	LOAD_REALMODE_ESP
+	LOCK_AND_LOAD_REALMODE_ESP
 
 	call	verify_cpu		# Verify the cpu supports long mode
 	testl   %eax, %eax		# Check for return code
@@ -106,7 +110,7 @@ SYM_CODE_START(sev_es_trampoline_start)
 	mov	%ax, %es
 	mov	%ax, %ss
 
-	LOAD_REALMODE_ESP
+	LOCK_AND_LOAD_REALMODE_ESP
 
 	jmp	.Lswitch_to_protected
 SYM_CODE_END(sev_es_trampoline_start)
@@ -189,7 +193,7 @@ SYM_CODE_START(pa_trampoline_compat)
 	 * In compatibility mode.  Prep ESP and DX for startup_32, then disable
 	 * paging and complete the switch to legacy 32-bit mode.
 	 */
-	LOAD_REALMODE_ESP
+	LOCK_AND_LOAD_REALMODE_ESP lock_pa=1
 	movw	$__KERNEL_DS, %dx
 
 	movl	$(CR0_STATE & ~X86_CR0_PG), %eax