* [PATCH net-next 0/8] qed*: Light L2 updates
@ 2017-06-08 16:13 Yuval Mintz
2017-06-08 16:13 ` [PATCH net-next 1/8] qed: LL2 to use packed information for tx Yuval Mintz
` (5 more replies)
0 siblings, 6 replies; 16+ messages in thread
From: Yuval Mintz @ 2017-06-08 16:13 UTC (permalink / raw)
To: netdev, davem; +Cc: linux-rdma, Michal.Kalderon, Yuval Mintz
This series does a major overhaul of the LL2 logic in qed.
The single biggest change done here is in #5 where we're changing
the API qed provides for LL2 [both internally in case of storage and
externally in case of RoCE] to become callback-based to allow cleaner
scalability in preperation to the future iWARP submission which would
aadd additional flavors of LL2. It's also the only patch in series
to modify !qed logic [qedr].
Patches prior to that mostly deal with refactoring LL2 code,
encapsulating varaious parameters into structure and re-ordering
of LL2 code. The latter patches add some small missing bits of LL2
ffunctionality.
Dave,
Please consider applying this series to `net-next'.
Thanks,
Yuval
Michal Kalderon (1):
qed*: LL2 callback operations
Yuval Mintz (7):
qed: LL2 to use packed information for tx
qed: Revise ll2 Rx completion
qed: Cleaner seperation of LL2 inputs
qed: LL2 code relocations
qed: No need for LL2 frags indication
qed: Call rx_release_cb() when flushing LL2
qed: collect GSI port statistics
drivers/infiniband/hw/qedr/main.c | 6 +-
drivers/infiniband/hw/qedr/qedr.h | 2 +
drivers/infiniband/hw/qedr/qedr_cm.c | 238 ++++++--
drivers/net/ethernet/qlogic/qed/qed.h | 1 -
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 882 +++++++++++++++--------------
drivers/net/ethernet/qlogic/qed/qed_ll2.h | 113 +---
drivers/net/ethernet/qlogic/qed/qed_roce.c | 295 +---------
drivers/net/ethernet/qlogic/qed/qed_roce.h | 43 --
include/linux/qed/qed_ll2_if.h | 126 +++++
include/linux/qed/qed_roce_if.h | 80 +--
10 files changed, 842 insertions(+), 944 deletions(-)
--
2.9.4
^ permalink raw reply [flat|nested] 16+ messages in thread
* [PATCH net-next 1/8] qed: LL2 to use packed information for tx
2017-06-08 16:13 [PATCH net-next 0/8] qed*: Light L2 updates Yuval Mintz
@ 2017-06-08 16:13 ` Yuval Mintz
[not found] ` <20170608161323.7597-2-Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
2017-06-08 16:13 ` [PATCH net-next 2/8] qed: Revise ll2 Rx completion Yuval Mintz
` (4 subsequent siblings)
5 siblings, 1 reply; 16+ messages in thread
From: Yuval Mintz @ 2017-06-08 16:13 UTC (permalink / raw)
To: netdev, davem; +Cc: linux-rdma, Michal.Kalderon, Yuval Mintz
First step in revising the LL2 interface, this declares
qed_ll2_tx_pkt_info as part of the ll2 interface, and uses it for
transmission instead of receiving lots of parameters.
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
---
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 127 +++++++++++++----------------
drivers/net/ethernet/qlogic/qed/qed_ll2.h | 41 ++--------
drivers/net/ethernet/qlogic/qed/qed_roce.c | 18 ++--
include/linux/qed/qed_ll2_if.h | 27 ++++++
4 files changed, 102 insertions(+), 111 deletions(-)
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
index f67ed6d..0b75f5d 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
@@ -740,12 +740,13 @@ static void
qed_ooo_submit_tx_buffers(struct qed_hwfn *p_hwfn,
struct qed_ll2_info *p_ll2_conn)
{
+ struct qed_ll2_tx_pkt_info tx_pkt;
struct qed_ooo_buffer *p_buffer;
- int rc;
u16 l4_hdr_offset_w;
dma_addr_t first_frag;
u16 parse_flags;
u8 bd_flags;
+ int rc;
/* Submit Tx buffers here */
while ((p_buffer = qed_ooo_get_ready_buffer(p_hwfn,
@@ -760,13 +761,18 @@ qed_ooo_submit_tx_buffers(struct qed_hwfn *p_hwfn,
SET_FIELD(bd_flags, CORE_TX_BD_DATA_FORCE_VLAN_MODE, 1);
SET_FIELD(bd_flags, CORE_TX_BD_DATA_L4_PROTOCOL, 1);
- rc = qed_ll2_prepare_tx_packet(p_hwfn, p_ll2_conn->my_id, 1,
- p_buffer->vlan, bd_flags,
- l4_hdr_offset_w,
- p_ll2_conn->conn.tx_dest, 0,
- first_frag,
- p_buffer->packet_length,
- p_buffer, true);
+ memset(&tx_pkt, 0, sizeof(tx_pkt));
+ tx_pkt.num_of_bds = 1;
+ tx_pkt.vlan = p_buffer->vlan;
+ tx_pkt.bd_flags = bd_flags;
+ tx_pkt.l4_hdr_offset_w = l4_hdr_offset_w;
+ tx_pkt.tx_dest = p_ll2_conn->conn.tx_dest;
+ tx_pkt.first_frag = first_frag;
+ tx_pkt.first_frag_len = p_buffer->packet_length;
+ tx_pkt.cookie = p_buffer;
+
+ rc = qed_ll2_prepare_tx_packet(p_hwfn, p_ll2_conn->my_id,
+ &tx_pkt, true);
if (rc) {
qed_ooo_put_ready_buffer(p_hwfn, p_hwfn->p_ooo_info,
p_buffer, false);
@@ -1593,20 +1599,18 @@ int qed_ll2_post_rx_buffer(struct qed_hwfn *p_hwfn,
static void qed_ll2_prepare_tx_packet_set(struct qed_hwfn *p_hwfn,
struct qed_ll2_tx_queue *p_tx,
struct qed_ll2_tx_packet *p_curp,
- u8 num_of_bds,
- dma_addr_t first_frag,
- u16 first_frag_len, void *p_cookie,
+ struct qed_ll2_tx_pkt_info *pkt,
u8 notify_fw)
{
list_del(&p_curp->list_entry);
- p_curp->cookie = p_cookie;
- p_curp->bd_used = num_of_bds;
+ p_curp->cookie = pkt->cookie;
+ p_curp->bd_used = pkt->num_of_bds;
p_curp->notify_fw = notify_fw;
p_tx->cur_send_packet = p_curp;
p_tx->cur_send_frag_num = 0;
- p_curp->bds_set[p_tx->cur_send_frag_num].tx_frag = first_frag;
- p_curp->bds_set[p_tx->cur_send_frag_num].frag_len = first_frag_len;
+ p_curp->bds_set[p_tx->cur_send_frag_num].tx_frag = pkt->first_frag;
+ p_curp->bds_set[p_tx->cur_send_frag_num].frag_len = pkt->first_frag_len;
p_tx->cur_send_frag_num++;
}
@@ -1614,32 +1618,33 @@ static void
qed_ll2_prepare_tx_packet_set_bd(struct qed_hwfn *p_hwfn,
struct qed_ll2_info *p_ll2,
struct qed_ll2_tx_packet *p_curp,
- u8 num_of_bds,
- enum core_tx_dest tx_dest,
- u16 vlan,
- u8 bd_flags,
- u16 l4_hdr_offset_w,
- enum core_roce_flavor_type roce_flavor,
- dma_addr_t first_frag,
- u16 first_frag_len)
+ struct qed_ll2_tx_pkt_info *pkt)
{
struct qed_chain *p_tx_chain = &p_ll2->tx_queue.txq_chain;
u16 prod_idx = qed_chain_get_prod_idx(p_tx_chain);
struct core_tx_bd *start_bd = NULL;
+ enum core_roce_flavor_type roce_flavor;
+ enum core_tx_dest tx_dest;
u16 bd_data = 0, frag_idx;
+ roce_flavor = (pkt->qed_roce_flavor == QED_LL2_ROCE) ? CORE_ROCE
+ : CORE_RROCE;
+
+ tx_dest = (pkt->tx_dest == QED_LL2_TX_DEST_NW) ? CORE_TX_DEST_NW
+ : CORE_TX_DEST_LB;
+
start_bd = (struct core_tx_bd *)qed_chain_produce(p_tx_chain);
- start_bd->nw_vlan_or_lb_echo = cpu_to_le16(vlan);
+ start_bd->nw_vlan_or_lb_echo = cpu_to_le16(pkt->vlan);
SET_FIELD(start_bd->bitfield1, CORE_TX_BD_L4_HDR_OFFSET_W,
- cpu_to_le16(l4_hdr_offset_w));
+ cpu_to_le16(pkt->l4_hdr_offset_w));
SET_FIELD(start_bd->bitfield1, CORE_TX_BD_TX_DST, tx_dest);
- bd_data |= bd_flags;
+ bd_data |= pkt->bd_flags;
SET_FIELD(bd_data, CORE_TX_BD_DATA_START_BD, 0x1);
- SET_FIELD(bd_data, CORE_TX_BD_DATA_NBDS, num_of_bds);
+ SET_FIELD(bd_data, CORE_TX_BD_DATA_NBDS, pkt->num_of_bds);
SET_FIELD(bd_data, CORE_TX_BD_DATA_ROCE_FLAV, roce_flavor);
start_bd->bd_data.as_bitfield = cpu_to_le16(bd_data);
- DMA_REGPAIR_LE(start_bd->addr, first_frag);
- start_bd->nbytes = cpu_to_le16(first_frag_len);
+ DMA_REGPAIR_LE(start_bd->addr, pkt->first_frag);
+ start_bd->nbytes = cpu_to_le16(pkt->first_frag_len);
DP_VERBOSE(p_hwfn,
(NETIF_MSG_TX_QUEUED | QED_MSG_LL2),
@@ -1648,17 +1653,17 @@ qed_ll2_prepare_tx_packet_set_bd(struct qed_hwfn *p_hwfn,
p_ll2->cid,
p_ll2->conn.conn_type,
prod_idx,
- first_frag_len,
- num_of_bds,
+ pkt->first_frag_len,
+ pkt->num_of_bds,
le32_to_cpu(start_bd->addr.hi),
le32_to_cpu(start_bd->addr.lo));
- if (p_ll2->tx_queue.cur_send_frag_num == num_of_bds)
+ if (p_ll2->tx_queue.cur_send_frag_num == pkt->num_of_bds)
return;
/* Need to provide the packet with additional BDs for frags */
for (frag_idx = p_ll2->tx_queue.cur_send_frag_num;
- frag_idx < num_of_bds; frag_idx++) {
+ frag_idx < pkt->num_of_bds; frag_idx++) {
struct core_tx_bd **p_bd = &p_curp->bds_set[frag_idx].txq_bd;
*p_bd = (struct core_tx_bd *)qed_chain_produce(p_tx_chain);
@@ -1726,21 +1731,13 @@ static void qed_ll2_tx_packet_notify(struct qed_hwfn *p_hwfn,
int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
u8 connection_handle,
- u8 num_of_bds,
- u16 vlan,
- u8 bd_flags,
- u16 l4_hdr_offset_w,
- enum qed_ll2_tx_dest e_tx_dest,
- enum qed_ll2_roce_flavor_type qed_roce_flavor,
- dma_addr_t first_frag,
- u16 first_frag_len, void *cookie, u8 notify_fw)
+ struct qed_ll2_tx_pkt_info *pkt,
+ bool notify_fw)
{
struct qed_ll2_tx_packet *p_curp = NULL;
struct qed_ll2_info *p_ll2_conn = NULL;
- enum core_roce_flavor_type roce_flavor;
struct qed_ll2_tx_queue *p_tx;
struct qed_chain *p_tx_chain;
- enum core_tx_dest tx_dest;
unsigned long flags;
int rc = 0;
@@ -1750,7 +1747,7 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
p_tx = &p_ll2_conn->tx_queue;
p_tx_chain = &p_tx->txq_chain;
- if (num_of_bds > CORE_LL2_TX_MAX_BDS_PER_PACKET)
+ if (pkt->num_of_bds > CORE_LL2_TX_MAX_BDS_PER_PACKET)
return -EIO;
spin_lock_irqsave(&p_tx->lock, flags);
@@ -1763,7 +1760,7 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
if (!list_empty(&p_tx->free_descq))
p_curp = list_first_entry(&p_tx->free_descq,
struct qed_ll2_tx_packet, list_entry);
- if (p_curp && qed_chain_get_elem_left(p_tx_chain) < num_of_bds)
+ if (p_curp && qed_chain_get_elem_left(p_tx_chain) < pkt->num_of_bds)
p_curp = NULL;
if (!p_curp) {
@@ -1771,26 +1768,10 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
goto out;
}
- tx_dest = e_tx_dest == QED_LL2_TX_DEST_NW ? CORE_TX_DEST_NW :
- CORE_TX_DEST_LB;
- if (qed_roce_flavor == QED_LL2_ROCE) {
- roce_flavor = CORE_ROCE;
- } else if (qed_roce_flavor == QED_LL2_RROCE) {
- roce_flavor = CORE_RROCE;
- } else {
- rc = -EINVAL;
- goto out;
- }
-
/* Prepare packet and BD, and perhaps send a doorbell to FW */
- qed_ll2_prepare_tx_packet_set(p_hwfn, p_tx, p_curp,
- num_of_bds, first_frag,
- first_frag_len, cookie, notify_fw);
- qed_ll2_prepare_tx_packet_set_bd(p_hwfn, p_ll2_conn, p_curp,
- num_of_bds, tx_dest,
- vlan, bd_flags, l4_hdr_offset_w,
- roce_flavor,
- first_frag, first_frag_len);
+ qed_ll2_prepare_tx_packet_set(p_hwfn, p_tx, p_curp, pkt, notify_fw);
+
+ qed_ll2_prepare_tx_packet_set_bd(p_hwfn, p_ll2_conn, p_curp, pkt);
qed_ll2_tx_packet_notify(p_hwfn, p_ll2_conn);
@@ -2245,6 +2226,7 @@ static int qed_ll2_stop(struct qed_dev *cdev)
static int qed_ll2_start_xmit(struct qed_dev *cdev, struct sk_buff *skb)
{
+ struct qed_ll2_tx_pkt_info pkt;
const skb_frag_t *frag;
int rc = -EINVAL, i;
dma_addr_t mapping;
@@ -2279,12 +2261,17 @@ static int qed_ll2_start_xmit(struct qed_dev *cdev, struct sk_buff *skb)
flags |= BIT(CORE_TX_BD_DATA_VLAN_INSERTION_SHIFT);
}
- rc = qed_ll2_prepare_tx_packet(QED_LEADING_HWFN(cdev),
- cdev->ll2->handle,
- 1 + skb_shinfo(skb)->nr_frags,
- vlan, flags, 0, QED_LL2_TX_DEST_NW,
- 0 /* RoCE FLAVOR */,
- mapping, skb->len, skb, 1);
+ memset(&pkt, 0, sizeof(pkt));
+ pkt.num_of_bds = 1 + skb_shinfo(skb)->nr_frags;
+ pkt.vlan = vlan;
+ pkt.bd_flags = flags;
+ pkt.tx_dest = QED_LL2_TX_DEST_NW;
+ pkt.first_frag = mapping;
+ pkt.first_frag_len = skb->len;
+ pkt.cookie = skb;
+
+ rc = qed_ll2_prepare_tx_packet(&cdev->hwfns[0], cdev->ll2->handle,
+ &pkt, 1);
if (rc)
goto err;
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.h b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
index 2c07d0e..96af50b 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
@@ -47,12 +47,6 @@
#define QED_MAX_NUM_OF_LL2_CONNECTIONS (4)
-enum qed_ll2_roce_flavor_type {
- QED_LL2_ROCE,
- QED_LL2_RROCE,
- MAX_QED_LL2_ROCE_FLAVOR_TYPE
-};
-
enum qed_ll2_conn_type {
QED_LL2_TYPE_FCOE,
QED_LL2_TYPE_ISCSI,
@@ -64,12 +58,6 @@ enum qed_ll2_conn_type {
MAX_QED_LL2_RX_CONN_TYPE
};
-enum qed_ll2_tx_dest {
- QED_LL2_TX_DEST_NW, /* Light L2 TX Destination to the Network */
- QED_LL2_TX_DEST_LB, /* Light L2 TX Destination to the Loopback */
- QED_LL2_TX_DEST_MAX
-};
-
struct qed_ll2_rx_packet {
struct list_head list_entry;
struct core_rx_bd_with_buff_len *rxq_bd;
@@ -216,35 +204,16 @@ int qed_ll2_post_rx_buffer(struct qed_hwfn *p_hwfn,
* to prepare Tx packet submission to FW.
*
* @param p_hwfn
- * @param connection_handle LL2 connection's handle obtained from
- * qed_ll2_require_connection
- * @param num_of_bds a number of requested BD equals a number of
- * fragments in Tx packet
- * @param vlan VLAN to insert to packet (if insertion set)
- * @param bd_flags
- * @param l4_hdr_offset_w L4 Header Offset from start of packet
- * (in words). This is needed if both l4_csum
- * and ipv6_ext are set
- * @param e_tx_dest indicates if the packet is to be transmitted via
- * loopback or to the network
- * @param first_frag
- * @param first_frag_len
- * @param cookie
- *
- * @param notify_fw
+ * @param connection_handle
+ * @param pkt - info regarding the tx packet
+ * @param notify_fw - issue doorbell to fw for this packet
*
* @return 0 on success, failure otherwise
*/
int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
u8 connection_handle,
- u8 num_of_bds,
- u16 vlan,
- u8 bd_flags,
- u16 l4_hdr_offset_w,
- enum qed_ll2_tx_dest e_tx_dest,
- enum qed_ll2_roce_flavor_type qed_roce_flavor,
- dma_addr_t first_frag,
- u16 first_frag_len, void *cookie, u8 notify_fw);
+ struct qed_ll2_tx_pkt_info *pkt,
+ bool notify_fw);
/**
* @brief qed_ll2_release_connection - releases resources
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index b9434b7..afc63c0 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -2937,6 +2937,7 @@ static int qed_roce_ll2_tx(struct qed_dev *cdev,
struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
struct qed_roce_ll2_info *roce_ll2 = hwfn->ll2;
enum qed_ll2_roce_flavor_type qed_roce_flavor;
+ struct qed_ll2_tx_pkt_info ll2_pkt;
u8 flags = 0;
int rc;
int i;
@@ -2955,11 +2956,18 @@ static int qed_roce_ll2_tx(struct qed_dev *cdev,
flags |= BIT(CORE_TX_BD_DATA_IP_CSUM_SHIFT);
/* Tx header */
- rc = qed_ll2_prepare_tx_packet(QED_LEADING_HWFN(cdev), roce_ll2->handle,
- 1 + pkt->n_seg, 0, flags, 0,
- QED_LL2_TX_DEST_NW,
- qed_roce_flavor, pkt->header.baddr,
- pkt->header.len, pkt, 1);
+ memset(&ll2_pkt, 0, sizeof(ll2_pkt));
+ ll2_pkt.num_of_bds = 1 + pkt->n_seg;
+ ll2_pkt.bd_flags = flags;
+ ll2_pkt.tx_dest = QED_LL2_TX_DEST_NW;
+ ll2_pkt.qed_roce_flavor = qed_roce_flavor;
+ ll2_pkt.first_frag = pkt->header.baddr;
+ ll2_pkt.first_frag_len = pkt->header.len;
+ ll2_pkt.cookie = pkt;
+
+ rc = qed_ll2_prepare_tx_packet(QED_LEADING_HWFN(cdev),
+ roce_ll2->handle,
+ &ll2_pkt, 1);
if (rc) {
DP_ERR(cdev, "roce ll2 tx: header failed (rc=%d)\n", rc);
return QED_ROCE_TX_HEAD_FAILURE;
diff --git a/include/linux/qed/qed_ll2_if.h b/include/linux/qed/qed_ll2_if.h
index 4fb4666..1a9a5b9 100644
--- a/include/linux/qed/qed_ll2_if.h
+++ b/include/linux/qed/qed_ll2_if.h
@@ -43,6 +43,18 @@
#include <linux/slab.h>
#include <linux/qed/qed_if.h>
+enum qed_ll2_roce_flavor_type {
+ QED_LL2_ROCE,
+ QED_LL2_RROCE,
+ MAX_QED_LL2_ROCE_FLAVOR_TYPE
+};
+
+enum qed_ll2_tx_dest {
+ QED_LL2_TX_DEST_NW, /* Light L2 TX Destination to the Network */
+ QED_LL2_TX_DEST_LB, /* Light L2 TX Destination to the Loopback */
+ QED_LL2_TX_DEST_MAX
+};
+
struct qed_ll2_stats {
u64 gsi_invalid_hdr;
u64 gsi_invalid_pkt_length;
@@ -67,6 +79,21 @@ struct qed_ll2_stats {
u64 sent_bcast_pkts;
};
+struct qed_ll2_tx_pkt_info {
+ u8 num_of_bds;
+ u16 vlan;
+ u8 bd_flags;
+ u16 l4_hdr_offset_w; /* from start of packet */
+ enum qed_ll2_tx_dest tx_dest;
+ enum qed_ll2_roce_flavor_type qed_roce_flavor;
+ dma_addr_t first_frag;
+ u16 first_frag_len;
+ bool enable_ip_cksum;
+ bool enable_l4_cksum;
+ bool calc_ip_len;
+ void *cookie;
+};
+
#define QED_LL2_UNUSED_HANDLE (0xff)
struct qed_ll2_cb_ops {
--
2.9.4
^ permalink raw reply related [flat|nested] 16+ messages in thread
* [PATCH net-next 2/8] qed: Revise ll2 Rx completion
2017-06-08 16:13 [PATCH net-next 0/8] qed*: Light L2 updates Yuval Mintz
2017-06-08 16:13 ` [PATCH net-next 1/8] qed: LL2 to use packed information for tx Yuval Mintz
@ 2017-06-08 16:13 ` Yuval Mintz
[not found] ` <20170608161323.7597-3-Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
2017-06-08 16:13 ` [PATCH net-next 4/8] qed: LL2 code relocations Yuval Mintz
` (3 subsequent siblings)
5 siblings, 1 reply; 16+ messages in thread
From: Yuval Mintz @ 2017-06-08 16:13 UTC (permalink / raw)
To: netdev, davem; +Cc: linux-rdma, Michal.Kalderon, Yuval Mintz
This introduces qed_ll2_comp_rx_data as a public struct
and moves handling of Rx packets in LL2 into using it.
