Skip to contentSkip to conversation

CSS Mesurement Block

PATCH Superseded Pierre Morel 2 participants 16 messages Mar 9, 2021 – Mar 11, 2021 KVM

v5 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
#0 lore raw

v5 cover CSS Mesurement Block

We tests the update of the Mesurement Block (MB) format 0 and format 1 using a serie of senseid requests.

*Warning*: One of the tests for format-1 will unexpectedly fail for QEMU elf unless the QEMU patch "css: SCHIB measurement block origin must be aligned" is applied.
With Protected Virtualization, the PGM is correctly recognized.

The MB format 1 is only provided if the Extended mesurement Block feature is available.

This feature is exposed by the CSS characteristics general features stored by the Store Channel Subsystem Characteristics CHSC command, consequently, we implement the CHSC instruction call and the SCSC CHSC command.

In order to ease the writing of new tests using:
- interrupt
- enablement of a subchannel
- multiple I/O on a subchannel

We do the following simplifications:
- we create a CSS initialization routine
- we register the I/O interrupt handler on CSS initialization
- we do not enable or disable a subchannel in the senseid test, assuming this test is done after the enable test, this allows to create traffic using the SSCH used by senseid.
- failures not part of the feature under test will stop the tests.
- we add a css_enabled() function to test if a subchannel is enabled.

*note*:
I rearranged the use of the senseid for the tests, by not modifying the existing test and having a dedicated senseid() function for the purpose of the tests.
I think that it is in the rigght way so I kept the RB and ACK on the simplification, there are less changes, if it is wrong from me I suppose I will see this in the comments.
Since the changed are moved inside the fmt0 test which is not approved for now I hope it is OK.

Regards,
Pierre

Pierre Morel (6):
s390x: css: Store CSS Characteristics
s390x: css: simplifications of the tests
s390x: css: extending the subchannel modifying functions
s390x: css: implementing Set CHannel Monitor
s390x: css: testing measurement block format 0
s390x: css: testing measurement block format 1

 lib/s390x/css.h     | 115 ++++++++++++++++++++-
 lib/s390x/css_lib.c | 236 ++++++++++++++++++++++++++++++++++++++++----
 s390x/css.c         | 216 ++++++++++++++++++++++++++++++++++++++--
 3 files changed, 540 insertions(+), 27 deletions(-)
Signature · 98 lines
2.17.1

changelog:

from v4:

- no modification of test_senseid() in the simplification patch,
  replaced by a new function in the fmt0 patch
  (Pierre)

- suppress report_info() in the interrupt routine.
  (Pierre)

- in chsc_scsc, rationalise the reserved area name.
  (Janosch)

- use assert for errors that should never happen
  like register_io_int_func()
  (Janosch)

- Changed css_general_feature to css_test_general_feature
  same for css_test_chsc_feature
  (Janosch)

- use report_abort() for hard failures in tests
  (Janosch, Connie)


from v3:

- stay coherent and use uintX_t types in css.h
  (Janosch)

- reworking test versus library to leave the reports
  mostly inside the tests
  (Janosh)

- Restructured some tests to use booleans instead of int
  this simplifies testing.
  (inspired by comments from Janosch)

- rewordings and orthograph (Measurement !!!)
  (Connie)

- stop measurement at the channel too before freeing the
  MB memory.
  (Connie)

- split out the subchannel-modifying functions into a
  separate patch
  (Connie)

- move msch_with_wrong_fmt1_mbo() from the library
  into the tests
  (Janosch)

- Suppress redundancy in the prefixes of format0/1 tests
  (Janosch)

from v2:

- stop measurement before freeing memory
  (Connie)

- added a css_disable_mb()
  (Connie)

- several rewriting of comments and commits
  (Connie)

- modified eroneous test for MB index for fmt0
  (Pierre)

- modified eroneous test for unaligned MBO for fmt1
  (Pierre)

- several small coding style issues corrected
  (Pierre)

from v1:

- check the return code of CHSC
  (Connie)

- reporting in css_init
  (Connie)

- added braces when a loop contains several statement
  (Thomas)

- changed retval to success in boolean function
  (Thomas)

- suppress goto retries
  (thomas)

- rewording and use correct return types in css_enabled
  (Janosch)
#1 lore raw

PATCH v5 1/6 s390x: css: Store CSS Characteristics

CSS characteristics exposes the features of the Channel SubSystem. Let's use Store Channel Subsystem Characteristics to retrieve the features of the CSS.

---
 lib/s390x/css.h     | 66 ++++++++++++++++++++++++++++++++
 lib/s390x/css_lib.c | 93 ++++++++++++++++++++++++++++++++++++++++++++-
 s390x/css.c         |  8 ++++
 3 files changed, 166 insertions(+), 1 deletion(-)
lib/s390x/css.h +66 -0
index 3e57445..fe68b7d 100644
@@ -288,4 +288,70 @@ int css_residual_count(unsigned int schid);
 void enable_io_isc(uint8_t isc);
 int wait_and_check_io_completion(int schid);
 
