Skip to contentSkip to conversation

x86: Add a test for AMD SEV-ES #VC handling

PATCH Superseded Varad Gautam 5 participants 11 messages May 31, 2021 – Jun 24, 2021

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86: Add a test for AMD SEV-ES #VC handling

Some vmexits on a SEV-ES guest need special handling within the guest before exiting to the hypervisor. This must happen within the guest's \#VC exception handler, triggered on every non automatic exit.

Add a KUnit based test to validate Linux's VC handling. The test:
1. installs a kretprobe on the #VC handler (sev_es_ghcb_hv_call, to access GHCB before/after the resulting VMGEXIT).
2. tiggers an NAE.
3. checks that the kretprobe was hit with the right exit_code available in GHCB.

Since it relies on kprobes, the test does not cover NMI contexts.

---
 arch/x86/Kconfig              |   9 +++
 arch/x86/kernel/Makefile      |   5 ++
 arch/x86/kernel/sev-test-vc.c | 142 ++++++++++++++++++++++++++++++++++
 3 files changed, 156 insertions(+)
 create mode 100644 arch/x86/kernel/sev-test-vc.c
arch/x86/Kconfig +9 -0
index 0045e1b441902..0a3c3f31813f1 100644
@@ -1543,6 +1543,15 @@ config AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT
 	  If set to N, then the encryption of system memory can be
 	  activated with the mem_encrypt=on command line option.
 
+config AMD_MEM_ENCRYPT_TEST_VC
+	bool "Test for AMD Secure Memory Encryption (SME) support"
+	depends on AMD_MEM_ENCRYPT
+	select KUNIT
+	select FUNCTION_TRACER
+	help
+	  Enable KUnit-based testing for the encryption of system memory
+	  using AMD SEV-ES. Currently only tests #VC handling.
+
 # Common NUMA Features
 config NUMA
 	bool "NUMA Memory Allocation and Scheduler Support"
arch/x86/kernel/Makefile +5 -0
index 0f66682ac02a6..360c5d33580ca 100644
@@ -23,6 +23,10 @@ CFLAGS_REMOVE_head64.o = -pg
 CFLAGS_REMOVE_sev.o = -pg
 endif
 
+ifdef CONFIG_AMD_MEM_ENCRYPT_TEST_VC
+CFLAGS_sev.o	+= -fno-ipa-sra
+endif
+
 KASAN_SANITIZE_head$(BITS).o				:= n
 KASAN_SANITIZE_dumpstack.o				:= n
 KASAN_SANITIZE_dumpstack_$(BITS).o			:= n
@@ -149,6 +153,7 @@ obj-$(CONFIG_UNWINDER_FRAME_POINTER)	+= unwind_frame.o
 obj-$(CONFIG_UNWINDER_GUESS)		+= unwind_guess.o
 
 obj-$(CONFIG_AMD_MEM_ENCRYPT)		+= sev.o
+obj-$(CONFIG_AMD_MEM_ENCRYPT_TEST_VC)	+= sev-test-vc.o
 ###
 # 64 bit specific files
 ifeq ($(CONFIG_X86_64),y)
arch/x86/kernel/sev-test-vc.c new file +142 -0

142 lines not on this page · raw message

Signature · 1 line
2.30.2
#1 lore raw

PATCH v2 1/1 x86: Add a test for AMD SEV-ES #VC handling

Some vmexits on a SEV-ES guest need special handling within the guest before exiting to the hypervisor. This must happen within the guest's \#VC exception handler, triggered on every non automatic exit.

Add a KUnit based test to validate Linux's VC handling. The test:
1. installs a kretprobe on the #VC handler (sev_es_ghcb_hv_call, to access GHCB before/after the resulting VMGEXIT).
2. tiggers an NAE.
3. checks that the kretprobe was hit with the right exit_code available in GHCB.

Since relying on kprobes, the test does not cover NMI contexts.

---
v2: Add a testcase for MMIO read/write.

 arch/x86/Kconfig              |   9 ++
 arch/x86/kernel/Makefile      |   5 ++
 arch/x86/kernel/sev-test-vc.c | 155 ++++++++++++++++++++++++++++++++++
 3 files changed, 169 insertions(+)
 create mode 100644 arch/x86/kernel/sev-test-vc.c
arch/x86/Kconfig +9 -0
index 0045e1b441902..0a3c3f31813f1 100644
@@ -1543,6 +1543,15 @@ config AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT
 	  If set to N, then the encryption of system memory can be
 	  activated with the mem_encrypt=on command line option.
 
+config AMD_MEM_ENCRYPT_TEST_VC
+	bool "Test for AMD Secure Memory Encryption (SME) support"
+	depends on AMD_MEM_ENCRYPT
+	select KUNIT
+	select FUNCTION_TRACER
+	help
+	  Enable KUnit-based testing for the encryption of system memory
+	  using AMD SEV-ES. Currently only tests #VC handling.
+
 # Common NUMA Features
 config NUMA
 	bool "NUMA Memory Allocation and Scheduler Support"
arch/x86/kernel/Makefile +5 -0
index 0f66682ac02a6..360c5d33580ca 100644
@@ -23,6 +23,10 @@ CFLAGS_REMOVE_head64.o = -pg
 CFLAGS_REMOVE_sev.o = -pg
 endif
 
+ifdef CONFIG_AMD_MEM_ENCRYPT_TEST_VC
+CFLAGS_sev.o	+= -fno-ipa-sra
+endif
+
 KASAN_SANITIZE_head$(BITS).o				:= n
 KASAN_SANITIZE_dumpstack.o				:= n
 KASAN_SANITIZE_dumpstack_$(BITS).o			:= n
@@ -149,6 +153,7 @@ obj-$(CONFIG_UNWINDER_FRAME_POINTER)	+= unwind_frame.o
 obj-$(CONFIG_UNWINDER_GUESS)		+= unwind_guess.o
 
 obj-$(CONFIG_AMD_MEM_ENCRYPT)		+= sev.o
+obj-$(CONFIG_AMD_MEM_ENCRYPT_TEST_VC)	+= sev-test-vc.o
 ###
 # 64 bit specific files
 ifeq ($(CONFIG_X86_64),y)
arch/x86/kernel/sev-test-vc.c new file +155 -0

155 lines not on this page · raw message

Signature · 1 line
2.30.2
#2 lore raw
Quoted 4 lines · Varad Gautam

Some vmexits on a SEV-ES guest need special handling within the guest before exiting to the hypervisor. This must happen within the guest's \#VC exception handler, triggered on every non automatic exit.

