| Message ID | 20190411055304.21429-1-thuth@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [kvm-unit-tests] s390x/skey: Skip the "iske" test when running under z/VM 6 | expand |
On 11.04.19 07:53, Thomas Huth wrote: > There is a known bug of z/VM 6 which causes the "iske" test to fail. > Since it is still sometimes useful to run the kvm-unit-tests in such > a nested environment, let's rather skip the "iske" test there instead > of always reporting a failure. I have seen the skey test fail as well, but only on 6.3. IIRC 6.4 was good. Do you have a bug number or an offical statement whatever (I did not found one). In general Ack. > > Signed-off-by: Thomas Huth <thuth@redhat.com> > --- > s390x/skey.c | 55 ++++++++++++++++++++++++++++++++++++++++++++++++---- > 1 file changed, 51 insertions(+), 4 deletions(-) > > diff --git a/s390x/skey.c b/s390x/skey.c > index f4894f1..b1e11af 100644 > --- a/s390x/skey.c > +++ b/s390x/skey.c > @@ -70,9 +70,51 @@ static void test_set(void) > skey.str.acc == ret.str.acc && skey.str.fp == ret.str.fp); > } > > +static inline int stsi(void *addr, int fc, int sel1, int sel2) > +{ > + register int r0 asm("0") = (fc << 28) | sel1; > + register int r1 asm("1") = sel2; > + int rc = 0; > + > + asm volatile( > + " stsi 0(%3)\n" > + " jz 0f\n" > + " lhi %1,-1\n" > + "0:\n" > + : "+d" (r0), "+d" (rc) > + : "d" (r1), "a" (addr) > + : "cc", "memory"); > + > + return rc; > +} > + > +/* Returns true if we are running under z/VM 6.x */ > +static bool check_for_zvm6(void) > +{ > + int dcbt; /* Descriptor block count */ > + int nr; > + static const unsigned char zvm6[] = { > + /* This is "z/VM 6" in EBCDIC */ > + 0xa9, 0x61, 0xe5, 0xd4, 0x40, 0x40, 0x40, 0x40, 0xf6 > + }; > + > + if (stsi(pagebuf, 3, 2, 2)) > + return false; > + > + dcbt = pagebuf[31] & 0xf; > + > + for (nr = 0; nr < dcbt; nr++) { > + if (!memcmp(&pagebuf[32 + nr * 64 + 24], zvm6, sizeof(zvm6))) > + return true; > + } > + > + return false; > +} > + > static void test_priv(void) > { > union skey skey; > + bool is_zvm6 = check_for_zvm6(); > > memset(pagebuf, 0, PAGE_SIZE * 2); > report_prefix_push("privileged"); > @@ -87,10 +129,15 @@ static void test_priv(void) > report("skey did not change on exception", skey.str.acc != 3); > > report_prefix_push("iske"); > - expect_pgm_int(); > - enter_pstate(); > - get_storage_key(page0); > - check_pgm_int_code(PGM_INT_CODE_PRIVILEGED_OPERATION); > + if (is_zvm6) { > + /* There is a known bug with z/VM 6, so skip the test there */ > + report_skip("not working on z/VM 6"); > + } else { > + expect_pgm_int(); > + enter_pstate(); > + get_storage_key(page0); > + check_pgm_int_code(PGM_INT_CODE_PRIVILEGED_OPERATION); > + } > report_prefix_pop(); > > report_prefix_pop(); >
On 11/04/2019 13.40, Christian Borntraeger wrote: > On 11.04.19 07:53, Thomas Huth wrote: >> There is a known bug of z/VM 6 which causes the "iske" test to fail. >> Since it is still sometimes useful to run the kvm-unit-tests in such >> a nested environment, let's rather skip the "iske" test there instead >> of always reporting a failure. > > I have seen the skey test fail as well, but only on 6.3. IIRC 6.4 was good. According to the information from /proc/sysinfo, we've got 6.4.0 here, and it is failing for me on that system. > Do you have a bug number or an offical statement whatever (I did not found > one). No, I only heard from Janosch that it should have been fixed in a later version ... so I was rather assuming that it has been fixed in z/VM 7 ? Thomas
diff --git a/s390x/skey.c b/s390x/skey.c index f4894f1..b1e11af 100644 --- a/s390x/skey.c +++ b/s390x/skey.c @@ -70,9 +70,51 @@ static void test_set(void) skey.str.acc == ret.str.acc && skey.str.fp == ret.str.fp); } +static inline int stsi(void *addr, int fc, int sel1, int sel2) +{ + register int r0 asm("0") = (fc << 28) | sel1; + register int r1 asm("1") = sel2; + int rc = 0; + + asm volatile( + " stsi 0(%3)\n" + " jz 0f\n" + " lhi %1,-1\n" + "0:\n" + : "+d" (r0), "+d" (rc) + : "d" (r1), "a" (addr) + : "cc", "memory"); + + return rc; +} + +/* Returns true if we are running under z/VM 6.x */ +static bool check_for_zvm6(void) +{ + int dcbt; /* Descriptor block count */ + int nr; + static const unsigned char zvm6[] = { + /* This is "z/VM 6" in EBCDIC */ + 0xa9, 0x61, 0xe5, 0xd4, 0x40, 0x40, 0x40, 0x40, 0xf6 + }; + + if (stsi(pagebuf, 3, 2, 2)) + return false; + + dcbt = pagebuf[31] & 0xf; + + for (nr = 0; nr < dcbt; nr++) { + if (!memcmp(&pagebuf[32 + nr * 64 + 24], zvm6, sizeof(zvm6))) + return true; + } + + return false; +} + static void test_priv(void) { union skey skey; + bool is_zvm6 = check_for_zvm6(); memset(pagebuf, 0, PAGE_SIZE * 2); report_prefix_push("privileged"); @@ -87,10 +129,15 @@ static void test_priv(void) report("skey did not change on exception", skey.str.acc != 3); report_prefix_push("iske"); - expect_pgm_int(); - enter_pstate(); - get_storage_key(page0); - check_pgm_int_code(PGM_INT_CODE_PRIVILEGED_OPERATION); + if (is_zvm6) { + /* There is a known bug with z/VM 6, so skip the test there */ + report_skip("not working on z/VM 6"); + } else { + expect_pgm_int(); + enter_pstate(); + get_storage_key(page0); + check_pgm_int_code(PGM_INT_CODE_PRIVILEGED_OPERATION); + } report_prefix_pop(); report_prefix_pop();
There is a known bug of z/VM 6 which causes the "iske" test to fail. Since it is still sometimes useful to run the kvm-unit-tests in such a nested environment, let's rather skip the "iske" test there instead of always reporting a failure. Signed-off-by: Thomas Huth <thuth@redhat.com> --- s390x/skey.c | 55 ++++++++++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 51 insertions(+), 4 deletions(-)