linux_dsm_epyc7002/net/smc/smc_core.c

715 lines
19 KiB
C
Raw Normal View History

/*
* Shared Memory Communications over RDMA (SMC-R) and RoCE
*
* Basic Transport Functions exploiting Infiniband API
*
* Copyright IBM Corp. 2016
*
* Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
*/
#include <linux/socket.h>
#include <linux/if_vlan.h>
#include <linux/random.h>
#include <linux/workqueue.h>
#include <net/tcp.h>
#include <net/sock.h>
#include <rdma/ib_verbs.h>
#include "smc.h"
#include "smc_clc.h"
#include "smc_core.h"
#include "smc_ib.h"
#include "smc_wr.h"
#include "smc_llc.h"
#include "smc_cdc.h"
#include "smc_close.h"
#define SMC_LGR_NUM_INCR 256
#define SMC_LGR_FREE_DELAY_SERV (600 * HZ)
#define SMC_LGR_FREE_DELAY_CLNT (SMC_LGR_FREE_DELAY_SERV + 10)
static u32 smc_lgr_num; /* unique link group number */
/* Register connection's alert token in our lookup structure.
* To use rbtrees we have to implement our own insert core.
* Requires @conns_lock
* @smc connection to register
* Returns 0 on success, != otherwise.
*/
static void smc_lgr_add_alert_token(struct smc_connection *conn)
{
struct rb_node **link, *parent = NULL;
u32 token = conn->alert_token_local;
link = &conn->lgr->conns_all.rb_node;
while (*link) {
struct smc_connection *cur = rb_entry(*link,
struct smc_connection, alert_node);
parent = *link;
if (cur->alert_token_local > token)
link = &parent->rb_left;
else
link = &parent->rb_right;
}
/* Put the new node there */
rb_link_node(&conn->alert_node, parent, link);
rb_insert_color(&conn->alert_node, &conn->lgr->conns_all);
}
/* Register connection in link group by assigning an alert token
* registered in a search tree.
* Requires @conns_lock
* Note that '0' is a reserved value and not assigned.
*/
static void smc_lgr_register_conn(struct smc_connection *conn)
{
struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
static atomic_t nexttoken = ATOMIC_INIT(0);
/* find a new alert_token_local value not yet used by some connection
* in this link group
*/
sock_hold(&smc->sk); /* sock_put in smc_lgr_unregister_conn() */
while (!conn->alert_token_local) {
conn->alert_token_local = atomic_inc_return(&nexttoken);
if (smc_lgr_find_conn(conn->alert_token_local, conn->lgr))
conn->alert_token_local = 0;
}
smc_lgr_add_alert_token(conn);
conn->lgr->conns_num++;
}
/* Unregister connection and reset the alert token of the given connection<
*/
static void __smc_lgr_unregister_conn(struct smc_connection *conn)
{
struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
struct smc_link_group *lgr = conn->lgr;
rb_erase(&conn->alert_node, &lgr->conns_all);
lgr->conns_num--;
conn->alert_token_local = 0;
conn->lgr = NULL;
sock_put(&smc->sk); /* sock_hold in smc_lgr_register_conn() */
}
/* Unregister connection and trigger lgr freeing if applicable
*/
static void smc_lgr_unregister_conn(struct smc_connection *conn)
{
struct smc_link_group *lgr = conn->lgr;
int reduced = 0;
write_lock_bh(&lgr->conns_lock);
if (conn->alert_token_local) {
reduced = 1;
__smc_lgr_unregister_conn(conn);
}
write_unlock_bh(&lgr->conns_lock);
if (!reduced || lgr->conns_num)
return;
/* client link group creation always follows the server link group
* creation. For client use a somewhat higher removal delay time,
* otherwise there is a risk of out-of-sync link groups.
