diff mbox series

x86/cpu: Use max host physical address if -cpu max option is applied

Message ID 20210113090430.26394-1-weijiang.yang@intel.com (mailing list archive)
State New, archived
Headers show
Series x86/cpu: Use max host physical address if -cpu max option is applied | expand

Commit Message

Yang Weijiang Jan. 13, 2021, 9:04 a.m. UTC
QEMU option -cpu max(max_features) means "Enables all features supported by
the accelerator in the current host", this looks true for all the features
except guest max physical address width, so add this patch to enable it.

Signed-off-by: Yang Weijiang <weijiang.yang@intel.com>
---
 target/i386/cpu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Paolo Bonzini Jan. 13, 2021, 10:06 a.m. UTC | #1
On 13/01/21 10:04, Yang Weijiang wrote:
> QEMU option -cpu max(max_features) means "Enables all features supported by
> the accelerator in the current host", this looks true for all the features
> except guest max physical address width, so add this patch to enable it.
> 
> Signed-off-by: Yang Weijiang <weijiang.yang@intel.com>
> ---
>   target/i386/cpu.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> index 35459a38bb..b5943406f7 100644
> --- a/target/i386/cpu.c
> +++ b/target/i386/cpu.c
> @@ -6673,7 +6673,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)
>                   warned = true;
>               }
>   
> -            if (cpu->host_phys_bits) {
> +            if (cpu->host_phys_bits || cpu->max_features) {
>                   /* The user asked for us to use the host physical bits */
>                   cpu->phys_bits = host_phys_bits;
>                   if (cpu->host_phys_bits_limit &&
> 

Can you check if this works?

diff --git a/target/i386/cpu.c b/target/i386/cpu.c
index 35459a38bb..72a79e6019 100644
--- a/target/i386/cpu.c
+++ b/target/i386/cpu.c
@@ -4319,6 +4319,7 @@ static void max_x86_cpu_initfn(Object *obj)
          if (lmce_supported()) {
              object_property_set_bool(OBJECT(cpu), "lmce", true, 
&error_abort);
          }
+        object_property_set_bool(OBJECT(cpu), "host-phys-bits", true, 
&error_abort);
      } else {
          object_property_set_str(OBJECT(cpu), "vendor", CPUID_VENDOR_AMD,
                                  &error_abort);


It should allow people to use -cpu max,host-phys-bits=false.

Thanks,

Paolo
Yang Weijiang Jan. 13, 2021, 3:04 p.m. UTC | #2
On Wed, Jan 13, 2021 at 11:06:09AM +0100, Paolo Bonzini wrote:
> On 13/01/21 10:04, Yang Weijiang wrote:
> > QEMU option -cpu max(max_features) means "Enables all features supported by
> > the accelerator in the current host", this looks true for all the features
> > except guest max physical address width, so add this patch to enable it.
> > 
> > Signed-off-by: Yang Weijiang <weijiang.yang@intel.com>
> > ---
> >   target/i386/cpu.c | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> > index 35459a38bb..b5943406f7 100644
> > --- a/target/i386/cpu.c
> > +++ b/target/i386/cpu.c
> > @@ -6673,7 +6673,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)
> >                   warned = true;
> >               }
> > -            if (cpu->host_phys_bits) {
> > +            if (cpu->host_phys_bits || cpu->max_features) {
> >                   /* The user asked for us to use the host physical bits */
> >                   cpu->phys_bits = host_phys_bits;
> >                   if (cpu->host_phys_bits_limit &&
> > 
> 
> Can you check if this works?
> 
Hi, Paolo,
Yes, below change works for kvm-unit-test/access. Would you add the
patch? 

Thanks!

> diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> index 35459a38bb..72a79e6019 100644
> --- a/target/i386/cpu.c
> +++ b/target/i386/cpu.c
> @@ -4319,6 +4319,7 @@ static void max_x86_cpu_initfn(Object *obj)
>          if (lmce_supported()) {
>              object_property_set_bool(OBJECT(cpu), "lmce", true,
> &error_abort);
>          }
> +        object_property_set_bool(OBJECT(cpu), "host-phys-bits", true,
> &error_abort);
>      } else {
>          object_property_set_str(OBJECT(cpu), "vendor", CPUID_VENDOR_AMD,
>                                  &error_abort);
> 
> 
> It should allow people to use -cpu max,host-phys-bits=false.
> 
> Thanks,
> 
> Paolo
Paolo Bonzini Jan. 13, 2021, 3:41 p.m. UTC | #3
Yes, thanks.

