| Message ID | 20171109210311.25655-6-jeremy.linton@arm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Thu, Nov 09, 2017 at 03:03:07PM -0600, Jeremy Linton wrote: > The /sys cache entries should support ACPI/PPTT generated cache > topology information. Lets detect ACPI systems and call > an arch specific cache_setup_acpi() routine to update the hardware > probed cache topology. > > For arm64, if ACPI is enabled, determine the max number of cache > levels and populate them using a PPTT table if one is available. > > Signed-off-by: Jeremy Linton <jeremy.linton@arm.com> > --- > arch/arm64/kernel/cacheinfo.c | 23 ++++++++++++++++++----- > drivers/acpi/pptt.c | 1 + > drivers/base/cacheinfo.c | 17 +++++++++++------ > include/linux/cacheinfo.h | 11 +++++++++-- > 4 files changed, 39 insertions(+), 13 deletions(-) > > diff --git a/arch/arm64/kernel/cacheinfo.c b/arch/arm64/kernel/cacheinfo.c > index 380f2e2fbed5..2e2cf0d312ba 100644 > --- a/arch/arm64/kernel/cacheinfo.c > +++ b/arch/arm64/kernel/cacheinfo.c > @@ -17,6 +17,7 @@ > * along with this program. If not, see <http://www.gnu.org/licenses/>. > */ > > +#include <linux/acpi.h> > #include <linux/cacheinfo.h> > #include <linux/of.h> > > @@ -44,9 +45,17 @@ static void ci_leaf_init(struct cacheinfo *this_leaf, > this_leaf->type = type; > } > > +#ifndef CONFIG_ACPI > +int acpi_find_last_cache_level(unsigned int cpu) > +{ > + /*ACPI kernels should be built with PPTT support*/ > + return 0; > +} > +#endif > + Why can't this be in generic acpi header #ifdef CONFIG_ACPI_PPTT int acpi_find_last_cache_level(unsigned int cpu) #else static int acpi_find_last_cache_level(unsigned int cpu) { return 0; } #endif or something similar to of_find_last_cache_level. I don't see why this has to be ARM64 specific. > static int __init_cache_level(unsigned int cpu) > { > - unsigned int ctype, level, leaves, of_level; > + unsigned int ctype, level, leaves, fw_level; > struct cpu_cacheinfo *this_cpu_ci = get_cpu_cacheinfo(cpu); > > for (level = 1, leaves = 0; level <= MAX_CACHE_LEVEL; level++) { > @@ -59,15 +68,19 @@ static int __init_cache_level(unsigned int cpu) > leaves += (ctype == CACHE_TYPE_SEPARATE) ? 2 : 1; > } > > - of_level = of_find_last_cache_level(cpu); > - if (level < of_level) { > + if (acpi_disabled) > + fw_level = of_find_last_cache_level(cpu); > + else > + fw_level = acpi_find_last_cache_level(cpu); > + > + if (level < fw_level) { > /* > * some external caches not specified in CLIDR_EL1 > * the information may be available in the device tree > * only unified external caches are considered here > */ > - leaves += (of_level - level); > - level = of_level; > + leaves += (fw_level - level); > + level = fw_level; > } > > this_cpu_ci->num_levels = level; Better to keep any arm64 related changes in a separate patch. > diff --git a/drivers/acpi/pptt.c b/drivers/acpi/pptt.c > index 9c9b8b4660e0..aa259502c4eb 100644 > --- a/drivers/acpi/pptt.c > +++ b/drivers/acpi/pptt.c > @@ -324,6 +324,7 @@ static void update_cache_properties(struct cacheinfo *this_leaf, > struct acpi_pptt_cache *found_cache, > struct acpi_pptt_processor *cpu_node) > { > + this_leaf->firmware_node = cpu_node; > if (found_cache->flags & ACPI_PPTT_SIZE_PROPERTY_VALID) > this_leaf->size = found_cache->size; > if (found_cache->flags & ACPI_PPTT_LINE_SIZE_VALID) > diff --git a/drivers/base/cacheinfo.c b/drivers/base/cacheinfo.c > index eb3af2739537..8eca279e50d1 100644 > --- a/drivers/base/cacheinfo.c > +++ b/drivers/base/cacheinfo.c > @@ -86,7 +86,7 @@ static int cache_setup_of_node(unsigned int cpu) > static