| Message ID | 20221101071048.29553-1-yangyicong@huawei.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Only generate cluster node in PPTT when specified | expand |
On Tue, Nov 01, 2022 at 03:10:42PM +0800, Yicong Yang wrote: > From: Yicong Yang <yangyicong@hisilicon.com> > > This series mainly change the policy for building a cluster topology node > in PPTT. Previously we'll always build a cluster node in PPTT without > asking the user, after this set the cluster node will be built only the > the user specify through "-smp clusters=X". > > One problem is related to this but not fully caused by this, see the > discussion in [*]. When booting the VM with `-smp 8` and 4 numa nodes, > the linux scheduling domains in the VM misses the NUMA domains. It's > because the MC level span extends to Cluster level (which is generated > by the Qemu by default) that spans all the cpus in the system, then the > scheduling domain building stops at MC level since it already includes all > the cpus. > > Considering cluster is an optional level and most platforms don't have it, > they may even don't realize this is built and a always build policy cannot > emulate the real topology on these platforms. So in this series improve the > policy to only generate cluster when the user explicitly want it. > > Update the tests and test tables accordingly. I think we can classify this as a bugfix and so allow after the freeze, however, this needs ack from ARM maintainers then. > [*] https://lore.kernel.org/lkml/2c079860-ee82-7719-d3d2-756192f41704@huawei.com/ > > Change since v3: > - Improve and attach the diff of the affected ACPI tables in the commit, and minor cleanups > Link: https://lore.kernel.org/qemu-devel/20221031090523.34146-1-yangyicong@huawei.com/ > > Change since v2: > - Add tag from Micheal, thanks > - Handle the tests changes with bios-tables-test-allowed-diff.h, Per Micheal > - Address the comments per Yanan > Link: https://lore.kernel.org/qemu-devel/20221027032613.18377-1-yangyicong@huawei.com/ > > Change since v1: > - Only includes the test tables which is really needed > - Enrich the commit > Link: https://lore.kernel.org/qemu-devel/20220922131143.58003-1-yangyicong@huawei.com/ > > Yicong Yang (6): > tests: virt: Allow changes to PPTT test table > hw/acpi/aml-build: Only generate cluster node in PPTT when specified > tests: virt: Update expected ACPI tables for virt test > tests: acpi: Add and whitelist *.topology blobs > tests: acpi: aarch64: Add topology test for aarch64 > tests: acpi: aarch64: Add *.topology tables > > hw/acpi/aml-build.c | 2 +- > hw/core/machine-smp.c | 2 ++ > include/hw/boards.h | 3 +++ > qemu-options.hx | 3 +++ > tests/data/acpi/virt/APIC.topology | Bin 0 -> 700 bytes > tests/data/acpi/virt/DSDT.topology | Bin 0 -> 5398 bytes > tests/data/acpi/virt/PPTT | Bin 96 -> 76 bytes > tests/data/acpi/virt/PPTT.topology | Bin 0 -> 336 bytes > tests/qtest/bios-tables-test.c | 19 +++++++++++++++++++ > 9 files changed, 28 insertions(+), 1 deletion(-) > create mode 100644 tests/data/acpi/virt/APIC.topology > create mode 100644 tests/data/acpi/virt/DSDT.topology > create mode 100644 tests/data/acpi/virt/PPTT.topology > > -- > 2.24.0
On 2022/11/2 16:17, Michael S. Tsirkin wrote: > On Tue, Nov 01, 2022 at 03:10:42PM +0800, Yicong Yang wrote: >> From: Yicong Yang <yangyicong@hisilicon.com> >> >> This series mainly change the policy for building a cluster topology node >> in PPTT. Previously we'll always build a cluster node in PPTT without >> asking the user, after this set the cluster node will be built only the >> the user specify through "-smp clusters=X". >> >> One problem is related to this but not fully caused by this, see the >> discussion in [*]. When booting the VM with `-smp 8` and 4 numa nodes, >> the linux scheduling domains in the VM misses the NUMA domains. It's >> because the MC level span extends to Cluster level (which is generated >> by the Qemu by default) that spans all the cpus in the system, then the >> scheduling domain building stops at MC level since it already includes all >> the cpus. >> >> Considering cluster is an optional level and most platforms don't have it, >> they may even don't realize this is built and a always build policy cannot >> emulate the real topology on these platforms. So in this series improve the >> policy to only generate cluster when the user