Skip to contentSkip to conversation

s390x: mvpg test

PATCH In review Claudio Imbrenda 2 participants 5 messages Mar 2, 2021 – Mar 8, 2021 KVM

lore
Share & tools
lore
#0 lore raw

v4 cover s390x: mvpg test

A simple unit test for the MVPG instruction.

The timeout is set to 10 seconds because the test should complete in a fraction of a second even on busy machines. If the test is run in VSIE and the host of the host is not handling MVPG properly, the test will probably hang.

Testing MVPG behaviour in VSIE is the main motivation for this test.

Anything related to storage keys is not tested.

v3->v4
* add memset after the first successful mvpg to make sure memory is really copied successfully
* add a comment and an additional prefix to the tests skipped when running in TCG

v2->v3
* fix copyright (2020 is over!)
* add the third patch to skip some known issues when running in TCG

v1->v2
* droppped patch 2 which introduced is_pgm();
* patch 1: replace a hardcoded value with the new macro SVC_LEAVE_PSTATE
* patch 2: clear_pgm_int() returns the old value, use that instad of is_pgm()

Claudio Imbrenda (3):
s390x: introduce leave_pstate to leave userspace
s390x: mvpg: simple test
s390x: mvpg: skip some tests when using TCG

 s390x/Makefile           |   1 +
 lib/s390x/asm/arch_def.h |   7 +
 lib/s390x/interrupt.c    |  12 +-
 s390x/mvpg.c             | 277 +++++++++++++++++++++++++++++++++++++++
 s390x/unittests.cfg      |   4 +
 5 files changed, 299 insertions(+), 2 deletions(-)
 create mode 100644 s390x/mvpg.c
Signature · 1 line
2.26.2
#1 lore raw

PATCH v4 1/3 s390x: introduce leave_pstate to leave userspace

In most testcases, we enter problem state (userspace) just to test if a privileged instruction causes a fault. In some cases, though, we need to test if an instruction works properly in userspace. This means that we do not expect a fault, and we need an orderly way to leave problem state afterwards.

This patch introduces a simple system based on the SVC instruction.

---
 lib/s390x/asm/arch_def.h |  7 +++++++
 lib/s390x/interrupt.c    | 12 ++++++++++--
 2 files changed, 17 insertions(+), 2 deletions(-)
lib/s390x/asm/arch_def.h +7 -0
index 9c4e330a..4cf8eb11 100644
@@ -173,6 +173,8 @@ struct cpuid {
 	uint64_t reserved : 15;
 };
 
+#define SVC_LEAVE_PSTATE 1
+
 static inline unsigned short stap(void)
 {
 	unsigned short cpu_address;
@@ -276,6 +278,11 @@ static inline void enter_pstate(void)
 	load_psw_mask(mask);
 }
 
+static inline void leave_pstate(void)
+{
+	asm volatile("	svc %0\n" : : "i" (SVC_LEAVE_PSTATE));
+}
+
 static inline int stsi(void *addr, int fc, int sel1, int sel2)
 {
 	register int r0 asm("0") = (fc << 28) | sel1;
lib/s390x/interrupt.c +10 -2
index 1ce36073..d0567845 100644
@@ -188,6 +188,14 @@ int unregister_io_int_func(void (*f)(void))
 
 void handle_svc_int(void)
 {
-	report_abort("Unexpected supervisor call interrupt: on cpu %d at %#lx",
-		     stap(), lc->svc_old_psw.addr);
+	uint16_t code = lc->svc_int_code;
+
+	switch (code) {
+	case SVC_LEAVE_PSTATE:
+		lc->svc_old_psw.mask &= ~PSW_MASK_PSTATE;
+		break;
+	default:
+		report_abort("Unexpected supervisor call interrupt: code %#x on cpu %d at %#lx",
+			      code, stap(), lc->svc_old_psw.addr);
+	}
 }
Signature · 1 line
2.26.2
#2 lore raw

PATCH v4 2/3 s390x: mvpg: simple test

In reply to Claudio Imbrenda

A simple unit test for the MVPG instruction.

The timeout is set to 10 seconds because the test should complete in a fraction of a second even on busy machines. If the test is run in VSIE and the host of the host is not handling MVPG properly, the test will probably hang.

Testing MVPG behaviour in VSIE is the main motivation for this test.

Anything related to storage keys is not tested.

