Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
To:     dave.hansen@intel.com, tglx@linutronix.de, mingo@redhat.com,
        bp@alien8.de
Cc:     decui@microsoft.com, rick.p.edgecombe@intel.com,
        sathyanarayanan.kuppuswamy@linux.intel.com, seanjc@google.com,
        thomas.lendacky@amd.com, x86@kernel.org,
        linux-kernel@vger.kernel.org,
        "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>,
        stable@vger.kernel.org
Subject: [PATCH 2/2] x86/tdx: Fix race between set_memory_encrypted() and load_unaligned_zeropad()
Date:   Fri, 26 May 2023 01:58:47 +0300
Message-Id: <20230525225847.28592-3-kirill.shutemov@linux.intel.com>
In-Reply-To: <20230525225847.28592-1-kirill.shutemov@linux.intel.com>
References: <20230525225847.28592-1-kirill.shutemov@linux.intel.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

Touching privately mapped GPA that is not properly converted to private
with MapGPA and accepted leads to unrecoverable exit to VMM.

load_unaligned_zeropad() can touch memory that is not owned by the
caller, but just happened to next after the owned memory.
This load_unaligned_zeropad() behaviour makes it important when kernel
asks VMM to convert a GPA from shared to private or back. Kernel must
never have a page mapped into direct mapping (and aliases) as private
when the GPA is already converted to shared or when GPA is not yet
converted to private.

guest.enc_status_change_prepare() called before adjusting direct mapping
and therefore it is responsible for converting the memory to private.

guest.enc_tlb_flush_required() called after adjusting direct mapping and
it converts the memory to shared.

It is okay to have a shared mapping of memory that is not converted
properly. handle_mmio() knows how to deal with load_unaligned_zeropad()
stepping on it.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Fixes: 7dbde7631629 ("x86/mm/cpa: Add support for TDX shared memory")
Cc: stable@vger.kernel.org
---
 arch/x86/coco/tdx/tdx.c | 56 ++++++++++++++++++++++++++++++++++++++---
 1 file changed, 53 insertions(+), 3 deletions(-)

diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c
index e146b599260f..84525df750d4 100644
--- a/arch/x86/coco/tdx/tdx.c
+++ b/arch/x86/coco/tdx/tdx.c
@@ -840,6 +840,30 @@ static bool tdx_enc_status_changed(unsigned long vaddr, int numpages, bool enc)
 	return true;
 }
 
+static bool tdx_enc_status_change_prepare(unsigned long vaddr, int numpages,
+					  bool enc)
+{
+	/*
+	 * Only handle shared->private conversion here.
+	 * See the comment in tdx_early_init().
+	 */
+	if (enc)
+		return tdx_enc_status_changed(vaddr, numpages, enc);
+	return true;
+}
+
+static bool tdx_enc_status_change_finish(unsigned long vaddr, int numpages,
+					 bool enc)
+{
+	/*
+	 * Only handle private->shared conversion here.
+	 * See the comment in tdx_early_init().
+	 */
+	if (!enc)
+		return tdx_enc_status_changed(vaddr, numpages, enc);
+	return true;
+}
+
 void __init tdx_early_init(void)
 {
 	u64 cc_mask;
@@ -867,9 +891,35 @@ void __init tdx_early_init(void)
 	 */
 	physical_mask &= cc_mask - 1;
 
-	x86_platform.guest.enc_cache_flush_required = tdx_cache_flush_required;
-	x86_platform.guest.enc_tlb_flush_required   = tdx_tlb_flush_required;
-	x86_platform.guest.enc_status_change_finish = tdx_enc_status_changed;
+	/*
+	 * Touching privately mapped GPA that is not properly converted to
+	 * private with MapGPA and accepted leads to unrecoverable exit
+	 * to VMM.
+	 *
+	 * load_unaligned_zeropad() can touch memory that is not owned by
+	 * the caller, but just happened to next after the owned memory.
+	 * This load_unaligned_zeropad() behaviour makes it important when
+	 * kernel asks VMM to convert a GPA from shared to private or back.
+	 * Kernel must never have a page mapped into direct mapping (and
+	 * aliases) as private when the GPA is already converted to shared or
+	 * when GPA is not yet converted to private.
+	 *
+	 * guest.enc_status_change_prepare() called before adjusting direct
+	 * mapping and therefore it is responsible for converting the memory
+	 * to private.
+	 *
+	 * guest.enc_tlb_flush_required() called after adjusting direct mapping
+	 * and it converts the memory to shared.
+	 *
+	 * It is okay to have a shared mapping of memory that is not converted
+	 * properly. handle_mmio() knows how to deal with load_unaligned_zeropad()
+	 * stepping on it.
+	 */
+	x86_platform.guest.enc_status_change_prepare = tdx_enc_status_change_prepare;
+	x86_platform.guest.enc_status_change_finish  = tdx_enc_status_change_finish;
+
+	x86_platform.guest.enc_cache_flush_required  = tdx_cache_flush_required;
+	x86_platform.guest.enc_tlb_flush_required    = tdx_tlb_flush_required;
 
 	pr_info("Guest detected\n");
 }
-- 
2.39.3