[kvm-unit-tests PATCH] s390x: Interception tests

[kvm-unit-tests PATCH] s390x: Interception tests

[Thomas Huth]

Certain CPU instructions will cause an exit of the virtual
machine. Run some of these instructions to check whether
they are emulated right by KVM (or QEMU).

Signed-off-by: Thomas Huth <thuth@redhat.com>
---
 Note: This is the initial framework for running interception tests
 ... more test will be added later (like testing exceptions once
 we've got the infrastructure for this in place).

 For running this test with QEMU TCG, you need to apply the
 patch for the TEST BLOCK instruction which has not been merged yet:
 https://lists.gnu.org/archive/html/qemu-devel/2017-05/msg04413.html

 s390x/Makefile    |  1 +
 s390x/intercept.c | 99 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 100 insertions(+)
 create mode 100644 s390x/intercept.c

diff --git a/s390x/Makefile b/s390x/Makefile
index e80add0..97745a1 100644
--- a/s390x/Makefile
+++ b/s390x/Makefile
@@ -1,4 +1,5 @@
 tests = $(TEST_DIR)/selftest.elf
+tests += $(TEST_DIR)/intercept.elf
 
 all: directories test_cases
 
diff --git a/s390x/intercept.c b/s390x/intercept.c
new file mode 100644
index 0000000..b0057e3
--- /dev/null
+++ b/s390x/intercept.c
@@ -0,0 +1,99 @@
+/*
+ * Interception tests - for s390x CPU instruction that cause a VM exit
+ *
+ * Copyright (c) 2017 Red Hat Inc
+ *
+ * Authors:
+ *  Thomas Huth <thuth@redhat.com>
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU Library General Public License version 2.
+ */
+#include <libcflat.h>
+#include <asm/page.h>
+
+static uint8_t pagebuf[PAGE_SIZE * 2] __attribute__((aligned(PAGE_SIZE)));
+
+/* Test the SET PREFIX and STORE PREFIX instructions */
+static void test_prefix(void)
+{
+	uint32_t old_prefix;
+	uint32_t new_prefix = (uint32_t)(intptr_t)pagebuf;
+
+	memset(pagebuf, 0, PAGE_SIZE * 2);
+
+	/*
+	 * Temporarily change the prefix page to our buffer, and store
+	 * some facility bits there ... at least some of them should be
+	 * set in our buffer afterwards.
+	 */
+	asm volatile (
+		" stpx	%0\n"
+		" spx	%1\n"
+		" stfl	0\n"
+		" spx	%0\n"
+		: "=m"(old_prefix) : "m"(new_prefix) : "memory");
+	report("spx + stfl", pagebuf[200] != 0);
+}
+
+/* Test the STORE CPU ADDRESS instruction */
+static void test_stap(void)
+{
+	uint16_t cpuid;
+
+	cpuid = 0xffff;
+	asm volatile ("stap %0\n" : "+m"(cpuid));
+	report("get cpu id", cpuid != 0xffff);
+}
+
+/* Test the TEST BLOCK instruction */
+static void test_testblock(void)
+{
+	int cc;
+
+	memset(pagebuf, 0xaa, PAGE_SIZE);
+
+	asm volatile (
+		" lghi	0,0\n"
+		" tb	%1\n"
+		" ipm	%0\n"
+		" srl	%0,28\n"
+		: "=d" (cc)
+		: "a"(pagebuf + 0x123)
+		: "memory", "0", "cc");
+	report("page cleared",
+	       cc == 0 && pagebuf[0] == 0 &&  pagebuf[PAGE_SIZE - 1] == 0);
+}
+
+struct {
+	const char *name;
+	void (*func)(void);
+} tests[] = {
+	{ "prefix", test_prefix },
+	{ "stap", test_stap },
+	{ "testblock", test_testblock },
+	{ NULL, NULL }
+};
+
+int main(int argc, char **argv)
+{
+	int all = 0;
+	int i;
+
+	report_prefix_push("intercept");
+
+	if (argc < 2 || (argc == 2 && !strcmp(argv[1], "all")))
+		all = 1;
+
+	for (i = 0; tests[i].name != NULL; i++) {
+		report_prefix_push(tests[i].name);
+		if (all || strcmp(argv[1], tests[i].name) == 0) {
+			tests[i].func();
+		}
+		report_prefix_pop();
+	}
+
+	report_prefix_pop();
+
+	return report_summary();
+}
-- 
1.8.3.1

Re: [kvm-unit-tests PATCH] s390x: Interception tests

[David Hildenbrand]

On 29.05.2017 12:06, Thomas Huth wrote:
> Certain CPU instructions will cause an exit of the virtual
> machine. Run some of these instructions to check whether
> they are emulated right by KVM (or QEMU).
> 
> Signed-off-by: Thomas Huth <thuth@redhat.com>
> ---
>  Note: This is the initial framework for running interception tests
>  ... more test will be added later (like testing exceptions once
>  we've got the infrastructure for this in place).