Quoted 8 lines

Add a KUnit based test to validate Linux's VC handling. The test:
1. installs a kretprobe on the #VC handler (sev_es_ghcb_hv_call, to access GHCB before/after the resulting VMGEXIT).
2. tiggers an NAE.
3. checks that the kretprobe was hit with the right exit_code available in GHCB.

Since relying on kprobes, the test does not cover NMI contexts.

I'm not very familiar with these types of tests, so just general feedback below.

Quoted 16 lines

---
v2: Add a testcase for MMIO read/write.

 arch/x86/Kconfig              |   9 ++
 arch/x86/kernel/Makefile      |   5 ++
 arch/x86/kernel/sev-test-vc.c | 155 ++++++++++++++++++++++++++++++++++
 3 files changed, 169 insertions(+)
 create mode 100644 arch/x86/kernel/sev-test-vc.c
arch/x86/Kconfig +0 -0
index 0045e1b441902..0a3c3f31813f1 100644

arch/x86/Kconfig config AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT · line 1547

Quoted 5 lines
arch/x86/Kconfig +2 -0
@@ -1543,6 +1543,15 @@ config AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT
 	  If set to N, then the encryption of system memory can be
 	  activated with the mem_encrypt=on command line option.
 
+config AMD_MEM_ENCRYPT_TEST_VC
+	bool "Test for AMD Secure Memory Encryption (SME) support"

I would change this to indicate that this is specifically for testing SEV-ES support. We messed up and didn't update the AMD_MEM_ENCRYPT entry when the SEV support was submitted.

Thanks,
Tom

Quoted 15 lines
arch/x86/Kconfig +7 -0
+	depends on AMD_MEM_ENCRYPT
+	select KUNIT
+	select FUNCTION_TRACER
+	help
+	  Enable KUnit-based testing for the encryption of system memory
+	  using AMD SEV-ES. Currently only tests #VC handling.
+
 # Common NUMA Features
 config NUMA
 	bool "NUMA Memory Allocation and Scheduler Support"
arch/x86/kernel/Makefile +0 -0
index 0f66682ac02a6..360c5d33580ca 100644

arch/x86/kernel/Makefile CFLAGS_REMOVE_head64.o = -pg · line 28

Quoted 6 lines
arch/x86/kernel/Makefile +3 -0
@@ -23,6 +23,10 @@ CFLAGS_REMOVE_head64.o = -pg
 CFLAGS_REMOVE_sev.o = -pg
 endif
 
+ifdef CONFIG_AMD_MEM_ENCRYPT_TEST_VC
+CFLAGS_sev.o	+= -fno-ipa-sra
+endif

Maybe add something in the commit message as to why this is needed.

Quoted 27 lines
arch/x86/kernel/Makefile +2 -0
+
 KASAN_SANITIZE_head$(BITS).o				:= n
 KASAN_SANITIZE_dumpstack.o				:= n
 KASAN_SANITIZE_dumpstack_$(BITS).o			:= n
@@ -149,6 +153,7 @@ obj-$(CONFIG_UNWINDER_FRAME_POINTER)	+= unwind_frame.o
 obj-$(CONFIG_UNWINDER_GUESS)		+= unwind_guess.o
 
 obj-$(CONFIG_AMD_MEM_ENCRYPT)		+= sev.o
+obj-$(CONFIG_AMD_MEM_ENCRYPT_TEST_VC)	+= sev-test-vc.o
 ###
 # 64 bit specific files
 ifeq ($(CONFIG_X86_64),y)
arch/x86/kernel/sev-test-vc.c new file +9 -0
new file mode 100644
index 0000000000000..2475270b844e8
@@ -0,0 +1,155 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2021 SUSE
+ *
+ * Author: Varad Gautam <[email protected]>
+ */
+
+#include <asm/cpufeature.h>
+#include <asm/debugreg.h>

arch/x86/kernel/sev-test-vc.c line 19

Quoted 10 lines
arch/x86/kernel/sev-test-vc.c new file +10 -0
+#include <asm/io.h>
+#include <asm/sev-common.h>
+#include <asm/svm.h>
+#include <kunit/test.h>
+#include <linux/kprobes.h>
+
+static struct kretprobe hv_call_krp;
+
+static int hv_call_krp_entry(struct kretprobe_instance *krpi,
+				    struct pt_regs *regs)

Align with the argument above.

arch/x86/kernel/sev-test-vc.c line 28

Quoted 9 lines
arch/x86/kernel/sev-test-vc.c new file +9 -0
+{
+	unsigned long ghcb_vaddr = regs_get_kernel_argument(regs, 0);
+	*((unsigned long *) krpi->data) = ghcb_vaddr;
+
+	return 0;
+}
+
+static int hv_call_krp_ret(struct kretprobe_instance *krpi,
+				    struct pt_regs *regs)

Align with the argument above.

Quoted 7 lines
arch/x86/kernel/sev-test-vc.c new file +7 -0
+{
+	unsigned long ghcb_vaddr = *((unsigned long *) krpi->data);
+	struct ghcb *ghcb = (struct ghcb *) ghcb_vaddr;
+	struct kunit *test = current->kunit_test;
+
+	if (test && strstr(test->name, "sev_es_") && test->priv)
+		cmpxchg((unsigned long *) test->priv, ghcb->save.sw_exit_code, 1);

arch/x86/kernel/sev-test-vc.c line 45

Quoted 10 lines
arch/x86/kernel/sev-test-vc.c new file +10 -0
+
+	return 0;
+}
+
+int sev_test_vc_init(struct kunit *test)
+{
+	int ret;
+
+	if (!sev_es_active()) {
+		pr_err("Not a SEV-ES guest. Skipping.");

Should this be a pr_info vs a pr_err?

Should you define a pr_fmt above for the pr_ messages?

Quoted 4 lines
arch/x86/kernel/sev-test-vc.c new file +4 -0
+		ret = -EINVAL;
+		goto out;
+	}
+

arch/x86/kernel/sev-test-vc.c line 59

Quoted 10 lines
arch/x86/kernel/sev-test-vc.c new file +10 -0
+	memset(&hv_call_krp, 0, sizeof(hv_call_krp));
+	hv_call_krp.entry_handler = hv_call_krp_entry;
+	hv_call_krp.handler = hv_call_krp_ret;
+	hv_call_krp.maxactive = 100;
+	hv_call_krp.data_size = sizeof(unsigned long);
+	hv_call_krp.kp.symbol_name = "sev_es_ghcb_hv_call";
+	hv_call_krp.kp.addr = 0;
+
+	ret = register_kretprobe(&hv_call_krp);
+	if (ret < 0) {

Should this just be "if (ret) {"? Can a positive number be returned and if so, what does it mean?