+/*
+ * CHSC definitions
+ */
+struct chsc_header {
+	uint16_t len;
+	uint16_t code;
+};
+
+/* Store Channel Subsystem Characteristics */
+struct chsc_scsc {
+	struct chsc_header req;
+	uint16_t req_fmt;
+	uint8_t cssid;
+	uint8_t reserved[9];
+	struct chsc_header res;
+	uint32_t res_fmt;
+	uint64_t general_char[255];
+	uint64_t chsc_char[254];
+};
+
+extern struct chsc_scsc *chsc_scsc;
+#define CHSC_SCSC	0x0010
+#define CHSC_SCSC_LEN	0x0010
+
+bool get_chsc_scsc(void);
+
+#define CSS_GENERAL_FEAT_BITLEN	(255 * 64)
+#define CSS_CHSC_FEAT_BITLEN	(254 * 64)
+
+#define CHSC_SCSC	0x0010
+#define CHSC_SCSC_LEN	0x0010
+
+#define CHSC_ERROR	0x0000
+#define CHSC_RSP_OK	0x0001
+#define CHSC_RSP_INVAL	0x0002
+#define CHSC_RSP_REQERR	0x0003
+#define CHSC_RSP_ENOCMD	0x0004
+#define CHSC_RSP_NODATA	0x0005
+#define CHSC_RSP_SUP31B	0x0006
+#define CHSC_RSP_EFRMT	0x0007
+#define CHSC_RSP_ECSSID	0x0008
+#define CHSC_RSP_ERFRMT	0x0009
+#define CHSC_RSP_ESSID	0x000A
+#define CHSC_RSP_EBUSY	0x000B
+#define CHSC_RSP_MAX	0x000B
+
+static inline int _chsc(void *p)
+{
+	int cc;
+
+	asm volatile(" .insn   rre,0xb25f0000,%2,0\n"
+		     " ipm     %0\n"
+		     " srl     %0,28\n"
+		     : "=d" (cc), "=m" (p)
+		     : "d" (p), "m" (p)
+		     : "cc");
+
+	return cc;
+}
+
+bool chsc(void *p, uint16_t code, uint16_t len);
+
+#include <bitops.h>
+#define css_general_feature(bit) test_bit_inv(bit, chsc_scsc->general_char)
+#define css_chsc_feature(bit) test_bit_inv(bit, chsc_scsc->chsc_char)
+
 #endif
lib/s390x/css_lib.c +92 -1

103 lines not on this page · raw message

s390x/css.c +8 -0
index 1a61a5c..12036b3 100644
@@ -14,6 +14,7 @@
 #include <string.h>
 #include <interrupt.h>
 #include <asm/arch_def.h>
+#include <alloc_page.h>
 
 #include <malloc_io.h>
 #include <css.h>
@@ -140,10 +141,17 @@ error_senseid:
 	unregister_io_int_func(css_irq_io);
 }
 
+static void css_init(void)
+{
+	report(get_chsc_scsc(), "Store Channel Characteristics");
+}
+
 static struct {
 	const char *name;
 	void (*func)(void);
 } tests[] = {
+	/* The css_init test is needed to initialize the CSS Characteristics */
+	{ "initialize CSS (chsc)", css_init },
 	{ "enumerate (stsch)", test_enumerate },
 	{ "enable (msch)", test_enable },
 	{ "sense (ssch/tsch)", test_sense },
Signature · 1 line
2.17.1
#2 lore raw

PATCH v5 2/6 s390x: css: simplifications of the tests

In reply to Pierre Morel

In order to ease the writing of tests based on:
- interrupt
- enabling a subchannel
- using multiple I/O on a channel without disabling it

We do the following simplifications:
- the I/O interrupt handler is registered on CSS initialization
- We do not enable again a subchannel in senseid if it is already enabled
- we add a css_enabled() function to test if a subchannel is enabled

---
 lib/s390x/css.h     |  1 +
 lib/s390x/css_lib.c | 37 ++++++++++++++++++++++++++-----------
 s390x/css.c         |  3 +++
 3 files changed, 30 insertions(+), 11 deletions(-)
lib/s390x/css.h +1 -0
index fe68b7d..71d91f0 100644
@@ -278,6 +278,7 @@ int css_enumerate(void);
 
 #define IO_SCH_ISC      3
 int css_enable(int schid, int isc);
+bool css_enabled(int schid);
 
 /* Library functions */
 int start_ccw1_chain(unsigned int sid, struct ccw1 *ccw);
lib/s390x/css_lib.c +26 -11
index f46e871..41134dc 100644
@@ -161,6 +161,31 @@ out:
 	return schid;
 }
 
+/*
+ * css_enabled: report if the subchannel is enabled
+ * @schid: Subchannel Identifier
+ * Return value:
+ *   true if the subchannel is enabled
+ *   false otherwise
+ */
+bool css_enabled(int schid)
+{
+	struct pmcw *pmcw = &schib.pmcw;
+	int cc;
+
+	cc = stsch(schid, &schib);
+	if (cc) {
+		report_info("stsch: updating sch %08x failed with cc=%d",
+			    schid, cc);
+		return false;
+	}
+
+	if (!(pmcw->flags & PMCW_ENABLE)) {
+		report_info("stsch: sch %08x not enabled", schid);
+		return false;
+	}
+	return true;
+}
 /*
  * css_enable: enable the subchannel with the specified ISC
  * @schid: Subchannel Identifier
@@ -210,18 +235,8 @@ retry:
 	/*
 	 * Read the SCHIB again to verify the enablement
 	 */
-	cc = stsch(schid, &schib);
-	if (cc) {
-		report_info("stsch: updating sch %08x failed with cc=%d",
-			    schid, cc);
-		return cc;
-	}
-
-	if ((pmcw->flags & flags) == flags) {
-		report_info("stsch: sch %08x successfully modified after %d retries",
-			    schid, retry_count);
+	if (css_enabled(schid))
 		return 0;
-	}
 
 	if (retry_count++ < MAX_ENABLE_RETRIES) {
 		mdelay(10); /* the hardware was not ready, give it some time */
s390x/css.c +3 -0
index 12036b3..069c2be 100644
@@ -143,6 +143,9 @@ error_senseid:
 
 static void css_init(void)
 {
+	assert(register_io_int_func(css_irq_io) == 0);
+	lowcore_ptr->io_int_param = 0;
+
 	report(get_chsc_scsc(), "Store Channel Characteristics");
 }
 
Signature · 1 line
2.17.1
#3 lore raw

PATCH v5 3/6 s390x: css: extending the subchannel modifying functions

In reply to Pierre Morel

To enable or disable measurement we will need specific modifications on the subchannel.

