Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Jan Beulich <jbeulich@suse.com>
Subject: [PATCH] x86/mm: keep __default_kernel_pte_mask in sync with
 __supported_pte_mask
To:     Dave Hansen <dave.hansen@linux.intel.com>,
        Andy Lutomirski <luto@kernel.org>,
        Peter Zijlstra <peterz@infradead.org>
Cc:     lkml <linux-kernel@vger.kernel.org>,
        Juergen Gross <jgross@suse.com>,
        Boris Ostrovsky <boris.ostrovsky@oracle.com>
Message-ID: <321effdb-06d5-6b63-e94c-64f83f771024@suse.com>
Date:   Wed, 27 May 2020 17:53:39 +0200
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:68.0) Gecko/20100101
 Thunderbird/68.8.0
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Both masks get applied in the process of e.g. set_fixmap() - the former
through use of PAGE_KERNEL, the latter by use of massage_pgprot(). Hence
forever since the introduction of the former there was a time window
(between x86_configure_nx() and the syncing of the two in
probe_page_size_mask(), as called from init_mem_mapping()) where fixmap
mappings would get established without NX set. For a 32-bit kernel
running in PV mode under Xen this meant a W+X mapping (and associated
warning) for its shared info page mapping established in
xen_pv_init_platform().

Signed-off-by: Jan Beulich <jbeulich@suse.com>

--- a/arch/x86/mm/setup_nx.c
+++ b/arch/x86/mm/setup_nx.c
@@ -33,10 +33,13 @@ early_param("noexec", noexec_setup);
 
 void x86_configure_nx(void)
 {
-	if (boot_cpu_has(X86_FEATURE_NX) && !disable_nx)
+	if (boot_cpu_has(X86_FEATURE_NX) && !disable_nx) {
 		__supported_pte_mask |= _PAGE_NX;
-	else
+		__default_kernel_pte_mask |= _PAGE_NX;
+	} else {
 		__supported_pte_mask &= ~_PAGE_NX;
+		__default_kernel_pte_mask &= ~_PAGE_NX;
+	}
 }
 
 void __init x86_report_nx(void)