| Message ID | 1386001205-11978-2-git-send-email-lorenzo.pieralisi@arm.com (mailing list archive) |
|---|---|
| State | Not Applicable, archived |
| Headers | show |
On Dec 2, 2013, at 10:20 AM, Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> wrote: > On ARM systems the cache topology cannot be probed at runtime, in > particular, it is impossible to probe which CPUs share a given cache > level. Power management software requires this knowledge to implement > optimized power down sequences, hence this patch adds a document that > defines the DT cache bindings for ARM systems. The bindings are compliant > with ePAPR (PowerPC bindings), and rely on the cache bindings already > standardized in the ePAPR v1.1 document; ARM required updates are underlined > in the binding document. > > Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> > --- > Documentation/devicetree/bindings/arm/cache.txt | 25 +++++++++++++++++++++++++ > 1 file changed, 25 insertions(+) > create mode 100644 Documentation/devicetree/bindings/arm/cache.txt > > diff --git a/Documentation/devicetree/bindings/arm/cache.txt b/Documentation/devicetree/bindings/arm/cache.txt > new file mode 100644 > index 0000000..009cddb > --- /dev/null > +++ b/Documentation/devicetree/bindings/arm/cache.txt > @@ -0,0 +1,25 @@ > +========================================== > +ARM processors cache binding description > +========================================== > + > +Device tree bindings for ARM processor caches adhere to the cache bindings > +described in [3], in section 3.8 for multi-level and shared caches. > + > +On ARM, internal caches cannot be described in the cpu node but require > +specific nodes marked with compatible string set to "cache" (see [3], > +section 3.8). can you explain why > + > +Furthermore the cache bindings in [3] require the following property update: > + > +- [Table 3.9] cache-level: This property of cache nodes must match the cache > + level encoded in the processors CLIDR (v7) and > + CLIDR_EL1 (v8) registers, as described in [1][2]. > + > +All other properties and rules apply. > + > +[1] ARMv7-AR Reference Manual > + http://infocenter.arm.com/help/index.jsp > +[2] ARMv8-A Reference Manual > + http://infocenter.arm.com/help/index.jsp > +[3] ePAPR standard > + https://www.power.org/documentation/epapr-version-1-1/ > -- > 1.8.4 > > > -- > To unsubscribe from this list: send the line "unsubscribe devicetree" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, Dec 02, 2013 at 05:28:41PM +0000, Kumar Gala wrote: > > On Dec 2, 2013, at 10:20 AM, Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> wrote: > > > On ARM systems the cache topology cannot be probed at runtime, in > > particular, it is impossible to probe which CPUs share a given cache > > level. Power management software requires this knowledge to implement > > optimized power down sequences, hence this patch adds a document that > > defines the DT cache bindings for ARM systems. The bindings are compliant > > with ePAPR (PowerPC bindings), and rely on the cache bindings already > > standardized in the ePAPR v1.1 document; ARM required updates are underlined > > in the binding document. > > > > Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> > > --- > > Documentation/devicetree/bindings/arm/cache.txt | 25 +++++++++++++++++++++++++ > > 1 file changed, 25 insertions(+) > > create mode 100644 Documentation/devicetree/bindings/arm/cache.txt > > > > diff --git a/Documentation/devicetree/bindings/arm/cache.txt b/Documentation/devicetree/bindings/arm/cache.txt > > new file mode 100644 > > index 0000000..009cddb > > --- /dev/null > > +++ b/Documentation/devicetree/bindings/arm/cache.txt > > @@ -0,0 +1,25 @@ > > +========================================== > > +ARM processors cache binding description > > +========================================== > > + > > +Device tree bindings for ARM processor caches adhere to the cache bindings > > +described in [3], in section 3.8 for multi-level and shared caches. > > + > > +On ARM, internal caches cannot be described in the cpu node but require > > +specific nodes marked with compatible string set to "cache" (see [3], > > +section 3.8). > > can you explain why ARM v7 and v8 processors have a concept of cache levels which is valid even for internal caches. So either we add a cache-level property to internal caches or we do not use them for ARM. On top of that the definition of internal caches in the ePAPR is not well defined (for ARM) (what if you have multiple levels of internal caches ? - do we describe just L1 in the cpu node ?). Overall I think that is better to describe every cache object with a node and a compatible string, and shared caches are just cache nodes pointed at by multiple nodes (and private caches are just pointed at by one phandle in the respective cpu node). No strong position on that though, provided every cache "object" on ARM has a cache-level attached to it. Lorenzo > > + > > +Furthermore the cache bindings in [3] require the following property update: > > + > > +- [Table 3.9] cache-level: This property of cache nodes must match the cache > > + level encoded