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
---
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 84 ++++++++++++++++++-------------
include/linux/qed/qed_ll2_if.h | 25 +++++++++
2 files changed, 75 insertions(+), 34 deletions(-)
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
index 0b75f5d..46b71a6 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
@@ -165,41 +165,33 @@ static void qed_ll2_kill_buffers(struct qed_dev *cdev)
}
static void qed_ll2b_complete_rx_packet(struct qed_hwfn *p_hwfn,
- u8 connection_handle,
- struct qed_ll2_rx_packet *p_pkt,
- struct core_rx_fast_path_cqe *p_cqe,
- bool b_last_packet)
+ struct qed_ll2_comp_rx_data *data)
{
- u16 packet_length = le16_to_cpu(p_cqe->packet_length);
- struct qed_ll2_buffer *buffer = p_pkt->cookie;
+ struct qed_ll2_buffer *buffer = data->cookie;
struct qed_dev *cdev = p_hwfn->cdev;
- u16 vlan = le16_to_cpu(p_cqe->vlan);
- u32 opaque_data_0, opaque_data_1;
- u8 pad = p_cqe->placement_offset;
dma_addr_t new_phys_addr;
struct sk_buff *skb;
bool reuse = false;
int rc = -EINVAL;
u8 *new_data;
- opaque_data_0 = le32_to_cpu(p_cqe->opaque_data.data[0]);
- opaque_data_1 = le32_to_cpu(p_cqe->opaque_data.data[1]);
-
DP_VERBOSE(p_hwfn,
(NETIF_MSG_RX_STATUS | QED_MSG_STORAGE | NETIF_MSG_PKTDATA),
"Got an LL2 Rx completion: [Buffer at phys 0x%llx, offset 0x%02x] Length 0x%04x Parse_flags 0x%04x vlan 0x%04x Opaque data [0x%08x:0x%08x]\n",
- (u64)p_pkt->rx_buf_addr, pad, packet_length,
- le16_to_cpu(p_cqe->parse_flags.flags), vlan,
- opaque_data_0, opaque_data_1);
+ (u64)data->rx_buf_addr,
+ data->u.placement_offset,
+ data->length.packet_length,
+ data->parse_flags,
+ data->vlan, data->opaque_data_0, data->opaque_data_1);
if ((cdev->dp_module & NETIF_MSG_PKTDATA) && buffer->data) {
print_hex_dump(KERN_INFO, "",
DUMP_PREFIX_OFFSET, 16, 1,
- buffer->data, packet_length, false);
+ buffer->data, data->length.packet_length, false);
}
/* Determine if data is valid */
- if (packet_length < ETH_HLEN)
+ if (data->length.packet_length < ETH_HLEN)
reuse = true;
/* Allocate a replacement for buffer; Reuse upon failure */
@@ -219,9 +211,9 @@ static void qed_ll2b_complete_rx_packet(struct qed_hwfn *p_hwfn,
goto out_post;
}
- pad += NET_SKB_PAD;
- skb_reserve(skb, pad);
- skb_put(skb, packet_length);
+ data->u.placement_offset += NET_SKB_PAD;
+ skb_reserve(skb, data->u.placement_offset);
+ skb_put(skb, data->length.packet_length);
skb_checksum_none_assert(skb);
/* Get parital ethernet information instead of eth_type_trans(),
@@ -232,10 +224,12 @@ static void qed_ll2b_complete_rx_packet(struct qed_hwfn *p_hwfn,
/* Pass SKB onward */
if (cdev->ll2->cbs && cdev->ll2->cbs->rx_cb) {
- if (vlan)
- __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan);
+ if (data->vlan)
+ __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
+ data->vlan);
cdev->ll2->cbs->rx_cb(cdev->ll2->cb_cookie, skb,
- opaque_data_0, opaque_data_1);
+ data->opaque_data_0,
+ data->opaque_data_1);
}
/* Update Buffer information and update FW producer */
@@ -472,33 +466,55 @@ qed_ll2_rxq_completion_gsi(struct qed_hwfn *p_hwfn,
return 0;
}
-static int qed_ll2_rxq_completion_reg(struct qed_hwfn *p_hwfn,
- struct qed_ll2_info *p_ll2_conn,
- union core_rx_cqe_union *p_cqe,
- unsigned long *p_lock_flags,
- bool b_last_cqe)
+static void qed_ll2_rxq_parse_reg(struct qed_hwfn *p_hwfn,
+ union core_rx_cqe_union *p_cqe,
+ struct qed_ll2_comp_rx_data *data)
+{
+ data->parse_flags = le16_to_cpu(p_cqe->rx_cqe_fp.parse_flags.flags);
+ data->length.packet_length =
+ le16_to_cpu(p_cqe->rx_cqe_fp.packet_length);
+ data->vlan = le16_to_cpu(p_cqe->rx_cqe_fp.vlan);
+ data->opaque_data_0 = le32_to_cpu(p_cqe->rx_cqe_fp.opaque_data.data[0]);
+ data->opaque_data_1 = le32_to_cpu(p_cqe->rx_cqe_fp.opaque_data.data[1]);
+ data->u.placement_offset = p_cqe->rx_cqe_fp.placement_offset;
+}
+
+static int
+qed_ll2_rxq_handle_completion(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn,
+ union core_rx_cqe_union *p_cqe,
+ unsigned long *p_lock_flags, bool b_last_cqe)
{
struct qed_ll2_rx_queue *p_rx = &p_ll2_conn->rx_queue;
struct qed_ll2_rx_packet *p_pkt = NULL;
+ struct qed_ll2_comp_rx_data data;
if (!list_empty(&p_rx->active_descq))
p_pkt = list_first_entry(&p_rx->active_descq,
struct qed_ll2_rx_packet, list_entry);
if (!p_pkt) {
DP_NOTICE(p_hwfn,
- "LL2 Rx completion but active_descq is empty\n");
+ "[%d] LL2 Rx completion but active_descq is empty\n",
+ p_ll2_conn->conn.conn_type);
+
return -EIO;
}
list_del(&p_pkt->list_entry);
+ qed_ll2_rxq_parse_reg(p_hwfn, p_cqe, &data);
if (qed_chain_consume(&p_rx->rxq_chain) != p_pkt->rxq_bd)
DP_NOTICE(p_hwfn,
"Mismatch between active_descq and the LL2 Rx chain\n");
+
list_add_tail(&p_pkt->list_entry, &p_rx->free_descq);
+ data.connection_handle = p_ll2_conn->my_id;
+ data.cookie = p_pkt->cookie;
+ data.rx_buf_addr = p_pkt->rx_buf_addr;
+ data.b_last_packet = b_last_cqe;
+
spin_unlock_irqrestore(&p_rx->lock, *p_lock_flags);
- qed_ll2b_complete_rx_packet(p_hwfn, p_ll2_conn->my_id,
- p_pkt, &p_cqe->rx_cqe_fp, b_last_cqe);
+ qed_ll2b_complete_rx_packet(p_hwfn, &data);
spin_lock_irqsave(&p_rx->lock, *p_lock_flags);
return 0;
@@ -538,9 +554,9 @@ static int qed_ll2_rxq_completion(struct qed_hwfn *p_hwfn, void *cookie)
cqe, flags, b_last_cqe);
break;
case CORE_RX_CQE_TYPE_REGULAR:
- rc = qed_ll2_rxq_completion_reg(p_hwfn, p_ll2_conn,
- cqe, &flags,
- b_last_cqe);
+ rc = qed_ll2_rxq_handle_completion(p_hwfn, p_ll2_conn,
+ cqe, &flags,
+ b_last_cqe);
break;
default:
rc = -EIO;
diff --git a/include/linux/qed/qed_ll2_if.h b/include/linux/qed/qed_ll2_if.h
index 1a9a5b9..5145278 100644
--- a/include/linux/qed/qed_ll2_if.h
+++ b/include/linux/qed/qed_ll2_if.h
@@ -79,6 +79,31 @@ struct qed_ll2_stats {
u64 sent_bcast_pkts;
};
+struct qed_ll2_comp_rx_data {
+ u8 connection_handle;
+ void *cookie;
+ dma_addr_t rx_buf_addr;
+ u16 parse_flags;
+ u16 vlan;
+ bool b_last_packet;
+
+ union {
+ u8 placement_offset;
+ u8 data_length_error;
+ } u;
+ union {
+ u16 packet_length;
+ u16 data_length;
+ } length;
+
+ u32 opaque_data_0;
+ u32 opaque_data_1;
+
+ /* GSI only */
+ u32 gid_dst[4];
+ u16 qp_id;
+};
+
struct qed_ll2_tx_pkt_info {
u8 num_of_bds;
u16 vlan;
--
2.9.4
^ permalink raw reply related [flat|nested] 16+ messages in thread
* [PATCH net-next 3/8] qed: Cleaner seperation of LL2 inputs
[not found] ` <20170608161323.7597-1-Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
@ 2017-06-08 16:13 ` Yuval Mintz
2017-06-08 16:13 ` [PATCH net-next 7/8] qed: Call rx_release_cb() when flushing LL2 Yuval Mintz
2017-06-08 16:13 ` [PATCH net-next 8/8] qed: collect GSI port statistics Yuval Mintz
2 siblings, 0 replies; 16+ messages in thread
From: Yuval Mintz @ 2017-06-08 16:13 UTC (permalink / raw)
To: netdev-u79uwXL29TY76Z2rM5mHXA, davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
Michal.Kalderon-YGCgFSpz5w/QT0dZR+AlfA, Yuval Mintz
A LL2 connection [qed_ll2_info] has a sub-structure of type qed_ll2_conn
that contain various inputs for ll2 acquisition, but the connection also
utilizes a couple of other inputs.
Restructure the input structure to include all the inputs and refactor
the code necessary to populate those.
Signed-off-by: Yuval Mintz <Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
---
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 379 +++++++++++++++--------------
drivers/net/ethernet/qlogic/qed/qed_ll2.h | 42 +---
drivers/net/ethernet/qlogic/qed/qed_roce.c | 31 +--
include/linux/qed/qed_ll2_if.h | 38 +++
4 files changed, 261 insertions(+), 229 deletions(-)
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
index 46b71a6..fcf4ea9 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
@@ -315,7 +315,7 @@ static void qed_ll2_txq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
list_del(&p_pkt->list_entry);
b_last_packet = list_empty(&p_tx->active_descq);
list_add_tail(&p_pkt->list_entry, &p_tx->free_descq);
- if (p_ll2_conn->conn.conn_type == QED_LL2_TYPE_ISCSI_OOO) {
+ if (p_ll2_conn->input.conn_type == QED_LL2_TYPE_ISCSI_OOO) {
struct qed_ooo_buffer *p_buffer;
p_buffer = (struct qed_ooo_buffer *)p_pkt->cookie;
@@ -327,7 +327,7 @@ static void qed_ll2_txq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
b_last_frag =
p_tx->cur_completing_bd_idx == p_pkt->bd_used;
tx_frag = p_pkt->bds_set[0].tx_frag;
- if (p_ll2_conn->conn.gsi_enable)
+ if (p_ll2_conn->input.gsi_enable)
qed_ll2b_release_tx_gsi_packet(p_hwfn,
p_ll2_conn->
my_id,
@@ -396,7 +396,7 @@ static int qed_ll2_txq_completion(struct qed_hwfn *p_hwfn, void *p_cookie)
spin_unlock_irqrestore(&p_tx->lock, flags);
tx_frag = p_pkt->bds_set[0].tx_frag;
- if (p_ll2_conn->conn.gsi_enable)
+ if (p_ll2_conn->input.gsi_enable)
qed_ll2b_complete_tx_gsi_packet(p_hwfn,
p_ll2_conn->my_id,
p_pkt->cookie,
@@ -495,7 +495,7 @@ qed_ll2_rxq_handle_completion(struct qed_hwfn *p_hwfn,
if (!p_pkt) {
DP_NOTICE(p_hwfn,
"[%d] LL2 Rx completion but active_descq is empty\n",
- p_ll2_conn->conn.conn_type);
+ p_ll2_conn->input.conn_type);
return -EIO;
}
@@ -522,7 +522,7 @@ qed_ll2_rxq_handle_completion(struct qed_hwfn *p_hwfn,
static int qed_ll2_rxq_completion(struct qed_hwfn *p_hwfn, void *cookie)
{
- struct qed_ll2_info *p_ll2_conn = cookie;
+ struct qed_ll2_info *p_ll2_conn = (struct qed_ll2_info *)cookie;
struct qed_ll2_rx_queue *p_rx = &p_ll2_conn->rx_queue;
union core_rx_cqe_union *cqe = NULL;
u16 cq_new_idx = 0, cq_old_idx = 0;
@@ -536,7 +536,9 @@ static int qed_ll2_rxq_completion(struct qed_hwfn *p_hwfn, void *cookie)
while (cq_new_idx != cq_old_idx) {
bool b_last_cqe = (cq_new_idx == cq_old_idx);
- cqe = qed_chain_consume(&p_rx->rcq_chain);
+ cqe =
+ (union core_rx_cqe_union *)
+ qed_chain_consume(&p_rx->rcq_chain);
cq_old_idx = qed_chain_get_cons_idx(&p_rx->rcq_chain);
DP_VERBOSE(p_hwfn,
@@ -591,7 +593,7 @@ static void qed_ll2_rxq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
list_move_tail(&p_pkt->list_entry, &p_rx->free_descq);
- if (p_ll2_conn->conn.conn_type == QED_LL2_TYPE_ISCSI_OOO) {
+ if (p_ll2_conn->input.conn_type == QED_LL2_TYPE_ISCSI_OOO) {
struct qed_ooo_buffer *p_buffer;
p_buffer = (struct qed_ooo_buffer *)p_pkt->cookie;
@@ -606,7 +608,6 @@ static void qed_ll2_rxq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
}
}
-#if IS_ENABLED(CONFIG_QED_ISCSI)
static u8 qed_ll2_convert_rx_parse_to_tx_flags(u16 parse_flags)
{
u8 bd_flags = 0;
@@ -782,7 +783,7 @@ qed_ooo_submit_tx_buffers(struct qed_hwfn *p_hwfn,
tx_pkt.vlan = p_buffer->vlan;
tx_pkt.bd_flags = bd_flags;
tx_pkt.l4_hdr_offset_w = l4_hdr_offset_w;
- tx_pkt.tx_dest = p_ll2_conn->conn.tx_dest;
+ tx_pkt.tx_dest = p_ll2_conn->tx_dest;
tx_pkt.first_frag = first_frag;
tx_pkt.first_frag_len = p_buffer->packet_length;
tx_pkt.cookie = p_buffer;
@@ -895,60 +896,11 @@ static int qed_ll2_lb_txq_completion(struct qed_hwfn *p_hwfn, void *p_cookie)
return 0;
}
-static int
-qed_ll2_acquire_connection_ooo(struct qed_hwfn *p_hwfn,
- struct qed_ll2_info *p_ll2_info,
- u16 rx_num_ooo_buffers, u16 mtu)
-{
- struct qed_ooo_buffer *p_buf = NULL;
- void *p_virt;
- u16 buf_idx;
- int rc = 0;
-
- if (p_ll2_info->conn.conn_type != QED_LL2_TYPE_ISCSI_OOO)
- return rc;
-
- if (!rx_num_ooo_buffers)
- return -EINVAL;
-
- for (buf_idx = 0; buf_idx < rx_num_ooo_buffers; buf_idx++) {
- p_buf = kzalloc(sizeof(*p_buf), GFP_KERNEL);
- if (!p_buf) {
- rc = -ENOMEM;
- goto out;
- }
-
- p_buf->rx_buffer_size = mtu + 26 + ETH_CACHE_LINE_SIZE;
- p_buf->rx_buffer_size = (p_buf->rx_buffer_size +
- ETH_CACHE_LINE_SIZE - 1) &
- ~(ETH_CACHE_LINE_SIZE - 1);
- p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
- p_buf->rx_buffer_size,
- &p_buf->rx_buffer_phys_addr,
- GFP_KERNEL);
- if (!p_virt) {
- kfree(p_buf);
- rc = -ENOMEM;
- goto out;
- }
-
- p_buf->rx_buffer_virt_addr = p_virt;
- qed_ooo_put_free_buffer(p_hwfn, p_hwfn->p_ooo_info, p_buf);
- }
-
- DP_VERBOSE(p_hwfn, QED_MSG_LL2,
- "Allocated [%04x] LL2 OOO buffers [each of size 0x%08x]\n",
- rx_num_ooo_buffers, p_buf->rx_buffer_size);
-
-out:
- return rc;
-}
-
static void
qed_ll2_establish_connection_ooo(struct qed_hwfn *p_hwfn,
struct qed_ll2_info *p_ll2_conn)
{
- if (p_ll2_conn->conn.conn_type != QED_LL2_TYPE_ISCSI_OOO)
+ if (p_ll2_conn->input.conn_type != QED_LL2_TYPE_ISCSI_OOO)
return;
qed_ooo_release_all_isles(p_hwfn, p_hwfn->p_ooo_info);
@@ -960,7 +912,7 @@ static void qed_ll2_release_connection_ooo(struct qed_hwfn *p_hwfn,
{
struct qed_ooo_buffer *p_buffer;
- if (p_ll2_conn->conn.conn_type != QED_LL2_TYPE_ISCSI_OOO)
+ if (p_ll2_conn->input.conn_type != QED_LL2_TYPE_ISCSI_OOO)
return;
qed_ooo_release_all_isles(p_hwfn, p_hwfn->p_ooo_info);
@@ -987,69 +939,11 @@ static void qed_ll2_stop_ooo(struct qed_dev *cdev)
*handle = QED_LL2_UNUSED_HANDLE;
}
-static int qed_ll2_start_ooo(struct qed_dev *cdev,
- struct qed_ll2_params *params)
-{
- struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
- u8 *handle = &hwfn->pf_params.iscsi_pf_params.ll2_ooo_queue_id;
- struct qed_ll2_conn ll2_info = { 0 };
- int rc;
-
- ll2_info.conn_type = QED_LL2_TYPE_ISCSI_OOO;
- ll2_info.mtu = params->mtu;
- ll2_info.rx_drop_ttl0_flg = params->drop_ttl0_packets;
- ll2_info.rx_vlan_removal_en = params->rx_vlan_stripping;
- ll2_info.tx_tc = OOO_LB_TC;
- ll2_info.tx_dest = CORE_TX_DEST_LB;
-
- rc = qed_ll2_acquire_connection(hwfn, &ll2_info,
- QED_LL2_RX_SIZE, QED_LL2_TX_SIZE,
- handle);
- if (rc) {
- DP_INFO(cdev, "Failed to acquire LL2 OOO connection\n");
- goto out;
- }
-
- rc = qed_ll2_establish_connection(hwfn, *handle);
- if (rc) {
- DP_INFO(cdev, "Failed to establist LL2 OOO connection\n");
- goto fail;
- }
-
- return 0;
-
-fail:
- qed_ll2_release_connection(hwfn, *handle);
-out:
- *handle = QED_LL2_UNUSED_HANDLE;
- return rc;
-}
-#else /* IS_ENABLED(CONFIG_QED_ISCSI) */
-static int qed_ll2_lb_rxq_completion(struct qed_hwfn *p_hwfn,
- void *p_cookie) { return -EINVAL; }
-static int qed_ll2_lb_txq_completion(struct qed_hwfn *p_hwfn,
- void *p_cookie) { return -EINVAL; }
-static inline int
-qed_ll2_acquire_connection_ooo(struct qed_hwfn *p_hwfn,
- struct qed_ll2_info *p_ll2_info,
- u16 rx_num_ooo_buffers, u16 mtu) { return 0; }
-static inline void
-qed_ll2_establish_connection_ooo(struct qed_hwfn *p_hwfn,
- struct qed_ll2_info *p_ll2_conn) { return; }
-static inline void
-qed_ll2_release_connection_ooo(struct qed_hwfn *p_hwfn,
- struct qed_ll2_info *p_ll2_conn) { return; }
-static inline void qed_ll2_stop_ooo(struct qed_dev *cdev) { return; }
-static inline int qed_ll2_start_ooo(struct qed_dev *cdev,
- struct qed_ll2_params *params)
- { return -EINVAL; }
-#endif /* IS_ENABLED(CONFIG_QED_ISCSI) */
-
static int qed_sp_ll2_rx_queue_start(struct qed_hwfn *p_hwfn,
struct qed_ll2_info *p_ll2_conn,
u8 action_on_error)
{
- enum qed_ll2_conn_type conn_type = p_ll2_conn->conn.conn_type;
+ enum qed_ll2_conn_type conn_type = p_ll2_conn->input.conn_type;
struct qed_ll2_rx_queue *p_rx = &p_ll2_conn->rx_queue;
struct core_rx_start_ramrod_data *p_ramrod = NULL;
struct qed_spq_entry *p_ent = NULL;
@@ -1075,16 +969,15 @@ static int qed_sp_ll2_rx_queue_start(struct qed_hwfn *p_hwfn,
p_ramrod->sb_index = p_rx->rx_sb_index;
p_ramrod->complete_event_flg = 1;
- p_ramrod->mtu = cpu_to_le16(p_ll2_conn->conn.mtu);
- DMA_REGPAIR_LE(p_ramrod->bd_base,
- p_rx->rxq_chain.p_phys_addr);
+ p_ramrod->mtu = cpu_to_le16(p_ll2_conn->input.mtu);
+ DMA_REGPAIR_LE(p_ramrod->bd_base, p_rx->rxq_chain.p_phys_addr);
cqe_pbl_size = (u16)qed_chain_get_page_cnt(&p_rx->rcq_chain);
p_ramrod->num_of_pbl_pages = cpu_to_le16(cqe_pbl_size);
DMA_REGPAIR_LE(p_ramrod->cqe_pbl_addr,
qed_chain_get_pbl_phys(&p_rx->rcq_chain));
