Thread (61 messages) flat view 61 messages, 3 authors, 21h ago

Re: [PATCH v4 9/9] mshv: set up own SynIC registers on a nested root partition

From: sashiko-bot@kernel.org
Date: 2026-08-31 12:09:42

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] Missing memory decryption for locally allocated L1VH SIRBP in Confidential VMs.
--

commit 0a5809fa9d9b76912e599a03beec81fda2326b1b
Author: Wei Hu [off-list ref]

mshv: set up own SynIC registers on a nested root partition

This patch modifies mshv_synic to set up its own SynIC registers on nested
root partitions instead of assuming VMBus has provisioned them. It also
updates the mapping of hypervisor-provided SIMP, SIEFP, and root SIRBP pages
to use MEMREMAP_DEC to support confidential hosts.
quoted hunk ↗ jump to hunk
diff --git a/drivers/hv/mshv_synic.c b/drivers/hv/mshv_synic.c
index 0fdbae1e053c3..8470cf958f21b 100644
--- a/drivers/hv/mshv_synic.c
+++ b/drivers/hv/mshv_synic.c
[ ... ]
quoted hunk ↗ jump to hunk
@@ -511,8 +524,8 @@ static int mshv_synic_cpu_init(unsigned int cpu)
 	sirbp.as_uint64 = hv_get_non_nested_msr(HV_MSR_SIRBP);
 
 	if (hv_root_partition()) {
-		event_ring_page = memremap(sirbp.base_sirbp_gpa << HV_HYP_PAGE_SHIFT,
-					   HV_HYP_PAGE_SIZE, MEMREMAP_WB);
+		event_ring_page =
+			mshv_synic_map_shared_page(sirbp.base_sirbp_gpa);
 
 		if (!event_ring_page)
 			goto cleanup_siefp;
[Severity: High]
Since this patch adds support for confidential hosts by mapping
hypervisor-provided pages with MEMREMAP_DEC, what happens to the locally
allocated SIRBP in the else branch for L1VH partitions?

If we look further down in mshv_synic_cpu_init():

	} else {
		/*
		 * On L1VH the hypervisor does not provide a SIRBP page.
		 * Allocate one and program its GPA into the MSR.
		 */
		event_ring_page = (struct hv_synic_event_ring_page *)
			get_zeroed_page(GFP_KERNEL);

In a Confidential VM, memory allocated via get_zeroed_page() is encrypted by
default. Because set_memory_decrypted() is never called here, will the
hypervisor be unable to access the event ring, causing a regression in event
delivery for L1VH?

For comparison, drivers/hv/hv.c decrypts its locally allocated SynIC pages
via hv_alloc_page(..., decrypt=true, ...).

[ ... ]
quoted hunk ↗ jump to hunk
@@ -595,8 +608,7 @@ static int mshv_synic_cpu_exit(unsigned int cpu)
 	struct hv_message_page *msg_page;
 	struct hv_synic_event_flags_page *event_flags_page;
 	struct hv_synic_event_ring_page *event_ring_page;
-	/* VMBus owns SIMP/SIEFP/SCONTROL when it is active */
-	bool vmbus_active = hv_vmbus_exists();
+	bool vmbus_active = mshv_synic_vmbus_owns_registers();
 
 	msg_page = READ_ONCE(spages->hyp_synic_message_page);
 	event_flags_page = READ_ONCE(spages->synic_event_flags_page);
[Severity: High]
If the allocation in mshv_synic_cpu_init() is updated to decrypt the page,
does the cleanup path in mshv_synic_cpu_exit() need a corresponding change?

Looking at the cleanup path in mshv_synic_cpu_exit():

	if (event_ring_page) {
		if (hv_root_partition())
			memunmap(event_ring_page);
		else
			free_page((unsigned long)event_ring_page);
	}

If the page is decrypted during initialization, will freeing it here without
first calling set_memory_encrypted() return a decrypted page to the kernel
allocator and cause a memory corruption regression?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260831112704.2851147-1-weh@linux.microsoft.com?part=9
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help