explicitly want it. >> >> Update the tests and test tables accordingly. > > I think we can classify this as a bugfix and so allow after Not quite sure about it as I regarded it as an improvement of the topology building policy. And the problem I met is not directly caused by the policy before this series. > the freeze, however, this needs ack from ARM maintainers then. sure. Will resend after the freeze. Thanks. > > >> [*] https://lore.kernel.org/lkml/2c079860-ee82-7719-d3d2-756192f41704@huawei.com/ >> >> Change since v3: >> - Improve and attach the diff of the affected ACPI tables in the commit, and minor cleanups >> Link: https://lore.kernel.org/qemu-devel/20221031090523.34146-1-yangyicong@huawei.com/ >> >> Change since v2: >> - Add tag from Micheal, thanks >> - Handle the tests changes with bios-tables-test-allowed-diff.h, Per Micheal >> - Address the comments per Yanan >> Link: https://lore.kernel.org/qemu-devel/20221027032613.18377-1-yangyicong@huawei.com/ >> >> Change since v1: >> - Only includes the test tables which is really needed >> - Enrich the commit >> Link: https://lore.kernel.org/qemu-devel/20220922131143.58003-1-yangyicong@huawei.com/ >> >> Yicong Yang (6): >> tests: virt: Allow changes to PPTT test table >> hw/acpi/aml-build: Only generate cluster node in PPTT when specified >> tests: virt: Update expected ACPI tables for virt test >> tests: acpi: Add and whitelist *.topology blobs >> tests: acpi: aarch64: Add topology test for aarch64 >> tests: acpi: aarch64: Add *.topology tables >> >> hw/acpi/aml-build.c | 2 +- >> hw/core/machine-smp.c | 2 ++ >> include/hw/boards.h | 3 +++ >> qemu-options.hx | 3 +++ >> tests/data/acpi/virt/APIC.topology | Bin 0 -> 700 bytes >> tests/data/acpi/virt/DSDT.topology | Bin 0 -> 5398 bytes >> tests/data/acpi/virt/PPTT | Bin 96 -> 76 bytes >> tests/data/acpi/virt/PPTT.topology | Bin 0 -> 336 bytes >> tests/qtest/bios-tables-test.c | 19 +++++++++++++++++++ >> 9 files changed, 28 insertions(+), 1 deletion(-) >> create mode 100644 tests/data/acpi/virt/APIC.topology >> create mode 100644 tests/data/acpi/virt/DSDT.topology >> create mode 100644 tests/data/acpi/virt/PPTT.topology >> >> -- >> 2.24.0 > > . >
On Wed, 2 Nov 2022 at 08:17, Michael S. Tsirkin <mst@redhat.com> wrote: > > On Tue, Nov 01, 2022 at 03:10:42PM +0800, Yicong Yang wrote: > > From: Yicong Yang <yangyicong@hisilicon.com> > > > > This series mainly change the policy for building a cluster topology node > > in PPTT. Previously we'll always build a cluster node in PPTT without > > asking the user, after this set the cluster node will be built only the > > the user specify through "-smp clusters=X". > > > > One problem is related to this but not fully caused by this, see the > > discussion in [*]. When booting the VM with `-smp 8` and 4 numa nodes, > > the linux scheduling domains in the VM misses the NUMA domains. It's > > because the MC level span extends to Cluster level (which is generated > > by the Qemu by default) that spans all the cpus in the system, then the > > scheduling domain building stops at MC level since it already includes all > > the cpus. > > > > Considering cluster is an optional level and most platforms don't have it, > > they may even don't realize this is built and a always build policy cannot > > emulate the real topology on these platforms. So in this series improve the > > policy to only generate cluster when the user explicitly want it. > > > > Update the tests and test tables accordingly. > > I think we can classify this as a bugfix and so allow after > the freeze, however, this needs ack from ARM maintainers then. I don't use, test or understand any of the ACPI related code :-) I'm happy to leave it up to your judgement whether this should go into this release or wait for 8.0. thanks -- PMM