Quoted 21 lines
arch/x86/kernel/sev-test-vc.c new file +21 -0
+		pr_err("Could not register kretprobe. Skipping.");
+		goto out;
+	}
+
+	test->priv = kunit_kzalloc(test, sizeof(unsigned long), GFP_KERNEL);
+	if (!test->priv) {
+		ret = -ENOMEM;
+		pr_err("Could not allocate. Skipping.");
+		goto out;
+	}
+
+out:
+	return ret;
+}
+
+void sev_test_vc_exit(struct kunit *test)
+{
+	if (test->priv)
+		kunit_kfree(test, test->priv);
+
+	if (hv_call_krp.kp.addr)

arch/x86/kernel/sev-test-vc.c line 90

Quoted 10 lines
arch/x86/kernel/sev-test-vc.c new file +10 -0
+		unregister_kretprobe(&hv_call_krp);
+}
+
+#define guarded_op(kt, ec, op)				\
+do {							\
+	struct kunit *t = (struct kunit *) kt;		\
+	smp_store_release((typeof(ec) *) t->priv, ec);	\
+	op;						\
+	KUNIT_EXPECT_EQ(t, (typeof(ec)) 1, 		\
+		(typeof(ec)) smp_load_acquire((typeof(ec) *) t->priv));	\

I usually like seeing all the '\' characters lined up, rather than having just the one hanging out.

Thanks,
Tom

Quoted 66 lines
arch/x86/kernel/sev-test-vc.c new file +65 -0
+} while(0)
+
+static void sev_es_nae_cpuid(struct kunit *test)
+{
+	unsigned int cpuid_fn = 0x8000001f;
+
+	guarded_op(test, SVM_EXIT_CPUID, native_cpuid_eax(cpuid_fn));
+}
+
+static void sev_es_nae_wbinvd(struct kunit *test)
+{
+	guarded_op(test, SVM_EXIT_WBINVD, wbinvd());
+}
+
+static void sev_es_nae_msr(struct kunit *test)
+{
+	guarded_op(test, SVM_EXIT_MSR, __rdmsr(MSR_IA32_TSC));
+}
+
+static void sev_es_nae_dr7_rw(struct kunit *test)
+{
+	guarded_op(test, SVM_EXIT_WRITE_DR7,
+		   native_set_debugreg(7, native_get_debugreg(7)));
+}
+
+static void sev_es_nae_ioio(struct kunit *test)
+{
+	unsigned long port = 0x80;
+	char val = 0;
+
+	guarded_op(test, SVM_EXIT_IOIO, val = inb(port));
+	guarded_op(test, SVM_EXIT_IOIO, outb(val, port));
+	guarded_op(test, SVM_EXIT_IOIO, insb(port, &val, sizeof(val)));
+	guarded_op(test, SVM_EXIT_IOIO, outsb(port, &val, sizeof(val)));
+}
+
+static void sev_es_nae_mmio(struct kunit *test)
+{
+	unsigned long lapic_ver_pa = 0xfee00030; /* APIC_DEFAULT_PHYS_BASE + APIC_LVR */
+	unsigned __iomem *lapic = ioremap(lapic_ver_pa, 0x4);
+	unsigned lapic_version = 0;
+
+	guarded_op(test, SVM_VMGEXIT_MMIO_READ, lapic_version = *lapic);
+	guarded_op(test, SVM_VMGEXIT_MMIO_WRITE, *lapic = lapic_version);
+
+	iounmap(lapic);
+}
+
+static struct kunit_case sev_test_vc_testcases[] = {
+	KUNIT_CASE(sev_es_nae_cpuid),
+	KUNIT_CASE(sev_es_nae_wbinvd),
+	KUNIT_CASE(sev_es_nae_msr),
+	KUNIT_CASE(sev_es_nae_dr7_rw),
+	KUNIT_CASE(sev_es_nae_ioio),
+	KUNIT_CASE(sev_es_nae_mmio),
+	{}
+};
+
+static struct kunit_suite sev_vc_test_suite = {
+	.name = "sev_test_vc",
+	.init = sev_test_vc_init,
+	.exit = sev_test_vc_exit,
+	.test_cases = sev_test_vc_testcases,
+};
+kunit_test_suite(sev_vc_test_suite);
maintainer
#3 lore raw
Quoted 5 lines · Tom Lendacky

arch/x86/kernel/Makefile CFLAGS_REMOVE_head64.o = -pg · line 28

Quoted 3 lines
arch/x86/kernel/Makefile +3 -0
+ifdef CONFIG_AMD_MEM_ENCRYPT_TEST_VC
+CFLAGS_sev.o	+= -fno-ipa-sra
+endif

Maybe add something in the commit message as to why this is needed.

Or even better - in a comment above it. Commit messages are harder to find when time passes and other commits touch this line and you have to go and do git archeology just to figure out why this was done...

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
#4 lore raw
In reply to Tom Lendacky
Quoted 36 lines · Tom Lendacky
Quoted 4 lines · Varad Gautam

Some vmexits on a SEV-ES guest need special handling within the guest before exiting to the hypervisor. This must happen within the guest's \#VC exception handler, triggered on every non automatic exit.

Quoted 8 lines

Add a KUnit based test to validate Linux's VC handling. The test:
1. installs a kretprobe on the #VC handler (sev_es_ghcb_hv_call, to access GHCB before/after the resulting VMGEXIT).
2. tiggers an NAE.
3. checks that the kretprobe was hit with the right exit_code available in GHCB.

Since relying on kprobes, the test does not cover NMI contexts.

I'm not very familiar with these types of tests, so just general feedback below.

Quoted 19 lines

---
v2: Add a testcase for MMIO read/write.

 arch/x86/Kconfig              |   9 ++
 arch/x86/kernel/Makefile      |   5 ++
 arch/x86/kernel/sev-test-vc.c | 155 ++++++++++++++++++++++++++++++++++
 3 files changed, 169 insertions(+)
 create mode 100644 arch/x86/kernel/sev-test-vc.c
arch/x86/Kconfig +0 -0
index 0045e1b441902..0a3c3f31813f1 100644
@@ -1543,6 +1543,15 @@ config AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT
 	  If set to N, then the encryption of system memory can be
 	  activated with the mem_encrypt=on command line option.
 
Quoted 7 lines

arch/x86/Kconfig config AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT · line 1547

Quoted 2 lines
arch/x86/Kconfig +2 -0
+config AMD_MEM_ENCRYPT_TEST_VC
+	bool "Test for AMD Secure Memory Encryption (SME) support"

I would change this to indicate that this is specifically for testing SEV-ES support. We messed up and didn't update the AMD_MEM_ENCRYPT entry when the SEV support was submitted.

Ack, changed naming everywhere to indicate that this is specific to SEV-ES.