inline bool cache_leaves_are_shared(struct cacheinfo *this_leaf, > struct cacheinfo *sib_leaf) > { > - return sib_leaf->of_node == this_leaf->of_node; > + return sib_leaf->firmware_node == this_leaf->firmware_node; > } > I am thinking it's better to even separate the DT and ACPI PPTT related changes. First refactor DT code renaming or adding new cacheinfo elements. Then add ACPI PPTT related changes. It helps to keep changes small at a time and easy for bisection in case of any issues. > /* OF properties to query for a given cache type */ > @@ -215,6 +215,11 @@ static inline bool cache_leaves_are_shared(struct cacheinfo *this_leaf, > } > #endif > > +int __weak cache_setup_acpi(unsigned int cpu) > +{ > + return -ENOTSUPP; > +} > + > static int cache_shared_cpu_map_setup(unsigned int cpu) > { > struct cpu_cacheinfo *this_cpu_ci = get_cpu_cacheinfo(cpu); > @@ -225,11 +230,11 @@ static int cache_shared_cpu_map_setup(unsigned int cpu) > if (this_cpu_ci->cpu_map_populated) > return 0; > > - if (of_have_populated_dt()) > + if (!acpi_disabled) > + ret = cache_setup_acpi(cpu); > + else if (of_have_populated_dt()) > ret = cache_setup_of_node(cpu); > - else if (!acpi_disabled) > - /* No cache property/hierarchy support yet in ACPI */ > - ret = -ENOTSUPP; > + > if (ret) > return ret; > > @@ -286,7 +291,7 @@ static void cache_shared_cpu_map_remove(unsigned int cpu) > > static void cache_override_properties(unsigned int cpu) > { > - if (of_have_populated_dt()) > + if (acpi_disabled && of_have_populated_dt()) > return cache_of_override_properties(cpu); > } > > diff --git a/include/linux/cacheinfo.h b/include/linux/cacheinfo.h > index 6a524bf6a06d..d1e9b8e01981 100644 > --- a/include/linux/cacheinfo.h > +++ b/include/linux/cacheinfo.h > @@ -36,6 +36,9 @@ enum cache_type { > * @of_node: if devicetree is used, this represents either the cpu node in > * case there's no explicit cache node or the cache node itself in the > * device tree > + * @firmware_node: Shared with of_node. When not using DT, this may contain > + * pointers to other firmware based values. Particularly ACPI/PPTT > + * unique values. > * @disable_sysfs: indicates whether this node is visible to the user via > * sysfs or not > * @priv: pointer to any private data structure specific to particular > @@ -64,8 +67,10 @@ struct cacheinfo { > #define CACHE_ALLOCATE_POLICY_MASK \ > (CACHE_READ_ALLOCATE | CACHE_WRITE_ALLOCATE) > #define CACHE_ID BIT(4) > - > - struct device_node *of_node; > + union { > + struct device_node *of_node; > + void *firmware_node; > + }; I would prefer struct device_node *of_node; changed to struct fwnode_handle *fwnode; You can then have struct pptt_fwnode { <.....> /*below fwnode allocated using acpi_alloc_fwnode_static */ struct fwnode_handle *fwnode; }; This gives a good starting point to abstract DT and ACPI. If not now, we can later implement fwnode.ops=pptt_cache_ops and then use get property for both DT and ACPI. -- Regards, Sudeep
On 11/20/2017 10:56 AM, Sudeep Holla wrote: (trimming) >> * case there's no explicit cache node or the cache node itself in the >> * device tree >> + * @firmware_node: Shared with of_node. When not using DT, this may contain >> + * pointers to other firmware based values. Particularly ACPI/PPTT >> + * unique values. >> * @disable_sysfs: indicates whether this node is visible to the user via >> * sysfs or not >> * @priv: pointer to any private data structure specific to particular >> @@ -64,8 +67,10 @@ struct cacheinfo { >> #define CACHE_ALLOCATE_POLICY_MASK \ >> (CACHE_READ_ALLOCATE | CACHE_WRITE_ALLOCATE) >> #define CACHE_ID BIT(4) >> - >> - struct device_node *of_node; >> + union { >> + struct device_node *of_node; >> + void *firmware_node; >> + }; > > I would prefer > struct device_node *of_node; > changed to > struct fwnode_handle *fwnode; > > You can then have > struct pptt_fwnode { > <.....