Re: [PATCH v3 1/7] ACPI/PPTT: Add Processor Properties Topology Table parsing

From: Jeremy Linton
Date: Thu Oct 19 2017 - 11:43:56 EST


On 10/19/2017 05:22 AM, Lorenzo Pieralisi wrote:
On Thu, Oct 12, 2017 at 02:48:50PM -0500, Jeremy Linton wrote:
ACPI 6.2 adds a new table, which describes how processing units
are related to each other in tree like fashion. Caches are
also sprinkled throughout the tree and describe the properties
of the caches in relation to other caches and processing units.

Add the code to parse the cache hierarchy and report the total
number of levels of cache for a given core using
acpi_find_last_cache_level() as well as fill out the individual
cores cache information with cache_setup_acpi() once the
cpu_cacheinfo structure has been populated by the arch specific
code.

Further, report peers in the topology using setup_acpi_cpu_topology()
to report a unique ID for each processing unit at a given level
in the tree. These unique id's can then be used to match related
processing units which exist as threads, COD (clusters
on die), within a given package, etc.

I think this patch should be split ((1) topology (2) cache), it is doing
too much which makes it hard to review.

If you look at the RFC, it only did cache parsing, the topology changes were added for v1. The cache bits are the ugly parts because they are walking up/down both the node tree, as well as the cache tree's attached to the nodes during the walk. Once that was in the place the addition of the cpu topology was trivial. But, trying to understand the cpu topology without first understanding the weird stuff done for the cache topology might not be the right way to approach this code.


[...]

+/* determine if the given node is a leaf node */
+static int acpi_pptt_leaf_node(struct acpi_table_header *table_hdr,
+ struct acpi_pptt_processor *node)
+{
+ struct acpi_subtable_header *entry;
+ unsigned long table_end;
+ u32 node_entry;
+ struct acpi_pptt_processor *cpu_node;
+
+ table_end = (unsigned long)table_hdr + table_hdr->length;
+ node_entry = (u32)((u8 *)node - (u8 *)table_hdr);
+ entry = (struct acpi_subtable_header *)((u8 *)table_hdr +
+ sizeof(struct acpi_table_pptt));
+
+ while (((unsigned long)entry) + sizeof(struct acpi_subtable_header) < table_end) {
+ cpu_node = (struct acpi_pptt_processor *)entry;
+ if ((entry->type == ACPI_PPTT_TYPE_PROCESSOR) &&
+ (cpu_node->parent == node_entry))
+ return 0;
+ entry = (struct acpi_subtable_header *)((u8 *)entry + entry->length);
+ }

A leaf node is a node with a valid acpi_id corresponding to an MADT
entry, right ? By the way, is this function really needed ?

Yes, because the only way to determine if it is a leaf node is to see if there are any references to it elsewhere in the table because the nodes point towards the root of the tree (rather than the other way).

This piece was the primary change for v1->v2.


+ return 1;
+}
+
+/*
+ * Find the subtable entry describing the provided processor
+ */
+static struct acpi_pptt_processor *acpi_find_processor_node(
+ struct acpi_table_header *table_hdr,
+ u32 acpi_cpu_id)
+{
+ struct acpi_subtable_header *entry;
+ unsigned long table_end;
+ struct acpi_pptt_processor *cpu_node;
+
+ table_end = (unsigned long)table_hdr + table_hdr->length;
+ entry = (struct acpi_subtable_header *)((u8 *)table_hdr +
+ sizeof(struct acpi_table_pptt));
+
+ /* find the processor structure associated with this cpuid */
+ while (((unsigned long)entry) + sizeof(struct acpi_subtable_header) < table_end) {
+ cpu_node = (struct acpi_pptt_processor *)entry;
+
+ if ((entry->type == ACPI_PPTT_TYPE_PROCESSOR) &&
+ acpi_pptt_leaf_node(table_hdr, cpu_node)) {

Is the leaf node check necessary ? Or you just need to check the
ACPI Processor ID valid flag (as discussed offline) ?

The valid flag doesn't mean anything for the leaf nodes, so its the only correct way of determining if the node _might_ have a valid madt/acpi ID. This actually should have the acpi_cpu_id checked as part of the if statement and the leaf node check below because doing it this way makes this parse n^2 instead of 2n. Of course in my mind, checking the id before we know it might be valid is backwards of the "logical" way to do it.


+ pr_debug("checking phy_cpu_id %d against acpi id %d\n",
+ acpi_cpu_id, cpu_node->acpi_processor_id);

Side note: I'd question (some of) these pr_debug() messages

+ if (acpi_cpu_id == cpu_node->acpi_processor_id) {
+ /* found the correct entry */
+ pr_debug("match found!\n");

Like this one for instance.