Quoted 21 lines

Thanks,
Tom

Quoted 18 lines
arch/x86/Kconfig +7 -0
+	depends on AMD_MEM_ENCRYPT
+	select KUNIT
+	select FUNCTION_TRACER
+	help
+	  Enable KUnit-based testing for the encryption of system memory
+	  using AMD SEV-ES. Currently only tests #VC handling.
+
 # Common NUMA Features
 config NUMA
 	bool "NUMA Memory Allocation and Scheduler Support"
arch/x86/kernel/Makefile +0 -0
index 0f66682ac02a6..360c5d33580ca 100644
@@ -23,6 +23,10 @@ CFLAGS_REMOVE_head64.o = -pg
 CFLAGS_REMOVE_sev.o = -pg
 endif
 
Quoted 6 lines

arch/x86/kernel/Makefile CFLAGS_REMOVE_head64.o = -pg · line 28

Quoted 3 lines
arch/x86/kernel/Makefile +3 -0
+ifdef CONFIG_AMD_MEM_ENCRYPT_TEST_VC
+CFLAGS_sev.o	+= -fno-ipa-sra
+endif

Maybe add something in the commit message as to why this is needed.

I added a comment here, ipa-sra adds an .isra.* suffix to function names, which breaks kprobes' lookup for "sev_es_ghcb_hv_call". The alternative here is to do an inexact search within kallsyms, but -fno-ipa-sra is simpler and assures that register_kprobes finds the right address.

Quoted 72 lines
Quoted 27 lines
arch/x86/kernel/Makefile +2 -0
+
 KASAN_SANITIZE_head$(BITS).o				:= n
 KASAN_SANITIZE_dumpstack.o				:= n
 KASAN_SANITIZE_dumpstack_$(BITS).o			:= n
@@ -149,6 +153,7 @@ obj-$(CONFIG_UNWINDER_FRAME_POINTER)	+= unwind_frame.o
 obj-$(CONFIG_UNWINDER_GUESS)		+= unwind_guess.o
 
 obj-$(CONFIG_AMD_MEM_ENCRYPT)		+= sev.o
+obj-$(CONFIG_AMD_MEM_ENCRYPT_TEST_VC)	+= sev-test-vc.o
 ###
 # 64 bit specific files
 ifeq ($(CONFIG_X86_64),y)
arch/x86/kernel/sev-test-vc.c new file +9 -0
new file mode 100644
index 0000000000000..2475270b844e8
@@ -0,0 +1,155 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2021 SUSE
+ *
+ * Author: Varad Gautam <[email protected]>
+ */
+
+#include <asm/cpufeature.h>
+#include <asm/debugreg.h>

arch/x86/kernel/sev-test-vc.c line 19

Quoted 10 lines
arch/x86/kernel/sev-test-vc.c new file +10 -0
+#include <asm/io.h>
+#include <asm/sev-common.h>
+#include <asm/svm.h>
+#include <kunit/test.h>
+#include <linux/kprobes.h>
+
+static struct kretprobe hv_call_krp;
+
+static int hv_call_krp_entry(struct kretprobe_instance *krpi,
+				    struct pt_regs *regs)

Align with the argument above.

arch/x86/kernel/sev-test-vc.c line 28

Quoted 9 lines
arch/x86/kernel/sev-test-vc.c new file +9 -0
+{
+	unsigned long ghcb_vaddr = regs_get_kernel_argument(regs, 0);
+	*((unsigned long *) krpi->data) = ghcb_vaddr;
+
+	return 0;
+}
+
+static int hv_call_krp_ret(struct kretprobe_instance *krpi,
+				    struct pt_regs *regs)

Align with the argument above.

Quoted 7 lines
arch/x86/kernel/sev-test-vc.c new file +7 -0
+{
+	unsigned long ghcb_vaddr = *((unsigned long *) krpi->data);
+	struct ghcb *ghcb = (struct ghcb *) ghcb_vaddr;
+	struct kunit *test = current->kunit_test;
+
+	if (test && strstr(test->name, "sev_es_") && test->priv)
+		cmpxchg((unsigned long *) test->priv, ghcb->save.sw_exit_code, 1);

arch/x86/kernel/sev-test-vc.c line 45

Quoted 10 lines
arch/x86/kernel/sev-test-vc.c new file +10 -0
+
+	return 0;
+}
+
+int sev_test_vc_init(struct kunit *test)
+{
+	int ret;
+
+	if (!sev_es_active()) {
+		pr_err("Not a SEV-ES guest. Skipping.");

Should this be a pr_info vs a pr_err?

Quoted 2 lines

Should you define a pr_fmt above for the pr_ messages?

Changed the prints to kunit_info(), which appends a per-test prefix.

Quoted 9 lines
Quoted 9 lines
arch/x86/kernel/sev-test-vc.c new file +9 -0
+		ret = -EINVAL;
+		goto out;
+	}
+
+	memset(&hv_call_krp, 0, sizeof(hv_call_krp));
+	hv_call_krp.entry_handler = hv_call_krp_entry;
+	hv_call_krp.handler = hv_call_krp_ret;
+	hv_call_krp.maxactive = 100;
+	hv_call_krp.data_size = sizeof(unsigned long);
Quoted 9 lines

arch/x86/kernel/sev-test-vc.c line 59

Quoted 5 lines
arch/x86/kernel/sev-test-vc.c new file +5 -0
+	hv_call_krp.kp.symbol_name = "sev_es_ghcb_hv_call";
+	hv_call_krp.kp.addr = 0;
+
+	ret = register_kretprobe(&hv_call_krp);
+	if (ret < 0) {

Should this just be "if (ret) {"? Can a positive number be returned and if so, what does it mean?

Yes, "if (ret) {" is better. Changed.

Thanks,
Varad

Quoted 105 lines
Quoted 21 lines
arch/x86/kernel/sev-test-vc.c new file +21 -0
+		pr_err("Could not register kretprobe. Skipping.");
+		goto out;
+	}
+
+	test->priv = kunit_kzalloc(test, sizeof(unsigned long), GFP_KERNEL);
+	if (!test->priv) {
+		ret = -ENOMEM;
+		pr_err("Could not allocate. Skipping.");
+		goto out;
+	}
+
+out:
+	return ret;
+}
+
+void sev_test_vc_exit(struct kunit *test)
+{
+	if (test->priv)
+		kunit_kfree(test, test->priv);
+
+	if (hv_call_krp.kp.addr)

arch/x86/kernel/sev-test-vc.c line 90

Quoted 10 lines
arch/x86/kernel/sev-test-vc.c new file +10 -0
+		unregister_kretprobe(&hv_call_krp);
+}
+
+#define guarded_op(kt, ec, op)				\
+do {							\
+	struct kunit *t = (struct kunit *) kt;		\
+	smp_store_release((typeof(ec) *) t->priv, ec);	\
+	op;						\
+	KUNIT_EXPECT_EQ(t, (typeof(ec)) 1, 		\
+		(typeof(ec)) smp_load_acquire((typeof(ec) *) t->priv));	\

I usually like seeing all the '\' characters lined up, rather than having just the one hanging out.

