Skip to contentSkip to conversation

KVM: selftests: Add simple SEV test

PATCH Superseded Peter Gonda 2 participants 24 messages Aug 29, 2022 – Oct 27, 2022

Subsystems (2)

v4 of 7 Superseded Latest is v7 →

lore
Share & tools
lore
#0 lore raw

v4 cover KVM: selftests: Add simple SEV test

This patch series continues the work Michael Roth has done in supporting SEV guests in selftests. It continues on top of the work Sean Christopherson has sent to support ucalls from SEV guests. Along with a very simple version of the SEV selftests Michael originally proposed.

V4
* Rebase ontop of seanjc@'s latest Ucall Pool series:
https://lore.kernel.org/linux-arm-kernel/[email protected]/
* Fix up review comments from seanjc
* Switch authorship on 2 patches because of significant changes, added
* Michael as suggested-by or originally-by.

V3
* Addressed more of andrew.jones@ in ucall patches.
* Fix build in non-x86 archs.

V2
* Dropped RFC tag
* Correctly separated Sean's ucall patches into 2 as originally intended.
* Addressed andrew.jones@ in ucall patches.
* Fixed ucall pool usage to work for other archs

V1
* https://lore.kernel.org/all/[email protected]/

Michael Roth (5):
KVM: selftests: move vm_phy_pages_alloc() earlier in file
KVM: selftests: sparsebit: add const where appropriate
KVM: selftests: add hooks for managing encrypted guest memory
KVM: selftests: handle encryption bits in page tables
KVM: selftests: add support for encrypted vm_vaddr_* allocations

Peter Gonda (3):
KVM: selftests: add library for creating/interacting with SEV guests
KVM: selftests: Update ucall pool to allocate from shared memory
KVM: selftests: Add simple sev vm testing

 tools/testing/selftests/kvm/.gitignore        |   1 +
 tools/testing/selftests/kvm/Makefile          |   2 +
 .../selftests/kvm/include/kvm_util_base.h     |  23 ++
 .../testing/selftests/kvm/include/sparsebit.h |  36 +--
 .../selftests/kvm/include/x86_64/sev.h        |  47 +++
 tools/testing/selftests/kvm/lib/kvm_util.c    | 267 +++++++++++++-----
 tools/testing/selftests/kvm/lib/sparsebit.c   |  48 ++--
 .../testing/selftests/kvm/lib/ucall_common.c  |   2 +-
 .../selftests/kvm/lib/x86_64/processor.c      |  15 +-
 tools/testing/selftests/kvm/lib/x86_64/sev.c  | 232 +++++++++++++++
 .../selftests/kvm/x86_64/sev_all_boot_test.c  | 127 +++++++++
 11 files changed, 674 insertions(+), 126 deletions(-)
 create mode 100644 tools/testing/selftests/kvm/include/x86_64/sev.h
 create mode 100644 tools/testing/selftests/kvm/lib/x86_64/sev.c
 create mode 100644 tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c
Signature · 1 line
2.37.2.672.g94769d06f0-goog
#1 lore raw

PATCH v4 1/8 KVM: selftests: move vm_phy_pages_alloc() earlier in file

From: Michael Roth <[email protected]>

Subsequent patches will break some of this code out into file-local helper functions, which will be used by functions like vm_vaddr_alloc(), which currently are defined earlier in the file, so a forward declaration would be needed.

Instead, move it earlier in the file, just above vm_vaddr_alloc() and and friends, which are the main users.

---
 tools/testing/selftests/kvm/lib/kvm_util.c | 145 ++++++++++-----------
 1 file changed, 72 insertions(+), 73 deletions(-)
tools/testing/selftests/kvm/lib/kvm_util.c +72 -73
index 846f9f6c5a17..06559994711e 100644
@@ -1100,6 +1100,78 @@ struct kvm_vcpu *__vm_vcpu_add(struct kvm_vm *vm, uint32_t vcpu_id)
 	return vcpu;
 }
 
+/*
+ * Physical Contiguous Page Allocator
+ *
+ * Input Args:
+ *   vm - Virtual Machine
+ *   num - number of pages
+ *   paddr_min - Physical address minimum
+ *   memslot - Memory region to allocate page from
+ *
+ * Output Args: None
+ *
+ * Return:
+ *   Starting physical address
+ *
+ * Within the VM specified by vm, locates a range of available physical
+ * pages at or above paddr_min. If found, the pages are marked as in use
+ * and their base address is returned. A TEST_ASSERT failure occurs if
+ * not enough pages are available at or above paddr_min.
+ */
+vm_paddr_t vm_phy_pages_alloc(struct kvm_vm *vm, size_t num,
+			      vm_paddr_t paddr_min, uint32_t memslot)
+{
+	struct userspace_mem_region *region;
+	sparsebit_idx_t pg, base;
+
+	TEST_ASSERT(num > 0, "Must allocate at least one page");
+
+	TEST_ASSERT((paddr_min % vm->page_size) == 0,
+		"Min physical address not divisible by page size.\n paddr_min: 0x%lx page_size: 0x%x",
+		paddr_min, vm->page_size);
+
+	region = memslot2region(vm, memslot);
+	base = pg = paddr_min >> vm->page_shift;
+
+	do {
+		for (; pg < base + num; ++pg) {
+			if (!sparsebit_is_set(region->unused_phy_pages, pg)) {
+				base = pg = sparsebit_next_set(region->unused_phy_pages, pg);
+				break;
+			}
+		}
+	} while (pg && pg != base + num);
+
+	if (pg == 0) {
+		fprintf(stderr,
+			"No guest physical page available, paddr_min: 0x%lx page_size: 0x%x memslot: %u\n",
+			paddr_min, vm->page_size, memslot);
+		fputs("---- vm dump ----\n", stderr);
+		vm_dump(stderr, vm, 2);
+		abort();
+	}
+
+	for (pg = base; pg < base + num; ++pg)
+		sparsebit_clear(region->unused_phy_pages, pg);
+
+	return base * vm->page_size;
+}
+
+vm_paddr_t vm_phy_page_alloc(struct kvm_vm *vm, vm_paddr_t paddr_min,
+			     uint32_t memslot)
+{
+	return vm_phy_pages_alloc(vm, 1, paddr_min, memslot);
+}
+
+/* Arbitrary minimum physical address used for virtual translation tables. */
+#define KVM_GUEST_PAGE_TABLE_MIN_PADDR 0x180000
+
+vm_paddr_t vm_alloc_page_table(struct kvm_vm *vm)
+{
+	return vm_phy_page_alloc(vm, KVM_GUEST_PAGE_TABLE_MIN_PADDR, 0);
+}
+
 /*
  * VM Virtual Address Unused Gap
  *
@@ -1746,79 +1818,6 @@ const char *exit_reason_str(unsigned int exit_reason)
 	return "Unknown";
 }
 
-/*
- * Physical Contiguous Page Allocator
- *
- * Input Args:
- *   vm - Virtual Machine
- *   num - number of pages
- *   paddr_min - Physical address minimum
- *   memslot - Memory region to allocate page from
- *
- * Output Args: None
- *
- * Return:
- *   Starting physical address
- *
- * Within the VM specified by vm, locates a range of available physical
- * pages at or above paddr_min. If found, the pages are marked as in use
- * and their base address is returned. A TEST_ASSERT failure occurs if
- * not enough pages are available at or above paddr_min.
- */
-vm_paddr_t vm_phy_pages_alloc(struct kvm_vm *vm, size_t num,
-			      vm_paddr_t paddr_min, uint32_t memslot)
-{
-	struct userspace_mem_region *region;
-	sparsebit_idx_t pg, base;
-
-	TEST_ASSERT(num > 0, "Must allocate at least one page");
-
-	TEST_ASSERT((paddr_min % vm->page_size) == 0, "Min physical address "
-		"not divisible by page size.\n"
-		"  paddr_min: 0x%lx page_size: 0x%x",
-		paddr_min, vm->page_size);
-
-	region = memslot2region(vm, memslot);
-	base = pg = paddr_min >> vm->page_shift;
-
-	do {
-		for (; pg < base + num; ++pg) {
-			if (!sparsebit_is_set(region->unused_phy_pages, pg)) {
-				base = pg = sparsebit_next_set(region->unused_phy_pages, pg);
-				break;
-			}
-		}
-	} while (pg && pg != base + num);
-
-	if (pg == 0) {
-		fprintf(stderr, "No guest physical page available, "
-			"paddr_min: 0x%lx page_size: 0x%x memslot: %u\n",
-			paddr_min, vm->page_size, memslot);
-		fputs("---- vm dump ----\n", stderr);
-		vm_dump(stderr, vm, 2);
-		abort();
-	}
-
-	for (pg = base; pg < base + num; ++pg)
-		sparsebit_clear(region->unused_phy_pages, pg);
-
-	return base * vm->page_size;
-}
-
-vm_paddr_t vm_phy_page_alloc(struct kvm_vm *vm, vm_paddr_t paddr_min,
-			     uint32_t memslot)
-{
-	return vm_phy_pages_alloc(vm, 1, paddr_min, memslot);
-}
-
-/* Arbitrary minimum physical address used for virtual translation tables. */
-#define KVM_GUEST_PAGE_TABLE_MIN_PADDR 0x180000
-
-vm_paddr_t vm_alloc_page_table(struct kvm_vm *vm)
-{
-	return vm_phy_page_alloc(vm, KVM_GUEST_PAGE_TABLE_MIN_PADDR, 0);
-}
-
 /*
  * Address Guest Virtual to Host Virtual
  *
Signature · 1 line
2.37.2.672.g94769d06f0-goog
#2 lore raw

Re: [V4 1/8] KVM: selftests: move vm_phy_pages_alloc() earlier in file

Quoted 9 lines · Peter Gonda

From: Michael Roth <[email protected]>

Subsequent patches will break some of this code out into file-local helper functions, which will be used by functions like vm_vaddr_alloc(), which currently are defined earlier in the file, so a forward declaration would be needed.

Instead, move it earlier in the file, just above vm_vaddr_alloc() and and friends, which are the main users.

Meh, the wrappers are one-liners, why not keep this where it is, keep it exposed, and then make the wrappers "static inline".

#3 lore raw

PATCH v4 2/8 KVM: selftests: sparsebit: add const where appropriate

In reply to Peter Gonda

From: Michael Roth <[email protected]>

Subsequent patches will introduce an encryption bitmap in kvm_util that would be useful to allow tests to access in read-only fashion. This will be done via a const sparsebit*. To avoid warnings or the need to add casts everywhere, add const to the various sparsebit functions that are applicable for read-only usage of sparsebit.

---
 .../testing/selftests/kvm/include/sparsebit.h | 36 +++++++-------
 tools/testing/selftests/kvm/lib/sparsebit.c   | 48 +++++++++----------
 2 files changed, 42 insertions(+), 42 deletions(-)
tools/testing/selftests/kvm/include/sparsebit.h +18 -18
index 12a9a4b9cead..fb5170d57fcb 100644
@@ -30,26 +30,26 @@ typedef uint64_t sparsebit_num_t;
 
 struct sparsebit *sparsebit_alloc(void);
 void sparsebit_free(struct sparsebit **sbitp);
-void sparsebit_copy(struct sparsebit *dstp, struct sparsebit *src);
+void sparsebit_copy(struct sparsebit *dstp, const struct sparsebit *src);
 
-bool sparsebit_is_set(struct sparsebit *sbit, sparsebit_idx_t idx);
-bool sparsebit_is_set_num(struct sparsebit *sbit,
+bool sparsebit_is_set(const struct sparsebit *sbit, sparsebit_idx_t idx);
+bool sparsebit_is_set_num(const struct sparsebit *sbit,
 			  sparsebit_idx_t idx, sparsebit_num_t num);
-bool sparsebit_is_clear(struct sparsebit *sbit, sparsebit_idx_t idx);
-bool sparsebit_is_clear_num(struct sparsebit *sbit,
+bool sparsebit_is_clear(const struct sparsebit *sbit, sparsebit_idx_t idx);
+bool sparsebit_is_clear_num(const struct sparsebit *sbit,
 			    sparsebit_idx_t idx, sparsebit_num_t num);
-sparsebit_num_t sparsebit_num_set(struct sparsebit *sbit);
-bool sparsebit_any_set(struct sparsebit *sbit);
-bool sparsebit_any_clear(struct sparsebit *sbit);
-bool sparsebit_all_set(struct sparsebit *sbit);
-bool sparsebit_all_clear(struct sparsebit *sbit);
-sparsebit_idx_t sparsebit_first_set(struct sparsebit *sbit);
-sparsebit_idx_t sparsebit_first_clear(struct sparsebit *sbit);
-sparsebit_idx_t sparsebit_next_set(struct sparsebit *sbit, sparsebit_idx_t prev);
-sparsebit_idx_t sparsebit_next_clear(struct sparsebit *sbit, sparsebit_idx_t prev);
-sparsebit_idx_t sparsebit_next_set_num(struct sparsebit *sbit,
+sparsebit_num_t sparsebit_num_set(const struct sparsebit *sbit);
+bool sparsebit_any_set(const struct sparsebit *sbit);
+bool sparsebit_any_clear(const struct sparsebit *sbit);
+bool sparsebit_all_set(const struct sparsebit *sbit);
+bool sparsebit_all_clear(const struct sparsebit *sbit);
+sparsebit_idx_t sparsebit_first_set(const struct sparsebit *sbit);
+sparsebit_idx_t sparsebit_first_clear(const struct sparsebit *sbit);
+sparsebit_idx_t sparsebit_next_set(const struct sparsebit *sbit, sparsebit_idx_t prev);
+sparsebit_idx_t sparsebit_next_clear(const struct sparsebit *sbit, sparsebit_idx_t prev);
+sparsebit_idx_t sparsebit_next_set_num(const struct sparsebit *sbit,
 				       sparsebit_idx_t start, sparsebit_num_t num);
-sparsebit_idx_t sparsebit_next_clear_num(struct sparsebit *sbit,
+sparsebit_idx_t sparsebit_next_clear_num(const struct sparsebit *sbit,
 					 sparsebit_idx_t start, sparsebit_num_t num);
 
 void sparsebit_set(struct sparsebit *sbitp, sparsebit_idx_t idx);
@@ -62,9 +62,9 @@ void sparsebit_clear_num(struct sparsebit *sbitp,
 			 sparsebit_idx_t start, sparsebit_num_t num);
 void sparsebit_clear_all(struct sparsebit *sbitp);
 
-void sparsebit_dump(FILE *stream, struct sparsebit *sbit,
+void sparsebit_dump(FILE *stream, const struct sparsebit *sbit,
 		    unsigned int indent);
-void sparsebit_validate_internal(struct sparsebit *sbit);
+void sparsebit_validate_internal(const struct sparsebit *sbit);
 
 #ifdef __cplusplus
 }
tools/testing/selftests/kvm/lib/sparsebit.c +24 -24

190 lines not on this page · raw message

Signature · 1 line
2.37.2.672.g94769d06f0-goog
#4 lore raw

PATCH v4 3/8 KVM: selftests: add hooks for managing encrypted guest memory

In reply to Peter Gonda

From: Michael Roth <[email protected]>

VM implementations that make use of encrypted memory need a way to configure things like the encryption/shared bit position for page table handling, the default encryption policy for internal allocations made by the core library, and a way to fetch the list/bitmap of encrypted pages to do the actual memory encryption. Add an interface to configure these parameters. Also introduce a sparsebit map to track allocations/mappings that should be treated as encrypted, and provide a way for VM implementations to retrieve it to handle operations related memory encryption.

