x86/xen: Delay get_cpu_cap until stack canary is established

Commit 2cc42bac ("x86-64/Xen: eliminate W+X mappings") introduced a
call to get_cpu_cap, which is fstack-protected.  This is works on x86-64
as commit 4f277295 ("x86/xen: init %gs very early to avoid page
faults with stack protector") ensures the stack protector is configured,
but it it did not cover x86-32.

Delay calling get_cpu_cap until after xen_setup_gdt has initialized the
stack canary.  Without this, a 32bit PV machine crashes early
in boot.
(XEN) Domain 0 (vcpu#0) crashed on cpu#0:
(XEN) ----[ Xen-4.6.6-xc  x86_64  debug=n  Tainted:    C ]----
(XEN) CPU:    0
(XEN) RIP:    e019:[<00000000c10362f8>]

And the PV kernel IP corresponds to init_scattered_cpuid_features
   0xc10362f8 <+24>:    mov    %gs:0x14,%eax

Fixes 2cc42bac ("x86-64/Xen: eliminate W+X mappings")
Signed-off-by: Jason Andryuk <jandryuk@gmail.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>

arch/x86/xen/enlighten_pv.c

@@ -1259,10 +1259,6 @@ asmlinkage __visible void __init xen_start_kernel(void)
 	 */
 	__userpte_alloc_gfp &= ~__GFP_HIGHMEM;
 
-	/* Work out if we support NX */
-	get_cpu_cap(&boot_cpu_data);
-	x86_configure_nx();
-
 	/* Get mfn list */
 	xen_build_dynamic_phys_to_machine();
 
@@ -1272,6 +1268,10 @@ asmlinkage __visible void __init xen_start_kernel(void)
 	 */
 	xen_setup_gdt(0);
 
+	/* Work out if we support NX */
+	get_cpu_cap(&boot_cpu_data);
+	x86_configure_nx();
+
 	xen_init_irq_ops();
 
 	/* Let's presume PV guests always boot on vCPU with id 0. */