Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0
0002 
0003 /* net/sched/sch_taprio.c    Time Aware Priority Scheduler
0004  *
0005  * Authors: Vinicius Costa Gomes <vinicius.gomes@intel.com>
0006  *
0007  */
0008 
0009 #include <linux/ethtool.h>
0010 #include <linux/types.h>
0011 #include <linux/slab.h>
0012 #include <linux/kernel.h>
0013 #include <linux/string.h>
0014 #include <linux/list.h>
0015 #include <linux/errno.h>
0016 #include <linux/skbuff.h>
0017 #include <linux/math64.h>
0018 #include <linux/module.h>
0019 #include <linux/spinlock.h>
0020 #include <linux/rcupdate.h>
0021 #include <linux/time.h>
0022 #include <net/netlink.h>
0023 #include <net/pkt_sched.h>
0024 #include <net/pkt_cls.h>
0025 #include <net/sch_generic.h>
0026 #include <net/sock.h>
0027 #include <net/tcp.h>
0028 
0029 static LIST_HEAD(taprio_list);
0030 static DEFINE_SPINLOCK(taprio_list_lock);
0031 
0032 #define TAPRIO_ALL_GATES_OPEN -1
0033 
0034 #define TXTIME_ASSIST_IS_ENABLED(flags) ((flags) & TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST)
0035 #define FULL_OFFLOAD_IS_ENABLED(flags) ((flags) & TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD)
0036 #define TAPRIO_FLAGS_INVALID U32_MAX
0037 
0038 struct sched_entry {
0039     struct list_head list;
0040 
0041     /* The instant that this entry "closes" and the next one
0042      * should open, the qdisc will make some effort so that no
0043      * packet leaves after this time.
0044      */
0045     ktime_t close_time;
0046     ktime_t next_txtime;
0047     atomic_t budget;
0048     int index;
0049     u32 gate_mask;
0050     u32 interval;
0051     u8 command;
0052 };
0053 
0054 struct sched_gate_list {
0055     struct rcu_head rcu;
0056     struct list_head entries;
0057     size_t num_entries;
0058     ktime_t cycle_close_time;
0059     s64 cycle_time;
0060     s64 cycle_time_extension;
0061     s64 base_time;
0062 };
0063 
0064 struct taprio_sched {
0065     struct Qdisc **qdiscs;
0066     struct Qdisc *root;
0067     u32 flags;
0068     enum tk_offsets tk_offset;
0069     int clockid;
0070     bool offloaded;
0071     atomic64_t picos_per_byte; /* Using picoseconds because for 10Gbps+
0072                     * speeds it's sub-nanoseconds per byte
0073                     */
0074 
0075     /* Protects the update side of the RCU protected current_entry */
0076     spinlock_t current_entry_lock;
0077     struct sched_entry __rcu *current_entry;
0078     struct sched_gate_list __rcu *oper_sched;
0079     struct sched_gate_list __rcu *admin_sched;
0080     struct hrtimer advance_timer;
0081     struct list_head taprio_list;
0082     struct sk_buff *(*dequeue)(struct Qdisc *sch);
0083     struct sk_buff *(*peek)(struct Qdisc *sch);
0084     u32 txtime_delay;
0085 };
0086 
0087 struct __tc_taprio_qopt_offload {
0088     refcount_t users;
0089     struct tc_taprio_qopt_offload offload;
0090 };
0091 
0092 static ktime_t sched_base_time(const struct sched_gate_list *sched)
0093 {
0094     if (!sched)
0095         return KTIME_MAX;
0096 
0097     return ns_to_ktime(sched->base_time);
0098 }
0099 
0100 static ktime_t taprio_mono_to_any(const struct taprio_sched *q, ktime_t mono)
0101 {
0102     /* This pairs with WRITE_ONCE() in taprio_parse_clockid() */
0103     enum tk_offsets tk_offset = READ_ONCE(q->tk_offset);
0104 
0105     switch (tk_offset) {
0106     case TK_OFFS_MAX:
0107         return mono;
0108     default:
0109         return ktime_mono_to_any(mono, tk_offset);
0110     }
0111 }
0112 
0113 static ktime_t taprio_get_time(const struct taprio_sched *q)
0114 {
0115     return taprio_mono_to_any(q, ktime_get());
0116 }
0117 
0118 static void taprio_free_sched_cb(struct rcu_head *head)
0119 {
0120     struct sched_gate_list *sched = container_of(head, struct sched_gate_list, rcu);
0121     struct sched_entry *entry, *n;
0122 
0123     list_for_each_entry_safe(entry, n, &sched->entries, list) {
0124         list_del(&entry->list);
0125         kfree(entry);
0126     }
0127 
0128     kfree(sched);
0129 }
0130 
0131 static void switch_schedules(struct taprio_sched *q,
0132                  struct sched_gate_list **admin,
0133                  struct sched_gate_list **oper)
0134 {
0135     rcu_assign_pointer(q->oper_sched, *admin);
0136     rcu_assign_pointer(q->admin_sched, NULL);
0137 
0138     if (*oper)
0139         call_rcu(&(*oper)->rcu, taprio_free_sched_cb);
0140 
0141     *oper = *admin;
0142     *admin = NULL;
0143 }
0144 
0145 /* Get how much time has been already elapsed in the current cycle. */
0146 static s32 get_cycle_time_elapsed(struct sched_gate_list *sched, ktime_t time)
0147 {
0148     ktime_t time_since_sched_start;
0149     s32 time_elapsed;
0150 
0151     time_since_sched_start = ktime_sub(time, sched->base_time);
0152     div_s64_rem(time_since_sched_start, sched->cycle_time, &time_elapsed);
0153 
0154     return time_elapsed;
0155 }
0156 
0157 static ktime_t get_interval_end_time(struct sched_gate_list *sched,
0158                      struct sched_gate_list *admin,
0159                      struct sched_entry *entry,
0160                      ktime_t intv_start)
0161 {
0162     s32 cycle_elapsed = get_cycle_time_elapsed(sched, intv_start);
0163     ktime_t intv_end, cycle_ext_end, cycle_end;
0164 
0165     cycle_end = ktime_add_ns(intv_start, sched->cycle_time - cycle_elapsed);
0166     intv_end = ktime_add_ns(intv_start, entry->interval);
0167     cycle_ext_end = ktime_add(cycle_end, sched->cycle_time_extension);
0168 
0169     if (ktime_before(intv_end, cycle_end))
0170         return intv_end;
0171     else if (admin && admin != sched &&
0172          ktime_after(admin->base_time, cycle_end) &&
0173          ktime_before(admin->base_time, cycle_ext_end))
0174         return admin->base_time;
0175     else
0176         return cycle_end;
0177 }
0178 
0179 static int length_to_duration(struct taprio_sched *q, int len)
0180 {
0181     return div_u64(len * atomic64_read(&q->picos_per_byte), PSEC_PER_NSEC);
0182 }
0183 
0184 /* Returns the entry corresponding to next available interval. If
0185  * validate_interval is set, it only validates whether the timestamp occurs
0186  * when the gate corresponding to the skb's traffic class is open.
0187  */
0188 static struct sched_entry *find_entry_to_transmit(struct sk_buff *skb,
0189                           struct Qdisc *sch,
0190                           struct sched_gate_list *sched,
0191                           struct sched_gate_list *admin,
0192                           ktime_t time,
0193                           ktime_t *interval_start,
0194                           ktime_t *interval_end,
0195                           bool validate_interval)
0196 {
0197     ktime_t curr_intv_start, curr_intv_end, cycle_end, packet_transmit_time;
0198     ktime_t earliest_txtime = KTIME_MAX, txtime, cycle, transmit_end_time;
0199     struct sched_entry *entry = NULL, *entry_found = NULL;
0200     struct taprio_sched *q = qdisc_priv(sch);
0201     struct net_device *dev = qdisc_dev(sch);
0202     bool entry_available = false;
0203     s32 cycle_elapsed;
0204     int tc, n;
0205 
0206     tc = netdev_get_prio_tc_map(dev, skb->priority);
0207     packet_transmit_time = length_to_duration(q, qdisc_pkt_len(skb));
0208 
0209     *interval_start = 0;
0210     *interval_end = 0;
0211 
0212     if (!sched)
0213         return NULL;
0214 
0215     cycle = sched->cycle_time;
0216     cycle_elapsed = get_cycle_time_elapsed(sched, time);
0217     curr_intv_end = ktime_sub_ns(time, cycle_elapsed);
0218     cycle_end = ktime_add_ns(curr_intv_end, cycle);
0219 
0220     list_for_each_entry(entry, &sched->entries, list) {
0221         curr_intv_start = curr_intv_end;
0222         curr_intv_end = get_interval_end_time(sched, admin, entry,
0223                               curr_intv_start);
0224 
0225         if (ktime_after(curr_intv_start, cycle_end))
0226             break;
0227 
0228         if (!(entry->gate_mask & BIT(tc)) ||
0229             packet_transmit_time > entry->interval)
0230             continue;
0231 
0232         txtime = entry->next_txtime;
0233 
0234         if (ktime_before(txtime, time) || validate_interval) {
0235             transmit_end_time = ktime_add_ns(time, packet_transmit_time);
0236             if ((ktime_before(curr_intv_start, time) &&
0237                  ktime_before(transmit_end_time, curr_intv_end)) ||
0238                 (ktime_after(curr_intv_start, time) && !validate_interval)) {
0239                 entry_found = entry;
0240                 *interval_start = curr_intv_start;
0241                 *interval_end = curr_intv_end;
0242                 break;
0243             } else if (!entry_available && !validate_interval) {
0244                 /* Here, we are just trying to find out the
0245                  * first available interval in the next cycle.