Paolo


Il mer 13 gen 2021, 15:52 Yang Weijiang <weijiang.yang@intel.com> ha
scritto:

> On Wed, Jan 13, 2021 at 11:06:09AM +0100, Paolo Bonzini wrote:
> > On 13/01/21 10:04, Yang Weijiang wrote:
> > > QEMU option -cpu max(max_features) means "Enables all features
> supported by
> > > the accelerator in the current host", this looks true for all the
> features
> > > except guest max physical address width, so add this patch to enable
> it.
> > >
> > > Signed-off-by: Yang Weijiang <weijiang.yang@intel.com>
> > > ---
> > >   target/i386/cpu.c | 2 +-
> > >   1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> > > index 35459a38bb..b5943406f7 100644
> > > --- a/target/i386/cpu.c
> > > +++ b/target/i386/cpu.c
> > > @@ -6673,7 +6673,7 @@ static void x86_cpu_realizefn(DeviceState *dev,
> Error **errp)
> > >                   warned = true;
> > >               }
> > > -            if (cpu->host_phys_bits) {
> > > +            if (cpu->host_phys_bits || cpu->max_features) {
> > >                   /* The user asked for us to use the host physical
> bits */
> > >                   cpu->phys_bits = host_phys_bits;
> > >                   if (cpu->host_phys_bits_limit &&
> > >
> >
> > Can you check if this works?
> >
> Hi, Paolo,
> Yes, below change works for kvm-unit-test/access. Would you add the
> patch?
>
> Thanks!
>
> > diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> > index 35459a38bb..72a79e6019 100644
> > --- a/target/i386/cpu.c
> > +++ b/target/i386/cpu.c
> > @@ -4319,6 +4319,7 @@ static void max_x86_cpu_initfn(Object *obj)
> >          if (lmce_supported()) {
> >              object_property_set_bool(OBJECT(cpu), "lmce", true,
> > &error_abort);
> >          }
> > +        object_property_set_bool(OBJECT(cpu), "host-phys-bits", true,
> > &error_abort);
> >      } else {
> >          object_property_set_str(OBJECT(cpu), "vendor", CPUID_VENDOR_AMD,
> >                                  &error_abort);
> >
> >
> > It should allow people to use -cpu max,host-phys-bits=false.
> >
> > Thanks,
> >
> > Paolo
>
>
Nathan Chancellor Jan. 24, 2021, 9:08 p.m. UTC | #4
On Wed, Jan 13, 2021 at 05:04:30PM +0800, Yang Weijiang wrote:
> QEMU option -cpu max(max_features) means "Enables all features supported by
> the accelerator in the current host", this looks true for all the features
> except guest max physical address width, so add this patch to enable it.
> 
> Signed-off-by: Yang Weijiang <weijiang.yang@intel.com>
> ---
>  target/i386/cpu.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> index 35459a38bb..b5943406f7 100644
> --- a/target/i386/cpu.c
> +++ b/target/i386/cpu.c
> @@ -6673,7 +6673,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)
>                  warned = true;
>              }
>  
> -            if (cpu->host_phys_bits) {
> +            if (cpu->host_phys_bits || cpu->max_features) {
>                  /* The user asked for us to use the host physical bits */
>                  cpu->phys_bits = host_phys_bits;
>                  if (cpu->host_phys_bits_limit &&
> -- 
> 2.17.2
> 
> 

Hi,

This patch as commit 5a140b255d ("x86/cpu: Use max host physical address
if -cpu max option is applied") prevents me from using '-cpu host' while
booting an i386_defconfig kernel.

$ qemu-system-i386 \
-append console=ttyS0 \
-cpu host \
-display none \
-enable-kvm \
-initrd rootfs.cpio \
-kernel bzImage \
-serial mon:stdio
qemu-system-i386: phys-bits should be between 32 and 36  (but is 48)

Am I expected to pass "-cpu host,host-phys-bits=false" now or did this
do something unexpected?