Working on a patch for that, will post that soon.

> 
>  For running this test with QEMU TCG, you need to apply the
>  patch for the TEST BLOCK instruction which has not been merged yet:
>  https://lists.gnu.org/archive/html/qemu-devel/2017-05/msg04413.html
> 
>  s390x/Makefile    |  1 +
>  s390x/intercept.c | 99 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
>  2 files changed, 100 insertions(+)
>  create mode 100644 s390x/intercept.c

I think you should add that to s390x/unittests.cfg as well. (running all
tests)

> 
> diff --git a/s390x/Makefile b/s390x/Makefile
> index e80add0..97745a1 100644
> --- a/s390x/Makefile
> +++ b/s390x/Makefile
> @@ -1,4 +1,5 @@
>  tests = $(TEST_DIR)/selftest.elf
> +tests += $(TEST_DIR)/intercept.elf
>  
>  all: directories test_cases
>  
> diff --git a/s390x/intercept.c b/s390x/intercept.c
> new file mode 100644
> index 0000000..b0057e3
> --- /dev/null
> +++ b/s390x/intercept.c
> @@ -0,0 +1,99 @@
> +/*
> + * Interception tests - for s390x CPU instruction that cause a VM exit
> + *
> + * Copyright (c) 2017 Red Hat Inc
> + *
> + * Authors:
> + *  Thomas Huth <thuth@redhat.com>
> + *
> + * This code is free software; you can redistribute it and/or modify it
> + * under the terms of the GNU Library General Public License version 2.
> + */
> +#include <libcflat.h>
> +#include <asm/page.h>
> +
> +static uint8_t pagebuf[PAGE_SIZE * 2] __attribute__((aligned(PAGE_SIZE)));
> +
> +/* Test the SET PREFIX and STORE PREFIX instructions */
> +static void test_prefix(void)
> +{
> +	uint32_t old_prefix;
> +	uint32_t new_prefix = (uint32_t)(intptr_t)pagebuf;
> +
> +	memset(pagebuf, 0, PAGE_SIZE * 2);
> +
> +	/*
> +	 * Temporarily change the prefix page to our buffer, and store
> +	 * some facility bits there ... at least some of them should be
> +	 * set in our buffer afterwards.
> +	 */
> +	asm volatile (
> +		" stpx	%0\n"
> +		" spx	%1\n"
> +		" stfl	0\n"

you could add another stpx to tmp_prefix here and test below if stpx
actually return the correct value (e.g. not simply always 0).

> +		" spx	%0\n"
> +		: "=m"(old_prefix) : "m"(new_prefix) : "memory");
> +	report("spx + stfl", pagebuf[200] != 0);

I have a patch that introduced the struct lowcore and generates
asm-offsets, will post that soon.

> +}
> +
> +/* Test the STORE CPU ADDRESS instruction */
> +static void test_stap(void)
> +{
> +	uint16_t cpuid;
> +
> +	cpuid = 0xffff;

you could initialize that directly.

> +	asm volatile ("stap %0\n" : "+m"(cpuid));
> +	report("get cpu id", cpuid != 0xffff);
> +}
> +
> +/* Test the TEST BLOCK instruction */
> +static void test_testblock(void)
> +{
> +	int cc;
> +
> +	memset(pagebuf, 0xaa, PAGE_SIZE);
> +
> +	asm volatile (
> +		" lghi	0,0\n"
> +		" tb	%1\n"
> +		" ipm	%0\n"
> +		" srl	%0,28\n"
> +		: "=d" (cc)
> +		: "a"(pagebuf + 0x123)
> +		: "memory", "0", "cc");
> +	report("page cleared",
> +	       cc == 0 && pagebuf[0] == 0 &&  pagebuf[PAGE_SIZE - 1] == 0);
> +}
> +
> +struct {
> +	const char *name;
> +	void (*func)(void);
> +} tests[] = {
> +	{ "prefix", test_prefix },
> +	{ "stap", test_stap },
> +	{ "testblock", test_testblock },
> +	{ NULL, NULL }
> +};
> +
> +int main(int argc, char **argv)
> +{
> +	int all = 0;
> +	int i;
> +
> +	report_prefix_push("intercept");
> +
> +	if (argc < 2 || (argc == 2 && !strcmp(argv[1], "all")))
> +		all = 1;
> +
> +	for (i = 0; tests[i].name != NULL; i++) {
> +		report_prefix_push(tests[i].name);
> +		if (all || strcmp(argv[1], tests[i].name) == 0) {
> +			tests[i].func();
> +		}
> +		report_prefix_pop();
> +	}
> +
> +	report_prefix_pop();
> +
> +	return report_summary();
> +}
> 

Looks good to me!