in the processors CLIDR (v7) and > > + CLIDR_EL1 (v8) registers, as described in [1][2]. > > + > > +All other properties and rules apply. > > + > > +[1] ARMv7-AR Reference Manual > > + http://infocenter.arm.com/help/index.jsp > > +[2] ARMv8-A Reference Manual > > + http://infocenter.arm.com/help/index.jsp > > +[3] ePAPR standard > > + https://www.power.org/documentation/epapr-version-1-1/ > > -- > > 1.8.4 > > > > > > -- > > To unsubscribe from this list: send the line "unsubscribe devicetree" in > > the body of a message to majordomo@vger.kernel.org > > More majordomo info at http://vger.kernel.org/majordomo-info.html > > -- > Employee of Qualcomm Innovation Center, Inc. > Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation > > -- To unsubscribe from this list: send the line "unsubscribe linux-pm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Dec 2, 2013, at 11:50 AM, Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> wrote: > On Mon, Dec 02, 2013 at 05:28:41PM +0000, Kumar Gala wrote: >> >> On Dec 2, 2013, at 10:20 AM, Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> wrote: >> >>> On ARM systems the cache topology cannot be probed at runtime, in >>> particular, it is impossible to probe which CPUs share a given cache >>> level. Power management software requires this knowledge to implement >>> optimized power down sequences, hence this patch adds a document that >>> defines the DT cache bindings for ARM systems. The bindings are compliant >>> with ePAPR (PowerPC bindings), and rely on the cache bindings already >>> standardized in the ePAPR v1.1 document; ARM required updates are underlined >>> in the binding document. >>> >>> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> >>> --- >>> Documentation/devicetree/bindings/arm/cache.txt | 25 +++++++++++++++++++++++++ >>> 1 file changed, 25 insertions(+) >>> create mode 100644 Documentation/devicetree/bindings/arm/cache.txt >>> >>> diff --git a/Documentation/devicetree/bindings/arm/cache.txt b/Documentation/devicetree/bindings/arm/cache.txt >>> new file mode 100644 >>> index 0000000..009cddb >>> --- /dev/null >>> +++ b/Documentation/devicetree/bindings/arm/cache.txt >>> @@ -0,0 +1,25 @@ >>> +========================================== >>> +ARM processors cache binding description >>> +========================================== >>> + >>> +Device tree bindings for ARM processor caches adhere to the cache bindings >>> +described in [3], in section 3.8 for multi-level and shared caches. >>> + >>> +On ARM, internal caches cannot be described in the cpu node but require >>> +specific nodes marked with compatible string set to "cache" (see [3], >>> +section 3.8). >> >> can you explain why > > ARM v7 and v8 processors have a concept of cache levels which is valid > even for internal caches. So either we add a cache-level property to > internal caches or we do not use them for ARM. There isn’t anything precluding a cachel-level property for internal caches. > On top of that the definition of internal caches in the ePAPR is not > well defined (for ARM) (what if you have multiple levels of internal > caches ? - do we describe just L1 in the cpu node ?). ePAPR has examples of multiple level’s of internal cache in the cpu node. > Overall I think that is better to describe every cache object with a node > and a compatible string, and shared caches are just cache nodes pointed at > by multiple nodes (and private caches are just pointed at by one phandle > in the respective cpu node). > > No strong position on that though, provided every cache "object" on ARM has a > cache-level attached to it. > > Lorenzo > >>> + >>> +Furthermore the cache bindings in [3] require the following property update: >>> + >>> +- [Table 3.9] cache-level: This property of cache nodes must match the cache >>> + level encoded in the processors CLIDR (v7) and >>> + CLIDR_EL1 (v8) registers, as described in [1][2]. >>> + >>> +All other properties and rules apply. >>> + >>> +[1] ARMv7-AR Reference Manual >>> + http://infocenter.arm.com/help/index.jsp >>> +[2] ARMv8-A Reference Manual >>> + http://infocenter.arm.com/help/index.jsp >>> +[3] ePAPR standard >>> + https://www.power.org/documentation/epapr-version-1-1/ >>> -- >>> 1.8.4 >>> >>> >>> -- >>> To unsubscribe from this list: send the line "unsubscribe devicetree" in >>> the body of a message to majordomo@vger.kernel.org >>> More majordomo info at http://vger.kernel.org/majordomo-info.html >> >> -- >> Employee of Qualcomm Innovation Center, Inc. >> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation >> >> > > -- > To unsubscribe from this list: send the line "unsubscribe devicetree" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, Dec 02, 2013 at 05:59:56PM +0000, Kumar Gala wrote: > > On Dec 2, 2013, at 11:50 AM, Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> wrote: > > > On Mon, Dec 02, 2013 at 05:28:41PM +0000, Kumar Gala wrote: > >> > >> On Dec 2, 2013, at 10:20 AM, Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> wrote: > >> > >>> On ARM systems the cache topology cannot be probed at runtime, in > >>> particular, it is impossible to probe which