Cheers,
Nathan
Yang Weijiang Jan. 25, 2021, 5:41 a.m. UTC | #5
On Sun, Jan 24, 2021 at 02:08:29PM -0700, Nathan Chancellor wrote:
> On Wed, Jan 13, 2021 at 05:04:30PM +0800, Yang Weijiang wrote:
> > QEMU option -cpu max(max_features) means "Enables all features supported by
> > the accelerator in the current host", this looks true for all the features
> > except guest max physical address width, so add this patch to enable it.
> > 
> > Signed-off-by: Yang Weijiang <weijiang.yang@intel.com>
> > ---
> >  target/i386/cpu.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> > index 35459a38bb..b5943406f7 100644
> > --- a/target/i386/cpu.c
> > +++ b/target/i386/cpu.c
> > @@ -6673,7 +6673,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)
> >                  warned = true;
> >              }
> >  
> > -            if (cpu->host_phys_bits) {
> > +            if (cpu->host_phys_bits || cpu->max_features) {
> >                  /* The user asked for us to use the host physical bits */
> >                  cpu->phys_bits = host_phys_bits;
> >                  if (cpu->host_phys_bits_limit &&
> > -- 
> > 2.17.2
> > 
> > 
> 
> Hi,
> 
> This patch as commit 5a140b255d ("x86/cpu: Use max host physical address
> if -cpu max option is applied") prevents me from using '-cpu host' while
> booting an i386_defconfig kernel.
> 
> $ qemu-system-i386 \
> -append console=ttyS0 \
> -cpu host \
> -display none \
> -enable-kvm \
> -initrd rootfs.cpio \
> -kernel bzImage \
> -serial mon:stdio
> qemu-system-i386: phys-bits should be between 32 and 36  (but is 48)
> 
> Am I expected to pass "-cpu host,host-phys-bits=false" now or did this
> do something unexpected?
>
Hi, Nathan,
Could you try Paolo's latest patch?

[PULL 03/31] x86/cpu: Use max host physical address if -cpu max option is applied

diff --git a/target/i386/cpu.c b/target/i386/cpu.c index 35459a38bb..72a79e6019 100644
--- a/target/i386/cpu.c
+++ b/target/i386/cpu.c
@@ -4319,6 +4319,7 @@ static void max_x86_cpu_initfn(Object *obj)
         if (lmce_supported()) {
             object_property_set_bool(OBJECT(cpu), "lmce", true, &error_abort);
         }
+        object_property_set_bool(OBJECT(cpu), "host-phys-bits", true,
+ &error_abort);
     } else {
         object_property_set_str(OBJECT(cpu), "vendor", CPUID_VENDOR_AMD,
                                 &error_abort);
> Cheers,
> Nathan
Nathan Chancellor Jan. 25, 2021, 7:10 a.m. UTC | #6
On Mon, Jan 25, 2021 at 01:41:50PM +0800, Yang Weijiang wrote:
> On Sun, Jan 24, 2021 at 02:08:29PM -0700, Nathan Chancellor wrote:
> > On Wed, Jan 13, 2021 at 05:04:30PM +0800, Yang Weijiang wrote:
> > > QEMU option -cpu max(max_features) means "Enables all features supported by
> > > the accelerator in the current host", this looks true for all the features
> > > except guest max physical address width, so add this patch to enable it.
> > > 
> > > Signed-off-by: Yang Weijiang <weijiang.yang@intel.com>
> > > ---
> > >  target/i386/cpu.c | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> > > index 35459a38bb..b5943406f7 100644
> > > --- a/target/i386/cpu.c
> > > +++ b/target/i386/cpu.c
> > > @@ -6673,7 +6673,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)
> > >                  warned = true;
> > >              }
> > >  
> > > -            if (cpu->host_phys_bits) {
> > > +            if (cpu->host_phys_bits || cpu->max_features) {
> > >                  /* The user asked for us to use the host physical bits */
> > >                  cpu->phys_bits = host_phys_bits;
> > >                  if (cpu->host_phys_bits_limit &&
> > > -- 
> > > 2.17.2
> > > 
> > > 
> > 
> > Hi,
> > 
> > This patch as commit 5a140b255d ("x86/cpu: Use max host physical address
> > if -cpu max option is applied") prevents me from using '-cpu host' while
> > booting an i386_defconfig kernel.
> > 
> > $ qemu-system-i386 \
> > -append console=ttyS0 \
> > -cpu host \
> > -display none \
> > -enable-kvm \
> > -initrd rootfs.cpio \
> > -kernel bzImage \
> > -serial mon:stdio
> > qemu-system-i386: phys-bits should be between 32 and 36  (but is 48)
> > 
> > Am I expected to pass "-cpu host,host-phys-bits=false" now or did this
> > do something unexpected?
> >
> Hi, Nathan,
> Could you try Paolo's latest patch?
> 
> [PULL 03/31] x86/cpu: Use max host physical address if -cpu max option is applied

Hi Yang,

That is the version of the patch I tried, which has been pulled into the
master branch.

Cheers,
Nathan
Paolo Bonzini Jan. 25, 2021, 10:42 a.m. UTC | #7
On 25/01/21 08:10, Nathan Chancellor wrote:
>>> This patch as commit 5a140b255d ("x86/cpu: Use max host physical address
>>> if -cpu max option is applied") prevents me from using '-cpu host' while
>>> booting an i386_defconfig kernel.
>>>
>>> $ qemu-system-i386 \
>>> -append console=ttyS0 \
>>> -cpu host \
>>> -display none \
>>> -enable-kvm \
>>> -initrd rootfs.cpio \
>>> -kernel bzImage \
>>> -serial mon:stdio
>>> qemu-system-i386: phys-bits should be between 32 and 36  (but is 48)
>>>
>>> Am I expected to pass "-cpu host,host-phys-bits=false" now or did this
>>> do something unexpected?

