[libvirt] [PATCH] cpu: Add support for al57 Intel features
We can start qemu with a "cpu,+la57" to set 57-bit vitrual address space. So VM can be aware that it need to enable 5-level paging. Corresponding QEMU commits: al57 6c7c3c21f95dd9af8a0691c0dd29b07247984122 --- src/cpu/cpu_map.xml | 3 +++ 1 file changed, 3 insertions(+) diff --git a/src/cpu/cpu_map.xml b/src/cpu/cpu_map.xml index e5da7a8..922a195 100644 --- a/src/cpu/cpu_map.xml +++ b/src/cpu/cpu_map.xml @@ -285,6 +285,9 @@ <feature name='ospke'> <cpuid eax_in='0x07' ecx_in='0x00' ecx='0x00000010'/> </feature> + <feature name='la57'> + <cpuid eax_in='0x07' ecx_in='0x00' ecx='0x00010000'/> + </feature> <feature name='avx512-4vnniw'> <cpuid eax_in='0x07' ecx_in='0x00' edx='0x00000004'/> -- 2.7.4
On 12/17/2017 06:02 PM, Shaohe Feng wrote:
We can start qemu with a "cpu,+la57" to set 57-bit vitrual address space. So VM can be aware that it need to enable 5-level paging.
Corresponding QEMU commits: al57 6c7c3c21f95dd9af8a0691c0dd29b07247984122 --- src/cpu/cpu_map.xml | 3 +++ 1 file changed, 3 insertions(+)
I think if you go through history of cpu_map.xml changes you'll find when a new feature is added there are tests added as well - this would thus seemingly need a test adjustment as well. John
diff --git a/src/cpu/cpu_map.xml b/src/cpu/cpu_map.xml index e5da7a8..922a195 100644 --- a/src/cpu/cpu_map.xml +++ b/src/cpu/cpu_map.xml @@ -285,6 +285,9 @@ <feature name='ospke'> <cpuid eax_in='0x07' ecx_in='0x00' ecx='0x00000010'/> </feature> + <feature name='la57'> + <cpuid eax_in='0x07' ecx_in='0x00' ecx='0x00010000'/> + </feature>
<feature name='avx512-4vnniw'> <cpuid eax_in='0x07' ecx_in='0x00' edx='0x00000004'/>
On 2018年01月05日 02:52, John Ferlan wrote:
On 12/17/2017 06:02 PM, Shaohe Feng wrote:
We can start qemu with a "cpu,+la57" to set 57-bit vitrual address space. So VM can be aware that it need to enable 5-level paging.
Corresponding QEMU commits: al57 6c7c3c21f95dd9af8a0691c0dd29b07247984122 --- src/cpu/cpu_map.xml | 3 +++ 1 file changed, 3 insertions(+)
I think if you go through history of cpu_map.xml changes you'll find when a new feature is added there are tests added as well - this would thus seemingly need a test adjustment as well.
Thanks John. Will add tests.
John
diff --git a/src/cpu/cpu_map.xml b/src/cpu/cpu_map.xml index e5da7a8..922a195 100644 --- a/src/cpu/cpu_map.xml +++ b/src/cpu/cpu_map.xml @@ -285,6 +285,9 @@ <feature name='ospke'> <cpuid eax_in='0x07' ecx_in='0x00' ecx='0x00000010'/> </feature> + <feature name='la57'> + <cpuid eax_in='0x07' ecx_in='0x00' ecx='0x00010000'/> + </feature>
<feature name='avx512-4vnniw'> <cpuid eax_in='0x07' ecx_in='0x00' edx='0x00000004'/>
On Mon, Jan 08, 2018 at 19:06:43 +0800, Feng, Shaohe wrote:
On 2018年01月05日 02:52, John Ferlan wrote:
On 12/17/2017 06:02 PM, Shaohe Feng wrote:
We can start qemu with a "cpu,+la57" to set 57-bit vitrual address space. So VM can be aware that it need to enable 5-level paging.