---
 s390x/Makefile      |   1 +
 s390x/mvpg.c        | 267 ++++++++++++++++++++++++++++++++++++++++++++
 s390x/unittests.cfg |   4 +
 3 files changed, 272 insertions(+)
 create mode 100644 s390x/mvpg.c
s390x/Makefile +1 -0
index 08d85c9f..770eaa0b 100644
@@ -20,6 +20,7 @@ tests += $(TEST_DIR)/sclp.elf
 tests += $(TEST_DIR)/css.elf
 tests += $(TEST_DIR)/uv-guest.elf
 tests += $(TEST_DIR)/sie.elf
+tests += $(TEST_DIR)/mvpg.elf
 
 tests_binary = $(patsubst %.elf,%.bin,$(tests))
 ifneq ($(HOST_KEY_DOCUMENT),)
s390x/mvpg.c new file +267 -0

267 lines not on this page · raw message

s390x/unittests.cfg +4 -0
index 2298be6c..9f81a608 100644
@@ -99,3 +99,7 @@ file = uv-guest.elf
 
 [sie]
 file = sie.elf
+
+[mvpg]
+file = mvpg.elf
+timeout = 10
Signature · 1 line
2.26.2
#3 lore raw

PATCH v4 3/3 s390x: mvpg: skip some tests when using TCG

In reply to Claudio Imbrenda

TCG is known to fail these tests, so add an explicit exception to skip them.

Once TCG has been fixed, it will be enough to revert this patch.

---
 s390x/mvpg.c | 34 ++++++++++++++++++++++------------
 1 file changed, 22 insertions(+), 12 deletions(-)
s390x/mvpg.c +22 -12
index 1776ec6e..5743d5b6 100644
@@ -20,6 +20,7 @@
 #include <smp.h>
 #include <alloc_page.h>
 #include <bitops.h>
+#include <vm.h>
 
 /* Used to build the appropriate test values for register 0 */
 #define KFC(x) ((x) << 10)
@@ -225,20 +226,29 @@ static void test_mmu_prot(void)
 	report(clear_pgm_int() == PGM_INT_CODE_PROTECTION, "destination read only");
 	fresh += PAGE_SIZE;
 
-	protect_page(fresh, PAGE_ENTRY_I);
-	cc = mvpg(CCO, fresh, source);
-	report(cc == 1, "destination invalid");
-	fresh += PAGE_SIZE;
+	/* Known issue in TCG: CCO flag is not honoured */
+	if (vm_is_tcg()) {
+		report_prefix_push("TCG");
+		report_skip("destination invalid");
+		report_skip("source invalid");
+		report_skip("source and destination invalid");
+		report_prefix_pop();
+	} else {
+		protect_page(fresh, PAGE_ENTRY_I);
+		cc = mvpg(CCO, fresh, source);
+		report(cc == 1, "destination invalid");
+		fresh += PAGE_SIZE;
 
-	protect_page(source, PAGE_ENTRY_I);
-	cc = mvpg(CCO, fresh, source);
-	report(cc == 2, "source invalid");
-	fresh += PAGE_SIZE;
+		protect_page(source, PAGE_ENTRY_I);
+		cc = mvpg(CCO, fresh, source);
+		report(cc == 2, "source invalid");
+		fresh += PAGE_SIZE;
 
-	protect_page(fresh, PAGE_ENTRY_I);
-	cc = mvpg(CCO, fresh, source);
-	report(cc == 2, "source and destination invalid");
-	fresh += PAGE_SIZE;
+		protect_page(fresh, PAGE_ENTRY_I);
+		cc = mvpg(CCO, fresh, source);
+		report(cc == 2, "source and destination invalid");
+		fresh += PAGE_SIZE;
+	}
 
 	unprotect_page(source, PAGE_ENTRY_I);
 	report_prefix_pop();
Signature · 1 line
2.26.2
#4 lore raw
In reply to Claudio Imbrenda
Quoted 10 lines · Claudio Imbrenda

A simple unit test for the MVPG instruction.

The timeout is set to 10 seconds because the test should complete in a fraction of a second even on busy machines. If the test is run in VSIE and the host of the host is not handling MVPG properly, the test will probably hang.

Testing MVPG behaviour in VSIE is the main motivation for this test.

Anything related to storage keys is not tested.

Thanks, picked.

Quoted 29 lines

v3->v4
* add memset after the first successful mvpg to make sure memory is really copied successfully
* add a comment and an additional prefix to the tests skipped when running in TCG

v2->v3
* fix copyright (2020 is over!)
* add the third patch to skip some known issues when running in TCG

v1->v2
* droppped patch 2 which introduced is_pgm();
* patch 1: replace a hardcoded value with the new macro SVC_LEAVE_PSTATE
* patch 2: clear_pgm_int() returns the old value, use that instad of is_pgm()

Claudio Imbrenda (3):
s390x: introduce leave_pstate to leave userspace
s390x: mvpg: simple test
s390x: mvpg: skip some tests when using TCG

 s390x/Makefile           |   1 +
 lib/s390x/asm/arch_def.h |   7 +
 lib/s390x/interrupt.c    |  12 +-
 s390x/mvpg.c             | 277 +++++++++++++++++++++++++++++++++++++++
 s390x/unittests.cfg      |   4 +
 5 files changed, 299 insertions(+), 2 deletions(-)
 create mode 100644 s390x/mvpg.c