| Message ID | 20191011145349.123425-1-xiaoyao.li@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v2] target/i386: Add Snowridge-v2 (noMPX) CPU model | expand |
On Fri, Oct 11, 2019 at 10:53:49PM +0800, Xiaoyao Li wrote: > Add new version of Snowridge CPU model that removes MPX feature. > > MPX support is being phased out by Intel. GCC has dropped it, Linux kernel > and kvm are also going to do that in the future. > > Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com> > --- > Changes in v2: > - Use CPU model versioning mechanism instead of machine-type compat > --- > target/i386/cpu.c | 13 +++++++++++++ > 1 file changed, 13 insertions(+) > > diff --git a/target/i386/cpu.c b/target/i386/cpu.c > index 44f1bbdcac76..27b0a17b46a8 100644 > --- a/target/i386/cpu.c > +++ b/target/i386/cpu.c > @@ -2793,6 +2793,19 @@ static X86CPUDefinition builtin_x86_defs[] = { > CPUID_6_EAX_ARAT, > .xlevel = 0x80000008, > .model_id = "Intel Atom Processor (SnowRidge)", > + .versions = (X86CPUVersionDefinition[]) { > + { .version = 1 }, > + { > + .version = 2, > + .alias = "Snowridge-noMPX", The intention is to stop creating new funny names for CPU model variations, now, and stick to -v1, -v2, -v3, etc. The .alias field is optional, and was added only for compatibility with the existing -noTSX and -IBRS CPU models. > + .props = (PropValue[]) { > + { "mpx", "off" }, > + { "model-id", "Intel Atom Processor (Snowridge, no MPX)" }, Do you think it's important to report a different model-id? I would keep it the same and only add mpx=off. > + { /* end of list */ }, > + }, > + }, > + { /* end of list */ }, > + }, > }, > { > .name = "KnightsMill", > -- > 2.19.1 >
On 10/12/2019 2:21 AM, Eduardo Habkost wrote: > On Fri, Oct 11, 2019 at 10:53:49PM +0800, Xiaoyao Li wrote: >> Add new version of Snowridge CPU model that removes MPX feature. >> >> MPX support is being phased out by Intel. GCC has dropped it, Linux kernel >> and kvm are also going to do that in the future. >> >> Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com> >> --- >> Changes in v2: >> - Use CPU model versioning mechanism instead of machine-type compat >> --- >> target/i386/cpu.c | 13 +++++++++++++ >> 1 file changed, 13 insertions(+) >> >> diff --git a/target/i386/cpu.c b/target/i386/cpu.c >> index 44f1bbdcac76..27b0a17b46a8 100644 >> --- a/target/i386/cpu.c >> +++ b/target/i386/cpu.c >> @@ -2793,6 +2793,19 @@ static X86CPUDefinition builtin_x86_defs[] = { >> CPUID_6_EAX_ARAT, >> .xlevel = 0x80000008, >> .model_id = "Intel Atom Processor (SnowRidge)", >> + .versions = (X86CPUVersionDefinition[]) { >> + { .version = 1 }, >> + { >> + .version = 2, >> + .alias = "Snowridge-noMPX", > > The intention is to stop creating new funny names for CPU model > variations, now, and stick to -v1, -v2, -v3, etc. > > The .alias field is optional, and was added only for > compatibility with the existing -noTSX and -IBRS CPU models. Got it. >> + .props = (PropValue[]) { >> + { "mpx", "off" }, >> + { "model-id", "Intel Atom Processor (Snowridge, no MPX)" }, > > Do you think it's important to report a different model-id? > I would keep it the same and only add mpx=off. I just want to let user know easily the differences between Snowridge-v1 and Snowridge-v2. Unfortunately, it seems ugly. When testing with Cascadelake-Server, it puzzles every time that which one should I choose between Cascadelake-Server-v1 and Cascadelake-Server-v2. From the output of "-cpu ?", I don't know the differences between them. Everytime I have to go to the source code to see the difference. Maybe there is a way to see/report the differences between different versions of the same CPU model that I just don't know? >> + { /* end of list */ }, >> + }, >> + }, >> + { /* end of list */ }, >> + }, >> }, >> { >> .name = "KnightsMill", >> -- >> 2.19.1 >> >