This one is a bit redundant, but I come from the school that I want to be able to debug a remote machine. Large blocks of silent code are a nightmare, particularly if you have a sysadmin level user driving the keyboard/etc.


+ return (struct acpi_pptt_processor *)entry;
+ }
+ }
+
+ if (entry->length == 0) {
+ pr_err("Invalid zero length subtable\n");
+ break;
+ }

This should be moved at the beginning of the loop.

Yah, the intention was to verify the next entry, but if its 0 then good point, the current one is probably invalid.


+ entry = (struct acpi_subtable_header *)
+ ((u8 *)entry + entry->length);
+ }
+
+ return NULL;
+}
+
+/*
+ * Given a acpi_pptt_processor node, walk up until we identify the
+ * package that the node is associated with or we run out of levels
+ * to request.
+ */
+static struct acpi_pptt_processor *acpi_find_processor_package_id(
+ struct acpi_table_header *table_hdr,
+ struct acpi_pptt_processor *cpu,
+ int level)
+{
+ struct acpi_pptt_processor *prev_node;
+
+ while (cpu && level && !(cpu->flags & ACPI_PPTT_PHYSICAL_PACKAGE)) {

I really do not understand what ACPI_PPTT_PHYSICAL_PACKAGE means and
more importantly, how it is actually used in this code.

?

Physical package maps to the package_id, which is generally defined to mean the "socket" and is used to terminate the cpu topology side of the parse.


This function is used to get a topology id (that is just a number for
a given topology level) for a given level starting from a given leaf
node.

This flag is the one decent part of the spec, because its the only level which actually is guaranteed to mean anything. Because the requirement that the sharability of cache nodes is described with general processor nodes it means that the number of nodes within a given leg of the tree is mostly meaningless because people sprinkle caches around the system, including potentially above the "socket" level.

Why do we care at all about ACPI_PPTT_PHYSICAL_PACKAGE ?

Because, it gives us a hard mapping to core siblings.


+ pr_debug("level %d\n", level);
+ prev_node = fetch_pptt_node(table_hdr, cpu->parent);
+ if (prev_node == NULL)
+ break;
+ cpu = prev_node;
+ level--;
+ }
+ return cpu;
+}
+
+static int acpi_parse_pptt(struct acpi_table_header *table_hdr, u32 acpi_cpu_id)
+{
+ int number_of_levels = 0;
+ struct acpi_pptt_processor *cpu;
+
+ cpu = acpi_find_processor_node(table_hdr, acpi_cpu_id);
+ if (cpu)
+ number_of_levels = acpi_process_node(table_hdr, cpu);
+
+ return number_of_levels;
+}
+
+#define ACPI_6_2_CACHE_TYPE_DATA (0x0)
+#define ACPI_6_2_CACHE_TYPE_INSTR (1<<2)
+#define ACPI_6_2_CACHE_TYPE_UNIFIED (1<<3)
+#define ACPI_6_2_CACHE_POLICY_WB (0x0)
+#define ACPI_6_2_CACHE_POLICY_WT (1<<4)
+#define ACPI_6_2_CACHE_READ_ALLOCATE (0x0)
+#define ACPI_6_2_CACHE_WRITE_ALLOCATE (0x01)
+#define ACPI_6_2_CACHE_RW_ALLOCATE (0x02)
+
+static u8 acpi_cache_type(enum cache_type type)
+{
+ switch (type) {
+ case CACHE_TYPE_DATA:
+ pr_debug("Looking for data cache\n");
+ return ACPI_6_2_CACHE_TYPE_DATA;
+ case CACHE_TYPE_INST:
+ pr_debug("Looking for instruction cache\n");
+ return ACPI_6_2_CACHE_TYPE_INSTR;
+ default:
+ pr_debug("Unknown cache type, assume unified\n");
+ case CACHE_TYPE_UNIFIED:
+ pr_debug("Looking for unified cache\n");
+ return ACPI_6_2_CACHE_TYPE_UNIFIED;
+ }
+}
+
+/* find the ACPI node describing the cache type/level for the given CPU */
+static struct acpi_pptt_cache *acpi_find_cache_node(
+ struct acpi_table_header *table_hdr, u32 acpi_cpu_id,
+ enum cache_type type, unsigned int level,
+ struct acpi_pptt_processor **node)
+{
+ int total_levels = 0;
+ struct acpi_pptt_cache *found = NULL;
+ struct acpi_pptt_processor *cpu_node;
+ u8 acpi_type = acpi_cache_type(type);
+
+ pr_debug("Looking for CPU %d's level %d cache type %d\n",
+ acpi_cpu_id, level, acpi_type);
+
+ cpu_node = acpi_find_processor_node(table_hdr, acpi_cpu_id);
+ if (!cpu_node)
+ return NULL;
+
+ do {
+ found = acpi_find_cache_level(table_hdr, cpu_node, &total_levels, level, acpi_type);
+ *node = cpu_node;
+ cpu_node = fetch_pptt_node(table_hdr, cpu_node->parent);
+ } while ((cpu_node) && (!found));
+
+ return found;
+}
+
+int acpi_find_last_cache_level(unsigned int cpu)
+{
+ u32 acpi_cpu_id;
+ struct acpi_table_header *table;
+ int number_of_levels = 0;
+ acpi_status status;
+
+ pr_debug("Cache Setup find last level cpu=%d\n", cpu);
+
+ acpi_cpu_id = acpi_cpu_get_madt_gicc(cpu)->uid;