On Tue, Nov 01, 2022 at 03:10:42PM +0800, Yicong Yang wrote: > From: Yicong Yang <yangyicong@hisilicon.com> > > This series mainly change the policy for building a cluster topology node > in PPTT. Previously we'll always build a cluster node in PPTT without > asking the user, after this set the cluster node will be built only the > the user specify through "-smp clusters=X". > > One problem is related to this but not fully caused by this, see the > discussion in [*]. When booting the VM with `-smp 8` and 4 numa nodes, > the linux scheduling domains in the VM misses the NUMA domains. It's > because the MC level span extends to Cluster level (which is generated > by the Qemu by default) that spans all the cpus in the system, then the > scheduling domain building stops at MC level since it already includes all > the cpus. > > Considering cluster is an optional level and most platforms don't have it, > they may even don't realize this is built and a always build policy cannot > emulate the real topology on these platforms. So in this series improve the > policy to only generate cluster when the user explicitly want it. > > Update the tests and test tables accordingly. To merge this to master we also need to update the new tests/data/acpi/virt/PPTT.acpihmatvirt I could do so myself but I'd rather you did the rebase and verified the diff manually. If the diff is the same no need to update commit log just the binaries. Thanks! > [*] https://lore.kernel.org/lkml/2c079860-ee82-7719-d3d2-756192f41704@huawei.com/ > > Change since v3: > - Improve and attach the diff of the affected ACPI tables in the commit, and minor cleanups > Link: https://lore.kernel.org/qemu-devel/20221031090523.34146-1-yangyicong@huawei.com/ > > Change since v2: > - Add tag from Micheal, thanks > - Handle the tests changes with bios-tables-test-allowed-diff.h, Per Micheal > - Address the comments per Yanan > Link: https://lore.kernel.org/qemu-devel/20221027032613.18377-1-yangyicong@huawei.com/ > > Change since v1: > - Only includes the test tables which is really needed > - Enrich the commit > Link: https://lore.kernel.org/qemu-devel/20220922131143.58003-1-yangyicong@huawei.com/ > > Yicong Yang (6): > tests: virt: Allow changes to PPTT test table > hw/acpi/aml-build: Only generate cluster node in PPTT when specified > tests: virt: Update expected ACPI tables for virt test > tests: acpi: Add and whitelist *.topology blobs > tests: acpi: aarch64: Add topology test for aarch64 > tests: acpi: aarch64: Add *.topology tables > > hw/acpi/aml-build.c | 2 +- > hw/core/machine-smp.c | 2 ++ > include/hw/boards.h | 3 +++ > qemu-options.hx | 3 +++ > tests/data/acpi/virt/APIC.topology | Bin 0 -> 700 bytes > tests/data/acpi/virt/DSDT.topology | Bin 0 -> 5398 bytes > tests/data/acpi/virt/PPTT | Bin 96 -> 76 bytes > tests/data/acpi/virt/PPTT.topology | Bin 0 -> 336 bytes > tests/qtest/bios-tables-test.c | 19 +++++++++++++++++++ > 9 files changed, 28 insertions(+), 1 deletion(-) > create mode 100644 tests/data/acpi/virt/APIC.topology > create mode 100644 tests/data/acpi/virt/DSDT.topology > create mode 100644 tests/data/acpi/virt/PPTT.topology > > -- > 2.24.0
On 2022/12/21 14:27, Michael S. Tsirkin wrote: > On Tue, Nov 01, 2022 at 03:10:42PM +0800, Yicong Yang wrote: >> From: Yicong Yang <yangyicong@hisilicon.com> >> >> This series mainly change the policy for building a cluster topology node >> in PPTT. Previously we'll always build a cluster node in PPTT without >> asking the user, after this set the cluster node will be built only the >> the user specify through "-smp clusters=X". >> >> One problem is related to this but not fully caused by this, see the >> discussion in [*]. When booting the VM with `-smp 8` and 4 numa nodes, >> the linux scheduling domains in the VM misses the NUMA domains. It's >> because the MC level span extends to Cluster level (which is generated >> by the Qemu by default) that spans all the cpus in the system, then the >> scheduling domain building stops at MC level since it already includes all >> the cpus. >> >> Considering cluster is an optional level and most platforms don't have it, >> they may even don't realize this is built and a always build policy cannot >> emulate the real topology on these platforms. So in this series improve the >> policy to only generate cluster when the user explicitly want it. >> >> Update the tests and test tables accordingly. > > To merge this to master we also need to update the new > tests/data/acpi/virt/PPTT.acpihmatvirt > > I could do so myself but I'd rather you did the rebase and verified > the diff manually. If the diff is the same no need to update > commit log just the binaries. > Thanks for the hint and reminder! Will rebase on the lastest and do the necessary updates. Will send an updated version tomorrow or late this week. Thanks, Yicong > > > >> [*] https://lore.kernel.org/lkml/2c079860-ee82-7719-d3d2-756192f41704@huawei.com/ >> >> Change