crypto: scomp - Allocate per-cpu buffers on first use
Herbert Xu
maintainercover crypto: scomp - Allocate per-cpu buffers on first use
Per-cpu buffers can be wasteful when the number of CPUs is large, especially if the buffer itself is likely to never be used. Reduce such wastage by only allocating them on first use of a particular CPU.
The first two patches create the gound-work so that scomp can safely flush work structs in its cra_destroy function. The last patch implements allocation on first use.
Herbert Xu (3):
crypto: api - Move alg destroy work from instance to template
crypto: api - Ensure cra_type->destroy is done in process context
crypto: scomp - Allocate per-cpu buffer on first use of each CPU
crypto/algapi.c | 48 +++++-- crypto/api.c | 10 -- crypto/internal.h | 3 +- crypto/scompress.c | 203 ++++++++++++++++++++++------ include/crypto/acompress.h | 6 - include/crypto/algapi.h | 5 +- include/crypto/internal/acompress.h | 1 - include/crypto/internal/scompress.h | 13 ++ 8 files changed, 214 insertions(+), 75 deletions(-)
Signature · 1 line
2.39.5
Herbert Xu
maintainerPATCH 1/3 crypto: api - Move alg destroy work from instance to template
Commit 9ae4577bc077 ("crypto: api - Use work queue in crypto_destroy_instance") introduced a work struct to free an instance after the last user goes away.
Move the delayed work from the instance into its template so that when the template is unregistered it can ensure that all its instances have been freed before returning.
--- crypto/algapi.c | 35 +++++++++++++++++++++++++++-------- include/crypto/algapi.h | 5 +++-- 2 files changed, 30 insertions(+), 10 deletions(-)
crypto/algapi.c +27 -8
@@ -71,12 +71,23 @@ static void crypto_free_instance(struct crypto_instance *inst)static void crypto_destroy_instance_workfn (struct work_struct *w) { - struct crypto_instance *inst = container_of(w, struct crypto_instance, + struct crypto_template *tmpl = container_of(w, struct crypto_template, free_work); -struct crypto_template *tmpl = inst->tmpl; +struct crypto_instance *inst; +struct hlist_node *n; +HLIST_HEAD (list); -crypto_free_instance (inst); -crypto_tmpl_put (tmpl); +down_write (&crypto_alg_sem); +hlist_for_each_entry_safe (inst, n, &tmpl->dead, list) { +if (refcount_read (&inst->alg.cra_refcnt) !=-1 ) +continue ; +hlist_del (&inst->list); +hlist_add_head (&inst->list, &list); + } +up_write (&crypto_alg_sem); + +hlist_for_each_entry_safe (inst, n, &list, list) +crypto_free_instance (inst); }static void crypto_destroy_instance (struct crypto_alg *alg)
@@ -84,9 +95,10 @@ static void crypto_destroy_instance(struct crypto_alg *alg)struct crypto_instance *inst =container_of (alg, struct crypto_instance, alg); +struct crypto_template *tmpl = inst->tmpl; -INIT_WORK (&inst->free_work, crypto_destroy_instance_workfn); -schedule_work (&inst->free_work); +refcount_set (&alg->cra_refcnt,-1 ); +schedule_work (&tmpl->free_work); }/*
@@ -132,14 +144,17 @@ static void crypto_remove_instance(struct crypto_instance *inst, inst->alg.cra_flags |=CRYPTO_ALG_DEAD ; -if (!tmpl || !crypto_tmpl_get (tmpl)) +if (!tmpl)return ; -list_move (&inst->alg.cra_list, list); +list_del_init (&inst->alg.cra_list);hlist_del (&inst->list); +hlist_add_head (&inst->list, &tmpl->dead); inst->alg.cra_destroy = crypto_destroy_instance;BUG_ON (!list_empty (&inst->alg.cra_users)); + +crypto_alg_put (&inst->alg); }/*
@@ -504,6 +519,8 @@ int crypto_register_template(struct crypto_template *tmpl)struct crypto_template *q;int err = -EEXIST ; +INIT_WORK (&tmpl->free_work, crypto_destroy_instance_workfn); +down_write (&crypto_alg_sem);crypto_check_module_sig (tmpl->module);
@@ -565,6 +582,8 @@ void crypto_unregister_template(struct crypto_template *tmpl)crypto_free_instance (inst); }crypto_remove_final (&users); + +flush_work (&tmpl->free_work); }EXPORT_SYMBOL_GPL (crypto_unregister_template);
include/crypto/algapi.h +3 -2
@@ -68,16 +68,17 @@ struct crypto_instance {struct crypto_spawn *spawns; }; -struct work_struct free_work; - void *__ctx[] CRYPTO_MINALIGN_ATTR; };struct crypto_template {struct list_head list;struct hlist_head instances; +struct hlist_head dead;struct module *module; +struct work_struct free_work; +int (*create)(struct crypto_template *tmpl,struct rtattr **tb);char name[CRYPTO_MAX_ALG_NAME ];
Signature · 1 line
2.39.5
Herbert Xu
maintainerPATCH 2/3 crypto: api - Ensure cra_type->destroy is done in process context
In reply to Herbert XuMove the cra_type->destroy call out of crypto_alg_put and into crypto_unregister_alg and crypto_free_instance. This ensures that it's always done in process context so calls such as flush_work can be done.