Thanks,
Tom

Quoted 66 lines
arch/x86/kernel/sev-test-vc.c new file +65 -0
+} while(0)
+
+static void sev_es_nae_cpuid(struct kunit *test)
+{
+	unsigned int cpuid_fn = 0x8000001f;
+
+	guarded_op(test, SVM_EXIT_CPUID, native_cpuid_eax(cpuid_fn));
+}
+
+static void sev_es_nae_wbinvd(struct kunit *test)
+{
+	guarded_op(test, SVM_EXIT_WBINVD, wbinvd());
+}
+
+static void sev_es_nae_msr(struct kunit *test)
+{
+	guarded_op(test, SVM_EXIT_MSR, __rdmsr(MSR_IA32_TSC));
+}
+
+static void sev_es_nae_dr7_rw(struct kunit *test)
+{
+	guarded_op(test, SVM_EXIT_WRITE_DR7,
+		   native_set_debugreg(7, native_get_debugreg(7)));
+}
+
+static void sev_es_nae_ioio(struct kunit *test)
+{
+	unsigned long port = 0x80;
+	char val = 0;
+
+	guarded_op(test, SVM_EXIT_IOIO, val = inb(port));
+	guarded_op(test, SVM_EXIT_IOIO, outb(val, port));
+	guarded_op(test, SVM_EXIT_IOIO, insb(port, &val, sizeof(val)));
+	guarded_op(test, SVM_EXIT_IOIO, outsb(port, &val, sizeof(val)));
+}
+
+static void sev_es_nae_mmio(struct kunit *test)
+{
+	unsigned long lapic_ver_pa = 0xfee00030; /* APIC_DEFAULT_PHYS_BASE + APIC_LVR */
+	unsigned __iomem *lapic = ioremap(lapic_ver_pa, 0x4);
+	unsigned lapic_version = 0;
+
+	guarded_op(test, SVM_VMGEXIT_MMIO_READ, lapic_version = *lapic);
+	guarded_op(test, SVM_VMGEXIT_MMIO_WRITE, *lapic = lapic_version);
+
+	iounmap(lapic);
+}
+
+static struct kunit_case sev_test_vc_testcases[] = {
+	KUNIT_CASE(sev_es_nae_cpuid),
+	KUNIT_CASE(sev_es_nae_wbinvd),
+	KUNIT_CASE(sev_es_nae_msr),
+	KUNIT_CASE(sev_es_nae_dr7_rw),
+	KUNIT_CASE(sev_es_nae_ioio),
+	KUNIT_CASE(sev_es_nae_mmio),
+	{}
+};
+
+static struct kunit_suite sev_vc_test_suite = {
+	.name = "sev_test_vc",
+	.init = sev_test_vc_init,
+	.exit = sev_test_vc_exit,
+	.test_cases = sev_test_vc_testcases,
+};
+kunit_test_suite(sev_vc_test_suite);
Signature · 7 lines
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5
90409 Nürnberg
Germany

HRB 36809, AG Nürnberg
Geschäftsführer: Felix Imendörffer
#5 lore raw

PATCH v3 1/1 x86: Add a test for AMD SEV-ES guest #VC handling

In reply to Varad Gautam

From: Varad Gautam <[email protected]>

Some vmexits on a SEV-ES guest need special handling within the guest before exiting to the hypervisor. This must happen within the guest's \#VC exception handler, triggered on every non automatic exit.

Add a KUnit based test to validate Linux's VC handling. The test:
1. installs a kretprobe on the #VC handler (sev_es_ghcb_hv_call, to access GHCB before/after the resulting VMGEXIT).
2. tiggers an NAE.
3. checks that the kretprobe was hit with the right exit_code available in GHCB.

Since relying on kprobes, the test does not cover NMI contexts.

---
 arch/x86/Kconfig                 |   9 ++
 arch/x86/kernel/Makefile         |   8 ++
 arch/x86/kernel/sev-es-test-vc.c | 155 +++++++++++++++++++++++++++++++
 3 files changed, 172 insertions(+)
 create mode 100644 arch/x86/kernel/sev-es-test-vc.c
arch/x86/Kconfig +9 -0
index 0045e1b441902..85b8ac450ba56 100644
@@ -1543,6 +1543,15 @@ config AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT
 	  If set to N, then the encryption of system memory can be
 	  activated with the mem_encrypt=on command line option.
 
+config AMD_SEV_ES_TEST_VC
+	bool "Test for AMD SEV-ES VC exception handling."
+	depends on AMD_MEM_ENCRYPT
+	select FUNCTION_TRACER
+	select KPROBES
+	select KUNIT
+	help
+	  Enable KUnit-based testing for AMD SEV-ES #VC exception handling.
+
 # Common NUMA Features
 config NUMA
 	bool "NUMA Memory Allocation and Scheduler Support"
arch/x86/kernel/Makefile +8 -0
index 0f66682ac02a6..1632d6156be6e 100644
@@ -23,6 +23,13 @@ CFLAGS_REMOVE_head64.o = -pg
 CFLAGS_REMOVE_sev.o = -pg
 endif
 
+# AMD_SEV_ES_TEST_VC registers a kprobe by function name. IPA-SRA creates
+# function copies and renames them to have an .isra suffix, which breaks kprobes'
+# lookup. Build with -fno-ipa-sra for the test.
+ifdef CONFIG_AMD_SEV_ES_TEST_VC
+CFLAGS_sev.o	+= -fno-ipa-sra
+endif
+
 KASAN_SANITIZE_head$(BITS).o				:= n
 KASAN_SANITIZE_dumpstack.o				:= n
 KASAN_SANITIZE_dumpstack_$(BITS).o			:= n
@@ -149,6 +156,7 @@ obj-$(CONFIG_UNWINDER_FRAME_POINTER)	+= unwind_frame.o
 obj-$(CONFIG_UNWINDER_GUESS)		+= unwind_guess.o
 
 obj-$(CONFIG_AMD_MEM_ENCRYPT)		+= sev.o
+obj-$(CONFIG_AMD_SEV_ES_TEST_VC)	+= sev-es-test-vc.o
 ###
 # 64 bit specific files
 ifeq ($(CONFIG_X86_64),y)
arch/x86/kernel/sev-es-test-vc.c new file +155 -0

155 lines not on this page · raw message

Signature · 1 line
2.30.2
Tested-by on v3 1/1
#6 lore raw

Re: [PATCH v3] x86: Add a test for AMD SEV-ES guest #VC handling

Quoted 13 lines · Varad Gautam

From: Varad Gautam <[email protected]>

Some vmexits on a SEV-ES guest need special handling within the guest before exiting to the hypervisor. This must happen within the guest's \#VC exception handler, triggered on every non automatic exit.