-- 

Thanks,

David

Re: [kvm-unit-tests PATCH] s390x: Interception tests

[Paolo Bonzini]

On 29/05/2017 12:06, Thomas Huth wrote:
> Certain CPU instructions will cause an exit of the virtual
> machine. Run some of these instructions to check whether
> they are emulated right by KVM (or QEMU).
> 
> Signed-off-by: Thomas Huth <thuth@redhat.com>
> ---
>  Note: This is the initial framework for running interception tests
>  ... more test will be added later (like testing exceptions once
>  we've got the infrastructure for this in place).

Since you are at it, would it make sense to add some timing
infrastructure and get something like vmexit.flat in the x86 tests?

Also, the change to unittests.cfg is missing.

Thanks,

Paolo

>  For running this test with QEMU TCG, you need to apply the
>  patch for the TEST BLOCK instruction which has not been merged yet:
>  https://lists.gnu.org/archive/html/qemu-devel/2017-05/msg04413.html
> 
>  s390x/Makefile    |  1 +
>  s390x/intercept.c | 99 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
>  2 files changed, 100 insertions(+)
>  create mode 100644 s390x/intercept.c
> 
> diff --git a/s390x/Makefile b/s390x/Makefile
> index e80add0..97745a1 100644
> --- a/s390x/Makefile
> +++ b/s390x/Makefile
> @@ -1,4 +1,5 @@
>  tests = $(TEST_DIR)/selftest.elf
> +tests += $(TEST_DIR)/intercept.elf
>  
>  all: directories test_cases
>  
> diff --git a/s390x/intercept.c b/s390x/intercept.c
> new file mode 100644
> index 0000000..b0057e3
> --- /dev/null
> +++ b/s390x/intercept.c
> @@ -0,0 +1,99 @@
> +/*
> + * Interception tests - for s390x CPU instruction that cause a VM exit
> + *
> + * Copyright (c) 2017 Red Hat Inc
> + *
> + * Authors:
> + *  Thomas Huth <thuth@redhat.com>
> + *
> + * This code is free software; you can redistribute it and/or modify it
> + * under the terms of the GNU Library General Public License version 2.
> + */
> +#include <libcflat.h>
> +#include <asm/page.h>
> +
> +static uint8_t pagebuf[PAGE_SIZE * 2] __attribute__((aligned(PAGE_SIZE)));
> +
> +/* Test the SET PREFIX and STORE PREFIX instructions */
> +static void test_prefix(void)
> +{
> +	uint32_t old_prefix;
> +	uint32_t new_prefix = (uint32_t)(intptr_t)pagebuf;
> +
> +	memset(pagebuf, 0, PAGE_SIZE * 2);
> +
> +	/*
> +	 * Temporarily change the prefix page to our buffer, and store
> +	 * some facility bits there ... at least some of them should be
> +	 * set in our buffer afterwards.
> +	 */
> +	asm volatile (
> +		" stpx	%0\n"
> +		" spx	%1\n"
> +		" stfl	0\n"
> +		" spx	%0\n"
> +		: "=m"(old_prefix) : "m"(new_prefix) : "memory");
> +	report("spx + stfl", pagebuf[200] != 0);
> +}
> +
> +/* Test the STORE CPU ADDRESS instruction */
> +static void test_stap(void)
> +{
> +	uint16_t cpuid;
> +
> +	cpuid = 0xffff;
> +	asm volatile ("stap %0\n" : "+m"(cpuid));
> +	report("get cpu id", cpuid != 0xffff);
> +}
> +
> +/* Test the TEST BLOCK instruction */
> +static void test_testblock(void)
> +{
> +	int cc;
> +
> +	memset(pagebuf, 0xaa, PAGE_SIZE);
> +
> +	asm volatile (
> +		" lghi	0,0\n"
> +		" tb	%1\n"
> +		" ipm	%0\n"
> +		" srl	%0,28\n"
> +		: "=d" (cc)
> +		: "a"(pagebuf + 0x123)
> +		: "memory", "0", "cc");
> +	report("page cleared",
> +	       cc == 0 && pagebuf[0] == 0 &&  pagebuf[PAGE_SIZE - 1] == 0);
> +}
> +
> +struct {
> +	const char *name;
> +	void (*func)(void);
> +} tests[] = {
> +	{ "prefix", test_prefix },
> +	{ "stap", test_stap },
> +	{ "testblock", test_testblock },
> +	{ NULL, NULL }
> +};
> +
> +int main(int argc, char **argv)
> +{
> +	int all = 0;
> +	int i;
> +
> +	report_prefix_push("intercept");
> +
> +	if (argc < 2 || (argc == 2 && !strcmp(argv[1], "all")))
> +		all = 1;
> +
> +	for (i = 0; tests[i].name != NULL; i++) {
> +		report_prefix_push(tests[i].name);
> +		if (all || strcmp(argv[1], tests[i].name) == 0) {
> +			tests[i].func();
> +		}
> +		report_prefix_pop();
> +	}
> +
> +	report_prefix_pop();
> +
> +	return report_summary();
> +}
> 