- p_ramrod->drop_ttl0_flg = p_ll2_conn->conn.rx_drop_ttl0_flg;
- p_ramrod->inner_vlan_removal_en = p_ll2_conn->conn.rx_vlan_removal_en;
+ p_ramrod->drop_ttl0_flg = p_ll2_conn->input.rx_drop_ttl0_flg;
+ p_ramrod->inner_vlan_removal_en = p_ll2_conn->input.rx_vlan_removal_en;
p_ramrod->queue_id = p_ll2_conn->queue_id;
p_ramrod->main_func_queue = (conn_type == QED_LL2_TYPE_ISCSI_OOO) ? 0
: 1;
@@ -1099,14 +992,14 @@ static int qed_sp_ll2_rx_queue_start(struct qed_hwfn *p_hwfn,
}
p_ramrod->action_on_error.error_type = action_on_error;
- p_ramrod->gsi_offload_flag = p_ll2_conn->conn.gsi_enable;
+ p_ramrod->gsi_offload_flag = p_ll2_conn->input.gsi_enable;
return qed_spq_post(p_hwfn, p_ent, NULL);
}
static int qed_sp_ll2_tx_queue_start(struct qed_hwfn *p_hwfn,
struct qed_ll2_info *p_ll2_conn)
{
- enum qed_ll2_conn_type conn_type = p_ll2_conn->conn.conn_type;
+ enum qed_ll2_conn_type conn_type = p_ll2_conn->input.conn_type;
struct qed_ll2_tx_queue *p_tx = &p_ll2_conn->tx_queue;
struct core_tx_start_ramrod_data *p_ramrod = NULL;
struct qed_spq_entry *p_ent = NULL;
@@ -1117,7 +1010,7 @@ static int qed_sp_ll2_tx_queue_start(struct qed_hwfn *p_hwfn,
if (!QED_LL2_TX_REGISTERED(p_ll2_conn))
return 0;
- if (p_ll2_conn->conn.conn_type == QED_LL2_TYPE_ISCSI_OOO)
+ if (p_ll2_conn->input.conn_type == QED_LL2_TYPE_ISCSI_OOO)
p_ll2_conn->tx_stats_en = 0;
else
p_ll2_conn->tx_stats_en = 1;
@@ -1138,7 +1031,7 @@ static int qed_sp_ll2_tx_queue_start(struct qed_hwfn *p_hwfn,
p_ramrod->sb_id = cpu_to_le16(qed_int_get_sp_sb_id(p_hwfn));
p_ramrod->sb_index = p_tx->tx_sb_index;
- p_ramrod->mtu = cpu_to_le16(p_ll2_conn->conn.mtu);
+ p_ramrod->mtu = cpu_to_le16(p_ll2_conn->input.mtu);
p_ramrod->stats_en = p_ll2_conn->tx_stats_en;
p_ramrod->stats_id = p_ll2_conn->tx_stats_id;
@@ -1147,7 +1040,7 @@ static int qed_sp_ll2_tx_queue_start(struct qed_hwfn *p_hwfn,
pbl_size = qed_chain_get_page_cnt(&p_tx->txq_chain);
p_ramrod->pbl_size = cpu_to_le16(pbl_size);
- switch (p_ll2_conn->conn.tx_tc) {
+ switch (p_ll2_conn->input.tx_tc) {
case LB_TC:
pq_id = qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_LB);
break;
@@ -1177,7 +1070,8 @@ static int qed_sp_ll2_tx_queue_start(struct qed_hwfn *p_hwfn,
DP_NOTICE(p_hwfn, "Unknown connection type: %d\n", conn_type);
}
- p_ramrod->gsi_offload_flag = p_ll2_conn->conn.gsi_enable;
+ p_ramrod->gsi_offload_flag = p_ll2_conn->input.gsi_enable;
+
return qed_spq_post(p_hwfn, p_ent, NULL);
}
@@ -1233,20 +1127,20 @@ static int qed_sp_ll2_tx_queue_stop(struct qed_hwfn *p_hwfn,
static int
qed_ll2_acquire_connection_rx(struct qed_hwfn *p_hwfn,
- struct qed_ll2_info *p_ll2_info, u16 rx_num_desc)
+ struct qed_ll2_info *p_ll2_info)
{
struct qed_ll2_rx_packet *p_descq;
u32 capacity;
int rc = 0;
- if (!rx_num_desc)
+ if (!p_ll2_info->input.rx_num_desc)
goto out;
rc = qed_chain_alloc(p_hwfn->cdev,
QED_CHAIN_USE_TO_CONSUME_PRODUCE,
QED_CHAIN_MODE_NEXT_PTR,
QED_CHAIN_CNT_TYPE_U16,
- rx_num_desc,
+ p_ll2_info->input.rx_num_desc,
sizeof(struct core_rx_bd),
&p_ll2_info->rx_queue.rxq_chain);
if (rc) {
@@ -1268,7 +1162,7 @@ qed_ll2_acquire_connection_rx(struct qed_hwfn *p_hwfn,
QED_CHAIN_USE_TO_CONSUME_PRODUCE,
QED_CHAIN_MODE_PBL,
QED_CHAIN_CNT_TYPE_U16,
- rx_num_desc,
+ p_ll2_info->input.rx_num_desc,
sizeof(struct core_rx_fast_path_cqe),
&p_ll2_info->rx_queue.rcq_chain);
if (rc) {
@@ -1278,28 +1172,27 @@ qed_ll2_acquire_connection_rx(struct qed_hwfn *p_hwfn,
DP_VERBOSE(p_hwfn, QED_MSG_LL2,
"Allocated LL2 Rxq [Type %08x] with 0x%08x buffers\n",
- p_ll2_info->conn.conn_type, rx_num_desc);
+ p_ll2_info->input.conn_type, p_ll2_info->input.rx_num_desc);
out:
return rc;
}
static int qed_ll2_acquire_connection_tx(struct qed_hwfn *p_hwfn,
- struct qed_ll2_info *p_ll2_info,
- u16 tx_num_desc)
+ struct qed_ll2_info *p_ll2_info)
{
struct qed_ll2_tx_packet *p_descq;
u32 capacity;
int rc = 0;
- if (!tx_num_desc)
+ if (!p_ll2_info->input.tx_num_desc)
goto out;
rc = qed_chain_alloc(p_hwfn->cdev,
QED_CHAIN_USE_TO_CONSUME_PRODUCE,
QED_CHAIN_MODE_PBL,
QED_CHAIN_CNT_TYPE_U16,
- tx_num_desc,
+ p_ll2_info->input.tx_num_desc,
sizeof(struct core_tx_bd),
&p_ll2_info->tx_queue.txq_chain);
if (rc)
@@ -1316,28 +1209,95 @@ static int qed_ll2_acquire_connection_tx(struct qed_hwfn *p_hwfn,
DP_VERBOSE(p_hwfn, QED_MSG_LL2,
"Allocated LL2 Txq [Type %08x] with 0x%08x buffers\n",
- p_ll2_info->conn.conn_type, tx_num_desc);
+ p_ll2_info->input.conn_type, p_ll2_info->input.tx_num_desc);
out:
if (rc)
DP_NOTICE(p_hwfn,
"Can't allocate memory for Tx LL2 with 0x%08x buffers\n",
- tx_num_desc);
+ p_ll2_info->input.tx_num_desc);
+ return rc;
+}
+
+static int
+qed_ll2_acquire_connection_ooo(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_info, u16 mtu)
+{
+ struct qed_ooo_buffer *p_buf = NULL;
+ void *p_virt;
+ u16 buf_idx;
+ int rc = 0;
+
+ if (p_ll2_info->input.conn_type != QED_LL2_TYPE_ISCSI_OOO)
+ return rc;
+
+ /* Correct number of requested OOO buffers if needed */
+ if (!p_ll2_info->input.rx_num_ooo_buffers) {
+ u16 num_desc = p_ll2_info->input.rx_num_desc;
+
+ if (!num_desc)
+ return -EINVAL;
+ p_ll2_info->input.rx_num_ooo_buffers = num_desc * 2;
+ }
+
+ for (buf_idx = 0; buf_idx < p_ll2_info->input.rx_num_ooo_buffers;
+ buf_idx++) {
+ p_buf = kzalloc(sizeof(*p_buf), GFP_KERNEL);
+ if (!p_buf) {
+ rc = -ENOMEM;
+ goto out;
+ }
+
+ p_buf->rx_buffer_size = mtu + 26 + ETH_CACHE_LINE_SIZE;
+ p_buf->rx_buffer_size = (p_buf->rx_buffer_size +
+ ETH_CACHE_LINE_SIZE - 1) &
+ ~(ETH_CACHE_LINE_SIZE - 1);
+ p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
+ p_buf->rx_buffer_size,
+ &p_buf->rx_buffer_phys_addr,
+ GFP_KERNEL);
+ if (!p_virt) {
+ kfree(p_buf);
+ rc = -ENOMEM;
+ goto out;
+ }
+
+ p_buf->rx_buffer_virt_addr = p_virt;
+ qed_ooo_put_free_buffer(p_hwfn, p_hwfn->p_ooo_info, p_buf);
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_LL2,
+ "Allocated [%04x] LL2 OOO buffers [each of size 0x%08x]\n",
+ p_ll2_info->input.rx_num_ooo_buffers, p_buf->rx_buffer_size);
+
+out:
return rc;
}
+static enum core_error_handle
+qed_ll2_get_error_choice(enum qed_ll2_error_handle err)
+{
+ switch (err) {
+ case QED_LL2_DROP_PACKET:
+ return LL2_DROP_PACKET;
+ case QED_LL2_DO_NOTHING:
+ return LL2_DO_NOTHING;
+ case QED_LL2_ASSERT:
+ return LL2_ASSERT;
+ default:
+ return LL2_DO_NOTHING;
+ }
+}
+
int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
- struct qed_ll2_conn *p_params,
- u16 rx_num_desc,
- u16 tx_num_desc,
- u8 *p_connection_handle)
+ struct qed_ll2_acquire_data *data)
{
qed_int_comp_cb_t comp_rx_cb, comp_tx_cb;
struct qed_ll2_info *p_ll2_info = NULL;
+ u8 i, *p_tx_max;
int rc;
- u8 i;
- if (!p_connection_handle || !p_hwfn->p_ll2_info)
+ if (!data->p_connection_handle || !p_hwfn->p_ll2_info)
return -EINVAL;
/* Find a free connection to be used */
@@ -1356,23 +1316,33 @@ int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
if (!p_ll2_info)
return -EBUSY;
- p_ll2_info->conn = *p_params;
+ memcpy(&p_ll2_info->input, &data->input, sizeof(p_ll2_info->input));
- rc = qed_ll2_acquire_connection_rx(p_hwfn, p_ll2_info, rx_num_desc);
+ p_ll2_info->tx_dest = (data->input.tx_dest == QED_LL2_TX_DEST_NW) ?
+ CORE_TX_DEST_NW : CORE_TX_DEST_LB;
+
+ /* Correct maximum number of Tx BDs */
+ p_tx_max = &p_ll2_info->input.tx_max_bds_per_packet;
+ if (*p_tx_max == 0)
+ *p_tx_max = CORE_LL2_TX_MAX_BDS_PER_PACKET;
+ else
+ *p_tx_max = min_t(u8, *p_tx_max,
+ CORE_LL2_TX_MAX_BDS_PER_PACKET);
+ rc = qed_ll2_acquire_connection_rx(p_hwfn, p_ll2_info);
if (rc)
goto q_allocate_fail;
- rc = qed_ll2_acquire_connection_tx(p_hwfn, p_ll2_info, tx_num_desc);
+ rc = qed_ll2_acquire_connection_tx(p_hwfn, p_ll2_info);
if (rc)
goto q_allocate_fail;
rc = qed_ll2_acquire_connection_ooo(p_hwfn, p_ll2_info,
- rx_num_desc * 2, p_params->mtu);
+ data->input.mtu);
if (rc)
goto q_allocate_fail;
/* Register callbacks for the Rx/Tx queues */
- if (p_params->conn_type == QED_LL2_TYPE_ISCSI_OOO) {
+ if (data->input.conn_type == QED_LL2_TYPE_ISCSI_OOO) {
comp_rx_cb = qed_ll2_lb_rxq_completion;
comp_tx_cb = qed_ll2_lb_txq_completion;
} else {
@@ -1380,7 +1350,7 @@ int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
comp_tx_cb = qed_ll2_txq_completion;
}
- if (rx_num_desc) {
+ if (data->input.rx_num_desc) {
qed_int_register_cb(p_hwfn, comp_rx_cb,
&p_hwfn->p_ll2_info[i],
&p_ll2_info->rx_queue.rx_sb_index,
@@ -1388,7 +1358,7 @@ int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
p_ll2_info->rx_queue.b_cb_registred = true;
}
- if (tx_num_desc) {
+ if (data->input.tx_num_desc) {
qed_int_register_cb(p_hwfn,
comp_tx_cb,
&p_hwfn->p_ll2_info[i],
@@ -1397,7 +1367,7 @@ int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
p_ll2_info->tx_queue.b_cb_registred = true;
}
- *p_connection_handle = i;
+ *data->p_connection_handle = i;
return rc;
q_allocate_fail:
@@ -1408,18 +1378,21 @@ int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
static int qed_ll2_establish_connection_rx(struct qed_hwfn *p_hwfn,
struct qed_ll2_info *p_ll2_conn)
{
+ enum qed_ll2_error_handle error_input;
+ enum core_error_handle error_mode;
u8 action_on_error = 0;
if (!QED_LL2_RX_REGISTERED(p_ll2_conn))
return 0;
DIRECT_REG_WR(p_ll2_conn->rx_queue.set_prod_addr, 0x0);
-
+ error_input = p_ll2_conn->input.ai_err_packet_too_big;
+ error_mode = qed_ll2_get_error_choice(error_input);
SET_FIELD(action_on_error,
- CORE_RX_ACTION_ON_ERROR_PACKET_TOO_BIG,
- p_ll2_conn->conn.ai_err_packet_too_big);
- SET_FIELD(action_on_error,
- CORE_RX_ACTION_ON_ERROR_NO_BUFF, p_ll2_conn->conn.ai_err_no_buf);
+ CORE_RX_ACTION_ON_ERROR_PACKET_TOO_BIG, error_mode);
+ error_input = p_ll2_conn->input.ai_err_no_buf;
+ error_mode = qed_ll2_get_error_choice(error_input);
+ SET_FIELD(action_on_error, CORE_RX_ACTION_ON_ERROR_NO_BUFF, error_mode);
return qed_sp_ll2_rx_queue_start(p_hwfn, p_ll2_conn, action_on_error);
}
@@ -1503,7 +1476,7 @@ int qed_ll2_establish_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
qed_ll2_establish_connection_ooo(p_hwfn, p_ll2_conn);
- if (p_ll2_conn->conn.conn_type == QED_LL2_TYPE_FCOE) {
+ if (p_ll2_conn->input.conn_type == QED_LL2_TYPE_FCOE) {
qed_llh_add_protocol_filter(p_hwfn, p_ptt,
0x8906, 0,
QED_LLH_FILTER_ETHERTYPE);
@@ -1667,7 +1640,7 @@ qed_ll2_prepare_tx_packet_set_bd(struct qed_hwfn *p_hwfn,
"LL2 [q 0x%02x cid 0x%08x type 0x%08x] Tx Producer at [0x%04x] - set with a %04x bytes %02x BDs buffer at %08x:%08x\n",
p_ll2->queue_id,
p_ll2->cid,
- p_ll2->conn.conn_type,
+ p_ll2->input.conn_type,
prod_idx,
pkt->first_frag_len,
pkt->num_of_bds,
@@ -1742,7 +1715,8 @@ static void qed_ll2_tx_packet_notify(struct qed_hwfn *p_hwfn,
(NETIF_MSG_TX_QUEUED | QED_MSG_LL2),
"LL2 [q 0x%02x cid 0x%08x type 0x%08x] Doorbelled [producer 0x%04x]\n",
p_ll2_conn->queue_id,
- p_ll2_conn->cid, p_ll2_conn->conn.conn_type, db_msg.spq_prod);
+ p_ll2_conn->cid,
+ p_ll2_conn->input.conn_type, db_msg.spq_prod);
}
int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
@@ -1866,10 +1840,10 @@ int qed_ll2_terminate_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
qed_ll2_rxq_flush(p_hwfn, connection_handle);
}
- if (p_ll2_conn->conn.conn_type == QED_LL2_TYPE_ISCSI_OOO)
+ if (p_ll2_conn->input.conn_type == QED_LL2_TYPE_ISCSI_OOO)
qed_ooo_release_all_isles(p_hwfn, p_hwfn->p_ooo_info);
- if (p_ll2_conn->conn.conn_type == QED_LL2_TYPE_FCOE) {
+ if (p_ll2_conn->input.conn_type == QED_LL2_TYPE_FCOE) {
qed_llh_remove_protocol_filter(p_hwfn, p_ptt,
0x8906, 0,
QED_LLH_FILTER_ETHERTYPE);
@@ -2054,11 +2028,68 @@ static void qed_ll2_register_cb_ops(struct qed_dev *cdev,
cdev->ll2->cb_cookie = cookie;
}
+static void qed_ll2_set_conn_data(struct qed_dev *cdev,
+ struct qed_ll2_acquire_data *data,
+ struct qed_ll2_params *params,
+ enum qed_ll2_conn_type conn_type,
+ u8 *handle, bool lb, u8 gsi_enable)
+{
+ memset(data, 0, sizeof(*data));
+
+ data->input.conn_type = conn_type;
+ data->input.mtu = params->mtu;
+ data->input.rx_num_desc = QED_LL2_RX_SIZE;
+ data->input.rx_drop_ttl0_flg = params->drop_ttl0_packets;
+ data->input.rx_vlan_removal_en = params->rx_vlan_stripping;
+ data->input.tx_num_desc = QED_LL2_TX_SIZE;
+ data->input.gsi_enable = gsi_enable;
+ data->p_connection_handle = handle;
+ if (lb) {
+ data->input.tx_tc = OOO_LB_TC;
+ data->input.tx_dest = QED_LL2_TX_DEST_LB;
+ } else {
+ data->input.tx_tc = 0;
+ data->input.tx_dest = QED_LL2_TX_DEST_NW;
+ }
+}
+
+static int qed_ll2_start_ooo(struct qed_dev *cdev,
+ struct qed_ll2_params *params)
+{
+ struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ u8 *handle = &hwfn->pf_params.iscsi_pf_params.ll2_ooo_queue_id;
+ struct qed_ll2_acquire_data data;
+ int rc;
+
+ qed_ll2_set_conn_data(cdev, &data, params,
+ QED_LL2_TYPE_ISCSI_OOO, handle, true, 0);
+
+ rc = qed_ll2_acquire_connection(hwfn, &data);
+ if (rc) {
+ DP_INFO(cdev, "Failed to acquire LL2 OOO connection\n");
+ goto out;
+ }
+
+ rc = qed_ll2_establish_connection(hwfn, *handle);
+ if (rc) {
+ DP_INFO(cdev, "Failed to establist LL2 OOO connection\n");
+ goto fail;
+ }
+
+ return 0;
+
+fail:
+ qed_ll2_release_connection(hwfn, *handle);
+out:
+ *handle = QED_LL2_UNUSED_HANDLE;
+ return rc;
+}
+
static int qed_ll2_start(struct qed_dev *cdev, struct qed_ll2_params *params)
{
- struct qed_ll2_conn ll2_info;
struct qed_ll2_buffer *buffer, *tmp_buffer;
enum qed_ll2_conn_type conn_type;
+ struct qed_ll2_acquire_data data;
struct qed_ptt *p_ptt;
int rc, i;
u8 gsi_enable = 1;
@@ -2106,20 +2137,10 @@ static int qed_ll2_start(struct qed_dev *cdev, struct qed_ll2_params *params)
conn_type = QED_LL2_TYPE_TEST;
}
- /* Prepare the temporary ll2 information */
- memset(&ll2_info, 0, sizeof(ll2_info));
-
- ll2_info.conn_type = conn_type;
- ll2_info.mtu = params->mtu;
- ll2_info.rx_drop_ttl0_flg = params->drop_ttl0_packets;
- ll2_info.rx_vlan_removal_en = params->rx_vlan_stripping;
- ll2_info.tx_tc = 0;
- ll2_info.tx_dest = CORE_TX_DEST_NW;
- ll2_info.gsi_enable = gsi_enable;
+ qed_ll2_set_conn_data(cdev, &data, params, conn_type,
+ &cdev->ll2->handle, false, gsi_enable);
- rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &ll2_info,
- QED_LL2_RX_SIZE, QED_LL2_TX_SIZE,
- &cdev->ll2->handle);
+ rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &data);
if (rc) {
DP_INFO(cdev, "Failed to acquire LL2 connection\n");
goto fail;
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.h b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
index 96af50b..3caaad5 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
@@ -47,17 +47,6 @@
#define QED_MAX_NUM_OF_LL2_CONNECTIONS (4)
-enum qed_ll2_conn_type {
- QED_LL2_TYPE_FCOE,
- QED_LL2_TYPE_ISCSI,
- QED_LL2_TYPE_TEST,
- QED_LL2_TYPE_ISCSI_OOO,
- QED_LL2_TYPE_RESERVED2,
- QED_LL2_TYPE_ROCE,
- QED_LL2_TYPE_RESERVED3,
- MAX_QED_LL2_RX_CONN_TYPE
-};
-
struct qed_ll2_rx_packet {
struct list_head list_entry;
struct core_rx_bd_with_buff_len *rxq_bd;
@@ -123,27 +112,17 @@ struct qed_ll2_tx_queue {
bool b_completing_packet;
};
-struct qed_ll2_conn {
- enum qed_ll2_conn_type conn_type;
- u16 mtu;
- u8 rx_drop_ttl0_flg;
- u8 rx_vlan_removal_en;
- u8 tx_tc;
- enum core_tx_dest tx_dest;
- enum core_error_handle ai_err_packet_too_big;
- enum core_error_handle ai_err_no_buf;
- u8 gsi_enable;
-};
-
struct qed_ll2_info {
/* Lock protecting the state of LL2 */
struct mutex mutex;
- struct qed_ll2_conn conn;
+
+ struct qed_ll2_acquire_data_inputs input;
u32 cid;
u8 my_id;
u8 queue_id;
u8 tx_stats_id;
bool b_active;
+ enum core_tx_dest tx_dest;
u8 tx_stats_en;
struct qed_ll2_rx_queue rx_queue;
struct qed_ll2_tx_queue tx_queue;
@@ -154,20 +133,13 @@ struct qed_ll2_info {
* starts rx & tx (if relevant) queues pair. Provides
* connecion handler as output parameter.