---
 .../selftests/kvm/include/kvm_util_base.h     | 17 ++++++
 tools/testing/selftests/kvm/lib/kvm_util.c    | 52 +++++++++++++++++--
 2 files changed, 66 insertions(+), 3 deletions(-)
tools/testing/selftests/kvm/include/kvm_util_base.h +17 -0
index 59d52b58a1a6..5ecde5ad4c2f 100644
@@ -33,6 +33,7 @@ typedef uint64_t vm_vaddr_t; /* Virtual Machine (Guest) virtual address */
 struct userspace_mem_region {
 	struct kvm_userspace_memory_region region;
 	struct sparsebit *unused_phy_pages;
+	struct sparsebit *encrypted_phy_pages;
 	int fd;
 	off_t offset;
 	void *host_mem;
@@ -65,6 +66,14 @@ struct userspace_mem_regions {
 	DECLARE_HASHTABLE(slot_hash, 9);
 };
 
+/* Memory encryption policy/configuration. */
+struct vm_memcrypt {
+	bool enabled;
+	int8_t enc_by_default;
+	bool has_enc_bit;
+	int8_t enc_bit;
+};
+
 struct kvm_vm {
 	int mode;
 	unsigned long type;
@@ -89,6 +98,7 @@ struct kvm_vm {
 	vm_vaddr_t idt;
 	vm_vaddr_t handlers;
 	uint32_t dirty_ring_size;
+	struct vm_memcrypt memcrypt;
 
 	/* Cache of information for binary stats interface */
 	int stats_fd;
@@ -849,4 +859,11 @@ static inline int __vm_disable_nx_huge_pages(struct kvm_vm *vm)
 	return __vm_enable_cap(vm, KVM_CAP_VM_DISABLE_NX_HUGE_PAGES, 0);
 }
 
+void vm_set_memory_encryption(struct kvm_vm *vm, bool enc_by_default, bool has_enc_bit,
+			      uint8_t enc_bit);
+
+const struct sparsebit *vm_get_encrypted_phy_pages(struct kvm_vm *vm, int slot,
+						   vm_paddr_t *gpa_start,
+						   uint64_t *size);
+
 #endif /* SELFTEST_KVM_UTIL_BASE_H */
tools/testing/selftests/kvm/lib/kvm_util.c +49 -3

96 lines not on this page · raw message

Signature · 1 line
2.37.2.672.g94769d06f0-goog
#5 lore raw

Re: [V4 3/8] KVM: selftests: add hooks for managing encrypted guest memory

tools/testing/selftests/kvm/lib/kvm_util.c struct kvm_vcpu *__vm_vcpu_add(struct kvm_vm *vm, uint32_t vcpu_id) · line 1126

Quoted 2 lines · Peter Gonda
tools/testing/selftests/kvm/lib/kvm_util.c +2 -0
+static vm_paddr_t
+_vm_phy_pages_alloc(struct kvm_vm *vm, size_t num, vm_paddr_t paddr_min,

Do not wrap before the function name. Linus has a nice explanation/rant on this[*]. Note to self, add a Vim macro for this...

[*] https://lore.kernel.org/all/CAHk-=wjoLAYG446ZNHfg=GhjSY6nFmuB_wA8fYd5iLBNXjo9Bw@mail.gmail.com

tools/testing/selftests/kvm/lib/kvm_util.c vm_paddr_t vm_phy_pages_alloc(struct kvm_vm *vm, size_t num, · line 1161

Quoted 12 lines
tools/testing/selftests/kvm/lib/kvm_util.c +3 -1
+		    uint32_t memslot, bool encrypt)
 {
 	struct userspace_mem_region *region;
 	sparsebit_idx_t pg, base;
@@ -1152,12 +1156,22 @@ vm_paddr_t vm_phy_pages_alloc(struct kvm_vm *vm, size_t num,
 		abort();
 	}
 
-	for (pg = base; pg < base + num; ++pg)
+	for (pg = base; pg < base + num; ++pg) {
 		sparsebit_clear(region->unused_phy_pages, pg);
+		if (encrypt)

prefer s/encrypt/private, and s/encrypted_phy_pages/private_phy_pages. pKVM doesn't rely on encryption, and it's not impossible that x86 will someday gain similar functionality. And "encrypted" is also technically wrong for SEV and TDX, as shared memory can also be encrypted with a common key.

tools/testing/selftests/kvm/lib/kvm_util.c vm_paddr_t vm_phy_pages_alloc(struct kvm_vm *vm, size_t num, · line 1172

Quoted 11 lines
tools/testing/selftests/kvm/lib/kvm_util.c +7 -0
+			sparsebit_set(region->encrypted_phy_pages, pg);
+	}
 
 	return base * vm->page_size;
 }
 
+vm_paddr_t vm_phy_pages_alloc(struct kvm_vm *vm, size_t num,
+			      vm_paddr_t paddr_min, uint32_t memslot)
+{
+	return _vm_phy_pages_alloc(vm, num, paddr_min, memslot,
+				   vm->memcrypt.enc_by_default);

enc_by_default yields a bizarre API. The behavior depends on whether or not the VM is protected, and whether or not the VM wants to protect memory by default.

For simplicity, IMO vm_phy_pages_alloc() should allocate memory as private if the VM supports protected memory, i.e. just have vm->protected or whatever and use that here.

tools/testing/selftests/kvm/lib/kvm_util.c void vm_dump(FILE *stream, struct kvm_vm *vm, uint8_t indent) · line 1758

Quoted 10 lines
tools/testing/selftests/kvm/lib/kvm_util.c +3 -0
+}
+
 vm_paddr_t vm_phy_page_alloc(struct kvm_vm *vm, vm_paddr_t paddr_min,
 			     uint32_t memslot)
 {
@@ -1741,6 +1755,10 @@ void vm_dump(FILE *stream, struct kvm_vm *vm, uint8_t indent)
 			region->host_mem);
 		fprintf(stream, "%*sunused_phy_pages: ", indent + 2, "");
 		sparsebit_dump(stream, region->unused_phy_pages, 0);
+		if (vm->memcrypt.enabled) {

vm->protected

Quoted 13 lines
tools/testing/selftests/kvm/lib/kvm_util.c +6 -0
+			fprintf(stream, "%*sencrypted_phy_pages: ", indent + 2, "");
+			sparsebit_dump(stream, region->encrypted_phy_pages, 0);
+		}
 	}
 	fprintf(stream, "%*sMapped Virtual Pages:\n", indent, "");
 	sparsebit_dump(stream, vm->vpages_mapped, indent + 2);
@@ -1989,3 +2007,31 @@ void __vm_get_stat(struct kvm_vm *vm, const char *stat_name, uint64_t *data,
 		break;
 	}
 }
+
+void vm_set_memory_encryption(struct kvm_vm *vm, bool enc_by_default, bool has_enc_bit,
+			      uint8_t enc_bit)

tools/testing/selftests/kvm/lib/kvm_util.c void __vm_get_stat(struct kvm_vm *vm, const char *stat_name, uint64_t *data, · line 2022

Quoted 10 lines
tools/testing/selftests/kvm/lib/kvm_util.c +10 -0
+{
+	vm->memcrypt.enabled = true;
+	vm->memcrypt.enc_by_default = enc_by_default;
+	vm->memcrypt.has_enc_bit = has_enc_bit;
+	vm->memcrypt.enc_bit = enc_bit;
+}
+
+const struct sparsebit *
+vm_get_encrypted_phy_pages(struct kvm_vm *vm, int slot, vm_paddr_t *gpa_start,
+			   uint64_t *size)

Bad wrap.

tools/testing/selftests/kvm/lib/kvm_util.c void __vm_get_stat(struct kvm_vm *vm, const char *stat_name, uint64_t *data, · line 2026

Quoted 4 lines
tools/testing/selftests/kvm/lib/kvm_util.c +4 -0
+{
+	struct userspace_mem_region *region;
+
+	if (!vm->memcrypt.enabled)

This seems rather silly, why not TEST_ASSERT()?

tools/testing/selftests/kvm/lib/kvm_util.c void __vm_get_stat(struct kvm_vm *vm, const char *stat_name, uint64_t *data, · line 2030

Quoted 4 lines
tools/testing/selftests/kvm/lib/kvm_util.c +4 -0
+		return NULL;
+
+	region = memslot2region(vm, slot);
+	if (!region)

Same here, TEST_ASSERT() seems more appropriate.

Actually, I can't envision a use outside of SEV. AFAIK, no other architecture does the whole "launch update" thing. I.e. just open code this in sev_encrypt(). The more generic API that will be useful for other VM types will be to query if a specific GPA is private vs. shared.

Quoted 10 lines
tools/testing/selftests/kvm/lib/kvm_util.c +7 -0
+		return NULL;
+
+	*size = region->region.memory_size;
+	*gpa_start = region->region.guest_phys_addr;
+
+	return region->encrypted_phy_pages;
+}
Signature · 1 line
2.37.2.672.g94769d06f0-goog
#6 lore raw

Re: [V4 3/8] KVM: selftests: add hooks for managing encrypted guest memory

Quoted 10 lines · Sean Christopherson

tools/testing/selftests/kvm/lib/kvm_util.c struct kvm_vcpu *__vm_vcpu_add(struct kvm_vm *vm, uint32_t vcpu_id) · line 1126

Quoted 2 lines · Peter Gonda
tools/testing/selftests/kvm/lib/kvm_util.c +2 -0
+static vm_paddr_t
+_vm_phy_pages_alloc(struct kvm_vm *vm, size_t num, vm_paddr_t paddr_min,

Do not wrap before the function name. Linus has a nice explanation/rant on this[*]. Note to self, add a Vim macro for this...

[*] https://lore.kernel.org/all/CAHk-=wjoLAYG446ZNHfg=GhjSY6nFmuB_wA8fYd5iLBNXjo9Bw@mail.gmail.com

Fixed. Thanks. I'll work on a fix to my VIM setup.

Quoted 8 lines
Quoted 8 lines
tools/testing/selftests/kvm/lib/kvm_util.c +1 -0
+                 uint32_t memslot, bool encrypt)
 {
      struct userspace_mem_region *region;
      sparsebit_idx_t pg, base;
@@ -1152,12 +1156,22 @@ vm_paddr_t vm_phy_pages_alloc(struct kvm_vm *vm, size_t num,
              abort();
      }
 
Quoted 9 lines

tools/testing/selftests/kvm/lib/kvm_util.c vm_paddr_t vm_phy_pages_alloc(struct kvm_vm *vm, size_t num, · line 1161

Quoted 4 lines
tools/testing/selftests/kvm/lib/kvm_util.c +2 -1
-     for (pg = base; pg < base + num; ++pg)
+     for (pg = base; pg < base + num; ++pg) {
              sparsebit_clear(region->unused_phy_pages, pg);
+             if (encrypt)

prefer s/encrypt/private, and s/encrypted_phy_pages/private_phy_pages. pKVM doesn't rely on encryption, and it's not impossible that x86 will someday gain similar functionality. And "encrypted" is also technically wrong for SEV and TDX, as shared memory can also be encrypted with a common key.

Makes sense. Private or protected sound better.

Quoted 16 lines

tools/testing/selftests/kvm/lib/kvm_util.c vm_paddr_t vm_phy_pages_alloc(struct kvm_vm *vm, size_t num, · line 1172

Quoted 12 lines
tools/testing/selftests/kvm/lib/kvm_util.c +7 -0
+                     sparsebit_set(region->encrypted_phy_pages, pg);
+     }
 
      return base * vm->page_size;
 }
 
+vm_paddr_t vm_phy_pages_alloc(struct kvm_vm *vm, size_t num,
+                           vm_paddr_t paddr_min, uint32_t memslot)
+{
+     return _vm_phy_pages_alloc(vm, num, paddr_min, memslot,
+                                vm->memcrypt.enc_by_default);

enc_by_default yields a bizarre API. The behavior depends on whether or not the VM is protected, and whether or not the VM wants to protect memory by default.

Quoted 3 lines

For simplicity, IMO vm_phy_pages_alloc() should allocate memory as private if the VM supports protected memory, i.e. just have vm->protected or whatever and use that here.

Removed "enc_by_default" concept.

Quoted 13 lines

tools/testing/selftests/kvm/lib/kvm_util.c void vm_dump(FILE *stream, struct kvm_vm *vm, uint8_t indent) · line 1758

Quoted 11 lines
tools/testing/selftests/kvm/lib/kvm_util.c +3 -0
+}
+
 vm_paddr_t vm_phy_page_alloc(struct kvm_vm *vm, vm_paddr_t paddr_min,
                           uint32_t memslot)
 {
@@ -1741,6 +1755,10 @@ void vm_dump(FILE *stream, struct kvm_vm *vm, uint8_t indent)
                      region->host_mem);
              fprintf(stream, "%*sunused_phy_pages: ", indent + 2, "");
              sparsebit_dump(stream, region->unused_phy_pages, 0);
+             if (vm->memcrypt.enabled) {

vm->protected

renamed memcrypt -> protected.

Quoted 18 lines
Quoted 18 lines
tools/testing/selftests/kvm/lib/kvm_util.c +10 -0
+                     fprintf(stream, "%*sencrypted_phy_pages: ", indent + 2, "");
+                     sparsebit_dump(stream, region->encrypted_phy_pages, 0);
+             }
      }
      fprintf(stream, "%*sMapped Virtual Pages:\n", indent, "");
      sparsebit_dump(stream, vm->vpages_mapped, indent + 2);
@@ -1989,3 +2007,31 @@ void __vm_get_stat(struct kvm_vm *vm, const char *stat_name, uint64_t *data,
              break;
      }
 }
+
+void vm_set_memory_encryption(struct kvm_vm *vm, bool enc_by_default, bool has_enc_bit,
+                           uint8_t enc_bit)
+{
+     vm->memcrypt.enabled = true;
+     vm->memcrypt.enc_by_default = enc_by_default;
+     vm->memcrypt.has_enc_bit = has_enc_bit;
Quoted 8 lines

tools/testing/selftests/kvm/lib/kvm_util.c void __vm_get_stat(struct kvm_vm *vm, const char *stat_name, uint64_t *data, · line 2022

Quoted 6 lines
tools/testing/selftests/kvm/lib/kvm_util.c +6 -0
+     vm->memcrypt.enc_bit = enc_bit;
+}
+
+const struct sparsebit *
+vm_get_encrypted_phy_pages(struct kvm_vm *vm, int slot, vm_paddr_t *gpa_start,
+                        uint64_t *size)

Bad wrap.

Fixed.

Quoted 13 lines

tools/testing/selftests/kvm/lib/kvm_util.c void __vm_get_stat(struct kvm_vm *vm, const char *stat_name, uint64_t *data, · line 2026

Quoted 5 lines
tools/testing/selftests/kvm/lib/kvm_util.c +4 -0
+{
+     struct userspace_mem_region *region;
+
+     if (!vm->memcrypt.enabled)

This seems rather silly, why not TEST_ASSERT()?

tools/testing/selftests/kvm/lib/kvm_util.c void __vm_get_stat(struct kvm_vm *vm, const char *stat_name, uint64_t *data, · line 2030

Quoted 4 lines
tools/testing/selftests/kvm/lib/kvm_util.c +4 -0
+             return NULL;
+
+     region = memslot2region(vm, slot);
+     if (!region)
Quoted 6 lines

Same here, TEST_ASSERT() seems more appropriate.