since v3: >> - Improve and attach the diff of the affected ACPI tables in the commit, and minor cleanups >> Link: https://lore.kernel.org/qemu-devel/20221031090523.34146-1-yangyicong@huawei.com/ >> >> Change since v2: >> - Add tag from Micheal, thanks >> - Handle the tests changes with bios-tables-test-allowed-diff.h, Per Micheal >> - Address the comments per Yanan >> Link: https://lore.kernel.org/qemu-devel/20221027032613.18377-1-yangyicong@huawei.com/ >> >> Change since v1: >> - Only includes the test tables which is really needed >> - Enrich the commit >> Link: https://lore.kernel.org/qemu-devel/20220922131143.58003-1-yangyicong@huawei.com/ >> >> Yicong Yang (6): >> tests: virt: Allow changes to PPTT test table >> hw/acpi/aml-build: Only generate cluster node in PPTT when specified >> tests: virt: Update expected ACPI tables for virt test >> tests: acpi: Add and whitelist *.topology blobs >> tests: acpi: aarch64: Add topology test for aarch64 >> tests: acpi: aarch64: Add *.topology tables >> >> hw/acpi/aml-build.c | 2 +- >> hw/core/machine-smp.c | 2 ++ >> include/hw/boards.h | 3 +++ >> qemu-options.hx | 3 +++ >> tests/data/acpi/virt/APIC.topology | Bin 0 -> 700 bytes >> tests/data/acpi/virt/DSDT.topology | Bin 0 -> 5398 bytes >> tests/data/acpi/virt/PPTT | Bin 96 -> 76 bytes >> tests/data/acpi/virt/PPTT.topology | Bin 0 -> 336 bytes >> tests/qtest/bios-tables-test.c | 19 +++++++++++++++++++ >> 9 files changed, 28 insertions(+), 1 deletion(-) >> create mode 100644 tests/data/acpi/virt/APIC.topology >> create mode 100644 tests/data/acpi/virt/DSDT.topology >> create mode 100644 tests/data/acpi/virt/PPTT.topology >> >> -- >> 2.24.0 > > . >
From: Yicong Yang <yangyicong@hisilicon.com> This series mainly change the policy for building a cluster topology node in PPTT. Previously we'll always build a cluster node in PPTT without asking the user, after this set the cluster node will be built only the the user specify through "-smp clusters=X". One problem is related to this but not fully caused by this, see the discussion in [*]. When booting the VM with `-smp 8` and 4 numa nodes, the linux scheduling domains in the VM misses the NUMA domains. It's because the MC level span extends to Cluster level (which is generated by the Qemu by default) that spans all the cpus in the system, then the scheduling domain building stops at MC level since it already includes all the cpus. Considering cluster is an optional level and most platforms don't have it, they may even don't realize this is built and a always build policy cannot emulate the real topology on these platforms. So in this series improve the policy to only generate cluster when the user explicitly want it. Update the tests and test tables accordingly. [*] https://lore.kernel.org/lkml/2c079860-ee82-7719-d3d2-756192f41704@huawei.com/ Change since v3: - Improve and attach the diff of the affected ACPI tables in the commit, and minor cleanups Link: https://lore.kernel.org/qemu-devel/20221031090523.34146-1-yangyicong@huawei.com/ Change since v2: - Add tag from Micheal, thanks - Handle the tests changes with bios-tables-test-allowed-diff.h, Per Micheal - Address the comments per Yanan Link: https://lore.kernel.org/qemu-devel/20221027032613.18377-1-yangyicong@huawei.com/ Change since v1: - Only includes the test tables which is really needed - Enrich the commit Link: https://lore.kernel.org/qemu-devel/20220922131143.58003-1-yangyicong@huawei.com/ Yicong Yang (6): tests: virt: Allow changes to PPTT test table hw/acpi/aml-build: Only generate cluster node in PPTT when specified tests: virt: Update expected ACPI tables for virt test tests: acpi: Add and whitelist *.topology blobs tests: acpi: aarch64: Add topology test for aarch64 tests: acpi: aarch64: Add *.topology tables hw/acpi/aml-build.c | 2 +- hw/core/machine-smp.c | 2 ++ include/hw/boards.h | 3 +++ qemu-options.hx | 3 +++ tests/data/acpi/virt/APIC.topology | Bin 0 -> 700 bytes tests/data/acpi/virt/DSDT.topology | Bin 0 -> 5398 bytes tests/data/acpi/virt/PPTT | Bin 96 -> 76 bytes tests/data/acpi/virt/PPTT.topology | Bin 0 -> 336 bytes tests/qtest/bios-tables-test.c | 19 +++++++++++++++++++ 9 files changed, 28 insertions(+), 1 deletion(-) create mode 100644 tests/data/acpi/virt/APIC.topology create mode 100644 tests/data/acpi/virt/DSDT.topology create mode 100644 tests/data/acpi/virt/PPTT.topology