---
 lib/s390x/css.h     |   9 +++-
 lib/s390x/css_lib.c | 100 ++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 108 insertions(+), 1 deletion(-)
lib/s390x/css.h +8 -1
index 71d91f0..3c50fa8 100644
@@ -82,6 +82,8 @@ struct pmcw {
 	uint32_t intparm;
 #define PMCW_DNV	0x0001
 #define PMCW_ENABLE	0x0080
+#define PMCW_MBUE	0x0010
+#define PMCW_DCTME	0x0008
 #define PMCW_ISC_MASK	0x3800
 #define PMCW_ISC_SHIFT	11
 	uint16_t flags;
@@ -94,6 +96,7 @@ struct pmcw {
 	uint8_t  pom;
 	uint8_t  pam;
 	uint8_t  chpid[8];
+#define PMCW_MBF1	0x0004
 	uint32_t flags2;
 };
 #define PMCW_CHANNEL_TYPE(pmcw) (pmcw->flags2 >> 21)
@@ -101,7 +104,8 @@ struct pmcw {
 struct schib {
 	struct pmcw pmcw;
 	struct scsw scsw;
-	uint8_t  md[12];
+	uint64_t mbo;
+	uint8_t  md[4];
 } __attribute__ ((aligned(4)));
 
 struct irb {
@@ -355,4 +359,7 @@ bool chsc(void *p, uint16_t code, uint16_t len);
 #define css_general_feature(bit) test_bit_inv(bit, chsc_scsc->general_char)
 #define css_chsc_feature(bit) test_bit_inv(bit, chsc_scsc->chsc_char)
 
+bool css_enable_mb(int sid, uint64_t mb, uint16_t mbi, uint16_t flg, bool fmt1);
+bool css_disable_mb(int schid);
+
 #endif
lib/s390x/css_lib.c +100 -0

106 lines not on this page · raw message

Signature · 1 line
2.17.1
#4 lore raw

PATCH v5 4/6 s390x: css: implementing Set CHannel Monitor

In reply to Pierre Morel

We implement the call of the Set CHannel Monitor instruction, starting the monitoring of the all Channel Sub System, and initializing channel subsystem monitoring.

Initial tests report the presence of the extended measurement block feature, and verify the error reporting of the hypervisor for SCHM.

---
 lib/s390x/css.h     | 16 ++++++++++++++--
 lib/s390x/css_lib.c |  4 ++--
 s390x/css.c         | 35 +++++++++++++++++++++++++++++++++++
 3 files changed, 51 insertions(+), 4 deletions(-)
lib/s390x/css.h +14 -2
index 3c50fa8..7158423 100644
@@ -309,6 +309,7 @@ struct chsc_scsc {
 	uint8_t reserved[9];
 	struct chsc_header res;
 	uint32_t res_fmt;
+#define CSSC_EXTENDED_MEASUREMENT_BLOCK 48
 	uint64_t general_char[255];
 	uint64_t chsc_char[254];
 };
@@ -356,8 +357,19 @@ static inline int _chsc(void *p)
 bool chsc(void *p, uint16_t code, uint16_t len);
 
 #include <bitops.h>
-#define css_general_feature(bit) test_bit_inv(bit, chsc_scsc->general_char)
-#define css_chsc_feature(bit) test_bit_inv(bit, chsc_scsc->chsc_char)
+#define css_test_general_feature(bit) test_bit_inv(bit, chsc_scsc->general_char)
+#define css_test_chsc_feature(bit) test_bit_inv(bit, chsc_scsc->chsc_char)
+
+#define SCHM_DCTM	1 /* activate Device Connection TiMe */
+#define SCHM_MBU	2 /* activate Measurement Block Update */
+
+static inline void schm(void *mbo, unsigned int flags)
+{
+	register void *__gpr2 asm("2") = mbo;
+	register long __gpr1 asm("1") = flags;
+
+	asm("schm" : : "d" (__gpr2), "d" (__gpr1));
+}
 
 bool css_enable_mb(int sid, uint64_t mb, uint16_t mbi, uint16_t flg, bool fmt1);
 bool css_disable_mb(int schid);
lib/s390x/css_lib.c +2 -2
index 77b39c7..95d9a78 100644
@@ -94,7 +94,7 @@ bool get_chsc_scsc(void)
 		return false;
 
 	for (i = 0, p = buffer; i < CSS_GENERAL_FEAT_BITLEN; i++) {
-		if (css_general_feature(i)) {
+		if (css_test_general_feature(i)) {
 			n = snprintf(p, sizeof(buffer), "%d,", i);
 			p += n;
 		}
@@ -102,7 +102,7 @@ bool get_chsc_scsc(void)
 	report_info("General features: %s", buffer);
 
 	for (i = 0, p = buffer; i < CSS_CHSC_FEAT_BITLEN; i++) {
-		if (css_chsc_feature(i)) {
+		if (css_test_chsc_feature(i)) {
 			n = snprintf(p, sizeof(buffer), "%d,", i);
 			p += n;
 		}
s390x/css.c +35 -0
index 069c2be..a763814 100644
@@ -149,6 +149,40 @@ static void css_init(void)
 	report(get_chsc_scsc(), "Store Channel Characteristics");
 }
 
+static void test_schm(void)
+{
+	if (css_test_general_feature(CSSC_EXTENDED_MEASUREMENT_BLOCK))
+		report_info("Extended measurement block available");
+
+	/* bits 59-63 of MB address must be 0  if MBU is defined */
+	report_prefix_push("Unaligned operand");
+	expect_pgm_int();
+	schm((void *)0x01, SCHM_MBU);
+	check_pgm_int_code(PGM_INT_CODE_OPERAND);
+	report_prefix_pop();
+
+	/* bits 36-61 of register 1 (flags) must be 0 */
+	report_prefix_push("Bad flags");
+	expect_pgm_int();
+	schm(NULL, 0xfffffffc);
+	check_pgm_int_code(PGM_INT_CODE_OPERAND);
+	report_prefix_pop();
+
+	/* SCHM is a privilege operation */
+	report_prefix_push("Privilege");
+	enter_pstate();
+	expect_pgm_int();
+	schm(NULL, SCHM_MBU);
+	check_pgm_int_code(PGM_INT_CODE_PRIVILEGED_OPERATION);
+	report_prefix_pop();
+
+	/* Normal operation */
+	report_prefix_push("Normal operation");
+	schm(NULL, SCHM_MBU);
+	report(1, "SCHM call without address");
+	report_prefix_pop();
+}
+
 static struct {
 	const char *name;
 	void (*func)(void);
@@ -158,6 +192,7 @@ static struct {
 	{ "enumerate (stsch)", test_enumerate },
 	{ "enable (msch)", test_enable },
 	{ "sense (ssch/tsch)", test_sense },
+	{ "measurement block (schm)", test_schm },
 	{ NULL, NULL }
 };
 
