From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753271Ab3BCBVm (ORCPT ); Sat, 2 Feb 2013 20:21:42 -0500 Received: from mail-pb0-f46.google.com ([209.85.160.46]:59936 "EHLO mail-pb0-f46.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752506Ab3BCBVU (ORCPT ); Sat, 2 Feb 2013 20:21:20 -0500 From: Tejun Heo To: akpm@linux-foundation.org Cc: linux-kernel@vger.kernel.org, rusty@rustcorp.com.au, bfields@fieldses.org, skinsbursky@parallels.com, ebiederm@xmission.com, jmorris@namei.org, axboe@kernel.dk, Tejun Heo Subject: [PATCH 05/62] idr: implement idr_preload[_end]() and idr_alloc() Date: Sat, 2 Feb 2013 17:20:06 -0800 Message-Id: <1359854463-2538-6-git-send-email-tj@kernel.org> X-Mailer: git-send-email 1.8.1 In-Reply-To: <1359854463-2538-1-git-send-email-tj@kernel.org> References: <1359854463-2538-1-git-send-email-tj@kernel.org> Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org The current idr interface is very cumbersome. * For all allocations, two function calls - idr_pre_get() and idr_get_new*() - should be made. * idr_pre_get() doesn't guarantee that the following idr_get_new*() will not fail from memory shortage. If idr_get_new*() returns -EAGAIN, the caller is expected to retry pre_get and allocation. * idr_get_new*() can't enforce upper limit. Upper limit can only be enforced by allocating and then freeing if above limit. * idr_layer buffer is unnecessarily per-idr. Each idr ends up keeping around MAX_IDR_FREE idr_layers. The memory consumed per idr is under two pages but it makes it difficult to make idr_layer larger. This patch implements the following new set of allocation functions. * idr_preload[_end]() - Similar to radix preload but doesn't fail. The first idr_alloc() inside preload section can be treated as if it were called with @gfp_mask used for idr_preload(). * idr_alloc() - Allocate an ID w/ lower and upper limits. Takes @gfp_flags and can be used w/o preloading. When used inside preloaded section, the allocation mask of preloading can be assumed. If idr_alloc() can be called from a context which allows sufficiently relaxed @gfp_mask, it can be used by itself. If, for example, idr_alloc() is called inside spinlock protected region, preloading can be used like the following. idr_preload(GFP_KERNEL); spin_lock(lock); id = idr_alloc(idr, ptr, start, end, GFP_NOWAIT); spin_unlock(lock); idr_preload_end(); if (id < 0) error; which is much simpler and less error-prone than idr_pre_get and idr_get_new*() loop. The new interface uses per-pcu idr_layer buffer and thus the number of idr's in the system doesn't affect the amount of memory used for preloading. idr_layer_alloc() is introduced to handle idr_layer allocations for both old and new ID allocation paths. This is a bit hairy now but the new interface is expected to replace the old and the internal implementation eventually will become simpler. Signed-off-by: Tejun Heo Cc: Rusty Russell --- include/linux/idr.h | 14 +++++ lib/idr.c | 168 +++++++++++++++++++++++++++++++++++++++++++++++++--- 2 files changed, 174 insertions(+), 8 deletions(-) diff --git a/include/linux/idr.h b/include/linux/idr.h index 92e84f7..8d3ffe1 100644 --- a/include/linux/idr.h +++ b/include/linux/idr.h @@ -94,15 +94,29 @@ struct idr { void *idr_find(struct idr *idp, int id); int idr_pre_get(struct idr *idp, gfp_t gfp_mask); int idr_get_new_above(struct idr *idp, void *ptr, int starting_id, int *id); +void idr_preload(gfp_t gfp_mask); +int idr_alloc(struct idr *idp, void *ptr, int start, int end, gfp_t gfp_mask); int idr_for_each(struct idr *idp, int (*fn)(int id, void *p, void *data), void *data); void *idr_get_next(struct idr *idp, int *nextid); void *idr_replace(struct idr *idp, void *ptr, int id); void idr_remove(struct idr *idp, int id); +void idr_free(struct idr *idp, int id); void idr_destroy(struct idr *idp); void idr_init(struct idr *idp); /** + * idr_preload_end - end preload section started with idr_preload() + * + * Each idr_preload() should be matched with an invocation of this + * function. See