Re: [kvm-unit-tests PATCH] s390x: Interception tests

[Thomas Huth]

On 29.05.2017 12:17, David Hildenbrand wrote:
> On 29.05.2017 12:06, Thomas Huth wrote:
[...]
>>  s390x/Makefile    |  1 +
>>  s390x/intercept.c | 99 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
>>  2 files changed, 100 insertions(+)
>>  create mode 100644 s390x/intercept.c
> 
> I think you should add that to s390x/unittests.cfg as well. (running all
> tests)

Ooops, right, forgot to add that :-/

>> diff --git a/s390x/intercept.c b/s390x/intercept.c
>> new file mode 100644
>> index 0000000..b0057e3
>> --- /dev/null
>> +++ b/s390x/intercept.c
>> @@ -0,0 +1,99 @@
>> +/*
>> + * Interception tests - for s390x CPU instruction that cause a VM exit
>> + *
>> + * Copyright (c) 2017 Red Hat Inc
>> + *
>> + * Authors:
>> + *  Thomas Huth <thuth@redhat.com>
>> + *
>> + * This code is free software; you can redistribute it and/or modify it
>> + * under the terms of the GNU Library General Public License version 2.
>> + */
>> +#include <libcflat.h>
>> +#include <asm/page.h>
>> +
>> +static uint8_t pagebuf[PAGE_SIZE * 2] __attribute__((aligned(PAGE_SIZE)));
>> +
>> +/* Test the SET PREFIX and STORE PREFIX instructions */
>> +static void test_prefix(void)
>> +{
>> +	uint32_t old_prefix;
>> +	uint32_t new_prefix = (uint32_t)(intptr_t)pagebuf;
>> +
>> +	memset(pagebuf, 0, PAGE_SIZE * 2);
>> +
>> +	/*
>> +	 * Temporarily change the prefix page to our buffer, and store
>> +	 * some facility bits there ... at least some of them should be
>> +	 * set in our buffer afterwards.
>> +	 */
>> +	asm volatile (
>> +		" stpx	%0\n"
>> +		" spx	%1\n"
>> +		" stfl	0\n"
> 
> you could add another stpx to tmp_prefix here and test below if stpx
> actually return the correct value (e.g. not simply always 0).

Good idea, will do that!

>> +		" spx	%0\n"
>> +		: "=m"(old_prefix) : "m"(new_prefix) : "memory");
>> +	report("spx + stfl", pagebuf[200] != 0);
> 
> I have a patch that introduced the struct lowcore and generates
> asm-offsets, will post that soon.

OK, I just saw your "generate asm offsets for the lowcore" on the list
... will have a look at that...

 Thomas

Re: [kvm-unit-tests PATCH] s390x: Interception tests

[Thomas Huth]

On 29.05.2017 13:01, Paolo Bonzini wrote:
> 
> On 29/05/2017 12:06, Thomas Huth wrote:
>> Certain CPU instructions will cause an exit of the virtual
>> machine. Run some of these instructions to check whether
>> they are emulated right by KVM (or QEMU).
>>
>> Signed-off-by: Thomas Huth <thuth@redhat.com>
>> ---
>>  Note: This is the initial framework for running interception tests
>>  ... more test will be added later (like testing exceptions once
>>  we've got the infrastructure for this in place).
> 
> Since you are at it, would it make sense to add some timing
> infrastructure and get something like vmexit.flat in the x86 tests?

You mean something to run the test for a certain amount of time or a
certain amount of iterations? Sure, yes, I can add that.

> Also, the change to unittests.cfg is missing.

Yes, missed that ... will add it...

 Thomas

Re: [kvm-unit-tests PATCH] s390x: Interception tests

[Paolo Bonzini]

On 29/05/2017 15:13, Thomas Huth wrote:
>>>  Note: This is the initial framework for running interception tests
>>>  ... more test will be added later (like testing exceptions once
>>>  we've got the infrastructure for this in place).
>>
>> Since you are at it, would it make sense to add some timing
>> infrastructure and get something like vmexit.flat in the x86 tests?
>
> You mean something to run the test for a certain amount of time or a
> certain amount of iterations? Sure, yes, I can add that.

Yes, see x86/vmexit.c for an example.

Paolo