Actually, I can't envision a use outside of SEV. AFAIK, no other architecture does the whole "launch update" thing. I.e. just open code this in sev_encrypt(). The more generic API that will be useful for other VM types will be to query if a specific GPA is private vs. shared.

Good point. I'll move this code into that sev_encrypt() flow and remove this function completely.

Quoted 11 lines
Quoted 11 lines
tools/testing/selftests/kvm/lib/kvm_util.c +7 -1
+             return NULL;
+
+     *size = region->region.memory_size;
+     *gpa_start = region->region.guest_phys_addr;
+
+     return region->encrypted_phy_pages;
+}
--

2.37.2.672.g94769d06f0-goog

#7 lore raw

PATCH v4 4/8 KVM: selftests: handle encryption bits in page tables

In reply to Peter Gonda

From: Michael Roth <[email protected]>

SEV guests rely on an encyption bit which resides within the range that current code treats as address bits. Guest code will expect these bits to be set appropriately in their page tables, whereas the rest of the kvm_util functions will generally expect these bits to not be present. Introduce addr_gpa2raw()/addr_raw2gpa() to add/remove these bits, then use them where appropriate.

---
 .../selftests/kvm/include/kvm_util_base.h     |  2 +
 tools/testing/selftests/kvm/lib/kvm_util.c    | 55 ++++++++++++++++++-
 .../selftests/kvm/lib/x86_64/processor.c      | 15 +++--
 3 files changed, 66 insertions(+), 6 deletions(-)
tools/testing/selftests/kvm/include/kvm_util_base.h +2 -0
index 5ecde5ad4c2f..dda8467d1434 100644
@@ -401,6 +401,8 @@ void *addr_gpa2hva(struct kvm_vm *vm, vm_paddr_t gpa);
 void *addr_gva2hva(struct kvm_vm *vm, vm_vaddr_t gva);
 vm_paddr_t addr_hva2gpa(struct kvm_vm *vm, void *hva);
 void *addr_gpa2alias(struct kvm_vm *vm, vm_paddr_t gpa);
+vm_paddr_t addr_raw2gpa(struct kvm_vm *vm, vm_vaddr_t gpa_raw);
+vm_paddr_t addr_gpa2raw(struct kvm_vm *vm, vm_vaddr_t gpa);
 
 void vcpu_run(struct kvm_vcpu *vcpu);
 int _vcpu_run(struct kvm_vcpu *vcpu);
tools/testing/selftests/kvm/lib/kvm_util.c +54 -1

69 lines not on this page · raw message

tools/testing/selftests/kvm/lib/x86_64/processor.c +10 -5
index 2e6e61bbe81b..b2df259ce706 100644
@@ -118,7 +118,7 @@ void virt_arch_pgd_alloc(struct kvm_vm *vm)
 
 	/* If needed, create page map l4 table. */
 	if (!vm->pgd_created) {
-		vm->pgd = vm_alloc_page_table(vm);
+		vm->pgd = addr_gpa2raw(vm, vm_alloc_page_table(vm));
 		vm->pgd_created = true;
 	}
 }
@@ -140,13 +140,15 @@ static uint64_t *virt_create_upper_pte(struct kvm_vm *vm,
 				       int target_level)
 {
 	uint64_t *pte = virt_get_pte(vm, pt_pfn, vaddr, current_level);
+	uint64_t paddr_raw = addr_gpa2raw(vm, paddr);
 
 	if (!(*pte & PTE_PRESENT_MASK)) {
 		*pte = PTE_PRESENT_MASK | PTE_WRITABLE_MASK;
 		if (current_level == target_level)
-			*pte |= PTE_LARGE_MASK | (paddr & PHYSICAL_PAGE_MASK);
+			*pte |= PTE_LARGE_MASK | (paddr_raw & PHYSICAL_PAGE_MASK);
 		else
-			*pte |= vm_alloc_page_table(vm) & PHYSICAL_PAGE_MASK;
+			*pte |= addr_gpa2raw(vm, vm_alloc_page_table(vm)) & PHYSICAL_PAGE_MASK;
+
 	} else {
 		/*
 		 * Entry already present.  Assert that the caller doesn't want
@@ -184,6 +186,8 @@ void __virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr, int level)
 		    "Physical address beyond maximum supported,\n"
 		    "  paddr: 0x%lx vm->max_gfn: 0x%lx vm->page_size: 0x%x",
 		    paddr, vm->max_gfn, vm->page_size);
+	TEST_ASSERT(addr_raw2gpa(vm, paddr) == paddr,
+		    "Unexpected bits in paddr: %lx", paddr);
 
 	/*
 	 * Allocate upper level page tables, if not already present.  Return
@@ -206,7 +210,8 @@ void __virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr, int level)
 	pte = virt_get_pte(vm, PTE_GET_PFN(*pde), vaddr, PG_LEVEL_4K);
 	TEST_ASSERT(!(*pte & PTE_PRESENT_MASK),
 		    "PTE already present for 4k page at vaddr: 0x%lx\n", vaddr);
-	*pte = PTE_PRESENT_MASK | PTE_WRITABLE_MASK | (paddr & PHYSICAL_PAGE_MASK);
+	*pte = PTE_PRESENT_MASK | PTE_WRITABLE_MASK |
+	       (addr_gpa2raw(vm, paddr) & PHYSICAL_PAGE_MASK);
 }
 
 void virt_arch_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr)
@@ -515,7 +520,7 @@ vm_paddr_t addr_arch_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva)
 	if (!(pte[index[0]] & PTE_PRESENT_MASK))
 		goto unmapped_gva;
 
-	return (PTE_GET_PFN(pte[index[0]]) * vm->page_size) + (gva & ~PAGE_MASK);
+	return addr_raw2gpa(vm, PTE_GET_PFN(pte[index[0]]) * vm->page_size) + (gva & ~PAGE_MASK);
 
 unmapped_gva:
 	TEST_FAIL("No mapping for vm virtual address, gva: 0x%lx", gva);
Signature · 1 line
2.37.2.672.g94769d06f0-goog
#8 lore raw

Re: [V4 4/8] KVM: selftests: handle encryption bits in page tables

Quoted 16 lines · Peter Gonda
tools/testing/selftests/kvm/lib/kvm_util.c +8 -0
index 53b9a509c1d5..de13be62d52d 100644
@@ -1388,6 +1388,58 @@ void virt_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr,
 	}
 }
 
+/*
+ * Mask off any special bits from raw GPA
+ *
+ * Input Args:
+ *   vm - Virtual Machine
+ *   gpa_raw - Raw VM physical address
+ *
+ * Output Args: None

tools/testing/selftests/kvm/lib/kvm_util.c void virt_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr, · line 1408

Quoted 10 lines
tools/testing/selftests/kvm/lib/kvm_util.c +10 -0
+ *
+ * Return:
+ *   GPA with special bits (e.g. shared/encrypted) masked off.
+ */
+vm_paddr_t addr_raw2gpa(struct kvm_vm *vm, vm_paddr_t gpa_raw)
+{
+	if (!vm->memcrypt.has_enc_bit)
+		return gpa_raw;
+
+	return gpa_raw & ~(1ULL << vm->memcrypt.enc_bit);

1. The notion of stealing GPA bits to tag the page should not be tied to memory encryption.

2. Assuming that the shared vs. private bit is active-high is wrong, e.g. TDX has active-low behavior.

3. "raw" is not super untuitive.

4. addr_gpa2raw() should not exist. It assumes the "raw" address always wants the encryption bit set.

5. enc_by_default is an SEV-centric flag that should not exist outside of x86.

I think the easiest and most familiar solution for #1 will be to borrow the kernel's tag/untag terminology for handling virtual addresses that can be tagged for ASAN, e.g. ARM's top-byte-ignore and x86's linear address masking (LAM).

So instead of addr_raw2gpa(), just do:

static inline vm_paddr_t vm_untag_gpa(struct kvm_vm *vm, vm_vaddr_t gpa)

  {
	return gpa & ~vm->gpa_tag_mask;

}

That way zero-allocating the VM will Just Work.

tools/testing/selftests/kvm/lib/x86_64/processor.c void virt_arch_pgd_alloc(struct kvm_vm *vm) · line 121

Quoted 10 lines
tools/testing/selftests/kvm/lib/x86_64/processor.c +1 -1
index 2e6e61bbe81b..b2df259ce706 100644
@@ -118,7 +118,7 @@ void virt_arch_pgd_alloc(struct kvm_vm *vm)
 
 	/* If needed, create page map l4 table. */
 	if (!vm->pgd_created) {
-		vm->pgd = vm_alloc_page_table(vm);
+		vm->pgd = addr_gpa2raw(vm, vm_alloc_page_table(vm));

Rather than add "struct vm_memcrypt", I think it makes sense to introduce "struct kvm_vm_arch" and then the x86 implementation can add pte_me_mask, similar to what KVM does for SPTEs. THen this code because

		vm->pgd = vm_alloc_page_table(vm) | vm->arch.pte_me_mask;

That will Just Work for TDX, because its pte_me_mask will be '0', even though it effectively has an encryption bit (active-low).

Quoted 7 lines
tools/testing/selftests/kvm/lib/x86_64/processor.c +0 -0
 		vm->pgd_created = true;
 	}
 }
@@ -140,13 +140,15 @@ static uint64_t *virt_create_upper_pte(struct kvm_vm *vm,
 				       int target_level)
 {
 	uint64_t *pte = virt_get_pte(vm, pt_pfn, vaddr, current_level);

tools/testing/selftests/kvm/lib/x86_64/processor.c static uint64_t *virt_create_upper_pte(struct kvm_vm *vm, · line 150

Quoted 10 lines
tools/testing/selftests/kvm/lib/x86_64/processor.c +3 -2
+	uint64_t paddr_raw = addr_gpa2raw(vm, paddr);
 
 	if (!(*pte & PTE_PRESENT_MASK)) {
 		*pte = PTE_PRESENT_MASK | PTE_WRITABLE_MASK;
 		if (current_level == target_level)
-			*pte |= PTE_LARGE_MASK | (paddr & PHYSICAL_PAGE_MASK);
+			*pte |= PTE_LARGE_MASK | (paddr_raw & PHYSICAL_PAGE_MASK);
 		else
-			*pte |= vm_alloc_page_table(vm) & PHYSICAL_PAGE_MASK;
+			*pte |= addr_gpa2raw(vm, vm_alloc_page_table(vm)) & PHYSICAL_PAGE_MASK;

Prefer to write this as:

			*pte |= vm_alloc_page_table(vm) & PHYSICAL_PAGE_MASK;
			*pte |= vm->arch.pte_me_mask;

so that selftests don't assume the encryption bit is stolen from the GPA.

Quoted 14 lines
tools/testing/selftests/kvm/lib/x86_64/processor.c +3 -0
+
 	} else {
 		/*
 		 * Entry already present.  Assert that the caller doesn't want
@@ -184,6 +186,8 @@ void __virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr, int level)
 		    "Physical address beyond maximum supported,\n"
 		    "  paddr: 0x%lx vm->max_gfn: 0x%lx vm->page_size: 0x%x",
 		    paddr, vm->max_gfn, vm->page_size);
+	TEST_ASSERT(addr_raw2gpa(vm, paddr) == paddr,
+		    "Unexpected bits in paddr: %lx", paddr);
 
 	/*
 	 * Allocate upper level page tables, if not already present.  Return

tools/testing/selftests/kvm/lib/x86_64/processor.c void __virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr, int level) · line 214

Quoted 6 lines
tools/testing/selftests/kvm/lib/x86_64/processor.c +2 -1
@@ -206,7 +210,8 @@ void __virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr, int level)
 	pte = virt_get_pte(vm, PTE_GET_PFN(*pde), vaddr, PG_LEVEL_4K);
 	TEST_ASSERT(!(*pte & PTE_PRESENT_MASK),
 		    "PTE already present for 4k page at vaddr: 0x%lx\n", vaddr);
-	*pte = PTE_PRESENT_MASK | PTE_WRITABLE_MASK | (paddr & PHYSICAL_PAGE_MASK);
+	*pte = PTE_PRESENT_MASK | PTE_WRITABLE_MASK |
+	       (addr_gpa2raw(vm, paddr) & PHYSICAL_PAGE_MASK);

And with the above suggestions, this can become something like:

	if (vm_is_gpa_encrypted(vm, paddr))
		*pte |= vm->arch.c_bit;
	else
		*pte |= vm->arch.s_bit;

where SEV sets c_bit and TDX sets s_bit.

tools/testing/selftests/kvm/lib/x86_64/processor.c vm_paddr_t addr_arch_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva) · line 523

Quoted 7 lines
tools/testing/selftests/kvm/lib/x86_64/processor.c +1 -1
 void virt_arch_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr)
@@ -515,7 +520,7 @@ vm_paddr_t addr_arch_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva)
 	if (!(pte[index[0]] & PTE_PRESENT_MASK))
 		goto unmapped_gva;
 
-	return (PTE_GET_PFN(pte[index[0]]) * vm->page_size) + (gva & ~PAGE_MASK);
+	return addr_raw2gpa(vm, PTE_GET_PFN(pte[index[0]]) * vm->page_size) + (gva & ~PAGE_MASK);

Aha! A use for my rework[*] of this mess! I don't think you need to take a dependency on that work, at a glance the conflicts should be minor, i.e. doesn't matter that much which lands first.

[*] https://lore.kernel.org/all/[email protected]

#9 lore raw

PATCH v4 5/8 KVM: selftests: add support for encrypted vm_vaddr_* allocations

In reply to Peter Gonda

From: Michael Roth <[email protected]>

The default policy for whether to handle allocations as encrypted or shared pages is currently determined by vm_phy_pages_alloc(), which in turn uses the policy defined by vm->memcrypt.enc_by_default.

Test programs may wish to allocate shared vaddrs for things like sharing memory with the guest. Since enc_by_default will be true in the case of SEV guests (since it's required in order to have the initial ELF binary and page table become part of the initial guest payload), an interface is needed to explicitly request shared pages.

Implement this by splitting the common code out from vm_vaddr_alloc() and introducing a new vm_vaddr_alloc_shared().