On Sat, Oct 12, 2019 at 09:15:56AM +0800, Xiaoyao Li wrote: > On 10/12/2019 2:21 AM, Eduardo Habkost wrote: > > On Fri, Oct 11, 2019 at 10:53:49PM +0800, Xiaoyao Li wrote: > > > Add new version of Snowridge CPU model that removes MPX feature. > > > > > > MPX support is being phased out by Intel. GCC has dropped it, Linux kernel > > > and kvm are also going to do that in the future. > > > > > > Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com> > > > --- > > > Changes in v2: > > > - Use CPU model versioning mechanism instead of machine-type compat > > > --- > > > target/i386/cpu.c | 13 +++++++++++++ > > > 1 file changed, 13 insertions(+) > > > > > > diff --git a/target/i386/cpu.c b/target/i386/cpu.c > > > index 44f1bbdcac76..27b0a17b46a8 100644 > > > --- a/target/i386/cpu.c > > > +++ b/target/i386/cpu.c > > > @@ -2793,6 +2793,19 @@ static X86CPUDefinition builtin_x86_defs[] = { > > > CPUID_6_EAX_ARAT, > > > .xlevel = 0x80000008, > > > .model_id = "Intel Atom Processor (SnowRidge)", > > > + .versions = (X86CPUVersionDefinition[]) { > > > + { .version = 1 }, > > > + { > > > + .version = 2, > > > + .alias = "Snowridge-noMPX", > > > > The intention is to stop creating new funny names for CPU model > > variations, now, and stick to -v1, -v2, -v3, etc. > > > > The .alias field is optional, and was added only for > > compatibility with the existing -noTSX and -IBRS CPU models. > > Got it. > > > > + .props = (PropValue[]) { > > > + { "mpx", "off" }, > > > + { "model-id", "Intel Atom Processor (Snowridge, no MPX)" }, > > > > Do you think it's important to report a different model-id? > > I would keep it the same and only add mpx=off. > > I just want to let user know easily the differences between Snowridge-v1 and > Snowridge-v2. Unfortunately, it seems ugly. > > When testing with Cascadelake-Server, it puzzles every time that which one > should I choose between Cascadelake-Server-v1 and Cascadelake-Server-v2. > From the output of "-cpu ?", I don't know the differences between them. > Everytime I have to go to the source code to see the difference. > > Maybe there is a way to see/report the differences between different > versions of the same CPU model that I just don't know? Good point. I forgot that model-id is also the model description in "-cpu ?". Well, it doesn't hurt to have a different model-id in v2 that's more informative. Feel free to keep the model-id line in v3. Thanks!
On 10/12/2019 9:21 AM, Eduardo Habkost wrote: > On Sat, Oct 12, 2019 at 09:15:56AM +0800, Xiaoyao Li wrote: >> On 10/12/2019 2:21 AM, Eduardo Habkost wrote: >>> On Fri, Oct 11, 2019 at 10:53:49PM +0800, Xiaoyao Li wrote: >>>> Add new version of Snowridge CPU model that removes MPX feature. >>>> >>>> MPX support is being phased out by Intel. GCC has dropped it, Linux kernel >>>> and kvm are also going to do that in the future. >>>> >>>> Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com> >>>> --- >>>> Changes in v2: >>>> - Use CPU model versioning mechanism instead of machine-type compat >>>> --- >>>> target/i386/cpu.c | 13 +++++++++++++ >>>> 1 file changed, 13 insertions(+) >>>> >>>> diff --git a/target/i386/cpu.c b/target/i386/cpu.c >>>> index 44f1bbdcac76..27b0a17b46a8 100644 >>>> --- a/target/i386/cpu.c >>>> +++ b/target/i386/cpu.c >>>> @@ -2793,6 +2793,19 @@ static X86CPUDefinition builtin_x86_defs[] = { >>>> CPUID_6_EAX_ARAT, >>>> .xlevel = 0x80000008, >>>> .model_id = "Intel Atom Processor (SnowRidge)", >>>> + .versions = (X86CPUVersionDefinition[]) { >>>> + { .version = 1 }, >>>> + { >>>> + .version = 2, >>>> + .alias = "Snowridge-noMPX", >>> >>> The intention is to stop creating new funny names for CPU model >>> variations, now, and stick to -v1, -v2, -v3, etc. >>> >>> The .alias field is optional, and was added only for >>> compatibility with the existing -noTSX and -IBRS CPU models. >> >> Got it. >> >>>> + .props = (PropValue[]) { >>>> + { "mpx", "off" }, >>>> + { "model-id", "Intel Atom Processor (Snowridge, no MPX)" }, >>> >>> Do you think it's important to report a different model-id? >>> I would keep it the same and only add mpx=off. >> >> I just want to let user know easily the differences between Snowridge-v1 and >> Snowridge-v2. Unfortunately, it seems ugly. >> >> When testing with Cascadelake-Server, it puzzles every time that which one >> should I choose between Cascadelake-Server-v1 and Cascadelake-Server-v2. >> From the output of "-cpu ?", I don't know the differences between them. >> Everytime I have to go to the source code to see the difference. >> >> Maybe there is a way to see/report the differences between different >> versions of the same CPU model that I just don't know? > > Good point. I forgot that model-id is also the model description > in "-cpu ?". > > Well, it doesn't hurt to have a different model-id in v2 that's > more informative. Feel free to keep the model-id line in v3. OK. I will send out the v3 patch keeping the model-id while removing the alias. BTW, do you have better idea to tell the differences among versions of the same CPU model?