*
- * @param p_hwfn
- * @param p_params Contain various configuration properties
- * @param rx_num_desc
- * @param tx_num_desc
- *
- * @param p_connection_handle Output container for LL2 connection's handle
*
- * @return 0 on success, failure otherwise
+ * @param p_hwfn
+ * @param data - describes connection parameters
+ * @return int
*/
int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
- struct qed_ll2_conn *p_params,
- u16 rx_num_desc,
- u16 tx_num_desc,
- u8 *p_connection_handle);
+ struct qed_ll2_acquire_data *data);
/**
* @brief qed_ll2_establish_connection - start previously
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index afc63c0..33abcf1 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -2818,7 +2818,7 @@ static int qed_roce_ll2_start(struct qed_dev *cdev,
{
struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
struct qed_roce_ll2_info *roce_ll2;
- struct qed_ll2_conn ll2_params;
+ struct qed_ll2_acquire_data data;
int rc;
if (!params) {
@@ -2844,25 +2844,26 @@ static int qed_roce_ll2_start(struct qed_dev *cdev,
DP_ERR(cdev, "qed roce ll2 start: failed memory allocation\n");
return -ENOMEM;
}
+
roce_ll2->handle = QED_LL2_UNUSED_HANDLE;
roce_ll2->cbs = params->cbs;
roce_ll2->cb_cookie = params->cb_cookie;
mutex_init(&roce_ll2->lock);
- memset(&ll2_params, 0, sizeof(ll2_params));
- ll2_params.conn_type = QED_LL2_TYPE_ROCE;
- ll2_params.mtu = params->mtu;
- ll2_params.rx_drop_ttl0_flg = true;
- ll2_params.rx_vlan_removal_en = false;
- ll2_params.tx_dest = CORE_TX_DEST_NW;
- ll2_params.ai_err_packet_too_big = LL2_DROP_PACKET;
- ll2_params.ai_err_no_buf = LL2_DROP_PACKET;
- ll2_params.gsi_enable = true;
-
- rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &ll2_params,
- params->max_rx_buffers,
- params->max_tx_buffers,
- &roce_ll2->handle);
+ memset(&data, 0, sizeof(data));
+ data.input.conn_type = QED_LL2_TYPE_ROCE;
+ data.input.mtu = params->mtu;
+ data.input.rx_num_desc = params->max_rx_buffers;
+ data.input.tx_num_desc = params->max_tx_buffers;
+ data.input.rx_drop_ttl0_flg = true;
+ data.input.rx_vlan_removal_en = false;
+ data.input.tx_dest = QED_LL2_TX_DEST_NW;
+ data.input.ai_err_packet_too_big = LL2_DROP_PACKET;
+ data.input.ai_err_no_buf = LL2_DROP_PACKET;
+ data.p_connection_handle = &roce_ll2->handle;
+ data.input.gsi_enable = true;
+
+ rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &data);
if (rc) {
DP_ERR(cdev,
"qed roce ll2 start: failed to acquire LL2 connection (rc=%d)\n",
diff --git a/include/linux/qed/qed_ll2_if.h b/include/linux/qed/qed_ll2_if.h
index 5145278..a8f58d9 100644
--- a/include/linux/qed/qed_ll2_if.h
+++ b/include/linux/qed/qed_ll2_if.h
@@ -43,6 +43,17 @@
#include <linux/slab.h>
#include <linux/qed/qed_if.h>
+enum qed_ll2_conn_type {
+ QED_LL2_TYPE_FCOE,
+ QED_LL2_TYPE_ISCSI,
+ QED_LL2_TYPE_TEST,
+ QED_LL2_TYPE_ISCSI_OOO,
+ QED_LL2_TYPE_RESERVED2,
+ QED_LL2_TYPE_ROCE,
+ QED_LL2_TYPE_RESERVED3,
+ MAX_QED_LL2_RX_CONN_TYPE
+};
+
enum qed_ll2_roce_flavor_type {
QED_LL2_ROCE,
QED_LL2_RROCE,
@@ -55,6 +66,12 @@ enum qed_ll2_tx_dest {
QED_LL2_TX_DEST_MAX
};
+enum qed_ll2_error_handle {
+ QED_LL2_DROP_PACKET,
+ QED_LL2_DO_NOTHING,
+ QED_LL2_ASSERT,
+};
+
struct qed_ll2_stats {
u64 gsi_invalid_hdr;
u64 gsi_invalid_pkt_length;
@@ -104,6 +121,27 @@ struct qed_ll2_comp_rx_data {
u16 qp_id;
};
+struct qed_ll2_acquire_data_inputs {
+ enum qed_ll2_conn_type conn_type;
+ u16 mtu;
+ u16 rx_num_desc;
+ u16 rx_num_ooo_buffers;
+ u8 rx_drop_ttl0_flg;
+ u8 rx_vlan_removal_en;
+ u16 tx_num_desc;
+ u8 tx_max_bds_per_packet;
+ u8 tx_tc;
+ enum qed_ll2_tx_dest tx_dest;
+ enum qed_ll2_error_handle ai_err_packet_too_big;
+ enum qed_ll2_error_handle ai_err_no_buf;
+ u8 gsi_enable;
+};
+
+struct qed_ll2_acquire_data {
+ struct qed_ll2_acquire_data_inputs input;
+ /* Output container for LL2 connection's handle */
+ u8 *p_connection_handle;
+};
struct qed_ll2_tx_pkt_info {
u8 num_of_bds;
u16 vlan;
--
2.9.4
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 16+ messages in thread
* [PATCH net-next 4/8] qed: LL2 code relocations
2017-06-08 16:13 [PATCH net-next 0/8] qed*: Light L2 updates Yuval Mintz
2017-06-08 16:13 ` [PATCH net-next 1/8] qed: LL2 to use packed information for tx Yuval Mintz
2017-06-08 16:13 ` [PATCH net-next 2/8] qed: Revise ll2 Rx completion Yuval Mintz
@ 2017-06-08 16:13 ` Yuval Mintz
2017-06-08 16:13 ` [PATCH net-next 5/8] qed*: LL2 callback operations Yuval Mintz
` (2 subsequent siblings)
5 siblings, 0 replies; 16+ messages in thread
From: Yuval Mintz @ 2017-06-08 16:13 UTC (permalink / raw)
To: netdev, davem; +Cc: linux-rdma, Michal.Kalderon, Yuval Mintz
Instead of having the OOO logic packetd, divide it with rest of code
according to establish/release flows.
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
---
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 58 +++++++++++++++----------------
1 file changed, 28 insertions(+), 30 deletions(-)
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
index fcf4ea9..f3aad61 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
@@ -896,36 +896,6 @@ static int qed_ll2_lb_txq_completion(struct qed_hwfn *p_hwfn, void *p_cookie)
return 0;
}
-static void
-qed_ll2_establish_connection_ooo(struct qed_hwfn *p_hwfn,
- struct qed_ll2_info *p_ll2_conn)
-{
- if (p_ll2_conn->input.conn_type != QED_LL2_TYPE_ISCSI_OOO)
- return;
-
- qed_ooo_release_all_isles(p_hwfn, p_hwfn->p_ooo_info);
- qed_ooo_submit_rx_buffers(p_hwfn, p_ll2_conn);
-}
-
-static void qed_ll2_release_connection_ooo(struct qed_hwfn *p_hwfn,
- struct qed_ll2_info *p_ll2_conn)
-{
- struct qed_ooo_buffer *p_buffer;
-
- if (p_ll2_conn->input.conn_type != QED_LL2_TYPE_ISCSI_OOO)
- return;
-
- qed_ooo_release_all_isles(p_hwfn, p_hwfn->p_ooo_info);
- while ((p_buffer = qed_ooo_get_free_buffer(p_hwfn,
- p_hwfn->p_ooo_info))) {
- dma_free_coherent(&p_hwfn->cdev->pdev->dev,
- p_buffer->rx_buffer_size,
- p_buffer->rx_buffer_virt_addr,
- p_buffer->rx_buffer_phys_addr);
- kfree(p_buffer);
- }
-}
-
static void qed_ll2_stop_ooo(struct qed_dev *cdev)
{
struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
@@ -1397,6 +1367,16 @@ static int qed_ll2_establish_connection_rx(struct qed_hwfn *p_hwfn,
return qed_sp_ll2_rx_queue_start(p_hwfn, p_ll2_conn, action_on_error);
}
+static void
+qed_ll2_establish_connection_ooo(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn)
+{
+ if (p_ll2_conn->input.conn_type != QED_LL2_TYPE_ISCSI_OOO)
+ return;
+
+ qed_ooo_release_all_isles(p_hwfn, p_hwfn->p_ooo_info);
+ qed_ooo_submit_rx_buffers(p_hwfn, p_ll2_conn);
+}
int qed_ll2_establish_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
{
struct qed_ll2_info *p_ll2_conn;
@@ -1857,6 +1837,24 @@ int qed_ll2_terminate_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
return rc;
}
+static void qed_ll2_release_connection_ooo(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn)
+{
+ struct qed_ooo_buffer *p_buffer;
+
+ if (p_ll2_conn->input.conn_type != QED_LL2_TYPE_ISCSI_OOO)
+ return;
+
+ qed_ooo_release_all_isles(p_hwfn, p_hwfn->p_ooo_info);
+ while ((p_buffer = qed_ooo_get_free_buffer(p_hwfn,
+ p_hwfn->p_ooo_info))) {
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ p_buffer->rx_buffer_size,
+ p_buffer->rx_buffer_virt_addr,
+ p_buffer->rx_buffer_phys_addr);
+ kfree(p_buffer);
+ }
+}
void qed_ll2_release_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
{
struct qed_ll2_info *p_ll2_conn = NULL;
--
2.9.4
^ permalink raw reply related [flat|nested] 16+ messages in thread
* [PATCH net-next 5/8] qed*: LL2 callback operations
2017-06-08 16:13 [PATCH net-next 0/8] qed*: Light L2 updates Yuval Mintz
` (2 preceding siblings ...)
2017-06-08 16:13 ` [PATCH net-next 4/8] qed: LL2 code relocations Yuval Mintz
@ 2017-06-08 16:13 ` Yuval Mintz
2017-06-08 16:13 ` [PATCH net-next 6/8] qed: No need for LL2 frags indication Yuval Mintz
[not found] ` <20170608161323.7597-1-Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
5 siblings, 0 replies; 16+ messages in thread
From: Yuval Mintz @ 2017-06-08 16:13 UTC (permalink / raw)
To: netdev, davem; +Cc: linux-rdma, Michal.Kalderon, Yuval Mintz
From: Michal Kalderon <Michal.Kalderon@cavium.com>
LL2 today is interrupt driven - when tx/rx completion arrives [or any
other indication], qed needs to operate on the connection and pass
the information to the protocol-driver [or internal qed consumer].
Since we have several flavors of ll2 employeed by the driver,
each handler needs to do an if-else to determine the right functionality
to use based on the connection type.
In order to make things more scalable [given that we're going to add
additional types of ll2 flavors] move the infrastrucutre into using
a callback-based approach - the callbacks would be provided as part
of the connection's initialization parameters.
Signed-off-by: Michal Kalderon <Michal.Kalderon@cavium.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
---
drivers/infiniband/hw/qedr/main.c | 6 +-
drivers/infiniband/hw/qedr/qedr.h | 2 +
drivers/infiniband/hw/qedr/qedr_cm.c | 238 ++++++++++++++++------
drivers/net/ethernet/qlogic/qed/qed.h | 1 -
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 197 +++++++++----------
drivers/net/ethernet/qlogic/qed/qed_ll2.h | 34 ++--
drivers/net/ethernet/qlogic/qed/qed_roce.c | 304 ++---------------------------
drivers/net/ethernet/qlogic/qed/qed_roce.h | 43 ----
include/linux/qed/qed_ll2_if.h | 36 ++++
include/linux/qed/qed_roce_if.h | 80 +++-----
10 files changed, 384 insertions(+), 557 deletions(-)
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 6a72095..485c1fe 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -886,9 +886,9 @@ static void qedr_mac_address_change(struct qedr_dev *dev)
memcpy(&sgid->raw[8], guid, sizeof(guid));
/* Update LL2 */
- rc = dev->ops->roce_ll2_set_mac_filter(dev->cdev,
- dev->gsi_ll2_mac_address,
- dev->ndev->dev_addr);
+ rc = dev->ops->ll2_set_mac_filter(dev->cdev,
+ dev->gsi_ll2_mac_address,
+ dev->ndev->dev_addr);
ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr);
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index aa08c76..80333ec 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -150,6 +150,8 @@ struct qedr_dev {
u32 dp_module;
u8 dp_level;
u8 num_hwfns;
+ u8 gsi_ll2_handle;
+
uint wq_multiplier;
u8 gsi_ll2_mac_address[ETH_ALEN];
int gsi_qp_created;
diff --git a/drivers/infiniband/hw/qedr/qedr_cm.c b/drivers/infiniband/hw/qedr/qedr_cm.c
index d86dbe8..eb3dce7 100644
--- a/drivers/infiniband/hw/qedr/qedr_cm.c
+++ b/drivers/infiniband/hw/qedr/qedr_cm.c
@@ -64,9 +64,14 @@ void qedr_store_gsi_qp_cq(struct qedr_dev *dev, struct qedr_qp *qp,
dev->gsi_qp = qp;
}
-void qedr_ll2_tx_cb(void *_qdev, struct qed_roce_ll2_packet *pkt)
+void qedr_ll2_complete_tx_packet(void *cxt,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment, bool b_last_packet)
{
- struct qedr_dev *dev = (struct qedr_dev *)_qdev;
+ struct qedr_dev *dev = (struct qedr_dev *)cxt;
+ struct qed_roce_ll2_packet *pkt = cookie;
struct qedr_cq *cq = dev->gsi_sqcq;
struct qedr_qp *qp = dev->gsi_qp;
unsigned long flags;
@@ -88,20 +93,26 @@ void qedr_ll2_tx_cb(void *_qdev, struct qed_roce_ll2_packet *pkt)
(*cq->ibcq.comp_handler) (&cq->ibcq, cq->ibcq.cq_context);
}
-void qedr_ll2_rx_cb(void *_dev, struct qed_roce_ll2_packet *pkt,
- struct qed_roce_ll2_rx_params *params)
+void qedr_ll2_complete_rx_packet(void *cxt,
+ struct qed_ll2_comp_rx_data *data)
{
- struct qedr_dev *dev = (struct qedr_dev *)_dev;
+ struct qedr_dev *dev = (struct qedr_dev *)cxt;
struct qedr_cq *cq = dev->gsi_rqcq;
struct qedr_qp *qp = dev->gsi_qp;
unsigned long flags;
spin_lock_irqsave(&qp->q_lock, flags);
- qp->rqe_wr_id[qp->rq.gsi_cons].rc = params->rc;
- qp->rqe_wr_id[qp->rq.gsi_cons].vlan_id = params->vlan_id;
- qp->rqe_wr_id[qp->rq.gsi_cons].sg_list[0].length = pkt->payload[0].len;
- ether_addr_copy(qp->rqe_wr_id[qp->rq.gsi_cons].smac, params->smac);
+ qp->rqe_wr_id[qp->rq.gsi_cons].rc = data->u.data_length_error ?
+ -EINVAL : 0;
+ qp->rqe_wr_id[qp->rq.gsi_cons].vlan_id = data->vlan;
+ /* note: length stands for data length i.e. GRH is excluded */
+ qp->rqe_wr_id[qp->rq.gsi_cons].sg_list[0].length =
+ data->length.data_length;
+ *((u32 *)&qp->rqe_wr_id[qp->rq.gsi_cons].smac[0]) =
+ ntohl(data->opaque_data_0);
+ *((u16 *)&qp->rqe_wr_id[qp->rq.gsi_cons].smac[4]) =
+ ntohs((u16)data->opaque_data_1);
qedr_inc_sw_gsi_cons(&qp->rq);
@@ -111,6 +122,14 @@ void qedr_ll2_rx_cb(void *_dev, struct qed_roce_ll2_packet *pkt,
(*cq->ibcq.comp_handler) (&cq->ibcq, cq->ibcq.cq_context);
}
+void qedr_ll2_release_rx_packet(void *cxt,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t rx_buf_addr, bool b_last_packet)
+{
+ /* Do nothing... */
+}
+
static void qedr_destroy_gsi_cq(struct qedr_dev *dev,
struct ib_qp_init_attr *attrs)
{
@@ -159,27 +178,159 @@ static inline int qedr_check_gsi_qp_attrs(struct qedr_dev *dev,
return 0;
}
+static int qedr_ll2_post_tx(struct qedr_dev *dev,
+ struct qed_roce_ll2_packet *pkt)
+{
+ enum qed_ll2_roce_flavor_type roce_flavor;
+ struct qed_ll2_tx_pkt_info ll2_tx_pkt;
+ int rc;
+ int i;
+
+ memset(&ll2_tx_pkt, 0, sizeof(ll2_tx_pkt));
+
+ roce_flavor = (pkt->roce_mode == ROCE_V1) ?