CPUs share a given cache > >>> level. Power management software requires this knowledge to implement > >>> optimized power down sequences, hence this patch adds a document that > >>> defines the DT cache bindings for ARM systems. The bindings are compliant > >>> with ePAPR (PowerPC bindings), and rely on the cache bindings already > >>> standardized in the ePAPR v1.1 document; ARM required updates are underlined > >>> in the binding document. > >>> > >>> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> > >>> --- > >>> Documentation/devicetree/bindings/arm/cache.txt | 25 +++++++++++++++++++++++++ > >>> 1 file changed, 25 insertions(+) > >>> create mode 100644 Documentation/devicetree/bindings/arm/cache.txt > >>> > >>> diff --git a/Documentation/devicetree/bindings/arm/cache.txt b/Documentation/devicetree/bindings/arm/cache.txt > >>> new file mode 100644 > >>> index 0000000..009cddb > >>> --- /dev/null > >>> +++ b/Documentation/devicetree/bindings/arm/cache.txt > >>> @@ -0,0 +1,25 @@ > >>> +========================================== > >>> +ARM processors cache binding description > >>> +========================================== > >>> + > >>> +Device tree bindings for ARM processor caches adhere to the cache bindings > >>> +described in [3], in section 3.8 for multi-level and shared caches. > >>> + > >>> +On ARM, internal caches cannot be described in the cpu node but require > >>> +specific nodes marked with compatible string set to "cache" (see [3], > >>> +section 3.8). > >> > >> can you explain why > > > > ARM v7 and v8 processors have a concept of cache levels which is valid > > even for internal caches. So either we add a cache-level property to > > internal caches or we do not use them for ARM. > > There isn't anything precluding a cachel-level property for internal caches. > > > On top of that the definition of internal caches in the ePAPR is not > > well defined (for ARM) (what if you have multiple levels of internal > > caches ? - do we describe just L1 in the cpu node ?). > > ePAPR has examples of multiple level's of internal cache in the cpu node. Where ? 3.7.3 (ePAPR v1.1) Internal (L1) Cache properties ? I do not think so. It does by using the definitions in 3.8 "Multi level and shared caches". So, since the cache bindings in 3.8 allow to describe a cache hierarchy on ARM, that's what I used. Declaring L1 in the CPU node buys us nothing, so I do not think there is a point in using it on ARM. That's why instead of changing the bindings for internal (L1) caches to require cache-level we should just ignore them IMHO. ePAPR v1.1 contains a bug (example page 47 - line 47) since a cpu node cannot contain a cache-level property, that's not a property of internal caches and it has to be declared in a node with: compatible = "cache"; unless my understanding of ePAPR is wrong. All I am saying is that multi-level cache bindings can be used to describe all cache hierarchies for ARM, we do not need internal (L1) cache bindings at all, it is just legacy stuff. Lorenzo -- To unsubscribe from this list: send the line "unsubscribe linux-pm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, Dec 02, 2013 at 04:20:04PM +0000, Lorenzo Pieralisi wrote: > On ARM systems the cache topology cannot be probed at runtime, in > particular, it is impossible to probe which CPUs share a given cache > level. Power management software requires this knowledge to implement > optimized power down sequences, hence this patch adds a document that > defines the DT cache bindings for ARM systems. The bindings are compliant > with ePAPR (PowerPC bindings), and rely on the cache bindings already > standardized in the ePAPR v1.1 document; ARM required updates are underlined > in the binding document. > > Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> > --- > Documentation/devicetree/bindings/arm/cache.txt | 25 +++++++++++++++++++++++++ > 1 file changed, 25 insertions(+) > create mode 100644 Documentation/devicetree/bindings/arm/cache.txt > > diff --git a/Documentation/devicetree/bindings/arm/cache.txt b/Documentation/devicetree/bindings/arm/cache.txt > new file mode 100644 > index 0000000..009cddb > --- /dev/null > +++ b/Documentation/devicetree/bindings/arm/cache.txt > @@ -0,0 +1,25 @@ > +========================================== > +ARM processors cache binding description > +========================================== > + > +Device tree bindings for ARM processor caches adhere to the cache bindings > +described in [3], in section 3.8 for multi-level and shared caches. For architected levels of cache, most of the cache properties can be probed through the system register interface. Should we be explicit about prohibiting properties in the DT describing cache characteristics that can be probed? > + > +On ARM, internal caches cannot be described in the cpu node but require > +specific nodes marked with compatible string set to "cache" (see [3], > +section 3.8). > + > +Furthermore the cache bindings in [3] require the following property update: > + > +- [Table 3.9] cache-level: This property of cache nodes must match the cache > + level encoded in the processors CLIDR (v7) and > + CLIDR_EL1 (v8) registers, as described in [1][2]. I still don't understand why a given cache definitely has the same level number assigned to it with respect to each CPU. Is that an architectural requirement? All those LoC, LoU, LoUIS properties etc. are probeable per-CPU not globally... I don't see why they couldn't be different. Cheers ---Dave -- To unsubscribe from this list: send the line "unsubscribe linux-pm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed, Dec 04, 2013 at 01:29:55PM +0000, Dave Martin wrote: > On Mon, Dec 02, 2013 at 04:20:04PM +0000, Lorenzo Pieralisi wrote: > > On ARM systems the cache topology cannot be probed at runtime, in > > particular, it is impossible to probe which CPUs share a given cache > > level. Power management software requires this knowledge to implement > > optimized power down sequences, hence this patch adds a document that > > defines the DT cache bindings for ARM systems. The bindings are compliant > > with ePAPR (PowerPC bindings), and rely on the cache bindings already > > standardized in the ePAPR v1.1 document; ARM required updates are underlined > > in the binding document. > > > > Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> > > --- > > Documentation/devicetree/bindings/arm/cache.txt | 25 +++++++++++++++++++++++++ > > 1 file changed, 25 insertions(+) > > create mode 100644 Documentation/devicetree/bindings/arm/cache.txt > > > > diff --git a/Documentation/devicetree/bindings/arm/cache.txt b/Documentation/devicetree/bindings/arm/cache.txt > > new file mode 100644 > > index 0000000..009cddb > > --- /dev/null > > +++ b/Documentation/devicetree/bindings/arm/cache.txt > > @@ -0,0 +1,25 @@ > > +========================================== > > +ARM processors cache binding description > > +========================================== > > + > > +Device tree bindings for ARM processor caches adhere to the cache bindings > > +described in [3], in section 3.8 for multi-level and shared caches. > > For architected levels of cache, most of the cache properties can be > probed through the system register interface. > > Should we be explicit about prohibiting properties in the DT describing > cache characteristics that can be probed? I think so, yes. > > +On ARM, internal caches cannot be described in the cpu node but require > > +specific nodes marked with compatible string set to "cache" (see [3], > > +section 3.8). > > + > > +Furthermore the cache bindings in [3] require the following property update: > > + > > +- [Table 3.9] cache-level: This property of cache nodes must match the cache > > + level encoded in the processors CLIDR (v7) and > > + CLIDR_EL1 (v8) registers, as described in [1][2]. > > I still don't understand why a given cache definitely has the same level > number assigned to it with respect to each CPU. Is that an architectural > requirement? > > All those LoC, LoU, LoUIS properties etc. are probeable per-CPU not > globally... I don't see why they couldn't be different. That's a good point, that definitely requires investigation. Thanks, Lorenzo -- To unsubscribe from this list: send the line "unsubscribe linux-pm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/Documentation/devicetree/bindings/arm/cache.txt b/Documentation/devicetree/bindings/arm/cache.txt new file mode 100644 index 0000000..009cddb --- /dev/null +++ b/Documentation/devicetree/bindings/arm/cache.txt @@ -0,0 +1,25 @@ +========================================== +ARM processors cache binding description +========================================== + +Device tree bindings for ARM processor caches adhere to the cache bindings +described in [3], in section 3.8 for multi-level and shared caches. + +On ARM, internal caches cannot be described in the cpu node but require +specific nodes marked with compatible string set to "cache" (see [3], +section 3.8). + +Furthermore the cache bindings in [3] require the following property update: + +- [Table 3.9] cache-level: This property of cache nodes must match the cache + level encoded in the processors CLIDR (v7) and + CLIDR_EL1 (v8) registers, as described in [1][2]. + +All other properties and rules apply. + +[1] ARMv7-AR Reference Manual + http://infocenter.arm.com/help/index.jsp +[2] ARMv8-A Reference Manual + http://infocenter.arm.com/help/index.jsp +[3] ePAPR standard + https://www.power.org/documentation/epapr-version-1-1/
On ARM systems the cache topology cannot be probed at runtime, in particular, it is impossible to probe which CPUs share a given cache level. Power management software requires this knowledge to implement optimized power down sequences, hence this patch adds a document that defines the DT cache bindings for ARM systems. The bindings are compliant with ePAPR (PowerPC bindings), and rely on the cache bindings already standardized in the ePAPR v1.1 document; ARM required updates are underlined in the binding document. Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> --- Documentation/devicetree/bindings/arm/cache.txt | 25 +++++++++++++++++++++++++ 1 file changed, 25 insertions(+) create mode 100644 Documentation/devicetree/bindings/arm/cache.txt