Skip to contentSkip to conversation

s390x: diag10: Fixup

PATCH In review Janosch Frank 4 participants 11 messages May 28, 2025

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

cover s390x: diag10: Fixup

While reviewing Claudio's patch set I found a problem that should've been caught by the diag10 test but wasn't.

Not only does the test never check the good case but it also doesn't fence against environments where diag10 is not available. The part that makes this problematic is the fact that this test only tests priv and spec PGMs. These PGMs are presented even if no diag10 support is provided since they are also part of the base diagnose architecture.

The tests currently succeed in TCG emulation and PV, both of which do not implement this specific diagnose.

Therefore this series fences TCG & PV as well as adding a check if the page has really been cleared.

Janosch Frank (2):
s390x: diag10: Fence tcg and pv environments
s390x: diag10: Check page clear

 s390x/diag10.c      | 26 ++++++++++++++++++++++++++
 s390x/unittests.cfg |  1 +
 2 files changed, 27 insertions(+)
Signature · 1 line
2.48.1
#1 lore raw

PATCH 1/2 s390x: diag10: Fence tcg and pv environments

Diag10 isn't supported under either of these environments so let's make sure that the test bails out accordingly.

---
 s390x/diag10.c      | 15 +++++++++++++++
 s390x/unittests.cfg |  1 +
 2 files changed, 16 insertions(+)
s390x/diag10.c +15 -0
index 579a7a5d..00725f58 100644
@@ -9,6 +9,8 @@
  */
 
 #include <libcflat.h>
+#include <uv.h>
+#include <hardware.h>
 #include <asm/asm-offsets.h>
 #include <asm/interrupt.h>
 #include <asm/page.h>
@@ -95,8 +97,21 @@ static void test_priv(void)
 int main(void)
 {
 	report_prefix_push("diag10");
+
+	if (host_is_tcg()) {
+		report_skip("Test unsupported under TCG");
+		goto out;
+	}
+	if (uv_os_is_guest()) {
+		report_skip("Test unsupported under PV");
+		goto out;
+	}
+
 	test_prefix();
 	test_params();
 	test_priv();
+
+out:
+	report_prefix_pop();
 	return report_summary();
 }
s390x/unittests.cfg +1 -0
index a9af6680..9c43ab2f 100644
@@ -51,6 +51,7 @@ extra_params = -device virtio-net-ccw
 
 [diag10]
 file = diag10.elf
+accel = kvm
 
 [diag308]
 file = diag308.elf
Signature · 1 line
2.48.1
#2 lore raw

Re: [kvm-unit-tests PATCH 1/2] s390x: diag10: Fence tcg and pv environments

Quoted 2 lines · On Wed, 28 May 2025 09:13:49 +0000 Janosch Frank <[email protected]> wrote:

Diag10 isn't supported under either of these environments so let's make sure that the test bails out accordingly.

does KVM always implement diag10?

is there no other way to check whether diag10 is available?

we could, for example, try to run it "correctly" and see whether we get a Specification exception, and then fence.

Quoted 54 lines
---
 s390x/diag10.c      | 15 +++++++++++++++
 s390x/unittests.cfg |  1 +
 2 files changed, 16 insertions(+)
s390x/diag10.c +15 -0
index 579a7a5d..00725f58 100644
@@ -9,6 +9,8 @@
  */
 
 #include <libcflat.h>
+#include <uv.h>
+#include <hardware.h>
 #include <asm/asm-offsets.h>
 #include <asm/interrupt.h>
 #include <asm/page.h>
@@ -95,8 +97,21 @@ static void test_priv(void)
 int main(void)
 {
 	report_prefix_push("diag10");
+
+	if (host_is_tcg()) {
+		report_skip("Test unsupported under TCG");
+		goto out;
+	}
+	if (uv_os_is_guest()) {
+		report_skip("Test unsupported under PV");
+		goto out;
+	}
+
 	test_prefix();
 	test_params();
 	test_priv();
+
+out:
+	report_prefix_pop();
 	return report_summary();
 }
s390x/unittests.cfg +1 -0
index a9af6680..9c43ab2f 100644
@@ -51,6 +51,7 @@ extra_params = -device virtio-net-ccw
 
 [diag10]
 file = diag10.elf
+accel = kvm
 
 [diag308]
 file = diag308.elf
#3 lore raw

Re: [kvm-unit-tests PATCH 1/2] s390x: diag10: Fence tcg and pv environments

Quoted 7 lines · Claudio Imbrenda
Quoted 2 lines · On Wed, 28 May 2025 09:13:49 +0000 Janosch Frank <[email protected]> wrote:

Diag10 isn't supported under either of these environments so let's make sure that the test bails out accordingly.

does KVM always implement diag10?