+ QED_LL2_ROCE : QED_LL2_RROCE;
+
+ if (pkt->roce_mode == ROCE_V2_IPV4)
+ ll2_tx_pkt.enable_ip_cksum = 1;
+
+ ll2_tx_pkt.num_of_bds = 1 /* hdr */ + pkt->n_seg;
+ ll2_tx_pkt.vlan = 0;
+ ll2_tx_pkt.tx_dest = pkt->tx_dest;
+ ll2_tx_pkt.qed_roce_flavor = roce_flavor;
+ ll2_tx_pkt.first_frag = pkt->header.baddr;
+ ll2_tx_pkt.first_frag_len = pkt->header.len;
+ ll2_tx_pkt.cookie = pkt;
+
+ /* tx header */
+ rc = dev->ops->ll2_prepare_tx_packet(dev->rdma_ctx,
+ dev->gsi_ll2_handle,
+ &ll2_tx_pkt, 1);
+ if (rc) {
+ /* TX failed while posting header - release resources */
+ dma_free_coherent(&dev->pdev->dev, pkt->header.len,
+ pkt->header.vaddr, pkt->header.baddr);
+ kfree(pkt);
+
+ DP_ERR(dev, "roce ll2 tx: header failed (rc=%d)\n", rc);
+ return rc;
+ }
+
+ /* tx payload */
+ for (i = 0; i < pkt->n_seg; i++) {
+ rc = dev->ops->ll2_set_fragment_of_tx_packet(
+ dev->rdma_ctx,
+ dev->gsi_ll2_handle,
+ pkt->payload[i].baddr,
+ pkt->payload[i].len);
+
+ if (rc) {
+ /* if failed not much to do here, partial packet has
+ * been posted we can't free memory, will need to wait
+ * for completion
+ */
+ DP_ERR(dev, "ll2 tx: payload failed (rc=%d)\n", rc);
+ return rc;
+ }
+ }
+
+ return 0;
+}
+
+int qedr_ll2_stop(struct qedr_dev *dev)
+{
+ int rc;
+
+ if (dev->gsi_ll2_handle == QED_LL2_UNUSED_HANDLE)
+ return 0;
+
+ /* remove LL2 MAC address filter */
+ rc = dev->ops->ll2_set_mac_filter(dev->cdev,
+ dev->gsi_ll2_mac_address, NULL);
+
+ rc = dev->ops->ll2_terminate_connection(dev->rdma_ctx,
+ dev->gsi_ll2_handle);
+ if (rc)
+ DP_ERR(dev, "Failed to terminate LL2 connection (rc=%d)\n", rc);
+
+ dev->ops->ll2_release_connection(dev->rdma_ctx, dev->gsi_ll2_handle);
+
+ dev->gsi_ll2_handle = QED_LL2_UNUSED_HANDLE;
+
+ return rc;
+}
+
+int qedr_ll2_start(struct qedr_dev *dev,
+ struct ib_qp_init_attr *attrs, struct qedr_qp *qp)
+{
+ struct qed_ll2_acquire_data data;
+ struct qed_ll2_cbs cbs;
+ int rc;
+
+ /* configure and start LL2 */
+ cbs.rx_comp_cb = qedr_ll2_complete_rx_packet;
+ cbs.tx_comp_cb = qedr_ll2_complete_tx_packet;
+ cbs.rx_release_cb = qedr_ll2_release_rx_packet;
+ cbs.tx_release_cb = qedr_ll2_complete_tx_packet;
+ cbs.cookie = dev;
+
+ memset(&data, 0, sizeof(data));
+ data.input.conn_type = QED_LL2_TYPE_ROCE;
+ data.input.mtu = dev->ndev->mtu;
+ data.input.rx_num_desc = attrs->cap.max_recv_wr;
+ data.input.rx_drop_ttl0_flg = true;
+ data.input.rx_vlan_removal_en = false;
+ data.input.tx_num_desc = attrs->cap.max_send_wr;
+ data.input.tx_tc = 0;
+ data.input.tx_dest = QED_LL2_TX_DEST_NW;
+ data.input.ai_err_packet_too_big = QED_LL2_DROP_PACKET;
+ data.input.ai_err_no_buf = QED_LL2_DROP_PACKET;
+ data.input.gsi_enable = 1;
+ data.p_connection_handle = &dev->gsi_ll2_handle;
+ data.cbs = &cbs;
+
+ rc = dev->ops->ll2_acquire_connection(dev->rdma_ctx, &data);
+ if (rc) {
+ DP_ERR(dev,
+ "ll2 start: failed to acquire LL2 connection (rc=%d)\n",
+ rc);
+ return rc;
+ }
+
+ rc = dev->ops->ll2_establish_connection(dev->rdma_ctx,
+ dev->gsi_ll2_handle);
+ if (rc) {
+ DP_ERR(dev,
+ "ll2 start: failed to establish LL2 connection (rc=%d)\n",
+ rc);
+ goto err1;
+ }
+
+ rc = dev->ops->ll2_set_mac_filter(dev->cdev, NULL, dev->ndev->dev_addr);
+ if (rc)
+ goto err2;
+
+ return 0;
+
+err2:
+ dev->ops->ll2_terminate_connection(dev->rdma_ctx, dev->gsi_ll2_handle);
+err1:
+ dev->ops->ll2_release_connection(dev->rdma_ctx, dev->gsi_ll2_handle);
+
+ return rc;
+}
+
struct ib_qp *qedr_create_gsi_qp(struct qedr_dev *dev,
struct ib_qp_init_attr *attrs,
struct qedr_qp *qp)
{
- struct qed_roce_ll2_params ll2_params;
int rc;
rc = qedr_check_gsi_qp_attrs(dev, attrs);
if (rc)
return ERR_PTR(rc);
- /* configure and start LL2 */
- memset(&ll2_params, 0, sizeof(ll2_params));
- ll2_params.max_tx_buffers = attrs->cap.max_send_wr;
- ll2_params.max_rx_buffers = attrs->cap.max_recv_wr;
- ll2_params.cbs.tx_cb = qedr_ll2_tx_cb;
- ll2_params.cbs.rx_cb = qedr_ll2_rx_cb;
- ll2_params.cb_cookie = (void *)dev;
- ll2_params.mtu = dev->ndev->mtu;
- ether_addr_copy(ll2_params.mac_address, dev->ndev->dev_addr);
- rc = dev->ops->roce_ll2_start(dev->cdev, &ll2_params);
+ rc = qedr_ll2_start(dev, attrs, qp);
if (rc) {
DP_ERR(dev, "create gsi qp: failed on ll2 start. rc=%d\n", rc);
return ERR_PTR(rc);
@@ -214,7 +365,7 @@ struct ib_qp *qedr_create_gsi_qp(struct qedr_dev *dev,
err:
kfree(qp->rqe_wr_id);
- rc = dev->ops->roce_ll2_stop(dev->cdev);
+ rc = qedr_ll2_stop(dev);
if (rc)
DP_ERR(dev, "create gsi qp: failed destroy on create\n");
@@ -223,15 +374,7 @@ struct ib_qp *qedr_create_gsi_qp(struct qedr_dev *dev,
int qedr_destroy_gsi_qp(struct qedr_dev *dev)
{
- int rc;
-
- rc = dev->ops->roce_ll2_stop(dev->cdev);
- if (rc)
- DP_ERR(dev, "destroy gsi qp: failed (rc=%d)\n", rc);
- else
- DP_DEBUG(dev, QEDR_MSG_GSI, "destroy gsi qp: success\n");
-
- return rc;
+ return qedr_ll2_stop(dev);
}
#define QEDR_MAX_UD_HEADER_SIZE (100)
@@ -421,7 +564,6 @@ int qedr_gsi_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
{
struct qed_roce_ll2_packet *pkt = NULL;
struct qedr_qp *qp = get_qedr_qp(ibqp);
- struct qed_roce_ll2_tx_params params;
struct qedr_dev *dev = qp->dev;
unsigned long flags;
int rc;
@@ -449,8 +591,6 @@ int qedr_gsi_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
goto err;
}
- memset(¶ms, 0, sizeof(params));
-
spin_lock_irqsave(&qp->q_lock, flags);
rc = qedr_gsi_build_packet(dev, qp, wr, &pkt);
@@ -459,7 +599,8 @@ int qedr_gsi_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
goto err;
}
- rc = dev->ops->roce_ll2_tx(dev->cdev, pkt, ¶ms);
+ rc = qedr_ll2_post_tx(dev, pkt);
+
if (!rc) {
qp->wqe_wr_id[qp->sq.prod].wr_id = wr->wr_id;
qedr_inc_sw_prod(&qp->sq);
@@ -467,17 +608,6 @@ int qedr_gsi_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
"gsi post send: opcode=%d, in_irq=%ld, irqs_disabled=%d, wr_id=%llx\n",
wr->opcode, in_irq(), irqs_disabled(), wr->wr_id);
} else {
- if (rc == QED_ROCE_TX_HEAD_FAILURE) {
- /* TX failed while posting header - release resources */
- dma_free_coherent(&dev->pdev->dev, pkt->header.len,
- pkt->header.vaddr, pkt->header.baddr);
- kfree(pkt);
- } else if (rc == QED_ROCE_TX_FRAG_FAILURE) {
- /* NTD since TX failed while posting a fragment. We will
- * release the resources on TX callback
- */
- }
-
DP_ERR(dev, "gsi post send: failed to transmit (rc=%d)\n", rc);
rc = -EAGAIN;
*bad_wr = wr;
@@ -504,10 +634,8 @@ int qedr_gsi_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
{
struct qedr_dev *dev = get_qedr_dev(ibqp->device);
struct qedr_qp *qp = get_qedr_qp(ibqp);
- struct qed_roce_ll2_buffer buf;
unsigned long flags;
- int status = 0;
- int rc;
+ int rc = 0;
if ((qp->state != QED_ROCE_QP_STATE_RTR) &&
(qp->state != QED_ROCE_QP_STATE_RTS)) {
@@ -518,8 +646,6 @@ int qedr_gsi_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
return -EINVAL;
}
- memset(&buf, 0, sizeof(buf));
-
spin_lock_irqsave(&qp->q_lock, flags);
while (wr) {
@@ -530,10 +656,12 @@ int qedr_gsi_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
goto err;
}
- buf.baddr = wr->sg_list[0].addr;
- buf.len = wr->sg_list[0].length;
-
- rc = dev->ops->roce_ll2_post_rx_buffer(dev->cdev, &buf, 0, 1);
+ rc = dev->ops->ll2_post_rx_buffer(dev->rdma_ctx,
+ dev->gsi_ll2_handle,
+ wr->sg_list[0].addr,
+ wr->sg_list[0].length,
+ 0 /* cookie */,
+ 1 /* notify_fw */);
if (rc) {
DP_ERR(dev,
"gsi post recv: failed to post rx buffer (rc=%d)\n",
@@ -553,7 +681,7 @@ int qedr_gsi_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
spin_unlock_irqrestore(&qp->q_lock, flags);
- return status;
+ return rc;
err:
spin_unlock_irqrestore(&qp->q_lock, flags);
*bad_wr = wr;
diff --git a/drivers/net/ethernet/qlogic/qed/qed.h b/drivers/net/ethernet/qlogic/qed/qed.h
index d7afc42..14b08ee 100644
--- a/drivers/net/ethernet/qlogic/qed/qed.h
+++ b/drivers/net/ethernet/qlogic/qed/qed.h
@@ -552,7 +552,6 @@ struct qed_hwfn {
#endif
struct z_stream_s *stream;
- struct qed_roce_ll2_info *ll2;
};
struct pci_params {
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
index f3aad61..b222b2b 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
@@ -89,13 +89,14 @@ struct qed_ll2_buffer {
dma_addr_t phys_addr;
};
-static void qed_ll2b_complete_tx_packet(struct qed_hwfn *p_hwfn,
+static void qed_ll2b_complete_tx_packet(void *cxt,
u8 connection_handle,
void *cookie,
dma_addr_t first_frag_addr,
bool b_last_fragment,
bool b_last_packet)
{
+ struct qed_hwfn *p_hwfn = cxt;
struct qed_dev *cdev = p_hwfn->cdev;
struct sk_buff *skb = cookie;
@@ -164,9 +165,9 @@ static void qed_ll2_kill_buffers(struct qed_dev *cdev)
qed_ll2_dealloc_buffer(cdev, buffer);
}
-static void qed_ll2b_complete_rx_packet(struct qed_hwfn *p_hwfn,
- struct qed_ll2_comp_rx_data *data)
+void qed_ll2b_complete_rx_packet(void *cxt, struct qed_ll2_comp_rx_data *data)
{
+ struct qed_hwfn *p_hwfn = cxt;
struct qed_ll2_buffer *buffer = data->cookie;
struct qed_dev *cdev = p_hwfn->cdev;
dma_addr_t new_phys_addr;
@@ -327,21 +328,12 @@ static void qed_ll2_txq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
b_last_frag =
p_tx->cur_completing_bd_idx == p_pkt->bd_used;
tx_frag = p_pkt->bds_set[0].tx_frag;
- if (p_ll2_conn->input.gsi_enable)
- qed_ll2b_release_tx_gsi_packet(p_hwfn,
- p_ll2_conn->
- my_id,
- p_pkt->cookie,
- tx_frag,
- b_last_frag,
- b_last_packet);
- else
- qed_ll2b_complete_tx_packet(p_hwfn,
- p_ll2_conn->my_id,
- p_pkt->cookie,
- tx_frag,
- b_last_frag,
- b_last_packet);
+ p_ll2_conn->cbs.tx_release_cb(p_ll2_conn->cbs.cookie,
+ p_ll2_conn->my_id,
+ p_pkt->cookie,
+ tx_frag,
+ b_last_frag,
+ b_last_packet);
}
}
}
@@ -354,7 +346,6 @@ static int qed_ll2_txq_completion(struct qed_hwfn *p_hwfn, void *p_cookie)
struct qed_ll2_tx_packet *p_pkt;
bool b_last_frag = false;
unsigned long flags;
- dma_addr_t tx_frag;
int rc = -EINVAL;
spin_lock_irqsave(&p_tx->lock, flags);
@@ -395,19 +386,13 @@ static int qed_ll2_txq_completion(struct qed_hwfn *p_hwfn, void *p_cookie)
list_add_tail(&p_pkt->list_entry, &p_tx->free_descq);
spin_unlock_irqrestore(&p_tx->lock, flags);
- tx_frag = p_pkt->bds_set[0].tx_frag;
- if (p_ll2_conn->input.gsi_enable)
- qed_ll2b_complete_tx_gsi_packet(p_hwfn,
- p_ll2_conn->my_id,
- p_pkt->cookie,
- tx_frag,
- b_last_frag, !num_bds);
- else
- qed_ll2b_complete_tx_packet(p_hwfn,
- p_ll2_conn->my_id,
- p_pkt->cookie,
- tx_frag,
- b_last_frag, !num_bds);
+
+ p_ll2_conn->cbs.tx_comp_cb(p_ll2_conn->cbs.cookie,
+ p_ll2_conn->my_id,
+ p_pkt->cookie,
+ p_pkt->bds_set[0].tx_frag,
+ b_last_frag, !num_bds);
+
spin_lock_irqsave(&p_tx->lock, flags);
}
@@ -418,52 +403,16 @@ static int qed_ll2_txq_completion(struct qed_hwfn *p_hwfn, void *p_cookie)
return rc;
}
-static int
-qed_ll2_rxq_completion_gsi(struct qed_hwfn *p_hwfn,
- struct qed_ll2_info *p_ll2_info,
- union core_rx_cqe_union *p_cqe,
- unsigned long lock_flags, bool b_last_cqe)
+static void qed_ll2_rxq_parse_gsi(struct qed_hwfn *p_hwfn,
+ union core_rx_cqe_union *p_cqe,
+ struct qed_ll2_comp_rx_data *data)
{
- struct qed_ll2_rx_queue *p_rx = &p_ll2_info->rx_queue;
- struct qed_ll2_rx_packet *p_pkt = NULL;
- u16 packet_length, parse_flags, vlan;
- u32 src_mac_addrhi;
- u16 src_mac_addrlo;
-
- if (!list_empty(&p_rx->active_descq))
- p_pkt = list_first_entry(&p_rx->active_descq,
- struct qed_ll2_rx_packet, list_entry);
- if (!p_pkt) {
- DP_NOTICE(p_hwfn,
- "GSI Rx completion but active_descq is empty\n");
- return -EIO;
- }
-
- list_del(&p_pkt->list_entry);
- parse_flags = le16_to_cpu(p_cqe->rx_cqe_gsi.parse_flags.flags);
- packet_length = le16_to_cpu(p_cqe->rx_cqe_gsi.data_length);
- vlan = le16_to_cpu(p_cqe->rx_cqe_gsi.vlan);
- src_mac_addrhi = le32_to_cpu(p_cqe->rx_cqe_gsi.src_mac_addrhi);
- src_mac_addrlo = le16_to_cpu(p_cqe->rx_cqe_gsi.src_mac_addrlo);
- if (qed_chain_consume(&p_rx->rxq_chain) != p_pkt->rxq_bd)
- DP_NOTICE(p_hwfn,
- "Mismatch between active_descq and the LL2 Rx chain\n");
- list_add_tail(&p_pkt->list_entry, &p_rx->free_descq);
-
- spin_unlock_irqrestore(&p_rx->lock, lock_flags);
- qed_ll2b_complete_rx_gsi_packet(p_hwfn,
- p_ll2_info->my_id,
- p_pkt->cookie,
- p_pkt->rx_buf_addr,
- packet_length,
- p_cqe->rx_cqe_gsi.data_length_error,
- parse_flags,
- vlan,
- src_mac_addrhi,
- src_mac_addrlo, b_last_cqe);
- spin_lock_irqsave(&p_rx->lock, lock_flags);
-
- return 0;
+ data->parse_flags = le16_to_cpu(p_cqe->rx_cqe_gsi.parse_flags.flags);
+ data->length.data_length = le16_to_cpu(p_cqe->rx_cqe_gsi.data_length);
+ data->vlan = le16_to_cpu(p_cqe->rx_cqe_gsi.vlan);
+ data->opaque_data_0 = le32_to_cpu(p_cqe->rx_cqe_gsi.src_mac_addrhi);
+ data->opaque_data_1 = le16_to_cpu(p_cqe->rx_cqe_gsi.src_mac_addrlo);
+ data->u.data_length_error = p_cqe->rx_cqe_gsi.data_length_error;
}
static void qed_ll2_rxq_parse_reg(struct qed_hwfn *p_hwfn,
@@ -501,7 +450,10 @@ qed_ll2_rxq_handle_completion(struct qed_hwfn *p_hwfn,
}
list_del(&p_pkt->list_entry);
- qed_ll2_rxq_parse_reg(p_hwfn, p_cqe, &data);
+ if (p_cqe->rx_cqe_sp.type == CORE_RX_CQE_TYPE_REGULAR)
+ qed_ll2_rxq_parse_reg(p_hwfn, p_cqe, &data);
+ else
+ qed_ll2_rxq_parse_gsi(p_hwfn, p_cqe, &data);
if (qed_chain_consume(&p_rx->rxq_chain) != p_pkt->rxq_bd)
DP_NOTICE(p_hwfn,
"Mismatch between active_descq and the LL2 Rx chain\n");
@@ -514,7 +466,8 @@ qed_ll2_rxq_handle_completion(struct qed_hwfn *p_hwfn,
data.b_last_packet = b_last_cqe;
spin_unlock_irqrestore(&p_rx->lock, *p_lock_flags);
- qed_ll2b_complete_rx_packet(p_hwfn, &data);
+ p_ll2_conn->cbs.rx_comp_cb(p_ll2_conn->cbs.cookie, &data);
+
spin_lock_irqsave(&p_rx->lock, *p_lock_flags);
return 0;
@@ -552,9 +505,6 @@ static int qed_ll2_rxq_completion(struct qed_hwfn *p_hwfn, void *cookie)
rc = -EINVAL;
break;
case CORE_RX_CQE_TYPE_GSI_OFFLOAD:
- rc = qed_ll2_rxq_completion_gsi(p_hwfn, p_ll2_conn,
- cqe, flags, b_last_cqe);
- break;
case CORE_RX_CQE_TYPE_REGULAR:
rc = qed_ll2_rxq_handle_completion(p_hwfn, p_ll2_conn,
cqe, &flags,
@@ -1244,6 +1194,23 @@ qed_ll2_acquire_connection_ooo(struct qed_hwfn *p_hwfn,
return rc;
}
+static int
+qed_ll2_set_cbs(struct qed_ll2_info *p_ll2_info, const struct qed_ll2_cbs *cbs)
+{
+ if (!cbs || (!cbs->rx_comp_cb ||
+ !cbs->rx_release_cb ||
+ !cbs->tx_comp_cb || !cbs->tx_release_cb || !cbs->cookie))
+ return -EINVAL;
+
+ p_ll2_info->cbs.rx_comp_cb = cbs->rx_comp_cb;
+ p_ll2_info->cbs.rx_release_cb = cbs->rx_release_cb;
+ p_ll2_info->cbs.tx_comp_cb = cbs->tx_comp_cb;
+ p_ll2_info->cbs.tx_release_cb = cbs->tx_release_cb;
+ p_ll2_info->cbs.cookie = cbs->cookie;
+
+ return 0;
+}
+
static enum core_error_handle
qed_ll2_get_error_choice(enum qed_ll2_error_handle err)
{
@@ -1259,9 +1226,9 @@ qed_ll2_get_error_choice(enum qed_ll2_error_handle err)
}
}
-int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
- struct qed_ll2_acquire_data *data)
+int qed_ll2_acquire_connection(void *cxt, struct qed_ll2_acquire_data *data)
{
+ struct qed_hwfn *p_hwfn = cxt;
qed_int_comp_cb_t comp_rx_cb, comp_tx_cb;
struct qed_ll2_info *p_ll2_info = NULL;
u8 i, *p_tx_max;
@@ -1298,6 +1265,13 @@ int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
else
*p_tx_max = min_t(u8, *p_tx_max,
CORE_LL2_TX_MAX_BDS_PER_PACKET);
+
+ rc = qed_ll2_set_cbs(p_ll2_info, data->cbs);
+ if (rc) {
+ DP_NOTICE(p_hwfn, "Invalid callback functions\n");
+ goto q_allocate_fail;
+ }
+
rc = qed_ll2_acquire_connection_rx(p_hwfn, p_ll2_info);
if (rc)
goto q_allocate_fail;
@@ -1377,8 +1351,10 @@ qed_ll2_establish_connection_ooo(struct qed_hwfn *p_hwfn,
qed_ooo_release_all_isles(p_hwfn, p_hwfn->p_ooo_info);
qed_ooo_submit_rx_buffers(p_hwfn, p_ll2_conn);
}
-int qed_ll2_establish_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
+
+int qed_ll2_establish_connection(void *cxt, u8 connection_handle)
{
+ struct qed_hwfn *p_hwfn = cxt;
struct qed_ll2_info *p_ll2_conn;
struct qed_ll2_rx_queue *p_rx;
struct qed_ll2_tx_queue *p_tx;
@@ -1505,11 +1481,12 @@ static void qed_ll2_post_rx_buffer_notify_fw(struct qed_hwfn *p_hwfn,
DIRECT_REG_WR(p_rx->set_prod_addr, *((u32 *)&rx_prod));
}
-int qed_ll2_post_rx_buffer(struct qed_hwfn *p_hwfn,
+int qed_ll2_post_rx_buffer(void *cxt,
u8 connection_handle,
dma_addr_t addr,
u16 buf_len, void *cookie, u8 notify_fw)
{
+ struct qed_hwfn *p_hwfn = cxt;
struct core_rx_bd_with_buff_len *p_curb = NULL;
struct qed_ll2_rx_packet *p_curp = NULL;
struct qed_ll2_info *p_ll2_conn;
@@ -1699,11 +1676,12 @@ static void qed_ll2_tx_packet_notify(struct qed_hwfn *p_hwfn,
p_ll2_conn->input.conn_type, db_msg.spq_prod);
}
-int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
+int qed_ll2_prepare_tx_packet(void *cxt,
u8 connection_handle,
struct qed_ll2_tx_pkt_info *pkt,
bool notify_fw)
{
+ struct qed_hwfn *p_hwfn = cxt;
struct qed_ll2_tx_packet *p_curp = NULL;
struct qed_ll2_info *p_ll2_conn = NULL;
struct qed_ll2_tx_queue *p_tx;
@@ -1750,11 +1728,12 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
return rc;
}
-int qed_ll2_set_fragment_of_tx_packet(struct qed_hwfn *p_hwfn,
+int qed_ll2_set_fragment_of_tx_packet(void *cxt,
u8 connection_handle,
dma_addr_t addr, u16 nbytes)
{
struct qed_ll2_tx_packet *p_cur_send_packet = NULL;
+ struct qed_hwfn *p_hwfn = cxt;
struct qed_ll2_info *p_ll2_conn = NULL;
u16 cur_send_frag_num = 0;
struct core_tx_bd *p_bd;
@@ -1789,8 +1768,9 @@ int qed_ll2_set_fragment_of_tx_packet(struct qed_hwfn *p_hwfn,
return 0;
}
-int qed_ll2_terminate_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
+int qed_ll2_terminate_connection(void *cxt, u8 connection_handle)
{
+ struct qed_hwfn *p_hwfn = cxt;
struct qed_ll2_info *p_ll2_conn = NULL;
int rc = -EINVAL;
struct qed_ptt *p_ptt;
@@ -1855,8 +1835,10 @@ static void qed_ll2_release_connection_ooo(struct qed_hwfn *p_hwfn,
kfree(p_buffer);
}
}
-void qed_ll2_release_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
+
+void qed_ll2_release_connection(void *cxt, u8 connection_handle)
{
+ struct qed_hwfn *p_hwfn = cxt;
struct qed_ll2_info *p_ll2_conn = NULL;
p_ll2_conn = qed_ll2_handle_sanity(p_hwfn, connection_handle);
@@ -1989,9 +1971,10 @@ static void _qed_ll2_get_pstats(struct qed_hwfn *p_hwfn,
p_stats->sent_bcast_pkts = HILO_64_REGPAIR(pstats.sent_bcast_pkts);
}
-int qed_ll2_get_stats(struct qed_hwfn *p_hwfn,
+int qed_ll2_get_stats(void *cxt,
u8 connection_handle, struct qed_ll2_stats *p_stats)
{
+ struct qed_hwfn *p_hwfn = cxt;
struct qed_ll2_info *p_ll2_conn = NULL;
struct qed_ptt *p_ptt;
@@ -2018,6 +2001,17 @@ int qed_ll2_get_stats(struct qed_hwfn *p_hwfn,
return 0;
}
+static void qed_ll2b_release_rx_packet(void *cxt,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t rx_buf_addr,
+ bool b_last_packet)
+{
+ struct qed_hwfn *p_hwfn = cxt;
+
+ qed_ll2_dealloc_buffer(p_hwfn->cdev, cookie);
+}
+
static void qed_ll2_register_cb_ops(struct qed_dev *cdev,
const struct qed_ll2_cb_ops *ops,
void *cookie)
@@ -2026,11 +2020,18 @@ static void qed_ll2_register_cb_ops(struct qed_dev *cdev,
cdev->ll2->cb_cookie = cookie;
}
+struct qed_ll2_cbs ll2_cbs = {
+ .rx_comp_cb = &qed_ll2b_complete_rx_packet,
+ .rx_release_cb = &qed_ll2b_release_rx_packet,
+ .tx_comp_cb = &qed_ll2b_complete_tx_packet,
+ .tx_release_cb = &qed_ll2b_complete_tx_packet,
+};
+
static void qed_ll2_set_conn_data(struct qed_dev *cdev,
struct qed_ll2_acquire_data *data,
struct qed_ll2_params *params,
enum qed_ll2_conn_type conn_type,
- u8 *handle, bool lb, u8 gsi_enable)
+ u8 *handle, bool lb)
{
memset(data, 0, sizeof(*data));
@@ -2040,8 +2041,10 @@ static void qed_ll2_set_conn_data(struct qed_dev *cdev,
data->input.rx_drop_ttl0_flg = params->drop_ttl0_packets;
data->input.rx_vlan_removal_en = params->rx_vlan_stripping;
data->input.tx_num_desc = QED_LL2_TX_SIZE;
- data->input.gsi_enable = gsi_enable;
data->p_connection_handle = handle;
+ data->cbs = &ll2_cbs;
+ ll2_cbs.cookie = QED_LEADING_HWFN(cdev);
+
if (lb) {
data->input.tx_tc = OOO_LB_TC;
data->input.tx_dest = QED_LL2_TX_DEST_LB;
@@ -2060,7 +2063,7 @@ static int qed_ll2_start_ooo(struct qed_dev *cdev,
int rc;
qed_ll2_set_conn_data(cdev, &data, params,
- QED_LL2_TYPE_ISCSI_OOO, handle, true, 0);
+ QED_LL2_TYPE_ISCSI_OOO, handle, true);
rc = qed_ll2_acquire_connection(hwfn, &data);
if (rc) {
@@ -2090,7 +2093,7 @@ static int qed_ll2_start(struct qed_dev *cdev, struct qed_ll2_params *params)
struct qed_ll2_acquire_data data;
struct qed_ptt *p_ptt;
int rc, i;
- u8 gsi_enable = 1;
+
/* Initialize LL2 locks & lists */
INIT_LIST_HEAD(&cdev->ll2->list);
@@ -2122,11 +2125,9 @@ static int qed_ll2_start(struct qed_dev *cdev, struct qed_ll2_params *params)
switch (QED_LEADING_HWFN(cdev)->hw_info.personality) {
case QED_PCI_FCOE:
conn_type = QED_LL2_TYPE_FCOE;
- gsi_enable = 0;
break;
case QED_PCI_ISCSI:
conn_type = QED_LL2_TYPE_ISCSI;
- gsi_enable = 0;
break;
case QED_PCI_ETH_ROCE:
conn_type = QED_LL2_TYPE_ROCE;
@@ -2136,7 +2137,7 @@ static int qed_ll2_start(struct qed_dev *cdev, struct qed_ll2_params *params)
}
qed_ll2_set_conn_data(cdev, &data, params, conn_type,
- &cdev->ll2->handle, false, gsi_enable);
+ &cdev->ll2->handle, false);
rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &data);
if (rc) {
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.h b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
index 3caaad5..a822528 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
@@ -126,6 +126,7 @@ struct qed_ll2_info {
u8 tx_stats_en;
struct qed_ll2_rx_queue rx_queue;
struct qed_ll2_tx_queue tx_queue;
+ struct qed_ll2_cbs cbs;
};
/**
@@ -134,30 +135,29 @@ struct qed_ll2_info {
* connecion handler as output parameter.
*
*
- * @param p_hwfn
+ * @param cxt - pointer to the hw-function [opaque to some]
* @param data - describes connection parameters
* @return int
*/
-int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
- struct qed_ll2_acquire_data *data);
+int qed_ll2_acquire_connection(void *cxt, struct qed_ll2_acquire_data *data);
/**
* @brief qed_ll2_establish_connection - start previously
* allocated LL2 queues pair
*
- * @param p_hwfn
+ * @param cxt - pointer to the hw-function [opaque to some]
* @param p_ptt
* @param connection_handle LL2 connection's handle obtained from
* qed_ll2_require_connection
*
* @return 0 on success, failure otherwise
*/
-int qed_ll2_establish_connection(struct qed_hwfn *p_hwfn, u8 connection_handle);
+int qed_ll2_establish_connection(void *cxt, u8 connection_handle);
/**
* @brief qed_ll2_post_rx_buffers - submit buffers to LL2 Rx queue.
*
- * @param p_hwfn
+ * @param cxt - pointer to the hw-function [opaque to some]
* @param connection_handle LL2 connection's handle obtained from
* qed_ll2_require_connection
* @param addr rx (physical address) buffers to submit
@@ -166,7 +166,7 @@ int qed_ll2_establish_connection(struct qed_hwfn *p_hwfn, u8 connection_handle);
*
* @return 0 on success, failure otherwise
*/
-int qed_ll2_post_rx_buffer(struct qed_hwfn *p_hwfn,
+int qed_ll2_post_rx_buffer(void *cxt,
u8 connection_handle,
dma_addr_t addr,
u16 buf_len, void *cookie, u8 notify_fw);
@@ -175,14 +175,14 @@ int qed_ll2_post_rx_buffer(struct qed_hwfn *p_hwfn,
* @brief qed_ll2_prepare_tx_packet - request for start Tx BD
* to prepare Tx packet submission to FW.
*
- * @param p_hwfn
+ * @param cxt - pointer to the hw-function [opaque to some]
* @param connection_handle
* @param pkt - info regarding the tx packet
* @param notify_fw - issue doorbell to fw for this packet
*
* @return 0 on success, failure otherwise
*/
-int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
+int qed_ll2_prepare_tx_packet(void *cxt,
u8 connection_handle,
struct qed_ll2_tx_pkt_info *pkt,
bool notify_fw);
@@ -191,18 +191,18 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
* @brief qed_ll2_release_connection - releases resources
* allocated for LL2 connection
*
- * @param p_hwfn
+ * @param cxt - pointer to the hw-function [opaque to some]
* @param connection_handle LL2 connection's handle obtained from
* qed_ll2_require_connection
*/
-void qed_ll2_release_connection(struct qed_hwfn *p_hwfn, u8 connection_handle);
+void qed_ll2_release_connection(void *cxt, u8 connection_handle);
/**
* @brief qed_ll2_set_fragment_of_tx_packet - provides fragments to fill
* Tx BD of BDs requested by
* qed_ll2_prepare_tx_packet
*
- * @param p_hwfn
+ * @param cxt - pointer to the hw-function [opaque to some]
* @param connection_handle LL2 connection's handle
* obtained from
* qed_ll2_require_connection
@@ -211,7 +211,7 @@ void qed_ll2_release_connection(struct qed_hwfn *p_hwfn, u8 connection_handle);
*
* @return 0 on success, failure otherwise
*/
-int qed_ll2_set_fragment_of_tx_packet(struct qed_hwfn *p_hwfn,
+int qed_ll2_set_fragment_of_tx_packet(void *cxt,
u8 connection_handle,
dma_addr_t addr, u16 nbytes);
@@ -219,27 +219,27 @@ int qed_ll2_set_fragment_of_tx_packet(struct qed_hwfn *p_hwfn,
* @brief qed_ll2_terminate_connection - stops Tx/Rx queues
*
*
- * @param p_hwfn
+ * @param cxt - pointer to the hw-function [opaque to some]
* @param connection_handle LL2 connection's handle
* obtained from
* qed_ll2_require_connection
*
* @return 0 on success, failure otherwise
*/
-int qed_ll2_terminate_connection(struct qed_hwfn *p_hwfn, u8 connection_handle);
+int qed_ll2_terminate_connection(void *cxt, u8 connection_handle);
/**
* @brief qed_ll2_get_stats - get LL2 queue's statistics
*
*
- * @param p_hwfn
+ * @param cxt - pointer to the hw-function [opaque to some]
* @param connection_handle LL2 connection's handle obtained from
* qed_ll2_require_connection
* @param p_stats
*
* @return 0 on success, failure otherwise
*/
-int qed_ll2_get_stats(struct qed_hwfn *p_hwfn,
+int qed_ll2_get_stats(void *cxt,
u8 connection_handle, struct qed_ll2_stats *p_stats);
/**
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index 33abcf1..4bc2f6c 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -35,7 +35,6 @@
#include <linux/delay.h>
#include <linux/dma-mapping.h>
#include <linux/errno.h>
-#include <linux/etherdevice.h>
#include <linux/if_ether.h>
#include <linux/if_vlan.h>
#include <linux/io.h>
@@ -65,6 +64,7 @@
#include "qed_sp.h"
#include "qed_roce.h"
#include "qed_ll2.h"
+#include <linux/qed/qed_ll2_if.h>
static void qed_roce_free_real_icid(struct qed_hwfn *p_hwfn, u16 icid);
@@ -2709,310 +2709,35 @@ static void qed_rdma_remove_user(void *rdma_cxt, u16 dpi)
spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
}
-void qed_ll2b_complete_tx_gsi_packet(struct qed_hwfn *p_hwfn,
- u8 connection_handle,
- void *cookie,
- dma_addr_t first_frag_addr,
- bool b_last_fragment, bool b_last_packet)
-{
- struct qed_roce_ll2_packet *packet = cookie;
- struct qed_roce_ll2_info *roce_ll2 = p_hwfn->ll2;
-
- roce_ll2->cbs.tx_cb(roce_ll2->cb_cookie, packet);
-}
-
-void qed_ll2b_release_tx_gsi_packet(struct qed_hwfn *p_hwfn,
- u8 connection_handle,
- void *cookie,
- dma_addr_t first_frag_addr,
- bool b_last_fragment, bool b_last_packet)
-{
- qed_ll2b_complete_tx_gsi_packet(p_hwfn, connection_handle,
- cookie, first_frag_addr,
- b_last_fragment, b_last_packet);
-}
-
-void qed_ll2b_complete_rx_gsi_packet(struct qed_hwfn *p_hwfn,
- u8 connection_handle,
- void *cookie,
- dma_addr_t rx_buf_addr,
- u16 data_length,
- u8 data_length_error,
- u16 parse_flags,
- u16 vlan,
- u32 src_mac_addr_hi,
- u16 src_mac_addr_lo, bool b_last_packet)
-{
- struct qed_roce_ll2_info *roce_ll2 = p_hwfn->ll2;
- struct qed_roce_ll2_rx_params params;
- struct qed_dev *cdev = p_hwfn->cdev;
- struct qed_roce_ll2_packet pkt;
-
- DP_VERBOSE(cdev,
- QED_MSG_LL2,
- "roce ll2 rx complete: bus_addr=%p, len=%d, data_len_err=%d\n",
- (void *)(uintptr_t)rx_buf_addr,
- data_length, data_length_error);
-
- memset(&pkt, 0, sizeof(pkt));
- pkt.n_seg = 1;
- pkt.payload[0].baddr = rx_buf_addr;
- pkt.payload[0].len = data_length;
-
- memset(¶ms, 0, sizeof(params));
- params.vlan_id = vlan;
- *((u32 *)¶ms.smac[0]) = ntohl(src_mac_addr_hi);
- *((u16 *)¶ms.smac[4]) = ntohs(src_mac_addr_lo);
-
- if (data_length_error) {
- DP_ERR(cdev,
- "roce ll2 rx complete: data length error %d, length=%d\n",
- data_length_error, data_length);
- params.rc = -EINVAL;
- }
-
- roce_ll2->cbs.rx_cb(roce_ll2->cb_cookie, &pkt, ¶ms);
-}
-
static int qed_roce_ll2_set_mac_filter(struct qed_dev *cdev,
u8 *old_mac_address,
u8 *new_mac_address)
{
- struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ struct qed_hwfn *p_hwfn = QED_LEADING_HWFN(cdev);
struct qed_ptt *p_ptt;
int rc = 0;
- if (!hwfn->ll2 || hwfn->ll2->handle == QED_LL2_UNUSED_HANDLE) {
- DP_ERR(cdev,
- "qed roce mac filter failed - roce_info/ll2 NULL\n");
- return -EINVAL;
- }
-
- p_ptt = qed_ptt_acquire(QED_LEADING_HWFN(cdev));
+ p_ptt = qed_ptt_acquire(p_hwfn);
if (!p_ptt) {
DP_ERR(cdev,
"qed roce ll2 mac filter set: failed to acquire PTT\n");
return -EINVAL;
}
- mutex_lock(&hwfn->ll2->lock);
if (old_mac_address)
- qed_llh_remove_mac_filter(QED_LEADING_HWFN(cdev), p_ptt,
- old_mac_address);
+ qed_llh_remove_mac_filter(p_hwfn, p_ptt, old_mac_address);
if (new_mac_address)
- rc = qed_llh_add_mac_filter(QED_LEADING_HWFN(cdev), p_ptt,
- new_mac_address);
- mutex_unlock(&hwfn->ll2->lock);
-
- qed_ptt_release(QED_LEADING_HWFN(cdev), p_ptt);
-
- if (rc)
- DP_ERR(cdev,
- "qed roce ll2 mac filter set: failed to add mac filter\n");
-
- return rc;
-}
-
-static int qed_roce_ll2_start(struct qed_dev *cdev,
- struct qed_roce_ll2_params *params)
-{
- struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
- struct qed_roce_ll2_info *roce_ll2;
- struct qed_ll2_acquire_data data;
- int rc;
-
- if (!params) {
- DP_ERR(cdev, "qed roce ll2 start: failed due to NULL params\n");
- return -EINVAL;
- }
- if (!params->cbs.tx_cb || !params->cbs.rx_cb) {
- DP_ERR(cdev,
- "qed roce ll2 start: failed due to NULL tx/rx. tx_cb=%p, rx_cb=%p\n",
- params->cbs.tx_cb, params->cbs.rx_cb);
- return -EINVAL;
- }
- if (!is_valid_ether_addr(params->mac_address)) {
- DP_ERR(cdev,
- "qed roce ll2 start: failed due to invalid Ethernet address %pM\n",
- params->mac_address);
- return -EINVAL;
- }
-
- /* Initialize */
- roce_ll2 = kzalloc(sizeof(*roce_ll2), GFP_ATOMIC);
- if (!roce_ll2) {
- DP_ERR(cdev, "qed roce ll2 start: failed memory allocation\n");
- return -ENOMEM;
- }
+ rc = qed_llh_add_mac_filter(p_hwfn, p_ptt, new_mac_address);
- roce_ll2->handle = QED_LL2_UNUSED_HANDLE;
- roce_ll2->cbs = params->cbs;
- roce_ll2->cb_cookie = params->cb_cookie;
- mutex_init(&roce_ll2->lock);
-
- memset(&data, 0, sizeof(data));
- data.input.conn_type = QED_LL2_TYPE_ROCE;
- data.input.mtu = params->mtu;
- data.input.rx_num_desc = params->max_rx_buffers;
- data.input.tx_num_desc = params->max_tx_buffers;
- data.input.rx_drop_ttl0_flg = true;
- data.input.rx_vlan_removal_en = false;
- data.input.tx_dest = QED_LL2_TX_DEST_NW;
- data.input.ai_err_packet_too_big = LL2_DROP_PACKET;
- data.input.ai_err_no_buf = LL2_DROP_PACKET;
- data.p_connection_handle = &roce_ll2->handle;
- data.input.gsi_enable = true;
-
- rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &data);
- if (rc) {
- DP_ERR(cdev,
- "qed roce ll2 start: failed to acquire LL2 connection (rc=%d)\n",
- rc);
- goto err;
- }
-
- rc = qed_ll2_establish_connection(QED_LEADING_HWFN(cdev),
- roce_ll2->handle);
- if (rc) {
- DP_ERR(cdev,
- "qed roce ll2 start: failed to establish LL2 connection (rc=%d)\n",
- rc);
- goto err1;
- }
-
- hwfn->ll2 = roce_ll2;
-
- rc = qed_roce_ll2_set_mac_filter(cdev, NULL, params->mac_address);
- if (rc) {
- hwfn->ll2 = NULL;
- goto err2;
- }
- ether_addr_copy(roce_ll2->mac_address, params->mac_address);
-
- return 0;
-
-err2:
- qed_ll2_terminate_connection(QED_LEADING_HWFN(cdev), roce_ll2->handle);
-err1:
- qed_ll2_release_connection(QED_LEADING_HWFN(cdev), roce_ll2->handle);
-err:
- kfree(roce_ll2);
- return rc;
-}
-
-static int qed_roce_ll2_stop(struct qed_dev *cdev)
-{
- struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
- struct qed_roce_ll2_info *roce_ll2 = hwfn->ll2;
- int rc;
-
- if (roce_ll2->handle == QED_LL2_UNUSED_HANDLE) {
- DP_ERR(cdev, "qed roce ll2 stop: cannot stop an unused LL2\n");
- return -EINVAL;
- }
-
- /* remove LL2 MAC address filter */
- rc = qed_roce_ll2_set_mac_filter(cdev, roce_ll2->mac_address, NULL);
- eth_zero_addr(roce_ll2->mac_address);
+ qed_ptt_release(p_hwfn, p_ptt);
- rc = qed_ll2_terminate_connection(QED_LEADING_HWFN(cdev),
- roce_ll2->handle);
if (rc)
DP_ERR(cdev,
- "qed roce ll2 stop: failed to terminate LL2 connection (rc=%d)\n",
- rc);
-
- qed_ll2_release_connection(QED_LEADING_HWFN(cdev), roce_ll2->handle);
-
- roce_ll2->handle = QED_LL2_UNUSED_HANDLE;
-
- kfree(roce_ll2);
+ "qed roce ll2 mac filter set: failed to add MAC filter\n");
return rc;
}
-static int qed_roce_ll2_tx(struct qed_dev *cdev,
- struct qed_roce_ll2_packet *pkt,
- struct qed_roce_ll2_tx_params *params)
-{
- struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
- struct qed_roce_ll2_info *roce_ll2 = hwfn->ll2;
- enum qed_ll2_roce_flavor_type qed_roce_flavor;
- struct qed_ll2_tx_pkt_info ll2_pkt;
- u8 flags = 0;
- int rc;
- int i;
-
- if (!pkt || !params) {
- DP_ERR(cdev,
- "roce ll2 tx: failed tx because one of the following is NULL - drv=%p, pkt=%p, params=%p\n",
- cdev, pkt, params);
- return -EINVAL;
- }
-
- qed_roce_flavor = (pkt->roce_mode == ROCE_V1) ? QED_LL2_ROCE
- : QED_LL2_RROCE;
-
- if (pkt->roce_mode == ROCE_V2_IPV4)
- flags |= BIT(CORE_TX_BD_DATA_IP_CSUM_SHIFT);
-
- /* Tx header */
- memset(&ll2_pkt, 0, sizeof(ll2_pkt));
- ll2_pkt.num_of_bds = 1 + pkt->n_seg;
- ll2_pkt.bd_flags = flags;
- ll2_pkt.tx_dest = QED_LL2_TX_DEST_NW;
- ll2_pkt.qed_roce_flavor = qed_roce_flavor;
- ll2_pkt.first_frag = pkt->header.baddr;
- ll2_pkt.first_frag_len = pkt->header.len;
- ll2_pkt.cookie = pkt;
-
- rc = qed_ll2_prepare_tx_packet(QED_LEADING_HWFN(cdev),
- roce_ll2->handle,
- &ll2_pkt, 1);
- if (rc) {
- DP_ERR(cdev, "roce ll2 tx: header failed (rc=%d)\n", rc);
- return QED_ROCE_TX_HEAD_FAILURE;
- }
-
- /* Tx payload */
- for (i = 0; i < pkt->n_seg; i++) {
- rc = qed_ll2_set_fragment_of_tx_packet(QED_LEADING_HWFN(cdev),
- roce_ll2->handle,
- pkt->payload[i].baddr,
- pkt->payload[i].len);
- if (rc) {
- /* If failed not much to do here, partial packet has
- * been posted * we can't free memory, will need to wait
- * for completion
- */
- DP_ERR(cdev,
- "roce ll2 tx: payload failed (rc=%d)\n", rc);
- return QED_ROCE_TX_FRAG_FAILURE;
- }
- }
-
- return 0;
-}
-
-static int qed_roce_ll2_post_rx_buffer(struct qed_dev *cdev,
- struct qed_roce_ll2_buffer *buf,
- u64 cookie, u8 notify_fw)
-{
- return qed_ll2_post_rx_buffer(QED_LEADING_HWFN(cdev),
- QED_LEADING_HWFN(cdev)->ll2->handle,
- buf->baddr, buf->len,
- (void *)(uintptr_t)cookie, notify_fw);
-}
-
-static int qed_roce_ll2_stats(struct qed_dev *cdev, struct qed_ll2_stats *stats)
-{
- struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
- struct qed_roce_ll2_info *roce_ll2 = hwfn->ll2;
-
- return qed_ll2_get_stats(QED_LEADING_HWFN(cdev),
- roce_ll2->handle, stats);
-}
-
static const struct qed_rdma_ops qed_rdma_ops_pass = {
.common = &qed_common_ops_pass,
.fill_dev_info = &qed_fill_rdma_dev_info,
@@ -3040,12 +2765,15 @@ static const struct qed_rdma_ops qed_rdma_ops_pass = {
.rdma_free_tid = &qed_rdma_free_tid,
.rdma_register_tid = &qed_rdma_register_tid,
.rdma_deregister_tid = &qed_rdma_deregister_tid,
- .roce_ll2_start = &qed_roce_ll2_start,
- .roce_ll2_stop = &qed_roce_ll2_stop,
- .roce_ll2_tx = &qed_roce_ll2_tx,
- .roce_ll2_post_rx_buffer = &qed_roce_ll2_post_rx_buffer,
- .roce_ll2_set_mac_filter = &qed_roce_ll2_set_mac_filter,
- .roce_ll2_stats = &qed_roce_ll2_stats,
+ .ll2_acquire_connection = &qed_ll2_acquire_connection,
+ .ll2_establish_connection = &qed_ll2_establish_connection,
+ .ll2_terminate_connection = &qed_ll2_terminate_connection,
+ .ll2_release_connection = &qed_ll2_release_connection,
+ .ll2_post_rx_buffer = &qed_ll2_post_rx_buffer,
+ .ll2_prepare_tx_packet = &qed_ll2_prepare_tx_packet,
+ .ll2_set_fragment_of_tx_packet = &qed_ll2_set_fragment_of_tx_packet,
+ .ll2_set_mac_filter = &qed_roce_ll2_set_mac_filter,
+ .ll2_get_stats = &qed_ll2_get_stats,
};
const struct qed_rdma_ops *qed_get_rdma_ops(void)
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.h b/drivers/net/ethernet/qlogic/qed/qed_roce.h
index 9742af5..94be3b5 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.h
@@ -170,53 +170,10 @@ struct qed_rdma_qp {
void qed_rdma_dpm_bar(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt);
void qed_roce_async_event(struct qed_hwfn *p_hwfn,
u8 fw_event_code, union rdma_eqe_data *rdma_data);
-void qed_ll2b_complete_tx_gsi_packet(struct qed_hwfn *p_hwfn,
- u8 connection_handle,
- void *cookie,
- dma_addr_t first_frag_addr,
- bool b_last_fragment, bool b_last_packet);
-void qed_ll2b_release_tx_gsi_packet(struct qed_hwfn *p_hwfn,
- u8 connection_handle,
- void *cookie,
- dma_addr_t first_frag_addr,
- bool b_last_fragment, bool b_last_packet);
-void qed_ll2b_complete_rx_gsi_packet(struct qed_hwfn *p_hwfn,
- u8 connection_handle,
- void *cookie,
- dma_addr_t rx_buf_addr,
- u16 data_length,
- u8 data_length_error,
- u16 parse_flags,
- u16 vlan,
- u32 src_mac_addr_hi,
- u16 src_mac_addr_lo, bool b_last_packet);
#else
static inline void qed_rdma_dpm_bar(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt) {}
static inline void qed_roce_async_event(struct qed_hwfn *p_hwfn,
u8 fw_event_code,
union rdma_eqe_data *rdma_data) {}
-static inline void qed_ll2b_complete_tx_gsi_packet(struct qed_hwfn *p_hwfn,
- u8 connection_handle,
- void *cookie,
- dma_addr_t first_frag_addr,
- bool b_last_fragment,
- bool b_last_packet) {}
-static inline void qed_ll2b_release_tx_gsi_packet(struct qed_hwfn *p_hwfn,
- u8 connection_handle,
- void *cookie,
- dma_addr_t first_frag_addr,
- bool b_last_fragment,
- bool b_last_packet) {}
-static inline void qed_ll2b_complete_rx_gsi_packet(struct qed_hwfn *p_hwfn,
- u8 connection_handle,
- void *cookie,
- dma_addr_t rx_buf_addr,
- u16 data_length,
- u8 data_length_error,
- u16 parse_flags,
- u16 vlan,
- u32 src_mac_addr_hi,
- u16 src_mac_addr_lo,
- bool b_last_packet) {}
#endif
#endif
diff --git a/include/linux/qed/qed_ll2_if.h b/include/linux/qed/qed_ll2_if.h
index a8f58d9..e03e9e3 100644
--- a/include/linux/qed/qed_ll2_if.h
+++ b/include/linux/qed/qed_ll2_if.h
@@ -121,6 +121,40 @@ struct qed_ll2_comp_rx_data {
u16 qp_id;
};
+typedef
+void (*qed_ll2_complete_rx_packet_cb)(void *cxt,
+ struct qed_ll2_comp_rx_data *data);
+
+typedef
+void (*qed_ll2_release_rx_packet_cb)(void *cxt,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t rx_buf_addr,
+ bool b_last_packet);
+
+typedef
+void (*qed_ll2_complete_tx_packet_cb)(void *cxt,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment,
+ bool b_last_packet);
+
+typedef
+void (*qed_ll2_release_tx_packet_cb)(void *cxt,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment, bool b_last_packet);
+
+struct qed_ll2_cbs {
+ qed_ll2_complete_rx_packet_cb rx_comp_cb;
+ qed_ll2_release_rx_packet_cb rx_release_cb;
+ qed_ll2_complete_tx_packet_cb tx_comp_cb;
+ qed_ll2_release_tx_packet_cb tx_release_cb;
+ void *cookie;
+};
+
struct qed_ll2_acquire_data_inputs {
enum qed_ll2_conn_type conn_type;
u16 mtu;
@@ -139,6 +173,8 @@ struct qed_ll2_acquire_data_inputs {
struct qed_ll2_acquire_data {
struct qed_ll2_acquire_data_inputs input;
+ const struct qed_ll2_cbs *cbs;
+
/* Output container for LL2 connection's handle */
u8 *p_connection_handle;
};
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
index cbb2ff0..8e70f5e 100644
--- a/include/linux/qed/qed_roce_if.h
+++ b/include/linux/qed/qed_roce_if.h
@@ -34,8 +34,6 @@
#include <linux/types.h>
#include <linux/delay.h>
#include <linux/list.h>
-#include <linux/mutex.h>
-#include <linux/pci.h>
#include <linux/slab.h>
#include <linux/qed/qed_if.h>
#include <linux/qed/qed_ll2_if.h>
@@ -491,42 +489,6 @@ struct qed_roce_ll2_packet {
enum qed_roce_ll2_tx_dest tx_dest;
};
-struct qed_roce_ll2_tx_params {
- int reserved;
-};
-
-struct qed_roce_ll2_rx_params {
- u16 vlan_id;
- u8 smac[ETH_ALEN];
- int rc;
-};
-
-struct qed_roce_ll2_cbs {
- void (*tx_cb)(void *pdev, struct qed_roce_ll2_packet *pkt);
-
- void (*rx_cb)(void *pdev, struct qed_roce_ll2_packet *pkt,
- struct qed_roce_ll2_rx_params *params);
-};
-
-struct qed_roce_ll2_params {
- u16 max_rx_buffers;
- u16 max_tx_buffers;
- u16 mtu;
- u8 mac_address[ETH_ALEN];
- struct qed_roce_ll2_cbs cbs;
- void *cb_cookie;
-};
-
-struct qed_roce_ll2_info {
- u8 handle;
- struct qed_roce_ll2_cbs cbs;
- u8 mac_address[ETH_ALEN];
- void *cb_cookie;
-
- /* Lock to protect ll2 */
- struct mutex lock;
-};
-
enum qed_rdma_type {
QED_RDMA_TYPE_ROCE,
};
@@ -579,26 +541,40 @@ struct qed_rdma_ops {
int (*rdma_query_qp)(void *rdma_cxt, struct qed_rdma_qp *qp,
struct qed_rdma_query_qp_out_params *oparams);
int (*rdma_destroy_qp)(void *rdma_cxt, struct qed_rdma_qp *qp);
+
int
(*rdma_register_tid)(void *rdma_cxt,
struct qed_rdma_register_tid_in_params *iparams);
+
int (*rdma_deregister_tid)(void *rdma_cxt, u32 itid);
int (*rdma_alloc_tid)(void *rdma_cxt, u32 *itid);
void (*rdma_free_tid)(void *rdma_cxt, u32 itid);
- int (*roce_ll2_start)(struct qed_dev *cdev,
- struct qed_roce_ll2_params *params);
- int (*roce_ll2_stop)(struct qed_dev *cdev);
- int (*roce_ll2_tx)(struct qed_dev *cdev,
- struct qed_roce_ll2_packet *packet,
- struct qed_roce_ll2_tx_params *params);
- int (*roce_ll2_post_rx_buffer)(struct qed_dev *cdev,
- struct qed_roce_ll2_buffer *buf,
- u64 cookie, u8 notify_fw);
- int (*roce_ll2_set_mac_filter)(struct qed_dev *cdev,
- u8 *old_mac_address,
- u8 *new_mac_address);
- int (*roce_ll2_stats)(struct qed_dev *cdev,
- struct qed_ll2_stats *stats);
+
+ int (*ll2_acquire_connection)(void *rdma_cxt,
+ struct qed_ll2_acquire_data *data);
+
+ int (*ll2_establish_connection)(void *rdma_cxt, u8 connection_handle);
+ int (*ll2_terminate_connection)(void *rdma_cxt, u8 connection_handle);
+ void (*ll2_release_connection)(void *rdma_cxt, u8 connection_handle);
+
+ int (*ll2_prepare_tx_packet)(void *rdma_cxt,
+ u8 connection_handle,
+ struct qed_ll2_tx_pkt_info *pkt,
+ bool notify_fw);
+
+ int (*ll2_set_fragment_of_tx_packet)(void *rdma_cxt,
+ u8 connection_handle,
+ dma_addr_t addr,
+ u16 nbytes);
+ int (*ll2_post_rx_buffer)(void *rdma_cxt, u8 connection_handle,
+ dma_addr_t addr, u16 buf_len, void *cookie,
+ u8 notify_fw);
+ int (*ll2_get_stats)(void *rdma_cxt,
+ u8 connection_handle,
+ struct qed_ll2_stats *p_stats);
+ int (*ll2_set_mac_filter)(struct qed_dev *cdev,
+ u8 *old_mac_address, u8 *new_mac_address);
+
};
const struct qed_rdma_ops *qed_get_rdma_ops(void);
--
2.9.4
^ permalink raw reply related [flat|nested] 16+ messages in thread
* [PATCH net-next 6/8] qed: No need for LL2 frags indication
2017-06-08 16:13 [PATCH net-next 0/8] qed*: Light L2 updates Yuval Mintz
` (3 preceding siblings ...)
2017-06-08 16:13 ` [PATCH net-next 5/8] qed*: LL2 callback operations Yuval Mintz
@ 2017-06-08 16:13 ` Yuval Mintz
[not found] ` <20170608161323.7597-1-Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
5 siblings, 0 replies; 16+ messages in thread
From: Yuval Mintz @ 2017-06-08 16:13 UTC (permalink / raw)
To: netdev, davem; +Cc: linux-rdma, Michal.Kalderon, Yuval Mintz
This is a legacy leftover; There's no current flow where 'frags_mapped'
would be set.
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
---
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 31 ++++++++-----------------------
1 file changed, 8 insertions(+), 23 deletions(-)
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
index b222b2b..2a68fb9 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
@@ -73,7 +73,6 @@ struct qed_cb_ll2_info {
int rx_cnt;
u32 rx_size;
u8 handle;
- bool frags_mapped;
/* Lock protecting LL2 buffer lists in sleepless context */
spinlock_t lock;
@@ -108,12 +107,6 @@ static void qed_ll2b_complete_tx_packet(void *cxt,
cdev->ll2->cbs->tx_cb(cdev->ll2->cb_cookie, skb,
b_last_fragment);
- if (cdev->ll2->frags_mapped)
- /* Case where mapped frags were received, need to
- * free skb with nr_frags marked as 0
- */
- skb_shinfo(skb)->nr_frags = 0;
-
dev_kfree_skb_any(skb);
}
@@ -2100,7 +2093,6 @@ static int qed_ll2_start(struct qed_dev *cdev, struct qed_ll2_params *params)
spin_lock_init(&cdev->ll2->lock);
cdev->ll2->rx_size = NET_SKB_PAD + ETH_HLEN +
L1_CACHE_BYTES + params->mtu;
- cdev->ll2->frags_mapped = params->frags_mapped;
/*Allocate memory for LL2 */
DP_INFO(cdev, "Allocating LL2 buffers of size %08x bytes\n",
@@ -2313,21 +2305,14 @@ static int qed_ll2_start_xmit(struct qed_dev *cdev, struct sk_buff *skb)
for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
frag = &skb_shinfo(skb)->frags[i];
- if (!cdev->ll2->frags_mapped) {
- mapping = skb_frag_dma_map(&cdev->pdev->dev, frag, 0,
- skb_frag_size(frag),
- DMA_TO_DEVICE);
-
- if (unlikely(dma_mapping_error(&cdev->pdev->dev,
- mapping))) {
- DP_NOTICE(cdev,
- "Unable to map frag - dropping packet\n");
- rc = -ENOMEM;
- goto err;
- }
- } else {
- mapping = page_to_phys(skb_frag_page(frag)) |
- frag->page_offset;
+
+ mapping = skb_frag_dma_map(&cdev->pdev->dev, frag, 0,
+ skb_frag_size(frag), DMA_TO_DEVICE);
+
+ if (unlikely(dma_mapping_error(&cdev->pdev->dev, mapping))) {
+ DP_NOTICE(cdev,
+ "Unable to map frag - dropping packet\n");
+ goto err;
}
rc = qed_ll2_set_fragment_of_tx_packet(QED_LEADING_HWFN(cdev),
--
2.9.4
^ permalink raw reply related [flat|nested] 16+ messages in thread
* [PATCH net-next 7/8] qed: Call rx_release_cb() when flushing LL2
[not found] ` <20170608161323.7597-1-Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
2017-06-08 16:13 ` [PATCH net-next 3/8] qed: Cleaner seperation of LL2 inputs Yuval Mintz
@ 2017-06-08 16:13 ` Yuval Mintz
2017-06-08 16:13 ` [PATCH net-next 8/8] qed: collect GSI port statistics Yuval Mintz
2 siblings, 0 replies; 16+ messages in thread
From: Yuval Mintz @ 2017-06-08 16:13 UTC (permalink / raw)
To: netdev-u79uwXL29TY76Z2rM5mHXA, davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
Michal.Kalderon-YGCgFSpz5w/QT0dZR+AlfA, Yuval Mintz
Driver to inform the connection owner that the its buffers are being
released as part of a flush.
