On 2/23/22 16:39, Janosch Frank wrote:
On 2/23/22 14:29, Nico Boehr wrote:
css_lib extensively uses STSCH, but two more cases deserve their own
tests:
- unaligned address for SCHIB. We check for misalignment by 1 and 2
bytes.
- channel not operational
- bit 47 in SID not set
- bit 5 of PMCW flags.
As per the principles of operation, bit 5 of the PMCW flags shall be
ignored by msch and always stored as zero by stsch.
Older QEMU versions require this bit to always be zero on msch,
which is why this test may fail. A fix is available in QEMU master
commit 2df59b73e086 ("s390x/css: fix PMCW invalid mask"). >
Here's the QEMU PMCW invalid mask fix:
https://lists.nongnu.org/archive/html/qemu-s390x/2021-12/msg00100.html
Signed-off-by: Nico Boehr <nrb@xxxxxxxxxxxxx>
---
s390x/css.c | 74 +++++++++++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 74 insertions(+)
diff --git a/s390x/css.c b/s390x/css.c
index a90a0cd64e2b..021eb12573c0 100644
--- a/s390x/css.c
+++ b/s390x/css.c
@@ -496,6 +496,78 @@ static void test_ssch(void)
report_prefix_pop();
}
+static void test_stsch(void)
+{
+ const int align_to = 4;
+ struct schib schib;
+ int cc;
+
+ if (!test_device_sid) {
+ report_skip("No device");
+ return;
+ }
+
+ report_prefix_push("Unaligned");
+ for (int i = 1; i < align_to; i *= 2) {
+ report_prefix_pushf("%d", i);
+
+ expect_pgm_int();
+ stsch(test_device_sid, (struct schib *)(alignment_test_page +
i));
+ check_pgm_int_code(PGM_INT_CODE_SPECIFICATION);
+
+ report_prefix_pop();
+ }
+ report_prefix_pop();
+
+ report_prefix_push("Invalid subchannel number");
+ cc = stsch(0x0001ffff, &schib);
+ report(cc == 3, "Channel not operational");
+ report_prefix_pop();
+
+ report_prefix_push("Bit 47 in SID is zero");
+ expect_pgm_int();
+ stsch(0x0000ffff, &schib);
+ check_pgm_int_code(PGM_INT_CODE_OPERAND);
+ report_prefix_pop();
Add a comment:
No matter if the multiple-subchannel-set facility is installed or not,
bit 47 always needs to be 1.
Do we have the MSS facility?
yes
If yes, could we disable it to test the 32-47 == 0x0001 case?
AFAIK it is not enabled in the KVM unit tests
We are able to enable it, it could be done with CHSC tests.
+}
+
+static void test_pmcw_bit5(void)
+{
+ int cc;
+ uint16_t old_pmcw_flags;
I need a comment here for further reference since that behavior is
documented at the description of the schib and not where STSCH is
described:
According to architecture MSCH does ignore bit 5 of the second word but
STSCH will store bit 5 as zero.
We could check if bits 0,1 and 6,7 are also zero but I'm not sure if
that's interesting since MSCH does not ignore those bits and should
result in an operand exception when trying to set them.
@Halil, @Pierre: Any opinions?
Yes we should check this.
We often do STSCH/MSCH in a row so I think it is interesting to check it.
+
+ cc = stsch(test_device_sid, &schib);
+ if (cc) {
+ report_fail("stsch: sch %08x failed with cc=%d",
test_device_sid, cc);
+ return;
+ }
+ old_pmcw_flags = schib.pmcw.flags;
+
+ report_prefix_push("Bit 5 set");
+
+ schib.pmcw.flags = old_pmcw_flags | BIT(15 - 5);
+ cc = msch(test_device_sid, &schib);
+ report(!cc, "MSCH cc == 0");
+
+ cc = stsch(test_device_sid, &schib);
+ report(!cc, "STSCH cc == 0");
+ report(!(schib.pmcw.flags & BIT(15 - 5)), "STSCH PMCW Bit 5 is
clear");
+
+ report_prefix_pop();
+
+ report_prefix_push("Bit 5 clear");
+
+ schib.pmcw.flags = old_pmcw_flags & ~BIT(15 - 5);
+ cc = msch(test_device_sid, &schib);
+ report(!cc, "MSCH cc == 0");
+
+ cc = stsch(test_device_sid, &schib);
+ report(!cc, "STSCH cc == 0");
+ report(!(schib.pmcw.flags & BIT(15 - 5)), "STSCH PMCW Bit 5 is
clear");
+
+ report_prefix_pop();
+}
+
static struct {
const char *name;
void (*func)(void);
@@ -511,6 +583,8 @@ static struct {
{ "msch", test_msch },
{ "stcrw", test_stcrw },
{ "ssch", test_ssch },
+ { "stsch", test_stsch },
+ { "pmcw bit 5 ignored", test_pmcw_bit5 },
{ NULL, NULL }
};
--
Pierre Morel
IBM Lab Boeblingen
