8f9e91ad03
These patches were generated from: https://github.com/raspberrypi/linux/commits/rpi-6.12.y With the following command: git format-patch -N v6.12.27..HEAD (HEAD -> 8d3206ee456a5ecdf9ddbfd8e5e231e4f0cd716e) Exceptions: - (def)configs patches - github workflows patches - applied & reverted patches - readme patches - wireless patches Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>
188 lines
5.1 KiB
Diff
188 lines
5.1 KiB
Diff
From b3c4a07fdd65476971040682b16ce43e872c7ea7 Mon Sep 17 00:00:00 2001
|
|
From: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
|
|
Date: Wed, 22 May 2024 14:12:42 +0100
|
|
Subject: [PATCH] mm/mempolicy: Add MPOL_RANDOM
|
|
|
|
To help work around certain memory controller limitations or similar, a
|
|
random NUMA allocation memory policy is added.
|
|
|
|
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
|
|
---
|
|
include/uapi/linux/mempolicy.h | 1 +
|
|
mm/mempolicy.c | 71 ++++++++++++++++++++++++++++++++++
|
|
2 files changed, 72 insertions(+)
|
|
|
|
--- a/include/uapi/linux/mempolicy.h
|
|
+++ b/include/uapi/linux/mempolicy.h
|
|
@@ -24,6 +24,7 @@ enum {
|
|
MPOL_LOCAL,
|
|
MPOL_PREFERRED_MANY,
|
|
MPOL_WEIGHTED_INTERLEAVE,
|
|
+ MPOL_RANDOM,
|
|
MPOL_MAX, /* always last member of enum */
|
|
};
|
|
|
|
--- a/mm/mempolicy.c
|
|
+++ b/mm/mempolicy.c
|
|
@@ -41,6 +41,9 @@
|
|
* preferred many Try a set of nodes first before normal fallback. This is
|
|
* similar to preferred without the special case.
|
|
*
|
|
+ * random Allocate memory from a random node out of allowed set of
|
|
+ * nodes.
|
|
+ *
|
|
* default Allocate on the local node first, or when on a VMA
|
|
* use the process policy. This is what Linux always did
|
|
* in a NUMA aware kernel and still does by, ahem, default.
|
|
@@ -452,6 +455,10 @@ static const struct mempolicy_operations
|
|
.create = mpol_new_nodemask,
|
|
.rebind = mpol_rebind_nodemask,
|
|
},
|
|
+ [MPOL_RANDOM] = {
|
|
+ .create = mpol_new_nodemask,
|
|
+ .rebind = mpol_rebind_nodemask,
|
|
+ },
|
|
};
|
|
|
|
static bool migrate_folio_add(struct folio *folio, struct list_head *foliolist,
|
|
@@ -900,6 +907,7 @@ static void get_policy_nodemask(struct m
|
|
case MPOL_PREFERRED:
|
|
case MPOL_PREFERRED_MANY:
|
|
case MPOL_WEIGHTED_INTERLEAVE:
|
|
+ case MPOL_RANDOM:
|
|
*nodes = pol->nodes;
|
|
break;
|
|
case MPOL_LOCAL:
|
|
@@ -1917,6 +1925,27 @@ static unsigned int interleave_nodes(str
|
|
return nid;
|
|
}
|
|
|
|
+static unsigned int read_once_policy_nodemask(struct mempolicy *pol, nodemask_t *mask);
|
|
+
|
|
+static unsigned int random_nodes(struct mempolicy *policy)
|
|
+{
|
|
+ unsigned int nid = first_node(policy->nodes);
|
|
+ unsigned int cpuset_mems_cookie;
|
|
+ nodemask_t nodemask;
|
|
+ unsigned int r;
|
|
+
|
|
+ r = get_random_u32_below(read_once_policy_nodemask(policy, &nodemask));
|
|
+
|
|
+ /* to prevent miscount, use tsk->mems_allowed_seq to detect rebind */
|
|
+ do {
|
|
+ cpuset_mems_cookie = read_mems_allowed_begin();
|
|
+ while (r--)
|
|
+ nid = next_node_in(nid, policy->nodes);
|
|
+ } while (read_mems_allowed_retry(cpuset_mems_cookie));
|
|
+
|
|
+ return nid;
|
|
+}
|
|
+
|
|
/*
|
|
* Depending on the memory policy provide a node from which to allocate the
|
|
* next slab entry.