Add a KUnit based test to validate Linux's VC handling. The test:
1. installs a kretprobe on the #VC handler (sev_es_ghcb_hv_call, to access GHCB before/after the resulting VMGEXIT).
2. tiggers an NAE.
3. checks that the kretprobe was hit with the right exit_code available in GHCB.

Quoted 7 lines

Since relying on kprobes, the test does not cover NMI contexts.

---
 arch/x86/Kconfig                 |   9 ++
 arch/x86/kernel/Makefile         |   8 ++
 arch/x86/kernel/sev-es-test-vc.c | 155 +++++++++++++++++++++++++++++++

This looks good to me except for the small comment below, thanks Varad. I ran it in an SEV-ES guest and I am seeing the test results in dmesg. Only thing I am missing is a 'rep movs' test for MMIO, but that can be added later, so

Btw, should we create a separate directory for such tests like /arch/x86/tests/ or something along those lines?

Quoted 7 lines
arch/x86/Kconfig +0 -0
index 0045e1b441902..85b8ac450ba56 100644
@@ -1543,6 +1543,15 @@ config AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT
 	  If set to N, then the encryption of system memory can be
 	  activated with the mem_encrypt=on command line option.

arch/x86/Kconfig config AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT · line 1554

Quoted 10 lines
arch/x86/Kconfig +9 -0
 
+config AMD_SEV_ES_TEST_VC
+	bool "Test for AMD SEV-ES VC exception handling."
+	depends on AMD_MEM_ENCRYPT
+	select FUNCTION_TRACER
+	select KPROBES
+	select KUNIT
+	help
+	  Enable KUnit-based testing for AMD SEV-ES #VC exception handling.
+

I think this should be in arch/x86/Kconfig.debug.

#7 lore raw

Re: [PATCH v3] x86: Add a test for AMD SEV-ES guest #VC handling

Quoted 27 lines · Joerg Roedel
Quoted 13 lines · Varad Gautam

From: Varad Gautam <[email protected]>

Some vmexits on a SEV-ES guest need special handling within the guest before exiting to the hypervisor. This must happen within the guest's \#VC exception handler, triggered on every non automatic exit.

Add a KUnit based test to validate Linux's VC handling. The test:
1. installs a kretprobe on the #VC handler (sev_es_ghcb_hv_call, to access GHCB before/after the resulting VMGEXIT).
2. tiggers an NAE.
3. checks that the kretprobe was hit with the right exit_code available in GHCB.

Quoted 7 lines

Since relying on kprobes, the test does not cover NMI contexts.

---
 arch/x86/Kconfig                 |   9 ++
 arch/x86/kernel/Makefile         |   8 ++
 arch/x86/kernel/sev-es-test-vc.c | 155 +++++++++++++++++++++++++++++++

This looks good to me except for the small comment below, thanks Varad. I ran it in an SEV-ES guest and I am seeing the test results in dmesg. Only thing I am missing is a 'rep movs' test for MMIO, but that can be added later, so

Quoted 5 lines

Btw, should we create a separate directory for such tests like /arch/x86/tests/ or something along those lines?

Thanks, I've sent out a v4 with the test moved to arch/x86/tests. This is now hidden behind a CONFIG_X86_TESTS, where more stuff can be plugged in.

Quoted 8 lines
Quoted 8 lines
arch/x86/Kconfig +0 -0
index 0045e1b441902..85b8ac450ba56 100644
@@ -1543,6 +1543,15 @@ config AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT
 	  If set to N, then the encryption of system memory can be
 	  activated with the mem_encrypt=on command line option.
 
Quoted 12 lines

arch/x86/Kconfig config AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT · line 1554

Quoted 9 lines
arch/x86/Kconfig +9 -0
+config AMD_SEV_ES_TEST_VC
+	bool "Test for AMD SEV-ES VC exception handling."
+	depends on AMD_MEM_ENCRYPT
+	select FUNCTION_TRACER
+	select KPROBES
+	select KUNIT
+	help
+	  Enable KUnit-based testing for AMD SEV-ES #VC exception handling.
+

I think this should be in arch/x86/Kconfig.debug.

Ack, also moved in v4.

Varad

Signature · 7 lines
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5
90409 Nürnberg
Germany

HRB 36809, AG Nürnberg
Geschäftsführer: Felix Imendörffer
#8 lore raw

PATCH v3 1/1 x86: Add a test for AMD SEV-ES guest #VC handling

In reply to Varad Gautam

Some vmexits on a SEV-ES guest need special handling within the guest before exiting to the hypervisor. This must happen within the guest's \#VC exception handler, triggered on every non automatic exit.

Add a KUnit based test to validate Linux's VC handling. The test:
1. installs a kretprobe on the #VC handler (sev_es_ghcb_hv_call, to access GHCB before/after the resulting VMGEXIT).
2. tiggers an NAE.
3. checks that the kretprobe was hit with the right exit_code available in GHCB.

Since relying on kprobes, the test does not cover NMI contexts.

---
 arch/x86/Kconfig                 |   9 ++
 arch/x86/kernel/Makefile         |   8 ++
 arch/x86/kernel/sev-es-test-vc.c | 155 +++++++++++++++++++++++++++++++
 3 files changed, 172 insertions(+)
 create mode 100644 arch/x86/kernel/sev-es-test-vc.c
arch/x86/Kconfig +9 -0
index 0045e1b441902..85b8ac450ba56 100644
@@ -1543,6 +1543,15 @@ config AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT
 	  If set to N, then the encryption of system memory can be
 	  activated with the mem_encrypt=on command line option.
 