0246                  */
0247                 entry_available = true;
0248                 entry_found = entry;
0249                 *interval_start = ktime_add_ns(curr_intv_start, cycle);
0250                 *interval_end = ktime_add_ns(curr_intv_end, cycle);
0251             }
0252         } else if (ktime_before(txtime, earliest_txtime) &&
0253                !entry_available) {
0254             earliest_txtime = txtime;
0255             entry_found = entry;
0256             n = div_s64(ktime_sub(txtime, curr_intv_start), cycle);
0257             *interval_start = ktime_add(curr_intv_start, n * cycle);
0258             *interval_end = ktime_add(curr_intv_end, n * cycle);
0259         }
0260     }
0261 
0262     return entry_found;
0263 }
0264 
0265 static bool is_valid_interval(struct sk_buff *skb, struct Qdisc *sch)
0266 {
0267     struct taprio_sched *q = qdisc_priv(sch);
0268     struct sched_gate_list *sched, *admin;
0269     ktime_t interval_start, interval_end;
0270     struct sched_entry *entry;
0271 
0272     rcu_read_lock();
0273     sched = rcu_dereference(q->oper_sched);
0274     admin = rcu_dereference(q->admin_sched);
0275 
0276     entry = find_entry_to_transmit(skb, sch, sched, admin, skb->tstamp,
0277                        &interval_start, &interval_end, true);
0278     rcu_read_unlock();
0279 
0280     return entry;
0281 }
0282 
0283 static bool taprio_flags_valid(u32 flags)
0284 {
0285     /* Make sure no other flag bits are set. */
0286     if (flags & ~(TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST |
0287               TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD))
0288         return false;
0289     /* txtime-assist and full offload are mutually exclusive */
0290     if ((flags & TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST) &&
0291         (flags & TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD))
0292         return false;
0293     return true;
0294 }
0295 
0296 /* This returns the tstamp value set by TCP in terms of the set clock. */
0297 static ktime_t get_tcp_tstamp(struct taprio_sched *q, struct sk_buff *skb)
0298 {
0299     unsigned int offset = skb_network_offset(skb);
0300     const struct ipv6hdr *ipv6h;
0301     const struct iphdr *iph;
0302     struct ipv6hdr _ipv6h;
0303 
0304     ipv6h = skb_header_pointer(skb, offset, sizeof(_ipv6h), &_ipv6h);
0305     if (!ipv6h)
0306         return 0;
0307 
0308     if (ipv6h->version == 4) {
0309         iph = (struct iphdr *)ipv6h;
0310         offset += iph->ihl * 4;
0311 
0312         /* special-case 6in4 tunnelling, as that is a common way to get
0313          * v6 connectivity in the home
0314          */
0315         if (iph->protocol == IPPROTO_IPV6) {
0316             ipv6h = skb_header_pointer(skb, offset,
0317                            sizeof(_ipv6h), &_ipv6h);
0318 
0319             if (!ipv6h || ipv6h->nexthdr != IPPROTO_TCP)
0320                 return 0;
0321         } else if (iph->protocol != IPPROTO_TCP) {
0322             return 0;
0323         }
0324     } else if (ipv6h->version == 6 && ipv6h->nexthdr != IPPROTO_TCP) {
0325         return 0;
0326     }
0327 
0328     return taprio_mono_to_any(q, skb->skb_mstamp_ns);
0329 }
0330 
0331 /* There are a few scenarios where we will have to modify the txtime from
0332  * what is read from next_txtime in sched_entry. They are:
0333  * 1. If txtime is in the past,
0334  *    a. The gate for the traffic class is currently open and packet can be
0335  *       transmitted before it closes, schedule the packet right away.
0336  *    b. If the gate corresponding to the traffic class is going to open later
0337  *       in the cycle, set the txtime of packet to the interval start.
0338  * 2. If txtime is in the future, there are packets corresponding to the
0339  *    current traffic class waiting to be transmitted. So, the following
0340  *    possibilities exist:
0341  *    a. We can transmit the packet before the window containing the txtime
0342  *       closes.
0343  *    b. The window might close before the transmission can be completed
0344  *       successfully. So, schedule the packet in the next open window.
0345  */
0346 static long get_packet_txtime(struct sk_buff *skb, struct Qdisc *sch)
0347 {
0348     ktime_t transmit_end_time, interval_end, interval_start, tcp_tstamp;
0349     struct taprio_sched *q = qdisc_priv(sch);
0350     struct sched_gate_list *sched, *admin;
0351     ktime_t minimum_time, now, txtime;
0352     int len, packet_transmit_time;
0353     struct sched_entry *entry;
0354     bool sched_changed;
0355 
0356     now = taprio_get_time(q);
0357     minimum_time = ktime_add_ns(now, q->txtime_delay);
0358 
0359     tcp_tstamp = get_tcp_tstamp(q, skb);
0360     minimum_time = max_t(ktime_t, minimum_time, tcp_tstamp);
0361 
0362     rcu_read_lock();
0363     admin = rcu_dereference(q->admin_sched);
0364     sched = rcu_dereference(q->oper_sched);
0365     if (admin && ktime_after(minimum_time, admin->base_time))
0366         switch_schedules(q, &admin, &sched);
0367 
0368     /* Until the schedule starts, all the queues are open */
0369     if (!sched || ktime_before(minimum_time, sched->base_time)) {
0370         txtime = minimum_time;
0371         goto done;
0372     }
0373 
0374     len = qdisc_pkt_len(skb);
0375     packet_transmit_time = length_to_duration(q, len);
0376 
0377     do {
0378         sched_changed = false;
0379 
0380         entry = find_entry_to_transmit(skb, sch, sched, admin,
0381                            minimum_time,
0382                            &interval_start, &interval_end,
0383                            false);
0384         if (!entry) {
0385             txtime = 0;
0386             goto done;
0387         }
0388 
0389         txtime = entry->next_txtime;
0390         txtime = max_t(ktime_t, txtime, minimum_time);
0391         txtime = max_t(ktime_t, txtime, interval_start);
0392 
0393         if (admin && admin != sched &&
0394             ktime_after(txtime, admin->base_time)) {
0395             sched = admin;
0396             sched_changed = true;
0397             continue;
0398         }
0399 
0400         transmit_end_time = ktime_add(txtime, packet_transmit_time);
0401         minimum_time = transmit_end_time;
0402 
0403         /* Update the txtime of current entry to the next time it's
0404          * interval starts.
0405          */
0406         if (ktime_after(transmit_end_time, interval_end))
0407             entry->next_txtime = ktime_add(interval_start, sched->cycle_time);
0408     } while (sched_changed || ktime_after(transmit_end_time, interval_end));
0409 
0410     entry->next_txtime = transmit_end_time;
0411 
0412 done:
0413     rcu_read_unlock();
0414     return txtime;
0415 }
0416 
0417 static int taprio_enqueue_one(struct sk_buff *skb, struct Qdisc *sch,
0418                   struct Qdisc *child, struct sk_buff **to_free)
0419 {
0420     struct taprio_sched *q = qdisc_priv(sch);
0421 
0422     /* sk_flags are only safe to use on full sockets. */
0423     if (skb->sk && sk_fullsock(skb->sk) && sock_flag(skb->sk, SOCK_TXTIME)) {
0424         if (!is_valid_interval(skb, sch))
0425             return qdisc_drop(skb, sch, to_free);
0426     } else if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
0427         skb->tstamp = get_packet_txtime(skb, sch);
0428         if (!skb->tstamp)
0429             return qdisc_drop(skb, sch, to_free);
0430     }
0431 
0432     qdisc_qstats_backlog_inc(sch, skb);
0433     sch->q.qlen++;
0434 
0435     return qdisc_enqueue(skb, child, to_free);
0436 }
0437 
0438 static int taprio_enqueue(struct sk_buff *skb, struct Qdisc *sch,
0439               struct sk_buff **to_free)
0440 {
0441     struct taprio_sched *q = qdisc_priv(sch);
0442     struct Qdisc *child;
0443     int queue;
0444 
0445     if (unlikely(FULL_OFFLOAD_IS_ENABLED(q->flags))) {
0446         WARN_ONCE(1, "Trying to enqueue skb into the root of a taprio qdisc configured with full offload\n");
0447         return qdisc_drop(skb, sch, to_free);
0448     }
0449 
0450     queue = skb_get_queue_mapping(skb);
0451 
0452     child = q->qdiscs[queue];
0453     if (unlikely(!child))
0454         return qdisc_drop(skb, sch, to_free);
0455 
0456     /* Large packets might not be transmitted when the transmission duration
0457      * exceeds any configured interval. Therefore, segment the skb into
0458      * smaller chunks. Skip it for the full offload case, as the driver
0459      * and/or the hardware is expected to handle this.