Corresponding QEMU commits: al57 6c7c3c21f95dd9af8a0691c0dd29b07247984122 --- src/cpu/cpu_map.xml | 3 +++ 1 file changed, 3 insertions(+)
I think if you go through history of cpu_map.xml changes you'll find when a new feature is added there are tests added as well - this would thus seemingly need a test adjustment as well.
Thanks John. Will add tests.
Adding such test usually means you go to tests/cputestdata/ directory and run "./cpu-gather.sh | ./cpu-parse.sh" which will produce several data files. Then you add a new test case in tests/cputest.c for this new CPU and regenerate the files with expected results. If you do this before adding the new feature, the patch which will then add it will nicely show the new feature gets added into the expected results of the new CPU test. Don't forget to install cpuid tool and make sure you have the latest upstream qemu installed in the system. Jirka
On 2018年01月09日 18:40, Jiri Denemark wrote:
On Mon, Jan 08, 2018 at 19:06:43 +0800, Feng, Shaohe wrote:
On 2018年01月05日 02:52, John Ferlan wrote:
On 12/17/2017 06:02 PM, Shaohe Feng wrote:
We can start qemu with a "cpu,+la57" to set 57-bit vitrual address space. So VM can be aware that it need to enable 5-level paging.
Corresponding QEMU commits: al57 6c7c3c21f95dd9af8a0691c0dd29b07247984122 --- src/cpu/cpu_map.xml | 3 +++ 1 file changed, 3 insertions(+)
I think if you go through history of cpu_map.xml changes you'll find when a new feature is added there are tests added as well - this would thus seemingly need a test adjustment as well. Thanks John. Will add tests. Adding such test usually means you go to tests/cputestdata/ directory and run "./cpu-gather.sh | ./cpu-parse.sh" which will produce several data files. Then you add a new test case in tests/cputest.c for this new CPU and regenerate the files with expected results. If you do this before adding the new feature, the patch which will then add it will nicely show the new feature gets added into the expected results of the new CPU test.
Don't forget to install cpuid tool and make sure you have the latest upstream qemu installed in the system.
Jirka
Hi Jirka, I went through the tests/cputestdata/ directory and found it’s for defining features for different CPU models. However, the la57 feature is not included in any current CPUs. So I think we can get waived for the tests changes? I check la57 on my Haswell platform it show $ qemu-system-x86_64 -enable-kvm -cpu Haswell,+la57,check warning: host doesn't support requested feature: CPUID.07H:ECX.la57 [bit 16] BR Shaohe Feng
On Fri, Jan 19, 2018 at 22:28:39 +0800, Feng, Shaohe wrote:
On 2018年01月09日 18:40, Jiri Denemark wrote:
On Mon, Jan 08, 2018 at 19:06:43 +0800, Feng, Shaohe wrote:
On 2018年01月05日 02:52, John Ferlan wrote:
On 12/17/2017 06:02 PM, Shaohe Feng wrote:
We can start qemu with a "cpu,+la57" to set 57-bit vitrual address space. So VM can be aware that it need to enable 5-level paging.
Corresponding QEMU commits: al57 6c7c3c21f95dd9af8a0691c0dd29b07247984122 --- src/cpu/cpu_map.xml | 3 +++ 1 file changed, 3 insertions(+)
I think if you go through history of cpu_map.xml changes you'll find when a new feature is added there are tests added as well - this would thus seemingly need a test adjustment as well. Thanks John. Will add tests. Adding such test usually means you go to tests/cputestdata/ directory and run "./cpu-gather.sh | ./cpu-parse.sh" which will produce several data files. Then you add a new test case in tests/cputest.c for this new CPU and regenerate the files with expected results. If you do this before adding the new feature, the patch which will then add it will nicely show the new feature gets added into the expected results of the new CPU test.
Don't forget to install cpuid tool and make sure you have the latest upstream qemu installed in the system.
