*/
/*
* Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, 2013, Intel Corporation.
+ *
+ * Copyright (c) 2012, 2015, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#ifndef HAVE_LIBCFS_CPT
-#ifndef __KERNEL__
-typedef unsigned long cpumask_t;
-typedef unsigned long nodemask_t;
-#endif
-
struct cfs_cpt_table {
/* # of CPU partitions */
int ctb_nparts;
*/
int cfs_cpu_ht_nsiblings(int cpu);
+/*
+ * allocate per-cpu-partition data, returned value is an array of pointers,
+ * variable can be indexed by CPU ID.
+ * cptab != NULL: size of array is number of CPU partitions
+ * cptab == NULL: size of array is number of HW cores
+ */
+void *cfs_percpt_alloc(struct cfs_cpt_table *cptab, unsigned int size);
+/*
+ * destory per-cpu-partition variable
+ */
+void cfs_percpt_free(void *vars);
+int cfs_percpt_number(void *vars);
+
+#define cfs_percpt_for_each(var, i, vars) \
+ for (i = 0; i < cfs_percpt_number(vars) && \
+ ((var) = (vars)[i]) != NULL; i++)
+
+/*
+ * percpu partition lock
+ *
+ * There are some use-cases like this in Lustre:
+ * . each CPU partition has it's own private data which is frequently changed,
+ * and mostly by the local CPU partition.
+ * . all CPU partitions share some global data, these data are rarely changed.
+ *
+ * LNet is typical example.
+ * CPU partition lock is designed for this kind of use-cases:
+ * . each CPU partition has it's own private lock
+ * . change on private data just needs to take the private lock
+ * . read on shared data just needs to take _any_ of private locks
+ * . change on shared data needs to take _all_ private locks,
+ * which is slow and should be really rare.
+ */
+enum {
+ CFS_PERCPT_LOCK_EX = -1, /* negative */
+};
+
+struct cfs_percpt_lock {
+ /* cpu-partition-table for this lock */
+ struct cfs_cpt_table *pcl_cptab;
+ /* exclusively locked */
+ unsigned int pcl_locked;
+ /* private lock table */
+ spinlock_t **pcl_locks;
+};
+
+/* return number of private locks */
+#define cfs_percpt_lock_num(pcl) cfs_cpt_number(pcl->pcl_cptab)
+
+/*
+ * create a cpu-partition lock based on CPU partition table \a cptab,
+ * each private lock has extra \a psize bytes padding data
+ */
+struct cfs_percpt_lock *cfs_percpt_lock_create(struct cfs_cpt_table *cptab,
+ struct lock_class_key *keys);
+/* destroy a cpu-partition lock */
+void cfs_percpt_lock_free(struct cfs_percpt_lock *pcl);
+
+/* lock private lock \a index of \a pcl */
+void cfs_percpt_lock(struct cfs_percpt_lock *pcl, int index);
+/* unlock private lock \a index of \a pcl */
+void cfs_percpt_unlock(struct cfs_percpt_lock *pcl, int index);
+
+#define CFS_PERCPT_LOCK_KEYS 256
+
+/* NB: don't allocate keys dynamically, lockdep needs them to be in ".data" */
+#define cfs_percpt_lock_alloc(cptab) \
+({ \
+ static struct lock_class_key ___keys[CFS_PERCPT_LOCK_KEYS]; \
+ struct cfs_percpt_lock *___lk; \
+ \
+ if (cfs_cpt_number(cptab) > CFS_PERCPT_LOCK_KEYS) \
+ ___lk = cfs_percpt_lock_create(cptab, NULL); \
+ else \
+ ___lk = cfs_percpt_lock_create(cptab, ___keys); \
+ ___lk; \
+})
+
+/**
+ * allocate \a nr_bytes of physical memory from a contiguous region with the
+ * properties of \a flags which are bound to the partition id \a cpt. This
+ * function should only be used for the case when only a few pages of memory
+ * are need.
+ */
+static inline void *
+cfs_cpt_malloc(struct cfs_cpt_table *cptab, int cpt, size_t nr_bytes,
+ gfp_t flags)
+{
+ return kmalloc_node(nr_bytes, flags,
+ cfs_cpt_spread_node(cptab, cpt));
+}
+
+/**
+ * allocate \a nr_bytes of virtually contiguous memory that is bound to the
+ * partition id \a cpt.
+ */
+static inline void *
+cfs_cpt_vzalloc(struct cfs_cpt_table *cptab, int cpt, size_t nr_bytes)
+{
+ /* vzalloc_node() sets __GFP_FS by default but no current Kernel
+ * exported entry-point allows for both a NUMA node specification
+ * and a custom allocation flags mask. This may be an issue since
+ * __GFP_FS usage can cause some deadlock situations in our code,
+ * like when memory reclaim started, within the same context of a
+ * thread doing FS operations, that can also attempt conflicting FS
+ * operations, ...
+ */
+ return vzalloc_node(nr_bytes, cfs_cpt_spread_node(cptab, cpt));
+}
+
+/**
+ * allocate a single page of memory with the properties of \a flags were
+ * that page is bound to the partition id \a cpt.
+ */
+static inline struct page *
+cfs_page_cpt_alloc(struct cfs_cpt_table *cptab, int cpt, gfp_t flags)
+{
+ return alloc_pages_node(cfs_cpt_spread_node(cptab, cpt), flags, 0);
+}
+
+/**
+ * allocate a chunck of memory from a memory pool that is bound to the
+ * partition id \a cpt with the properites of \a flags.
+ */
+static inline void *
+cfs_mem_cache_cpt_alloc(struct kmem_cache *cachep, struct cfs_cpt_table *cptab,
+ int cpt, gfp_t flags)
+{
+ return kmem_cache_alloc_node(cachep, flags,
+ cfs_cpt_spread_node(cptab, cpt));
+}
+
/**
* iterate over all CPU partitions in \a cptab
*/
git://git.whamcloud.com / fs / lustre-release.git / blobdiff