Re: [kvm-unit-tests PATCH v1 2/2] s390x: add migration test for storage keys

[Claudio Imbrenda <imbrenda@linux.ibm.com>]

On Fri, 13 May 2022 13:04:34 +0200
Janis Schoetterl-Glausch <scgl@linux.ibm.com> wrote:

> On 5/12/22 16:01, Nico Boehr wrote:
> > Upon migration, we expect storage keys being set by the guest to be preserved,
> > so add a test for it.
> > 
> > We keep 128 pages and set predictable storage keys. Then, we migrate and check
> > they can be read back and the respective access restrictions are in place when
> > the access key in the PSW doesn't match.
> > 
> > TCG currently doesn't implement key-controlled protection, see
> > target/s390x/mmu_helper.c, function mmu_handle_skey(), hence add the relevant
> > tests as xfails.
> > 
> > Signed-off-by: Nico Boehr <nrb@linux.ibm.com>
> > ---
> >  s390x/Makefile         |  1 +
> >  s390x/migration-skey.c | 98 ++++++++++++++++++++++++++++++++++++++++++
> >  s390x/unittests.cfg    |  4 ++
> >  3 files changed, 103 insertions(+)
> >  create mode 100644 s390x/migration-skey.c
> > 
> > diff --git a/s390x/Makefile b/s390x/Makefile
> > index a8e04aa6fe4d..f8ea594b641d 100644
> > --- a/s390x/Makefile
> > +++ b/s390x/Makefile
> > @@ -32,6 +32,7 @@ tests += $(TEST_DIR)/epsw.elf
> >  tests += $(TEST_DIR)/adtl-status.elf
> >  tests += $(TEST_DIR)/migration.elf
> >  tests += $(TEST_DIR)/pv-attest.elf
> > +tests += $(TEST_DIR)/migration-skey.elf
> >  
> >  pv-tests += $(TEST_DIR)/pv-diags.elf
> >  
> > diff --git a/s390x/migration-skey.c b/s390x/migration-skey.c
> > new file mode 100644
> > index 000000000000..6f3053d8ab40
> > --- /dev/null
> > +++ b/s390x/migration-skey.c
> > @@ -0,0 +1,98 @@
> > +/* SPDX-License-Identifier: GPL-2.0-only */
> > +/*
> > + * Storage Key migration tests
> > + *
> > + * Copyright IBM Corp. 2022
> > + *
> > + * Authors:
> > + *  Nico Boehr <nrb@linux.ibm.com>
> > + */
> > +
> > +#include <libcflat.h>
> > +#include <asm/facility.h>
> > +#include <asm/page.h>
> > +#include <asm/mem.h>
> > +#include <asm/interrupt.h>
> > +#include <hardware.h>
> > +
> > +#define NUM_PAGES 128
> > +static uint8_t pagebuf[NUM_PAGES][PAGE_SIZE] __attribute__((aligned(PAGE_SIZE)));
> > +
> > +static void test_migration(void)
> > +{
> > +	int i, key_to_set;
> > +	uint8_t *page;
> > +	union skey expected_key, actual_key, mismatching_key;  
> 
> I would tend to scope those to the bodies of the respective loop,
> but I don't know if that's in accordance with the coding style.

I don't think this is specified explicitly; personally I have a light
preference for declaring everything upfront (like here), but again,
this is not a big deal for me (and maybe Janosch and Thomas should
also chime in and tell what their preference is)

> > +
> > +	for (i = 0; i < NUM_PAGES; i++) {
> > +		/*
> > +		 * Storage keys are 7 bit, lowest bit is always returned as zero
> > +		 * by iske
> > +		 */
> > +		key_to_set = i * 2;
> > +		set_storage_key(pagebuf + i, key_to_set, 1);  
> 
> Why not just pagebuf[i]?
> > +	}
> > +
> > +	puts("Please migrate me, then press return\n");
> > +	(void)getchar();
> > +
> > +	for (i = 0; i < NUM_PAGES; i++) {
> > +		report_prefix_pushf("page %d", i);
> > +
> > +		page = &pagebuf[i][0];
> > +		actual_key.val = get_storage_key(page);
> > +		expected_key.val = i * 2;
> > +
> > +		/* ignore reference bit */
> > +		actual_key.str.rf = 0;
> > +		expected_key.str.rf = 0;
> > +
> > +		report(actual_key.val == expected_key.val, "expected_key=0x%x actual_key=0x%x", expected_key.val, actual_key.val);
> > +
> > +		/* ensure access key doesn't match storage key and is never zero */
> > +		mismatching_key.str.acc = expected_key.str.acc < 15 ? expected_key.str.acc + 1 : 1;
> > +		*page = 0xff;
> > +
> > +		expect_pgm_int();
> > +		asm volatile (
> > +			/* set access key */
> > +			"spka 0(%[mismatching_key])\n"
> > +			/* try to write page */
> > +			"mvi 0(%[page]), 42\n"
> > +			/* reset access key */
> > +			"spka 0\n"
> > +			:
> > +			: [mismatching_key] "a"(mismatching_key.val),
> > +			  [page] "a"(page)
> > +			: "memory"
> > +		);
> > +		check_pgm_int_code_xfail(host_is_tcg(), PGM_INT_CODE_PROTECTION);
> > +		report_xfail(host_is_tcg(), *page == 0xff, "no store occured");  
> 
> What are you testing with this bit? If storage keys are really effective after the migration?
> I'm wondering if using tprot would not be better, it should simplify the code a lot.
> Plus you'd easily test for fetch protection, too.

on the other hand you could have tprot successful, but then not honour
the protection it indicates (I don't know how TPROT is implemented in
TCG)

to be fair, this test is only about checking that storage keys are
correctly migrated, maybe the check for actual protection is out of
scope

> > +
> > +		report_prefix_pop();
> > +	}
> > +}
> > +
> > +int main(void)
> > +{
> > +	report_prefix_push("migration-skey");
> > +	if (test_facility(169)) {
> > +		report_skip("storage key removal facility is active");
> > +
> > +		/*
> > +		 * If we just exit and don't ask migrate_cmd to migrate us, it
> > +		 * will just hang forever. Hence, also ask for migration when we
> > +		 * skip this test alltogether.  
> 
> s/alltogether/altogether/
> 
> > +		 */
> > +		puts("Please migrate me, then press return\n");
> > +		(void)getchar();
> > +
> > +		goto done;
> > +	}
> > +
> > +	test_migration();
> > +
> > +done:
> > +	report_prefix_pop();
> > +	return report_summary();
> > +}
> > diff --git a/s390x/unittests.cfg b/s390x/unittests.cfg
> > index b456b2881448..1e851d8e3dd8 100644
> > --- a/s390x/unittests.cfg
> > +++ b/s390x/unittests.cfg
> > @@ -176,3 +176,7 @@ extra_params = -cpu qemu,gs=off,vx=off
> >  file = migration.elf
> >  groups = migration
> >  smp = 2
> > +
> > +[migration-skey]
> > +file = migration-skey.elf
> > +groups = migration  
>