*/
mod_delayed_work(system_wq, &lgr->free_work,
lgr->role == SMC_CLNT ? SMC_LGR_FREE_DELAY_CLNT :
SMC_LGR_FREE_DELAY_SERV);
}
static void smc_lgr_free_work(struct work_struct *work)
{
struct smc_link_group *lgr = container_of(to_delayed_work(work),
struct smc_link_group,
free_work);
bool conns;
spin_lock_bh(&smc_lgr_list.lock);
read_lock_bh(&lgr->conns_lock);
conns = RB_EMPTY_ROOT(&lgr->conns_all);
read_unlock_bh(&lgr->conns_lock);
if (!conns) { /* number of lgr connections is no longer zero */
spin_unlock_bh(&smc_lgr_list.lock);
return;
}
list_del_init(&lgr->list); /* remove from smc_lgr_list */
spin_unlock_bh(&smc_lgr_list.lock);
smc_lgr_free(lgr);
}
/* create a new SMC link group */
static int smc_lgr_create(struct smc_sock *smc, __be32 peer_in_addr,
struct smc_ib_device *smcibdev, u8 ibport,
char *peer_systemid, unsigned short vlan_id)
{
struct smc_link_group *lgr;
struct smc_link *lnk;
u8 rndvec[3];
int rc = 0;
int i;
lgr = kzalloc(sizeof(*lgr), GFP_KERNEL);
if (!lgr) {
rc = -ENOMEM;
goto out;
}
lgr->role = smc->listen_smc ? SMC_SERV : SMC_CLNT;
lgr->sync_err = false;
lgr->daddr = peer_in_addr;
memcpy(lgr->peer_systemid, peer_systemid, SMC_SYSTEMID_LEN);
lgr->vlan_id = vlan_id;
rwlock_init(&lgr->sndbufs_lock);
rwlock_init(&lgr->rmbs_lock);
for (i = 0; i < SMC_RMBE_SIZES; i++) {
INIT_LIST_HEAD(&lgr->sndbufs[i]);
INIT_LIST_HEAD(&lgr->rmbs[i]);
}
smc_lgr_num += SMC_LGR_NUM_INCR;
memcpy(&lgr->id, (u8 *)&smc_lgr_num, SMC_LGR_ID_SIZE);
INIT_DELAYED_WORK(&lgr->free_work, smc_lgr_free_work);
lgr->conns_all = RB_ROOT;
lnk = &lgr->lnk[SMC_SINGLE_LINK];
/* initialize link */
lnk->smcibdev = smcibdev;
lnk->ibport = ibport;
lnk->path_mtu = smcibdev->pattr[ibport - 1].active_mtu;
if (!smcibdev->initialized)
smc_ib_setup_per_ibdev(smcibdev);
get_random_bytes(rndvec, sizeof(rndvec));
lnk->psn_initial = rndvec[0] + (rndvec[1] << 8) + (rndvec[2] << 16);
rc = smc_wr_alloc_link_mem(lnk);
if (rc)
goto free_lgr;
rc = smc_ib_create_protection_domain(lnk);
if (rc)
goto free_link_mem;
rc = smc_ib_create_queue_pair(lnk);
if (rc)
goto dealloc_pd;
rc = smc_wr_create_link(lnk);
if (rc)
goto destroy_qp;
init_completion(&lnk->llc_confirm);
init_completion(&lnk->llc_confirm_resp);
smc->conn.lgr = lgr;
rwlock_init(&lgr->conns_lock);
spin_lock_bh(&smc_lgr_list.lock);
list_add(&lgr->list, &smc_lgr_list.list);
spin_unlock_bh(&smc_lgr_list.lock);
return 0;
destroy_qp:
smc_ib_destroy_queue_pair(lnk);
dealloc_pd:
smc_ib_dealloc_protection_domain(lnk);
free_link_mem:
smc_wr_free_link_mem(lnk);
free_lgr:
kfree(lgr);
out:
return rc;
}
static void smc_buf_unuse(struct smc_connection *conn)
{
if (conn->sndbuf_desc) {
conn->sndbuf_desc->used = 0;
conn->sndbuf_size = 0;