This would break !ARM64.


+ status = acpi_get_table(ACPI_SIG_PPTT, 0, &table);
+ if (ACPI_FAILURE(status)) {
+ pr_err_once("No PPTT table found, cache topology may be inaccurate\n");

Yup, as in a way this does too... Without writing the binding code for another arch where that line is isn't clear at the moment. Part of the reason I put this in the arm64 directory.


+ } else {
+ number_of_levels = acpi_parse_pptt(table, acpi_cpu_id);
+ acpi_put_table(table);
+ }
+ pr_debug("Cache Setup find last level level=%d\n", number_of_levels);
+
+ return number_of_levels;
+}
+
+/*
+ * The ACPI spec implies that the fields in the cache structures are used to
+ * extend and correct the information probed from the hardware. In the case
+ * of arm64 the CCSIDR probing has been removed because it might be incorrect.
+ */
+static void update_cache_properties(struct cacheinfo *this_leaf,
+ struct acpi_pptt_cache *found_cache,
+ struct acpi_pptt_processor *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)
+ this_leaf->coherency_line_size = found_cache->line_size;
+ if (found_cache->flags & ACPI_PPTT_NUMBER_OF_SETS_VALID)
+ this_leaf->number_of_sets = found_cache->number_of_sets;
+ if (found_cache->flags & ACPI_PPTT_ASSOCIATIVITY_VALID)
+ this_leaf->ways_of_associativity = found_cache->associativity;
+ if (found_cache->flags & ACPI_PPTT_WRITE_POLICY_VALID)
+ switch (found_cache->attributes & ACPI_PPTT_MASK_WRITE_POLICY) {
+ case ACPI_6_2_CACHE_POLICY_WT:
+ this_leaf->attributes = CACHE_WRITE_THROUGH;
+ break;
+ case ACPI_6_2_CACHE_POLICY_WB:
+ this_leaf->attributes = CACHE_WRITE_BACK;
+ break;
+ default:
+ pr_err("Unknown ACPI cache policy %d\n",
+ found_cache->attributes & ACPI_PPTT_MASK_WRITE_POLICY);
+ }
+ if (found_cache->flags & ACPI_PPTT_ALLOCATION_TYPE_VALID)
+ switch (found_cache->attributes & ACPI_PPTT_MASK_ALLOCATION_TYPE) {
+ case ACPI_6_2_CACHE_READ_ALLOCATE:
+ this_leaf->attributes |= CACHE_READ_ALLOCATE;
+ break;
+ case ACPI_6_2_CACHE_WRITE_ALLOCATE:
+ this_leaf->attributes |= CACHE_WRITE_ALLOCATE;
+ break;
+ case ACPI_6_2_CACHE_RW_ALLOCATE:
+ this_leaf->attributes |=
+ CACHE_READ_ALLOCATE|CACHE_WRITE_ALLOCATE;
+ break;
+ default:
+ pr_err("Unknown ACPI cache allocation policy %d\n",
+ found_cache->attributes & ACPI_PPTT_MASK_ALLOCATION_TYPE);
+ }
+}
+
+static void cache_setup_acpi_cpu(struct acpi_table_header *table,
+ unsigned int cpu)
+{
+ struct acpi_pptt_cache *found_cache;
+ struct cpu_cacheinfo *this_cpu_ci = get_cpu_cacheinfo(cpu);
+ u32 acpi_cpu_id = acpi_cpu_get_madt_gicc(cpu)->uid;

Ditto.

+ struct cacheinfo *this_leaf;
+ unsigned int index = 0;
+ struct acpi_pptt_processor *cpu_node = NULL;
+
+ while (index < get_cpu_cacheinfo(cpu)->num_leaves) {
+ this_leaf = this_cpu_ci->info_list + index;
+ found_cache = acpi_find_cache_node(table, acpi_cpu_id,
+ this_leaf->type,
+ this_leaf->level,
+ &cpu_node);
+ pr_debug("found = %p %p\n", found_cache, cpu_node);
+ if (found_cache)
+ update_cache_properties(this_leaf,
+ found_cache,
+ cpu_node);
+
+ index++;
+ }
+}
+
+static int topology_setup_acpi_cpu(struct acpi_table_header *table,
+ unsigned int cpu, int level)
+{
+ struct acpi_pptt_processor *cpu_node;
+ u32 acpi_cpu_id = acpi_cpu_get_madt_gicc(cpu)->uid;

Ditto.

+ cpu_node = acpi_find_processor_node(table, acpi_cpu_id);
+ if (cpu_node) {
+ cpu_node = acpi_find_processor_package_id(table, cpu_node, level);

If level is 0 there is nothing to do here.

+ /* Only the first level has a guaranteed id */
+ if (level == 0)
+ return cpu_node->acpi_processor_id;
+ return (int)((u8 *)cpu_node - (u8 *)table);

Please explain to me the rationale behind this. To me acpi_processor_id
is as good as the cpu_node offset in the table to describe the topology
id at a given level, why special case level 0.

Level 0 is the only level guaranteed to have something set in the acpi_processor_id field. Its possible that values exist in nodes above this one, but they must _all_ be flagged and have matching container ids, and nothing in the spec requires that. Meaning that we need a guaranteed way to generate ids. This was added between v2->v3 after the discussion about making the ids a little nicer for the user.



On top of that, with this ID scheme, we would end up with
thread/core/cluster id potentially being non-sequential values
(depending on the PPTT table layout) which should not be a problem but
we'd better check how people are using them.

The thread (or core, depending on which is the 0 level) will have firmware provided Ids, everything else gets somewhat random looking but consistent ids. I commented earlier in this series that "normalizing" them is totally doable, although at the moment really only the physical_id is user visible and that should probably be normalized outside of this module in the arm64 topology parser if we want to actually do it. I'm not sure its worth the effort at least not as part of the general PPTT changes.



+ }
+ pr_err_once("PPTT table found, but unable to locate core for %d\n",
+ cpu);
+ return -ENOENT;
+}
+
+/*
+ * simply assign a ACPI cache entry to each known CPU cache entry
+ * determining which entries are shared is done later.

Add a kerneldoc style comment for an external interface.

That is a good point.


+ */
+int cache_setup_acpi(unsigned int cpu)
+{
+ struct acpi_table_header *table;
+ acpi_status status;
+
+ pr_debug("Cache Setup ACPI cpu %d\n", cpu);
+
+ status = acpi_get_table(ACPI_SIG_PPTT, 0, &table);
+ if (ACPI_FAILURE(status)) {
+ pr_err_once("No PPTT table found, cache topology may be inaccurate\n");
+ return -ENOENT;
+ }
+
+ cache_setup_acpi_cpu(table, cpu);
+ acpi_put_table(table);
+
+ return status;
+}
+
+/*
+ * Determine a topology unique ID for each thread/core/cluster/socket/etc.
+ * This ID can then be used to group peers.

Ditto.

+ */
+int setup_acpi_cpu_topology(unsigned int cpu, int level)
+{
+ struct acpi_table_header *table;
+ acpi_status status;
+ int retval;
+
+ status = acpi_get_table(ACPI_SIG_PPTT, 0, &table);
+ if (ACPI_FAILURE(status)) {
+ pr_err_once("No PPTT table found, cpu topology may be inaccurate\n");
+ return -ENOENT;
+ }
+ retval = topology_setup_acpi_cpu(table, cpu, level);
+ pr_debug("Topology Setup ACPI cpu %d, level %d ret = %d\n",
+ cpu, level, retval);
+ acpi_put_table(table);
+
+ return retval;

This value is just a token - with no HW meaning whatsoever and that's
where I question the ACPI_PPTT_PHYSICAL_PACKAGE flag usage in retrieving
it, you are not looking for a packageid (which has no meaning whatsoever
anyway and I wonder why it was added to the specs at all) you are
looking for an id at a given level.

If you look at the next patch in the series, to get the top level I pass an arbitrary large value as the "level" which should terminate on the PHYSICAL_PACKAGE rather than any intermediate nodes.



I will comment on the cache code separately - which deserves to
be in a separate patch to simplify the review, I avoided repeating
already reported review comments.

Lorenzo