x86/xen: don't do PV iret hypercall through hypercall page

Instead of jumping to the Xen hypercall page for doing the iret
hypercall, directly code the required sequence in xen-asm.S.

This is done in preparation of no longer using hypercall page at all,
as it has shown to cause problems with speculation mitigations.

This is part of XSA-466 / CVE-2024-53241.

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

arch/x86/xen/xen-asm.S

@@ -176,7 +176,6 @@ SYM_CODE_START(xen_early_idt_handler_array)
 SYM_CODE_END(xen_early_idt_handler_array)
 	__FINIT
 
-hypercall_iret = hypercall_page + __HYPERVISOR_iret * 32
 /*
  * Xen64 iret frame:
  *
@@ -186,17 +185,28 @@ hypercall_iret = hypercall_page + __HYPERVISOR_iret * 32
  *	cs
  *	rip		<-- standard iret frame
  *
- *	flags
+ *	flags		<-- xen_iret must push from here on
  *
- *	rcx		}
- *	r11		}<-- pushed by hypercall page
- * rsp->rax		}
+ *	rcx
+ *	r11
+ * rsp->rax
  */
+.macro xen_hypercall_iret
+	pushq $0	/* Flags */
+	push %rcx
+	push %r11
+	push %rax
+	mov  $__HYPERVISOR_iret, %eax
+	syscall		/* Do the IRET. */
+#ifdef CONFIG_MITIGATION_SLS
+	int3
+#endif
+.endm
+
 SYM_CODE_START(xen_iret)
 	UNWIND_HINT_UNDEFINED
 	ANNOTATE_NOENDBR
-	pushq $0
-	jmp hypercall_iret
+	xen_hypercall_iret
 SYM_CODE_END(xen_iret)
 
 /*
@@ -301,8 +311,7 @@ SYM_CODE_START(xen_entry_SYSENTER_compat)
 	ENDBR
 	lea 16(%rsp), %rsp	/* strip %rcx, %r11 */
 	mov $-ENOSYS, %rax
-	pushq $0
-	jmp hypercall_iret
+	xen_hypercall_iret
 SYM_CODE_END(xen_entry_SYSENTER_compat)
 SYM_CODE_END(xen_entry_SYSCALL_compat)