---
 .../selftests/kvm/include/kvm_util_base.h     |  1 +
 tools/testing/selftests/kvm/lib/kvm_util.c    | 21 +++++++++++++++----
 2 files changed, 18 insertions(+), 4 deletions(-)
tools/testing/selftests/kvm/include/kvm_util_base.h +1 -0
index dda8467d1434..489e8c833e5f 100644
@@ -392,6 +392,7 @@ void vm_mem_region_move(struct kvm_vm *vm, uint32_t slot, uint64_t new_gpa);
 void vm_mem_region_delete(struct kvm_vm *vm, uint32_t slot);
 struct kvm_vcpu *__vm_vcpu_add(struct kvm_vm *vm, uint32_t vcpu_id);
 vm_vaddr_t vm_vaddr_alloc(struct kvm_vm *vm, size_t sz, vm_vaddr_t vaddr_min);
+vm_vaddr_t vm_vaddr_alloc_shared(struct kvm_vm *vm, size_t sz, vm_vaddr_t vaddr_min);
 vm_vaddr_t vm_vaddr_alloc_pages(struct kvm_vm *vm, int nr_pages);
 vm_vaddr_t vm_vaddr_alloc_page(struct kvm_vm *vm);
 
tools/testing/selftests/kvm/lib/kvm_util.c +17 -4
index de13be62d52d..ffdf39a5b12d 100644
@@ -1273,12 +1273,13 @@ static vm_vaddr_t vm_vaddr_unused_gap(struct kvm_vm *vm, size_t sz,
 }
 
 /*
- * VM Virtual Address Allocate
+ * VM Virtual Address Allocate Shared/Encrypted
  *
  * Input Args:
  *   vm - Virtual Machine
  *   sz - Size in bytes
  *   vaddr_min - Minimum starting virtual address
+ *   encrypt - Whether the region should be handled as encrypted
  *
  * Output Args: None
  *
@@ -1291,13 +1292,15 @@ static vm_vaddr_t vm_vaddr_unused_gap(struct kvm_vm *vm, size_t sz,
  * a unique set of pages, with the minimum real allocation being at least
  * a page.
  */
-vm_vaddr_t vm_vaddr_alloc(struct kvm_vm *vm, size_t sz, vm_vaddr_t vaddr_min)
+static vm_vaddr_t
+_vm_vaddr_alloc(struct kvm_vm *vm, size_t sz, vm_vaddr_t vaddr_min, bool encrypt)
 {
 	uint64_t pages = (sz >> vm->page_shift) + ((sz % vm->page_size) != 0);
 
 	virt_pgd_alloc(vm);
-	vm_paddr_t paddr = vm_phy_pages_alloc(vm, pages,
-					      KVM_UTIL_MIN_PFN * vm->page_size, 0);
+	vm_paddr_t paddr = _vm_phy_pages_alloc(vm, pages,
+					       KVM_UTIL_MIN_PFN * vm->page_size,
+					       0, encrypt);
 
 	/*
 	 * Find an unused range of virtual page addresses of at least
@@ -1318,6 +1321,16 @@ vm_vaddr_t vm_vaddr_alloc(struct kvm_vm *vm, size_t sz, vm_vaddr_t vaddr_min)
 	return vaddr_start;
 }
 
+vm_vaddr_t vm_vaddr_alloc(struct kvm_vm *vm, size_t sz, vm_vaddr_t vaddr_min)
+{
+	return _vm_vaddr_alloc(vm, sz, vaddr_min, vm->memcrypt.enc_by_default);
+}
+
+vm_vaddr_t vm_vaddr_alloc_shared(struct kvm_vm *vm, size_t sz, vm_vaddr_t vaddr_min)
+{
+	return _vm_vaddr_alloc(vm, sz, vaddr_min, false);
+}
+
 /*
  * VM Virtual Address Allocate Pages
  *
Signature · 1 line
2.37.2.672.g94769d06f0-goog
#10 lore raw

PATCH v4 6/8 KVM: selftests: add library for creating/interacting with SEV guests

In reply to Peter Gonda

Add interfaces to allow tests to create/manage SEV guests. The additional state associated with these guests is encapsulated in a new struct sev_vm, which is a light wrapper around struct kvm_vm. These VMs will use vm_set_memory_encryption() and vm_get_encrypted_phy_pages() under the covers to configure and sync up with the core kvm_util library on what should/shouldn't be treated as encrypted memory.

---
 tools/testing/selftests/kvm/Makefile          |   1 +
 .../selftests/kvm/include/kvm_util_base.h     |   3 +
 .../selftests/kvm/include/x86_64/sev.h        |  47 ++++
 tools/testing/selftests/kvm/lib/x86_64/sev.c  | 232 ++++++++++++++++++
 4 files changed, 283 insertions(+)
 create mode 100644 tools/testing/selftests/kvm/include/x86_64/sev.h
 create mode 100644 tools/testing/selftests/kvm/lib/x86_64/sev.c
tools/testing/selftests/kvm/Makefile +1 -0
index 23649c5d42fc..0a70e50f0498 100644
@@ -56,6 +56,7 @@ LIBKVM_x86_64 += lib/x86_64/processor.c
 LIBKVM_x86_64 += lib/x86_64/svm.c
 LIBKVM_x86_64 += lib/x86_64/ucall.c
 LIBKVM_x86_64 += lib/x86_64/vmx.c
+LIBKVM_x86_64 += lib/x86_64/sev.c
 
 LIBKVM_aarch64 += lib/aarch64/gic.c
 LIBKVM_aarch64 += lib/aarch64/gic_v3.c
tools/testing/selftests/kvm/include/kvm_util_base.h +3 -0
index 489e8c833e5f..0927e262623d 100644
@@ -69,6 +69,7 @@ struct userspace_mem_regions {
 /* Memory encryption policy/configuration. */
 struct vm_memcrypt {
 	bool enabled;
+	bool encrypted;
 	int8_t enc_by_default;
 	bool has_enc_bit;
 	int8_t enc_bit;
@@ -831,6 +832,8 @@ vm_paddr_t addr_arch_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva);
 
 static inline vm_paddr_t addr_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva)
 {
+	TEST_ASSERT(!vm->memcrypt.encrypted,
+		    "Encrypted guests have their page tables encrypted so gva2gpa conversions are not possible.");
 	return addr_arch_gva2gpa(vm, gva);
 }
 
tools/testing/selftests/kvm/include/x86_64/sev.h new file +47 -0

47 lines not on this page · raw message

tools/testing/selftests/kvm/lib/x86_64/sev.c new file +232 -0

232 lines not on this page · raw message

Signature · 1 line
2.37.2.672.g94769d06f0-goog
#11 lore raw

Re: [V4 6/8] KVM: selftests: add library for creating/interacting with SEV guests

Quoted 43 lines · Peter Gonda

Add interfaces to allow tests to create/manage SEV guests. The additional state associated with these guests is encapsulated in a new struct sev_vm, which is a light wrapper around struct kvm_vm. These VMs will use vm_set_memory_encryption() and vm_get_encrypted_phy_pages() under the covers to configure and sync up with the core kvm_util library on what should/shouldn't be treated as encrypted memory.

---
 tools/testing/selftests/kvm/Makefile          |   1 +
 .../selftests/kvm/include/kvm_util_base.h     |   3 +
 .../selftests/kvm/include/x86_64/sev.h        |  47 ++++
 tools/testing/selftests/kvm/lib/x86_64/sev.c  | 232 ++++++++++++++++++
 4 files changed, 283 insertions(+)
 create mode 100644 tools/testing/selftests/kvm/include/x86_64/sev.h
 create mode 100644 tools/testing/selftests/kvm/lib/x86_64/sev.c
tools/testing/selftests/kvm/Makefile +1 -0
index 23649c5d42fc..0a70e50f0498 100644
@@ -56,6 +56,7 @@ LIBKVM_x86_64 += lib/x86_64/processor.c
 LIBKVM_x86_64 += lib/x86_64/svm.c
 LIBKVM_x86_64 += lib/x86_64/ucall.c
 LIBKVM_x86_64 += lib/x86_64/vmx.c
+LIBKVM_x86_64 += lib/x86_64/sev.c
 
 LIBKVM_aarch64 += lib/aarch64/gic.c
 LIBKVM_aarch64 += lib/aarch64/gic_v3.c
tools/testing/selftests/kvm/include/kvm_util_base.h +1 -0
index 489e8c833e5f..0927e262623d 100644
@@ -69,6 +69,7 @@ struct userspace_mem_regions {
 /* Memory encryption policy/configuration. */
 struct vm_memcrypt {
 	bool enabled;
+	bool encrypted;
 	int8_t enc_by_default;
 	bool has_enc_bit;
 	int8_t enc_bit;

tools/testing/selftests/kvm/include/kvm_util_base.h vm_paddr_t addr_arch_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva); · line 835

Quoted 4 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +1 -0
@@ -831,6 +832,8 @@ vm_paddr_t addr_arch_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva);
 
 static inline vm_paddr_t addr_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva)
 {
+	TEST_ASSERT(!vm->memcrypt.encrypted,

vm->protected

tools/testing/selftests/kvm/include/kvm_util_base.h vm_paddr_t addr_arch_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva); · line 836

Quoted 1 line
tools/testing/selftests/kvm/include/kvm_util_base.h +1 -0
+		    "Encrypted guests have their page tables encrypted so gva2gpa conversions are not possible.");

Unnecessarily verbose, e.g.

		    "Protected VMs have private, inaccessible page tables");

tools/testing/selftests/kvm/include/kvm_util_base.h vm_paddr_t addr_arch_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva);

Quoted 4 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +4 -0
+#define CPUID_MEM_ENC_LEAF 0x8000001f
+#define CPUID_EBX_CBIT_MASK 0x3f
+
+/* Common SEV helpers/accessors. */

Please drop this comment and the "Local helpers" and "SEV VM implementation" comments below. There's 0% chance these comments will stay fresh as code is added and moved around. They also add no value IMO, e.g. "static" makes it quite obvious it's a local function, and sev_* vs. sev_es_*. vs. sev_snp_* namespacing takes care of the rest.

Quoted 58 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +58 -0
+void sev_ioctl(int sev_fd, int cmd, void *data)
+{
+	int ret;
+	struct sev_issue_cmd arg;
+
+	arg.cmd = cmd;
+	arg.data = (unsigned long)data;
+	ret = ioctl(sev_fd, SEV_ISSUE_CMD, &arg);
+	TEST_ASSERT(ret == 0,
+		    "SEV ioctl %d failed, error: %d, fw_error: %d",
+		    cmd, ret, arg.error);
+}
+
+void kvm_sev_ioctl(struct sev_vm *sev, int cmd, void *data)
+{
+	struct kvm_sev_cmd arg = {0};
+	int ret;
+
+	arg.id = cmd;
+	arg.sev_fd = sev->fd;
+	arg.data = (__u64)data;
+
+	ret = ioctl(sev->vm->fd, KVM_MEMORY_ENCRYPT_OP, &arg);
+	TEST_ASSERT(ret == 0,
+		    "SEV KVM ioctl %d failed, rc: %i errno: %i (%s), fw_error: %d",
+		    cmd, ret, errno, strerror(errno), arg.error);
+}
+
+/* Local helpers. */
+
+static void sev_register_user_region(struct sev_vm *sev, void *hva, uint64_t size)
+{
+	struct kvm_enc_region range = {0};
+	int ret;
+
+	pr_debug("%s: hva: %p, size: %lu\n", __func__, hva, size);
+
+	range.addr = (__u64)hva;
+	range.size = size;
+
+	ret = ioctl(sev->vm->fd, KVM_MEMORY_ENCRYPT_REG_REGION, &range);
+	TEST_ASSERT(ret == 0, "failed to register user range, errno: %i\n", errno);
+}
+
+static void sev_encrypt_phy_range(struct sev_vm *sev, vm_paddr_t gpa, uint64_t size)
+{
+	struct kvm_sev_launch_update_data ksev_update_data = {0};
+
+	pr_debug("%s: addr: 0x%lx, size: %lu\n", __func__, gpa, size);
+
+	ksev_update_data.uaddr = (__u64)addr_gpa2hva(sev->vm, gpa);
+	ksev_update_data.len = size;
+
+	kvm_sev_ioctl(sev, KVM_SEV_LAUNCH_UPDATE_DATA, &ksev_update_data);
+}
+
+static void sev_encrypt(struct sev_vm *sev)
+{

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 10 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +10 -0
+	const struct sparsebit *enc_phy_pages;
+	struct kvm_vm *vm = sev->vm;
+	sparsebit_idx_t pg = 0;
+	vm_paddr_t gpa_start;
+	uint64_t memory_size;
+	int ctr;
+	struct userspace_mem_region *region;
+
+	hash_for_each(vm->regions.slot_hash, ctr, region, slot_node) {
+		enc_phy_pages = vm_get_encrypted_phy_pages(

Please don't wrap after the opening paranthesis unless it's really, really necessary. More for future reference since I think vm_get_encrypted_phy_pages() should be open coded here. E.g. in this case, the "enc_phy_" prefix doesn't add much value, and dropping that makes the code easier to read overall.

		pages = vm_get_encrypted_phy_pages(sev->vm, region->region.slot,
						   &gpa_start, &memory_size);

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 5 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +5 -0
+			sev->vm, region->region.slot, &gpa_start, &memory_size);
+		TEST_ASSERT(enc_phy_pages,
+			    "Unable to retrieve encrypted pages bitmap");
+		while (pg < (memory_size / vm->page_size)) {
+			sparsebit_idx_t pg_cnt;

s/pg_cnt/nr_pages

Quoted 25 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +25 -0
+
+			if (sparsebit_is_clear(enc_phy_pages, pg)) {
+				pg = sparsebit_next_set(enc_phy_pages, pg);
+				if (!pg)
+					break;
+			}
+
+			pg_cnt = sparsebit_next_clear(enc_phy_pages, pg) - pg;
+			if (pg_cnt <= 0)
+				pg_cnt = 1;
+
+			sev_encrypt_phy_range(sev,
+					      gpa_start + pg * vm->page_size,
+					      pg_cnt * vm->page_size);
+			pg += pg_cnt;
+		}
+	}
+
+	sev->vm->memcrypt.encrypted = true;
+}
+
+/* SEV VM implementation. */
+
+static struct sev_vm *sev_vm_alloc(struct kvm_vm *vm)
+{

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 10 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +10 -0
+	struct sev_user_data_status sev_status;
+	uint32_t eax, ebx, ecx, edx;
+	struct sev_vm *sev;
+	int sev_fd;
+
+	sev_fd = open(SEV_DEV_PATH, O_RDWR);
+	if (sev_fd < 0) {
+		pr_info("Failed to open SEV device, path: %s, error: %d, skipping test.\n",
+			SEV_DEV_PATH, sev_fd);
+		return NULL;

Printing "skipping test" is wrong as there's no guarantee the caller is going to skip the test. E.g. the sole user in this series asserts, i.e. fails the test.

I also think that waiting until VM allocation to perform these sanity checks is flawed. Rather do these checks every time, add helpers to query SEV and SEV-ES support, and then use TEST_REQUIRE() to actually skip tests that require support, e.g.

	TEST_REQUIRE(kvm_is_sev_supported());

or

	TEST_REQUIRE(kvm_is_sev_es_supported());

Then this helper can simply assert that opening SEV_DEV_PATH succeeds.

Quoted 4 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +4 -0
+	}
+
+	sev_ioctl(sev_fd, SEV_PLATFORM_STATUS, &sev_status);
+

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 10 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +10 -0
+	if (!(sev_status.api_major > SEV_FW_REQ_VER_MAJOR ||
+	      (sev_status.api_major == SEV_FW_REQ_VER_MAJOR &&
+	       sev_status.api_minor >= SEV_FW_REQ_VER_MINOR))) {
+		pr_info("SEV FW version too old. Have API %d.%d (build: %d), need %d.%d, skipping test.\n",
+			sev_status.api_major, sev_status.api_minor, sev_status.build,
+			SEV_FW_REQ_VER_MAJOR, SEV_FW_REQ_VER_MINOR);
+		return NULL;
+	}
+
+	sev = calloc(1, sizeof(*sev));

TEST_ASSERT(sev, ...)

