[PATCH v4 3/3] x86/hyperv: Make encrypted/decrypted changes safe for load_unaligned_zeropad()

From: mhkelley58@gmail.com
To: tglx@linutronix.de, mingo@redhat.com, bp@alien8.de,
	dave.hansen@linux.intel.com, x86@kernel.org, hpa@zytor.com,
	kirill.shutemov@linux.intel.com, haiyangz@microsoft.com,
	wei.liu@kernel.org, decui@microsoft.com, luto@kernel.org,
	peterz@infradead.org, akpm@linux-foundation.org,
	urezki@gmail.com, hch@infradead.org, lstoakes@gmail.com,
	thomas.lendacky@amd.com, ardb@kernel.org, jroedel@suse.de,
	seanjc@google.com, rick.p.edgecombe@intel.com,
	sathyanarayanan.kuppuswamy@linux.intel.com,
	linux-kernel@vger.kernel.org, linux-coco@lists.linux.dev,
	linux-hyperv@vger.kernel.org, linux-mm@kvack.org
Subject: [PATCH v4 3/3] x86/hyperv: Make encrypted/decrypted changes safe for load_unaligned_zeropad()
Date: Mon, 15 Jan 2024 18:20:08 -0800	[thread overview]
Message-ID: <20240116022008.1023398-4-mhklinux@outlook.com> (raw)
In-Reply-To: <20240116022008.1023398-1-mhklinux@outlook.com>

From: Michael Kelley <mhklinux@outlook.com>

In a CoCo VM, when transitioning memory from encrypted to decrypted, or
vice versa, the caller of set_memory_encrypted() or set_memory_decrypted()
is responsible for ensuring the memory isn't in use and isn't referenced
while the transition is in progress.  The transition has multiple steps,
and the memory is in an inconsistent state until all steps are complete.
A reference while the state is inconsistent could result in an exception
that can't be cleanly fixed up.

However, the kernel load_unaligned_zeropad() mechanism could cause a stray
reference that can't be prevented by the caller of set_memory_encrypted()
or set_memory_decrypted(), so there's specific code to handle this case.
But a CoCo VM running on Hyper-V may be configured to run with a paravisor,
with the #VC or #VE exception routed to the paravisor. There's no
architectural way to forward the exceptions back to the guest kernel, and
in such a case, the load_unaligned_zeropad() specific code doesn't work.

To avoid this problem, mark pages as "not present" while a transition
is in progress. If load_unaligned_zeropad() causes a stray reference, a
normal page fault is generated instead of #VC or #VE, and the
page-fault-based fixup handlers for load_unaligned_zeropad() resolve the
reference. When the encrypted/decrypted transition is complete, mark the
pages as "present" again.

Signed-off-by: Michael Kelley <mhklinux@outlook.com>
Reviewed-by: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@linux.intel.com>
---
 arch/x86/hyperv/ivm.c | 53 +++++++++++++++++++++++++++++++++++++++----
 1 file changed, 49 insertions(+), 4 deletions(-)

diff --git a/arch/x86/hyperv/ivm.c b/arch/x86/hyperv/ivm.c
index 851107c77f4d..95036feb95e7 100644
--- a/arch/x86/hyperv/ivm.c
+++ b/arch/x86/hyperv/ivm.c
@@ -15,6 +15,7 @@
 #include <asm/io.h>
 #include <asm/coco.h>
 #include <asm/mem_encrypt.h>
+#include <asm/set_memory.h>
 #include <asm/mshyperv.h>
 #include <asm/hypervisor.h>
 #include <asm/mtrr.h>
@@ -502,6 +503,31 @@ static int hv_mark_gpa_visibility(u16 count, const u64 pfn[],
 		return -EFAULT;
 }
 
+/*
+ * When transitioning memory between encrypted and decrypted, the caller
+ * of set_memory_encrypted() or set_memory_decrypted() is responsible for
+ * ensuring that the memory isn't in use and isn't referenced while the
+ * transition is in progress.  The transition has multiple steps, and the
+ * memory is in an inconsistent state until all steps are complete. A
+ * reference while the state is inconsistent could result in an exception
+ * that can't be cleanly fixed up.
+ *
+ * But the Linux kernel load_unaligned_zeropad() mechanism could cause a
+ * stray reference that can't be prevented by the caller, so Linux has
+ * specific code to handle this case. But when the #VC and #VE exceptions
+ * routed to a paravisor, the specific code doesn't work. To avoid this
+ * problem, mark the pages as "not present" while the transition is in
+ * progress. If load_unaligned_zeropad() causes a stray reference, a normal
+ * page fault is generated instead of #VC or #VE, and the page-fault-based
+ * handlers for load_unaligned_zeropad() resolve the reference.  When the
+ * transition is complete, hv_vtom_set_host_visibility() marks the pages
+ * as "present" again.
+ */
+static bool hv_vtom_clear_present(unsigned long kbuffer, int pagecount, bool enc)
+{
+	return !set_memory_np(kbuffer, pagecount);
+}
+
 /*
  * hv_vtom_set_host_visibility - Set specified memory visible to host.
  *
@@ -522,8 +548,10 @@ static bool hv_vtom_set_host_visibility(unsigned long kbuffer, int pagecount, bo
 	int i, pfn;
 
 	pfn_array = kmalloc(HV_HYP_PAGE_SIZE, GFP_KERNEL);
-	if (!pfn_array)
-		return false;
+	if (!pfn_array) {
+		result = false;
+		goto err_set_memory_p;
+	}
 
 	for (i = 0, pfn = 0; i < pagecount; i++) {
 		/*
@@ -548,14 +576,30 @@ static bool hv_vtom_set_host_visibility(unsigned long kbuffer, int pagecount, bo
 		}
 	}
 
- err_free_pfn_array:
+err_free_pfn_array:
 	kfree(pfn_array);
+
+err_set_memory_p:
+	/*
+	 * Set the PTE PRESENT bits again to revert what hv_vtom_clear_present()
+	 * did. Do this even if there is an error earlier in this function in
+	 * order to avoid leaving the memory range in a "broken" state. Setting
+	 * the PRESENT bits shouldn't fail, but return an error if it does.
+	 */
+	if (set_memory_p(kbuffer, pagecount))
+		result = false;
+
 	return result;
 }
 
 static bool hv_vtom_tlb_flush_required(bool private)
 {
-	return true;
+	/*
+	 * Since hv_vtom_clear_present() marks the PTEs as "not present"
+	 * and flushes the TLB, they can't be in the TLB. That makes the
+	 * flush controlled by this function redundant, so return "false".
+	 */
+	return false;
 }
 
 static bool hv_vtom_cache_flush_required(void)
@@ -618,6 +662,7 @@ void __init hv_vtom_init(void)
 	x86_platform.hyper.is_private_mmio = hv_is_private_mmio;
 	x86_platform.guest.enc_cache_flush_required = hv_vtom_cache_flush_required;
 	x86_platform.guest.enc_tlb_flush_required = hv_vtom_tlb_flush_required;
+	x86_platform.guest.enc_status_change_prepare = hv_vtom_clear_present;
 	x86_platform.guest.enc_status_change_finish = hv_vtom_set_host_visibility;
 
 	/* Set WB as the default cache mode. */
-- 
2.25.1

