diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c index 3220b488dd1e..f100bd958b88 100644 --- a/drivers/net/bonding/bond_main.c +++ b/drivers/net/bonding/bond_main.c @@ -79,7 +79,6 @@ #include #include #include -#include #include "bonding.h" #include "bond_3ad.h" #include "bond_alb.h" @@ -3596,8 +3595,9 @@ static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int sl */ static u32 bond_rr_gen_slave_id(struct bonding *bond) { - int packets_per_slave = bond->params.packets_per_slave; u32 slave_id; + struct reciprocal_value reciprocal_packets_per_slave; + int packets_per_slave = bond->params.packets_per_slave; switch (packets_per_slave) { case 0: @@ -3607,8 +3607,10 @@ static u32 bond_rr_gen_slave_id(struct bonding *bond) slave_id = bond->rr_tx_counter; break; default: + reciprocal_packets_per_slave = + bond->params.reciprocal_packets_per_slave; slave_id = reciprocal_divide(bond->rr_tx_counter, - packets_per_slave); + reciprocal_packets_per_slave); break; } bond->rr_tx_counter++; @@ -4343,10 +4345,18 @@ static int bond_check_params(struct bond_params *params) params->resend_igmp = resend_igmp; params->min_links = min_links; params->lp_interval = lp_interval; - if (packets_per_slave > 1) - params->packets_per_slave = reciprocal_value(packets_per_slave); - else - params->packets_per_slave = packets_per_slave; + params->packets_per_slave = packets_per_slave; + if (packets_per_slave > 0) { + params->reciprocal_packets_per_slave = + reciprocal_value(packets_per_slave); + } else { + /* reciprocal_packets_per_slave is unused if + * packets_per_slave is 0 or 1, just initialize it + */ + params->reciprocal_packets_per_slave = + (struct reciprocal_value) { 0 }; + } + if (primary) { strncpy(params->primary, primary, IFNAMSIZ); params->primary[IFNAMSIZ - 1] = 0; diff --git a/drivers/net/bonding/bond_netlink.c b/drivers/net/bonding/bond_netlink.c index 21c648854a8c..e8526552790c 100644 --- a/drivers/net/bonding/bond_netlink.c +++ b/drivers/net/bonding/bond_netlink.c @@ -19,7 +19,6 @@ #include #include #include -#include #include "bonding.h" int bond_get_slave(struct net_device *slave_dev, struct sk_buff *skb) @@ -452,9 +451,6 @@ static int bond_fill_info(struct sk_buff *skb, goto nla_put_failure; packets_per_slave = bond->params.packets_per_slave; - if (packets_per_slave > 1) - packets_per_slave = reciprocal_value(packets_per_slave); - if (nla_put_u32(skb, IFLA_BOND_PACKETS_PER_SLAVE, packets_per_slave)) goto nla_put_failure; diff --git a/drivers/net/bonding/bond_options.c b/drivers/net/bonding/bond_options.c index 945a6668da83..85e434886f2e 100644 --- a/drivers/net/bonding/bond_options.c +++ b/drivers/net/bonding/bond_options.c @@ -16,7 +16,6 @@ #include #include #include -#include #include "bonding.h" int bond_option_mode_set(struct bonding *bond, int mode) @@ -671,11 +670,17 @@ int bond_option_packets_per_slave_set(struct bonding *bond, pr_warn("%s: Warning: packets_per_slave has effect only in balance-rr mode\n", bond->dev->name); - if (packets_per_slave > 1) - bond->params.packets_per_slave = + bond->params.packets_per_slave = packets_per_slave; + if (packets_per_slave > 0) { + bond->params.reciprocal_packets_per_slave = reciprocal_value(packets_per_slave); - else - bond->params.packets_per_slave = packets_per_slave; + } else { + /* reciprocal_packets_per_slave is unused if + * packets_per_slave is 0 or 1, just initialize it + */ + bond->params.reciprocal_packets_per_slave = + (struct reciprocal_value) { 0 }; + } return 0; } diff --git a/drivers/net/bonding/bond_sysfs.c b/drivers/net/bonding/bond_sysfs.c index 011f163c2c67..c083e9a66ece 100644 --- a/drivers/net/bonding/bond_sysfs.c +++ b/drivers/net/bonding/bond_sysfs.c @@ -39,7 +39,6 @@ #include #include #include -#include #include "bonding.h" @@ -1374,10 +1373,6 @@ static ssize_t bonding_show_packets_per_slave(struct device *d, { struct bonding *bond = to_bond(d); unsigned int packets_per_slave = bond->params.packets_per_slave; - - if (packets_per_slave > 1) - packets_per_slave = reciprocal_value(packets_per_slave); - return sprintf(buf, "%u\n", packets_per_slave); } diff --git a/drivers/net/bonding/bonding.h b/drivers/net/bonding/bonding.h index 8a935f8f2b3c..0a616c41dc94 100644 --- a/drivers/net/bonding/bonding.h +++ b/drivers/net/bonding/bonding.h @@ -23,6 +23,8 @@ #include #include #include +#include + #include "bond_3ad.h" #include "bond_alb.h" @@ -171,6 +173,7 @@ struct bond_params { int resend_igmp; int lp_interval; int packets_per_slave; + struct reciprocal_value reciprocal_packets_per_slave; }; struct bond_parm_tbl { diff --git a/include/linux/flex_array.h b/include/linux/flex_array.h index 6843cf193a44..b6efb0c64408 100644 --- a/include/linux/flex_array.h +++ b/include/linux/flex_array.h @@ -2,6 +2,7 @@ #define _FLEX_ARRAY_H #include +#include #include #define FLEX_ARRAY_PART_SIZE PAGE_SIZE @@ -22,7 +23,7 @@ struct flex_array { int element_size; int total_nr_elements; int elems_per_part; - u32 reciprocal_elems; + struct reciprocal_value reciprocal_elems; struct flex_array_part *parts[]; }; /* diff --git a/include/linux/reciprocal_div.h b/include/linux/reciprocal_div.h index f9c90b33285b..8c5a3fb6c6c5 100644 --- a/include/linux/reciprocal_div.h +++ b/include/linux/reciprocal_div.h @@ -4,29 +4,32 @@ #include /* - * This file describes reciprocical division. + * This algorithm is based on the paper "Division by Invariant + * Integers Using Multiplication" by Torbjörn Granlund and Peter + * L. Montgomery. * - * This optimizes the (A/B) problem, when A and B are two u32 - * and B is a known value (but not known at compile time) + * The assembler implementation from Agner Fog, which this code is + * based on, can be found here: + * http://www.agner.org/optimize/asmlib.zip * - * The math principle used is : - * Let RECIPROCAL_VALUE(B) be (((1LL << 32) + (B - 1))/ B) - * Then A / B = (u32)(((u64)(A) * (R)) >> 32) - * - * This replaces a divide by a multiply (and a shift), and - * is generally less expensive in CPU cycles. + * This optimization for A/B is helpful if the divisor B is mostly + * runtime invariant. The reciprocal of B is calculated in the + * slow-path with reciprocal_value(). The fast-path can then just use + * a much faster multiplication operation with a variable dividend A + * to calculate the division A/B. */ -/* - * Computes the reciprocal value (R) for the value B of the divisor. - * Should not be called before each reciprocal_divide(), - * or else the performance is slower than a normal divide. - */ -extern u32 reciprocal_value(u32 B); +struct reciprocal_value { + u32 m; + u8 sh1, sh2; +}; +struct reciprocal_value reciprocal_value(u32 d); -static inline u32 reciprocal_divide(u32 A, u32 R) +static inline u32 reciprocal_divide(u32 a, struct reciprocal_value R) { - return (u32)(((u64)A * R) >> 32); + u32 t = (u32)(((u64)a * R.m) >> 32); + return (t + ((a - t) >> R.sh1)) >> R.sh2; } -#endif + +#endif /* _LINUX_RECIPROCAL_DIV_H */ diff --git a/include/linux/slab_def.h b/include/linux/slab_def.h index 09bfffb08a56..96e8abae19a9 100644 --- a/include/linux/slab_def.h +++ b/include/linux/slab_def.h @@ -1,6 +1,8 @@ #ifndef _LINUX_SLAB_DEF_H #define _LINUX_SLAB_DEF_H +#include + /* * Definitions unique to the original Linux SLAB allocator. */ @@ -12,7 +14,7 @@ struct kmem_cache { unsigned int shared; unsigned int size; - u32 reciprocal_buffer_size; + struct reciprocal_value reciprocal_buffer_size; /* 2) touched by every alloc & free from the backend */ unsigned int flags; /* constant flags */ diff --git a/include/net/red.h b/include/net/red.h index 168bb2f495f2..76e0b5f922c6 100644 --- a/include/net/red.h +++ b/include/net/red.h @@ -130,7 +130,8 @@ struct red_parms { u32 qth_max; /* Max avg length threshold: Wlog scaled */ u32 Scell_max; u32 max_P; /* probability, [0 .. 1.0] 32 scaled */ - u32 max_P_reciprocal; /* reciprocal_value(max_P / qth_delta) */ + /* reciprocal_value(max_P / qth_delta) */ + struct reciprocal_value max_P_reciprocal; u32 qth_delta; /* max_th - min_th */ u32 target_min; /* min_th + 0.4*(max_th - min_th) */ u32 target_max; /* min_th + 0.6*(max_th - min_th) */ diff --git a/lib/flex_array.c b/lib/flex_array.c index 6948a6692fc4..2eed22fa507c 100644 --- a/lib/flex_array.c +++ b/lib/flex_array.c @@ -90,8 +90,8 @@ struct flex_array *flex_array_alloc(int element_size, unsigned int total, { struct flex_array *ret; int elems_per_part = 0; - int reciprocal_elems = 0; int max_size = 0; + struct reciprocal_value reciprocal_elems = { 0 }; if (element_size) { elems_per_part = FLEX_ARRAY_ELEMENTS_PER_PART(element_size); @@ -119,6 +119,11 @@ EXPORT_SYMBOL(flex_array_alloc); static int fa_element_to_part_nr(struct flex_array *fa, unsigned int element_nr) { + /* + * if element_size == 0 we don't get here, so we never touch + * the zeroed fa->reciprocal_elems, which would yield invalid + * results + */ return reciprocal_divide(element_nr, fa->reciprocal_elems); } diff --git a/lib/reciprocal_div.c b/lib/reciprocal_div.c index 75510e94f7d0..464152410c51 100644 --- a/lib/reciprocal_div.c +++ b/lib/reciprocal_div.c @@ -1,11 +1,27 @@ +#include #include #include #include -u32 reciprocal_value(u32 k) +/* + * For a description of the algorithm please have a look at + * include/linux/reciprocal_div.h + */ + +struct reciprocal_value reciprocal_value(u32 d) { - u64 val = (1LL << 32) + (k - 1); - do_div(val, k); - return (u32)val; + struct reciprocal_value R; + u64 m; + int l; + + l = fls(d - 1); + m = ((1ULL << 32) * ((1ULL << l) - d)); + do_div(m, d); + ++m; + R.m = (u32)m; + R.sh1 = min(l, 1); + R.sh2 = max(l - 1, 0); + + return R; } EXPORT_SYMBOL(reciprocal_value); diff --git a/net/sched/sch_netem.c b/net/sched/sch_netem.c index a2bfc371b44a..de1059af6da1 100644 --- a/net/sched/sch_netem.c +++ b/net/sched/sch_netem.c @@ -91,7 +91,7 @@ struct netem_sched_data { u64 rate; s32 packet_overhead; u32 cell_size; - u32 cell_size_reciprocal; + struct reciprocal_value cell_size_reciprocal; s32 cell_overhead; struct crndstate { @@ -725,9 +725,11 @@ static void get_rate(struct Qdisc *sch, const struct nlattr *attr) q->rate = r->rate; q->packet_overhead = r->packet_overhead; q->cell_size = r->cell_size; + q->cell_overhead = r->cell_overhead; if (q->cell_size) q->cell_size_reciprocal = reciprocal_value(q->cell_size); - q->cell_overhead = r->cell_overhead; + else + q->cell_size_reciprocal = (struct reciprocal_value) { 0 }; } static int get_loss_clg(struct Qdisc *sch, const struct nlattr *attr)