idr_preload() for details. + */ +static inline void idr_preload_end(void) +{ + preempt_enable(); +} + +/** * idr_get_new - allocate new idr entry * @idp: idr handle * @ptr: pointer you want associated with the id diff --git a/lib/idr.c b/lib/idr.c index b13aae5..9f8b3e6 100644 --- a/lib/idr.c +++ b/lib/idr.c @@ -35,8 +35,12 @@ #include #include #include +#include +#include static struct kmem_cache *idr_layer_cache; +static DEFINE_PER_CPU(struct idr_layer *, idr_preload_head); +static DEFINE_PER_CPU(int, idr_preload_cnt); static DEFINE_SPINLOCK(simple_ida_lock); static struct idr_layer *get_from_free_list(struct idr *idp) @@ -54,6 +58,48 @@ static struct idr_layer *get_from_free_list(struct idr *idp) return(p); } +/** + * idr_layer_alloc - allocate a new idr_layer + * @gfp_mask: allocation mask + * @layer_idr: optional idr to allocate from + * + * If @layer_idr is %NULL, fetch one from the per-cpu preload buffer or + * directly allocate one using @gfp_mask. If @layer_idr is not %NULL, + * fetch an idr_layer from @idr->id_free. + * + * @layer_idr is to maintain backward compatibility with the old alloc + * interface - idr_pre_get() and idr_get_new*() - and will be removed + * together with per-pool preload buffer. + */ +static struct idr_layer *idr_layer_alloc(gfp_t gfp_mask, struct idr *layer_idr) +{ + /* this is the old path, bypass to get_from_free_list() */ + if (layer_idr) + return get_from_free_list(layer_idr); + + /* + * Try to fetch one from the per-cpu preload buffer if in process + * context. See idr_preload() for details. + */ + if (!in_interrupt()) { + struct idr_layer *new; + + preempt_disable(); + new = __this_cpu_read(idr_preload_head); + if (new) { + __this_cpu_write(idr_preload_head, new->ary[0]); + __this_cpu_dec(idr_preload_cnt); + new->ary[0] = NULL; + preempt_enable(); + return new; + } + preempt_enable(); + } + + /* try to allocate directly from kmem_cache */ + return kmem_cache_zalloc(idr_layer_cache, gfp_mask); +} + static void idr_layer_rcu_free(struct rcu_head *head) { struct idr_layer *layer; @@ -139,6 +185,8 @@ EXPORT_SYMBOL(idr_pre_get); * @starting_id: id to start search at * @id: pointer to the allocated handle * @pa: idr_layer[MAX_IDR_LEVEL] used as backtrack buffer + * @gfp_mask: allocation mask for idr_layer_alloc() + * @layer_idr: optional idr passed to idr_layer_alloc() * * Allocate an id in range [@starting_id, INT_MAX] from @idp without * growing its depth. Returns @@ -148,7 +196,8 @@ EXPORT_SYMBOL(idr_pre_get); * -ENOSPC if the id space is exhausted, * -ENOMEM if more idr_layers need to be allocated. */ -static int sub_alloc(struct idr *idp, int *starting_id, struct idr_layer **pa) +static int sub_alloc(struct idr *idp, int *starting_id, struct idr_layer **pa, + gfp_t gfp_mask, struct idr *layer_idr) { int n, m, sh; struct idr_layer *p, *new; @@ -202,7 +251,7 @@ static int sub_alloc(struct idr *idp, int *starting_id, struct idr_layer **pa) * Create the layer below if it is missing. */ if (!p->ary[m]) { - new = get_from_free_list(idp); + new = idr_layer_alloc(gfp_mask, layer_idr); if (!new) return -ENOMEM; new->layer = l-1; @@ -218,7 +267,8 @@ static int sub_alloc(struct idr *idp, int *starting_id, struct idr_layer **pa) } static int idr_get_empty_slot(struct idr *idp, int starting_id, - struct idr_layer **pa) + struct idr_layer **pa, gfp_t gfp_mask, + struct idr *layer_idr) { struct idr_layer *p, *new; int layers, v, id; @@ -229,7 +279,7 @@ build_up: p = idp->top; layers = idp->layers; if (unlikely(!p)) { - if (!(p = get_from_free_list(idp))) + if (!(p = idr_layer_alloc(gfp_mask, layer_idr))) return -ENOMEM; p->layer = 0; layers = 1; @@ -248,7 +298,7 @@ build_up: p->layer++; continue; } - if (!(new = get_from_free_list(idp))) { + if (!