--- crypto/algapi.c | 13 ++++++++++--- crypto/api.c | 10 ---------- crypto/internal.h | 3 +-- 3 files changed, 11 insertions(+), 15 deletions(-)
crypto/algapi.c +10 -3
@@ -66,7 +66,13 @@ static int crypto_check_alg(struct crypto_alg *alg)static void crypto_free_instance (struct crypto_instance *inst) { - inst->alg.cra_type->free (inst); +struct crypto_alg *alg = &inst->alg; +const struct crypto_type *type; + + type = alg->cra_type; +if (type->destroy) + type->destroy (alg); + type->free (inst); }static void crypto_destroy_instance_workfn (struct work_struct *w)
@@ -150,7 +156,6 @@ static void crypto_remove_instance(struct crypto_instance *inst,list_del_init (&inst->alg.cra_list);hlist_del (&inst->list);hlist_add_head (&inst->list, &tmpl->dead); - inst->alg.cra_destroy = crypto_destroy_instance;BUG_ON (!list_empty (&inst->alg.cra_users));
@@ -479,7 +484,8 @@ void crypto_unregister_alg(struct crypto_alg *alg)if (WARN_ON (refcount_read (&alg->cra_refcnt) !=1 ))return ; -crypto_alg_put (alg); +if (alg->cra_type && alg->cra_type->destroy) + alg->cra_type->destroy (alg);crypto_remove_final (&list); }
@@ -637,6 +643,7 @@ int crypto_register_instance(struct crypto_template *tmpl, inst->alg.cra_module = tmpl->module; inst->alg.cra_flags |=CRYPTO_ALG_INSTANCE ; + inst->alg.cra_destroy = crypto_destroy_instance;down_write (&crypto_alg_sem);
crypto/api.c +0 -10
@@ -703,15 +703,5 @@ void crypto_req_done(void *data, int err) }EXPORT_SYMBOL_GPL (crypto_req_done); -void crypto_destroy_alg (struct crypto_alg *alg) -{ -if (alg->cra_type && alg->cra_type->destroy) - alg->cra_type->destroy (alg); - -if (alg->cra_destroy) - alg->cra_destroy (alg); -} -EXPORT_SYMBOL_GPL (crypto_destroy_alg); -MODULE_DESCRIPTION ("Cryptographic core API" );MODULE_LICENSE ("GPL" );
crypto/internal.h +1 -2
@@ -128,7 +128,6 @@ void *crypto_create_tfm_node(struct crypto_alg *alg, const struct crypto_type *frontend, int node);void *crypto_clone_tfm (const struct crypto_type *frontend,struct crypto_tfm *otfm); -void crypto_destroy_alg (struct crypto_alg *alg);static inline void *crypto_create_tfm (struct crypto_alg *alg,const struct crypto_type *frontend)
@@ -166,7 +165,7 @@ static inline struct crypto_alg *crypto_alg_get(struct crypto_alg *alg)static inline void crypto_alg_put (struct crypto_alg *alg) {if (refcount_dec_and_test (&alg->cra_refcnt)) -crypto_destroy_alg (alg); + alg->cra_destroy (alg); }static inline int crypto_tmpl_get (struct crypto_template *tmpl)
Signature · 1 line
2.39.5
Herbert Xu
maintainerPATCH 3/3 crypto: scomp - Allocate per-cpu buffer on first use of each CPU
In reply to Herbert XuPer-cpu buffers can be wasteful when the number of CPUs is large, especially if the buffer itself is likely to never be used. Reduce such wastage by only allocating them on first use of a particular CPU.