}
if (conn->rmb_desc) {
conn->rmb_desc->reused = true;
conn->rmb_desc->used = 0;
conn->rmbe_size = 0;
}
}
/* remove a finished connection from its link group */
void smc_conn_free(struct smc_connection *conn)
{
struct smc_link_group *lgr = conn->lgr;
if (!lgr)
return;
smc_cdc_tx_dismiss_slots(conn);
smc_lgr_unregister_conn(conn);
smc_buf_unuse(conn);
}
static void smc_link_clear(struct smc_link *lnk)
{
lnk->peer_qpn = 0;
smc_ib_modify_qp_reset(lnk);
smc_wr_free_link(lnk);
smc_ib_destroy_queue_pair(lnk);
smc_ib_dealloc_protection_domain(lnk);
smc_wr_free_link_mem(lnk);
}
static void smc_buf_free(struct smc_buf_desc *buf_desc, struct smc_link *lnk,
bool is_rmb)
{
if (is_rmb) {
if (buf_desc->mr_rx[SMC_SINGLE_LINK])
smc_ib_put_memory_region(
buf_desc->mr_rx[SMC_SINGLE_LINK]);
smc_ib_buf_unmap_sg(lnk->smcibdev, buf_desc,
DMA_FROM_DEVICE);
} else {
smc_ib_buf_unmap_sg(lnk->smcibdev, buf_desc,
DMA_TO_DEVICE);
}
sg_free_table(&buf_desc->sgt[SMC_SINGLE_LINK]);
if (buf_desc->cpu_addr)
free_pages((unsigned long)buf_desc->cpu_addr, buf_desc->order);
kfree(buf_desc);
}
static void __smc_lgr_free_bufs(struct smc_link_group *lgr, bool is_rmb)
{
struct smc_link *lnk = &lgr->lnk[SMC_SINGLE_LINK];
struct smc_buf_desc *buf_desc, *bf_desc;
struct list_head *buf_list;
int i;
for (i = 0; i < SMC_RMBE_SIZES; i++) {
if (is_rmb)
buf_list = &lgr->rmbs[i];
else
buf_list = &lgr->sndbufs[i];
list_for_each_entry_safe(buf_desc, bf_desc, buf_list,
list) {
list_del(&buf_desc->list);
smc_buf_free(buf_desc, lnk, is_rmb);
}
}
}
static void smc_lgr_free_bufs(struct smc_link_group *lgr)
{
/* free send buffers */
__smc_lgr_free_bufs(lgr, false);
/* free rmbs */
__smc_lgr_free_bufs(lgr, true);
}
/* remove a link group */
void smc_lgr_free(struct smc_link_group *lgr)
{
smc_lgr_free_bufs(lgr);
smc_link_clear(&lgr->lnk[SMC_SINGLE_LINK]);
kfree(lgr);
}
/* terminate linkgroup abnormally */
void smc_lgr_terminate(struct smc_link_group *lgr)
{
struct smc_connection *conn;
struct smc_sock *smc;
struct rb_node *node;
spin_lock_bh(&smc_lgr_list.lock);
if (list_empty(&lgr->list)) {
/* termination already triggered */
spin_unlock_bh(&smc_lgr_list.lock);
return;
}
/* do not use this link group for new connections */
list_del_init(&lgr->list);
spin_unlock_bh(&smc_lgr_list.lock);
write_lock_bh(&lgr->conns_lock);
node = rb_first(&lgr->conns_all);
while (node) {
conn = rb_entry(node, struct smc_connection, alert_node);
smc = container_of(conn, struct smc_sock, conn);
sock_hold(&smc->sk);
__smc_lgr_unregister_conn(conn);
schedule_work(&conn->close_work);
sock_put(&smc->sk);
node = rb_first(&lgr->conns_all);
}
write_unlock_bh(&lgr->conns_lock);
}
/* Determine vlan of internal TCP socket.