> > /*below fwnode allocated using acpi_alloc_fwnode_static */ > struct fwnode_handle *fwnode; > }; > > This gives a good starting point to abstract DT and ACPI. > > If not now, we can later implement fwnode.ops=pptt_cache_ops and then > use get property for both DT and ACPI. I'm obviously confused why this keeps coming up. On the surface it sounds like a good idea. But then, given that I've actually implemented a portion of it, what becomes clear is that the PPTT isn't a good match. Converting the OF routines to use the fwnode is fairly straightforward, but that doesn't help the ACPI situation other than to create a lot of misleading code (and the possibility of creating nonstandard DSDT entries). The fact that this hasn't been done for other tables MADT/SLIT/SRAT/etc makes me wonder why we should do it for the PPTT? Particularly, when one considers fwnode is more a DSDT<->DT abstraction and thus has a lot of API surface that simply doesn't make any sense given the PPTT binary tree structure. Given that most of the fwnode routines are translating string properties (for example fwnode_property_read_string()) it might be possible to build a translator of some form which takes DT style properties and attempts to map them to the ACPI PPTT tree. What this adds I can't fathom, beyond the fact that suddenly the fwnode interface is a partial/brittle implementation where a large subset of the fwnode_operations will tend to be degenerate cases. The result likely will be a poorly implemented translator which breaks or is meaningless over a large part of the fwnode API surface.
On 20/11/17 18:02, Jeremy Linton wrote: > On 11/20/2017 10:56 AM, Sudeep Holla wrote: > > (trimming) > >>> * case there's no explicit cache node or the cache node >>> itself in the * device tree + * @firmware_node: Shared with >>> of_node. When not using DT, this may contain + * pointers to >>> other firmware based values. Particularly ACPI/PPTT + * unique >>> values. * @disable_sysfs: indicates whether this node is visible >>> to the user via * sysfs or not * @priv: pointer to any private >>> data structure specific to particular @@ -64,8 +67,10 @@ struct >>> cacheinfo { #define CACHE_ALLOCATE_POLICY_MASK \ >>> (CACHE_READ_ALLOCATE | CACHE_WRITE_ALLOCATE) #define CACHE_ID >>> BIT(4) - - struct device_node *of_node; + union { + >>> struct device_node *of_node; + void *firmware_node; + >>> }; >> >> I would prefer struct device_node *of_node; changed to struct >> fwnode_handle *fwnode; >> >> You can then have struct pptt_fwnode { <.....> /*below fwnode >> allocated using acpi_alloc_fwnode_static */ struct fwnode_handle >> *fwnode; }; >> >> This gives a good starting point to abstract DT and ACPI. >> >> If not now, we can later implement fwnode.ops=pptt_cache_ops and >> then use get property for both DT and ACPI. > > > I'm obviously confused why this keeps coming up. On the surface it > sounds like a good idea. But then, given that I've actually > implemented a portion of it, what becomes clear is that the PPTT > isn't a good match. Fair enough. > Converting the OF routines to use the fwnode is fairly > straightforward, but that doesn't help the ACPI situation other than > to create a lot of misleading code (and the possibility of creating > nonstandard DSDT entries). The fact that this hasn't been done for > other tables MADT/SLIT/SRAT/etc makes me wonder why we should do it > for the PPTT? > IRQ/IORT does use it. If we don't want to use it fine. But the union doesn't make sense and breaks the flow many other subsystems follow. Hence I raised. Sorry, I hadn't followed the last revision/discussion on this, my bad. But I had this thought since the beginning, hence I brought this up. > Particularly, when one considers fwnode is more a DSDT<->DT > abstraction and thus has a lot of API surface that simply doesn't > make any sense given the PPTT binary tree structure. Given that most > of the fwnode routines are translating string properties (for > example fwnode_property_read_string()) it might be possible to build > a translator of some form which takes DT style properties and > attempts to map them to the ACPI PPTT tree. What this adds I can't > fathom, beyond the fact that suddenly the fwnode interface is a > partial/brittle implementation where a large subset of the > fwnode_operations will tend to be degenerate cases. The result likely > will be a poorly implemented translator which breaks or is > meaningless over a large part of the fwnode API surface. Sure, I just mentioned ops thing, but that's optional. I just didn't like the union which has of_node and void ptr instead of fwhandle. I am fine if many agree that it's bad idea to use fwhandle here.
Hi, BTW: Thanks for looking at this! On 11/20/2017 12:14 PM, Sudeep Holla wrote: > > > On 20/11/17 18:02, Jeremy Linton wrote: >> On 11/20/2017 10:56 AM, Sudeep Holla wrote: >> >> (trimming) >> >>>> * case there's no explicit cache node or the cache node >>>> itself in the * device tree + * @firmware_node: Shared with >>>> of_node. When not using DT, this may contain + * pointers to >>>> other firmware based values. Particularly ACPI/PPTT + * unique >>>> values. * @disable_sysfs: indicates whether this node is visible >>>> to the user via * sysfs or not * @priv: pointer to any private >>>> data structure specific to particular @@ -64,8 +67,10 @@ struct >>>> cacheinfo { #define CACHE_ALLOCATE_POLICY_MASK \ >>>> (CACHE_READ_ALLOCATE | CACHE_WRITE_ALLOCATE) #define CACHE_ID >>>> BIT(4) - - struct device_node *of_node; + union { + >>>> struct device_node *of_node; + void *firmware_node; + >>>> }; >>> >>> I would prefer struct device_node *of_node; changed to struct >>> fwnode_handle *fwnode; >>> >>> You can then have struct pptt_fwnode { <.....> /*below fwnode >>> allocated using acpi_alloc_fwnode_static */ struct fwnode_handle >>> *fwnode; }; >>> >>> This gives a good starting point to abstract DT and ACPI. >>> >>> If not now, we can later implement fwnode.ops=pptt_cache_ops and >>> then use get property for both DT and ACPI. >> >> >> I'm obviously confused why this keeps coming up. On the surface it >> sounds like a good idea. But then, given that I've actually >> implemented a portion of it, what becomes clear is that the PPTT >> isn't a good match. > > Fair enough. > >> Converting the OF routines to use the fwnode is fairly >> straightforward, but that doesn't help the ACPI situation other than >> to create a lot of misleading code (and the possibility of creating >> nonstandard DSDT entries). The fact that this hasn't been done for >> other tables MADT/SLIT/SRAT/etc makes me wonder why we should do it >> for the PPTT? >> > > IRQ/IORT does use it. If we don't want to use it fine. But the union > doesn't make sense and breaks the flow many other subsystems follow. > Hence I raised. Sorry, I hadn't