On Thu, Jan 16, 2020 at 01:58:56PM +0100, Jack Wang wrote: > From: Jack Wang <jinpu.wang@xxxxxxxxxxxxxxx> > > This is main functionality of rtrs-client module, which manages > set of RDMA connections for each rtrs session, does multipathing, > load balancing and failover of RDMA requests. > > Signed-off-by: Danil Kipnis <danil.kipnis@xxxxxxxxxxxxxxx> > Signed-off-by: Jack Wang <jinpu.wang@xxxxxxxxxxxxxxx> > --- > drivers/infiniband/ulp/rtrs/rtrs-clt.c | 2967 ++++++++++++++++++++++++ > 1 file changed, 2967 insertions(+) > create mode 100644 drivers/infiniband/ulp/rtrs/rtrs-clt.c > > diff --git a/drivers/infiniband/ulp/rtrs/rtrs-clt.c b/drivers/infiniband/ulp/rtrs/rtrs-clt.c > new file mode 100644 > index 000000000000..717d19d4d930 > --- /dev/null > +++ b/drivers/infiniband/ulp/rtrs/rtrs-clt.c > @@ -0,0 +1,2967 @@ > +// SPDX-License-Identifier: GPL-2.0-or-later > +/* > + * RDMA Transport Layer > + * > + * Copyright (c) 2014 - 2018 ProfitBricks GmbH. All rights reserved. > + * > + * Copyright (c) 2018 - 2019 1&1 IONOS Cloud GmbH. All rights reserved. > + * > + * Copyright (c) 2019 - 2020 1&1 IONOS SE. All rights reserved. Please no extra lines between Copyright lines. > + */ > + > +#undef pr_fmt > +#define pr_fmt(fmt) KBUILD_MODNAME " L" __stringify(__LINE__) ": " fmt I never understood this pr_fmt() thing, do we really need it? > + > +#include <linux/module.h> > +#include <linux/rculist.h> > +#include <linux/blkdev.h> /* for BLK_MAX_SEGMENT_SIZE */ > + > +#include "rtrs-clt.h" > +#include "rtrs-log.h" > + > +#define RTRS_CONNECT_TIMEOUT_MS 30000 > + > +MODULE_DESCRIPTION("RDMA Transport Client"); > +MODULE_LICENSE("GPL"); > + > +static ushort nr_cons_per_session; > +module_param(nr_cons_per_session, ushort, 0444); > +MODULE_PARM_DESC(nr_cons_per_session, > + "Number of connections per session. (default: nr_cpu_ids)"); > + > +static int retry_cnt = 7; > +module_param_named(retry_cnt, retry_cnt, int, 0644); > +MODULE_PARM_DESC(retry_cnt, > + "Number of times to send the message if the remote side didn't respond with Ack or Nack (default: 7, min: " > + __stringify(MIN_RTR_CNT) ", max: " > + __stringify(MAX_RTR_CNT) ")"); > + > +static int __read_mostly noreg_cnt; > +module_param_named(noreg_cnt, noreg_cnt, int, 0444); > +MODULE_PARM_DESC(noreg_cnt, > + "Max number of SG entries when MR registration does not happen (default: 0)"); We don't like modules in new code. > + > +static const struct rtrs_rdma_dev_pd_ops dev_pd_ops; > +static struct rtrs_rdma_dev_pd dev_pd = { > + .ops = &dev_pd_ops > +}; > + > +static struct workqueue_struct *rtrs_wq; > +static struct class *rtrs_clt_dev_class; > + > +static inline bool rtrs_clt_is_connected(const struct rtrs_clt *clt) > +{ > + struct rtrs_clt_sess *sess; > + bool connected = false; > + > + rcu_read_lock(); > + list_for_each_entry_rcu(sess, &clt->paths_list, s.entry) > + connected |= (READ_ONCE(sess->state) == RTRS_CLT_CONNECTED); > + rcu_read_unlock(); > + > + return connected; > +} > + > +static inline struct rtrs_permit * > +__rtrs_get_permit(struct rtrs_clt *clt, enum rtrs_clt_con_type con_type) > +{ > + size_t max_depth = clt->queue_depth; > + struct rtrs_permit *permit; > + int