(new = idr_layer_alloc(gfp_mask, layer_idr))) { /* * The allocation failed. If we built part of * the structure tear it down. @@ -272,7 +322,7 @@ build_up: } rcu_assign_pointer(idp->top, p); idp->layers = layers; - v = sub_alloc(idp, &id, pa); + v = sub_alloc(idp, &id, pa, gfp_mask, layer_idr); if (v == -EAGAIN) goto build_up; return(v); @@ -312,7 +362,7 @@ int idr_get_new_above(struct idr *idp, void *ptr, int starting_id, int *id) struct idr_layer *pa[MAX_IDR_LEVEL]; int rv; - rv = idr_get_empty_slot(idp, starting_id, pa); + rv = idr_get_empty_slot(idp, starting_id, pa, 0, idp); if (rv < 0) return rv == -ENOMEM ? -EAGAIN : rv; @@ -322,6 +372,108 @@ int idr_get_new_above(struct idr *idp, void *ptr, int starting_id, int *id) } EXPORT_SYMBOL(idr_get_new_above); +/** + * idr_preload - preload for idr_alloc() + * @gfp_mask: allocation mask to use for preloading + * + * Preload per-cpu layer buffer for idr_alloc(). Can only be used from + * process context and each idr_preload() invocation should be matched with + * idr_preload_end(). The first idr_alloc() in the preloaded section can + * be treated as if it is invoked with @gfp_mask used for preloading. This + * allows applying more permissive allocation masks for idrs protected by + * spinlocks. + * + * For example, if idr_alloc() below fails, the failure can be treated as + * if idr_alloc() were called with GFP_KERNEL rather than GFP_NOWAIT. + * + * idr_preload(GFP_KERNEL); + * spin_lock(lock); + * + * id = idr_alloc(idr, ptr, start, end, GFP_NOWAIT); + * + * spin_unlock(lock); + * idr_preload_end(); + * if (id < 0) + * error; + */ +void idr_preload(gfp_t gfp_mask) +{ + /* + * Consuming preload buffer from non-process context breaks preload + * allocation guarantee. Disallow usage from those contexts. + */ + WARN_ON_ONCE(in_interrupt()); + might_sleep_if(gfp_mask & __GFP_WAIT); + + preempt_disable(); + + /* + * idr_alloc() is likely to succeed w/o full idr_layer buffer and + * return value from idr_alloc() needs to be checked for failure + * anyway. Silently give up if allocation fails. The caller can + * treat failures from idr_alloc() as if idr_alloc() were called + * with @gfp_mask which should be enough. + */ + while (__this_cpu_read(idr_preload_cnt) < MAX_IDR_FREE) { + struct idr_layer *new; + + preempt_enable(); + new = kmem_cache_zalloc(idr_layer_cache, gfp_mask); + preempt_disable(); + if (!new) + break; + + /* link the new one to per-cpu preload list */ + new->ary[0] = __this_cpu_read(idr_preload_head); + __this_cpu_write(idr_preload_head, new); + __this_cpu_inc(idr_preload_cnt); + } +} +EXPORT_SYMBOL(idr_preload); + +/** + * idr_alloc - allocate new idr entry + * @idr: the (initialized) idr + * @ptr: pointer to be associated with the new id + * @start: the minimum id (inclusive) + * @end: the maximum id (exclusive, 0 for max) + * @gfp_mask: memory allocation flags + * + * Allocate an id in [start, end) and associate it with @ptr. If no ID is + * available in the specified range, returns -ENOSPC. On memory allocation + * failure, returns -ENOMEM. + * + * The user is responsible for exclusively synchronizing all operations + * which may modify @idr. However, read-only accesses such as idr_find() + * or iteration can be performed under RCU read lock provided the user + * destroys @ptr in RCU-safe way after removal from idr. + */ +int idr_alloc(struct idr *idr, void *ptr, int start, int end, gfp_t gfp_mask) +{ + int max = end ? end - 1 : INT_MAX; /* inclusive upper limit */ + struct idr_layer *pa[MAX_IDR_LEVEL]; + int id; + + might_sleep_if(gfp_mask & __GFP_WAIT); + + /* sanity checks */ + if (WARN_ON_ONCE(start < 0 || end < 0)) + return -EINVAL; + if (unlikely(max < start)) + return -ENOSPC; + + /* allocate id */ + id = idr_get_empty_slot(idr, start, pa, gfp_mask, NULL); + if (unlikely(id < 0)) + return id; + if (unlikely(id > max)) + return -ENOSPC; + + idr_fill_slot(ptr, id, pa); + return id; +} +EXPORT_SYMBOL_GPL(idr_alloc); + static void idr_remove_warning(int id) { printk(KERN_WARNING @@ -769,7 +921,7 @@ int ida_get_new_above(struct ida *ida, int starting_id, int *p_id) restart: /* get vacant slot */ - t = idr_get_empty_slot(&ida->idr, idr_id, pa); + t = idr_get_empty_slot(&ida->idr, idr_id, pa, 0, &ida->idr); if (t < 0) return t == -ENOMEM ? -EAGAIN : t; -- 1.8.1