followed the last revision/discussion on > this, my bad. But I had this thought since the beginning, hence I > brought this up. > >> Particularly, when one considers fwnode is more a DSDT<->DT >> abstraction and thus has a lot of API surface that simply doesn't >> make any sense given the PPTT binary tree structure. Given that most >> of the fwnode routines are translating string properties (for >> example fwnode_property_read_string()) it might be possible to build >> a translator of some form which takes DT style properties and >> attempts to map them to the ACPI PPTT tree. What this adds I can't >> fathom, beyond the fact that suddenly the fwnode interface is a >> partial/brittle implementation where a large subset of the >> fwnode_operations will tend to be degenerate cases. The result likely >> will be a poorly implemented translator which breaks or is >> meaningless over a large part of the fwnode API surface. > > Sure, I just mentioned ops thing, but that's optional. I just didn't > like the union which has of_node and void ptr instead of fwhandle. I am > fine if many agree that it's bad idea to use fwhandle here. So, if we say the union is bad, as is a common fwnode_handle, shall I just make the "firmware_node" (pptt_node?) standalone? That adds a if (acpi) check in cache_leaves_are_shared() which is the only place that the cache topology code does anything with the ACPI field. Also, if you missed it there is a further patch which overrides the cache type field if everything else on the PPTT node is valid and the cache type is NONE. http://linux-arm.org/git?p=linux-jlinton.git;a=log;h=refs/heads/pptt_v4 finally, I will split out the of_node/fw_node, and move the #ifdef ACPI somewhere else.
diff --git a/arch/arm64/kernel/cacheinfo.c b/arch/arm64/kernel/cacheinfo.c index 380f2e2fbed5..2e2cf0d312ba 100644 --- a/arch/arm64/kernel/cacheinfo.c +++ b/arch/arm64/kernel/cacheinfo.c @@ -17,6 +17,7 @@ * along with this program. If not, see <http://www.gnu.org/licenses/>. */ +#include <linux/acpi.h> #include <linux/cacheinfo.h> #include <linux/of.h> @@ -44,9 +45,17 @@ static void ci_leaf_init(struct cacheinfo *this_leaf, this_leaf->type = type; } +#ifndef CONFIG_ACPI +int acpi_find_last_cache_level(unsigned int cpu) +{ + /*ACPI kernels should be built with PPTT support*/ + return 0; +} +#endif + static int __init_cache_level(unsigned int cpu) { - unsigned int ctype, level, leaves, of_level; + unsigned int ctype, level, leaves, fw_level; struct cpu_cacheinfo *this_cpu_ci = get_cpu_cacheinfo(cpu); for (level = 1, leaves = 0; level <= MAX_CACHE_LEVEL; level++) { @@ -59,15 +68,19 @@ static int __init_cache_level(unsigned int cpu) leaves += (ctype == CACHE_TYPE_SEPARATE) ? 2 : 1; } - of_level = of_find_last_cache_level(cpu); - if (level < of_level) { + if (acpi_disabled) + fw_level = of_find_last_cache_level(cpu); + else + fw_level = acpi_find_last_cache_level(cpu); + + if (level < fw_level) { /* * some external caches not specified in CLIDR_EL1 * the information may be available in the device tree * only unified external caches are considered here */ - leaves += (of_level - level); - level = of_level; + leaves += (fw_level - level); + level = fw_level; } this_cpu_ci->num_levels = level; diff --git a/drivers/acpi/pptt.c b/drivers/acpi/pptt.c index 9c9b8b4660e0..aa259502c4eb 100644 --- a/drivers/acpi/pptt.c +++ b/drivers/acpi/pptt.c @@ -324,6 +324,7 @@ static void update_cache_properties(struct cacheinfo *this_leaf, struct acpi_pptt_cache *found_cache, struct acpi_pptt_processor *cpu_node) { + this_leaf->firmware_node = cpu_node; if (found_cache->flags & ACPI_PPTT_SIZE_PROPERTY_VALID) this_leaf->size = found_cache->size; if (found_cache->flags & ACPI_PPTT_LINE_SIZE_VALID) diff --git a/drivers/base/cacheinfo.c b/drivers/base/cacheinfo.c index eb3af2739537..8eca279e50d1 100644 --- a/drivers/base/cacheinfo.c +++ b/drivers/base/cacheinfo.c @@ -86,7 +86,7 @@ static int cache_setup_of_node(unsigned int cpu) static inline bool cache_leaves_are_shared(struct cacheinfo *this_leaf, struct cacheinfo *sib_leaf) { - return sib_leaf->of_node == this_leaf->of_node; + return sib_leaf->firmware_node == this_leaf->firmware_node; } /* OF properties to query for a given cache type */ @@ -215,6 +215,11 @@ static inline bool cache_leaves_are_shared(struct cacheinfo *this_leaf, } #endif +int __weak cache_setup_acpi(unsigned int cpu) +{ + return -ENOTSUPP; +} + static int cache_shared_cpu_map_setup(unsigned int cpu) { struct cpu_cacheinfo *this_cpu_ci = get_cpu_cacheinfo(cpu); @@ -225,11 +230,11 @@ static int cache_shared_cpu_map_setup(unsigned int cpu) if (this_cpu_ci->cpu_map_populated) return 0; - if (of_have_populated_dt()) + if (!acpi_disabled) + ret = cache_setup_acpi(cpu); + else if (of_have_populated_dt()) ret = cache_setup_of_node(cpu); - else if (!acpi_disabled) - /* No cache property/hierarchy support yet in ACPI */ - ret = -ENOTSUPP; + if (ret) return ret; @@ -286,7 +291,7 @@ static void cache_shared_cpu_map_remove(unsigned int cpu) static void cache_override_properties(unsigned int cpu) { - if (of_have_populated_dt()) + if (acpi_disabled && of_have_populated_dt()) return cache_of_override_properties(cpu); } diff --git a/include/linux/cacheinfo.h b/include/linux/cacheinfo.h index 6a524bf6a06d..d1e9b8e01981 100644 --- a/include/linux/cacheinfo.h +++ b/include/linux/cacheinfo.h @@ -36,6 +36,9 @@ enum cache_type { * @of_node: if devicetree is used, this represents either the cpu node in * case there's no explicit cache node or the cache node itself in the * device tree + * @firmware_node: Shared with of_node. When not using DT, this may contain + * pointers to other firmware based values. Particularly ACPI/PPTT + * unique values. * @disable_sysfs: indicates whether this node is visible to the user via * sysfs or not * @priv: pointer to any private data structure specific to particular @@ -64,8 +67,10 @@ struct cacheinfo { #define CACHE_ALLOCATE_POLICY_MASK \ (CACHE_READ_ALLOCATE | CACHE_WRITE_ALLOCATE) #define CACHE_ID BIT(4) - - struct device_node *of_node; + union { + struct device_node *of_node; + void *firmware_node; + }; bool disable_sysfs; void *priv; }; @@ -98,6 +103,8 @@ int func(unsigned int cpu) \ struct cpu_cacheinfo *get_cpu_cacheinfo(unsigned int cpu); int init_cache_level(unsigned int cpu); int populate_cache_leaves(unsigned int cpu); +int cache_setup_acpi(unsigned int cpu); +int acpi_find_last_cache_level(unsigned int cpu); const struct attribute_group *cache_get_priv_group(struct cacheinfo *this_leaf);
The /sys cache entries should support ACPI/PPTT generated cache topology information. Lets detect ACPI systems and call an arch specific cache_setup_acpi() routine to update the hardware probed cache topology. For arm64, if ACPI is enabled, determine the max number of cache levels and populate them using a PPTT table if one is available. Signed-off-by: Jeremy Linton <jeremy.linton@arm.com> --- arch/arm64/kernel/cacheinfo.c | 23 ++++++++++++++++++----- drivers/acpi/pptt.c | 1 + drivers/base/cacheinfo.c | 17 +++++++++++------ include/linux/cacheinfo.h | 11 +++++++++-- 4 files changed, 39 insertions(+), 13 deletions(-)