cpu, bit; > + > + /* Combined with cq_vector, we pin the IO to the the cpu it comes */ > + cpu = get_cpu(); > + do { > + bit = find_first_zero_bit(clt->permits_map, max_depth); > + if (unlikely(bit >= max_depth)) { > + put_cpu(); > + return NULL; > + } > + > + } while (unlikely(test_and_set_bit_lock(bit, clt->permits_map))); > + put_cpu(); > + > + permit = GET_PERMIT(clt, bit); > + WARN_ON(permit->mem_id != bit); > + permit->cpu_id = cpu; > + permit->con_type = con_type; > + > + return permit; > +} > + > +static inline void __rtrs_put_permit(struct rtrs_clt *clt, > + struct rtrs_permit *permit) > +{ > + clear_bit_unlock(permit->mem_id, clt->permits_map); > +} > + > +struct rtrs_permit *rtrs_clt_get_permit(struct rtrs_clt *clt, > + enum rtrs_clt_con_type con_type, > + int can_wait) > +{ > + struct rtrs_permit *permit; > + DEFINE_WAIT(wait); > + > + permit = __rtrs_get_permit(clt, con_type); > + if (likely(permit) || !can_wait) > + return permit; > + > + do { > + prepare_to_wait(&clt->permits_wait, &wait, > + TASK_UNINTERRUPTIBLE); > + permit = __rtrs_get_permit(clt, con_type); > + if (likely(permit)) > + break; > + > + io_schedule(); > + } while (1); > + > + finish_wait(&clt->permits_wait, &wait); > + > + return permit; > +} > +EXPORT_SYMBOL(rtrs_clt_get_permit); > + > +void rtrs_clt_put_permit(struct rtrs_clt *clt, struct rtrs_permit *permit) > +{ > + if (WARN_ON(!test_bit(permit->mem_id, clt->permits_map))) > + return; > + > + __rtrs_put_permit(clt, permit); > + > + /* > + * Putting a permit is a barrier, so we will observe > + * new entry in the wait list, no worries. > + */ > + if (waitqueue_active(&clt->permits_wait)) Where do you put permit? Does it include barrier? > + wake_up(&clt->permits_wait); > +} > +EXPORT_SYMBOL(rtrs_clt_put_permit); > + > +struct rtrs_permit *rtrs_permit_from_pdu(void *pdu) > +{ > + return pdu - sizeof(struct rtrs_permit); C standard doesn't allow pointer arithmetic on void*. > +} > +EXPORT_SYMBOL(rtrs_permit_from_pdu); > + > +void *rtrs_permit_to_pdu(struct rtrs_permit *permit) > +{ > + return permit + 1; > +} > +EXPORT_SYMBOL(rtrs_permit_to_pdu); > + > +/** > + * rtrs_permit_to_clt_con() - returns RDMA connection pointer by the permit > + * @sess: client session pointer > + * @permit: permit for the allocation of the RDMA buffer > + * Note: > + * IO connection starts from 1. > + * 0 connection is for user messages. > + */ > +static > +struct rtrs_clt_con *rtrs_permit_to_clt_con(struct rtrs_clt_sess *sess, > + struct rtrs_permit *permit) > +{ > + int id = 0; > + > + if (likely(permit->con_type == RTRS_IO_CON)) > + id = (permit->cpu_id % (sess->s.con_num - 1)) + 1; > + > + return to_clt_con(sess->s.con[id]); > +} > + > +static bool __rtrs_clt_change_state(struct rtrs_clt_sess *sess, > + enum rtrs_clt_state new_state) > +{ > + enum rtrs_clt_state old_state; > + bool changed = false; > + > + lockdep_assert_held(&sess->state_wq.lock); > + > + old_state = sess->state; > + switch (new_state) { > + case RTRS_CLT_CONNECTING: > + switch (old_state) { Double switch is better to be avoided. > + case RTRS_CLT_RECONNECTING: > + changed = true; > + /* FALLTHRU */ > + default: > + break; > + } > + break; .... Thanks