0460      */
0461     if (skb_is_gso(skb) && !FULL_OFFLOAD_IS_ENABLED(q->flags)) {
0462         unsigned int slen = 0, numsegs = 0, len = qdisc_pkt_len(skb);
0463         netdev_features_t features = netif_skb_features(skb);
0464         struct sk_buff *segs, *nskb;
0465         int ret;
0466 
0467         segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
0468         if (IS_ERR_OR_NULL(segs))
0469             return qdisc_drop(skb, sch, to_free);
0470 
0471         skb_list_walk_safe(segs, segs, nskb) {
0472             skb_mark_not_on_list(segs);
0473             qdisc_skb_cb(segs)->pkt_len = segs->len;
0474             slen += segs->len;
0475 
0476             ret = taprio_enqueue_one(segs, sch, child, to_free);
0477             if (ret != NET_XMIT_SUCCESS) {
0478                 if (net_xmit_drop_count(ret))
0479                     qdisc_qstats_drop(sch);
0480             } else {
0481                 numsegs++;
0482             }
0483         }
0484 
0485         if (numsegs > 1)
0486             qdisc_tree_reduce_backlog(sch, 1 - numsegs, len - slen);
0487         consume_skb(skb);
0488 
0489         return numsegs > 0 ? NET_XMIT_SUCCESS : NET_XMIT_DROP;
0490     }
0491 
0492     return taprio_enqueue_one(skb, sch, child, to_free);
0493 }
0494 
0495 static struct sk_buff *taprio_peek_soft(struct Qdisc *sch)
0496 {
0497     struct taprio_sched *q = qdisc_priv(sch);
0498     struct net_device *dev = qdisc_dev(sch);
0499     struct sched_entry *entry;
0500     struct sk_buff *skb;
0501     u32 gate_mask;
0502     int i;
0503 
0504     rcu_read_lock();
0505     entry = rcu_dereference(q->current_entry);
0506     gate_mask = entry ? entry->gate_mask : TAPRIO_ALL_GATES_OPEN;
0507     rcu_read_unlock();
0508 
0509     if (!gate_mask)
0510         return NULL;
0511 
0512     for (i = 0; i < dev->num_tx_queues; i++) {
0513         struct Qdisc *child = q->qdiscs[i];
0514         int prio;
0515         u8 tc;
0516 
0517         if (unlikely(!child))
0518             continue;
0519 
0520         skb = child->ops->peek(child);
0521         if (!skb)
0522             continue;
0523 
0524         if (TXTIME_ASSIST_IS_ENABLED(q->flags))
0525             return skb;
0526 
0527         prio = skb->priority;
0528         tc = netdev_get_prio_tc_map(dev, prio);
0529 
0530         if (!(gate_mask & BIT(tc)))
0531             continue;
0532 
0533         return skb;
0534     }
0535 
0536     return NULL;
0537 }
0538 
0539 static struct sk_buff *taprio_peek_offload(struct Qdisc *sch)
0540 {
0541     WARN_ONCE(1, "Trying to peek into the root of a taprio qdisc configured with full offload\n");
0542 
0543     return NULL;
0544 }
0545 
0546 static struct sk_buff *taprio_peek(struct Qdisc *sch)
0547 {
0548     struct taprio_sched *q = qdisc_priv(sch);
0549 
0550     return q->peek(sch);
0551 }
0552 
0553 static void taprio_set_budget(struct taprio_sched *q, struct sched_entry *entry)
0554 {
0555     atomic_set(&entry->budget,
0556            div64_u64((u64)entry->interval * PSEC_PER_NSEC,
0557                  atomic64_read(&q->picos_per_byte)));
0558 }
0559 
0560 static struct sk_buff *taprio_dequeue_soft(struct Qdisc *sch)
0561 {
0562     struct taprio_sched *q = qdisc_priv(sch);
0563     struct net_device *dev = qdisc_dev(sch);
0564     struct sk_buff *skb = NULL;
0565     struct sched_entry *entry;
0566     u32 gate_mask;
0567     int i;
0568 
0569     rcu_read_lock();
0570     entry = rcu_dereference(q->current_entry);
0571     /* if there's no entry, it means that the schedule didn't
0572      * start yet, so force all gates to be open, this is in
0573      * accordance to IEEE 802.1Qbv-2015 Section 8.6.9.4.5
0574      * "AdminGateStates"
0575      */
0576     gate_mask = entry ? entry->gate_mask : TAPRIO_ALL_GATES_OPEN;
0577 
0578     if (!gate_mask)
0579         goto done;
0580 
0581     for (i = 0; i < dev->num_tx_queues; i++) {
0582         struct Qdisc *child = q->qdiscs[i];
0583         ktime_t guard;
0584         int prio;
0585         int len;
0586         u8 tc;
0587 
0588         if (unlikely(!child))
0589             continue;
0590 
0591         if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
0592             skb = child->ops->dequeue(child);
0593             if (!skb)
0594                 continue;
0595             goto skb_found;
0596         }
0597 
0598         skb = child->ops->peek(child);
0599         if (!skb)
0600             continue;
0601 
0602         prio = skb->priority;
0603         tc = netdev_get_prio_tc_map(dev, prio);
0604 
0605         if (!(gate_mask & BIT(tc))) {
0606             skb = NULL;
0607             continue;
0608         }
0609 
0610         len = qdisc_pkt_len(skb);
0611         guard = ktime_add_ns(taprio_get_time(q),
0612                      length_to_duration(q, len));
0613 
0614         /* In the case that there's no gate entry, there's no
0615          * guard band ...
0616          */
0617         if (gate_mask != TAPRIO_ALL_GATES_OPEN &&
0618             ktime_after(guard, entry->close_time)) {
0619             skb = NULL;
0620             continue;
0621         }
0622 
0623         /* ... and no budget. */
0624         if (gate_mask != TAPRIO_ALL_GATES_OPEN &&
0625             atomic_sub_return(len, &entry->budget) < 0) {
0626             skb = NULL;
0627             continue;
0628         }
0629 
0630         skb = child->ops->dequeue(child);
0631         if (unlikely(!skb))
0632             goto done;
0633 
0634 skb_found:
0635         qdisc_bstats_update(sch, skb);
0636         qdisc_qstats_backlog_dec(sch, skb);
0637         sch->q.qlen--;
0638 
0639         goto done;
0640     }
0641 
0642 done:
0643     rcu_read_unlock();
0644 
0645     return skb;
0646 }
0647 
0648 static struct sk_buff *taprio_dequeue_offload(struct Qdisc *sch)
0649 {
0650     WARN_ONCE(1, "Trying to dequeue from the root of a taprio qdisc configured with full offload\n");
0651 
0652     return NULL;
0653 }
0654 
0655 static struct sk_buff *taprio_dequeue(struct Qdisc *sch)
0656 {
0657     struct taprio_sched *q = qdisc_priv(sch);
0658 
0659     return q->dequeue(sch);
0660 }
0661 
0662 static bool should_restart_cycle(const struct sched_gate_list *oper,
0663                  const struct sched_entry *entry)
0664 {
0665     if (list_is_last(&entry->list, &oper->entries))
0666         return true;
0667 
0668     if (ktime_compare(entry->close_time, oper->cycle_close_time) == 0)
0669         return true;
0670 
0671     return false;
0672 }
0673 
0674 static bool should_change_schedules(const struct sched_gate_list *admin,
0675                     const struct sched_gate_list *oper,
0676                     ktime_t close_time)
0677 {
0678     ktime_t next_base_time, extension_time;
0679 
0680     if (!admin)
0681         return false;
0682 
0683     next_base_time = sched_base_time(admin);
0684 
0685     /* This is the simple case, the close_time would fall after
0686      * the next schedule base_time.
0687      */
0688     if (ktime_compare(next_base_time, close_time) <= 0)
0689         return true;
0690 
0691     /* This is the cycle_time_extension case, if the close_time
0692      * plus the amount that can be extended would fall after the
0693      * next schedule base_time, we can extend the current schedule
0694      * for that amount.
0695      */
0696     extension_time = ktime_add_ns(close_time, oper->cycle_time_extension);
0697 
0698     /* FIXME: the IEEE 802.1Q-2018 Specification isn't clear about
0699      * how precisely the extension should be made. So after
0700      * conformance testing, this logic may change.
0701      */
0702     if (ktime_compare(next_base_time, extension_time) <= 0)
0703         return true;
0704 
0705     return false;
0706 }
0707 
0708 static enum hrtimer_restart advance_sched(struct hrtimer *timer)
0709 {
0710     struct taprio_sched *q = container_of(timer, struct taprio_sched,
0711                           advance_timer);
0712     struct sched_gate_list *oper, *admin;
0713     struct sched_entry *entry, *next;
0714     struct Qdisc *sch = q->root;
0715     ktime_t close_time;
0716 
0717     spin_lock(&q->current_entry_lock);
0718     entry = rcu_dereference_protected(q->current_entry,
0719                       lockdep_is_held(&q->current_entry_lock));
0720     oper = rcu_dereference_protected(q->oper_sched,
0721                      lockdep_is_held(&q->current_entry_lock));
0722     admin = rcu_dereference_protected(q->admin_sched,
0723                       lockdep_is_held(&q->current_entry_lock));
0724 
0725     if (!oper)
0726         switch_schedules(q, &admin, &oper);
0727 
0728     /* This can happen in two cases: 1. this is the very first run
0729      * of this function (i.e. we weren't running any schedule
0730      * previously); 2. The previous schedule just ended. The first
0731      * entry of all schedules are pre-calculated during the
0732      * schedule initialization.
0733      */
0734     if (unlikely(!entry || entry->close_time == oper->base_time)) {
0735         next = list_first_entry(&oper->entries, struct sched_entry,
0736                     list);
0737         close_time = next->close_time;
0738         goto first_run;
0739     }
0740 
0741     if (should_restart_cycle(oper, entry)) {
0742         next = list_first_entry(&oper->entries, struct sched_entry,
0743                     list);
0744         oper->cycle_close_time = ktime_add_ns(oper->cycle_close_time,
0745                               oper->cycle_time);
0746     } else {
0747         next = list_next_entry(entry, list);
0748     }
0749 
0750     close_time = ktime_add_ns(entry->close_time, next->interval);
0751     close_time = min_t(ktime_t, close_time, oper->cycle_close_time);
0752 
0753     if (should_change_schedules(admin, oper, close_time)) {
0754         /* Set things so the next time this runs, the new
0755          * schedule runs.