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 6 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +6 -0
+	sev->fd = sev_fd;
+	sev->vm = vm;
+
+	/* Get encryption bit via CPUID. */
+	cpuid(CPUID_MEM_ENC_LEAF, &eax, &ebx, &ecx, &edx);
+	sev->enc_bit = ebx & CPUID_EBX_CBIT_MASK;

Oh hey, another series of mine[*] that you can leverage :-)

[*] https://lore.kernel.org/all/[email protected]

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 12 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +12 -0
+
+	return sev;
+}
+
+void sev_vm_free(struct sev_vm *sev)
+{
+	kvm_vm_free(sev->vm);
+	close(sev->fd);
+	free(sev);
+}
+
+struct sev_vm *sev_vm_create(uint32_t policy, uint64_t npages)

The migration test already defines sev_vm_create(). That conflict needs to be resolved.

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 6 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +6 -0
+{
+	struct sev_vm *sev;
+	struct kvm_vm *vm;
+
+	/* Need to handle memslots after init, and after setting memcrypt. */
+	vm = vm_create_barebones();

Do not use vm_create_barebones(). That API is only to be used for tests that do not intend to run vCPUs.

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 8 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +8 -0
+	sev = sev_vm_alloc(vm);
+	if (!sev)
+		return NULL;
+	sev->sev_policy = policy;
+
+	kvm_sev_ioctl(sev, KVM_SEV_INIT, NULL);
+
+	vm->vpages_mapped = sparsebit_alloc();

This is unnecessary and leaks memory, vm->vpages_mapped is allocated by ____vm_create().

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 5 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +5 -0
+	vm_set_memory_encryption(vm, true, true, sev->enc_bit);
+	pr_info("SEV cbit: %d\n", sev->enc_bit);
+	vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS, 0, 0, npages, 0);
+	sev_register_user_region(sev, addr_gpa2hva(vm, 0),
+				 npages * vm->page_size);

Burying sev_register_user_region() in here is not going to be maintainble. I think the best away to handle this is to add an arch hook in vm_userspace_mem_region_add() and automatically register regions when they're created.

And with that, I believe sev_vm_create() can go away entirely and the SEV encryption stuff can be handled via a new vm_guest_mode. ____vm_create() already has a gross __x86_64__ hook that we can tweak, e.g.

tools/testing/selftests/kvm/lib/kvm_util.c +2 -1
index 54b8d8825f5d..2d6cbca2c01a 100644
@@ -238,9 +238,10 @@ struct kvm_vm *____vm_create(enum vm_guest_mode mode, uint64_t nr_pages)
        case VM_MODE_P36V47_16K:
                vm->pgtable_levels = 3;
                break;
+       case VM_MODE_PXXV48_4K_SEV:
        case VM_MODE_PXXV48_4K:
 #ifdef __x86_64__
-               kvm_get_cpu_address_width(&vm->pa_bits, &vm->va_bits);
+               kvm_init_vm_address_properties(vm);
                /*
                 * Ignore KVM support for 5-level paging (vm->va_bits == 57),
                 * it doesn't take effect unless a CR4.LA57 is set, which it

Then kvm_init_vm_address_properties() can pivot on vm->mode to deal with SEV specific stuff.

Quoted 44 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +44 -0
+
+	pr_info("SEV guest created, policy: 0x%x, size: %lu KB\n",
+		sev->sev_policy, npages * vm->page_size / 1024);
+
+	return sev;
+}
+
+void sev_vm_launch(struct sev_vm *sev)
+{
+	struct kvm_sev_launch_start ksev_launch_start = {0};
+	struct kvm_sev_guest_status ksev_status;
+
+	ksev_launch_start.policy = sev->sev_policy;
+	kvm_sev_ioctl(sev, KVM_SEV_LAUNCH_START, &ksev_launch_start);
+	kvm_sev_ioctl(sev, KVM_SEV_GUEST_STATUS, &ksev_status);
+	TEST_ASSERT(ksev_status.policy == sev->sev_policy, "Incorrect guest policy.");
+	TEST_ASSERT(ksev_status.state == SEV_GSTATE_LUPDATE,
+		    "Unexpected guest state: %d", ksev_status.state);
+
+	ucall_init(sev->vm, 0);
+
+	sev_encrypt(sev);
+}
+
+void sev_vm_launch_measure(struct sev_vm *sev, uint8_t *measurement)
+{
+	struct kvm_sev_launch_measure ksev_launch_measure;
+	struct kvm_sev_guest_status ksev_guest_status;
+
+	ksev_launch_measure.len = 256;
+	ksev_launch_measure.uaddr = (__u64)measurement;
+	kvm_sev_ioctl(sev, KVM_SEV_LAUNCH_MEASURE, &ksev_launch_measure);
+
+	/* Measurement causes a state transition, check that. */
+	kvm_sev_ioctl(sev, KVM_SEV_GUEST_STATUS, &ksev_guest_status);
+	TEST_ASSERT(ksev_guest_status.state == SEV_GSTATE_LSECRET,
+		    "Unexpected guest state: %d", ksev_guest_status.state);
+}
+
+void sev_vm_launch_finish(struct sev_vm *sev)
+{
+	struct kvm_sev_guest_status ksev_status;
+
+	kvm_sev_ioctl(sev, KVM_SEV_GUEST_STATUS, &ksev_status);

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 10 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +10 -0
+	TEST_ASSERT(ksev_status.state == SEV_GSTATE_LUPDATE ||
+		    ksev_status.state == SEV_GSTATE_LSECRET,
+		    "Unexpected guest state: %d", ksev_status.state);
+
+	kvm_sev_ioctl(sev, KVM_SEV_LAUNCH_FINISH, NULL);
+
+	kvm_sev_ioctl(sev, KVM_SEV_GUEST_STATUS, &ksev_status);
+	TEST_ASSERT(ksev_status.state == SEV_GSTATE_RUNNING,
+		    "Unexpected guest state: %d", ksev_status.state);
+}

Rather than force each test to invoke these via something like setup_test_common(), add the same kvm_arch_vm_post_create() hook that Vishal is likely going to add, and then automatically do all of the launch+measure+finish stuff for non-barebones VMs. That will let SEV/SEV-ES tests use __vm_create_with_vcpus() and __vm_create().

And it'd be a little gross, but I think it'd be wortwhile to add another layer to the "one_vcpu" helpers to make things even more convenient, e.g.

struct kvm_vm *____vm_create_with_one_vcpu(enum vm_guest_mode mode,
struct kvm_vcpu **vcpu, uint64_t extra_mem_pages, void *guest_code)

{
	struct kvm_vcpu *vcpus[1];
	struct kvm_vm *vm;

	vm = __vm_create_with_vcpus(mode, 1, extra_mem_pages, guest_code, vcpus);

	*vcpu = vcpus[0];
	return vm;
}

static inline struct kvm_vm *__vm_create_with_one_vcpu(struct kvm_vcpu **vcpu,
uint64_t extra_mem_pages, void *guest_code)

{
	return ____vm_create_with_one_vcpu(VM_MODE_DEFAULT, vcpu,
					   extra_mem_pages, guest_code);
}

static inline struct kvm_vm *____vm_create_with_one_vcpu(enum vm_guest_mode mode,
struct kvm_vcpu **vcpu, uint64_t extra_mem_pages, void *guest_code)
____vm_create_with_one_vcpu

tools/testing/selftests/kvm/lib/kvm_util.c +2 -3
index dafe4471a6c7..593dfadb662e 100644
@@ -298,9 +298,8 @@ struct kvm_vm *__vm_create(enum vm_guest_mode mode, uint32_t nr_runnable_vcpus,
 
        kvm_vm_elf_load(vm, program_invocation_name);
 
-#ifdef __x86_64__
-       vm_create_irqchip(vm);
-#endif
+       kvm_arch_vm_post_create(vm);
+
        return vm;
 }
 

[*] https://lore.kernel.org/all/[email protected]

#12 lore raw

Re: [V4 6/8] KVM: selftests: add library for creating/interacting with SEV guests

Quoted 418 lines · Sean Christopherson
Quoted 43 lines · Peter Gonda

Add interfaces to allow tests to create/manage SEV guests. The additional state associated with these guests is encapsulated in a new struct sev_vm, which is a light wrapper around struct kvm_vm. These VMs will use vm_set_memory_encryption() and vm_get_encrypted_phy_pages() under the covers to configure and sync up with the core kvm_util library on what should/shouldn't be treated as encrypted memory.

---
 tools/testing/selftests/kvm/Makefile          |   1 +
 .../selftests/kvm/include/kvm_util_base.h     |   3 +
 .../selftests/kvm/include/x86_64/sev.h        |  47 ++++
 tools/testing/selftests/kvm/lib/x86_64/sev.c  | 232 ++++++++++++++++++
 4 files changed, 283 insertions(+)
 create mode 100644 tools/testing/selftests/kvm/include/x86_64/sev.h
 create mode 100644 tools/testing/selftests/kvm/lib/x86_64/sev.c
tools/testing/selftests/kvm/Makefile +1 -0
index 23649c5d42fc..0a70e50f0498 100644
@@ -56,6 +56,7 @@ LIBKVM_x86_64 += lib/x86_64/processor.c
 LIBKVM_x86_64 += lib/x86_64/svm.c
 LIBKVM_x86_64 += lib/x86_64/ucall.c
 LIBKVM_x86_64 += lib/x86_64/vmx.c
+LIBKVM_x86_64 += lib/x86_64/sev.c
 
 LIBKVM_aarch64 += lib/aarch64/gic.c
 LIBKVM_aarch64 += lib/aarch64/gic_v3.c
tools/testing/selftests/kvm/include/kvm_util_base.h +1 -0
index 489e8c833e5f..0927e262623d 100644
@@ -69,6 +69,7 @@ struct userspace_mem_regions {
 /* Memory encryption policy/configuration. */
 struct vm_memcrypt {
      bool enabled;
+     bool encrypted;
      int8_t enc_by_default;
      bool has_enc_bit;
      int8_t enc_bit;

tools/testing/selftests/kvm/include/kvm_util_base.h vm_paddr_t addr_arch_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva); · line 835

Quoted 4 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +1 -0
@@ -831,6 +832,8 @@ vm_paddr_t addr_arch_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva);
 
 static inline vm_paddr_t addr_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva)
 {
+     TEST_ASSERT(!vm->memcrypt.encrypted,

vm->protected

tools/testing/selftests/kvm/include/kvm_util_base.h vm_paddr_t addr_arch_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva); · line 836

Quoted 1 line
tools/testing/selftests/kvm/include/kvm_util_base.h +1 -0
+                 "Encrypted guests have their page tables encrypted so gva2gpa conversions are not possible.");

Unnecessarily verbose, e.g.

                    "Protected VMs have private, inaccessible page tables");

tools/testing/selftests/kvm/include/kvm_util_base.h vm_paddr_t addr_arch_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva);

Quoted 4 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +4 -0
+#define CPUID_MEM_ENC_LEAF 0x8000001f
+#define CPUID_EBX_CBIT_MASK 0x3f
+
+/* Common SEV helpers/accessors. */

Please drop this comment and the "Local helpers" and "SEV VM implementation" comments below. There's 0% chance these comments will stay fresh as code is added and moved around. They also add no value IMO, e.g. "static" makes it quite obvious it's a local function, and sev_* vs. sev_es_*. vs. sev_snp_* namespacing takes care of the rest.

Quoted 58 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +58 -0
+void sev_ioctl(int sev_fd, int cmd, void *data)
+{
+     int ret;
+     struct sev_issue_cmd arg;
+
+     arg.cmd = cmd;
+     arg.data = (unsigned long)data;
+     ret = ioctl(sev_fd, SEV_ISSUE_CMD, &arg);
+     TEST_ASSERT(ret == 0,
+                 "SEV ioctl %d failed, error: %d, fw_error: %d",
+                 cmd, ret, arg.error);
+}
+
+void kvm_sev_ioctl(struct sev_vm *sev, int cmd, void *data)
+{
+     struct kvm_sev_cmd arg = {0};
+     int ret;
+
+     arg.id = cmd;
+     arg.sev_fd = sev->fd;
+     arg.data = (__u64)data;
+
+     ret = ioctl(sev->vm->fd, KVM_MEMORY_ENCRYPT_OP, &arg);
+     TEST_ASSERT(ret == 0,
+                 "SEV KVM ioctl %d failed, rc: %i errno: %i (%s), fw_error: %d",
+                 cmd, ret, errno, strerror(errno), arg.error);
+}
+
+/* Local helpers. */
+
+static void sev_register_user_region(struct sev_vm *sev, void *hva, uint64_t size)
+{
+     struct kvm_enc_region range = {0};
+     int ret;
+
+     pr_debug("%s: hva: %p, size: %lu\n", __func__, hva, size);
+
+     range.addr = (__u64)hva;
+     range.size = size;
+
+     ret = ioctl(sev->vm->fd, KVM_MEMORY_ENCRYPT_REG_REGION, &range);
+     TEST_ASSERT(ret == 0, "failed to register user range, errno: %i\n", errno);
+}
+
+static void sev_encrypt_phy_range(struct sev_vm *sev, vm_paddr_t gpa, uint64_t size)
+{
+     struct kvm_sev_launch_update_data ksev_update_data = {0};
+
+     pr_debug("%s: addr: 0x%lx, size: %lu\n", __func__, gpa, size);
+
+     ksev_update_data.uaddr = (__u64)addr_gpa2hva(sev->vm, gpa);
+     ksev_update_data.len = size;
+
+     kvm_sev_ioctl(sev, KVM_SEV_LAUNCH_UPDATE_DATA, &ksev_update_data);
+}
+
+static void sev_encrypt(struct sev_vm *sev)
+{

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 10 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +10 -0
+     const struct sparsebit *enc_phy_pages;
+     struct kvm_vm *vm = sev->vm;
+     sparsebit_idx_t pg = 0;
+     vm_paddr_t gpa_start;
+     uint64_t memory_size;
+     int ctr;
+     struct userspace_mem_region *region;
+
+     hash_for_each(vm->regions.slot_hash, ctr, region, slot_node) {
+             enc_phy_pages = vm_get_encrypted_phy_pages(

Please don't wrap after the opening paranthesis unless it's really, really necessary. More for future reference since I think vm_get_encrypted_phy_pages() should be open coded here. E.g. in this case, the "enc_phy_" prefix doesn't add much value, and dropping that makes the code easier to read overall.

                pages = vm_get_encrypted_phy_pages(sev->vm, region->region.slot,
                                                   &gpa_start, &memory_size);

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 5 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +5 -0
+                     sev->vm, region->region.slot, &gpa_start, &memory_size);
+             TEST_ASSERT(enc_phy_pages,
+                         "Unable to retrieve encrypted pages bitmap");
+             while (pg < (memory_size / vm->page_size)) {
+                     sparsebit_idx_t pg_cnt;

s/pg_cnt/nr_pages

Quoted 25 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +25 -0
+
+                     if (sparsebit_is_clear(enc_phy_pages, pg)) {
+                             pg = sparsebit_next_set(enc_phy_pages, pg);
+                             if (!pg)
+                                     break;
+                     }
+
+                     pg_cnt = sparsebit_next_clear(enc_phy_pages, pg) - pg;
+                     if (pg_cnt <= 0)
+                             pg_cnt = 1;
+
+                     sev_encrypt_phy_range(sev,
+                                           gpa_start + pg * vm->page_size,
+                                           pg_cnt * vm->page_size);
+                     pg += pg_cnt;
+             }
+     }
+
+     sev->vm->memcrypt.encrypted = true;
+}
+
+/* SEV VM implementation. */
+
+static struct sev_vm *sev_vm_alloc(struct kvm_vm *vm)
+{

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 10 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +10 -0
+     struct sev_user_data_status sev_status;
+     uint32_t eax, ebx, ecx, edx;
+     struct sev_vm *sev;
+     int sev_fd;
+
+     sev_fd = open(SEV_DEV_PATH, O_RDWR);
+     if (sev_fd < 0) {
+             pr_info("Failed to open SEV device, path: %s, error: %d, skipping test.\n",
+                     SEV_DEV_PATH, sev_fd);
+             return NULL;

Printing "skipping test" is wrong as there's no guarantee the caller is going to skip the test. E.g. the sole user in this series asserts, i.e. fails the test.