Signature · 1 line
2.17.1
#5 lore raw

Re: [kvm-unit-tests PATCH v5 4/6] s390x: css: implementing Set CHannel Monitor

Quoted 29 lines · On Tue, 9 Mar 2021 13:51:15 +0100 Pierre Morel <[email protected]> wrote:

We implement the call of the Set CHannel Monitor instruction, starting the monitoring of the all Channel Sub System, and initializing channel subsystem monitoring.

Initial tests report the presence of the extended measurement block feature, and verify the error reporting of the hypervisor for SCHM.

---
 lib/s390x/css.h     | 16 ++++++++++++++--
 lib/s390x/css_lib.c |  4 ++--
 s390x/css.c         | 35 +++++++++++++++++++++++++++++++++++
 3 files changed, 51 insertions(+), 4 deletions(-)
lib/s390x/css.h +1 -0
index 3c50fa8..7158423 100644
@@ -309,6 +309,7 @@ struct chsc_scsc {
 	uint8_t reserved[9];
 	struct chsc_header res;
 	uint32_t res_fmt;
+#define CSSC_EXTENDED_MEASUREMENT_BLOCK 48
 	uint64_t general_char[255];
 	uint64_t chsc_char[254];
 };

lib/s390x/css.h static inline int _chsc(void *p) · line 361

Quoted 7 lines
lib/s390x/css.h +2 -2
@@ -356,8 +357,19 @@ static inline int _chsc(void *p)
 bool chsc(void *p, uint16_t code, uint16_t len);
 
 #include <bitops.h>
-#define css_general_feature(bit) test_bit_inv(bit, chsc_scsc->general_char)
-#define css_chsc_feature(bit) test_bit_inv(bit, chsc_scsc->chsc_char)
+#define css_test_general_feature(bit) test_bit_inv(bit, chsc_scsc->general_char)
+#define css_test_chsc_feature(bit) test_bit_inv(bit, chsc_scsc->chsc_char)

I think the renaming belongs in patch 1?

Quoted 19 lines
lib/s390x/css.h +11 -0
+
+#define SCHM_DCTM	1 /* activate Device Connection TiMe */
+#define SCHM_MBU	2 /* activate Measurement Block Update */
+
+static inline void schm(void *mbo, unsigned int flags)
+{
+	register void *__gpr2 asm("2") = mbo;
+	register long __gpr1 asm("1") = flags;
+
+	asm("schm" : : "d" (__gpr2), "d" (__gpr1));
+}
 
 bool css_enable_mb(int sid, uint64_t mb, uint16_t mbi, uint16_t flg, bool fmt1);
 bool css_disable_mb(int schid);
lib/s390x/css_lib.c +0 -0
index 77b39c7..95d9a78 100644

lib/s390x/css_lib.c bool get_chsc_scsc(void) · line 97

Quoted 5 lines
lib/s390x/css_lib.c +1 -1
@@ -94,7 +94,7 @@ bool get_chsc_scsc(void)
 		return false;
 
 	for (i = 0, p = buffer; i < CSS_GENERAL_FEAT_BITLEN; i++) {
-		if (css_general_feature(i)) {
+		if (css_test_general_feature(i)) {

and here...

lib/s390x/css_lib.c bool get_chsc_scsc(void) · line 105

Quoted 9 lines
lib/s390x/css_lib.c +1 -1
 			n = snprintf(p, sizeof(buffer), "%d,", i);
 			p += n;
 		}
@@ -102,7 +102,7 @@ bool get_chsc_scsc(void)
 	report_info("General features: %s", buffer);
 
 	for (i = 0, p = buffer; i < CSS_CHSC_FEAT_BITLEN; i++) {
-		if (css_chsc_feature(i)) {
+		if (css_test_chsc_feature(i)) {

...and here.

Quoted 3 lines
lib/s390x/css_lib.c +0 -0
 			n = snprintf(p, sizeof(buffer), "%d,", i);
 			p += n;
 		}
#6 lore raw

Re: [kvm-unit-tests PATCH v5 4/6] s390x: css: implementing Set CHannel Monitor

Quoted 34 lines · Cornelia Huck
Quoted 31 lines · On Tue, 9 Mar 2021 13:51:15 +0100 Pierre Morel <[email protected]> wrote:

We implement the call of the Set CHannel Monitor instruction, starting the monitoring of the all Channel Sub System, and initializing channel subsystem monitoring.

Initial tests report the presence of the extended measurement block feature, and verify the error reporting of the hypervisor for SCHM.