On start-up allocate a single buffer on the first possible CPU. For every other CPU a work struct will be scheduled on first use to allocate the buffer for that CPU. Until the allocation succeeds simply use the first CPU's buffer which is protected under a spin lock.
--- crypto/scompress.c | 203 ++++++++++++++++++++++------ include/crypto/acompress.h | 6 - include/crypto/internal/acompress.h | 1 - include/crypto/internal/scompress.h | 13 ++ 4 files changed, 173 insertions(+), 50 deletions(-)
crypto/scompress.c +160 -43
@@ -10,6 +10,7 @@#include <crypto/internal/acompress.h> #include <crypto/internal/scompress.h> #include <crypto/scatterwalk.h> +#include <linux/cpumask.h> #include <linux/cryptouser.h> #include <linux/err.h> #include <linux/highmem.h>
@@ -20,7 +21,7 @@#include <linux/seq_file.h> #include <linux/slab.h> #include <linux/string.h> -#include <linux/vmalloc.h> +#include <linux/workqueue.h> #include <net/netlink.h> #include "compress.h"
@@ -44,6 +45,10 @@ static const struct crypto_type crypto_scomp_type;static int scomp_scratch_users;static DEFINE_MUTEX (scomp_lock); +static cpumask_t scomp_scratch_want; +static void scomp_scratch_workfn (struct work_struct *work); +static DECLARE_WORK(scomp_scratch_work, scomp_scratch_workfn); +static int __maybe_unused crypto_scomp_report (struct sk_buff *skb,struct crypto_alg *alg) {
@@ -74,36 +79,57 @@ static void crypto_scomp_free_scratches(void) scratch =per_cpu_ptr (&scomp_scratch, i);free_page (scratch->saddr); -vfree (scratch->dst); +kvfree (scratch->dst); scratch->src =NULL ; scratch->dst =NULL ; } } +static int scomp_alloc_scratch (struct scomp_scratch *scratch,int cpu) +{ +int node =cpu_to_node (cpu); +struct page *page; +void *mem; + + mem =kvmalloc_node (SCOMP_SCRATCH_SIZE ,GFP_KERNEL , node); +if (!mem) +return -ENOMEM ; + page =alloc_pages_node (node,GFP_KERNEL ,0 ); +if (!page) { +kvfree (mem); +return -ENOMEM ; + } +spin_lock_bh (&scratch->lock); + scratch->src =page_address (page); + scratch->dst = mem; +spin_unlock_bh (&scratch->lock); +return 0 ; +} + +static void scomp_scratch_workfn (struct work_struct *work) +{ +int cpu; + +for_each_cpu (cpu, &scomp_scratch_want) { +struct scomp_scratch *scratch; + + scratch =per_cpu_ptr (&scomp_scratch, cpu); +if (scratch->src) +continue ; +if (scomp_alloc_scratch (scratch, cpu)) +break ; + +cpumask_clear_cpu (cpu, &scomp_scratch_want); + } +} +static int crypto_scomp_alloc_scratches (void ) { +unsigned int i =cpumask_first (cpu_possible_mask);struct scomp_scratch *scratch; -int i; -for_each_possible_cpu (i) { -struct page *page; -void *mem; - - scratch =per_cpu_ptr (&scomp_scratch, i); - - page =alloc_pages_node (cpu_to_node (i),GFP_KERNEL ,0 ); -if (!page) - goto error; - scratch->src =page_address (page); - mem =vmalloc_node (SCOMP_SCRATCH_SIZE ,cpu_to_node (i)); -if (!mem) - goto error; - scratch->dst = mem; - } -return 0 ; -error: -crypto_scomp_free_scratches (); -return -ENOMEM ; + scratch =per_cpu_ptr (&scomp_scratch, i); +return scomp_alloc_scratch (scratch, i); }static void scomp_free_streams (struct scomp_alg *alg)
@@ -115,7 +141,7 @@ static void scomp_free_streams(struct scomp_alg *alg)struct crypto_acomp_stream *ps =per_cpu_ptr (stream, i);if (!ps->ctx) -break ; +continue ; alg->free_ctx (ps->ctx); }
@@ -126,21 +152,26 @@ static void scomp_free_streams(struct scomp_alg *alg)static int scomp_alloc_streams (struct scomp_alg *alg) {struct crypto_acomp_stream __percpu *stream; -int i; +struct crypto_acomp_stream *ps; +unsigned int i; +void *ctx; stream = alloc_percpu(struct crypto_acomp_stream);if (!stream)return -ENOMEM ; + ctx = alg->alloc_ctx (); +if (IS_ERR (ctx)) { +free_percpu (stream); +return PTR_ERR (ctx); + } + + i =cpumask_first (cpu_possible_mask); + ps =per_cpu_ptr (stream, i); + ps->ctx = ctx; +for_each_possible_cpu (i) { -struct crypto_acomp_stream *ps =per_cpu_ptr (stream, i); - - ps->ctx = alg->alloc_ctx (); -if (IS_ERR (ps->ctx)) { -scomp_free_streams (alg); -return PTR_ERR (ps->ctx); - } - + ps =per_cpu_ptr (stream, i);spin_lock_init (&ps->lock); }
@@ -148,6 +179,33 @@ static int scomp_alloc_streams(struct scomp_alg *alg)return 0 ; } +static void scomp_stream_workfn (struct work_struct *work) +{ + struct scomp_alg *alg = container_of(work, struct scomp_alg, + stream_work); +struct crypto_acomp_stream __percpu *stream = alg->stream; +int cpu; + +for_each_cpu (cpu, &alg->stream_want) { +struct crypto_acomp_stream *ps; +void *ctx; + + ps =per_cpu_ptr (stream, cpu); +if (ps->ctx) +continue ; + + ctx = alg->alloc_ctx (); +if (IS_ERR (ctx)) +break ; + +spin_lock_bh (&ps->lock); + ps->ctx = ctx; +spin_unlock_bh (&ps->lock); + +cpumask_clear_cpu (cpu, &alg->stream_want); + } +} +static int crypto_scomp_init_tfm (struct crypto_tfm *tfm) {struct scomp_alg *alg =crypto_scomp_alg (__crypto_scomp_tfm (tfm));
@@ -171,13 +229,67 @@ static int crypto_scomp_init_tfm(struct crypto_tfm *tfm)return ret; } +static struct scomp_scratch *scomp_lock_scratch_bh (void )__acquires (scratch) +{ +int cpu =raw_smp_processor_id (); +struct scomp_scratch *scratch; + + scratch =per_cpu_ptr (&scomp_scratch, cpu); +spin_lock_bh (&scratch->lock); +if (likely (scratch->src)) +return scratch; +spin_unlock (&scratch->lock); + +cpumask_set_cpu (cpu, &scomp_scratch_want); +schedule_work (&scomp_scratch_work); + + scratch =per_cpu_ptr (&scomp_scratch,cpumask_first (cpu_possible_mask)); +spin_lock (&scratch->lock); +return scratch; +} + +static inline void scomp_unlock_scratch_bh (struct scomp_scratch *scratch) +__releases (scratch) +{ +spin_unlock_bh (&scratch->lock); +} + +static struct crypto_acomp_stream *scomp_lock_stream (struct crypto_scomp *tfm) +__acquires (stream) +{ +struct scomp_alg *alg =crypto_scomp_alg (tfm); +struct crypto_acomp_stream __percpu *stream; +int cpu =raw_smp_processor_id (); +struct crypto_acomp_stream *ps; + + stream = alg->stream; + ps =per_cpu_ptr (stream, cpu); +spin_lock (&ps->lock); +if (likely (ps->ctx)) +return ps; +spin_unlock (&ps->lock); + +cpumask_set_cpu (cpu, &alg->stream_want); +schedule_work (&alg->stream_work); + + ps =per_cpu_ptr (stream,cpumask_first (cpu_possible_mask)); +spin_lock (&ps->lock); +return ps; +} + +static inline void scomp_unlock_stream (struct crypto_acomp_stream *stream) +__releases (stream) +{ +spin_unlock (&stream->lock); +} +static int scomp_acomp_comp_decomp (struct acomp_req *req,int dir) { -struct scomp_scratch *scratch =raw_cpu_ptr (&scomp_scratch);struct crypto_acomp *tfm =crypto_acomp_reqtfm (req);struct crypto_scomp **tfm_ctx =acomp_tfm_ctx (tfm);struct crypto_scomp *scomp = *tfm_ctx;struct crypto_acomp_stream *stream; +struct scomp_scratch *scratch;unsigned int slen = req->slen;unsigned int dlen = req->dlen;struct page *spage, *dpage;
@@ -194,6 +306,8 @@ static int scomp_acomp_comp_decomp(struct acomp_req *req, int dir)if (!req->dst || !dlen)return -EINVAL ; + scratch =scomp_lock_scratch_bh (); +if (acomp_request_src_isvirt (req)) src = req->svirt;else {