* @vlan_id: address to store the determined vlan id into
*/
static int smc_vlan_by_tcpsk(struct socket *clcsock, unsigned short *vlan_id)
{
struct dst_entry *dst = sk_dst_get(clcsock->sk);
int rc = 0;
*vlan_id = 0;
if (!dst) {
rc = -ENOTCONN;
goto out;
}
if (!dst->dev) {
rc = -ENODEV;
goto out_rel;
}
if (is_vlan_dev(dst->dev))
*vlan_id = vlan_dev_vlan_id(dst->dev);
out_rel:
dst_release(dst);
out:
return rc;
}
/* determine the link gid matching the vlan id of the link group */
static int smc_link_determine_gid(struct smc_link_group *lgr)
{
struct smc_link *lnk = &lgr->lnk[SMC_SINGLE_LINK];
struct ib_gid_attr gattr;
union ib_gid gid;
int i;
if (!lgr->vlan_id) {
lnk->gid = lnk->smcibdev->gid[lnk->ibport - 1];
return 0;
}
for (i = 0; i < lnk->smcibdev->pattr[lnk->ibport - 1].gid_tbl_len;
i++) {
if (ib_query_gid(lnk->smcibdev->ibdev, lnk->ibport, i, &gid,
&gattr))
continue;
if (gattr.ndev &&
(vlan_dev_vlan_id(gattr.ndev) == lgr->vlan_id)) {
lnk->gid = gid;
return 0;
}
}
return -ENODEV;
}
/* create a new SMC connection (and a new link group if necessary) */
int smc_conn_create(struct smc_sock *smc, __be32 peer_in_addr,
struct smc_ib_device *smcibdev, u8 ibport,
struct smc_clc_msg_local *lcl, int srv_first_contact)
{
struct smc_connection *conn = &smc->conn;
struct smc_link_group *lgr;
unsigned short vlan_id;
enum smc_lgr_role role;
int local_contact = SMC_FIRST_CONTACT;
int rc = 0;
role = smc->listen_smc ? SMC_SERV : SMC_CLNT;
rc = smc_vlan_by_tcpsk(smc->clcsock, &vlan_id);
if (rc)
return rc;
if ((role == SMC_CLNT) && srv_first_contact)
/* create new link group as well */
goto create;
/* determine if an existing link group can be reused */
spin_lock_bh(&smc_lgr_list.lock);
list_for_each_entry(lgr, &smc_lgr_list.list, list) {
write_lock_bh(&lgr->conns_lock);
if (!memcmp(lgr->peer_systemid, lcl->id_for_peer,
SMC_SYSTEMID_LEN) &&
!memcmp(lgr->lnk[SMC_SINGLE_LINK].peer_gid, &lcl->gid,
SMC_GID_SIZE) &&
!memcmp(lgr->lnk[SMC_SINGLE_LINK].peer_mac, lcl->mac,
sizeof(lcl->mac)) &&
!lgr->sync_err &&
(lgr->role == role) &&
(lgr->vlan_id == vlan_id) &&
((role == SMC_CLNT) ||
(lgr->conns_num < SMC_RMBS_PER_LGR_MAX))) {
/* link group found */
local_contact = SMC_REUSE_CONTACT;
conn->lgr = lgr;
smc_lgr_register_conn(conn); /* add smc conn to lgr */
write_unlock_bh(&lgr->conns_lock);
break;
}
write_unlock_bh(&lgr->conns_lock);
}
spin_unlock_bh(&smc_lgr_list.lock);
if (role == SMC_CLNT && !srv_first_contact &&
(local_contact == SMC_FIRST_CONTACT)) {
/* Server reuses a link group, but Client wants to start
* a new one
* send out_of_sync decline, reason synchr. error
*/
return -ENOLINK;
}
create:
if (local_contact == SMC_FIRST_CONTACT) {
rc = smc_lgr_create(smc, peer_in_addr, smcibdev, ibport,
lcl->id_for_peer, vlan_id);
if (rc)
goto out;
smc_lgr_register_conn(conn); /* add smc conn to lgr */
rc = smc_link_determine_gid(conn->lgr);
}
conn->local_tx_ctrl.common.type = SMC_CDC_MSG_TYPE;
conn->local_tx_ctrl.len = sizeof(struct smc_cdc_msg);
#ifndef KERNEL_HAS_ATOMIC64
spin_lock_init(&conn->acurs_lock);
#endif
out:
return rc ? rc : local_contact;
}
/* try to reuse a sndbuf or rmb description slot for a certain