I also think that waiting until VM allocation to perform these sanity checks is flawed. Rather do these checks every time, add helpers to query SEV and SEV-ES support, and then use TEST_REQUIRE() to actually skip tests that require support, e.g.

        TEST_REQUIRE(kvm_is_sev_supported());

or

        TEST_REQUIRE(kvm_is_sev_es_supported());

Then this helper can simply assert that opening SEV_DEV_PATH succeeds.

Quoted 4 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +4 -0
+     }
+
+     sev_ioctl(sev_fd, SEV_PLATFORM_STATUS, &sev_status);
+

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 10 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +10 -0
+     if (!(sev_status.api_major > SEV_FW_REQ_VER_MAJOR ||
+           (sev_status.api_major == SEV_FW_REQ_VER_MAJOR &&
+            sev_status.api_minor >= SEV_FW_REQ_VER_MINOR))) {
+             pr_info("SEV FW version too old. Have API %d.%d (build: %d), need %d.%d, skipping test.\n",
+                     sev_status.api_major, sev_status.api_minor, sev_status.build,
+                     SEV_FW_REQ_VER_MAJOR, SEV_FW_REQ_VER_MINOR);
+             return NULL;
+     }
+
+     sev = calloc(1, sizeof(*sev));

TEST_ASSERT(sev, ...)

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 6 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +6 -0
+     sev->fd = sev_fd;
+     sev->vm = vm;
+
+     /* Get encryption bit via CPUID. */
+     cpuid(CPUID_MEM_ENC_LEAF, &eax, &ebx, &ecx, &edx);
+     sev->enc_bit = ebx & CPUID_EBX_CBIT_MASK;

Oh hey, another series of mine[*] that you can leverage :-)

[*] https://lore.kernel.org/all/[email protected]

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 12 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +12 -0
+
+     return sev;
+}
+
+void sev_vm_free(struct sev_vm *sev)
+{
+     kvm_vm_free(sev->vm);
+     close(sev->fd);
+     free(sev);
+}
+
+struct sev_vm *sev_vm_create(uint32_t policy, uint64_t npages)

The migration test already defines sev_vm_create(). That conflict needs to be resolved.

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 6 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +6 -0
+{
+     struct sev_vm *sev;
+     struct kvm_vm *vm;
+
+     /* Need to handle memslots after init, and after setting memcrypt. */
+     vm = vm_create_barebones();

Do not use vm_create_barebones(). That API is only to be used for tests that do not intend to run vCPUs.

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 8 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +8 -0
+     sev = sev_vm_alloc(vm);
+     if (!sev)
+             return NULL;
+     sev->sev_policy = policy;
+
+     kvm_sev_ioctl(sev, KVM_SEV_INIT, NULL);
+
+     vm->vpages_mapped = sparsebit_alloc();

This is unnecessary and leaks memory, vm->vpages_mapped is allocated by ____vm_create().

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 5 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +5 -0
+     vm_set_memory_encryption(vm, true, true, sev->enc_bit);
+     pr_info("SEV cbit: %d\n", sev->enc_bit);
+     vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS, 0, 0, npages, 0);
+     sev_register_user_region(sev, addr_gpa2hva(vm, 0),
+                              npages * vm->page_size);

Burying sev_register_user_region() in here is not going to be maintainble. I think the best away to handle this is to add an arch hook in vm_userspace_mem_region_add() and automatically register regions when they're created.

And with that, I believe sev_vm_create() can go away entirely and the SEV encryption stuff can be handled via a new vm_guest_mode. ____vm_create() already has a gross __x86_64__ hook that we can tweak, e.g.

tools/testing/selftests/kvm/lib/kvm_util.c +2 -1
index 54b8d8825f5d..2d6cbca2c01a 100644
@@ -238,9 +238,10 @@ struct kvm_vm *____vm_create(enum vm_guest_mode mode, uint64_t nr_pages)
        case VM_MODE_P36V47_16K:
                vm->pgtable_levels = 3;
                break;
+       case VM_MODE_PXXV48_4K_SEV:
        case VM_MODE_PXXV48_4K:
 #ifdef __x86_64__
-               kvm_get_cpu_address_width(&vm->pa_bits, &vm->va_bits);
+               kvm_init_vm_address_properties(vm);
                /*
                 * Ignore KVM support for 5-level paging (vm->va_bits == 57),
                 * it doesn't take effect unless a CR4.LA57 is set, which it

Then kvm_init_vm_address_properties() can pivot on vm->mode to deal with SEV specific stuff.

Quoted 44 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +44 -0
+
+     pr_info("SEV guest created, policy: 0x%x, size: %lu KB\n",
+             sev->sev_policy, npages * vm->page_size / 1024);
+
+     return sev;
+}
+
+void sev_vm_launch(struct sev_vm *sev)
+{
+     struct kvm_sev_launch_start ksev_launch_start = {0};
+     struct kvm_sev_guest_status ksev_status;
+
+     ksev_launch_start.policy = sev->sev_policy;
+     kvm_sev_ioctl(sev, KVM_SEV_LAUNCH_START, &ksev_launch_start);
+     kvm_sev_ioctl(sev, KVM_SEV_GUEST_STATUS, &ksev_status);
+     TEST_ASSERT(ksev_status.policy == sev->sev_policy, "Incorrect guest policy.");
+     TEST_ASSERT(ksev_status.state == SEV_GSTATE_LUPDATE,
+                 "Unexpected guest state: %d", ksev_status.state);
+
+     ucall_init(sev->vm, 0);
+
+     sev_encrypt(sev);
+}
+
+void sev_vm_launch_measure(struct sev_vm *sev, uint8_t *measurement)
+{
+     struct kvm_sev_launch_measure ksev_launch_measure;
+     struct kvm_sev_guest_status ksev_guest_status;
+
+     ksev_launch_measure.len = 256;
+     ksev_launch_measure.uaddr = (__u64)measurement;
+     kvm_sev_ioctl(sev, KVM_SEV_LAUNCH_MEASURE, &ksev_launch_measure);
+
+     /* Measurement causes a state transition, check that. */
+     kvm_sev_ioctl(sev, KVM_SEV_GUEST_STATUS, &ksev_guest_status);
+     TEST_ASSERT(ksev_guest_status.state == SEV_GSTATE_LSECRET,
+                 "Unexpected guest state: %d", ksev_guest_status.state);
+}
+
+void sev_vm_launch_finish(struct sev_vm *sev)
+{
+     struct kvm_sev_guest_status ksev_status;
+
+     kvm_sev_ioctl(sev, KVM_SEV_GUEST_STATUS, &ksev_status);

tools/testing/selftests/kvm/include/kvm_util_base.h

Quoted 10 lines
tools/testing/selftests/kvm/include/kvm_util_base.h +10 -0
+     TEST_ASSERT(ksev_status.state == SEV_GSTATE_LUPDATE ||
+                 ksev_status.state == SEV_GSTATE_LSECRET,
+                 "Unexpected guest state: %d", ksev_status.state);
+
+     kvm_sev_ioctl(sev, KVM_SEV_LAUNCH_FINISH, NULL);
+
+     kvm_sev_ioctl(sev, KVM_SEV_GUEST_STATUS, &ksev_status);
+     TEST_ASSERT(ksev_status.state == SEV_GSTATE_RUNNING,
+                 "Unexpected guest state: %d", ksev_status.state);
+}

Rather than force each test to invoke these via something like setup_test_common(), add the same kvm_arch_vm_post_create() hook that Vishal is likely going to add, and then automatically do all of the launch+measure+finish stuff for non-barebones VMs. That will let SEV/SEV-ES tests use __vm_create_with_vcpus() and __vm_create().

And it'd be a little gross, but I think it'd be wortwhile to add another layer to the "one_vcpu" helpers to make things even more convenient, e.g.

struct kvm_vm *____vm_create_with_one_vcpu(enum vm_guest_mode mode,
struct kvm_vcpu **vcpu, uint64_t extra_mem_pages, void *guest_code)

{
        struct kvm_vcpu *vcpus[1];
        struct kvm_vm *vm;

        vm = __vm_create_with_vcpus(mode, 1, extra_mem_pages, guest_code, vcpus);

        *vcpu = vcpus[0];
        return vm;
}

static inline struct kvm_vm *__vm_create_with_one_vcpu(struct kvm_vcpu **vcpu,
uint64_t extra_mem_pages, void *guest_code)
{
return ____vm_create_with_one_vcpu(VM_MODE_DEFAULT, vcpu,
extra_mem_pages, guest_code);
}

static inline struct kvm_vm *____vm_create_with_one_vcpu(enum vm_guest_mode mode,
struct kvm_vcpu **vcpu, uint64_t extra_mem_pages, void *guest_code)
____vm_create_with_one_vcpu

tools/testing/selftests/kvm/lib/kvm_util.c +2 -3
index dafe4471a6c7..593dfadb662e 100644
@@ -298,9 +298,8 @@ struct kvm_vm *__vm_create(enum vm_guest_mode mode, uint32_t nr_runnable_vcpus,
 
        kvm_vm_elf_load(vm, program_invocation_name);
 
-#ifdef __x86_64__
-       vm_create_irqchip(vm);
-#endif
+       kvm_arch_vm_post_create(vm);
+
        return vm;
 }
 
Quoted 1 line

[*] https://lore.kernel.org/all/[email protected]

This refactor sounds good, working on this with a few changes.

Instead of kvm_init_vm_address_properties() as you suggested I've added this:

@@ -272,6 +275,8 @@ struct kvm_vm *____vm_create(enum vm_guest_mode
mode, uint64_t nr_pages)

                vm->type = KVM_VM_TYPE_ARM_IPA_SIZE(vm->pa_bits);

#endif

+ kvm_init_vm_arch(vm);
+

        vm_open(vm);

        /* Limit to VA-bit canonical virtual addresses. */

And I need to put kvm_arch_vm_post_create() after the vCPUs are created because the ordering we need is: KVM_SEV_INIT -> Create vCPUS -> KVM_SEV_LAUNCH_FINISH.

#13 lore raw

Re: [V4 6/8] KVM: selftests: add library for creating/interacting with SEV guests

Quoted 24 lines · Peter Gonda
Quoted 23 lines · Sean Christopherson

And with that, I believe sev_vm_create() can go away entirely and the SEV encryption stuff can be handled via a new vm_guest_mode. ____vm_create() already has a gross __x86_64__ hook that we can tweak, e.g.

tools/testing/selftests/kvm/lib/kvm_util.c +2 -1
index 54b8d8825f5d..2d6cbca2c01a 100644
@@ -238,9 +238,10 @@ struct kvm_vm *____vm_create(enum vm_guest_mode mode, uint64_t nr_pages)
        case VM_MODE_P36V47_16K:
                vm->pgtable_levels = 3;
                break;
+       case VM_MODE_PXXV48_4K_SEV:
        case VM_MODE_PXXV48_4K:
 #ifdef __x86_64__
-               kvm_get_cpu_address_width(&vm->pa_bits, &vm->va_bits);
+               kvm_init_vm_address_properties(vm);
                /*
                 * Ignore KVM support for 5-level paging (vm->va_bits == 57),
                 * it doesn't take effect unless a CR4.LA57 is set, which it

Then kvm_init_vm_address_properties() can pivot on vm->mode to deal with SEV specific stuff.

...

Review struct kvm_vm *____vm_create(enum vm_guest_mode · line 279

Quoted 10 lines

This refactor sounds good, working on this with a few changes.

Instead of kvm_init_vm_address_properties() as you suggested I've added this:

(quoted diff) +1 -0
@@ -272,6 +275,8 @@ struct kvm_vm *____vm_create(enum vm_guest_mode
 mode, uint64_t nr_pages)
                vm->type = KVM_VM_TYPE_ARM_IPA_SIZE(vm->pa_bits);
 #endif
 
+       kvm_init_vm_arch(vm);

Why? I'm not necessarily opposed to adding kvm_init_vm_arch(), but since x86 "needs" a dedicated hook to unpack the mode, why not piggyback that one?

Quoted 8 lines
tools/testing/selftests/kvm/lib/kvm_util.c +1 -0
+
        vm_open(vm);
 
        /* Limit to VA-bit canonical virtual addresses. */

And I need to put kvm_arch_vm_post_create() after the vCPUs are created because the ordering we need is: KVM_SEV_INIT -> Create vCPUS -> KVM_SEV_LAUNCH_FINISH.

Hrm, that's annoying. Please don't use kvm_arch_vm_post_create() as the name, that's a better fit for what Vishal is doing since the "vm_post_create()" implies that it's called for "all" VM creation paths, where "all" means "everything except barebones VMs". E.g. in Vishal's series, kvm_arch_vm_post_create() can be used to drop the vm_create_irqchip() call in common code. In your case, IIUC the hook will be invoked from __vm_create_with_vcpus().

I'm a little hesitant to have an arch hook for this case since it can't be all-or-nothing (again, ignoring barebones VMs). If a "finalize" arch hook is added, then arguably tests that do __vm_create() and manually add vCPUs should call the arch hook, i.e. we'd be adding maintenance burden to tests that in all likelihood don't care about SEV and never will.

It's somewhat unfortunate, but dedicated vm_sev_create_with_one_vcpu() and and vm_sev_create_with_vcpus() wrappers is probably the least awful solution.

#14 lore raw

Re: [V4 6/8] KVM: selftests: add library for creating/interacting with SEV guests

Quoted 38 lines · Sean Christopherson
Quoted 24 lines · Peter Gonda
Quoted 23 lines · Sean Christopherson

And with that, I believe sev_vm_create() can go away entirely and the SEV encryption stuff can be handled via a new vm_guest_mode. ____vm_create() already has a gross __x86_64__ hook that we can tweak, e.g.

tools/testing/selftests/kvm/lib/kvm_util.c +2 -1
index 54b8d8825f5d..2d6cbca2c01a 100644
@@ -238,9 +238,10 @@ struct kvm_vm *____vm_create(enum vm_guest_mode mode, uint64_t nr_pages)
        case VM_MODE_P36V47_16K:
                vm->pgtable_levels = 3;
                break;
+       case VM_MODE_PXXV48_4K_SEV:
        case VM_MODE_PXXV48_4K:
 #ifdef __x86_64__
-               kvm_get_cpu_address_width(&vm->pa_bits, &vm->va_bits);
+               kvm_init_vm_address_properties(vm);
                /*
                 * Ignore KVM support for 5-level paging (vm->va_bits == 57),
                 * it doesn't take effect unless a CR4.LA57 is set, which it

Then kvm_init_vm_address_properties() can pivot on vm->mode to deal with SEV specific stuff.

...

Quoted 9 lines

This refactor sounds good, working on this with a few changes.