0756          */
0757         close_time = sched_base_time(admin);
0758         switch_schedules(q, &admin, &oper);
0759     }
0760 
0761     next->close_time = close_time;
0762     taprio_set_budget(q, next);
0763 
0764 first_run:
0765     rcu_assign_pointer(q->current_entry, next);
0766     spin_unlock(&q->current_entry_lock);
0767 
0768     hrtimer_set_expires(&q->advance_timer, close_time);
0769 
0770     rcu_read_lock();
0771     __netif_schedule(sch);
0772     rcu_read_unlock();
0773 
0774     return HRTIMER_RESTART;
0775 }
0776 
0777 static const struct nla_policy entry_policy[TCA_TAPRIO_SCHED_ENTRY_MAX + 1] = {
0778     [TCA_TAPRIO_SCHED_ENTRY_INDEX]     = { .type = NLA_U32 },
0779     [TCA_TAPRIO_SCHED_ENTRY_CMD]       = { .type = NLA_U8 },
0780     [TCA_TAPRIO_SCHED_ENTRY_GATE_MASK] = { .type = NLA_U32 },
0781     [TCA_TAPRIO_SCHED_ENTRY_INTERVAL]  = { .type = NLA_U32 },
0782 };
0783 
0784 static const struct nla_policy taprio_policy[TCA_TAPRIO_ATTR_MAX + 1] = {
0785     [TCA_TAPRIO_ATTR_PRIOMAP]          = {
0786         .len = sizeof(struct tc_mqprio_qopt)
0787     },
0788     [TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST]           = { .type = NLA_NESTED },
0789     [TCA_TAPRIO_ATTR_SCHED_BASE_TIME]            = { .type = NLA_S64 },
0790     [TCA_TAPRIO_ATTR_SCHED_SINGLE_ENTRY]         = { .type = NLA_NESTED },
0791     [TCA_TAPRIO_ATTR_SCHED_CLOCKID]              = { .type = NLA_S32 },
0792     [TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME]           = { .type = NLA_S64 },
0793     [TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION] = { .type = NLA_S64 },
0794     [TCA_TAPRIO_ATTR_FLAGS]                      = { .type = NLA_U32 },
0795     [TCA_TAPRIO_ATTR_TXTIME_DELAY]           = { .type = NLA_U32 },
0796 };
0797 
0798 static int fill_sched_entry(struct taprio_sched *q, struct nlattr **tb,
0799                 struct sched_entry *entry,
0800                 struct netlink_ext_ack *extack)
0801 {
0802     int min_duration = length_to_duration(q, ETH_ZLEN);
0803     u32 interval = 0;
0804 
0805     if (tb[TCA_TAPRIO_SCHED_ENTRY_CMD])
0806         entry->command = nla_get_u8(
0807             tb[TCA_TAPRIO_SCHED_ENTRY_CMD]);
0808 
0809     if (tb[TCA_TAPRIO_SCHED_ENTRY_GATE_MASK])
0810         entry->gate_mask = nla_get_u32(
0811             tb[TCA_TAPRIO_SCHED_ENTRY_GATE_MASK]);
0812 
0813     if (tb[TCA_TAPRIO_SCHED_ENTRY_INTERVAL])
0814         interval = nla_get_u32(
0815             tb[TCA_TAPRIO_SCHED_ENTRY_INTERVAL]);
0816 
0817     /* The interval should allow at least the minimum ethernet
0818      * frame to go out.
0819      */
0820     if (interval < min_duration) {
0821         NL_SET_ERR_MSG(extack, "Invalid interval for schedule entry");
0822         return -EINVAL;
0823     }
0824 
0825     entry->interval = interval;
0826 
0827     return 0;
0828 }
0829 
0830 static int parse_sched_entry(struct taprio_sched *q, struct nlattr *n,
0831                  struct sched_entry *entry, int index,
0832                  struct netlink_ext_ack *extack)
0833 {
0834     struct nlattr *tb[TCA_TAPRIO_SCHED_ENTRY_MAX + 1] = { };
0835     int err;
0836 
0837     err = nla_parse_nested_deprecated(tb, TCA_TAPRIO_SCHED_ENTRY_MAX, n,
0838                       entry_policy, NULL);
0839     if (err < 0) {
0840         NL_SET_ERR_MSG(extack, "Could not parse nested entry");
0841         return -EINVAL;
0842     }
0843 
0844     entry->index = index;
0845 
0846     return fill_sched_entry(q, tb, entry, extack);
0847 }
0848 
0849 static int parse_sched_list(struct taprio_sched *q, struct nlattr *list,
0850                 struct sched_gate_list *sched,
0851                 struct netlink_ext_ack *extack)
0852 {
0853     struct nlattr *n;
0854     int err, rem;
0855     int i = 0;
0856 
0857     if (!list)
0858         return -EINVAL;
0859 
0860     nla_for_each_nested(n, list, rem) {
0861         struct sched_entry *entry;
0862 
0863         if (nla_type(n) != TCA_TAPRIO_SCHED_ENTRY) {
0864             NL_SET_ERR_MSG(extack, "Attribute is not of type 'entry'");
0865             continue;
0866         }
0867 
0868         entry = kzalloc(sizeof(*entry), GFP_KERNEL);
0869         if (!entry) {
0870             NL_SET_ERR_MSG(extack, "Not enough memory for entry");
0871             return -ENOMEM;
0872         }
0873 
0874         err = parse_sched_entry(q, n, entry, i, extack);
0875         if (err < 0) {
0876             kfree(entry);
0877             return err;
0878         }
0879 
0880         list_add_tail(&entry->list, &sched->entries);
0881         i++;
0882     }
0883 
0884     sched->num_entries = i;
0885 
0886     return i;
0887 }
0888 
0889 static int parse_taprio_schedule(struct taprio_sched *q, struct nlattr **tb,
0890                  struct sched_gate_list *new,
0891                  struct netlink_ext_ack *extack)
0892 {
0893     int err = 0;
0894 
0895     if (tb[TCA_TAPRIO_ATTR_SCHED_SINGLE_ENTRY]) {
0896         NL_SET_ERR_MSG(extack, "Adding a single entry is not supported");
0897         return -ENOTSUPP;
0898     }
0899 
0900     if (tb[TCA_TAPRIO_ATTR_SCHED_BASE_TIME])
0901         new->base_time = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_BASE_TIME]);
0902 
0903     if (tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION])
0904         new->cycle_time_extension = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION]);
0905 
0906     if (tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME])
0907         new->cycle_time = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME]);
0908 
0909     if (tb[TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST])
0910         err = parse_sched_list(q, tb[TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST],
0911                        new, extack);
0912     if (err < 0)
0913         return err;
0914 
0915     if (!new->cycle_time) {
0916         struct sched_entry *entry;
0917         ktime_t cycle = 0;
0918 
0919         list_for_each_entry(entry, &new->entries, list)
0920             cycle = ktime_add_ns(cycle, entry->interval);
0921 
0922         if (!cycle) {
0923             NL_SET_ERR_MSG(extack, "'cycle_time' can never be 0");
0924             return -EINVAL;
0925         }
0926 
0927         new->cycle_time = cycle;
0928     }
0929 
0930     return 0;
0931 }
0932 
0933 static int taprio_parse_mqprio_opt(struct net_device *dev,
0934                    struct tc_mqprio_qopt *qopt,
0935                    struct netlink_ext_ack *extack,
0936                    u32 taprio_flags)
0937 {
0938     int i, j;
0939 
0940     if (!qopt && !dev->num_tc) {
0941         NL_SET_ERR_MSG(extack, "'mqprio' configuration is necessary");
0942         return -EINVAL;
0943     }
0944 
0945     /* If num_tc is already set, it means that the user already
0946      * configured the mqprio part
0947      */
0948     if (dev->num_tc)
0949         return 0;
0950 
0951     /* Verify num_tc is not out of max range */
0952     if (qopt->num_tc > TC_MAX_QUEUE) {
0953         NL_SET_ERR_MSG(extack, "Number of traffic classes is outside valid range");
0954         return -EINVAL;
0955     }
0956 
0957     /* taprio imposes that traffic classes map 1:n to tx queues */
0958     if (qopt->num_tc > dev->num_tx_queues) {
0959         NL_SET_ERR_MSG(extack, "Number of traffic classes is greater than number of HW queues");
0960         return -EINVAL;
0961     }
0962 
0963     /* Verify priority mapping uses valid tcs */
0964     for (i = 0; i <= TC_BITMASK; i++) {
0965         if (qopt->prio_tc_map[i] >= qopt->num_tc) {
0966             NL_SET_ERR_MSG(extack, "Invalid traffic class in priority to traffic class mapping");
0967             return -EINVAL;
0968         }
0969     }
0970 
0971     for (i = 0; i < qopt->num_tc; i++) {
0972         unsigned int last = qopt->offset[i] + qopt->count[i];
0973 
0974         /* Verify the queue count is in tx range being equal to the
0975          * real_num_tx_queues indicates the last queue is in use.
