| Message ID | 20171215101651.13911-5-david@gibson.dropbear.id.au |
|---|---|
| State | New |
| Headers | show |
| Series | Improvements to pxe-test | expand |
On 15.12.2017 11:16, David Gibson wrote: > Previously virtio-net was only tested for ppc64 in "slow" mode. That > doesn't make much sense since virtio-net is used much more often in > practice than the spapr-vlan device which was tested always. So, move > virtio-net to always be tested on ppc64. Actually, I just moved virtio-net on ppc64 to the slow category a couple of months ago: https://git.qemu.org/?p=qemu.git;a=commitdiff;h=ab06ec43577177a442e8e5 This has been done since some people complained that SLOF is incredibly slow in in TCG mode and thus there is the risk of hitting the timeout on slow systems if they are overloaded. Things will hopefully get a little bit better with the "Use tcg_gen_lookup_and_goto_ptr" patch that you've queued in your ppc-for-2.12 branch, but still, the test is very slow on ppc64, so I'd rather keep it in the "slow" category. (virtio-net is already tested in big endian mode on s390x anyway, so we at least got some endianess test coverage for that NIC already). > We had no tests at all for the q35 machine, which doesn't seem wise given > it's increasing prominence. Add a couple of tests for it, including > testing the newer e1000e adapter. That sounds like a good idea! I wonder whether we should also test some "old" machine versions (in "slow" mode only, of course), like "pc-0.12" for example, to make sure that there are no regressions with old machines? Thomas
On Fri, Dec 15, 2017 at 11:51:18AM +0100, Thomas Huth wrote: > On 15.12.2017 11:16, David Gibson wrote: > > Previously virtio-net was only tested for ppc64 in "slow" mode. That > > doesn't make much sense since virtio-net is used much more often in > > practice than the spapr-vlan device which was tested always. So, move > > virtio-net to always be tested on ppc64. > > Actually, I just moved virtio-net on ppc64 to the slow category a couple > of months ago: > > https://git.qemu.org/?p=qemu.git;a=commitdiff;h=ab06ec43577177a442e8e5 > > This has been done since some people complained that SLOF is incredibly > slow in in TCG mode and thus there is the risk of hitting the timeout on > slow systems if they are overloaded. Hm. If we're really concerned about the speed, I'm more inclined to move spapr-vlan to slow than virtio-net. virtio-net is now a much more common case. > Things will hopefully get a little bit better with the "Use > tcg_gen_lookup_and_goto_ptr" patch that you've queued in your > ppc-for-2.12 branch, Just now merged, as it happens. > but still, the test is very slow on ppc64, so I'd > rather keep it in the "slow" category. (virtio-net is already tested in > big endian mode on s390x anyway, so we at least got some endianess test > coverage for that NIC already). TBH, I see this as exercising the firmware netboot features more than the device model. > > > We had no tests at all for the q35 machine, which doesn't seem wise given > > it's increasing prominence. Add a couple of tests for it, including > > testing the newer e1000e adapter. > > That sounds like a good idea! I wonder whether we should also test some > "old" machine versions (in "slow" mode only, of course), like "pc-0.12" > for example, to make sure that there are no regressions with old > machines? Maybe. Not really my department.
diff --git a/tests/pxe-test.c b/tests/pxe-test.c index e7a0610070..16e2520940 100644 --- a/tests/pxe-test.c +++ b/tests/pxe-test.c @@ -31,6 +31,8 @@ typedef struct testdef { static testdef_t x86_tests[] = { { "pc", "e1000" }, { "pc", "virtio", "-device virtio-net-pci,netdev=" NETNAME }, + { "q35", "e1000e" }, + { "q35", "virtio", "-device virtio-net-pci,netdev=" NETNAME }, { NULL }, }; @@ -44,11 +46,11 @@ static testdef_t x86_tests_slow[] = { static testdef_t ppc64_tests[] = { { "pseries", "spapr-vlan" }, + { "pseries", "virtio", "-device virtio-net-pci,netdev=" NETNAME }, { NULL }, }; static testdef_t ppc64_tests_slow[] = { - { "pseries", "virtio", "-device virtio-net-pci,netdev=" NETNAME }, { "pseries", "e1000" }, { NULL }, };
Previously virtio-net was only tested for ppc64 in "slow" mode. That doesn't make much sense since virtio-net is used much more often in practice than the spapr-vlan device which was tested always. So, move virtio-net to always be tested on ppc64. We had no tests at all for the q35 machine, which doesn't seem wise given it's increasing prominence. Add a couple of tests for it, including testing the newer e1000e adapter. Signed-off-by: David Gibson <david@gibson.dropbear.id.au> --- tests/pxe-test.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)