Instead of kvm_init_vm_address_properties() as you suggested I've added this:

(quoted diff) +0 -0
@@ -272,6 +275,8 @@ struct kvm_vm *____vm_create(enum vm_guest_mode
 mode, uint64_t nr_pages)
                vm->type = KVM_VM_TYPE_ARM_IPA_SIZE(vm->pa_bits);
 #endif
 
Quoted 5 lines

Review struct kvm_vm *____vm_create(enum vm_guest_mode · line 279

Quoted 1 line
(quoted diff) +1 -0
+       kvm_init_vm_arch(vm);

Why? I'm not necessarily opposed to adding kvm_init_vm_arch(), but since x86 "needs" a dedicated hook to unpack the mode, why not piggyback that one?

Well I since I need to do more than just kvm_init_vm_address_properties() I thought the more generic name would be better. We need to allocate kvm_vm_arch, find the c-bit, and call KVM_SEV_INIT. I can put it back in that switch case if thats better, thoughts?

Quoted 16 lines
Quoted 8 lines
tools/testing/selftests/kvm/lib/kvm_util.c +1 -0
+
        vm_open(vm);
 
        /* Limit to VA-bit canonical virtual addresses. */

And I need to put kvm_arch_vm_post_create() after the vCPUs are created because the ordering we need is: KVM_SEV_INIT -> Create vCPUS -> KVM_SEV_LAUNCH_FINISH.

Hrm, that's annoying. Please don't use kvm_arch_vm_post_create() as the name, that's a better fit for what Vishal is doing since the "vm_post_create()" implies that it's called for "all" VM creation paths, where "all" means "everything except barebones VMs". E.g. in Vishal's series, kvm_arch_vm_post_create() can be used to drop the vm_create_irqchip() call in common code. In your case, IIUC the hook will be invoked from __vm_create_with_vcpus().

Quoted 8 lines

I'm a little hesitant to have an arch hook for this case since it can't be all-or-nothing (again, ignoring barebones VMs). If a "finalize" arch hook is added, then arguably tests that do __vm_create() and manually add vCPUs should call the arch hook, i.e. we'd be adding maintenance burden to tests that in all likelihood don't care about SEV and never will.

It's somewhat unfortunate, but dedicated vm_sev_create_with_one_vcpu() and and vm_sev_create_with_vcpus() wrappers is probably the least awful solution.

Make sense. I think we can go back to your suggestion of kvm_init_vm_address_properties() above since we can now do all the KVM_SEV_* stuff. I think this means we don't need to add VM_MODE_PXXV48_4K_SEV since we can set up the c-bit from inside of vm_sev_create_*(), thoughts?

#15 lore raw

Re: [V4 6/8] KVM: selftests: add library for creating/interacting with SEV guests

Quoted 46 lines · Peter Gonda
Quoted 11 lines · Sean Christopherson
Quoted 9 lines · Peter Gonda

This refactor sounds good, working on this with a few changes.

Instead of kvm_init_vm_address_properties() as you suggested I've added this:

(quoted diff) +0 -0
@@ -272,6 +275,8 @@ struct kvm_vm *____vm_create(enum vm_guest_mode
 mode, uint64_t nr_pages)
                vm->type = KVM_VM_TYPE_ARM_IPA_SIZE(vm->pa_bits);
 #endif
 
Quoted 5 lines

Review struct kvm_vm *____vm_create(enum vm_guest_mode · line 279

Quoted 1 line
(quoted diff) +1 -0
+       kvm_init_vm_arch(vm);

Why? I'm not necessarily opposed to adding kvm_init_vm_arch(), but since x86 "needs" a dedicated hook to unpack the mode, why not piggyback that one?

Well I since I need to do more than just kvm_init_vm_address_properties() I thought the more generic name would be better. We need to allocate kvm_vm_arch, find the c-bit, and call KVM_SEV_INIT. I can put it back in that switch case if thats better, thoughts?

Quoted 22 lines
Quoted 8 lines
tools/testing/selftests/kvm/lib/kvm_util.c +1 -0
+
        vm_open(vm);
 
        /* Limit to VA-bit canonical virtual addresses. */

And I need to put kvm_arch_vm_post_create() after the vCPUs are created because the ordering we need is: KVM_SEV_INIT -> Create vCPUS -> KVM_SEV_LAUNCH_FINISH.

Hrm, that's annoying. Please don't use kvm_arch_vm_post_create() as the name, that's a better fit for what Vishal is doing since the "vm_post_create()" implies that it's called for "all" VM creation paths, where "all" means "everything except barebones VMs". E.g. in Vishal's series, kvm_arch_vm_post_create() can be used to drop the vm_create_irqchip() call in common code. In your case, IIUC the hook will be invoked from __vm_create_with_vcpus().

I'm a little hesitant to have an arch hook for this case since it can't be all-or-nothing (again, ignoring barebones VMs). If a "finalize" arch hook is added, then arguably tests that do __vm_create() and manually add vCPUs should call the arch hook, i.e. we'd be adding maintenance burden to tests that in all likelihood don't care about SEV and never will.

Quoted 8 lines
Quoted 2 lines

It's somewhat unfortunate, but dedicated vm_sev_create_with_one_vcpu() and and vm_sev_create_with_vcpus() wrappers is probably the least awful solution.

Make sense. I think we can go back to your suggestion of kvm_init_vm_address_properties() above since we can now do all the KVM_SEV_* stuff. I think this means we don't need to add VM_MODE_PXXV48_4K_SEV since we can set up the c-bit from inside of vm_sev_create_*(), thoughts?

Configuring the C-bit inside vm_sev_create_*() won't work (at least not well). The C-bit needs to be known before kvm_vm_elf_load(), i.e. can't be handled after __vm_create(), and needs to be tracked inside the VM, i.e. can't be handled before __vm_create().

The proposed kvm_init_vm_address_properties() seems like the best fit since the C-bit (and TDX's S-bit) is stolen from GPA space, i.e. directly affects the other values computed in that path.

As for the kvm_vm_arch allocation ugliness, when we talked off-list I didn't consider the need to allocate in kvm_init_vm_address_properties(). That's quite gross, especially since the pointer will be larger than the thing being allocated.

With that in mind, adding .../include/<arch>/kvm_util.h so that "struct kvm_vm_arch" can be defined and referenced directly doesn't seem so bad. Having to stub in the struct for the other architectures is annoying, but not the end of the world.

#16 lore raw

Re: [V4 6/8] KVM: selftests: add library for creating/interacting with SEV guests

Quoted 66 lines · Sean Christopherson
Quoted 46 lines · Peter Gonda
Quoted 11 lines · Sean Christopherson
Quoted 9 lines · Peter Gonda

This refactor sounds good, working on this with a few changes.

Instead of kvm_init_vm_address_properties() as you suggested I've added this:

(quoted diff) +0 -0
@@ -272,6 +275,8 @@ struct kvm_vm *____vm_create(enum vm_guest_mode
 mode, uint64_t nr_pages)
                vm->type = KVM_VM_TYPE_ARM_IPA_SIZE(vm->pa_bits);
 #endif
 
Quoted 5 lines

Review struct kvm_vm *____vm_create(enum vm_guest_mode · line 279

Quoted 1 line
(quoted diff) +1 -0
+       kvm_init_vm_arch(vm);

Why? I'm not necessarily opposed to adding kvm_init_vm_arch(), but since x86 "needs" a dedicated hook to unpack the mode, why not piggyback that one?

Well I since I need to do more than just kvm_init_vm_address_properties() I thought the more generic name would be better. We need to allocate kvm_vm_arch, find the c-bit, and call KVM_SEV_INIT. I can put it back in that switch case if thats better, thoughts?

Quoted 22 lines
Quoted 8 lines
tools/testing/selftests/kvm/lib/kvm_util.c +1 -0
+
        vm_open(vm);
 
        /* Limit to VA-bit canonical virtual addresses. */

And I need to put kvm_arch_vm_post_create() after the vCPUs are created because the ordering we need is: KVM_SEV_INIT -> Create vCPUS -> KVM_SEV_LAUNCH_FINISH.

Hrm, that's annoying. Please don't use kvm_arch_vm_post_create() as the name, that's a better fit for what Vishal is doing since the "vm_post_create()" implies that it's called for "all" VM creation paths, where "all" means "everything except barebones VMs". E.g. in Vishal's series, kvm_arch_vm_post_create() can be used to drop the vm_create_irqchip() call in common code. In your case, IIUC the hook will be invoked from __vm_create_with_vcpus().

I'm a little hesitant to have an arch hook for this case since it can't be all-or-nothing (again, ignoring barebones VMs). If a "finalize" arch hook is added, then arguably tests that do __vm_create() and manually add vCPUs should call the arch hook, i.e. we'd be adding maintenance burden to tests that in all likelihood don't care about SEV and never will.

Quoted 8 lines
Quoted 2 lines

It's somewhat unfortunate, but dedicated vm_sev_create_with_one_vcpu() and and vm_sev_create_with_vcpus() wrappers is probably the least awful solution.

Make sense. I think we can go back to your suggestion of kvm_init_vm_address_properties() above since we can now do all the KVM_SEV_* stuff. I think this means we don't need to add VM_MODE_PXXV48_4K_SEV since we can set up the c-bit from inside of vm_sev_create_*(), thoughts?

Configuring the C-bit inside vm_sev_create_*() won't work (at least not well). The C-bit needs to be known before kvm_vm_elf_load(), i.e. can't be handled after __vm_create(), and needs to be tracked inside the VM, i.e. can't be handled before __vm_create().

The proposed kvm_init_vm_address_properties() seems like the best fit since the C-bit (and TDX's S-bit) is stolen from GPA space, i.e. directly affects the other values computed in that path.

Quoted 7 lines

As for the kvm_vm_arch allocation ugliness, when we talked off-list I didn't consider the need to allocate in kvm_init_vm_address_properties(). That's quite gross, especially since the pointer will be larger than the thing being allocated.

With that in mind, adding .../include/<arch>/kvm_util.h so that "struct kvm_vm_arch" can be defined and referenced directly doesn't seem so bad. Having to stub in the struct for the other architectures is annoying, but not the end of the world.

I'll make "struct kvm_vm_arch" a non pointer member, so adding /include/<arch>/kvm_util.h files.

But I think we do not need VM_MODE_PXXV48_4K_SEV, see:

struct kvm_vm *vm_sev_create_with_one_vcpu(uint32_t policy, void *guest_code,
struct kvm_vcpu **cpu)

{
        enum vm_guest_mode mode = VM_MODE_PXXV48_4K;
        uint64_t nr_pages = vm_nr_pages_required(mode, 1, 0);
        struct kvm_vm *vm;
        uint8_t measurement[512];
        int i;

        vm = ____vm_create(mode, nr_pages);

        kvm_sev_ioctl(vm, KVM_SEV_INIT, NULL);

        configure_sev_pte_masks(vm);

        *cpu = vm_vcpu_add(vm, 0, guest_code);
        kvm_vm_elf_load(vm, program_invocation_name);

        sev_vm_launch(vm, policy);

        /* Dump the initial measurement. A test to actually verify it
would be nice. */
        sev_vm_launch_measure(vm, measurement);
        pr_info("guest measurement: ");
        for (i = 0; i < 32; ++i)
                pr_info("%02x", measurement[i]);
        pr_info("\n");

        sev_vm_launch_finish(vm);

        return vm;
}
#17 lore raw

Re: [V4 6/8] KVM: selftests: add library for creating/interacting with SEV guests

Quoted 19 lines · Peter Gonda
Quoted 18 lines · Sean Christopherson
Quoted 2 lines · Peter Gonda

I think this means we don't need to add VM_MODE_PXXV48_4K_SEV since we can set up the c-bit from inside of vm_sev_create_*(), thoughts?

Configuring the C-bit inside vm_sev_create_*() won't work (at least not well). The C-bit needs to be known before kvm_vm_elf_load(), i.e. can't be handled after __vm_create(), and needs to be tracked inside the VM, i.e. can't be handled before __vm_create().

The proposed kvm_init_vm_address_properties() seems like the best fit since the C-bit (and TDX's S-bit) is stolen from GPA space, i.e. directly affects the other values computed in that path.

As for the kvm_vm_arch allocation ugliness, when we talked off-list I didn't consider the need to allocate in kvm_init_vm_address_properties(). That's quite gross, especially since the pointer will be larger than the thing being allocated.

Quoted 8 lines
Quoted 3 lines

With that in mind, adding .../include/<arch>/kvm_util.h so that "struct kvm_vm_arch" can be defined and referenced directly doesn't seem so bad. Having to stub in the struct for the other architectures is annoying, but not the end of the world.

I'll make "struct kvm_vm_arch" a non pointer member, so adding /include/<arch>/kvm_util.h files.

But I think we do not need VM_MODE_PXXV48_4K_SEV, see:

I really don't want to open code __vm_create() with a slight tweak. E.g. the below code will be broken by Ricardo's series to add memslot0 is moved out of ____vm_create()[1], and kinda sorta be broken again by Vishal's series to add an arch hook to __vm_create()[2].

AFAICT, there is no requirement that KVM_SEV_INIT be called before computing the C-Bit, the only requirement is that KVM_SEV_INIT is called before adding vCPUs.

[1] https://lore.kernel.org/all/[email protected]
[2] https://lore.kernel.org/all/[email protected]

Quoted 32 lines

struct kvm_vm *vm_sev_create_with_one_vcpu(uint32_t policy, void *guest_code,
struct kvm_vcpu **cpu)

{
        enum vm_guest_mode mode = VM_MODE_PXXV48_4K;
        uint64_t nr_pages = vm_nr_pages_required(mode, 1, 0);
        struct kvm_vm *vm;
        uint8_t measurement[512];
        int i;

        vm = ____vm_create(mode, nr_pages);

        kvm_sev_ioctl(vm, KVM_SEV_INIT, NULL);

        configure_sev_pte_masks(vm);

        *cpu = vm_vcpu_add(vm, 0, guest_code);
        kvm_vm_elf_load(vm, program_invocation_name);

        sev_vm_launch(vm, policy);

        /* Dump the initial measurement. A test to actually verify it
would be nice. */
        sev_vm_launch_measure(vm, measurement);
        pr_info("guest measurement: ");
        for (i = 0; i < 32; ++i)
                pr_info("%02x", measurement[i]);
        pr_info("\n");

        sev_vm_launch_finish(vm);

        return vm;
}
#18 lore raw

Re: [V4 6/8] KVM: selftests: add library for creating/interacting with SEV guests

Quoted 35 lines · Sean Christopherson
Quoted 19 lines · Peter Gonda
Quoted 18 lines · Sean Christopherson
Quoted 2 lines · Peter Gonda

I think this means we don't need to add VM_MODE_PXXV48_4K_SEV since we can set up the c-bit from inside of vm_sev_create_*(), thoughts?

Configuring the C-bit inside vm_sev_create_*() won't work (at least not well). The C-bit needs to be known before kvm_vm_elf_load(), i.e. can't be handled after __vm_create(), and needs to be tracked inside the VM, i.e. can't be handled before __vm_create().

The proposed kvm_init_vm_address_properties() seems like the best fit since the C-bit (and TDX's S-bit) is stolen from GPA space, i.e. directly affects the other values computed in that path.

As for the kvm_vm_arch allocation ugliness, when we talked off-list I didn't consider the need to allocate in kvm_init_vm_address_properties(). That's quite gross, especially since the pointer will be larger than the thing being allocated.