0976          */
0977         if (qopt->offset[i] >= dev->num_tx_queues ||
0978             !qopt->count[i] ||
0979             last > dev->real_num_tx_queues) {
0980             NL_SET_ERR_MSG(extack, "Invalid queue in traffic class to queue mapping");
0981             return -EINVAL;
0982         }
0983 
0984         if (TXTIME_ASSIST_IS_ENABLED(taprio_flags))
0985             continue;
0986 
0987         /* Verify that the offset and counts do not overlap */
0988         for (j = i + 1; j < qopt->num_tc; j++) {
0989             if (last > qopt->offset[j]) {
0990                 NL_SET_ERR_MSG(extack, "Detected overlap in the traffic class to queue mapping");
0991                 return -EINVAL;
0992             }
0993         }
0994     }
0995 
0996     return 0;
0997 }
0998 
0999 static int taprio_get_start_time(struct Qdisc *sch,
1000                  struct sched_gate_list *sched,
1001                  ktime_t *start)
1002 {
1003     struct taprio_sched *q = qdisc_priv(sch);
1004     ktime_t now, base, cycle;
1005     s64 n;
1006 
1007     base = sched_base_time(sched);
1008     now = taprio_get_time(q);
1009 
1010     if (ktime_after(base, now)) {
1011         *start = base;
1012         return 0;
1013     }
1014 
1015     cycle = sched->cycle_time;
1016 
1017     /* The qdisc is expected to have at least one sched_entry.  Moreover,
1018      * any entry must have 'interval' > 0. Thus if the cycle time is zero,
1019      * something went really wrong. In that case, we should warn about this
1020      * inconsistent state and return error.
1021      */
1022     if (WARN_ON(!cycle))
1023         return -EFAULT;
1024 
1025     /* Schedule the start time for the beginning of the next
1026      * cycle.
1027      */
1028     n = div64_s64(ktime_sub_ns(now, base), cycle);
1029     *start = ktime_add_ns(base, (n + 1) * cycle);
1030     return 0;
1031 }
1032 
1033 static void setup_first_close_time(struct taprio_sched *q,
1034                    struct sched_gate_list *sched, ktime_t base)
1035 {
1036     struct sched_entry *first;
1037     ktime_t cycle;
1038 
1039     first = list_first_entry(&sched->entries,
1040                  struct sched_entry, list);
1041 
1042     cycle = sched->cycle_time;
1043 
1044     /* FIXME: find a better place to do this */
1045     sched->cycle_close_time = ktime_add_ns(base, cycle);
1046 
1047     first->close_time = ktime_add_ns(base, first->interval);
1048     taprio_set_budget(q, first);
1049     rcu_assign_pointer(q->current_entry, NULL);
1050 }
1051 
1052 static void taprio_start_sched(struct Qdisc *sch,
1053                    ktime_t start, struct sched_gate_list *new)
1054 {
1055     struct taprio_sched *q = qdisc_priv(sch);
1056     ktime_t expires;
1057 
1058     if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1059         return;
1060 
1061     expires = hrtimer_get_expires(&q->advance_timer);
1062     if (expires == 0)
1063         expires = KTIME_MAX;
1064 
1065     /* If the new schedule starts before the next expiration, we
1066      * reprogram it to the earliest one, so we change the admin
1067      * schedule to the operational one at the right time.
1068      */
1069     start = min_t(ktime_t, start, expires);
1070 
1071     hrtimer_start(&q->advance_timer, start, HRTIMER_MODE_ABS);
1072 }
1073 
1074 static void taprio_set_picos_per_byte(struct net_device *dev,
1075                       struct taprio_sched *q)
1076 {
1077     struct ethtool_link_ksettings ecmd;
1078     int speed = SPEED_10;
1079     int picos_per_byte;
1080     int err;
1081 
1082     err = __ethtool_get_link_ksettings(dev, &ecmd);
1083     if (err < 0)
1084         goto skip;
1085 
1086     if (ecmd.base.speed && ecmd.base.speed != SPEED_UNKNOWN)
1087         speed = ecmd.base.speed;
1088 
1089 skip:
1090     picos_per_byte = (USEC_PER_SEC * 8) / speed;
1091 
1092     atomic64_set(&q->picos_per_byte, picos_per_byte);
1093     netdev_dbg(dev, "taprio: set %s's picos_per_byte to: %lld, linkspeed: %d\n",
1094            dev->name, (long long)atomic64_read(&q->picos_per_byte),
1095            ecmd.base.speed);
1096 }
1097 
1098 static int taprio_dev_notifier(struct notifier_block *nb, unsigned long event,
1099                    void *ptr)
1100 {
1101     struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1102     struct net_device *qdev;
1103     struct taprio_sched *q;
1104     bool found = false;
1105 
1106     ASSERT_RTNL();
1107 
1108     if (event != NETDEV_UP && event != NETDEV_CHANGE)
1109         return NOTIFY_DONE;
1110 
1111     spin_lock(&taprio_list_lock);
1112     list_for_each_entry(q, &taprio_list, taprio_list) {
1113         qdev = qdisc_dev(q->root);
1114         if (qdev == dev) {
1115             found = true;
1116             break;
1117         }
1118     }
1119     spin_unlock(&taprio_list_lock);
1120 
1121     if (found)
1122         taprio_set_picos_per_byte(dev, q);
1123 
1124     return NOTIFY_DONE;
1125 }
1126 
1127 static void setup_txtime(struct taprio_sched *q,
1128              struct sched_gate_list *sched, ktime_t base)
1129 {
1130     struct sched_entry *entry;
1131     u32 interval = 0;
1132 
1133     list_for_each_entry(entry, &sched->entries, list) {
1134         entry->next_txtime = ktime_add_ns(base, interval);
1135         interval += entry->interval;
1136     }
1137 }
1138 
1139 static struct tc_taprio_qopt_offload *taprio_offload_alloc(int num_entries)
1140 {
1141     struct __tc_taprio_qopt_offload *__offload;
1142 
1143     __offload = kzalloc(struct_size(__offload, offload.entries, num_entries),
1144                 GFP_KERNEL);
1145     if (!__offload)
1146         return NULL;
1147 
1148     refcount_set(&__offload->users, 1);
1149 
1150     return &__offload->offload;
1151 }
1152 
1153 struct tc_taprio_qopt_offload *taprio_offload_get(struct tc_taprio_qopt_offload
1154                           *offload)
1155 {
1156     struct __tc_taprio_qopt_offload *__offload;
1157 
1158     __offload = container_of(offload, struct __tc_taprio_qopt_offload,
1159                  offload);
1160 
1161     refcount_inc(&__offload->users);
1162 
1163     return offload;
1164 }
1165 EXPORT_SYMBOL_GPL(taprio_offload_get);
1166 
1167 void taprio_offload_free(struct tc_taprio_qopt_offload *offload)
1168 {
1169     struct __tc_taprio_qopt_offload *__offload;
1170 
1171     __offload = container_of(offload, struct __tc_taprio_qopt_offload,
1172                  offload);
1173 
1174     if (!refcount_dec_and_test(&__offload->users))
1175         return;
1176 
1177     kfree(__offload);
1178 }
1179 EXPORT_SYMBOL_GPL(taprio_offload_free);
1180 
1181 /* The function will only serve to keep the pointers to the "oper" and "admin"
1182  * schedules valid in relation to their base times, so when calling dump() the
1183  * users looks at the right schedules.
1184  * When using full offload, the admin configuration is promoted to oper at the
1185  * base_time in the PHC time domain.  But because the system time is not
1186  * necessarily in sync with that, we can't just trigger a hrtimer to call
1187  * switch_schedules at the right hardware time.
1188  * At the moment we call this by hand right away from taprio, but in the future
1189  * it will be useful to create a mechanism for drivers to notify taprio of the
1190  * offload state (PENDING, ACTIVE, INACTIVE) so it can be visible in dump().
1191  * This is left as TODO.
1192  */
1193 static void taprio_offload_config_changed(struct taprio_sched *q)
1194 {
1195     struct sched_gate_list *oper, *admin;
1196 
1197     spin_lock(&q->current_entry_lock);
1198 
1199     oper = rcu_dereference_protected(q->oper_sched,
1200                      lockdep_is_held(&q->current_entry_lock));
1201     admin = rcu_dereference_protected(q->admin_sched,
1202                       lockdep_is_held(&q->current_entry_lock));
1203 
1204     switch_schedules(q, &admin, &oper);
1205 
1206     spin_unlock(&q->current_entry_lock);
1207 }
1208 
1209 static u32 tc_map_to_queue_mask(struct net_device *dev, u32 tc_mask)
1210 {
1211     u32 i, queue_mask = 0;
1212 
1213     for (i = 0; i < dev->num_tc; i++) {
1214         u32 offset, count;
1215 
1216         if (!(tc_mask & BIT(i)))
1217             continue;
1218 
1219         offset = dev->tc_to_txq[i].offset;
1220         count = dev->tc_to_txq[i].count;
1221 
1222         queue_mask |= GENMASK(offset + count - 1, offset);
1223     }
1224 
1225     return queue_mask;
1226 }
1227 
1228 static void taprio_sched_to_offload(struct net_device *dev,
1229                     struct sched_gate_list *sched,
1230                     struct tc_taprio_qopt_offload *offload)
1231 {
1232     struct sched_entry *entry;
1233     int i = 0;
1234 
1235     offload->base_time = sched->base_time;
1236     offload->cycle_time = sched->cycle_time;
1237     offload->cycle_time_extension = sched->cycle_time_extension;
1238 
1239     list_for_each_entry(entry, &sched->entries, list) {
1240         struct tc_taprio_sched_entry *e = &offload->entries[i];
1241 
1242         e->command = entry->command;
1243         e->interval = entry->interval;
1244         e->gate_mask = tc_map_to_queue_mask(dev, entry->gate_mask);
1245 
1246         i++;
1247     }
1248 
1249     offload->num_entries = i;
1250 }
1251 
1252 static int taprio_enable_offload(struct net_device *dev,
1253                  struct taprio_sched *q,
1254                  struct sched_gate_list *sched,
1255                  struct netlink_ext_ack *extack)
1256 {
1257     const struct net_device_ops *ops = dev->netdev_ops;
1258     struct tc_taprio_qopt_offload *offload;
1259     int err = 0;
1260 
1261     if (!ops->ndo_setup_tc) {
1262         NL_SET_ERR_MSG(extack,
1263                    "Device does not support taprio offload");
1264         return -EOPNOTSUPP;
1265     }
1266 
1267     offload = taprio_offload_alloc(sched->num_entries);
1268     if (!offload) {
1269         NL_SET_ERR_MSG(extack,
1270                    "Not enough memory for enabling offload mode");
1271         return -ENOMEM;
1272     }
1273     offload->enable = 1;
1274     taprio_sched_to_offload(dev, sched, offload);
1275 
1276     err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_TAPRIO, offload);
1277     if (err < 0) {
1278         NL_SET_ERR_MSG(extack,
1279                    "Device failed to setup taprio offload");
1280         goto done;
1281     }
1282 
1283     q->offloaded = true;
1284 
1285 done:
1286     taprio_offload_free(offload);
1287 
1288     return err;
1289 }
1290 
1291 static int taprio_disable_offload(struct net_device *dev,
1292                   struct taprio_sched *q,
1293                   struct netlink_ext_ack *extack)
1294 {
1295     const struct net_device_ops *ops = dev->netdev_ops;
1296     struct tc_taprio_qopt_offload *offload;
1297     int err;
1298 
1299     if (!q->offloaded)
1300         return 0;
1301 
1302     offload = taprio_offload_alloc(0);
1303     if (!offload) {
1304         NL_SET_ERR_MSG(extack,
1305                    "Not enough memory to disable offload mode");
1306         return -ENOMEM;
1307     }
1308     offload->enable = 0;
1309 
1310     err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_TAPRIO, offload);
1311     if (err < 0) {
1312         NL_SET_ERR_MSG(extack,
1313                    "Device failed to disable offload");
1314         goto out;
1315     }
1316 
1317     q->offloaded = false;
1318 
1319 out:
1320     taprio_offload_free(offload);
1321 
1322     return err;
1323 }
1324 
1325 /* If full offload is enabled, the only possible clockid is the net device's
1326  * PHC. For that reason, specifying a clockid through netlink is incorrect.