It was introduced October of 2011 and there's no way of disabling it that I can see.

Quoted 2 lines

is there no other way to check whether diag10 is available?

The z/VM CP programming services doesn't specify a feature bit or any other way to check as far as I can see.

Quoted 3 lines

we could, for example, try to run it "correctly" and see whether we get a Specification exception, and then fence.

We could move the content test to the top and bail out if there's a spec. That would allow us to handle the addition of diag10 support for TCG without a test change.

But:
How likely is it that we want to implement diag10 in TCG?
My guess is that adding it to PV is even less likely.
Not running them under PV and TCG gives us back some CI time although it's likely minuscule.

Reviewed-by
#4 lore raw

Re: [kvm-unit-tests PATCH 1/2] s390x: diag10: Fence tcg and pv environments

Quoted 26 lines · On Wed, 28 May 2025 13:13:45 +0200 Janosch Frank <[email protected]> wrote:
Quoted 7 lines · Claudio Imbrenda
Quoted 2 lines · On Wed, 28 May 2025 09:13:49 +0000 Janosch Frank <[email protected]> wrote:

Diag10 isn't supported under either of these environments so let's make sure that the test bails out accordingly.

does KVM always implement diag10?

It was introduced October of 2011 and there's no way of disabling it that I can see.

Quoted 2 lines

is there no other way to check whether diag10 is available?

The z/VM CP programming services doesn't specify a feature bit or any other way to check as far as I can see.

Quoted 3 lines

we could, for example, try to run it "correctly" and see whether we get a Specification exception, and then fence.

We could move the content test to the top and bail out if there's a spec. That would allow us to handle the addition of diag10 support for TCG without a test change.

Quoted 5 lines

But:
How likely is it that we want to implement diag10 in TCG? My guess is that adding it to PV is even less likely.
Not running them under PV and TCG gives us back some CI time although it's likely minuscule.

fair points

#5 lore raw

Re: [kvm-unit-tests PATCH 1/2] s390x: diag10: Fence tcg and pv environments

In reply to Janosch Frank
Quoted 29 lines · Janosch Frank

Diag10 isn't supported under either of these environments so let's make sure that the test bails out accordingly.

---
  s390x/diag10.c      | 15 +++++++++++++++
  s390x/unittests.cfg |  1 +
  2 files changed, 16 insertions(+)
s390x/diag10.c +5 -0
index 579a7a5d..00725f58 100644
@@ -9,6 +9,8 @@
   */
 
  #include <libcflat.h>
+#include <uv.h>
+#include <hardware.h>
  #include <asm/asm-offsets.h>
  #include <asm/interrupt.h>
  #include <asm/page.h>
@@ -95,8 +97,21 @@ static void test_priv(void)
  int main(void)
  {
  	report_prefix_push("diag10");
+
+	if (host_is_tcg()) {
+		report_skip("Test unsupported under TCG");

s390x/diag10.c static void test_priv(void) · line 112

Quoted 10 lines
s390x/diag10.c +7 -0
+		goto out;
+	}
+	if (uv_os_is_guest()) {
+		report_skip("Test unsupported under PV");
+		goto out;
+	}
+
  	test_prefix();
  	test_params();
  	test_priv();

Would it make sense to run test_priv() at least for TCG/PV, too?

Thomas

#6 lore raw

Re: [kvm-unit-tests PATCH 1/2] s390x: diag10: Fence tcg and pv environments

Quoted 34 lines · Thomas Huth
Quoted 33 lines · Janosch Frank

Diag10 isn't supported under either of these environments so let's make sure that the test bails out accordingly.