---
  lib/s390x/css.h     | 16 ++++++++++++++--
  lib/s390x/css_lib.c |  4 ++--
  s390x/css.c         | 35 +++++++++++++++++++++++++++++++++++
  3 files changed, 51 insertions(+), 4 deletions(-)
lib/s390x/css.h +1 -0
index 3c50fa8..7158423 100644
@@ -309,6 +309,7 @@ struct chsc_scsc {
  	uint8_t reserved[9];
  	struct chsc_header res;
  	uint32_t res_fmt;
+#define CSSC_EXTENDED_MEASUREMENT_BLOCK 48
  	uint64_t general_char[255];
  	uint64_t chsc_char[254];
  };
@@ -356,8 +357,19 @@ static inline int _chsc(void *p)
  bool chsc(void *p, uint16_t code, uint16_t len);
 
Quoted 7 lines

lib/s390x/css.h static inline int _chsc(void *p) · line 361

Quoted 5 lines
lib/s390x/css.h +2 -2
  #include <bitops.h>
-#define css_general_feature(bit) test_bit_inv(bit, chsc_scsc->general_char)
-#define css_chsc_feature(bit) test_bit_inv(bit, chsc_scsc->chsc_char)
+#define css_test_general_feature(bit) test_bit_inv(bit, chsc_scsc->general_char)
+#define css_test_chsc_feature(bit) test_bit_inv(bit, chsc_scsc->chsc_char)

I think the renaming belongs in patch 1?

grr, yes obviously :)

thanks,
Pierre

Signature · 2 lines
Pierre Morel
IBM Lab Boeblingen
#7 lore raw

PATCH v5 5/6 s390x: css: testing measurement block format 0

In reply to Pierre Morel

We test the update of the measurement block format 0, the measurement block origin is calculated from the mbo argument used by the SCHM instruction and the offset calculated using the measurement block index of the SCHIB.

---
 lib/s390x/css.h     | 12 ++++++
 lib/s390x/css_lib.c |  4 --
 s390x/css.c         | 95 ++++++++++++++++++++++++++++++++++++++++-----
 3 files changed, 98 insertions(+), 13 deletions(-)
lib/s390x/css.h +12 -0
index 7158423..335bc70 100644
@@ -374,4 +374,16 @@ static inline void schm(void *mbo, unsigned int flags)
 bool css_enable_mb(int sid, uint64_t mb, uint16_t mbi, uint16_t flg, bool fmt1);
 bool css_disable_mb(int schid);
 
+struct measurement_block_format0 {
+	uint16_t ssch_rsch_count;
+	uint16_t sample_count;
+	uint32_t device_connect_time;
+	uint32_t function_pending_time;
+	uint32_t device_disconnect_time;
+	uint32_t cu_queuing_time;
+	uint32_t device_active_only_time;
+	uint32_t device_busy_time;
+	uint32_t initial_cmd_resp_time;
+};
+
 #endif
lib/s390x/css_lib.c +0 -4
index 95d9a78..8f09383 100644
@@ -365,10 +365,6 @@ void css_irq_io(void)
 		       lowcore_ptr->io_int_param, sid);
 		goto pop;
 	}
-	report_info("subsys_id_word: %08x io_int_param %08x io_int_word %08x",
-			lowcore_ptr->subsys_id_word,
-			lowcore_ptr->io_int_param,
-			lowcore_ptr->io_int_word);
 	report_prefix_pop();
 
 	report_prefix_push("tsch");
s390x/css.c +86 -9

124 lines not on this page · raw message

Signature · 1 line
2.17.1
#8 lore raw

Re: [kvm-unit-tests PATCH v5 5/6] s390x: css: testing measurement block format 0

Quoted 12 lines · On Tue, 9 Mar 2021 13:51:16 +0100 Pierre Morel <[email protected]> wrote:

We test the update of the measurement block format 0, the measurement block origin is calculated from the mbo argument used by the SCHM instruction and the offset calculated using the measurement block index of the SCHIB.

---
 lib/s390x/css.h     | 12 ++++++
 lib/s390x/css_lib.c |  4 --
 s390x/css.c         | 95 ++++++++++++++++++++++++++++++++++++++++-----
 3 files changed, 98 insertions(+), 13 deletions(-)

(...)

lib/s390x/css_lib.c void css_irq_io(void) · line 371

Quoted 12 lines
lib/s390x/css_lib.c +0 -4
index 95d9a78..8f09383 100644
@@ -365,10 +365,6 @@ void css_irq_io(void)
 		       lowcore_ptr->io_int_param, sid);
 		goto pop;
 	}
-	report_info("subsys_id_word: %08x io_int_param %08x io_int_word %08x",
-			lowcore_ptr->subsys_id_word,
-			lowcore_ptr->io_int_param,
-			lowcore_ptr->io_int_word);

Hm, why are you removing it? If you are doing some general cleanup, it probably belongs into patch 2?

Quoted 8 lines
lib/s390x/css_lib.c +0 -0
 	report_prefix_pop();
 
 	report_prefix_push("tsch");
s390x/css.c +0 -0
index a763814..b63826e 100644

s390x/css.c static void test_sense(void) · line 82

Quoted 9 lines
s390x/css.c +0 -6
@@ -74,18 +74,12 @@ static void test_sense(void)
 		return;
 	}
 
-	ret = register_io_int_func(css_irq_io);
-	if (ret) {
-		report(0, "Could not register IRQ handler");
-		return;
-	}
-

This (and the cleanup changes) definitely belongs into patch 2.

Quoted 17 lines
s390x/css.c +2 -3
 	lowcore_ptr->io_int_param = 0;
 
 	senseid = alloc_io_mem(sizeof(*senseid), 0);
 	if (!senseid) {
 		report(0, "Allocation of senseid");
-		goto error_senseid;
+		return;
 	}
 
 	ccw = ccw_alloc(CCW_CMD_SENSE_ID, senseid, sizeof(*senseid), CCW_F_SLI);
@@ -137,8 +131,24 @@ error:
 	free_io_mem(ccw, sizeof(*ccw));
 error_ccw:
 	free_io_mem(senseid, sizeof(*senseid));
-error_senseid:
-	unregister_io_int_func(css_irq_io);
+}

s390x/css.c error: · line 144

Quoted 10 lines
s390x/css.c +10 -0
+
+static void sense_id(void)
+{
+	struct ccw1 *ccw;
+
+	senseid = alloc_io_mem(sizeof(*senseid), 0);
+	assert(senseid);
+
+	ccw = ccw_alloc(CCW_CMD_SENSE_ID, senseid, sizeof(*senseid), CCW_F_SLI);
+	assert(ccw);

You're allocating senseid and ccw every time... wouldn't it be better to allocate them once and pass them in as a parameter? (Not that it should matter much, I guess.)

Quoted 91 lines
s390x/css.c +74 -0
+
+	assert(!start_ccw1_chain(test_device_sid, ccw));
+
+	assert(wait_and_check_io_completion(test_device_sid) >= 0);
+
+	free_io_mem(ccw, sizeof(*ccw));
+	free_io_mem(senseid, sizeof(*senseid));
 }
 
 static void css_init(void)
@@ -183,6 +193,72 @@ static void test_schm(void)
 	report_prefix_pop();
 }
 
+#define SCHM_UPDATE_CNT 10
+static bool start_measuring(uint64_t mbo, uint16_t mbi, bool fmt1)
+{
+	int i;
+
+	if (!css_enable_mb(test_device_sid, mbo, mbi, PMCW_MBUE, fmt1)) {
+		report_abort("Enabling measurement block failed");
+		return false;
+	}
+
+	for (i = 0; i < SCHM_UPDATE_CNT; i++)
+		sense_id();
+
+	return true;
+}
+
+/*
+ * test_schm_fmt0:
+ * With measurement block format 0 a memory space is shared
+ * by all subchannels, each subchannel can provide an index
+ * for the measurement block facility to store the measurements.
+ */
+static void test_schm_fmt0(void)
+{
+	struct measurement_block_format0 *mb0;
+	int shared_mb_size = 2 * sizeof(struct measurement_block_format0);
+
+	if (!test_device_sid) {
+		report_skip("No device");
+		return;
+	}
+
+	/* Allocate zeroed Measurement block */
+	mb0 = alloc_io_mem(shared_mb_size, 0);
+	if (!mb0) {
+		report_abort("measurement_block_format0 allocation failed");
+		return;
+	}
+
+	schm(NULL, 0); /* Stop any previous measurement */
+	schm(mb0, SCHM_MBU);
+
+	/* Expect success */
+	report_prefix_push("Valid MB address and index 0");
+	report(start_measuring(0, 0, false) &&
+	       mb0->ssch_rsch_count == SCHM_UPDATE_CNT,
+	       "SSCH measured %d", mb0->ssch_rsch_count);
+	report_prefix_pop();
+
+	/* Clear the measurement block for the next test */
+	memset(mb0, 0, shared_mb_size);
+
+	/* Expect success */
+	report_prefix_push("Valid MB address and index 1");
+	if (start_measuring(0, 1, false))
+		report(mb0[1].ssch_rsch_count == SCHM_UPDATE_CNT,
+		       "SSCH measured %d", mb0[1].ssch_rsch_count);
+	report_prefix_pop();
+
+	/* Stop the measurement */
+	css_disable_mb(test_device_sid);
+	schm(NULL, 0);
+
+	free_io_mem(mb0, shared_mb_size);
+}
+
 static struct {
 	const char *name;
 	void (*func)(void);
@@ -193,6 +269,7 @@ static struct {
 	{ "enable (msch)", test_enable },
 	{ "sense (ssch/tsch)", test_sense },
 	{ "measurement block (schm)", test_schm },
+	{ "measurement block format0", test_schm_fmt0 },
 	{ NULL, NULL }
 };
 
#9 lore raw

Re: [kvm-unit-tests PATCH v5 5/6] s390x: css: testing measurement block format 0

Quoted 18 lines · Cornelia Huck
Quoted 12 lines · On Tue, 9 Mar 2021 13:51:16 +0100 Pierre Morel <[email protected]> wrote:

We test the update of the measurement block format 0, the measurement block origin is calculated from the mbo argument used by the SCHM instruction and the offset calculated using the measurement block index of the SCHIB.

---
  lib/s390x/css.h     | 12 ++++++
  lib/s390x/css_lib.c |  4 --
  s390x/css.c         | 95 ++++++++++++++++++++++++++++++++++++++++-----
  3 files changed, 98 insertions(+), 13 deletions(-)

(...)

Quoted 15 lines

lib/s390x/css_lib.c void css_irq_io(void) · line 371

Quoted 12 lines
lib/s390x/css_lib.c +0 -4
index 95d9a78..8f09383 100644
@@ -365,10 +365,6 @@ void css_irq_io(void)
  		       lowcore_ptr->io_int_param, sid);
  		goto pop;
  	}
-	report_info("subsys_id_word: %08x io_int_param %08x io_int_word %08x",
-			lowcore_ptr->subsys_id_word,
-			lowcore_ptr->io_int_param,
-			lowcore_ptr->io_int_word);

Hm, why are you removing it? If you are doing some general cleanup, it probably belongs into patch 2?

Yes, right.
I remove it because there are too many informations shown when the trafic becomes more intense. this was not intended to be here first, I must have forget to remove it after debugging during the css series round.

Quoted 12 lines
Quoted 12 lines
lib/s390x/css_lib.c +0 -0
  	report_prefix_pop();
 
  	report_prefix_push("tsch");
s390x/css.c +0 -0
index a763814..b63826e 100644
@@ -74,18 +74,12 @@ static void test_sense(void)
  		return;
  	}
 
Quoted 8 lines

s390x/css.c static void test_sense(void) · line 82

Quoted 6 lines
s390x/css.c +0 -6
-	ret = register_io_int_func(css_irq_io);
-	if (ret) {
-		report(0, "Could not register IRQ handler");
-		return;
-	}
-

This (and the cleanup changes) definitely belongs into patch 2.

yes too

Quoted 24 lines
Quoted 24 lines
s390x/css.c +8 -3
  	lowcore_ptr->io_int_param = 0;
 
  	senseid = alloc_io_mem(sizeof(*senseid), 0);
  	if (!senseid) {
  		report(0, "Allocation of senseid");
-		goto error_senseid;
+		return;
  	}
 
  	ccw = ccw_alloc(CCW_CMD_SENSE_ID, senseid, sizeof(*senseid), CCW_F_SLI);
@@ -137,8 +131,24 @@ error:
  	free_io_mem(ccw, sizeof(*ccw));
  error_ccw:
  	free_io_mem(senseid, sizeof(*senseid));
-error_senseid:
-	unregister_io_int_func(css_irq_io);
+}
+
+static void sense_id(void)
+{
+	struct ccw1 *ccw;
+
+	senseid = alloc_io_mem(sizeof(*senseid), 0);
Quoted 8 lines

s390x/css.c error: · line 144

Quoted 4 lines
s390x/css.c +4 -0
+	assert(senseid);
+
+	ccw = ccw_alloc(CCW_CMD_SENSE_ID, senseid, sizeof(*senseid), CCW_F_SLI);
+	assert(ccw);

You're allocating senseid and ccw every time... wouldn't it be better to allocate them once and pass them in as a parameter? (Not that it should matter much, I guess.)