1327  * For txtime-assist, it is implicitly assumed that the device's PHC is kept
1328  * in sync with the specified clockid via a user space daemon such as phc2sys.
1329  * For both software taprio and txtime-assist, the clockid is used for the
1330  * hrtimer that advances the schedule and hence mandatory.
1331  */
1332 static int taprio_parse_clockid(struct Qdisc *sch, struct nlattr **tb,
1333                 struct netlink_ext_ack *extack)
1334 {
1335     struct taprio_sched *q = qdisc_priv(sch);
1336     struct net_device *dev = qdisc_dev(sch);
1337     int err = -EINVAL;
1338 
1339     if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1340         const struct ethtool_ops *ops = dev->ethtool_ops;
1341         struct ethtool_ts_info info = {
1342             .cmd = ETHTOOL_GET_TS_INFO,
1343             .phc_index = -1,
1344         };
1345 
1346         if (tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]) {
1347             NL_SET_ERR_MSG(extack,
1348                        "The 'clockid' cannot be specified for full offload");
1349             goto out;
1350         }
1351 
1352         if (ops && ops->get_ts_info)
1353             err = ops->get_ts_info(dev, &info);
1354 
1355         if (err || info.phc_index < 0) {
1356             NL_SET_ERR_MSG(extack,
1357                        "Device does not have a PTP clock");
1358             err = -ENOTSUPP;
1359             goto out;
1360         }
1361     } else if (tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]) {
1362         int clockid = nla_get_s32(tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]);
1363         enum tk_offsets tk_offset;
1364 
1365         /* We only support static clockids and we don't allow
1366          * for it to be modified after the first init.
1367          */
1368         if (clockid < 0 ||
1369             (q->clockid != -1 && q->clockid != clockid)) {
1370             NL_SET_ERR_MSG(extack,
1371                        "Changing the 'clockid' of a running schedule is not supported");
1372             err = -ENOTSUPP;
1373             goto out;
1374         }
1375 
1376         switch (clockid) {
1377         case CLOCK_REALTIME:
1378             tk_offset = TK_OFFS_REAL;
1379             break;
1380         case CLOCK_MONOTONIC:
1381             tk_offset = TK_OFFS_MAX;
1382             break;
1383         case CLOCK_BOOTTIME:
1384             tk_offset = TK_OFFS_BOOT;
1385             break;
1386         case CLOCK_TAI:
1387             tk_offset = TK_OFFS_TAI;
1388             break;
1389         default:
1390             NL_SET_ERR_MSG(extack, "Invalid 'clockid'");
1391             err = -EINVAL;
1392             goto out;
1393         }
1394         /* This pairs with READ_ONCE() in taprio_mono_to_any */
1395         WRITE_ONCE(q->tk_offset, tk_offset);
1396 
1397         q->clockid = clockid;
1398     } else {
1399         NL_SET_ERR_MSG(extack, "Specifying a 'clockid' is mandatory");
1400         goto out;
1401     }
1402 
1403     /* Everything went ok, return success. */
1404     err = 0;
1405 
1406 out:
1407     return err;
1408 }
1409 
1410 static int taprio_mqprio_cmp(const struct net_device *dev,
1411                  const struct tc_mqprio_qopt *mqprio)
1412 {
1413     int i;
1414 
1415     if (!mqprio || mqprio->num_tc != dev->num_tc)
1416         return -1;
1417 
1418     for (i = 0; i < mqprio->num_tc; i++)
1419         if (dev->tc_to_txq[i].count != mqprio->count[i] ||
1420             dev->tc_to_txq[i].offset != mqprio->offset[i])
1421             return -1;
1422 
1423     for (i = 0; i <= TC_BITMASK; i++)
1424         if (dev->prio_tc_map[i] != mqprio->prio_tc_map[i])
1425             return -1;
1426 
1427     return 0;
1428 }
1429 
1430 /* The semantics of the 'flags' argument in relation to 'change()'
1431  * requests, are interpreted following two rules (which are applied in
1432  * this order): (1) an omitted 'flags' argument is interpreted as
1433  * zero; (2) the 'flags' of a "running" taprio instance cannot be
1434  * changed.
1435  */
1436 static int taprio_new_flags(const struct nlattr *attr, u32 old,
1437                 struct netlink_ext_ack *extack)
1438 {
1439     u32 new = 0;
1440 
1441     if (attr)
1442         new = nla_get_u32(attr);
1443 
1444     if (old != TAPRIO_FLAGS_INVALID && old != new) {
1445         NL_SET_ERR_MSG_MOD(extack, "Changing 'flags' of a running schedule is not supported");
1446         return -EOPNOTSUPP;
1447     }
1448 
1449     if (!taprio_flags_valid(new)) {
1450         NL_SET_ERR_MSG_MOD(extack, "Specified 'flags' are not valid");
1451         return -EINVAL;
1452     }
1453 
1454     return new;
1455 }
1456 
1457 static int taprio_change(struct Qdisc *sch, struct nlattr *opt,
1458              struct netlink_ext_ack *extack)
1459 {
1460     struct nlattr *tb[TCA_TAPRIO_ATTR_MAX + 1] = { };
1461     struct sched_gate_list *oper, *admin, *new_admin;
1462     struct taprio_sched *q = qdisc_priv(sch);
1463     struct net_device *dev = qdisc_dev(sch);
1464     struct tc_mqprio_qopt *mqprio = NULL;
1465     unsigned long flags;
1466     ktime_t start;
1467     int i, err;
1468 
1469     err = nla_parse_nested_deprecated(tb, TCA_TAPRIO_ATTR_MAX, opt,
1470                       taprio_policy, extack);
1471     if (err < 0)
1472         return err;
1473 
1474     if (tb[TCA_TAPRIO_ATTR_PRIOMAP])
1475         mqprio = nla_data(tb[TCA_TAPRIO_ATTR_PRIOMAP]);
1476 
1477     err = taprio_new_flags(tb[TCA_TAPRIO_ATTR_FLAGS],
1478                    q->flags, extack);
1479     if (err < 0)
1480         return err;
1481 
1482     q->flags = err;
1483 
1484     err = taprio_parse_mqprio_opt(dev, mqprio, extack, q->flags);
1485     if (err < 0)
1486         return err;
1487 
1488     new_admin = kzalloc(sizeof(*new_admin), GFP_KERNEL);
1489     if (!new_admin) {
1490         NL_SET_ERR_MSG(extack, "Not enough memory for a new schedule");
1491         return -ENOMEM;
1492     }
1493     INIT_LIST_HEAD(&new_admin->entries);
1494 
1495     rcu_read_lock();
1496     oper = rcu_dereference(q->oper_sched);
1497     admin = rcu_dereference(q->admin_sched);
1498     rcu_read_unlock();
1499 
1500     /* no changes - no new mqprio settings */
1501     if (!taprio_mqprio_cmp(dev, mqprio))
1502         mqprio = NULL;
1503 
1504     if (mqprio && (oper || admin)) {
1505         NL_SET_ERR_MSG(extack, "Changing the traffic mapping of a running schedule is not supported");
1506         err = -ENOTSUPP;
1507         goto free_sched;
1508     }
1509 
1510     err = parse_taprio_schedule(q, tb, new_admin, extack);
1511     if (err < 0)
1512         goto free_sched;
1513 
1514     if (new_admin->num_entries == 0) {
1515         NL_SET_ERR_MSG(extack, "There should be at least one entry in the schedule");
1516         err = -EINVAL;
1517         goto free_sched;
1518     }
1519 
1520     err = taprio_parse_clockid(sch, tb, extack);
1521     if (err < 0)
1522         goto free_sched;
1523 
1524     taprio_set_picos_per_byte(dev, q);
1525 
1526     if (mqprio) {
1527         err = netdev_set_num_tc(dev, mqprio->num_tc);
1528         if (err)
1529             goto free_sched;
1530         for (i = 0; i < mqprio->num_tc; i++)
1531             netdev_set_tc_queue(dev, i,
1532                         mqprio->count[i],
1533                         mqprio->offset[i]);
1534 
1535         /* Always use supplied priority mappings */
1536         for (i = 0; i <= TC_BITMASK; i++)
1537             netdev_set_prio_tc_map(dev, i,
1538                            mqprio->prio_tc_map[i]);
1539     }
1540 
1541     if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1542         err = taprio_enable_offload(dev, q, new_admin, extack);
1543     else
1544         err = taprio_disable_offload(dev, q, extack);
1545     if (err)
1546         goto free_sched;
1547 
1548     /* Protects against enqueue()/dequeue() */
1549     spin_lock_bh(qdisc_lock(sch));
1550 
1551     if (tb[TCA_TAPRIO_ATTR_TXTIME_DELAY]) {
1552         if (!TXTIME_ASSIST_IS_ENABLED(q->flags)) {
1553             NL_SET_ERR_MSG_MOD(extack, "txtime-delay can only be set when txtime-assist mode is enabled");
1554             err = -EINVAL;
1555             goto unlock;
1556         }
1557 
1558         q->txtime_delay = nla_get_u32(tb[TCA_TAPRIO_ATTR_TXTIME_DELAY]);
1559     }
1560 
1561     if (!TXTIME_ASSIST_IS_ENABLED(q->flags) &&
1562         !FULL_OFFLOAD_IS_ENABLED(q->flags) &&
1563         !hrtimer_active(&q->advance_timer)) {
1564         hrtimer_init(&q->advance_timer, q->clockid, HRTIMER_MODE_ABS);
1565         q->advance_timer.function = advance_sched;
1566     }
1567 
1568     if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1569         q->dequeue = taprio_dequeue_offload;
1570         q->peek = taprio_peek_offload;
1571     } else {
1572         /* Be sure to always keep the function pointers
1573          * in a consistent state.
1574          */
1575         q->dequeue = taprio_dequeue_soft;
1576         q->peek = taprio_peek_soft;
1577     }
1578 
1579     err = taprio_get_start_time(sch, new_admin, &start);
1580     if (err < 0) {
1581         NL_SET_ERR_MSG(extack, "Internal error: failed get start time");
1582         goto unlock;
1583     }
1584 
1585     setup_txtime(q, new_admin, start);
1586 
1587     if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
1588         if (!oper) {
1589             rcu_assign_pointer(q->oper_sched, new_admin);
1590             err = 0;
1591             new_admin = NULL;
1592             goto unlock;
1593         }
1594 
1595         rcu_assign_pointer(q->admin_sched, new_admin);
1596         if (admin)
1597             call_rcu(&admin->rcu, taprio_free_sched_cb);
1598     } else {
1599         setup_first_close_time(q, new_admin, start);
1600 
1601         /* Protects against advance_sched() */
1602         spin_lock_irqsave(&q->current_entry_lock, flags);
1603 
1604         taprio_start_sched(sch, start, new_admin);
1605 
1606         rcu_assign_pointer(q->admin_sched, new_admin);
1607         if (admin)
1608             call_rcu(&admin->rcu, taprio_free_sched_cb);
1609 
1610         spin_unlock_irqrestore(&q->current_entry_lock, flags);
1611 
1612         if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1613             taprio_offload_config_changed(q);
1614     }
1615 
1616     new_admin = NULL;
1617     err = 0;
1618 
1619 unlock:
1620     spin_unlock_bh(qdisc_lock(sch));
1621 
1622 free_sched:
1623     if (new_admin)
1624         call_rcu(&new_admin->rcu, taprio_free_sched_cb);
1625 
1626     return err;
1627 }
1628 
1629 static void taprio_reset(struct Qdisc *sch)
1630 {
1631     struct taprio_sched *q = qdisc_priv(sch);
1632     struct net_device *dev = qdisc_dev(sch);
1633     int i;
1634 
1635     hrtimer_cancel(&q->advance_timer);
1636     if (q->qdiscs) {
1637         for (i = 0; i < dev->num_tx_queues; i++)
1638             if (q->qdiscs[i])
1639                 qdisc_reset(q->qdiscs[i]);
1640     }
1641     sch->qstats.backlog = 0;
1642     sch->q.qlen = 0;
1643 }
1644 
1645 static void taprio_destroy(struct Qdisc *sch)
1646 {
1647     struct taprio_sched *q = qdisc_priv(sch);
1648     struct net_device *dev = qdisc_dev(sch);
1649     unsigned int i;
1650 
1651     spin_lock(&taprio_list_lock);
1652     list_del(&q->taprio_list);
1653     spin_unlock(&taprio_list_lock);
1654 
1655     /* Note that taprio_reset() might not be called if an error
1656      * happens in qdisc_create(), after taprio_init() has been called.
1657      */
1658     hrtimer_cancel(&q->advance_timer);
1659 
1660     taprio_disable_offload(dev, q, NULL);
1661 
1662     if (q->qdiscs) {
1663         for (i = 0; i < dev->num_tx_queues; i++)
1664             qdisc_put(q->qdiscs[i]);
1665 
1666         kfree(q->qdiscs);
1667     }
1668     q->qdiscs = NULL;
1669 
1670     netdev_reset_tc(dev);
1671 
1672     if (q->oper_sched)
1673         call_rcu(&q->oper_sched->rcu, taprio_free_sched_cb);
1674 
1675     if (q->admin_sched)
1676         call_rcu(&q->admin_sched->rcu, taprio_free_sched_cb);
1677 }
1678 
1679 static int taprio_init(struct Qdisc *sch, struct nlattr *opt,
1680                struct netlink_ext_ack *extack)
1681 {
1682     struct taprio_sched *q = qdisc_priv(sch);
1683     struct net_device *dev = qdisc_dev(sch);
1684     int i;
1685 
1686     spin_lock_init(&q->current_entry_lock);
1687 
1688     hrtimer_init(&q->advance_timer, CLOCK_TAI, HRTIMER_MODE_ABS);
1689     q->advance_timer.function = advance_sched;
1690 
1691     q->dequeue = taprio_dequeue_soft;
1692     q->peek = taprio_peek_soft;
1693 
1694     q->root = sch;
1695 
1696     /* We only support static clockids. Use an invalid value as default
1697      * and get the valid one on taprio_change().
1698      */
1699     q->clockid = -1;
1700     q->flags = TAPRIO_FLAGS_INVALID;
1701 
1702     spin_lock(&taprio_list_lock);
1703     list_add(&q->taprio_list, &taprio_list);
1704     spin_unlock(&taprio_list_lock);
1705 
1706     if (sch->parent != TC_H_ROOT)
1707         return -EOPNOTSUPP;
1708 
1709     if (!netif_is_multiqueue(dev))
1710         return -EOPNOTSUPP;
1711 
1712     /* pre-allocate qdisc, attachment can't fail */
1713     q->qdiscs = kcalloc(dev->num_tx_queues,
1714                 sizeof(q->qdiscs[0]),
1715                 GFP_KERNEL);
1716 
1717     if (!q->qdiscs)
1718         return -ENOMEM;
1719 
1720     if (!opt)
1721         return -EINVAL;
1722 
1723     for (i = 0; i < dev->num_tx_queues; i++) {
1724         struct netdev_queue *dev_queue;
1725         struct Qdisc *qdisc;
1726 
1727         dev_queue = netdev_get_tx_queue(dev, i);
1728         qdisc = qdisc_create_dflt(dev_queue,
1729                       &pfifo_qdisc_ops,
1730                       TC_H_MAKE(TC_H_MAJ(sch->handle),
1731                             TC_H_MIN(i + 1)),
1732                       extack);
1733         if (!qdisc)
1734             return -ENOMEM;
1735 
1736         if (i < dev->real_num_tx_queues)
1737             qdisc_hash_add(qdisc, false);
1738 
1739         q->qdiscs[i] = qdisc;
1740     }
1741 
1742     return taprio_change(sch, opt, extack);
1743 }
1744 
1745 static void taprio_attach(struct Qdisc *sch)
1746 {
1747     struct taprio_sched *q = qdisc_priv(sch);
1748     struct net_device *dev = qdisc_dev(sch);
1749     unsigned int ntx;
1750 
1751     /* Attach underlying qdisc */
1752     for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
1753         struct Qdisc *qdisc = q->qdiscs[ntx];
1754         struct Qdisc *old;
1755 
1756         if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1757             qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
1758             old = dev_graft_qdisc(qdisc->dev_queue, qdisc);
1759         } else {
1760             old = dev_graft_qdisc(qdisc->dev_queue, sch);
1761             qdisc_refcount_inc(sch);
1762         }
1763         if (old)
1764             qdisc_put(old);
1765     }
1766 
1767     /* access to the child qdiscs is not needed in offload mode */
1768     if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1769         kfree(q->qdiscs);
1770         q->qdiscs = NULL;
1771     }
1772 }
1773 
1774 static struct netdev_queue *taprio_queue_get(struct Qdisc *sch,
1775                          unsigned long cl)
1776 {
1777     struct net_device *dev = qdisc_dev(sch);
1778     unsigned long ntx = cl - 1;
1779 
1780     if (ntx >= dev->num_tx_queues)
1781         return NULL;
1782 
1783     return netdev_get_tx_queue(dev, ntx);
1784 }
1785 
1786 static int taprio_graft(struct Qdisc *sch, unsigned long cl,
1787             struct Qdisc *new, struct Qdisc **old,
1788             struct netlink_ext_ack *extack)
1789 {
1790     struct taprio_sched *q = qdisc_priv(sch);
1791     struct net_device *dev = qdisc_dev(sch);
1792     struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1793 
1794     if (!dev_queue)
1795         return -EINVAL;
1796 
1797     if (dev->flags & IFF_UP)
1798         dev_deactivate(dev);
1799 
1800     if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1801         *old = dev_graft_qdisc(dev_queue, new);
1802     } else {
1803         *old = q->qdiscs[cl - 1];
1804         q->qdiscs[cl - 1] = new;
1805     }
1806 
1807     if (new)
1808         new->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
1809 
1810     if (dev->flags & IFF_UP)
1811         dev_activate(dev);
1812 
1813     return 0;
1814 }
1815 
1816 static int dump_entry(struct sk_buff *msg,
1817               const struct sched_entry *entry)
1818 {
1819     struct nlattr *item;
1820 
1821     item = nla_nest_start_noflag(msg, TCA_TAPRIO_SCHED_ENTRY);
1822     if (!item)
1823         return -ENOSPC;