+config AMD_SEV_ES_TEST_VC
+	bool "Test for AMD SEV-ES VC exception handling."
+	depends on AMD_MEM_ENCRYPT
+	select FUNCTION_TRACER
+	select KPROBES
+	select KUNIT
+	help
+	  Enable KUnit-based testing for AMD SEV-ES #VC exception handling.
+
 # Common NUMA Features
 config NUMA
 	bool "NUMA Memory Allocation and Scheduler Support"
arch/x86/kernel/Makefile +8 -0
index 0f66682ac02a6..1632d6156be6e 100644
@@ -23,6 +23,13 @@ CFLAGS_REMOVE_head64.o = -pg
 CFLAGS_REMOVE_sev.o = -pg
 endif
 
+# AMD_SEV_ES_TEST_VC registers a kprobe by function name. IPA-SRA creates
+# function copies and renames them to have an .isra suffix, which breaks kprobes'
+# lookup. Build with -fno-ipa-sra for the test.
+ifdef CONFIG_AMD_SEV_ES_TEST_VC
+CFLAGS_sev.o	+= -fno-ipa-sra
+endif
+
 KASAN_SANITIZE_head$(BITS).o				:= n
 KASAN_SANITIZE_dumpstack.o				:= n
 KASAN_SANITIZE_dumpstack_$(BITS).o			:= n
@@ -149,6 +156,7 @@ obj-$(CONFIG_UNWINDER_FRAME_POINTER)	+= unwind_frame.o
 obj-$(CONFIG_UNWINDER_GUESS)		+= unwind_guess.o
 
 obj-$(CONFIG_AMD_MEM_ENCRYPT)		+= sev.o
+obj-$(CONFIG_AMD_SEV_ES_TEST_VC)	+= sev-es-test-vc.o
 ###
 # 64 bit specific files
 ifeq ($(CONFIG_X86_64),y)
arch/x86/kernel/sev-es-test-vc.c new file +155 -0

155 lines not on this page · raw message

Signature · 1 line
2.30.2
#9 lore raw

PATCH v4 1/1 x86: Add a test for AMD SEV-ES #VC handling

In reply to Varad Gautam

Some vmexits on a SEV-ES guest need special handling within the guest before exiting to the hypervisor. This must happen within the guest's \#VC exception handler, triggered on every non automatic exit.

Add a KUnit based test to validate Linux's VC handling, and introduce a new CONFIG_X86_TESTS to cover such tests. The test:
1. installs a kretprobe on the #VC handler (sev_es_ghcb_hv_call, to access GHCB before/after the resulting VMGEXIT).
2. tiggers an NAE.
3. checks that the kretprobe was hit with the right exit_code available in GHCB.

Since relying on kprobes, the test does not cover NMI contexts.

---
v4: Move this test to arch/x86/tests/, enabled by CONFIG_X86_TESTS.

 arch/x86/Kbuild                 |   2 +
 arch/x86/Kconfig.debug          |  15 ++++
 arch/x86/kernel/Makefile        |   7 ++
 arch/x86/tests/Makefile         |   3 +
 arch/x86/tests/sev-es-test-vc.c | 155 ++++++++++++++++++++++++++++++++
 5 files changed, 182 insertions(+)
 create mode 100644 arch/x86/tests/Makefile
 create mode 100644 arch/x86/tests/sev-es-test-vc.c
arch/x86/Kbuild +2 -0
index 30dec019756b9..965cfcbd12f67 100644
@@ -25,3 +25,5 @@ obj-y += platform/
 obj-y += net/
 
 obj-$(CONFIG_KEXEC_FILE) += purgatory/
+
+obj-$(CONFIG_X86_TESTS) += tests/
arch/x86/Kconfig.debug +15 -0
index 80b57e7f49477..6f63069fff972 100644
@@ -282,3 +282,18 @@ endchoice
 config FRAME_POINTER
 	depends on !UNWINDER_ORC && !UNWINDER_GUESS
 	bool
+
+config X86_TESTS
+	bool "Tests for x86"
+	help
+	    This enables building the tests under arch/x86/tests.
+
+config AMD_SEV_ES_TEST_VC
+	bool "Test for AMD SEV-ES VC exception handling"
+	depends on AMD_MEM_ENCRYPT
+	depends on X86_TESTS
+	select FUNCTION_TRACER
+	select KPROBES
+	select KUNIT
+	help
+	  Enable KUnit-based testing for AMD SEV-ES #VC exception handling.
arch/x86/kernel/Makefile +7 -0
index 0f66682ac02a6..bf1c4dc525ac6 100644
@@ -23,6 +23,13 @@ CFLAGS_REMOVE_head64.o = -pg
 CFLAGS_REMOVE_sev.o = -pg
 endif
 
+# AMD_SEV_ES_TEST_VC registers a kprobe by function name. IPA-SRA creates
+# function copies and renames them to have an .isra suffix, which breaks kprobes'
+# lookup. Build with -fno-ipa-sra for the test.
+ifdef CONFIG_AMD_SEV_ES_TEST_VC
+CFLAGS_sev.o	+= -fno-ipa-sra
+endif
+
 KASAN_SANITIZE_head$(BITS).o				:= n
 KASAN_SANITIZE_dumpstack.o				:= n
 KASAN_SANITIZE_dumpstack_$(BITS).o			:= n
arch/x86/tests/Makefile new file +3 -0
new file mode 100644
index 0000000000000..fa79c435d7843
@@ -0,0 +1,3 @@
+# SPDX-License-Identifier: GPL-2.0
+
+obj-$(CONFIG_AMD_SEV_ES_TEST_VC)	+= sev-es-test-vc.o
arch/x86/tests/sev-es-test-vc.c new file +155 -0

155 lines not on this page · raw message

Signature · 1 line
2.30.2
maintainer
#10 lore raw

On Wed, Jun 16, 2021 at 11:15:38AM +0200, Varad Gautam wrote:

Subject: Re: [PATCH v4] x86: Add a test for AMD SEV-ES #VC handling

Change your subject prefix to "x86/test: ... "

Quoted 18 lines

Some vmexits on a SEV-ES guest need special handling within the guest before exiting to the hypervisor. This must happen within the guest's \#VC exception handler, triggered on every non automatic exit.

Add a KUnit based test to validate Linux's VC handling, and introduce a new CONFIG_X86_TESTS to cover such tests. The test:
1. installs a kretprobe on the #VC handler (sev_es_ghcb_hv_call, to access GHCB before/after the resulting VMGEXIT).
2. tiggers an NAE.
3. checks that the kretprobe was hit with the right exit_code available in GHCB.

Since relying on kprobes, the test does not cover NMI contexts.

---
v4: Move this test to arch/x86/tests/, enabled by CONFIG_X86_TESTS.

Quoted 8 lines
 arch/x86/Kbuild                 |   2 +
 arch/x86/Kconfig.debug          |  15 ++++
 arch/x86/kernel/Makefile        |   7 ++
 arch/x86/tests/Makefile         |   3 +
 arch/x86/tests/sev-es-test-vc.c | 155 ++++++++++++++++++++++++++++++++
 5 files changed, 182 insertions(+)
 create mode 100644 arch/x86/tests/Makefile
 create mode 100644 arch/x86/tests/sev-es-test-vc.c

Please integrate scripts/checkpatch.pl into your patch creation workflow. Some of the warnings/errors *actually* make sense.