On Sat, Oct 12, 2019 at 09:31:25AM +0800, Xiaoyao Li wrote: > On 10/12/2019 9:21 AM, Eduardo Habkost wrote: > > On Sat, Oct 12, 2019 at 09:15:56AM +0800, Xiaoyao Li wrote: > > > On 10/12/2019 2:21 AM, Eduardo Habkost wrote: > > > > On Fri, Oct 11, 2019 at 10:53:49PM +0800, Xiaoyao Li wrote: > > > > > Add new version of Snowridge CPU model that removes MPX feature. > > > > > > > > > > MPX support is being phased out by Intel. GCC has dropped it, Linux kernel > > > > > and kvm are also going to do that in the future. > > > > > > > > > > Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com> > > > > > --- > > > > > Changes in v2: > > > > > - Use CPU model versioning mechanism instead of machine-type compat > > > > > --- > > > > > target/i386/cpu.c | 13 +++++++++++++ > > > > > 1 file changed, 13 insertions(+) > > > > > > > > > > diff --git a/target/i386/cpu.c b/target/i386/cpu.c > > > > > index 44f1bbdcac76..27b0a17b46a8 100644 > > > > > --- a/target/i386/cpu.c > > > > > +++ b/target/i386/cpu.c > > > > > @@ -2793,6 +2793,19 @@ static X86CPUDefinition builtin_x86_defs[] = { > > > > > CPUID_6_EAX_ARAT, > > > > > .xlevel = 0x80000008, > > > > > .model_id = "Intel Atom Processor (SnowRidge)", > > > > > + .versions = (X86CPUVersionDefinition[]) { > > > > > + { .version = 1 }, > > > > > + { > > > > > + .version = 2, > > > > > + .alias = "Snowridge-noMPX", > > > > > > > > The intention is to stop creating new funny names for CPU model > > > > variations, now, and stick to -v1, -v2, -v3, etc. > > > > > > > > The .alias field is optional, and was added only for > > > > compatibility with the existing -noTSX and -IBRS CPU models. > > > > > > Got it. > > > > > > > > + .props = (PropValue[]) { > > > > > + { "mpx", "off" }, > > > > > + { "model-id", "Intel Atom Processor (Snowridge, no MPX)" }, > > > > > > > > Do you think it's important to report a different model-id? > > > > I would keep it the same and only add mpx=off. > > > > > > I just want to let user know easily the differences between Snowridge-v1 and > > > Snowridge-v2. Unfortunately, it seems ugly. > > > > > > When testing with Cascadelake-Server, it puzzles every time that which one > > > should I choose between Cascadelake-Server-v1 and Cascadelake-Server-v2. > > > From the output of "-cpu ?", I don't know the differences between them. > > > Everytime I have to go to the source code to see the difference. > > > > > > Maybe there is a way to see/report the differences between different > > > versions of the same CPU model that I just don't know? > > > > Good point. I forgot that model-id is also the model description > > in "-cpu ?". > > > > Well, it doesn't hurt to have a different model-id in v2 that's > > more informative. Feel free to keep the model-id line in v3. > > OK. I will send out the v3 patch keeping the model-id while removing the > alias. > > BTW, do you have better idea to tell the differences among versions of the > same CPU model? We could add a new field for a human-readable description of the CPU model version, and print that field on "-cpu help" if set. We could also try to generate a description automatically (e.g. automatically describe SnowRidge-v2 as "SnowRidge-v1 plus mpx" based on .props). I'm not sure if it would be worth the effort, though.
diff --git a/target/i386/cpu.c b/target/i386/cpu.c index 44f1bbdcac76..27b0a17b46a8 100644 --- a/target/i386/cpu.c +++ b/target/i386/cpu.c @@ -2793,6 +2793,19 @@ static X86CPUDefinition builtin_x86_defs[] = { CPUID_6_EAX_ARAT, .xlevel = 0x80000008, .model_id = "Intel Atom Processor (SnowRidge)", + .versions = (X86CPUVersionDefinition[]) { + { .version = 1 }, + { + .version = 2, + .alias = "Snowridge-noMPX", + .props = (PropValue[]) { + { "mpx", "off" }, + { "model-id", "Intel Atom Processor (Snowridge, no MPX)" }, + { /* end of list */ }, + }, + }, + { /* end of list */ }, + }, }, { .name = "KnightsMill",
Add new version of Snowridge CPU model that removes MPX feature. MPX support is being phased out by Intel. GCC has dropped it, Linux kernel and kvm are also going to do that in the future. Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com> --- Changes in v2: - Use CPU model versioning mechanism instead of machine-type compat --- target/i386/cpu.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)