---
   s390x/diag10.c      | 15 +++++++++++++++
   s390x/unittests.cfg |  1 +
   2 files changed, 16 insertions(+)
s390x/diag10.c +9 -0
index 579a7a5d..00725f58 100644
@@ -9,6 +9,8 @@
    */
 
   #include <libcflat.h>
+#include <uv.h>
+#include <hardware.h>
   #include <asm/asm-offsets.h>
   #include <asm/interrupt.h>
   #include <asm/page.h>
@@ -95,8 +97,21 @@ static void test_priv(void)
   int main(void)
   {
   	report_prefix_push("diag10");
+
+	if (host_is_tcg()) {
+		report_skip("Test unsupported under TCG");
+		goto out;
+	}
+	if (uv_os_is_guest()) {
+		report_skip("Test unsupported under PV");
Quoted 8 lines

s390x/diag10.c static void test_priv(void) · line 112

Quoted 6 lines
s390x/diag10.c +3 -0
+		goto out;
+	}
+
   	test_prefix();
   	test_params();
   	test_priv();

Would it make sense to run test_priv() at least for TCG/PV, too?

Could do but I think such tests should go into a general diag test.

The KVM code tests priv before checking the subcode and going into the specific handler. So if the priv test on diag44 succeeds, then it should succeed on any other diag even if it's not implemented.

Under PV we'd be testing HW and/or FW instead of QEMU/KVM.

#8 lore raw

PATCH 2/2 s390x: diag10: Check page clear

In reply to Janosch Frank

We should get a new page after we discarded the page. So let's check for that.

---
 s390x/diag10.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
s390x/diag10.c +11 -0
index 00725f58..b68481ad 100644
@@ -94,6 +94,16 @@ static void test_priv(void)
 	report_prefix_pop();
 }
 
+static void test_content(void)
+{
+	report_prefix_push("content");
+	memset((void *)page0, 0x42, PAGE_SIZE);
+	memset((void *)page1, 0, PAGE_SIZE);
+	diag10(page0, page0);
+	report(!memcmp((void *)page0, (void *)page1, PAGE_SIZE), "Page cleared");
+	report_prefix_pop();
+}
+
 int main(void)
 {
 	report_prefix_push("diag10");
@@ -110,6 +120,7 @@ int main(void)
 	test_prefix();
 	test_params();
 	test_priv();
+	test_content();
 
 out:
 	report_prefix_pop();
Signature · 1 line
2.48.1
Reviewed-by Tested-by on v1 2/2
#9 lore raw

Re: [kvm-unit-tests PATCH 2/2] s390x: diag10: Check page clear

Quoted 4 lines · On Wed, 28 May 2025 09:13:50 +0000 Janosch Frank <[email protected]> wrote:

We should get a new page after we discarded the page. So let's check for that.

Quoted 33 lines
---
 s390x/diag10.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
s390x/diag10.c +11 -0
index 00725f58..b68481ad 100644
@@ -94,6 +94,16 @@ static void test_priv(void)
 	report_prefix_pop();
 }
 
+static void test_content(void)
+{
+	report_prefix_push("content");
+	memset((void *)page0, 0x42, PAGE_SIZE);
+	memset((void *)page1, 0, PAGE_SIZE);
+	diag10(page0, page0);
+	report(!memcmp((void *)page0, (void *)page1, PAGE_SIZE), "Page cleared");
+	report_prefix_pop();
+}
+
 int main(void)
 {
 	report_prefix_push("diag10");
@@ -110,6 +120,7 @@ int main(void)
 	test_prefix();
 	test_params();
 	test_priv();
+	test_content();
 
 out:
 	report_prefix_pop();
Reviewed-by on v1 2/2
#10 lore raw

Re: [kvm-unit-tests PATCH 2/2] s390x: diag10: Check page clear

In reply to Janosch Frank
Quoted 15 lines · Janosch Frank

We should get a new page after we discarded the page. So let's check for that.

---
 s390x/diag10.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
s390x/diag10.c +0 -0
index 00725f58..b68481ad 100644
@@ -94,6 +94,16 @@ static void test_priv(void)
 	report_prefix_pop();
 }

s390x/diag10.c static void test_priv(void) · line 105

Quoted 10 lines
s390x/diag10.c +9 -0
 
+static void test_content(void)
+{
+	report_prefix_push("content");
+	memset((void *)page0, 0x42, PAGE_SIZE);
+	memset((void *)page1, 0, PAGE_SIZE);
+	diag10(page0, page0);
+	report(!memcmp((void *)page0, (void *)page1, PAGE_SIZE), "Page cleared");
+	report_prefix_pop();
+}

Would be nice to test that we don't clear too much, but this is clearly better than what we had before so: