Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge branch 'net-wangxun-complete-ethtool-coalesce-options'
Jiawen Wu says:
====================
net: wangxun: complete ethtool coalesce options
Support to use adaptive RX coalescing. Change the default RX coalesce
usecs and limit the range of parameters for various types of devices,
according to their hardware design.
====================
Link: https://patch.msgid.link/20250821023408.53472-1-jiawenwu@trustnetic.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
+155
-27
+1
drivers/net/ethernet/wangxun/Kconfig
+1
drivers/net/ethernet/wangxun/Kconfig
+35
-20
drivers/net/ethernet/wangxun/libwx/wx_ethtool.c
+35
-20
drivers/net/ethernet/wangxun/libwx/wx_ethtool.c
···
303
303
else
304
304
ec->rx_coalesce_usecs = wx->rx_itr_setting >> 2;
305
305
306
+
if (wx->adaptive_itr) {
307
+
ec->use_adaptive_rx_coalesce = 1;
308
+
ec->use_adaptive_tx_coalesce = 1;
309
+
}
310
+
306
311
/* if in mixed tx/rx queues per vector mode, report only rx settings */
307
312
if (wx->q_vector[0]->tx.count && wx->q_vector[0]->rx.count)
308
313
return 0;
···
339
334
return -EOPNOTSUPP;
340
335
}
341
336
342
-
if (ec->tx_max_coalesced_frames_irq)
343
-
wx->tx_work_limit = ec->tx_max_coalesced_frames_irq;
337
+
if (ec->tx_max_coalesced_frames_irq > U16_MAX ||
338
+
!ec->tx_max_coalesced_frames_irq)
339
+
return -EINVAL;
340
+
341
+
wx->tx_work_limit = ec->tx_max_coalesced_frames_irq;
344
342
345
343
switch (wx->mac.type) {
346
344
case wx_mac_sp:
347
345
max_eitr = WX_SP_MAX_EITR;
346
+
rx_itr_param = WX_20K_ITR;
347
+
tx_itr_param = WX_12K_ITR;
348
348
break;
349
349
case wx_mac_aml:
350
350
case wx_mac_aml40:
351
351
max_eitr = WX_AML_MAX_EITR;
352
+
rx_itr_param = WX_20K_ITR;
353
+
tx_itr_param = WX_12K_ITR;
352
354
break;
353
355
default:
354
356
max_eitr = WX_EM_MAX_EITR;
357
+
rx_itr_param = WX_7K_ITR;
358
+
tx_itr_param = WX_7K_ITR;
355
359
break;
356
360
}
357
361
···
368
354
(ec->tx_coalesce_usecs > (max_eitr >> 2)))
369
355
return -EINVAL;
370
356
357
+
if (ec->use_adaptive_rx_coalesce) {
358
+
wx->adaptive_itr = true;
359
+
wx->rx_itr_setting = 1;
360
+
wx->tx_itr_setting = 1;
361
+
return 0;
362
+
}
363
+
371
364
if (ec->rx_coalesce_usecs > 1)
372
365
wx->rx_itr_setting = ec->rx_coalesce_usecs << 2;
373
366
else
374
367
wx->rx_itr_setting = ec->rx_coalesce_usecs;
375
-
376
-
if (wx->rx_itr_setting == 1)
377
-
rx_itr_param = WX_20K_ITR;
378
-
else
379
-
rx_itr_param = wx->rx_itr_setting;
380
368
381
369
if (ec->tx_coalesce_usecs > 1)
382
370
wx->tx_itr_setting = ec->tx_coalesce_usecs << 2;
383
371
else
384
372
wx->tx_itr_setting = ec->tx_coalesce_usecs;
385
373
386
-
if (wx->tx_itr_setting == 1) {
387
-
switch (wx->mac.type) {
388
-
case wx_mac_sp:
389
-
case wx_mac_aml:
390
-
case wx_mac_aml40:
391
-
tx_itr_param = WX_12K_ITR;
392
-
break;
393
-
default:
394
-
tx_itr_param = WX_20K_ITR;
395
-
break;
396
-
}
397
-
} else {
398
-
tx_itr_param = wx->tx_itr_setting;
374
+
if (wx->adaptive_itr) {
375
+
wx->adaptive_itr = false;
376
+
wx->rx_itr_setting = rx_itr_param;
377
+
wx->tx_itr_setting = tx_itr_param;
378
+
} else if (wx->rx_itr_setting == 1 || wx->tx_itr_setting == 1) {
379
+
wx->adaptive_itr = true;
399
380
}
381
+
382
+
if (wx->rx_itr_setting != 1)
383
+
rx_itr_param = wx->rx_itr_setting;
384
+
385
+
if (wx->tx_itr_setting != 1)
386
+
tx_itr_param = wx->tx_itr_setting;
400
387
401
388
/* mixed Rx/Tx */
402
389
if (wx->q_vector[0]->tx.count && wx->q_vector[0]->rx.count)
+102
-1
drivers/net/ethernet/wangxun/libwx/wx_lib.c
+102
-1
drivers/net/ethernet/wangxun/libwx/wx_lib.c
···
16
16
#include "wx_lib.h"
17
17
#include "wx_ptp.h"
18
18
#include "wx_hw.h"
19
+
#include "wx_vf_lib.h"
19
20
20
21
/* Lookup table mapping the HW PTYPE to the bit field for decoding */
21
22
static struct wx_dec_ptype wx_ptype_lookup[256] = {
···
833
832
return !!budget;
834
833
}
835
834
835
+
static void wx_update_rx_dim_sample(struct wx_q_vector *q_vector)
836
+
{
837
+
struct dim_sample sample = {};
838
+
839
+
dim_update_sample(q_vector->total_events,
840
+
q_vector->rx.total_packets,
841
+
q_vector->rx.total_bytes,
842
+
&sample);
843
+
844
+
net_dim(&q_vector->rx.dim, &sample);
845
+
}
846
+
847
+
static void wx_update_tx_dim_sample(struct wx_q_vector *q_vector)
848
+
{
849
+
struct dim_sample sample = {};
850
+
851
+
dim_update_sample(q_vector->total_events,
852
+
q_vector->tx.total_packets,
853
+
q_vector->tx.total_bytes,
854
+
&sample);
855
+
856
+
net_dim(&q_vector->tx.dim, &sample);
857
+
}
858
+
859
+
static void wx_update_dim_sample(struct wx_q_vector *q_vector)
860
+
{
861
+
wx_update_rx_dim_sample(q_vector);
862
+
wx_update_tx_dim_sample(q_vector);
863
+
}
864
+
836
865
/**
837
866
* wx_poll - NAPI polling RX/TX cleanup routine
838
867
* @napi: napi struct with our devices info in it
···
909
878
910
879
/* all work done, exit the polling mode */
911
880
if (likely(napi_complete_done(napi, work_done))) {
881
+
if (wx->adaptive_itr)
882
+
wx_update_dim_sample(q_vector);
912
883
if (netif_running(wx->netdev))
913
884
wx_intr_enable(wx, WX_INTR_Q(q_vector->v_idx));
914
885
}
···
1624
1591
}
1625
1592
EXPORT_SYMBOL(wx_xmit_frame);
1626
1593
1594
+
static void wx_set_itr(struct wx_q_vector *q_vector)
1595
+
{
1596
+
struct wx *wx = q_vector->wx;
1597
+
u32 new_itr;
1598
+
1599
+
if (!wx->adaptive_itr)
1600
+
return;
1601
+
1602
+
/* use the smallest value of new ITR delay calculations */
1603
+
new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
1604
+
new_itr <<= 2;
1605
+
1606
+
if (new_itr != q_vector->itr) {
1607
+
/* save the algorithm value here */
1608
+
q_vector->itr = new_itr;
1609
+
1610
+
if (wx->pdev->is_virtfn)
1611
+
wx_write_eitr_vf(q_vector);
1612
+
else
1613
+
wx_write_eitr(q_vector);
1614
+
}
1615
+
}
1616
+
1617
+
static void wx_rx_dim_work(struct work_struct *work)
1618
+
{
1619
+
struct dim *dim = container_of(work, struct dim, work);
1620
+
struct dim_cq_moder rx_moder;
1621
+
struct wx_ring_container *rx;
1622
+
struct wx_q_vector *q_vector;
1623
+
1624
+
rx = container_of(dim, struct wx_ring_container, dim);
1625
+
1626
+
rx_moder = net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1627
+
rx->itr = rx_moder.usec;
1628
+
1629
+
q_vector = container_of(rx, struct wx_q_vector, rx);
1630
+
wx_set_itr(q_vector);
1631
+
1632
+
dim->state = DIM_START_MEASURE;
1633
+
}
1634
+
1635
+
static void wx_tx_dim_work(struct work_struct *work)
1636
+
{
1637
+
struct dim *dim = container_of(work, struct dim, work);
1638
+
struct dim_cq_moder tx_moder;
1639
+
struct wx_ring_container *tx;
1640
+
struct wx_q_vector *q_vector;
1641
+
1642
+
tx = container_of(dim, struct wx_ring_container, dim);
1643
+
1644
+
tx_moder = net_dim_get_tx_moderation(dim->mode, dim->profile_ix);
1645
+
tx->itr = tx_moder.usec;
1646
+
1647
+
q_vector = container_of(tx, struct wx_q_vector, tx);
1648
+
wx_set_itr(q_vector);
1649
+
1650
+
dim->state = DIM_START_MEASURE;
1651
+
}
1652
+
1627
1653
void wx_napi_enable_all(struct wx *wx)
1628
1654
{
1629
1655
struct wx_q_vector *q_vector;
···
1690
1598
1691
1599
for (q_idx = 0; q_idx < wx->num_q_vectors; q_idx++) {
1692
1600
q_vector = wx->q_vector[q_idx];
1601
+
1602
+
INIT_WORK(&q_vector->rx.dim.work, wx_rx_dim_work);
1603
+
INIT_WORK(&q_vector->tx.dim.work, wx_tx_dim_work);
1604
+
q_vector->rx.dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_CQE;
1605
+
q_vector->tx.dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_CQE;
1693
1606
napi_enable(&q_vector->napi);
1694
1607
}
1695
1608
}
···
1708
1611
for (q_idx = 0; q_idx < wx->num_q_vectors; q_idx++) {
1709
1612
q_vector = wx->q_vector[q_idx];
1710
1613
napi_disable(&q_vector->napi);
1614
+
disable_work_sync(&q_vector->rx.dim.work);
1615
+
disable_work_sync(&q_vector->tx.dim.work);
1711
1616
}
1712
1617
}
1713
1618
EXPORT_SYMBOL(wx_napi_disable_all);
···
2296
2197
struct wx_q_vector *q_vector = data;
2297
2198
2298
2199
/* EIAM disabled interrupts (on this vector) for us */
2299
-
if (q_vector->rx.ring || q_vector->tx.ring)
2200
+
if (q_vector->rx.ring || q_vector->tx.ring) {
2300
2201
napi_schedule_irqoff(&q_vector->napi);
2202
+
q_vector->total_events++;
2203
+
}
2301
2204
2302
2205
return IRQ_HANDLED;
2303
2206
}
+5
drivers/net/ethernet/wangxun/libwx/wx_type.h
+5
drivers/net/ethernet/wangxun/libwx/wx_type.h
···
10
10
#include <linux/netdevice.h>
11
11
#include <linux/if_vlan.h>
12
12
#include <linux/phylink.h>
13
+
#include <linux/dim.h>
13
14
#include <net/ip.h>
14
15
15
16
#define WX_NCSI_SUP 0x8000
···
1034
1033
unsigned int total_packets; /* total packets processed this int */
1035
1034
u8 count; /* total number of rings in vector */
1036
1035
u8 itr; /* current ITR setting for ring */
1036
+
struct dim dim; /* data for net_dim algorithm */
1037
1037
};
1038
1038
struct wx_ring {
1039