* buffer size; if not available, return NULL
*/
static inline
struct smc_buf_desc *smc_buf_get_slot(struct smc_link_group *lgr,
int compressed_bufsize,
rwlock_t *lock,
struct list_head *buf_list)
{
struct smc_buf_desc *buf_slot;
read_lock_bh(lock);
list_for_each_entry(buf_slot, buf_list, list) {
if (cmpxchg(&buf_slot->used, 0, 1) == 0) {
read_unlock_bh(lock);
return buf_slot;
}
}
read_unlock_bh(lock);
return NULL;
}
/* one of the conditions for announcing a receiver's current window size is
* that it "results in a minimum increase in the window size of 10% of the
* receive buffer space" [RFC7609]
*/
static inline int smc_rmb_wnd_update_limit(int rmbe_size)
{
return min_t(int, rmbe_size / 10, SOCK_MIN_SNDBUF / 2);
}
static struct smc_buf_desc *smc_new_buf_create(struct smc_link_group *lgr,
bool is_rmb, int bufsize)
{
struct smc_buf_desc *buf_desc;
struct smc_link *lnk;
int rc;
/* try to alloc a new buffer */
buf_desc = kzalloc(sizeof(*buf_desc), GFP_KERNEL);
if (!buf_desc)
return ERR_PTR(-ENOMEM);
buf_desc->cpu_addr =
(void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN |
__GFP_NOMEMALLOC |
__GFP_NORETRY | __GFP_ZERO,
get_order(bufsize));
if (!buf_desc->cpu_addr) {
kfree(buf_desc);
return ERR_PTR(-EAGAIN);
}
buf_desc->order = get_order(bufsize);
/* build the sg table from the pages */
lnk = &lgr->lnk[SMC_SINGLE_LINK];
rc = sg_alloc_table(&buf_desc->sgt[SMC_SINGLE_LINK], 1,
GFP_KERNEL);
if (rc) {
smc_buf_free(buf_desc, lnk, is_rmb);
return ERR_PTR(rc);
}
sg_set_buf(buf_desc->sgt[SMC_SINGLE_LINK].sgl,
buf_desc->cpu_addr, bufsize);
/* map sg table to DMA address */
rc = smc_ib_buf_map_sg(lnk->smcibdev, buf_desc,
is_rmb ? DMA_FROM_DEVICE : DMA_TO_DEVICE);
/* SMC protocol depends on mapping to one DMA address only */
if (rc != 1) {
smc_buf_free(buf_desc, lnk, is_rmb);
return ERR_PTR(-EAGAIN);
}
/* create a new memory region for the RMB */
if (is_rmb) {
rc = smc_ib_get_memory_region(lnk->roce_pd,
IB_ACCESS_REMOTE_WRITE |
IB_ACCESS_LOCAL_WRITE,
buf_desc);
if (rc) {
smc_buf_free(buf_desc, lnk, is_rmb);
return ERR_PTR(rc);
}
}
return buf_desc;
}
static int __smc_buf_create(struct smc_sock *smc, bool is_rmb)
{
struct smc_connection *conn = &smc->conn;
struct smc_link_group *lgr = conn->lgr;
struct smc_buf_desc *buf_desc = NULL;
struct list_head *buf_list;
int bufsize, bufsize_short;
int sk_buf_size;
rwlock_t *lock;
if (is_rmb)
/* use socket recv buffer size (w/o overhead) as start value */
sk_buf_size = smc->sk.sk_rcvbuf / 2;
else
/* use socket send buffer size (w/o overhead) as start value */
sk_buf_size = smc->sk.sk_sndbuf / 2;
for (bufsize_short = smc_compress_bufsize(smc->sk.sk_sndbuf / 2);
bufsize_short >= 0; bufsize_short--) {
if (is_rmb) {
lock = &lgr->rmbs_lock;
buf_list = &lgr->rmbs[bufsize_short];
} else {
lock = &lgr->sndbufs_lock;
buf_list = &lgr->sndbufs[bufsize_short];
}
bufsize = smc_uncompress_bufsize(bufsize_short);
if ((1 << get_order(bufsize)) > SG_MAX_SINGLE_ALLOC)
continue;
/* check for reusable slot in the link group */
buf_desc = smc_buf_get_slot(lgr, bufsize_short, lock, buf_list);
if (buf_desc) {
memset(buf_desc->cpu_addr, 0, bufsize);