Signed-off-by: Yuval Mintz <Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
---
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 13 +++++++------
1 file changed, 7 insertions(+), 6 deletions(-)
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
index 2a68fb9..c6172a77e 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
@@ -525,10 +525,6 @@ static void qed_ll2_rxq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
p_rx = &p_ll2_conn->rx_queue;
while (!list_empty(&p_rx->active_descq)) {
- dma_addr_t rx_buf_addr;
- void *cookie;
- bool b_last;
-
p_pkt = list_first_entry(&p_rx->active_descq,
struct qed_ll2_rx_packet, list_entry);
if (!p_pkt)
@@ -543,10 +539,15 @@ static void qed_ll2_rxq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
qed_ooo_put_free_buffer(p_hwfn, p_hwfn->p_ooo_info,
p_buffer);
} else {
- rx_buf_addr = p_pkt->rx_buf_addr;
- cookie = p_pkt->cookie;
+ dma_addr_t rx_buf_addr = p_pkt->rx_buf_addr;
+ void *cookie = p_pkt->cookie;
+ bool b_last;
b_last = list_empty(&p_rx->active_descq);
+ p_ll2_conn->cbs.rx_release_cb(p_ll2_conn->cbs.cookie,
+ p_ll2_conn->my_id,
+ cookie,
+ rx_buf_addr, b_last);
}
}
}
--
2.9.4
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 16+ messages in thread
* [PATCH net-next 8/8] qed: collect GSI port statistics
[not found] ` <20170608161323.7597-1-Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
2017-06-08 16:13 ` [PATCH net-next 3/8] qed: Cleaner seperation of LL2 inputs Yuval Mintz
2017-06-08 16:13 ` [PATCH net-next 7/8] qed: Call rx_release_cb() when flushing LL2 Yuval Mintz
@ 2017-06-08 16:13 ` Yuval Mintz
2 siblings, 0 replies; 16+ messages in thread
From: Yuval Mintz @ 2017-06-08 16:13 UTC (permalink / raw)
To: netdev-u79uwXL29TY76Z2rM5mHXA, davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA,
Michal.Kalderon-YGCgFSpz5w/QT0dZR+AlfA, Yuval Mintz
The LL2 statistics already have place holders for these, but haven't
populated them so far.
Signed-off-by: Yuval Mintz <Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
---
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 23 +++++++++++++++++++++++
1 file changed, 23 insertions(+)
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
index c6172a77e..0e26193 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
@@ -1902,6 +1902,27 @@ void qed_ll2_free(struct qed_hwfn *p_hwfn)
p_hwfn->p_ll2_info = NULL;
}
+static void _qed_ll2_get_port_stats(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt,
+ struct qed_ll2_stats *p_stats)
+{
+ struct core_ll2_port_stats port_stats;
+
+ memset(&port_stats, 0, sizeof(port_stats));
+ qed_memcpy_from(p_hwfn, p_ptt, &port_stats,
+ BAR0_MAP_REG_TSDM_RAM +
+ TSTORM_LL2_PORT_STAT_OFFSET(MFW_PORT(p_hwfn)),
+ sizeof(port_stats));
+
+ p_stats->gsi_invalid_hdr = HILO_64_REGPAIR(port_stats.gsi_invalid_hdr);
+ p_stats->gsi_invalid_pkt_length =
+ HILO_64_REGPAIR(port_stats.gsi_invalid_pkt_length);
+ p_stats->gsi_unsupported_pkt_typ =
+ HILO_64_REGPAIR(port_stats.gsi_unsupported_pkt_typ);
+ p_stats->gsi_crcchksm_error =
+ HILO_64_REGPAIR(port_stats.gsi_crcchksm_error);
+}
+
static void _qed_ll2_get_tstats(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt,
struct qed_ll2_info *p_ll2_conn,
@@ -1986,6 +2007,8 @@ int qed_ll2_get_stats(void *cxt,
return -EINVAL;
}
+ if (p_ll2_conn->input.gsi_enable)
+ _qed_ll2_get_port_stats(p_hwfn, p_ptt, p_stats);
_qed_ll2_get_tstats(p_hwfn, p_ptt, p_ll2_conn, p_stats);
_qed_ll2_get_ustats(p_hwfn, p_ptt, p_ll2_conn, p_stats);
if (p_ll2_conn->tx_stats_en)
--
2.9.4
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 16+ messages in thread
* Re: [PATCH net-next 1/8] qed: LL2 to use packed information for tx
[not found] ` <20170608161323.7597-2-Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
@ 2017-06-08 23:23 ` David Miller
2017-06-09 7:08 ` Mintz, Yuval
2017-06-09 7:28 ` David Laight
0 siblings, 2 replies; 16+ messages in thread
From: David Miller @ 2017-06-08 23:23 UTC (permalink / raw)
To: Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA
Cc: netdev-u79uwXL29TY76Z2rM5mHXA, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
Michal.Kalderon-YGCgFSpz5w/QT0dZR+AlfA
From: Yuval Mintz <Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
Date: Thu, 8 Jun 2017 19:13:16 +0300
> @@ -67,6 +79,21 @@ struct qed_ll2_stats {
> u64 sent_bcast_pkts;
> };
>
> +struct qed_ll2_tx_pkt_info {
> + u8 num_of_bds;
> + u16 vlan;
> + u8 bd_flags;
> + u16 l4_hdr_offset_w; /* from start of packet */
> + enum qed_ll2_tx_dest tx_dest;
> + enum qed_ll2_roce_flavor_type qed_roce_flavor;
> + dma_addr_t first_frag;
> + u16 first_frag_len;
> + bool enable_ip_cksum;
> + bool enable_l4_cksum;
> + bool calc_ip_len;
> + void *cookie;
> +};
> +
This layout is extremely inefficient, with lots of padding in between
struct members.
Group small u8 members and u16 members together so that they consume
full 32-bit areas so you can eliminate all of the padding.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [PATCH net-next 2/8] qed: Revise ll2 Rx completion
[not found] ` <20170608161323.7597-3-Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
@ 2017-06-08 23:24 ` David Miller
0 siblings, 0 replies; 16+ messages in thread
From: David Miller @ 2017-06-08 23:24 UTC (permalink / raw)
To: Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA
Cc: netdev-u79uwXL29TY76Z2rM5mHXA, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
Michal.Kalderon-YGCgFSpz5w/QT0dZR+AlfA
From: Yuval Mintz <Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
Date: Thu, 8 Jun 2017 19:13:17 +0300
> +struct qed_ll2_comp_rx_data {
> + u8 connection_handle;
> + void *cookie;
> + dma_addr_t rx_buf_addr;
> + u16 parse_flags;
> + u16 vlan;
> + bool b_last_packet;
> +
> + union {
> + u8 placement_offset;
> + u8 data_length_error;
> + } u;
> + union {
> + u16 packet_length;
> + u16 data_length;
> + } length;
> +
> + u32 opaque_data_0;
> + u32 opaque_data_1;
> +
> + /* GSI only */
> + u32 gid_dst[4];
> + u16 qp_id;
> +};
> +
Again, a lot of unnecessary padding in the structure due to suboptimal
member ordering.
Please fix this.
Thank you.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 16+ messages in thread
* RE: [PATCH net-next 1/8] qed: LL2 to use packed information for tx
2017-06-08 23:23 ` David Miller
@ 2017-06-09 7:08 ` Mintz, Yuval
[not found] ` <BLUPR0701MB200415C3D80FB06774CF2C7B8DCE0-v5ruerSQ/oj1oiTf6dR/AE5OhdzP3rhOnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
2017-06-09 7:28 ` David Laight
1 sibling, 1 reply; 16+ messages in thread
From: Mintz, Yuval @ 2017-06-09 7:08 UTC (permalink / raw)
To: David Miller
Cc: netdev@vger.kernel.org, linux-rdma@vger.kernel.org,
Kalderon, Michal
> > +struct qed_ll2_tx_pkt_info {
> > + u8 num_of_bds;
> > + u16 vlan;
> > + u8 bd_flags;
> > + u16 l4_hdr_offset_w; /* from start of packet */
> > + enum qed_ll2_tx_dest tx_dest;
> > + enum qed_ll2_roce_flavor_type qed_roce_flavor;
> > + dma_addr_t first_frag;
> > + u16 first_frag_len;
> > + bool enable_ip_cksum;
> > + bool enable_l4_cksum;
> > + bool calc_ip_len;
> > + void *cookie;
> > +};
> > +
>
> This layout is extremely inefficient, with lots of padding in between struct
> members.
>
> Group small u8 members and u16 members together so that they consume
> full 32-bit areas so you can eliminate all of the padding.
While I can mend the holes, the total structure size doesn't really change:
struct qed_ll2_tx_pkt_info {
u8 num_of_bds; /* 0 1 */
/* XXX 1 byte hole, try to pack */
u16 vlan; /* 2 2 */
u8 bd_flags; /* 4 1 */
/* XXX 1 byte hole, try to pack */
u16 l4_hdr_offset_w; /* 6 2 */
enum qed_ll2_tx_dest tx_dest; /* 8 4 */
enum qed_ll2_roce_flavor_type qed_roce_flavor; /* 12 4 */
dma_addr_t first_frag; /* 16 8 */
u16 first_frag_len; /* 24 2 */
bool enable_ip_cksum; /* 26 1 */
bool enable_l4_cksum; /* 27 1 */
bool calc_ip_len; /* 28 1 */
/* XXX 3 bytes hole, try to pack */
void * cookie; /* 32 8 */
/* size: 40, cachelines: 1, members: 12 */
/* sum members: 35, holes: 3, sum holes: 5 */
/* last cacheline: 40 bytes */
};
Becomes:
struct qed_ll2_tx_pkt_info {
void * cookie; /* 0 8 */
dma_addr_t first_frag; /* 8 8 */
enum qed_ll2_tx_dest tx_dest; /* 16 4 */
enum qed_ll2_roce_flavor_type qed_roce_flavor; /* 20 4 */
u16 vlan; /* 24 2 */
u16 l4_hdr_offset_w; /* 26 2 */
u16 first_frag_len; /* 28 2 */
u8 num_of_bds; /* 30 1 */
u8 bd_flags; /* 31 1 */
bool enable_ip_cksum; /* 32 1 */
bool enable_l4_cksum; /* 33 1 */
bool calc_ip_len; /* 34 1 */
/* size: 40, cachelines: 1, members: 12 */
/* padding: 5 */
/* last cacheline: 40 bytes */
};
I'm going to send the changed version in V2 as as there's no harm to it,
[+ we *can* reduce the size of qed_ll2_comp_rx_data you commented
for patch #2 in series]
But one thing I thought of asking - do we consider layouts of relatively
insignificant structures to be some golden coding standard?
^ permalink raw reply [flat|nested] 16+ messages in thread
* RE: [PATCH net-next 1/8] qed: LL2 to use packed information for tx
2017-06-08 23:23 ` David Miller
2017-06-09 7:08 ` Mintz, Yuval
@ 2017-06-09 7:28 ` David Laight
2017-06-09 7:51 ` Mintz, Yuval
1 sibling, 1 reply; 16+ messages in thread
From: David Laight @ 2017-06-09 7:28 UTC (permalink / raw)
To: 'David Miller', Yuval.Mintz@cavium.com
Cc: netdev@vger.kernel.org, linux-rdma@vger.kernel.org,
Michal.Kalderon@cavium.com
From: David Miller
> Sent: 09 June 2017 00:24
>
> From: Yuval Mintz <Yuval.Mintz@cavium.com>
> Date: Thu, 8 Jun 2017 19:13:16 +0300
>
> > @@ -67,6 +79,21 @@ struct qed_ll2_stats {
> > u64 sent_bcast_pkts;
> > };
> >
> > +struct qed_ll2_tx_pkt_info {
> > + u8 num_of_bds;
> > + u16 vlan;
> > + u8 bd_flags;
> > + u16 l4_hdr_offset_w; /* from start of packet */
> > + enum qed_ll2_tx_dest tx_dest;
> > + enum qed_ll2_roce_flavor_type qed_roce_flavor;
> > + dma_addr_t first_frag;
> > + u16 first_frag_len;
> > + bool enable_ip_cksum;
> > + bool enable_l4_cksum;
> > + bool calc_ip_len;
> > + void *cookie;
> > +};
> > +
>
> This layout is extremely inefficient, with lots of padding in between
> struct members.
>
> Group small u8 members and u16 members together so that they consume
> full 32-bit areas so you can eliminate all of the padding.
I'd also query the use of u16 sizes/lengths, any arithmetic on u16 (and u8)
variables is likely to generate extra code (on non-x86).
You are using 32 bits for the 'enum' - I bet the values fit in 8 bits,
so aren't really worried about size.
If size did matter you can easily get the above down to 32 bytes.
David
^ permalink raw reply [flat|nested] 16+ messages in thread
* RE: [PATCH net-next 1/8] qed: LL2 to use packed information for tx
2017-06-09 7:28 ` David Laight
@ 2017-06-09 7:51 ` Mintz, Yuval
[not found] ` <BLUPR0701MB2004CA6F1F262EC4E815E6B58DCE0-v5ruerSQ/oj1oiTf6dR/AE5OhdzP3rhOnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
0 siblings, 1 reply; 16+ messages in thread
From: Mintz, Yuval @ 2017-06-09 7:51 UTC (permalink / raw)
To: David Laight, 'David Miller'
Cc: netdev@vger.kernel.org, linux-rdma@vger.kernel.org,
Kalderon, Michal
> -----Original Message-----
> From: David Laight [mailto:David.Laight@ACULAB.COM]
> Sent: Friday, June 09, 2017 10:28 AM
> To: 'David Miller' <davem@davemloft.net>; Mintz, Yuval
> <Yuval.Mintz@cavium.com>
> Cc: netdev@vger.kernel.org; linux-rdma@vger.kernel.org; Kalderon, Michal
> <Michal.Kalderon@cavium.com>
> Subject: RE: [PATCH net-next 1/8] qed: LL2 to use packed information for tx
>
> From: David Miller
> > Sent: 09 June 2017 00:24
> >
> > From: Yuval Mintz <Yuval.Mintz@cavium.com>
> > Date: Thu, 8 Jun 2017 19:13:16 +0300
> >
> > > @@ -67,6 +79,21 @@ struct qed_ll2_stats {
> > > u64 sent_bcast_pkts;
> > > };
> > >
> > > +struct qed_ll2_tx_pkt_info {
> > > + u8 num_of_bds;
> > > + u16 vlan;
> > > + u8 bd_flags;
> > > + u16 l4_hdr_offset_w; /* from start of packet */
> > > + enum qed_ll2_tx_dest tx_dest;
> > > + enum qed_ll2_roce_flavor_type qed_roce_flavor;
> > > + dma_addr_t first_frag;
> > > + u16 first_frag_len;
> > > + bool enable_ip_cksum;
> > > + bool enable_l4_cksum;
> > > + bool calc_ip_len;
> > > + void *cookie;
> > > +};
> > > +
> >
> > This layout is extremely inefficient, with lots of padding in between
> > struct members.
> >
> > Group small u8 members and u16 members together so that they consume
> > full 32-bit areas so you can eliminate all of the padding.
>
> I'd also query the use of u16 sizes/lengths, any arithmetic on u16 (and u8)
> variables is likely to generate extra code (on non-x86).
> You are using 32 bits for the 'enum' - I bet the values fit in 8 bits, so aren't
> really worried about size.
>
> If size did matter you can easily get the above down to 32 bytes.
You're right, and that's exactly the point - since this is not data-path critical
I don't see why the size/efficiency should matter [greatly].
^ permalink raw reply [flat|nested] 16+ messages in thread
* RE: [PATCH net-next 1/8] qed: LL2 to use packed information for tx
[not found] ` <BLUPR0701MB2004CA6F1F262EC4E815E6B58DCE0-v5ruerSQ/oj1oiTf6dR/AE5OhdzP3rhOnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
@ 2017-06-09 8:23 ` David Laight
0 siblings, 0 replies; 16+ messages in thread
From: David Laight @ 2017-06-09 8:23 UTC (permalink / raw)
To: 'Mintz, Yuval', 'David Miller'
Cc: netdev-u79uwXL29TY76Z2rM5mHXA@public.gmane.org,
linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org,
Kalderon, Michal
From: Mintz, Yuval
> Sent: 09 June 2017 08:52
> > From: David Laight [mailto:David.Laight-ZS65k/vG3HxXrIkS9f7CXA@public.gmane.org]
> > Sent: Friday, June 09, 2017 10:28 AM
> > To: 'David Miller' <davem-fT/PcQaiUtIeIZ0/mPfg9Q@public.gmane.org>; Mintz, Yuval
> > <Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
> > Cc: netdev-u79uwXL29TY76Z2rM5mHXA@public.gmane.org; linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org; Kalderon, Michal
> > <Michal.Kalderon-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
> > Subject: RE: [PATCH net-next 1/8] qed: LL2 to use packed information for tx
> >
> > From: David Miller
> > > Sent: 09 June 2017 00:24
> > >
> > > From: Yuval Mintz <Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
> > > Date: Thu, 8 Jun 2017 19:13:16 +0300
> > >
> > > > @@ -67,6 +79,21 @@ struct qed_ll2_stats {
> > > > u64 sent_bcast_pkts;
> > > > };
> > > >
> > > > +struct qed_ll2_tx_pkt_info {
> > > > + u8 num_of_bds;
> > > > + u16 vlan;
> > > > + u8 bd_flags;
> > > > + u16 l4_hdr_offset_w; /* from start of packet */
> > > > + enum qed_ll2_tx_dest tx_dest;
> > > > + enum qed_ll2_roce_flavor_type qed_roce_flavor;
> > > > + dma_addr_t first_frag;
> > > > + u16 first_frag_len;
> > > > + bool enable_ip_cksum;
> > > > + bool enable_l4_cksum;
> > > > + bool calc_ip_len;
> > > > + void *cookie;
> > > > +};
> > > > +
> > >
> > > This layout is extremely inefficient, with lots of padding in between
> > > struct members.
> > >
> > > Group small u8 members and u16 members together so that they consume
> > > full 32-bit areas so you can eliminate all of the padding.
> >
> > I'd also query the use of u16 sizes/lengths, any arithmetic on u16 (and u8)
> > variables is likely to generate extra code (on non-x86).
> > You are using 32 bits for the 'enum' - I bet the values fit in 8 bits, so aren't
> > really worried about size.
> >
> > If size did matter you can easily get the above down to 32 bytes.
>
> You're right, and that's exactly the point - since this is not data-path critical
> I don't see why the size/efficiency should matter [greatly].
It is just good practise so that it happens automatically when it does matter.
Just swapping 'vlan' and 'bd_flags' would make it look much better.
David
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [PATCH net-next 1/8] qed: LL2 to use packed information for tx
[not found] ` <BLUPR0701MB200415C3D80FB06774CF2C7B8DCE0-v5ruerSQ/oj1oiTf6dR/AE5OhdzP3rhOnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
@ 2017-06-09 13:59 ` David Miller
0 siblings, 0 replies; 16+ messages in thread
From: David Miller @ 2017-06-09 13:59 UTC (permalink / raw)
To: Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA
Cc: netdev-u79uwXL29TY76Z2rM5mHXA, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
Michal.Kalderon-YGCgFSpz5w/QT0dZR+AlfA
From: "Mintz, Yuval" <Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
Date: Fri, 9 Jun 2017 07:08:54 +0000
> But one thing I thought of asking - do we consider layouts of relatively
> insignificant structures to be some golden coding standard?
It just shows lack of thought when writing things.
I want to see code that shows the author put some care into it
and didn't just slap things together.
Thank you.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 16+ messages in thread
end of thread, other threads:[~2017-06-09 13:59 UTC | newest]
Thread overview: 16+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2017-06-08 16:13 [PATCH net-next 0/8] qed*: Light L2 updates Yuval Mintz
2017-06-08 16:13 ` [PATCH net-next 1/8] qed: LL2 to use packed information for tx Yuval Mintz
[not found] ` <20170608161323.7597-2-Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
2017-06-08 23:23 ` David Miller
2017-06-09 7:08 ` Mintz, Yuval
[not found] ` <BLUPR0701MB200415C3D80FB06774CF2C7B8DCE0-v5ruerSQ/oj1oiTf6dR/AE5OhdzP3rhOnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
2017-06-09 13:59 ` David Miller
2017-06-09 7:28 ` David Laight
2017-06-09 7:51 ` Mintz, Yuval
[not found] ` <BLUPR0701MB2004CA6F1F262EC4E815E6B58DCE0-v5ruerSQ/oj1oiTf6dR/AE5OhdzP3rhOnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
2017-06-09 8:23 ` David Laight
2017-06-08 16:13 ` [PATCH net-next 2/8] qed: Revise ll2 Rx completion Yuval Mintz
[not found] ` <20170608161323.7597-3-Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
2017-06-08 23:24 ` David Miller
2017-06-08 16:13 ` [PATCH net-next 4/8] qed: LL2 code relocations Yuval Mintz
2017-06-08 16:13 ` [PATCH net-next 5/8] qed*: LL2 callback operations Yuval Mintz
2017-06-08 16:13 ` [PATCH net-next 6/8] qed: No need for LL2 frags indication Yuval Mintz
[not found] ` <20170608161323.7597-1-Yuval.Mintz-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
2017-06-08 16:13 ` [PATCH net-next 3/8] qed: Cleaner seperation of LL2 inputs Yuval Mintz
2017-06-08 16:13 ` [PATCH net-next 7/8] qed: Call rx_release_cb() when flushing LL2 Yuval Mintz
2017-06-08 16:13 ` [PATCH net-next 8/8] qed: collect GSI port statistics Yuval Mintz
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).