1039
struct wx_ring *next; /* pointer to next ring in q_vector */
···
1090
1088
struct wx_ring_container rx, tx;
1091
1089
struct napi_struct napi;
1092
1090
struct rcu_head rcu; /* to avoid race with update stats on free */
1091
+
1092
+
u16 total_events; /* number of interrupts processed */
1093
1093
1094
1094
char name[IFNAMSIZ + 17];
1095
1095
···
1272
1268
int num_rx_queues;
1273
1269
u16 rx_itr_setting;
1274
1270
u16 rx_work_limit;
1271
+
bool adaptive_itr;
1275
1272
1276
1273
int num_q_vectors; /* current number of q_vectors for device */
1277
1274
int max_q_vectors; /* upper limit of q_vectors for device */
+1
-1
drivers/net/ethernet/wangxun/libwx/wx_vf_lib.c
+1
-1
drivers/net/ethernet/wangxun/libwx/wx_vf_lib.c
+1
drivers/net/ethernet/wangxun/libwx/wx_vf_lib.h
+1
drivers/net/ethernet/wangxun/libwx/wx_vf_lib.h
+2
-1
drivers/net/ethernet/wangxun/ngbe/ngbe_ethtool.c
+2
-1
drivers/net/ethernet/wangxun/ngbe/ngbe_ethtool.c
···
115
115
116
116
static const struct ethtool_ops ngbe_ethtool_ops = {
117
117
.supported_coalesce_params = ETHTOOL_COALESCE_USECS |
118
-
ETHTOOL_COALESCE_TX_MAX_FRAMES_IRQ,
118
+
ETHTOOL_COALESCE_TX_MAX_FRAMES_IRQ |
119
+
ETHTOOL_COALESCE_USE_ADAPTIVE,
119
120
.get_drvinfo = wx_get_drvinfo,
120
121
.get_link = ethtool_op_get_link,
121
122
.get_link_ksettings = wx_get_link_ksettings,
+3
-3
drivers/net/ethernet/wangxun/ngbe/ngbe_main.c
+3
-3
drivers/net/ethernet/wangxun/ngbe/ngbe_main.c
···
119
119
num_online_cpus());
120
120
wx->rss_enabled = true;
121
121
122
-
/* enable itr by default in dynamic mode */
123
-
wx->rx_itr_setting = 1;
124
-
wx->tx_itr_setting = 1;
122
+
wx->adaptive_itr = false;
123
+
wx->rx_itr_setting = WX_7K_ITR;
124
+
wx->tx_itr_setting = WX_7K_ITR;
125
125
126
126
/* set default ring sizes */
127
127
wx->tx_ring_count = NGBE_DEFAULT_TXD;
+1
drivers/net/ethernet/wangxun/ngbevf/ngbevf_main.c
+1
drivers/net/ethernet/wangxun/ngbevf/ngbevf_main.c
+2
-1
drivers/net/ethernet/wangxun/txgbe/txgbe_ethtool.c
+2
-1
drivers/net/ethernet/wangxun/txgbe/txgbe_ethtool.c
···
538
538
539
539
static const struct ethtool_ops txgbe_ethtool_ops = {
540
540
.supported_coalesce_params = ETHTOOL_COALESCE_USECS |
541
-
ETHTOOL_COALESCE_TX_MAX_FRAMES_IRQ,
541
+
ETHTOOL_COALESCE_TX_MAX_FRAMES_IRQ |
542
+
ETHTOOL_COALESCE_USE_ADAPTIVE,
542
543
.get_drvinfo = wx_get_drvinfo,
543
544
.nway_reset = wx_nway_reset,
544
545
.get_link = ethtool_op_get_link,
+1
drivers/net/ethernet/wangxun/txgbe/txgbe_main.c
+1
drivers/net/ethernet/wangxun/txgbe/txgbe_main.c
+1
drivers/net/ethernet/wangxun/txgbevf/txgbevf_main.c
+1
drivers/net/ethernet/wangxun/txgbevf/txgbevf_main.c
···
144
144
wx->mac.max_tx_queues = TXGBEVF_MAX_TX_QUEUES;
145
145
wx->mac.max_rx_queues = TXGBEVF_MAX_RX_QUEUES;
146
146
/* Enable dynamic interrupt throttling rates */
147
+
wx->adaptive_itr = true;
147
148
wx->rx_itr_setting = 1;
148
149
wx->tx_itr_setting = 1;
149
150
/* set default ring sizes */