break; /* found reusable slot */
}
buf_desc = smc_new_buf_create(lgr, is_rmb, bufsize);
if (PTR_ERR(buf_desc) == -ENOMEM)
break;
if (IS_ERR(buf_desc))
continue;
buf_desc->used = 1;
write_lock_bh(lock);
list_add(&buf_desc->list, buf_list);
write_unlock_bh(lock);
break; /* found */
}
if (IS_ERR(buf_desc))
return -ENOMEM;
if (is_rmb) {
conn->rmb_desc = buf_desc;
conn->rmbe_size = bufsize;
conn->rmbe_size_short = bufsize_short;
smc->sk.sk_rcvbuf = bufsize * 2;
atomic_set(&conn->bytes_to_rcv, 0);
conn->rmbe_update_limit = smc_rmb_wnd_update_limit(bufsize);
} else {
conn->sndbuf_desc = buf_desc;
conn->sndbuf_size = bufsize;
smc->sk.sk_sndbuf = bufsize * 2;
atomic_set(&conn->sndbuf_space, bufsize);
}
return 0;
}
void smc_sndbuf_sync_sg_for_cpu(struct smc_connection *conn)
{
struct smc_link_group *lgr = conn->lgr;
smc_ib_sync_sg_for_cpu(lgr->lnk[SMC_SINGLE_LINK].smcibdev,
conn->sndbuf_desc, DMA_TO_DEVICE);
}
void smc_sndbuf_sync_sg_for_device(struct smc_connection *conn)
{
struct smc_link_group *lgr = conn->lgr;
smc_ib_sync_sg_for_device(lgr->lnk[SMC_SINGLE_LINK].smcibdev,
conn->sndbuf_desc, DMA_TO_DEVICE);
}
void smc_rmb_sync_sg_for_cpu(struct smc_connection *conn)
{
struct smc_link_group *lgr = conn->lgr;
smc_ib_sync_sg_for_cpu(lgr->lnk[SMC_SINGLE_LINK].smcibdev,
conn->rmb_desc, DMA_FROM_DEVICE);
}
void smc_rmb_sync_sg_for_device(struct smc_connection *conn)
{
struct smc_link_group *lgr = conn->lgr;
smc_ib_sync_sg_for_device(lgr->lnk[SMC_SINGLE_LINK].smcibdev,
conn->rmb_desc, DMA_FROM_DEVICE);
}
/* create the send and receive buffer for an SMC socket;
* receive buffers are called RMBs;
* (even though the SMC protocol allows more than one RMB-element per RMB,
* the Linux implementation uses just one RMB-element per RMB, i.e. uses an
* extra RMB for every connection in a link group
*/
int smc_buf_create(struct smc_sock *smc)
{
int rc;
/* create send buffer */
rc = __smc_buf_create(smc, false);
if (rc)
return rc;
/* create rmb */
rc = __smc_buf_create(smc, true);
if (rc)
smc_buf_free(smc->conn.sndbuf_desc,
&smc->conn.lgr->lnk[SMC_SINGLE_LINK], false);
return rc;
}
static inline int smc_rmb_reserve_rtoken_idx(struct smc_link_group *lgr)
{
int i;
for_each_clear_bit(i, lgr->rtokens_used_mask, SMC_RMBS_PER_LGR_MAX) {
if (!test_and_set_bit(i, lgr->rtokens_used_mask))
return i;
}
return -ENOSPC;
}
/* save rkey and dma_addr received from peer during clc handshake */
int smc_rmb_rtoken_handling(struct smc_connection *conn,
struct smc_clc_msg_accept_confirm *clc)
{
u64 dma_addr = be64_to_cpu(clc->rmb_dma_addr);
struct smc_link_group *lgr = conn->lgr;
u32 rkey = ntohl(clc->rmb_rkey);
int i;
for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
if ((lgr->rtokens[i][SMC_SINGLE_LINK].rkey == rkey) &&
(lgr->rtokens[i][SMC_SINGLE_LINK].dma_addr == dma_addr) &&
test_bit(i, lgr->rtokens_used_mask)) {
conn->rtoken_idx = i;
return 0;
}
}
conn->rtoken_idx = smc_rmb_reserve_rtoken_idx(lgr);
if (conn->rtoken_idx < 0)
return conn->rtoken_idx;
lgr->rtokens[conn->rtoken_idx][SMC_SINGLE_LINK].rkey = rkey;
lgr->rtokens[conn->rtoken_idx][SMC_SINGLE_LINK].dma_addr = dma_addr;
return 0;
}