From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from dispatch1-us1.ppe-hosted.com ([67.231.154.164]:44022 "EHLO dispatch1-us1.ppe-hosted.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932068AbeCIPES (ORCPT ); Fri, 9 Mar 2018 10:04:18 -0500 From: Edward Cree Subject: [PATCH ethtool 2/3] ethtool: add support for extra RSS contexts and RSS steering filters To: "John W. Linville" CC: References: Message-ID: <219f1d93-46b9-ba70-266e-21ea34887034@solarflare.com> Date: Fri, 9 Mar 2018 15:04:12 +0000 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset="utf-8" Content-Language: en-GB Content-Transfer-Encoding: 7bit Sender: netdev-owner@vger.kernel.org List-ID: RSS contexts can be created on a device with -X ... context new, modified with -X ... context N, and deleted with -X ... context N delete. N-tuple filters can be directed at those contexts with -N ... context N. Signed-off-by: Edward Cree --- ethtool.c | 141 ++++++++++++++++++++++++++++++++++++++++++++++++++++--------- internal.h | 4 +- rxclass.c | 58 +++++++++++++++++++++---- 3 files changed, 173 insertions(+), 30 deletions(-) diff --git a/ethtool.c b/ethtool.c index c870a2b..5ea9f05 100644 --- a/ethtool.c +++ b/ethtool.c @@ -1506,7 +1506,7 @@ static void dump_features(const struct feature_defs *defs, static int dump_rxfhash(int fhash, u64 val) { - switch (fhash) { + switch (fhash & ~FLOW_RSS) { case TCP_V4_FLOW: fprintf(stdout, "TCP over IPV4 flows"); break; @@ -3517,11 +3517,20 @@ static int do_srxclass(struct cmd_context *ctx) if (ctx->argc < 2) exit_bad_args(); - if (ctx->argc == 3 && !strcmp(ctx->argp[0], "rx-flow-hash")) { + if (!strcmp(ctx->argp[0], "rx-flow-hash")) { int rx_fhash_set; u32 rx_fhash_val; struct ethtool_rxnfc nfccmd; + bool flow_rss = false; + if (ctx->argc == 5) { + if (strcmp(ctx->argp[3], "context")) + exit_bad_args(); + flow_rss = true; + nfccmd.rss_context = get_u32(ctx->argp[4], 0); + } else if (ctx->argc != 3) { + exit_bad_args(); + } rx_fhash_set = rxflow_str_to_type(ctx->argp[1]); if (!rx_fhash_set) exit_bad_args(); @@ -3531,16 +3540,19 @@ static int do_srxclass(struct cmd_context *ctx) nfccmd.cmd = ETHTOOL_SRXFH; nfccmd.flow_type = rx_fhash_set; nfccmd.data = rx_fhash_val; + if (flow_rss) + nfccmd.flow_type |= FLOW_RSS; err = send_ioctl(ctx, &nfccmd); if (err < 0) perror("Cannot change RX network flow hashing options"); } else if (!strcmp(ctx->argp[0], "flow-type")) { struct ethtool_rx_flow_spec rx_rule_fs; + __u32 rss_context = 0; ctx->argc--; ctx->argp++; - if (rxclass_parse_ruleopts(ctx, &rx_rule_fs) < 0) + if (rxclass_parse_ruleopts(ctx, &rx_rule_fs, &rss_context) < 0) exit_bad_args(); /* attempt to add rule via N-tuple specifier */ @@ -3549,7 +3561,7 @@ static int do_srxclass(struct cmd_context *ctx) return 0; /* attempt to add rule via network flow classifier */ - err = rxclass_rule_ins(ctx, &rx_rule_fs); + err = rxclass_rule_ins(ctx, &rx_rule_fs, rss_context); if (err < 0) { fprintf(stderr, "Cannot insert" " classification rule\n"); @@ -3578,8 +3590,18 @@ static int do_grxclass(struct cmd_context *ctx) struct ethtool_rxnfc nfccmd; int err; - if (ctx->argc == 2 && !strcmp(ctx->argp[0], "rx-flow-hash")) { + if (ctx->argc > 0 && !strcmp(ctx->argp[0], "rx-flow-hash")) { int rx_fhash_get; + bool flow_rss = false; + + if (ctx->argc == 4) { + if (strcmp(ctx->argp[2], "context")) + exit_bad_args(); + flow_rss = true; + nfccmd.rss_context = get_u32(ctx->argp[3], 0); + } else if (ctx->argc != 2) { + exit_bad_args(); + } rx_fhash_get = rxflow_str_to_type(ctx->argp[1]); if (!rx_fhash_get) @@ -3587,11 +3609,17 @@ static int do_grxclass(struct cmd_context *ctx) nfccmd.cmd = ETHTOOL_GRXFH; nfccmd.flow_type = rx_fhash_get; + if (flow_rss) + nfccmd.flow_type |= FLOW_RSS; err = send_ioctl(ctx, &nfccmd); - if (err < 0) + if (err < 0) { perror("Cannot get RX network flow hashing options"); - else + } else { + if (flow_rss) + fprintf(stdout, "For RSS context %u:\n", + nfccmd.rss_context); dump_rxfhash(rx_fhash_get, nfccmd.data); + } } else if (ctx->argc == 2 && !strcmp(ctx->argp[0], "rule")) { int rx_class_rule_get = get_uint_range(ctx->argp[1], 0, INT_MAX); @@ -3683,10 +3711,23 @@ static int do_grxfh(struct cmd_context *ctx) struct ethtool_rxfh rss_head = {0}; struct ethtool_rxnfc ring_count; struct ethtool_rxfh *rss; + u32 rss_context = 0; u32 i, indir_bytes; + int arg_num = 0; char *hkey; int err; + while (arg_num < ctx->argc) { + if (!strcmp(ctx->argp[arg_num], "context")) { + ++arg_num; + rss_context = get_int_range(ctx->argp[arg_num], 0, 1, + ETH_RXFH_CONTEXT_ALLOC - 1); + ++arg_num; + } else { + exit_bad_args(); + } + } + ring_count.cmd = ETHTOOL_GRXRINGS; err = send_ioctl(ctx, &ring_count); if (err < 0) { @@ -3695,6 +3736,7 @@ static int do_grxfh(struct cmd_context *ctx) } rss_head.cmd = ETHTOOL_GRSSH; + rss_head.rss_context = rss_context; err = send_ioctl(ctx, &rss_head); if (err < 0 && errno == EOPNOTSUPP) { return do_grxfhindir(ctx, &ring_count); @@ -3712,6 +3754,7 @@ static int do_grxfh(struct cmd_context *ctx) } rss->cmd = ETHTOOL_GRSSH; + rss->rss_context = rss_context; rss->indir_size = rss_head.indir_size; rss->key_size = rss_head.key_size; err = send_ioctl(ctx, rss); @@ -3872,6 +3915,8 @@ static int do_srxfh(struct cmd_context *ctx) u32 req_hfunc = 0; u32 entry_size = sizeof(rss_head.rss_config[0]); u32 num_weights = 0; + u32 rss_context = 0; + int delete = 0; if (ctx->argc < 1) exit_bad_args(); @@ -3907,6 +3952,18 @@ static int do_srxfh(struct cmd_context *ctx) if (!req_hfunc_name) exit_bad_args(); ++arg_num; + } else if (!strcmp(ctx->argp[arg_num], "context")) { + ++arg_num; + if(!strcmp(ctx->argp[arg_num], "new")) + rss_context = ETH_RXFH_CONTEXT_ALLOC; + else + rss_context = get_int_range( + ctx->argp[arg_num], 0, 1, + ETH_RXFH_CONTEXT_ALLOC - 1); + ++arg_num; + } else if (!strcmp(ctx->argp[arg_num], "delete")) { + ++arg_num; + delete = 1; } else { exit_bad_args(); } @@ -3930,6 +3987,41 @@ static int do_srxfh(struct cmd_context *ctx) return 1; } + if (rxfhindir_default && rss_context) { + fprintf(stderr, + "Default and context options are mutually exclusive\n"); + return 1; + } + + if (delete && !rss_context) { + fprintf(stderr, "Delete option requires context option\n"); + return 1; + } + + if (delete && rxfhindir_weight) { + fprintf(stderr, + "Delete and weight options are mutually exclusive\n"); + return 1; + } + + if (delete && rxfhindir_equal) { + fprintf(stderr, + "Delete and equal options are mutually exclusive\n"); + return 1; + } + + if (delete && rxfhindir_default) { + fprintf(stderr, + "Delete and default options are mutually exclusive\n"); + return 1; + } + + if (delete && rxfhindir_key) { + fprintf(stderr, + "Delete and hkey options are mutually exclusive\n"); + return 1; + } + ring_count.cmd = ETHTOOL_GRXRINGS; err = send_ioctl(ctx, &ring_count); if (err < 0) { @@ -3940,7 +4032,7 @@ static int do_srxfh(struct cmd_context *ctx) rss_head.cmd = ETHTOOL_GRSSH; err = send_ioctl(ctx, &rss_head); if (err < 0 && errno == EOPNOTSUPP && !rxfhindir_key && - !req_hfunc_name) { + !req_hfunc_name && !rss_context) { return do_srxfhindir(ctx, rxfhindir_default, rxfhindir_equal, rxfhindir_weight, num_weights); } else if (err < 0) { @@ -3988,14 +4080,19 @@ static int do_srxfh(struct cmd_context *ctx) goto free; } rss->cmd = ETHTOOL_SRSSH; - rss->indir_size = rss_head.indir_size; - rss->key_size = rss_head.key_size; + rss->rss_context = rss_context; rss->hfunc = req_hfunc; - - if (fill_indir_table(&rss->indir_size, rss->rss_config, rxfhindir_default, - rxfhindir_equal, rxfhindir_weight, num_weights)) { - err = 1; - goto free; + if (delete) { + rss->indir_size = rss->key_size = 0; + } else { + rss->indir_size = rss_head.indir_size; + rss->key_size = rss_head.key_size; + if (fill_indir_table(&rss->indir_size, rss->rss_config, + rxfhindir_default, rxfhindir_equal, + rxfhindir_weight, num_weights)) { + err = 1; + goto free; + } } if (hkey) @@ -4008,6 +4105,8 @@ static int do_srxfh(struct cmd_context *ctx) if (err < 0) { perror("Cannot set RX flow hash configuration"); err = 1; + } else if (rss_context == ETH_RXFH_CONTEXT_ALLOC) { + printf("New RSS context is %d\n", rss->rss_context); } free: @@ -5023,12 +5122,12 @@ static const struct option { { "-n|-u|--show-nfc|--show-ntuple", 1, do_grxclass, "Show Rx network flow classification options or rules", " [ rx-flow-hash tcp4|udp4|ah4|esp4|sctp4|" - "tcp6|udp6|ah6|esp6|sctp6 |\n" + "tcp6|udp6|ah6|esp6|sctp6 [context %d] |\n" " rule %d ]\n" }, { "-N|-U|--config-nfc|--config-ntuple", 1, do_srxclass, "Configure Rx network flow classification options or rules", " rx-flow-hash tcp4|udp4|ah4|esp4|sctp4|" - "tcp6|udp6|ah6|esp6|sctp6 m|v|t|s|d|f|n|r... |\n" + "tcp6|udp6|ah6|esp6|sctp6 m|v|t|s|d|f|n|r... [context %d] |\n" " flow-type ether|ip4|tcp4|udp4|sctp4|ah4|esp4|" "ip6|tcp6|udp6|ah6|esp6|sctp6\n" " [ src %x:%x:%x:%x:%x:%x [m %x:%x:%x:%x:%x:%x] ]\n" @@ -5047,17 +5146,21 @@ static const struct option { " [ user-def %x [m %x] ]\n" " [ dst-mac %x:%x:%x:%x:%x:%x [m %x:%x:%x:%x:%x:%x] ]\n" " [ action %d ] | [ vf %d queue %d ]\n" + " [ context %d ]\n" " [ loc %d]] |\n" " delete %d\n" }, { "-T|--show-time-stamping", 1, do_tsinfo, "Show time stamping capabilities" }, { "-x|--show-rxfh-indir|--show-rxfh", 1, do_grxfh, - "Show Rx flow hash indirection table and/or RSS hash key" }, + "Show Rx flow hash indirection table and/or RSS hash key", + " [ context %d ]\n" }, { "-X|--set-rxfh-indir|--rxfh", 1, do_srxfh, "Set Rx flow hash indirection table and/or RSS hash key", + " [ context %d|new ]\n" " [ equal N | weight W0 W1 ... | default ]\n" " [ hkey %x:%x:%x:%x:%x:.... ]\n" - " [ hfunc FUNC ]\n" }, + " [ hfunc FUNC ]\n" + " [ delete ]\n" }, { "-f|--flash", 1, do_flash, "Flash firmware image from the specified file to a region on the device", " FILENAME [ REGION-NUMBER-TO-FLASH ]\n" }, diff --git a/internal.h b/internal.h index 4e658ea..913f4eb 100644 --- a/internal.h +++ b/internal.h @@ -332,11 +332,11 @@ int vmxnet3_dump_regs(struct ethtool_drvinfo *info, struct ethtool_regs *regs); /* Rx flow classification */ int rxclass_parse_ruleopts(struct cmd_context *ctx, - struct ethtool_rx_flow_spec *fsp); + struct ethtool_rx_flow_spec *fsp, __u32 *rss_context); int rxclass_rule_getall(struct cmd_context *ctx); int rxclass_rule_get(struct cmd_context *ctx, __u32 loc); int rxclass_rule_ins(struct cmd_context *ctx, - struct ethtool_rx_flow_spec *fsp); + struct ethtool_rx_flow_spec *fsp, __u32 rss_context); int rxclass_rule_del(struct cmd_context *ctx, __u32 loc); /* Module EEPROM parsing code */ diff --git a/rxclass.c b/rxclass.c index e05031b..ce4b382 100644 --- a/rxclass.c +++ b/rxclass.c @@ -94,14 +94,15 @@ static void rxclass_print_nfc_spec_ext(struct ethtool_rx_flow_spec *fsp) } } -static void rxclass_print_nfc_rule(struct ethtool_rx_flow_spec *fsp) +static void rxclass_print_nfc_rule(struct ethtool_rx_flow_spec *fsp, + __u32 rss_context) { unsigned char *smac, *smacm, *dmac, *dmacm; __u32 flow_type; fprintf(stdout, "Filter: %d\n", fsp->location); - flow_type = fsp->flow_type & ~(FLOW_EXT | FLOW_MAC_EXT); + flow_type = fsp->flow_type & ~(FLOW_EXT | FLOW_MAC_EXT | FLOW_RSS); invert_flow_mask(fsp); @@ -247,6 +248,9 @@ static void rxclass_print_nfc_rule(struct ethtool_rx_flow_spec *fsp) rxclass_print_nfc_spec_ext(fsp); + if (fsp->flow_type & FLOW_RSS) + fprintf(stdout, "\tRSS Context ID: %u\n", rss_context); + if (fsp->ring_cookie != RX_CLS_FLOW_DISC) { u64 vf = ethtool_get_flow_spec_ring_vf(fsp->ring_cookie); u64 queue = ethtool_get_flow_spec_ring(fsp->ring_cookie); @@ -269,10 +273,11 @@ static void rxclass_print_nfc_rule(struct ethtool_rx_flow_spec *fsp) fprintf(stdout, "\n"); } -static void rxclass_print_rule(struct ethtool_rx_flow_spec *fsp) +static void rxclass_print_rule(struct ethtool_rx_flow_spec *fsp, + __u32 rss_context) { /* print the rule in this location */ - switch (fsp->flow_type & ~(FLOW_EXT | FLOW_MAC_EXT)) { + switch (fsp->flow_type & ~(FLOW_EXT | FLOW_MAC_EXT | FLOW_RSS)) { case TCP_V4_FLOW: case UDP_V4_FLOW: case SCTP_V4_FLOW: @@ -285,11 +290,11 @@ static void rxclass_print_rule(struct ethtool_rx_flow_spec *fsp) case ESP_V6_FLOW: case IPV6_USER_FLOW: case ETHER_FLOW: - rxclass_print_nfc_rule(fsp); + rxclass_print_nfc_rule(fsp, rss_context); break; case IPV4_USER_FLOW: if (fsp->h_u.usr_ip4_spec.ip_ver == ETH_RX_NFC_IP4) - rxclass_print_nfc_rule(fsp); + rxclass_print_nfc_rule(fsp, rss_context); else /* IPv6 uses IPV6_USER_FLOW */ fprintf(stderr, "IPV4_USER_FLOW with wrong ip_ver\n"); break; @@ -333,7 +338,7 @@ int rxclass_rule_get(struct cmd_context *ctx, __u32 loc) } /* display rule */ - rxclass_print_rule(&nfccmd.fs); + rxclass_print_rule(&nfccmd.fs, (__u32)nfccmd.rss_context); return err; } @@ -563,7 +568,7 @@ out: } int rxclass_rule_ins(struct cmd_context *ctx, - struct ethtool_rx_flow_spec *fsp) + struct ethtool_rx_flow_spec *fsp, __u32 rss_context) { struct ethtool_rxnfc nfccmd; __u32 loc = fsp->location; @@ -581,6 +586,7 @@ int rxclass_rule_ins(struct cmd_context *ctx, /* notify netdev of new rule */ nfccmd.cmd = ETHTOOL_SRXCLSRLINS; + nfccmd.rss_context = rss_context; nfccmd.fs = *fsp; err = send_ioctl(ctx, &nfccmd); if (err < 0) @@ -1254,7 +1260,7 @@ static int rxclass_get_mask(char *str, unsigned char *p, } int rxclass_parse_ruleopts(struct cmd_context *ctx, - struct ethtool_rx_flow_spec *fsp) + struct ethtool_rx_flow_spec *fsp, __u32 *rss_context) { const struct rule_opts *options; unsigned char *p = (unsigned char *)fsp; @@ -1343,6 +1349,40 @@ int rxclass_parse_ruleopts(struct cmd_context *ctx, for (i = 1; i < argc;) { const struct rule_opts *opt; int idx; + + /* special handling for 'context %d' as it doesn't go in + * the struct ethtool_rx_flow_spec + */ + if (!strcmp(argp[i], "context")) { + unsigned long long val; + + i++; + if (i >= argc) { + fprintf(stderr, "'context' missing value\n"); + return -1; + } + + if (rxclass_get_ulong(argp[i], &val, 32)) { + fprintf(stderr, "Invalid context value[%s]\n", + argp[i]); + return -1; + } + + /* Can't use the ALLOC special value as the context ID + * of a filter to insert + */ + if ((__u32)val == ETH_RXFH_CONTEXT_ALLOC) { + fprintf(stderr, "Bad context value %x\n", + (__u32)val); + return -1; + } + + *rss_context = (__u32)val; + fsp->flow_type |= FLOW_RSS; + i++; + continue; + } + for (opt = options, idx = 0; idx < n_opts; idx++, opt++) { char mask_name[16];