Jirka
Hi Jirka,
I went through the tests/cputestdata/ directory and found it’s for defining features for different CPU models.
However, the la57 feature is not included in any current CPUs. So I think we can get waived for the tests changes?
Hmm, I expected you had access to such host for testing... Anyway, ACK to the patch then with the following diff squashed in (apparently you didn't run make check): diff --git a/tests/domaincapsschemadata/qemu_2.9.0-tcg.x86_64.xml b/tests/domaincapsschemadata/qemu_2.9.0-tcg.x86_64.xml index 7ff998907d..6c2bbac190 100644 --- a/tests/domaincapsschemadata/qemu_2.9.0-tcg.x86_64.xml +++ b/tests/domaincapsschemadata/qemu_2.9.0-tcg.x86_64.xml @@ -43,6 +43,7 @@ <feature policy='require' name='clwb'/> <feature policy='require' name='pku'/> <feature policy='require' name='ospke'/> + <feature policy='require' name='la57'/> <feature policy='require' name='xsaveopt'/> <feature policy='require' name='xgetbv1'/> <feature policy='require' name='mmxext'/>
On 2018年01月25日 22:35, Jiri Denemark wrote:
On Fri, Jan 19, 2018 at 22:28:39 +0800, Feng, Shaohe wrote:
On 2018年01月09日 18:40, Jiri Denemark wrote:
On Mon, Jan 08, 2018 at 19:06:43 +0800, Feng, Shaohe wrote:
On 2018年01月05日 02:52, John Ferlan wrote:
On 12/17/2017 06:02 PM, Shaohe Feng wrote:
We can start qemu with a "cpu,+la57" to set 57-bit vitrual address space. So VM can be aware that it need to enable 5-level paging.
Corresponding QEMU commits: al57 6c7c3c21f95dd9af8a0691c0dd29b07247984122 --- src/cpu/cpu_map.xml | 3 +++ 1 file changed, 3 insertions(+)
I think if you go through history of cpu_map.xml changes you'll find when a new feature is added there are tests added as well - this would thus seemingly need a test adjustment as well. Thanks John. Will add tests. Adding such test usually means you go to tests/cputestdata/ directory and run "./cpu-gather.sh | ./cpu-parse.sh" which will produce several data files. Then you add a new test case in tests/cputest.c for this new CPU and regenerate the files with expected results. If you do this before adding the new feature, the patch which will then add it will nicely show the new feature gets added into the expected results of the new CPU test.
Don't forget to install cpuid tool and make sure you have the latest upstream qemu installed in the system.
Jirka Hi Jirka,
I went through the tests/cputestdata/ directory and found it’s for defining features for different CPU models.
However, the la57 feature is not included in any current CPUs. So I think we can get waived for the tests changes? Hmm, I expected you had access to such host for testing... I have checked with the KVM/qemu developer, the also have no such host. They do their development on an simulator.
But I did set up the libvirt env, start a VM with la57 feature, and check la57 argument in the qemu command line.
Anyway, ACK to the patch then with the following diff squashed in (apparently you didn't run make check):
diff --git a/tests/domaincapsschemadata/qemu_2.9.0-tcg.x86_64.xml b/tests/domaincapsschemadata/qemu_2.9.0-tcg.x86_64.xml index 7ff998907d..6c2bbac190 100644 --- a/tests/domaincapsschemadata/qemu_2.9.0-tcg.x86_64.xml +++ b/tests/domaincapsschemadata/qemu_2.9.0-tcg.x86_64.xml @@ -43,6 +43,7 @@ <feature policy='require' name='clwb'/> <feature policy='require' name='pku'/> <feature policy='require' name='ospke'/> + <feature policy='require' name='la57'/> <feature policy='require' name='xsaveopt'/> <feature policy='require' name='xgetbv1'/> <feature policy='require' name='mmxext'/>
participants (4)
-
Feng, Shaohe -
Jiri Denemark -
John Ferlan -
Shaohe Feng