OK, since I rework the patch I can change this too I guess. Thanks.

Regards,
Pierre

Signature · 2 lines
Pierre Morel
IBM Lab Boeblingen
#10 lore raw

PATCH v5 6/6 s390x: css: testing measurement block format 1

In reply to Pierre Morel

Measurement block format 1 is made available by the extended measurement block facility and is indicated in the SCHIB by the bit in the PMCW.

The MBO is specified in the SCHIB of each channel and the MBO defined by the SCHM instruction is ignored.

The test of the MB format 1 is just skipped if the feature is not available.

---
 lib/s390x/css.h     | 15 +++++++++
 lib/s390x/css_lib.c |  2 +-
 s390x/css.c         | 75 +++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 91 insertions(+), 1 deletion(-)
lib/s390x/css.h +15 -0
index 335bc70..7e3d261 100644
@@ -107,6 +107,7 @@ struct schib {
 	uint64_t mbo;
 	uint8_t  md[4];
 } __attribute__ ((aligned(4)));
+extern struct schib schib;
 
 struct irb {
 	struct scsw scsw;
@@ -386,4 +387,18 @@ struct measurement_block_format0 {
 	uint32_t initial_cmd_resp_time;
 };
 
+struct measurement_block_format1 {
+	uint32_t ssch_rsch_count;
+	uint32_t sample_count;
+	uint32_t device_connect_time;
+	uint32_t function_pending_time;
+	uint32_t device_disconnect_time;
+	uint32_t cu_queuing_time;
+	uint32_t device_active_only_time;
+	uint32_t device_busy_time;
+	uint32_t initial_cmd_resp_time;
+	uint32_t irq_delay_time;
+	uint32_t irq_prio_delay_time;
+};
+
 #endif
lib/s390x/css_lib.c +1 -1
index 8f09383..efc7057 100644
@@ -19,7 +19,7 @@
 #include <malloc_io.h>
 #include <css.h>
 
-static struct schib schib;
+struct schib schib;
 struct chsc_scsc *chsc_scsc;
 
 static const char * const chsc_rsp_description[] = {
s390x/css.c +75 -0

87 lines not on this page · raw message

Signature · 1 line
2.17.1
Reviewed-by on v5 6/6
#11 lore raw

Re: [kvm-unit-tests PATCH v5 6/6] s390x: css: testing measurement block format 1

Quoted 10 lines · On Tue, 9 Mar 2021 13:51:17 +0100 Pierre Morel <[email protected]> wrote:

Measurement block format 1 is made available by the extended measurement block facility and is indicated in the SCHIB by the bit in the PMCW.

The MBO is specified in the SCHIB of each channel and the MBO defined by the SCHM instruction is ignored.

The test of the MB format 1 is just skipped if the feature is not available.

Quoted 6 lines
---
 lib/s390x/css.h     | 15 +++++++++
 lib/s390x/css_lib.c |  2 +-
 s390x/css.c         | 75 +++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 91 insertions(+), 1 deletion(-)
#12 lore raw

Re: [kvm-unit-tests PATCH v5 6/6] s390x: css: testing measurement block format 1

Quoted 13 lines · Cornelia Huck
Quoted 10 lines · On Tue, 9 Mar 2021 13:51:17 +0100 Pierre Morel <[email protected]> wrote:

Measurement block format 1 is made available by the extended measurement block facility and is indicated in the SCHIB by the bit in the PMCW.

The MBO is specified in the SCHIB of each channel and the MBO defined by the SCHM instruction is ignored.

The test of the MB format 1 is just skipped if the feature is not available.

Quoted 9 lines
Quoted 6 lines
---
  lib/s390x/css.h     | 15 +++++++++
  lib/s390x/css_lib.c |  2 +-
  s390x/css.c         | 75 +++++++++++++++++++++++++++++++++++++++++++++
  3 files changed, 91 insertions(+), 1 deletion(-)

Thanks,
Pierre

Signature · 2 lines
Pierre Morel
IBM Lab Boeblingen
#13 lore raw
In reply to Pierre Morel
Quoted 6 lines · On Tue, 9 Mar 2021 13:51:11 +0100 Pierre Morel <[email protected]> wrote:

We tests the update of the Mesurement Block (MB) format 0 and format 1 using a serie of senseid requests.

*Warning*: One of the tests for format-1 will unexpectedly fail for QEMU elf unless the QEMU patch "css: SCHIB measurement block origin must be aligned" is applied.

That one has hit QEMU master by now.

Quoted 24 lines

With Protected Virtualization, the PGM is correctly recognized.

The MB format 1 is only provided if the Extended mesurement Block feature is available.

This feature is exposed by the CSS characteristics general features stored by the Store Channel Subsystem Characteristics CHSC command, consequently, we implement the CHSC instruction call and the SCSC CHSC command.

In order to ease the writing of new tests using:
- interrupt
- enablement of a subchannel
- multiple I/O on a subchannel

We do the following simplifications:
- we create a CSS initialization routine
- we register the I/O interrupt handler on CSS initialization
- we do not enable or disable a subchannel in the senseid test, assuming this test is done after the enable test, this allows to create traffic using the SSCH used by senseid.
- failures not part of the feature under test will stop the tests.
- we add a css_enabled() function to test if a subchannel is enabled.

