uma: implement provisional api for per-cpu zones

Per-cpu zone allocations are very rarely done compared to regular zones.
The intent is to avoid pessimizing the latter case with per-cpu specific
code.

In particular contrary to the claim in r334824, M_ZERO is sometimes being
used for such zones. But the zeroing method is completely different and
braching on it in the fast path for regular zones is a waste of time.
This commit is contained in:
Mateusz Guzik 2018-06-08 21:40:03 +00:00
parent 4f63fbc955
commit 4e180881ae
Notes: svn2git 2020-12-20 02:59:44 +00:00
svn path=/head/; revision=334858
4 changed files with 46 additions and 4 deletions

View File

@ -2957,7 +2957,7 @@ key_newsav(const struct sadb_msghdr *mhp, struct secasindex *saidx,
goto done;
}
mtx_init(sav->lock, "ipsec association", NULL, MTX_DEF);
sav->lft_c = uma_zalloc(V_key_lft_zone, M_NOWAIT);
sav->lft_c = uma_zalloc_pcpu(V_key_lft_zone, M_NOWAIT);
if (sav->lft_c == NULL) {
*errp = ENOBUFS;
goto done;
@ -3049,7 +3049,7 @@ key_newsav(const struct sadb_msghdr *mhp, struct secasindex *saidx,
free(sav->lock, M_IPSEC_MISC);
}
if (sav->lft_c != NULL)
uma_zfree(V_key_lft_zone, sav->lft_c);
uma_zfree_pcpu(V_key_lft_zone, sav->lft_c);
free(sav, M_IPSEC_SA), sav = NULL;
}
if (sah != NULL)

View File

@ -208,7 +208,7 @@ ipfw_alloc_rule(struct ip_fw_chain *chain, size_t rulesize)
struct ip_fw *rule;
rule = malloc(rulesize, M_IPFW, M_WAITOK | M_ZERO);
rule->cntr = uma_zalloc(V_ipfw_cntr_zone, M_WAITOK | M_ZERO);
rule->cntr = uma_zalloc_pcpu(V_ipfw_cntr_zone, M_WAITOK | M_ZERO);
return (rule);
}
@ -217,7 +217,7 @@ static void
free_rule(struct ip_fw *rule)
{
uma_zfree(V_ipfw_cntr_zone, rule->cntr);
uma_zfree_pcpu(V_ipfw_cntr_zone, rule->cntr);
free(rule, M_IPFW);
}

View File

@ -333,6 +333,7 @@ void uma_zdestroy(uma_zone_t zone);
*/
void *uma_zalloc_arg(uma_zone_t zone, void *arg, int flags);
void *uma_zalloc_pcpu_arg(uma_zone_t zone, void *arg, int flags);
/*
* Allocate an item from a specific NUMA domain. This uses a slow path in
@ -354,6 +355,7 @@ void *uma_zalloc_domain(uma_zone_t zone, void *arg, int domain, int flags);
*
*/
static __inline void *uma_zalloc(uma_zone_t zone, int flags);
static __inline void *uma_zalloc_pcpu(uma_zone_t zone, int flags);
static __inline void *
uma_zalloc(uma_zone_t zone, int flags)
@ -361,6 +363,12 @@ uma_zalloc(uma_zone_t zone, int flags)
return uma_zalloc_arg(zone, NULL, flags);
}
static __inline void *
uma_zalloc_pcpu(uma_zone_t zone, int flags)
{
return uma_zalloc_pcpu_arg(zone, NULL, flags);
}
/*
* Frees an item back into the specified zone.
*
@ -374,6 +382,7 @@ uma_zalloc(uma_zone_t zone, int flags)
*/
void uma_zfree_arg(uma_zone_t zone, void *item, void *arg);
void uma_zfree_pcpu_arg(uma_zone_t zone, void *item, void *arg);
/*
* Frees an item back to the specified zone's domain specific pool.
@ -392,6 +401,7 @@ void uma_zfree_domain(uma_zone_t zone, void *item, void *arg);
*
*/
static __inline void uma_zfree(uma_zone_t zone, void *item);
static __inline void uma_zfree_pcpu(uma_zone_t zone, void *item);
static __inline void
uma_zfree(uma_zone_t zone, void *item)
@ -399,6 +409,12 @@ uma_zfree(uma_zone_t zone, void *item)
uma_zfree_arg(zone, item, NULL);
}
static __inline void
uma_zfree_pcpu(uma_zone_t zone, void *item)
{
uma_zfree_pcpu_arg(zone, item, NULL);
}
/*
* Wait until the specified zone can allocate an item.
*/

View File

@ -2230,6 +2230,32 @@ uma_zwait(uma_zone_t zone)
uma_zfree(zone, item);
}
void *
uma_zalloc_pcpu_arg(uma_zone_t zone, void *udata, int flags)
{
void *item;
int i;
MPASS(zone->uz_flags & UMA_ZONE_PCPU);
item = uma_zalloc_arg(zone, udata, flags &~ M_ZERO);
if (item != NULL && (flags & M_ZERO)) {
CPU_FOREACH(i)
bzero(zpcpu_get_cpu(item, i), zone->uz_size);
}
return (item);
}
/*
* A stub while both regular and pcpu cases are identical.
*/
void
uma_zfree_pcpu_arg(uma_zone_t zone, void *item, void *udata)
{
MPASS(zone->uz_flags & UMA_ZONE_PCPU);
uma_zfree_arg(zone, item, udata);
}
/* See uma.h */
void *
uma_zalloc_arg(uma_zone_t zone, void *udata, int flags)