[PATCH v4 7/7] powerpc/pseries: Add support for FORM2 associativity
David Gibson
david at gibson.dropbear.id.au
Thu Jun 24 13:08:47 AEST 2021
On Thu, Jun 17, 2021 at 10:21:05PM +0530, Aneesh Kumar K.V wrote:
> PAPR interface currently supports two different ways of communicating resource
> grouping details to the OS. These are referred to as Form 0 and Form 1
> associativity grouping. Form 0 is the older format and is now considered
> deprecated. This patch adds another resource grouping named FORM2.
>
> Signed-off-by: Daniel Henrique Barboza <danielhb413 at gmail.com>
> Signed-off-by: Aneesh Kumar K.V <aneesh.kumar at linux.ibm.com>
> ---
> Documentation/powerpc/associativity.rst | 135 ++++++++++++++++++++
> arch/powerpc/include/asm/firmware.h | 3 +-
> arch/powerpc/include/asm/prom.h | 1 +
> arch/powerpc/kernel/prom_init.c | 3 +-
> arch/powerpc/mm/numa.c | 149 +++++++++++++++++++++-
> arch/powerpc/platforms/pseries/firmware.c | 1 +
> 6 files changed, 286 insertions(+), 6 deletions(-)
> create mode 100644 Documentation/powerpc/associativity.rst
>
> diff --git a/Documentation/powerpc/associativity.rst b/Documentation/powerpc/associativity.rst
> new file mode 100644
> index 000000000000..93be604ac54d
> --- /dev/null
> +++ b/Documentation/powerpc/associativity.rst
> @@ -0,0 +1,135 @@
> +============================
> +NUMA resource associativity
> +=============================
> +
> +Associativity represents the groupings of the various platform resources into
> +domains of substantially similar mean performance relative to resources outside
> +of that domain. Resources subsets of a given domain that exhibit better
> +performance relative to each other than relative to other resources subsets
> +are represented as being members of a sub-grouping domain. This performance
> +characteristic is presented in terms of NUMA node distance within the Linux kernel.
> +From the platform view, these groups are also referred to as domains.
> +
> +PAPR interface currently supports different ways of communicating these resource
> +grouping details to the OS. These are referred to as Form 0, Form 1 and Form2
> +associativity grouping. Form 0 is the older format and is now considered deprecated.
> +
> +Hypervisor indicates the type/form of associativity used via "ibm,arcitecture-vec-5 property".
> +Bit 0 of byte 5 in the "ibm,architecture-vec-5" property indicates usage of Form 0 or Form 1.
> +A value of 1 indicates the usage of Form 1 associativity. For Form 2 associativity
> +bit 2 of byte 5 in the "ibm,architecture-vec-5" property is used.
> +
> +Form 0
> +-----
> +Form 0 associativity supports only two NUMA distance (LOCAL and REMOTE).
> +
> +Form 1
> +-----
> +With Form 1 a combination of ibm,associativity-reference-points and ibm,associativity
> +device tree properties are used to determine the NUMA distance between resource groups/domains.
> +
> +The “ibm,associativity” property contains one or more lists of numbers (domainID)
> +representing the resource’s platform grouping domains.
> +
> +The “ibm,associativity-reference-points” property contains one or more list of numbers
> +(domainID index) that represents the 1 based ordinal in the associativity lists.
> +The list of domainID index represnets increasing hierachy of
> resource grouping.
Typo "represnets". Also s/hierachy/hierarchy/
> +
> +ex:
> +{ primary domainID index, secondary domainID index, tertiary domainID index.. }
> +Linux kernel uses the domainID at the primary domainID index as the NUMA node id.
> +Linux kernel computes NUMA distance between two domains by recursively comparing
> +if they belong to the same higher-level domains. For mismatch at every higher
> +level of the resource group, the kernel doubles the NUMA distance between the
> +comparing domains.
The Form1 description is still kinda confusing, but I don't really
care. Form1 *is* confusing, it's Form2 that I hope will be clearer.
> +
> +Form 2
> +-------
> +Form 2 associativity format adds separate device tree properties representing NUMA node distance
> +thereby making the node distance computation flexible. Form 2 also allows flexible primary
> +domain numbering. With numa distance computation now detached from the index value of
> +"ibm,associativity" property, Form 2 allows a large number of primary domain ids at the
> +same domainID index representing resource groups of different performance/latency characteristics.
So, see you've removed the special handling of secondary IDs for pmem
- big improvement, thanks. IIUC, in this revised version, for Form2
there's really no reason for ibm,associativity-reference-points to
have >1 entry. Is that right?
In Form2 everything revolves around the primary domain ID - so much
that I suggest we come up with a short name for it. How about just
"node id" since that's how Linux uses it.
> +Hypervisor indicates the usage of FORM2 associativity using bit 2 of byte 5 in the
> +"ibm,architecture-vec-5" property.
> +
> +"ibm,numa-lookup-index-table" property contains one or more list numbers representing
> +the domainIDs present in the system. The offset of the domainID in this property is considered
> +the domainID index.
The lookup-index-table is all about compactifying the representation
of the distance matrix. Because node ids are sparse, the matrix of
distances is also effectively sparse. You don't want to have a huge
matrix in the DT with mostly meaningless entries, so you use the
lookup table to compact the node ids.
It's a bit confusing though, because you now have two representations
of exactly the same information the "full" node id (== primary
domainID) and the "short" node id (==domainID lookup table offset).
I can see three ways to clarify this:
A) Give the short node id an explicit name rather than difficult to
parse "domainID index" or "domainID offset" which gets easily
muddled with other uses of index and offset. Use that name
throughout.
B) Eliminate the short ID entirely, and use an explicit sparse
matrix representation for the distance table e.g. a list of
(nodeA, nodeB, distance) tuples
That does have the disadvantage that it's not structurally
guaranteed that you have entries for every pair of "active" node
ids.
C) Eliminate the "long ID" entirely. In the Form2 world, is there
actually any point allowing sparse node ids? In this case you'd
have the node id read from associativity and use that directly to
index the distance table
> +prop-encoded-array: The number N of the domainIDs encoded as with encode-int, followed by
> +N domainID encoded as with encode-int
> +
> +For ex:
> +ibm,numa-lookup-index-table = {4, 0, 8, 250, 252}, domainID index
> for domainID 8 is 1.
As noted "domainID index" is confusing here, because it's different
from the "domainID index" values in reference-points.
> +
> +"ibm,numa-distance-table" property contains one or more list of numbers representing the NUMA
> +distance between resource groups/domains present in the system.
> +
> +prop-encoded-array: The number N of the distance values encoded as with encode-int, followed by
> +N distance values encoded as with encode-bytes. The max distance value we could encode is 255.
The N here always has to be the square of the N in the
lookup-index-table, yes? In which case do we actually need to include
it?
> +For ex:
> +ibm,numa-lookup-index-table = {3, 0, 8, 40}
> +ibm,numa-distance-table = {9, 10, 20, 80, 20, 10, 160, 80, 160, 10}
> +
> + | 0 8 40
> +--|------------
> + |
> +0 | 10 20 80
> + |
> +8 | 20 10 160
> + |
> +40| 80 160 10
> +
> +
> +"ibm,associativity" property for resources in node 0, 8 and 40
> +
> +{ 3, 6, 7, 0 }
> +{ 3, 6, 9, 8 }
> +{ 3, 6, 7, 40}
> +
> +With "ibm,associativity-reference-points" { 0x3 }
> +
> +Each resource (drcIndex) now also supports additional optional device tree properties.
> +These properties are marked optional because the platform can choose not to export
> +them and provide the system topology details using the earlier defined device tree
> +properties alone. The optional device tree properties are used when adding new resources
> +(DLPAR) and when the platform didn't provide the topology details of the domain which
> +contains the newly added resource during boot.
In what circumstances would you envisage a hot-add creating a new NUMA
node, as opposed to adding resources to an existing (possibly empty)
node? Do you need any provision for removing NUMA nodes again?
> +"ibm,numa-lookup-index" property contains a number representing the domainID index to be used
> +when building the NUMA distance of the numa node to which this resource belongs. This can
> +be looked at as the index at which this new domainID would have appeared in
> +"ibm,numa-lookup-index-table" if the domain was present during boot. The domainID
Again, confusing use of "index" where it's used both for offsets in
ibm,associativity properties and for offsets in ibm,numa-lookup-index-table
> +of the new resource can be obtained from the existing "ibm,associativity" property. This
> +can be used to build distance information of a newly onlined NUMA node via DLPAR operation.
> +The value is 1 based array index value.
> +
> +prop-encoded-array: An integer encoded as with encode-int specifying the domainID index
> +
> +"ibm,numa-distance" property contains one or more list of numbers presenting the NUMA distance
> +from this resource domain to other resources.
> +
> +prop-encoded-array: The number N of the distance values encoded as with encode-int, followed by
> +N distance values encoded as with encode-bytes. The max distance value we could encode is 255.
Again, does N have to equal 2 * (existing number of nodes + 1)? If so
you should say so, if not you should specify how incomplete
information is interpreted.
> +For ex:
> +ibm,associativity = { 4, 5, 10, 50}
> +ibm,numa-lookup-index = { 4 }
> +ibm,numa-distance = {8, 160, 255, 80, 10, 160, 255, 80, 10}
> +
> +resulting in a new toplogy as below.
> + | 0 8 40 50
> +--|------------------
> + |
> +0 | 10 20 80 160
> + |
> +8 | 20 10 160 255
> + |
> +40| 80 160 10 80
> + |
> +50| 160 255 80 10
> +
> diff --git a/arch/powerpc/include/asm/firmware.h b/arch/powerpc/include/asm/firmware.h
> index 60b631161360..97a3bd9ffeb9 100644
> --- a/arch/powerpc/include/asm/firmware.h
> +++ b/arch/powerpc/include/asm/firmware.h
> @@ -53,6 +53,7 @@
> #define FW_FEATURE_ULTRAVISOR ASM_CONST(0x0000004000000000)
> #define FW_FEATURE_STUFF_TCE ASM_CONST(0x0000008000000000)
> #define FW_FEATURE_RPT_INVALIDATE ASM_CONST(0x0000010000000000)
> +#define FW_FEATURE_FORM2_AFFINITY ASM_CONST(0x0000020000000000)
>
> #ifndef __ASSEMBLY__
>
> @@ -73,7 +74,7 @@ enum {
> FW_FEATURE_HPT_RESIZE | FW_FEATURE_DRMEM_V2 |
> FW_FEATURE_DRC_INFO | FW_FEATURE_BLOCK_REMOVE |
> FW_FEATURE_PAPR_SCM | FW_FEATURE_ULTRAVISOR |
> - FW_FEATURE_RPT_INVALIDATE,
> + FW_FEATURE_RPT_INVALIDATE | FW_FEATURE_FORM2_AFFINITY,
> FW_FEATURE_PSERIES_ALWAYS = 0,
> FW_FEATURE_POWERNV_POSSIBLE = FW_FEATURE_OPAL | FW_FEATURE_ULTRAVISOR,
> FW_FEATURE_POWERNV_ALWAYS = 0,
> diff --git a/arch/powerpc/include/asm/prom.h b/arch/powerpc/include/asm/prom.h
> index df9fec9d232c..5c80152e8f18 100644
> --- a/arch/powerpc/include/asm/prom.h
> +++ b/arch/powerpc/include/asm/prom.h
> @@ -149,6 +149,7 @@ extern int of_read_drc_info_cell(struct property **prop,
> #define OV5_XCMO 0x0440 /* Page Coalescing */
> #define OV5_FORM1_AFFINITY 0x0580 /* FORM1 NUMA affinity */
> #define OV5_PRRN 0x0540 /* Platform Resource Reassignment */
> +#define OV5_FORM2_AFFINITY 0x0520 /* Form2 NUMA affinity */
> #define OV5_HP_EVT 0x0604 /* Hot Plug Event support */
> #define OV5_RESIZE_HPT 0x0601 /* Hash Page Table resizing */
> #define OV5_PFO_HW_RNG 0x1180 /* PFO Random Number Generator */
> diff --git a/arch/powerpc/kernel/prom_init.c b/arch/powerpc/kernel/prom_init.c
> index 64b9593038a7..496fdac54c29 100644
> --- a/arch/powerpc/kernel/prom_init.c
> +++ b/arch/powerpc/kernel/prom_init.c
> @@ -1070,7 +1070,8 @@ static const struct ibm_arch_vec ibm_architecture_vec_template __initconst = {
> #else
> 0,
> #endif
> - .associativity = OV5_FEAT(OV5_FORM1_AFFINITY) | OV5_FEAT(OV5_PRRN),
> + .associativity = OV5_FEAT(OV5_FORM1_AFFINITY) | OV5_FEAT(OV5_PRRN) |
> + OV5_FEAT(OV5_FORM2_AFFINITY),
> .bin_opts = OV5_FEAT(OV5_RESIZE_HPT) | OV5_FEAT(OV5_HP_EVT),
> .micro_checkpoint = 0,
> .reserved0 = 0,
> diff --git a/arch/powerpc/mm/numa.c b/arch/powerpc/mm/numa.c
> index d32729f235b8..5a7d94960fb7 100644
> --- a/arch/powerpc/mm/numa.c
> +++ b/arch/powerpc/mm/numa.c
> @@ -56,12 +56,17 @@ static int n_mem_addr_cells, n_mem_size_cells;
>
> #define FORM0_AFFINITY 0
> #define FORM1_AFFINITY 1
> +#define FORM2_AFFINITY 2
> static int affinity_form;
>
> #define MAX_DISTANCE_REF_POINTS 4
> static int max_associativity_domain_index;
> static const __be32 *distance_ref_points;
> static int distance_lookup_table[MAX_NUMNODES][MAX_DISTANCE_REF_POINTS];
> +static int numa_distance_table[MAX_NUMNODES][MAX_NUMNODES] = {
> + [0 ... MAX_NUMNODES - 1] = { [0 ... MAX_NUMNODES - 1] = -1 }
> +};
> +static int numa_id_index_table[MAX_NUMNODES];
>
> /*
> * Allocate node_to_cpumask_map based on number of available nodes
> @@ -166,6 +171,27 @@ static void unmap_cpu_from_node(unsigned long cpu)
> }
> #endif /* CONFIG_HOTPLUG_CPU || CONFIG_PPC_SPLPAR */
>
> +/*
> + * With FORM2 if we are not using logical domain ids, we will find
> + * both primary and seconday domains with same value. Hence always
> + * start comparison from secondary domains
IIUC, in this new draft, secondary domains are no longer a meaningful
thing in Form2, so this comment and code seem outdated.
> + */
> +static int __cpu_form2_distance(__be32 *cpu1_assoc, __be32 *cpu2_assoc)
> +{
> + int dist = 0;
> +
> + int i, index;
> +
> + for (i = 1; i < max_associativity_domain_index; i++) {
> + index = be32_to_cpu(distance_ref_points[i]);
> + if (cpu1_assoc[index] == cpu2_assoc[index])
> + break;
> + dist++;
> + }
> +
> + return dist;
> +}
> +
> static int __cpu_form1_distance(__be32 *cpu1_assoc, __be32 *cpu2_assoc)
> {
> int dist = 0;
> @@ -178,7 +204,6 @@ static int __cpu_form1_distance(__be32 *cpu1_assoc, __be32 *cpu2_assoc)
> break;
> dist++;
> }
> -
> return dist;
> }
>
> @@ -186,8 +211,9 @@ int cpu_distance(__be32 *cpu1_assoc, __be32 *cpu2_assoc)
> {
> /* We should not get called with FORM0 */
> VM_WARN_ON(affinity_form == FORM0_AFFINITY);
> -
> - return __cpu_form1_distance(cpu1_assoc, cpu2_assoc);
> + if (affinity_form == FORM1_AFFINITY)
> + return __cpu_form1_distance(cpu1_assoc, cpu2_assoc);
> + return __cpu_form2_distance(cpu1_assoc, cpu2_assoc);
> }
>
> /* must hold reference to node during call */
> @@ -201,7 +227,9 @@ int __node_distance(int a, int b)
> int i;
> int distance = LOCAL_DISTANCE;
>
> - if (affinity_form == FORM0_AFFINITY)
> + if (affinity_form == FORM2_AFFINITY)
> + return numa_distance_table[a][b];
> + else if (affinity_form == FORM0_AFFINITY)
> return ((a == b) ? LOCAL_DISTANCE : REMOTE_DISTANCE);
>
> for (i = 0; i < max_associativity_domain_index; i++) {
> @@ -303,15 +331,116 @@ static void initialize_form1_numa_distance(struct device_node *node)
>
> /*
> * Used to update distance information w.r.t newly added node.
> + * ibm,numa-lookup-index -> 4
> + * ibm,numa-distance -> {5, 20, 40, 60, 80, 10 }
> */
> void update_numa_distance(struct device_node *node)
> {
> + int i, nid, other_nid, other_nid_index = 0;
> + const __be32 *numa_indexp;
> + const __u8 *numa_distancep;
> + int numa_index, max_numa_index, numa_distance;
> +
> if (affinity_form == FORM0_AFFINITY)
> return;
> else if (affinity_form == FORM1_AFFINITY) {
> initialize_form1_numa_distance(node);
> return;
> }
> + /* FORM2 affinity */
> +
> + nid = of_node_to_nid_single(node);
> + if (nid == NUMA_NO_NODE)
> + return;
> +
> + /* Already initialized */
> + if (numa_distance_table[nid][nid] != -1)
IIUC, this should exactly correspond to the case where the new
resource lies in an existing NUMA node, yes?
> + return;
> + /*
> + * update node distance if not already populated.
> + */
> + numa_distancep = of_get_property(node, "ibm,numa-distance", NULL);
> + if (!numa_distancep)
> + return;
> +
> + numa_indexp = of_get_property(node, "ibm,numa-lookup-index", NULL);
> + if (!numa_indexp)
> + return;
> +
> + numa_index = of_read_number(numa_indexp, 1);
> + /*
> + * update the numa_id_index_table. Device tree look at index table as
> + * 1 based array indexing.
> + */
> + numa_id_index_table[numa_index - 1] = nid;
> +
> + max_numa_index = of_read_number((const __be32 *)numa_distancep, 1);
> + VM_WARN_ON(max_numa_index != 2 * numa_index);
You WARN_ON(), but you don't actually bail out to avoid indexing past
the end of the property below.
AFAICT none of this really works unless numa_index == (previous number
of numa nodes + 1), which makes all the use of these different
variables kind of confusing. If you had a variable that you just set
equal to the new number of numa nodes (previous number plus the one
being added), then I think you can replace all numa_index and
max_numa_index references with that and it will be clearer.
> + /* Skip the size which is encoded int */
> + numa_distancep += sizeof(__be32);
> +
> + /*
> + * First fill the distance information from other node to this node.
> + */
> + other_nid_index = 0;
> + for (i = 0; i < numa_index; i++) {
> + numa_distance = numa_distancep[i];
> + other_nid = numa_id_index_table[other_nid_index++];
> + numa_distance_table[other_nid][nid] = numa_distance;
> + }
> +
> + other_nid_index = 0;
> + for (; i < max_numa_index; i++) {
Again, splitting this loop and carrying i over seems a confusing way
to code this. It's basically two loops of N, one writing a row of the
distance matrix, one writing a column (other_nid will even go through
the same values in each loop).
> + numa_distance = numa_distancep[i];
> + other_nid = numa_id_index_table[other_nid_index++];
> + numa_distance_table[nid][other_nid] = numa_distance;
> + }
> +}
> +
> +/*
> + * ibm,numa-lookup-index-table= {N, domainid1, domainid2, ..... domainidN}
> + * ibm,numa-distance-table = { N, 1, 2, 4, 5, 1, 6, .... N elements}
> + */
> +static void initialize_form2_numa_distance_lookup_table(struct device_node *root)
> +{
> + const __u8 *numa_dist_table;
> + const __be32 *numa_lookup_index;
> + int numa_dist_table_length;
> + int max_numa_index, distance_index;
> + int i, curr_row = 0, curr_column = 0;
> +
> + numa_lookup_index = of_get_property(root, "ibm,numa-lookup-index-table", NULL);
> + max_numa_index = of_read_number(&numa_lookup_index[0], 1);
max_numa_index here has a different meaning to max_numa_index in the
previous function, which is pointlessly confusing.
> + /* first element of the array is the size and is encode-int */
> + numa_dist_table = of_get_property(root, "ibm,numa-distance-table", NULL);
> + numa_dist_table_length = of_read_number((const __be32 *)&numa_dist_table[0], 1);
> + /* Skip the size which is encoded int */
> + numa_dist_table += sizeof(__be32);
> +
> + pr_debug("numa_dist_table_len = %d, numa_dist_indexes_len = %d \n",
> + numa_dist_table_length, max_numa_index);
> +
> + for (i = 0; i < max_numa_index; i++)
> + /* +1 skip the max_numa_index in the property */
> + numa_id_index_table[i] = of_read_number(&numa_lookup_index[i + 1], 1);
> +
> +
> + VM_WARN_ON(numa_dist_table_length != max_numa_index * max_numa_index);
Again, you don't actually bail out in this case. And if it has to
have this value, what's the point of encoding it into the property.
> + for (distance_index = 0; distance_index < numa_dist_table_length; distance_index++) {
> + int nodeA = numa_id_index_table[curr_row];
> + int nodeB = numa_id_index_table[curr_column++];
You've already (sort of) verified that the distance table has size
N^2, in which case you can just to a simple two dimensional loop
rather than having to do ugly calculations of row and column.
> +
> + numa_distance_table[nodeA][nodeB] = numa_dist_table[distance_index];
> +
> + pr_debug("dist[%d][%d]=%d ", nodeA, nodeB, numa_distance_table[nodeA][nodeB]);
> + if (curr_column >= max_numa_index) {
> + curr_row++;
> + /* reset the column */
> + curr_column = 0;
> + }
> + }
> }
>
> static int __init find_primary_domain_index(void)
> @@ -324,6 +453,9 @@ static int __init find_primary_domain_index(void)
> */
> if (firmware_has_feature(FW_FEATURE_OPAL)) {
> affinity_form = FORM1_AFFINITY;
> + } else if (firmware_has_feature(FW_FEATURE_FORM2_AFFINITY)) {
> + dbg("Using form 2 affinity\n");
> + affinity_form = FORM2_AFFINITY;
> } else if (firmware_has_feature(FW_FEATURE_FORM1_AFFINITY)) {
> dbg("Using form 1 affinity\n");
> affinity_form = FORM1_AFFINITY;
> @@ -368,8 +500,17 @@ static int __init find_primary_domain_index(void)
>
> index = of_read_number(&distance_ref_points[1], 1);
> } else {
> + /*
> + * Both FORM1 and FORM2 affinity find the primary domain details
> + * at the same offset.
> + */
> index = of_read_number(distance_ref_points, 1);
> }
> + /*
> + * If it is FORM2 also initialize the distance table here.
> + */
> + if (affinity_form == FORM2_AFFINITY)
> + initialize_form2_numa_distance_lookup_table(root);
>
> /*
> * Warn and cap if the hardware supports more than
> diff --git a/arch/powerpc/platforms/pseries/firmware.c b/arch/powerpc/platforms/pseries/firmware.c
> index 5d4c2bc20bba..f162156b7b68 100644
> --- a/arch/powerpc/platforms/pseries/firmware.c
> +++ b/arch/powerpc/platforms/pseries/firmware.c
> @@ -123,6 +123,7 @@ vec5_fw_features_table[] = {
> {FW_FEATURE_PRRN, OV5_PRRN},
> {FW_FEATURE_DRMEM_V2, OV5_DRMEM_V2},
> {FW_FEATURE_DRC_INFO, OV5_DRC_INFO},
> + {FW_FEATURE_FORM2_AFFINITY, OV5_FORM2_AFFINITY},
> };
>
> static void __init fw_vec5_feature_init(const char *vec5, unsigned long len)
--
David Gibson | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_
| _way_ _around_!
http://www.ozlabs.org/~dgibson
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <http://lists.ozlabs.org/pipermail/linuxppc-dev/attachments/20210624/ba7e7c61/attachment-0001.sig>
More information about the Linuxppc-dev
mailing list