|
|
@@ -1962,6 +1991,9 @@ unsigned int mempolicy_slab_node(void)
|
|
case MPOL_LOCAL:
|
|
return node;
|
|
|
|
+ case MPOL_RANDOM:
|
|
+ return random_nodes(policy);
|
|
+
|
|
default:
|
|
BUG();
|
|
}
|
|
@@ -2042,6 +2074,33 @@ static unsigned int interleave_nid(struc
|
|
return nid;
|
|
}
|
|
|
|
+static unsigned int random_nid(struct mempolicy *pol,
|
|
+ struct vm_area_struct *vma,
|
|
+ pgoff_t ilx)
|
|
+{
|
|
+ nodemask_t nodemask;
|
|
+ unsigned int r, nnodes;
|
|
+ int i, nid;
|
|
+
|
|
+ nnodes = read_once_policy_nodemask(pol, &nodemask);
|
|
+ if (!nnodes)
|
|
+ return numa_node_id();
|
|
+
|
|
+ /*
|
|
+ * QQQ
|
|
+ * Can we say hash of vma+ilx is sufficiently random but still
|
|
+ * stable in case of reliance on stable, as it appears is with
|
|
+ * mpol_misplaced and interleaving?
|
|
+ */
|
|
+ r = hash_long((unsigned long)vma + ilx,
|
|
+ ilog2(roundup_pow_of_two(nnodes)));
|
|
+
|
|
+ nid = first_node(nodemask);
|
|
+ for (i = 0; i < r; i++)
|
|
+ nid = next_node(nid, nodemask);
|
|
+ return nid;
|
|
+}
|
|
+
|
|
/*
|
|
* Return a nodemask representing a mempolicy for filtering nodes for
|
|
* page allocation, together with preferred node id (or the input node id).
|
|
@@ -2085,6 +2144,9 @@ static nodemask_t *policy_nodemask(gfp_t
|
|
weighted_interleave_nodes(pol) :
|
|
weighted_interleave_nid(pol, ilx);
|
|
break;
|
|
+ case MPOL_RANDOM:
|
|
+ *nid = random_nodes(pol);
|
|
+ break;
|
|
}
|
|
|
|
return nodemask;
|
|
@@ -2153,6 +2215,7 @@ bool init_nodemask_of_mempolicy(nodemask
|
|
case MPOL_BIND:
|
|
case MPOL_INTERLEAVE:
|
|
case MPOL_WEIGHTED_INTERLEAVE:
|
|
+ case MPOL_RANDOM:
|
|
*mask = mempolicy->nodes;
|
|
break;
|
|
|
|
@@ -2633,6 +2696,7 @@ bool __mpol_equal(struct mempolicy *a, s
|
|
case MPOL_PREFERRED:
|
|
case MPOL_PREFERRED_MANY:
|
|
case MPOL_WEIGHTED_INTERLEAVE:
|
|
+ case MPOL_RANDOM:
|
|
return !!nodes_equal(a->nodes, b->nodes);
|
|
case MPOL_LOCAL:
|
|
return true;
|
|
@@ -2824,6 +2888,10 @@ int mpol_misplaced(struct folio *folio,
|
|
polnid = zonelist_node_idx(z);
|
|
break;
|
|
|
|
+ case MPOL_RANDOM:
|
|
+ polnid = random_nid(pol, vma, ilx);
|
|
+ break;
|
|
+
|
|
default:
|
|
BUG();
|
|
}
|
|
@@ -3169,6 +3237,7 @@ static const char * const policy_modes[]
|
|
[MPOL_WEIGHTED_INTERLEAVE] = "weighted interleave",
|
|
[MPOL_LOCAL] = "local",
|
|
[MPOL_PREFERRED_MANY] = "prefer (many)",
|
|
+ [MPOL_RANDOM] = "random",
|
|
};
|
|
|
|
/**
|
|
@@ -3231,6 +3300,7 @@ int mpol_parse_str(char *str, struct mem
|
|
break;
|
|
case MPOL_INTERLEAVE:
|
|
case MPOL_WEIGHTED_INTERLEAVE:
|
|
+ case MPOL_RANDOM:
|
|
/*
|
|
* Default to online nodes with memory if no nodelist
|
|
*/
|
|
@@ -3375,6 +3445,7 @@ void mpol_to_str(char *buffer, int maxle
|
|
case MPOL_BIND:
|
|
case MPOL_INTERLEAVE:
|
|
case MPOL_WEIGHTED_INTERLEAVE:
|
|
+ case MPOL_RANDOM:
|
|
nodes = pol->nodes;
|
|
break;
|
|
default:
|