1824 
1825     if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_INDEX, entry->index))
1826         goto nla_put_failure;
1827 
1828     if (nla_put_u8(msg, TCA_TAPRIO_SCHED_ENTRY_CMD, entry->command))
1829         goto nla_put_failure;
1830 
1831     if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_GATE_MASK,
1832             entry->gate_mask))
1833         goto nla_put_failure;
1834 
1835     if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_INTERVAL,
1836             entry->interval))
1837         goto nla_put_failure;
1838 
1839     return nla_nest_end(msg, item);
1840 
1841 nla_put_failure:
1842     nla_nest_cancel(msg, item);
1843     return -1;
1844 }
1845 
1846 static int dump_schedule(struct sk_buff *msg,
1847              const struct sched_gate_list *root)
1848 {
1849     struct nlattr *entry_list;
1850     struct sched_entry *entry;
1851 
1852     if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_BASE_TIME,
1853             root->base_time, TCA_TAPRIO_PAD))
1854         return -1;
1855 
1856     if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME,
1857             root->cycle_time, TCA_TAPRIO_PAD))
1858         return -1;
1859 
1860     if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION,
1861             root->cycle_time_extension, TCA_TAPRIO_PAD))
1862         return -1;
1863 
1864     entry_list = nla_nest_start_noflag(msg,
1865                        TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST);
1866     if (!entry_list)
1867         goto error_nest;
1868 
1869     list_for_each_entry(entry, &root->entries, list) {
1870         if (dump_entry(msg, entry) < 0)
1871             goto error_nest;
1872     }
1873 
1874     nla_nest_end(msg, entry_list);
1875     return 0;
1876 
1877 error_nest:
1878     nla_nest_cancel(msg, entry_list);
1879     return -1;
1880 }
1881 
1882 static int taprio_dump(struct Qdisc *sch, struct sk_buff *skb)
1883 {
1884     struct taprio_sched *q = qdisc_priv(sch);
1885     struct net_device *dev = qdisc_dev(sch);
1886     struct sched_gate_list *oper, *admin;
1887     struct tc_mqprio_qopt opt = { 0 };
1888     struct nlattr *nest, *sched_nest;
1889     unsigned int i;
1890 
1891     rcu_read_lock();
1892     oper = rcu_dereference(q->oper_sched);
1893     admin = rcu_dereference(q->admin_sched);
1894 
1895     opt.num_tc = netdev_get_num_tc(dev);
1896     memcpy(opt.prio_tc_map, dev->prio_tc_map, sizeof(opt.prio_tc_map));
1897 
1898     for (i = 0; i < netdev_get_num_tc(dev); i++) {
1899         opt.count[i] = dev->tc_to_txq[i].count;
1900         opt.offset[i] = dev->tc_to_txq[i].offset;
1901     }
1902 
1903     nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1904     if (!nest)
1905         goto start_error;
1906 
1907     if (nla_put(skb, TCA_TAPRIO_ATTR_PRIOMAP, sizeof(opt), &opt))
1908         goto options_error;
1909 
1910     if (!FULL_OFFLOAD_IS_ENABLED(q->flags) &&
1911         nla_put_s32(skb, TCA_TAPRIO_ATTR_SCHED_CLOCKID, q->clockid))
1912         goto options_error;
1913 
1914     if (q->flags && nla_put_u32(skb, TCA_TAPRIO_ATTR_FLAGS, q->flags))
1915         goto options_error;
1916 
1917     if (q->txtime_delay &&
1918         nla_put_u32(skb, TCA_TAPRIO_ATTR_TXTIME_DELAY, q->txtime_delay))
1919         goto options_error;
1920 
1921     if (oper && dump_schedule(skb, oper))
1922         goto options_error;
1923 
1924     if (!admin)
1925         goto done;
1926 
1927     sched_nest = nla_nest_start_noflag(skb, TCA_TAPRIO_ATTR_ADMIN_SCHED);
1928     if (!sched_nest)
1929         goto options_error;
1930 
1931     if (dump_schedule(skb, admin))
1932         goto admin_error;
1933 
1934     nla_nest_end(skb, sched_nest);
1935 
1936 done:
1937     rcu_read_unlock();
1938 
1939     return nla_nest_end(skb, nest);
1940 
1941 admin_error:
1942     nla_nest_cancel(skb, sched_nest);
1943 
1944 options_error:
1945     nla_nest_cancel(skb, nest);
1946 
1947 start_error:
1948     rcu_read_unlock();
1949     return -ENOSPC;
1950 }
1951 
1952 static struct Qdisc *taprio_leaf(struct Qdisc *sch, unsigned long cl)
1953 {
1954     struct taprio_sched *q = qdisc_priv(sch);
1955     struct net_device *dev = qdisc_dev(sch);
1956     unsigned int ntx = cl - 1;
1957 
1958     if (ntx >= dev->num_tx_queues)
1959         return NULL;
1960 
1961     return q->qdiscs[ntx];
1962 }
1963 
1964 static unsigned long taprio_find(struct Qdisc *sch, u32 classid)
1965 {
1966     unsigned int ntx = TC_H_MIN(classid);
1967 
1968     if (!taprio_queue_get(sch, ntx))
1969         return 0;
1970     return ntx;
1971 }
1972 
1973 static int taprio_dump_class(struct Qdisc *sch, unsigned long cl,
1974                  struct sk_buff *skb, struct tcmsg *tcm)
1975 {
1976     struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1977 
1978     tcm->tcm_parent = TC_H_ROOT;
1979     tcm->tcm_handle |= TC_H_MIN(cl);
1980     tcm->tcm_info = dev_queue->qdisc_sleeping->handle;
1981 
1982     return 0;
1983 }
1984 
1985 static int taprio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
1986                    struct gnet_dump *d)
1987     __releases(d->lock)
1988     __acquires(d->lock)
1989 {
1990     struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1991 
1992     sch = dev_queue->qdisc_sleeping;
1993     if (gnet_stats_copy_basic(d, NULL, &sch->bstats, true) < 0 ||
1994         qdisc_qstats_copy(d, sch) < 0)
1995         return -1;
1996     return 0;
1997 }
1998 
1999 static void taprio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
2000 {
2001     struct net_device *dev = qdisc_dev(sch);
2002     unsigned long ntx;
2003 
2004     if (arg->stop)
2005         return;
2006 
2007     arg->count = arg->skip;
2008     for (ntx = arg->skip; ntx < dev->num_tx_queues; ntx++) {
2009         if (arg->fn(sch, ntx + 1, arg) < 0) {
2010             arg->stop = 1;
2011             break;
2012         }
2013         arg->count++;
2014     }
2015 }
2016 
2017 static struct netdev_queue *taprio_select_queue(struct Qdisc *sch,
2018                         struct tcmsg *tcm)
2019 {
2020     return taprio_queue_get(sch, TC_H_MIN(tcm->tcm_parent));
2021 }
2022 
2023 static const struct Qdisc_class_ops taprio_class_ops = {
2024     .graft      = taprio_graft,
2025     .leaf       = taprio_leaf,
2026     .find       = taprio_find,
2027     .walk       = taprio_walk,
2028     .dump       = taprio_dump_class,
2029     .dump_stats = taprio_dump_class_stats,
2030     .select_queue   = taprio_select_queue,
2031 };
2032 
2033 static struct Qdisc_ops taprio_qdisc_ops __read_mostly = {
2034     .cl_ops     = &taprio_class_ops,
2035     .id     = "taprio",
2036     .priv_size  = sizeof(struct taprio_sched),
2037     .init       = taprio_init,
2038     .change     = taprio_change,
2039     .destroy    = taprio_destroy,
2040     .reset      = taprio_reset,
2041     .attach     = taprio_attach,
2042     .peek       = taprio_peek,
2043     .dequeue    = taprio_dequeue,
2044     .enqueue    = taprio_enqueue,
2045     .dump       = taprio_dump,
2046     .owner      = THIS_MODULE,
2047 };
2048 
2049 static struct notifier_block taprio_device_notifier = {
2050     .notifier_call = taprio_dev_notifier,
2051 };
2052 
2053 static int __init taprio_module_init(void)
2054 {
2055     int err = register_netdevice_notifier(&taprio_device_notifier);
2056 
2057     if (err)
2058         return err;
2059 
2060     return register_qdisc(&taprio_qdisc_ops);
2061 }
2062 
2063 static void __exit taprio_module_exit(void)
2064 {
2065     unregister_qdisc(&taprio_qdisc_ops);
2066     unregister_netdevice_notifier(&taprio_device_notifier);
2067 }
2068 
2069 module_init(taprio_module_init);
2070 module_exit(taprio_module_exit);
2071 MODULE_LICENSE("GPL");
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!