@@ -218,6 +332,9 @@ static int scomp_acomp_comp_decomp(struct acomp_req *req, int dir)break ; src =kmap_local_page (spage) + soff; }while (0 ); + +if (src == scratch->src) +memcpy_from_sglist (scratch->src, req->src,0 , slen); }if (acomp_request_dst_isvirt (req))
@@ -250,13 +367,7 @@ static int scomp_acomp_comp_decomp(struct acomp_req *req, int dir) dlen =min (dlen, max); } -spin_lock_bh (&scratch->lock); - -if (src == scratch->src) -memcpy_from_sglist (scratch->src, req->src,0 , slen); - - stream =raw_cpu_ptr (crypto_scomp_alg (scomp)->stream); -spin_lock (&stream->lock); + stream =scomp_lock_stream (scomp);if (dir) ret =crypto_scomp_compress (scomp, src, slen, dst, &dlen, stream->ctx);
@@ -267,8 +378,8 @@ static int scomp_acomp_comp_decomp(struct acomp_req *req, int dir)if (dst == scratch->dst)memcpy_to_sglist (req->dst,0 , dst, dlen); -spin_unlock (&stream->lock); -spin_unlock_bh (&scratch->lock); +scomp_unlock_stream (stream); +scomp_unlock_scratch_bh (scratch); req->dlen = dlen;
@@ -319,6 +430,7 @@ static void crypto_exit_scomp_ops_async(struct crypto_tfm *tfm)crypto_free_scomp (*ctx); +flush_work (&scomp_scratch_work);mutex_lock (&scomp_lock);if (!--scomp_scratch_users)crypto_scomp_free_scratches ();
@@ -352,7 +464,10 @@ int crypto_init_scomp_ops_async(struct crypto_tfm *tfm)static void crypto_scomp_destroy (struct crypto_alg *alg) { -scomp_free_streams (__crypto_scomp_alg (alg)); +struct scomp_alg *scomp =__crypto_scomp_alg (alg); + +cancel_work_sync (&scomp->stream_work); +scomp_free_streams (scomp); }static const struct crypto_type crypto_scomp_type = {
@@ -378,6 +493,8 @@ static void scomp_prepare_alg(struct scomp_alg *alg)comp_prepare_alg (&alg->calg); base->cra_flags |=CRYPTO_ALG_REQ_CHAIN ; + +INIT_WORK (&alg->stream_work, scomp_stream_workfn); }int crypto_register_scomp (struct scomp_alg *alg)
include/crypto/acompress.h +0 -6
@@ -130,14 +130,8 @@ struct crypto_acomp {struct crypto_tfm base; }; -struct crypto_acomp_stream { -spinlock_t lock; -void *ctx; -}; -#define COMP_ALG_COMMON { \ struct crypto_alg base; \ - struct crypto_acomp_stream __percpu *stream; \ }struct comp_alg_common COMP_ALG_COMMON ;
include/crypto/internal/acompress.h +0 -1
@@ -37,7 +37,6 @@* * @reqsize: Context size for (de)compression requests * @base: Common crypto API algorithm data structure -* @stream: Per-cpu memory for algorithm * @calg: Cmonn algorithm data structure shared with scomp */ struct acomp_alg {
include/crypto/internal/scompress.h +13 -0
@@ -11,6 +11,8 @@#include <crypto/acompress.h> #include <crypto/algapi.h> +#include <linux/cpumask_types.h> +#include <linux/workqueue_types.h> struct acomp_req ;
@@ -18,6 +20,11 @@ struct crypto_scomp {struct crypto_tfm base; }; +struct crypto_acomp_stream { +spinlock_t lock; +void *ctx; +}; +/** * struct scomp_alg - synchronous compression algorithm *
@@ -27,6 +34,8 @@ struct crypto_scomp {* @decompress: Function performs a de-compress operation * @base: Common crypto API algorithm data structure * @stream: Per-cpu memory for algorithm +* @stream_work: Work struct to allocate stream memmory +* @stream_want: CPU mask for allocating stream memory * @calg: Cmonn algorithm data structure shared with acomp */ struct scomp_alg {
@@ -39,6 +48,10 @@ struct scomp_alg { unsigned int slen, u8 *dst, unsigned int *dlen, void *ctx); +struct crypto_acomp_stream __percpu *stream; +struct work_struct stream_work; +cpumask_t stream_want; +union {struct COMP_ALG_COMMON ;struct comp_alg_common calg;
Signature · 1 line
2.39.5