There's a bunch of things to fix I've pasted at the end of this mail.

Quoted 15 lines
arch/x86/Kbuild +2 -0
index 30dec019756b9..965cfcbd12f67 100644
@@ -25,3 +25,5 @@ obj-y += platform/
 obj-y += net/
 
 obj-$(CONFIG_KEXEC_FILE) += purgatory/
+
+obj-$(CONFIG_X86_TESTS) += tests/
arch/x86/Kconfig.debug +0 -0
index 80b57e7f49477..6f63069fff972 100644

arch/x86/Kconfig.debug endchoice · line 290

Quoted 9 lines
arch/x86/Kconfig.debug +6 -0
@@ -282,3 +282,18 @@ endchoice
 config FRAME_POINTER
 	depends on !UNWINDER_ORC && !UNWINDER_GUESS
 	bool
+
+config X86_TESTS
+	bool "Tests for x86"
+	help
+	    This enables building the tests under arch/x86/tests.
+

All those tests should probably be in a

if X86_TESTS

arch/x86/Kconfig.debug endchoice · line 299

Quoted 9 lines
arch/x86/Kconfig.debug +9 -0
+config AMD_SEV_ES_TEST_VC
+	bool "Test for AMD SEV-ES VC exception handling"
+	depends on AMD_MEM_ENCRYPT
+	depends on X86_TESTS
+	select FUNCTION_TRACER
+	select KPROBES
+	select KUNIT
+	help
+	  Enable KUnit-based testing for AMD SEV-ES #VC exception handling.

endif # X86_TESTS

so that you don't have the X86_TESTS dependency in each test explicitly.

arch/x86/tests/sev-es-test-vc.c

Quoted 32 lines
arch/x86/kernel/Makefile +7 -0
index 0f66682ac02a6..bf1c4dc525ac6 100644
@@ -23,6 +23,13 @@ CFLAGS_REMOVE_head64.o = -pg
 CFLAGS_REMOVE_sev.o = -pg
 endif
 
+# AMD_SEV_ES_TEST_VC registers a kprobe by function name. IPA-SRA creates
+# function copies and renames them to have an .isra suffix, which breaks kprobes'
+# lookup. Build with -fno-ipa-sra for the test.
+ifdef CONFIG_AMD_SEV_ES_TEST_VC
+CFLAGS_sev.o	+= -fno-ipa-sra
+endif
+
 KASAN_SANITIZE_head$(BITS).o				:= n
 KASAN_SANITIZE_dumpstack.o				:= n
 KASAN_SANITIZE_dumpstack_$(BITS).o			:= n
arch/x86/tests/Makefile new file +3 -0
new file mode 100644
index 0000000000000..fa79c435d7843
@@ -0,0 +1,3 @@
+# SPDX-License-Identifier: GPL-2.0
+
+obj-$(CONFIG_AMD_SEV_ES_TEST_VC)	+= sev-es-test-vc.o
arch/x86/tests/sev-es-test-vc.c new file +0 -0
new file mode 100644
index 0000000000000..98dc38572ed5d

Can this be called sev-tests.c and simply collect all SEV-related tests in it?

Quoted 26 lines
arch/x86/tests/sev-es-test-vc.c new file +25 -0
@@ -0,0 +1,155 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2021 SUSE
+ *
+ * Author: Varad Gautam <[email protected]>
+ */
+
+#include <asm/cpufeature.h>
+#include <asm/debugreg.h>
+#include <asm/io.h>
+#include <asm/sev-common.h>
+#include <asm/svm.h>
+#include <kunit/test.h>
+#include <linux/kprobes.h>
+
+static struct kretprobe hv_call_krp;
+
+static int hv_call_krp_entry(struct kretprobe_instance *krpi,
+			     struct pt_regs *regs)
+{
+	unsigned long ghcb_vaddr = regs_get_kernel_argument(regs, 0);
+	*((unsigned long *) krpi->data) = ghcb_vaddr;
+
+	return 0;
+}

arch/x86/tests/sev-es-test-vc.c line 35

Quoted 10 lines
arch/x86/tests/sev-es-test-vc.c new file +10 -0
+
+static int hv_call_krp_ret(struct kretprobe_instance *krpi,
+			   struct pt_regs *regs)
+{
+	unsigned long ghcb_vaddr = *((unsigned long *) krpi->data);
+	struct ghcb *ghcb = (struct ghcb *) ghcb_vaddr;
+	struct kunit *test = current->kunit_test;
+
+	if (test && strstr(test->name, "sev_es_") && test->priv)
+		cmpxchg((unsigned long *) test->priv, ghcb->save.sw_exit_code, 1);

Why is this needed? Normal assignment won't do?

Also, that "1" is magic as it is used in KUNIT_EXPECT_EQ below so maybe have it defined with a nice telling name.

IOW, why are those memory barriers needed?

...

Some checkpatch issues:

WARNING: 'tiggers' may be misspelled - perhaps 'triggers'? #74:
2. tiggers an NAE.
^^^^^^^

WARNING: please write a paragraph that describes the config symbol fully #112: FILE: arch/x86/Kconfig.debug:286:
+config X86_TESTS

WARNING: added, moved or deleted file(s), does MAINTAINERS need updating? #145:
new file mode 100644

WARNING: memory barrier without comment
#245: FILE: arch/x86/tests/sev-es-test-vc.c:87:
+ smp_store_release((typeof(ec) *) t->priv, ec); \

WARNING: please, no space before tabs
#247: FILE: arch/x86/tests/sev-es-test-vc.c:89:
+^IKUNIT_EXPECT_EQ(t, (typeof(ec)) 1, ^I^I^I^I\$

WARNING: memory barrier without comment
#248: FILE: arch/x86/tests/sev-es-test-vc.c:90:
+ (typeof(ec)) smp_load_acquire((typeof(ec) *) t->priv)); \

ERROR: space required before the open parenthesis '(' #249: FILE: arch/x86/tests/sev-es-test-vc.c:91:
+} while(0)

WARNING: Prefer 'unsigned int' to bare use of 'unsigned' #289: FILE: arch/x86/tests/sev-es-test-vc.c:131:
+ unsigned lapic_version = 0;

total: 1 errors, 7 warnings, 194 lines checked

NOTE: For some of the reported defects, checkpatch may be able to
mechanically convert to the typical style using --fix or --fix-inplace.

"[PATCH v4] x86: Add a test for AMD SEV-ES #VC handling" has style problems, please review.

NOTE: If any of the errors are false positives, please report
them to the maintainer, see CHECKPATCH in MAINTAINERS.
/tmp/varad.01 has problems, exiting...

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette