Commit 71e6fae3a9 fixed an issue with FORM2 affinity guests with NUMA
nodes in which the distance info is absent in
machine_state->numa_state->nodes. This happens when QEMU adds a default
NUMA node and when the user adds NUMA nodes without specifying the
distances.
During the discussions of the forementioned patch [1] it was found that
FORM1 guests were behaving in a strange way in the same scenario, with
the kernel seeing the distances between the nodes as '160', as we can
see in this example with 4 NUMA nodes without distance information:
$ numactl -H
available: 4 nodes (0-3)
(...)
node distances:
node 0 1 2 3
0: 10 160 160 160
1: 160 10 160 160
2: 160 160 10 160
3: 160 160 160 10
Turns out that we have the same problem with FORM1 guests - we are
calculating associativity domain using zeroed values. And as it also
turns out, the solution from 71e6fae3a9 applies to FORM1 as well.
This patch creates a wrapper called 'get_numa_distance' that contains
the logic used in FORM2 to define node distances when this information
is absent. This helper is then used in all places where we need to read
distance information from machine_state->numa_state->nodes. That way
we'll guarantee that the NUMA node distance is always being curated
before being used.
After this patch, the FORM1 guest mentioned above will have the
following topology:
$ numactl -H
available: 4 nodes (0-3)
(...)
node distances:
node 0 1 2 3
0: 10 20 20 20
1: 20 10 20 20
2: 20 20 10 20
3: 20 20 20 10
This is compatible with what FORM2 guests and other archs do in this
case.
[1] https://lists.gnu.org/archive/html/qemu-devel/2021-11/msg01960.html
Fixes: 690fbe4295 ("spapr_numa: consider user input when defining associativity")
CC: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
CC: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
A configuration that specifies multiple nodes without distance info
results in the non-local points in the FORM2 matrix having a distance of
0. This causes Linux to complain "Invalid distance value range" because
a node distance is smaller than the local distance.
Fix this by building a simple local / remote fallback for points where
distance information is missing.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Message-Id: <20211105135137.1584840-1-npiggin@gmail.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This patch has a handful of modifications for the recent added
FORM2 support:
- to not allocate more than the necessary size in 'distance_table'.
At this moment the array is oversized due to allocating uint32_t for
all elements, when most of them fits in an uint8_t. Fix it by
changing the array to uint8_t and allocating the exact size;
- use stl_be_p() to store the uint32_t at the start of 'distance_table';
- use sizeof(uint32_t) to skip the uint32_t length when populating the
distances;
- use the NUMA_DISTANCE_MIN macro from sysemu/numa.h to avoid hardcoding
the local distance value.
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210922122852.130054-2-danielhb413@gmail.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
numa_complete_configuration() in hw/core/numa.c always adds a NUMA node
for the pSeries machine if none was specified, but without node distance
information for the single node created.
NUMA FORM1 affinity code didn't rely on numa_state information to do its
job, but FORM2 does. As is now, this is the result of a pSeries guest
with NUMA FORM2 affinity when no NUMA nodes is specified:
$ numactl -H
available: 1 nodes (0)
node 0 cpus: 0
node 0 size: 16222 MB
node 0 free: 15681 MB
No distance information available.
This can be amended in spapr_numa_FORM2_write_rtas_tables(). We're
enforcing that the local distance (the distance to the node to itself) is
always 10. This allows for the proper creation of the NUMA distance tables,
fixing the output of 'numactl -H' in the guest:
$ numactl -H
available: 1 nodes (0)
node 0 cpus: 0
node 0 size: 16222 MB
node 0 free: 15685 MB
node distances:
node 0
0: 10
CC: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210920174947.556324-8-danielhb413@gmail.com>
Acked-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The main feature of FORM2 affinity support is the separation of NUMA
distances from ibm,associativity information. This allows for a more
flexible and straightforward NUMA distance assignment without relying on
complex associations between several levels of NUMA via
ibm,associativity matches. Another feature is its extensibility. This base
support contains the facilities for NUMA distance assignment, but in the
future more facilities will be added for latency, performance, bandwidth
and so on.
This patch implements the base FORM2 affinity support as follows:
- the use of FORM2 associativity is indicated by using bit 2 of byte 5
of ibm,architecture-vec-5. A FORM2 aware guest can choose to use FORM1
or FORM2 affinity. Setting both forms will default to FORM2. We're not
advertising FORM2 for pseries-6.1 and older machine versions to prevent
guest visible changes in those;
- ibm,associativity-reference-points has a new semantic. Instead of
being used to calculate distances via NUMA levels, it's now used to
indicate the primary domain index in the ibm,associativity domain of
each resource. In our case it's set to {0x4}, matching the position
where we already place logical_domain_id;
- two new RTAS DT artifacts are introduced: ibm,numa-lookup-index-table
and ibm,numa-distance-table. The index table is used to list all the
NUMA logical domains of the platform, in ascending order, and allows for
spartial NUMA configurations (although QEMU ATM doesn't support that).
ibm,numa-distance-table is an array that contains all the distances from
the first NUMA node to all other nodes, then the second NUMA node
distances to all other nodes and so on;
- get_max_dist_ref_points(), get_numa_assoc_size() and get_associativity()
now checks for OV5_FORM2_AFFINITY and returns FORM2 values if the guest
selected FORM2 affinity during CAS.
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210920174947.556324-7-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
FORM2 NUMA affinity is prepared to deal with empty (memory/cpu less)
NUMA nodes. This is used by the DAX KMEM driver to locate a PAPR SCM
device that has a different latency than the original NUMA node from the
regular memory. FORM2 is also able to deal with asymmetric NUMA
distances gracefully, something that our FORM1 implementation doesn't
do.
Move these FORM1 verifications to a new function and wait until after
CAS, when we're sure that we're sticking with FORM1, to enforce them.
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210920174947.556324-6-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Introducing a new NUMA affinity, FORM2, requires a new mechanism to
switch between affinity modes after CAS. Also, we want FORM2 data
structures and functions to be completely separated from the existing
FORM1 code, allowing us to avoid adding new code that inherits the
existing complexity of FORM1.
The idea of switching values used by the write_dt() functions in
spapr_numa.c was already introduced in the previous patch, and
the same approach will be used when dealing with the FORM1 and FORM2
arrays.
We can accomplish that by that by renaming the existing numa_assoc_array
to FORM1_assoc_array, which now is used exclusively to handle FORM1 affinity
data. A new helper get_associativity() is then introduced to be used by the
write_dt() functions to retrieve the current ibm,associativity array of
a given node, after considering affinity selection that might have been
done during CAS. All code that was using numa_assoc_array now needs to
retrieve the array by calling this function.
This will allow for an easier plug of FORM2 data later on.
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210920174947.556324-5-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The next preliminary step to introduce NUMA FORM2 affinity is to make
the existing code independent of FORM1 macros and values, i.e.
MAX_DISTANCE_REF_POINTS, NUMA_ASSOC_SIZE and VCPU_ASSOC_SIZE. This patch
accomplishes that by doing the following:
- move the NUMA related macros from spapr.h to spapr_numa.c where they
are used. spapr.h gets instead a 'NUMA_NODES_MAX_NUM' macro that is used
to refer to the maximum number of NUMA nodes, including GPU nodes, that
the machine can support;
- MAX_DISTANCE_REF_POINTS and NUMA_ASSOC_SIZE are renamed to
FORM1_DIST_REF_POINTS and FORM1_NUMA_ASSOC_SIZE. These FORM1 specific
macros are used in FORM1 init functions;
- code that uses MAX_DISTANCE_REF_POINTS now retrieves the
max_dist_ref_points value using get_max_dist_ref_points().
NUMA_ASSOC_SIZE is replaced by get_numa_assoc_size() and VCPU_ASSOC_SIZE
is replaced by get_vcpu_assoc_size(). These functions are used by the
generic device tree functions and h_home_node_associativity() and will
allow them to switch between FORM1 and FORM2 without changing their core
logic.
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210920174947.556324-4-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
When first introduced, 'legacy_numa' was a way to refer to guests that
either wouldn't be affected by associativity domain calculations, namely
the ones with only 1 NUMA node, and pre 5.2 guests that shouldn't be
affected by it because it would be an userspace change. Calling these
cases 'legacy_numa' was a convenient way to label these cases.
We're about to introduce a new NUMA affinity, FORM2, and this concept
of 'legacy_numa' is now a bit misleading because, although it is called
'legacy' it is in fact a FORM1 exclusive contraint.
This patch removes spapr_machine_using_legacy_numa() and open code the
conditions in each caller. While we're at it, move the chunk inside
spapr_numa_FORM1_affinity_init() that sets all numa_assoc_array domains
with 'node_id' to spapr_numa_define_FORM1_domains(). This chunk was
being executed if !pre_5_2_numa_associativity and num_nodes => 1, the
same conditions in which spapr_numa_define_FORM1_domains() is called
shortly after.
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210920174947.556324-3-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The upcoming FORM2 NUMA affinity will support asymmetric NUMA topologies
and doesn't need be concerned with all the legacy support for older
pseries FORM1 guests.
We're also not going to calculate associativity domains based on numa
distance (via spapr_numa_define_associativity_domains) since the
distances will be written directly into new DT properties.
Let's split FORM1 code into its own functions to allow for easier
insertion of FORM2 logic later on.
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210920174947.556324-2-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The current logic for calculating 'maxdomain' making it a sum of
numa_state->num_nodes with spapr->gpu_numa_id. spapr->gpu_numa_id is
used as a index to determine the next available NUMA id that a
given NVGPU can use.
The problem is that the initial value of gpu_numa_id, for any topology
that has more than one NUMA node, is equal to numa_state->num_nodes.
This means that our maxdomain will always be, at least, twice the
amount of existing NUMA nodes. This means that a guest with 4 NUMA
nodes will end up with the following max-associativity-domains:
rtas/ibm,max-associativity-domains
00000004 00000008 00000008 00000008 00000008
This overtuning of maxdomains doesn't go unnoticed in the guest, being
detected in SLUB during boot:
dmesg | grep SLUB
[ 0.000000] SLUB: HWalign=128, Order=0-3, MinObjects=0, CPUs=4, Nodes=8
SLUB is detecting 8 total nodes, with 4 nodes being online.
This patch fixes ibm,max-associativity-domains by considering the amount
of NVGPUs NUMA nodes presented in the guest, instead of just
spapr->gpu_numa_id.
Reported-by: Cédric Le Goater <clg@kaod.org>
Tested-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210128174213.1349181-4-danielhb413@gmail.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
We'll need to check the initial value given to spapr->gpu_numa_id when
building the rtas DT, so put it in a helper for easier access and to
avoid repetition.
Tested-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210128174213.1349181-3-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This function is used only in spapr_numa.c.
Tested-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210128174213.1349181-2-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
A new function called spapr_numa_define_associativity_domains()
is created to calculate the associativity domains and change
the associativity arrays considering user input. This is how
the associativity domain between two NUMA nodes A and B is
calculated:
- get the distance D between them
- get the correspondent NUMA level 'n_level' for D. This is done
via a helper called spapr_numa_get_numa_level()
- all associativity arrays were initialized with their own
numa_ids, and we're calculating the distance in node_id ascending
order, starting from node id 0 (the first node retrieved by
numa_state). This will have a cascade effect in the algorithm because
the associativity domains that node 0 defines will be carried over to
other nodes, and node 1 associativities will be carried over after
taking node 0 associativities into account, and so on. This
happens because we'll assign assoc_src as the associativity domain
of dst as well, for all NUMA levels beyond and including n_level.
The PPC kernel expects the associativity domains of the first node
(node id 0) to be always 0 [1], and this algorithm will grant that
by default.
Ultimately, all of this results in a best effort approximation for
the actual NUMA distances the user input in the command line. Given
the nature of how PAPR itself interprets NUMA distances versus the
expectations risen by how ACPI SLIT works, there might be better
algorithms but, in the end, it'll also result in another way to
approximate what the user really wanted.
To keep this commit message no longer than it already is, the next
patch will update the existing documentation in ppc-spapr-numa.rst
with more in depth details and design considerations/drawbacks.
[1] https://lore.kernel.org/linuxppc-dev/5e8fbea3-8faf-0951-172a-b41a2138fbcf@gmail.com/
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20201007172849.302240-5-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This is the first guest visible change introduced in
spapr_numa.c. The previous settings of both reference-points
and maxdomains were too restrictive, but enough for the
existing associativity we're setting in the resources.
We'll change that in the following patches, populating the
associativity arrays based on user input. For those changes
to be effective, reference-points and maxdomains must be
more flexible. After this patch, we'll have 4 distinct
levels of NUMA (0x4, 0x3, 0x2, 0x1) and maxdomains will
allow for any type of configuration the user intends to
do - under the scope and limitations of PAPR itself, of
course.
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20201007172849.302240-4-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The pSeries machine does not support asymmetrical NUMA
configurations. This doesn't make much of a different
since we're not using user input for pSeries NUMA setup,
but this will change in the next patches.
To avoid breaking existing setups, gate this change by
checking for legacy NUMA support.
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20201007172849.302240-3-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The current implementation of h_home_node_associativity hard codes
the values of associativity domains of the vcpus. Let's make
it consider the values already initialized in spapr->numa_assoc_array,
via the spapr_numa_get_vcpu_assoc() helper.
We want to set it and forget it, and for that we also need to
assert that we don't overflow the registers of the hypercall.
>From R4 to R9 we can squeeze in 12 associativity domains for
vcpus, so let's assert that VCPU_ASSOC_SIZE -1 isn't greater
than that.
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20200904172422.617460-4-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The work to be done in h_home_node_associativity() intersects
with what is already done in spapr_numa_fixup_cpu_dt(). This
patch creates a new helper, spapr_numa_get_vcpu_assoc(), to
be used for both spapr_numa_fixup_cpu_dt() and
h_home_node_associativity().
While we're at it, use memcpy() instead of loop assignment
to created the returned array.
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20200904172422.617460-3-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The implementation of this hypercall will be modified to use
spapr->numa_assoc_arrays input. Moving it to spapr_numa.c makes
make more sense.
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20200904172422.617460-2-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The NVLink2 GPUs works like a regular NUMA node with its
own associativity values, regardless of user input.
This can be handled inside spapr_numa_associativity_init(),
initializing NVGPU_MAX_NUM associativity arrays that can
be used by the GPUs.
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20200903220639.563090-5-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
In a similar fashion as the previous patch, let's move the
handling of ibm,associativity-lookup-arrays from spapr.c to
spapr_numa.c. A spapr_numa_write_assoc_lookup_arrays() helper was
created, and spapr_dt_dynamic_reconfiguration_memory() can now
use it to advertise the lookup-arrays.
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20200903220639.563090-4-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Vcpus have an additional paramenter to be appended, vcpu_id. This
also changes the size of the of property itself, which is being
represented in index 0 of numa_assoc_array[cpu->node_id],
and defaults to MAX_DISTANCE_REF_POINTS for all cases but
vcpus.
All this logic makes more sense in spapr_numa.c, where we handle
everything NUMA and associativity. A new helper spapr_numa_fixup_cpu_dt()
was added, and spapr.c uses it the same way as it was using the former
spapr_fixup_cpu_numa_dt().
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20200903220639.563090-3-danielhb413@gmail.com>
[dwg: Correct uint to int type, which can break windows builds]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The next step to centralize all NUMA/associativity handling in
the spapr machine is to create a 'one stop place' for all
things ibm,associativity.
This patch introduces numa_assoc_array, a 2 dimensional array
that will store all ibm,associativity arrays of all NUMA nodes.
This array is initialized in a new spapr_numa_associativity_init()
function, called in spapr_machine_init(). It is being initialized
with the same values used in other ibm,associativity properties
around spapr files (i.e. all zeros, last value is node_id).
The idea is to remove all hardcoded definitions and FDT writes
of ibm,associativity arrays, doing instead a call to the new
helper spapr_numa_write_associativity_dt() helper, that will
be able to write the DT with the correct values.
We'll start small, handling the trivial cases first. The
remaining instances of ibm,associativity will be handled
next.
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20200903220639.563090-2-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
We're going to make changes in how spapr handles all
ibm,associativity* related properties to enhance our current NUMA
support.
At this moment we have associativity code scattered all around
spapr_* files, with hardcoded values and array sizes. This
makes it harder to change any NUMA specific parameters in
the future. Having everything in the same place allows not
only for easier tuning, but also easier understanding since all
NUMA related code is on the same file.
This patch introduces a new file to gather all NUMA/associativity
handling code in spapr, spapr_numa.c. To get things started, let's
remove associativity-reference-points and max-associativity-domains
code from spapr_dt_rtas() to a new helper called spapr_numa_write_rtas_dt().
This will decouple spapr_dt_rtas() from the NUMA changes that
are going to happen in those two properties.
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20200901125645.118026-2-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