Quoted 9 lines

*note*:
I rearranged the use of the senseid for the tests, by not modifying the existing test and having a dedicated senseid() function for the purpose of the tests.
I think that it is in the rigght way so I kept the RB and ACK on the simplification, there are less changes, if it is wrong from me I suppose I will see this in the comments.
Since the changed are moved inside the fmt0 test which is not approved for now I hope it is OK.

I'll double-check, but I think it should be ok.

Quoted 17 lines

Regards,
Pierre

Pierre Morel (6):
s390x: css: Store CSS Characteristics
s390x: css: simplifications of the tests
s390x: css: extending the subchannel modifying functions
s390x: css: implementing Set CHannel Monitor
s390x: css: testing measurement block format 0
s390x: css: testing measurement block format 1

 lib/s390x/css.h     | 115 ++++++++++++++++++++-
 lib/s390x/css_lib.c | 236 ++++++++++++++++++++++++++++++++++++++++----
 s390x/css.c         | 216 ++++++++++++++++++++++++++++++++++++++--
 3 files changed, 540 insertions(+), 27 deletions(-)
#14 lore raw
Quoted 46 lines · On Tue, 9 Mar 2021 17:54:01 +0100 Cornelia Huck <[email protected]> wrote:
Quoted 6 lines · On Tue, 9 Mar 2021 13:51:11 +0100 Pierre Morel <[email protected]> wrote:

We tests the update of the Mesurement Block (MB) format 0 and format 1 using a serie of senseid requests.

*Warning*: One of the tests for format-1 will unexpectedly fail for QEMU elf unless the QEMU patch "css: SCHIB measurement block origin must be aligned" is applied.

That one has hit QEMU master by now.

Quoted 24 lines

With Protected Virtualization, the PGM is correctly recognized.

The MB format 1 is only provided if the Extended mesurement Block feature is available.

This feature is exposed by the CSS characteristics general features stored by the Store Channel Subsystem Characteristics CHSC command, consequently, we implement the CHSC instruction call and the SCSC CHSC command.

In order to ease the writing of new tests using:
- interrupt
- enablement of a subchannel
- multiple I/O on a subchannel

We do the following simplifications:
- we create a CSS initialization routine
- we register the I/O interrupt handler on CSS initialization
- we do not enable or disable a subchannel in the senseid test, assuming this test is done after the enable test, this allows to create traffic using the SSCH used by senseid.
- failures not part of the feature under test will stop the tests.
- we add a css_enabled() function to test if a subchannel is enabled.

Quoted 9 lines

*note*:
I rearranged the use of the senseid for the tests, by not modifying the existing test and having a dedicated senseid() function for the purpose of the tests.
I think that it is in the rigght way so I kept the RB and ACK on the simplification, there are less changes, if it is wrong from me I suppose I will see this in the comments.
Since the changed are moved inside the fmt0 test which is not approved for now I hope it is OK.

Quoted 1 line

I'll double-check, but I think it should be ok.

...that said, I found some reordering issues, but nothing major; generally, it looks good to me.

Quoted 19 lines
Quoted 18 lines

Regards,
Pierre

Pierre Morel (6):
s390x: css: Store CSS Characteristics
s390x: css: simplifications of the tests
s390x: css: extending the subchannel modifying functions
s390x: css: implementing Set CHannel Monitor
s390x: css: testing measurement block format 0
s390x: css: testing measurement block format 1

 lib/s390x/css.h     | 115 ++++++++++++++++++++-
 lib/s390x/css_lib.c | 236 ++++++++++++++++++++++++++++++++++++++++----
 s390x/css.c         | 216 ++++++++++++++++++++++++++++++++++++++--
 3 files changed, 540 insertions(+), 27 deletions(-)
#15 lore raw
Quoted 49 lines · Cornelia Huck
Quoted 46 lines · On Tue, 9 Mar 2021 17:54:01 +0100 Cornelia Huck <[email protected]> wrote:
Quoted 6 lines · On Tue, 9 Mar 2021 13:51:11 +0100 Pierre Morel <[email protected]> wrote:

We tests the update of the Mesurement Block (MB) format 0 and format 1 using a serie of senseid requests.

*Warning*: One of the tests for format-1 will unexpectedly fail for QEMU elf unless the QEMU patch "css: SCHIB measurement block origin must be aligned" is applied.

That one has hit QEMU master by now.

Quoted 33 lines

With Protected Virtualization, the PGM is correctly recognized.

The MB format 1 is only provided if the Extended mesurement Block feature is available.

This feature is exposed by the CSS characteristics general features stored by the Store Channel Subsystem Characteristics CHSC command, consequently, we implement the CHSC instruction call and the SCSC CHSC command.

In order to ease the writing of new tests using:
- interrupt
- enablement of a subchannel
- multiple I/O on a subchannel

We do the following simplifications:
- we create a CSS initialization routine
- we register the I/O interrupt handler on CSS initialization
- we do not enable or disable a subchannel in the senseid test, assuming this test is done after the enable test, this allows to create traffic using the SSCH used by senseid.
- failures not part of the feature under test will stop the tests.
- we add a css_enabled() function to test if a subchannel is enabled.

*note*:
I rearranged the use of the senseid for the tests, by not modifying the existing test and having a dedicated senseid() function for the purpose of the tests.
I think that it is in the rigght way so I kept the RB and ACK on the simplification, there are less changes, if it is wrong from me I suppose I will see this in the comments.
Since the changed are moved inside the fmt0 test which is not approved for now I hope it is OK.

Quoted 4 lines
Quoted 1 line

I'll double-check, but I think it should be ok.

...that said, I found some reordering issues, but nothing major; generally, it looks good to me.

Thanks,
so I do the reordering and respin for hopefully the right series :)

Regards,
Pierre

Signature · 2 lines
Pierre Morel
IBM Lab Boeblingen