Yes, it's setting the LM bit for a 32-bit guest.

Does this work for you?

diff --git a/target/i386/cpu.c b/target/i386/cpu.c
index 72a79e6019..70df57337f 100644
--- a/target/i386/cpu.c
+++ b/target/i386/cpu.c
@@ -5081,6 +5081,11 @@ static uint64_t 
x86_cpu_get_supported_feature_word(FeatureWord w,
      } else {
          return ~0;
      }
+#ifdef TARGET_I386
+    if (wi->cpuid.eax = 0x80000001) {
+        r &= ~CPUID_EXT2_LM;
+    }
+#endif
      if (migratable_only) {
          r &= x86_cpu_get_migratable_flags(w);
      }


Paolo

>> Hi, Nathan,
>> Could you try Paolo's latest patch?
>>
>> [PULL 03/31] x86/cpu: Use max host physical address if -cpu max option is applied
> 
> Hi Yang,
> 
> That is the version of the patch I tried, which has been pulled into the
> master branch.
> 
> Cheers,
> Nathan
>
Nathan Chancellor Jan. 25, 2021, 6:19 p.m. UTC | #8
On Mon, Jan 25, 2021 at 11:42:43AM +0100, Paolo Bonzini wrote:
> On 25/01/21 08:10, Nathan Chancellor wrote:
> > > > This patch as commit 5a140b255d ("x86/cpu: Use max host physical address
> > > > if -cpu max option is applied") prevents me from using '-cpu host' while
> > > > booting an i386_defconfig kernel.
> > > > 
> > > > $ qemu-system-i386 \
> > > > -append console=ttyS0 \
> > > > -cpu host \
> > > > -display none \
> > > > -enable-kvm \
> > > > -initrd rootfs.cpio \
> > > > -kernel bzImage \
> > > > -serial mon:stdio
> > > > qemu-system-i386: phys-bits should be between 32 and 36  (but is 48)
> > > > 
> > > > Am I expected to pass "-cpu host,host-phys-bits=false" now or did this
> > > > do something unexpected?
> 
> Yes, it's setting the LM bit for a 32-bit guest.
> 
> Does this work for you?

Yes, it does!

Tested-by: Nathan Chancellor <natechancellor@gmail.com>

> diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> index 72a79e6019..70df57337f 100644
> --- a/target/i386/cpu.c
> +++ b/target/i386/cpu.c
> @@ -5081,6 +5081,11 @@ static uint64_t
> x86_cpu_get_supported_feature_word(FeatureWord w,
>      } else {
>          return ~0;
>      }
> +#ifdef TARGET_I386
> +    if (wi->cpuid.eax = 0x80000001) {

This should be a '==':

../target/i386/cpu.c: In function ‘x86_cpu_get_supported_feature_word’:
../target/i386/cpu.c:5085:9: error: suggest parentheses around
assignment used as truth value [-Werror=parentheses]
 5085 |     if (wi->cpuid.eax = 0x80000001) {
      |         ^~
cc1: all warnings being treated as errors

> +        r &= ~CPUID_EXT2_LM;
> +    }
> +#endif
>      if (migratable_only) {
>          r &= x86_cpu_get_migratable_flags(w);
>      }
> 
> 
> Paolo
> 
> > > Hi, Nathan,
> > > Could you try Paolo's latest patch?
> > > 
> > > [PULL 03/31] x86/cpu: Use max host physical address if -cpu max option is applied
> > 
> > Hi Yang,
> > 
> > That is the version of the patch I tried, which has been pulled into the
> > master branch.
> > 
> > Cheers,
> > Nathan
> > 
V>
diff mbox series

Patch

diff --git a/target/i386/cpu.c b/target/i386/cpu.c
index 35459a38bb..b5943406f7 100644
--- a/target/i386/cpu.c
+++ b/target/i386/cpu.c
@@ -6673,7 +6673,7 @@  static void x86_cpu_realizefn(DeviceState *dev, Error **errp)
                 warned = true;
             }
 
-            if (cpu->host_phys_bits) {
+            if (cpu->host_phys_bits || cpu->max_features) {
                 /* The user asked for us to use the host physical bits */
                 cpu->phys_bits = host_phys_bits;
                 if (cpu->host_phys_bits_limit &&