Quoted 8 lines
Quoted 3 lines

With that in mind, adding .../include/<arch>/kvm_util.h so that "struct kvm_vm_arch" can be defined and referenced directly doesn't seem so bad. Having to stub in the struct for the other architectures is annoying, but not the end of the world.

I'll make "struct kvm_vm_arch" a non pointer member, so adding /include/<arch>/kvm_util.h files.

But I think we do not need VM_MODE_PXXV48_4K_SEV, see:

I really don't want to open code __vm_create() with a slight tweak. E.g. the below code will be broken by Ricardo's series to add memslot0 is moved out of ____vm_create()[1], and kinda sorta be broken again by Vishal's series to add an arch hook to __vm_create()[2].

Quoted 5 lines

AFAICT, there is no requirement that KVM_SEV_INIT be called before computing the C-Bit, the only requirement is that KVM_SEV_INIT is called before adding vCPUs.

[1] https://lore.kernel.org/all/[email protected]
[2] https://lore.kernel.org/all/[email protected]

Oh I misunderstood your suggestion above.

I should make KVM_SEV_INIT happen from kvm_arch_vm_post_create(). Add VM_MODE_PXXV48_4K_SEV for c-bit setting inside of kvm_init_vm_address_properties().

Inside of vm_sev_create_with_one_vcpu() I use __vm_create_with_vcpus(), then call KVM_SEV_LAUNCH_FINISH.

Is that correct?

Quoted 33 lines
Quoted 33 lines

struct kvm_vm *vm_sev_create_with_one_vcpu(uint32_t policy, void *guest_code,
struct kvm_vcpu **cpu)

{
        enum vm_guest_mode mode = VM_MODE_PXXV48_4K;
        uint64_t nr_pages = vm_nr_pages_required(mode, 1, 0);
        struct kvm_vm *vm;
        uint8_t measurement[512];
        int i;

        vm = ____vm_create(mode, nr_pages);

        kvm_sev_ioctl(vm, KVM_SEV_INIT, NULL);

        configure_sev_pte_masks(vm);

        *cpu = vm_vcpu_add(vm, 0, guest_code);
        kvm_vm_elf_load(vm, program_invocation_name);

        sev_vm_launch(vm, policy);

        /* Dump the initial measurement. A test to actually verify it
would be nice. */
        sev_vm_launch_measure(vm, measurement);
        pr_info("guest measurement: ");
        for (i = 0; i < 32; ++i)
                pr_info("%02x", measurement[i]);
        pr_info("\n");

        sev_vm_launch_finish(vm);

        return vm;
}
#19 lore raw

Re: [V4 6/8] KVM: selftests: add library for creating/interacting with SEV guests

Quoted 44 lines · Peter Gonda
Quoted 35 lines · Sean Christopherson
Quoted 19 lines · Peter Gonda
Quoted 18 lines · Sean Christopherson
Quoted 2 lines · Peter Gonda

I think this means we don't need to add VM_MODE_PXXV48_4K_SEV since we can set up the c-bit from inside of vm_sev_create_*(), thoughts?

Configuring the C-bit inside vm_sev_create_*() won't work (at least not well). The C-bit needs to be known before kvm_vm_elf_load(), i.e. can't be handled after __vm_create(), and needs to be tracked inside the VM, i.e. can't be handled before __vm_create().

The proposed kvm_init_vm_address_properties() seems like the best fit since the C-bit (and TDX's S-bit) is stolen from GPA space, i.e. directly affects the other values computed in that path.

As for the kvm_vm_arch allocation ugliness, when we talked off-list I didn't consider the need to allocate in kvm_init_vm_address_properties(). That's quite gross, especially since the pointer will be larger than the thing being allocated.

Quoted 8 lines
Quoted 3 lines

With that in mind, adding .../include/<arch>/kvm_util.h so that "struct kvm_vm_arch" can be defined and referenced directly doesn't seem so bad. Having to stub in the struct for the other architectures is annoying, but not the end of the world.

I'll make "struct kvm_vm_arch" a non pointer member, so adding /include/<arch>/kvm_util.h files.

But I think we do not need VM_MODE_PXXV48_4K_SEV, see:

I really don't want to open code __vm_create() with a slight tweak. E.g. the below code will be broken by Ricardo's series to add memslot0 is moved out of ____vm_create()[1], and kinda sorta be broken again by Vishal's series to add an arch hook to __vm_create()[2].

Quoted 5 lines

AFAICT, there is no requirement that KVM_SEV_INIT be called before computing the C-Bit, the only requirement is that KVM_SEV_INIT is called before adding vCPUs.

[1] https://lore.kernel.org/all/[email protected]
[2] https://lore.kernel.org/all/[email protected]

Oh I misunderstood your suggestion above.

Quoted 8 lines

I should make KVM_SEV_INIT happen from kvm_arch_vm_post_create(). Add VM_MODE_PXXV48_4K_SEV for c-bit setting inside of kvm_init_vm_address_properties().

Inside of vm_sev_create_with_one_vcpu() I use __vm_create_with_vcpus(), then call KVM_SEV_LAUNCH_FINISH.

Is that correct?

Yep, I'm pretty sure that was what I was thinking.

#20 lore raw

Re: [V4 6/8] KVM: selftests: add library for creating/interacting with SEV guests

Quoted 50 lines · Sean Christopherson
Quoted 48 lines · Peter Gonda
Quoted 35 lines · Sean Christopherson
Quoted 19 lines · Peter Gonda
Quoted 18 lines · Sean Christopherson
Quoted 2 lines · Peter Gonda

I think this means we don't need to add VM_MODE_PXXV48_4K_SEV since we can set up the c-bit from inside of vm_sev_create_*(), thoughts?

Configuring the C-bit inside vm_sev_create_*() won't work (at least not well). The C-bit needs to be known before kvm_vm_elf_load(), i.e. can't be handled after __vm_create(), and needs to be tracked inside the VM, i.e. can't be handled before __vm_create().

The proposed kvm_init_vm_address_properties() seems like the best fit since the C-bit (and TDX's S-bit) is stolen from GPA space, i.e. directly affects the other values computed in that path.

As for the kvm_vm_arch allocation ugliness, when we talked off-list I didn't consider the need to allocate in kvm_init_vm_address_properties(). That's quite gross, especially since the pointer will be larger than the thing being allocated.

Quoted 8 lines
Quoted 3 lines

With that in mind, adding .../include/<arch>/kvm_util.h so that "struct kvm_vm_arch" can be defined and referenced directly doesn't seem so bad. Having to stub in the struct for the other architectures is annoying, but not the end of the world.

I'll make "struct kvm_vm_arch" a non pointer member, so adding /include/<arch>/kvm_util.h files.

But I think we do not need VM_MODE_PXXV48_4K_SEV, see:

I really don't want to open code __vm_create() with a slight tweak. E.g. the below code will be broken by Ricardo's series to add memslot0 is moved out of ____vm_create()[1], and kinda sorta be broken again by Vishal's series to add an arch hook to __vm_create()[2].

Quoted 5 lines

AFAICT, there is no requirement that KVM_SEV_INIT be called before computing the C-Bit, the only requirement is that KVM_SEV_INIT is called before adding vCPUs.

[1] https://lore.kernel.org/all/[email protected]
[2] https://lore.kernel.org/all/[email protected]

Oh I misunderstood your suggestion above.

I should make KVM_SEV_INIT happen from kvm_arch_vm_post_create(). Add VM_MODE_PXXV48_4K_SEV for c-bit setting inside of kvm_init_vm_address_properties().

Quoted 6 lines
Quoted 4 lines

Inside of vm_sev_create_with_one_vcpu() I use __vm_create_with_vcpus(), then call KVM_SEV_LAUNCH_FINISH.

Is that correct?

Yep, I'm pretty sure that was what I was thinking.

Duh! Should have realized that at first. You said you were going to rebase your selftest series. I'll follow up with a V6 ontop of that rebase.

#21 lore raw

PATCH v4 7/8 KVM: selftests: Update ucall pool to allocate from shared memory

In reply to Peter Gonda

Update the per VM ucall_header allocation from vm_vaddr_alloc() to vm_vaddr_alloc_shared(). This allows encrypted guests to use ucall pools by placing their shared ucall structures in unencrypted (shared) memory. No behavior change for non encrypted guests.

---
 tools/testing/selftests/kvm/lib/ucall_common.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/testing/selftests/kvm/lib/ucall_common.c +1 -1
index cc79d497b6b4..8e164c47e2fb 100644
@@ -22,7 +22,7 @@ void ucall_init(struct kvm_vm *vm, vm_paddr_t mmio_gpa)
 	vm_vaddr_t vaddr;
 	int i;
 
-	vaddr = vm_vaddr_alloc(vm, sizeof(*hdr), KVM_UTIL_MIN_VADDR);
+	vaddr = vm_vaddr_alloc_shared(vm, sizeof(*hdr), KVM_UTIL_MIN_VADDR);
 	hdr = (struct ucall_header *)addr_gva2hva(vm, vaddr);
 	memset(hdr, 0, sizeof(*hdr));
 
Signature · 1 line
2.37.2.672.g94769d06f0-goog
#22 lore raw

PATCH v4 8/8 KVM: selftests: Add simple sev vm testing

In reply to Peter Gonda

A very simple of booting SEV guests that checks related CPUID bits. This is a stripped down version of "[PATCH v2 08/13] KVM: selftests: add SEV boot tests" from Michael but much simpler.

---
 tools/testing/selftests/kvm/.gitignore        |   1 +
 tools/testing/selftests/kvm/Makefile          |   1 +
 .../selftests/kvm/x86_64/sev_all_boot_test.c  | 127 ++++++++++++++++++
 3 files changed, 129 insertions(+)
 create mode 100644 tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c
tools/testing/selftests/kvm/.gitignore +1 -0
index d625a3f83780..ca57969a0923 100644
@@ -33,6 +33,7 @@
 /x86_64/pmu_event_filter_test
 /x86_64/set_boot_cpu_id
 /x86_64/set_sregs_test
+/x86_64/sev_all_boot_test
 /x86_64/sev_migrate_tests
 /x86_64/smm_test
 /x86_64/state_test
tools/testing/selftests/kvm/Makefile +1 -0
index 0a70e50f0498..10a7ea09ea98 100644
@@ -123,6 +123,7 @@ TEST_GEN_PROGS_x86_64 += x86_64/tsc_msrs_test
 TEST_GEN_PROGS_x86_64 += x86_64/vmx_pmu_caps_test
 TEST_GEN_PROGS_x86_64 += x86_64/xen_shinfo_test
 TEST_GEN_PROGS_x86_64 += x86_64/xen_vmcall_test
+TEST_GEN_PROGS_x86_64 += x86_64/sev_all_boot_test
 TEST_GEN_PROGS_x86_64 += x86_64/sev_migrate_tests
 TEST_GEN_PROGS_x86_64 += x86_64/amx_test
 TEST_GEN_PROGS_x86_64 += x86_64/max_vcpuid_cap_test
tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c new file +127 -0

127 lines not on this page · raw message

Signature · 1 line
2.37.2.672.g94769d06f0-goog
#23 lore raw

Re: [V4 8/8] KVM: selftests: Add simple sev vm testing

tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c line 49

Quoted 11 lines · Peter Gonda
tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c new file +11 -0
+	for (i = 0; i <= NR_SYNCS; ++i) {
+		vcpu_run(vcpu);
+		switch (get_ucall(vcpu, &uc)) {
+		case UCALL_SYNC:
+			continue;
+		case UCALL_DONE:
+			return;
+		case UCALL_ABORT:
+			TEST_ASSERT(false, "%s at %s:%ld\n\tvalues: %#lx, %#lx",
+				    (const char *)uc.args[0], __FILE__,
+				    uc.args[1], uc.args[2], uc.args[3]);

REPORT_GUEST_ASSERT

tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c line 53

Quoted 4 lines
tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c new file +4 -0
+		default:
+			TEST_ASSERT(
+				false, "Unexpected exit: %s",
+				exit_reason_str(vcpu->run->exit_reason));

TEST_FAIL

Quoted 5 lines
tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c new file +5 -0
+		}
+	}
+}
+
+static void guest_sev_code(void)

tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c line 68

Quoted 10 lines
tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c new file +10 -0
+{
+	uint32_t eax, ebx, ecx, edx;
+	uint64_t sev_status;
+
+	GUEST_SYNC(1);
+
+	GUEST_ASSERT(this_cpu_has(X86_FEATURE_SEV));
+
+	sev_status = rdmsr(MSR_AMD64_SEV);
+	GUEST_ASSERT((sev_status & 0x1) == 1);

Add a #define for the magic 0x1. And the "== 1" is unnecessary. Hmm, and the querying of SEV/SEV-ES/etc status in MSR_AMD64_SEV should be done in helpers.

Quoted 4 lines
tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c new file +4 -0
+
+	GUEST_DONE();
+}
+

tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c line 82

Quoted 10 lines
tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c new file +10 -0
+static struct sev_vm *setup_test_common(void *guest_code, uint64_t policy,
+					struct kvm_vcpu **vcpu)
+{
+	uint8_t measurement[512];
+	struct sev_vm *sev;
+	struct kvm_vm *vm;
+	int i;
+
+	sev = sev_vm_create(policy, TOTAL_PAGES);
+	TEST_ASSERT(sev, "sev_vm_create() failed to create VM\n");

See earlier comments on using TEST_REQUIRE.

Quoted 8 lines
tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c new file +8 -0
+
+	vm = sev->vm;
+
+	/* Set up VCPU and initial guest kernel. */
+	*vcpu = vm_vcpu_add(vm, 0, guest_code);
+	kvm_vm_elf_load(vm, program_invocation_name);
+
+	/* Allocations/setup done. Encrypt initial guest payload. */

tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c line 100

Quoted 10 lines
tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c new file +10 -0
+	sev_vm_launch(sev);
+
+	/* Dump the initial measurement. A test to actually verify it would be nice. */
+	sev_vm_launch_measure(sev, measurement);
+	pr_info("guest measurement: ");
+	for (i = 0; i < 32; ++i)
+		pr_info("%02x", measurement[i]);
+	pr_info("\n");
+
+	sev_vm_launch_finish(sev);

I believe this entire function gets replaced with

	vm = ____vm_create_with_one_vcpu(VM_MODE_PXXV48_4K_SEV, &vcpu, 0,
					 guest_code);

tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c line 112

Quoted 12 lines
tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c new file +12 -0
+
+	return sev;
+}
+
+static void test_sev(void *guest_code, uint64_t policy)
+{
+	struct sev_vm *sev;
+	struct kvm_vcpu *vcpu;
+
+	sev = setup_test_common(guest_code, policy, &vcpu);
+
+	/* Guest is ready to run. Do the tests. */

Comment doesn't add much value.

tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c line 115

Quoted 3 lines
tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c new file +3 -0
+	guest_run_loop(vcpu);
+
+	pr_info("guest ran successfully\n");

Meh, the test passing is a good indication the test ran successfully.

tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c line 122

Quoted 7 lines
tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c new file +7 -0
+
+	sev_vm_free(sev);
+}
+
+int main(int argc, char *argv[])
+{
+	/* SEV tests */

Again, stating the obvious.

Quoted 8 lines
tools/testing/selftests/kvm/x86_64/sev_all_boot_test.c new file +5 -0
+	test_sev(guest_sev_code, SEV_POLICY_NO_DBG);
+	test_sev(guest_sev_code, 0);
+
+	return 0;
+}
Signature · 1 line
2.37.2.672.g94769d06f0-goog