Hi Zenghui,
On 3/30/20 2:06 PM, Zenghui Yu wrote:
> Hi Eric,
>
> On 2020/3/20 17:24, Eric Auger wrote:
>> Add two new migration tests. One testing the migration of
>> a topology where collection were unmapped. The second test
>> checks the migration of the pending table.
>>
>> Signed-off-by: Eric Auger <eric.auger@xxxxxxxxxx>
>
> [...]
>
>> @@ -659,6 +678,15 @@ static int its_prerequisites(int nb_cpus)
>> return 0;
>> }
>> +static void set_lpi(struct its_device *dev, u32 eventid, u32 physid,
>> + struct its_collection *col)
>> +{
>> + assert(dev && col);
>> +
>> + its_send_mapti(dev, physid, eventid, col);
>> + gicv3_lpi_set_config(physid, LPI_PROP_DEFAULT);
>> +}
>
> I'd say we can just drop this helper and open-code it anywhere
> necessarily. The name 'set_lpi' doesn't tell me too much about
> what has been implemented inside the helper.
>
>> +
>> /*
>> * Setup the configuration for those mappings:
>> * dev_id=2 event=20 -> vcpu 3, intid=8195
>> @@ -790,6 +818,108 @@ static void test_its_migration(void)
>> its_send_int(dev7, 255);
>> check_lpi_stats("dev7/eventid=255 triggers LPI 8196 on PE #2
>> after migration");
>> }
>> +
>> +static void test_migrate_unmapped_collection(void)
>> +{
>> + struct its_collection *col;
>> + struct its_device *dev2, *dev7;
>> + int pe0 = 0;
>> + u8 config;
>> +
>> + if (its_setup1())
>> + return;
>> +
>> + col = its_create_collection(pe0, pe0);
>> + dev2 = its_get_device(2);
>> + dev7 = its_get_device(7);
>> +
>> + /* MAPTI with the collection unmapped */
>> + set_lpi(dev2, 0, 8192, col);
>
> ... and it's only invoked here.
>
>> +
>> + puts("Now migrate the VM, then press a key to continue...\n");
>> + (void)getchar();
>> + report_info("Migration complete");
>> +
>> + /* on the destination, map the collection */
>> + its_send_mapc(col, true);
>> + its_send_invall(col);
>> +
>> + lpi_stats_expect(2, 8196);
>> + its_send_int(dev7, 255);
>> + check_lpi_stats("dev7/eventid= 255 triggered LPI 8196 on PE #2");
>> +
>> + config = gicv3_lpi_get_config(8192);
>> + report(config == LPI_PROP_DEFAULT,
>> + "Config of LPI 8192 was properly migrated");
>> +
>> + lpi_stats_expect(pe0, 8192);
>> + its_send_int(dev2, 0);
>> + check_lpi_stats("dev2/eventid = 0 triggered LPI 8192 on PE0");
>> +}
>> +
>> +static void test_its_pending_migration(void)
>> +{
>> + struct its_device *dev;
>> + struct its_collection *collection[2];
>> + int *expected = malloc(nr_cpus * sizeof(int));
>> + int pe0 = nr_cpus - 1, pe1 = nr_cpus - 2;
>> + u64 pendbaser;
>> + void *ptr;
>> + int i;
>> +
>> + if (its_prerequisites(4))
>> + return;
>> +
>> + dev = its_create_device(2 /* dev id */, 8 /* nb_ites */);
>> + its_send_mapd(dev, true);
>> +
>> + collection[0] = its_create_collection(pe0, pe0);
>> + collection[1] = its_create_collection(pe1, pe1);
>> + its_send_mapc(collection[0], true);
>> + its_send_mapc(collection[1], true);
>> +
>> + /* disable lpi at redist level */
>> + gicv3_lpi_rdist_disable(pe0);
>> + gicv3_lpi_rdist_disable(pe1);
>> +
>> + /* lpis are interleaved inbetween the 2 PEs */
>> + for (i = 0; i < 256; i++) {
>> + struct its_collection *col = i % 2 ? collection[0] :
>> + collection[1];
>> + int vcpu = col->target_address >> 16;
>> +
>> + its_send_mapti(dev, LPI(i), i, col);
>> + gicv3_lpi_set_config(LPI(i), LPI_PROP_DEFAULT);
>> + gicv3_lpi_set_clr_pending(vcpu, LPI(i), true);
>> + }
>> + its_send_invall(collection[0]);
>> + its_send_invall(collection[1]);
>> +
>> + /* Clear the PTZ bit on each pendbaser */
>> +
>> + expected[pe0] = 128;
>> + expected[pe1] = 128;
>
> Do we need to initialize expected[] for other PEs? Or it has always
> been zeroed by the kvm-unit-tests implementation of malloc()?
>
>> +
>> + ptr = gicv3_data.redist_base[pe0] + GICR_PENDBASER;
>> + pendbaser = readq(ptr);
>> + writeq(pendbaser & ~GICR_PENDBASER_PTZ, ptr);
>> +
>> + ptr = gicv3_data.redist_base[pe1] + GICR_PENDBASER;
>> + pendbaser = readq(ptr);
>> + writeq(pendbaser & ~GICR_PENDBASER_PTZ, ptr);
>> +
>> + gicv3_lpi_rdist_enable(pe0);
>> + gicv3_lpi_rdist_enable(pe1);
>
> I don't know how the migration gets implemented in kvm-unit-tests.
> But is there any guarantee that the LPIs will only be triggered on the
> destination side? As once the EnableLPIs bit becomes 1, VGIC will start
> reading the pending bit in guest memory and potentially injecting LPIs
> into the target vcpu (in the source side).
I expect some LPIs to hit on source and some others to hit on the
destination. To me, this does not really matter as long as the handlers
gets called and accumulate the stats. Given the number of LPIs, we will
at least test the migration of some of the pending bits and especially
adjacent ones. It does work as it allows to test your fix:
ca185b260951 KVM: arm/arm64: vgic: Don't rely on the wrong pending table
Thanks
Eric
>
>> +
>> + puts("Now migrate the VM, then press a key to continue...\n");
>> + (void)getchar();
>> + report_info("Migration complete");
>> +
>> + /* let's wait for the 256 LPIs to be handled */
>> + mdelay(1000);
>> +
>> + check_lpi_hits(expected, "128 LPIs on both PE0 and PE1 after
>> migration");
>> +}
>
>
> Thanks,
> Zenghui
>
>
