Commit 22adfe0a authored by Santosh Rastapur's avatar Santosh Rastapur Committed by David S. Miller

cxgb4: Add T5 write combining support

This patch implements a low latency Write Combining (aka Write Coalescing) work
request path. PCIE maps User Space Doorbell BAR2 region writes to the new
interface to SGE. SGE pulls a new message from PCIE new interface and if its a
coalesced write work request then pushes it for processing. This patch copies
coalesced work request to memory mapped BAR2 space.
Signed-off-by: default avatarSantosh Rastapur <santosh@chelsio.com>
Signed-off-by: default avatarVipul Pandya <vipul@chelsio.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
parent 251f9e88
......@@ -439,6 +439,7 @@ struct sge_txq {
spinlock_t db_lock;
int db_disabled;
unsigned short db_pidx;
u64 udb;
};
struct sge_eth_txq { /* state for an SGE Ethernet Tx queue */
......@@ -543,6 +544,7 @@ enum chip_type {
struct adapter {
void __iomem *regs;
void __iomem *bar2;
struct pci_dev *pdev;
struct device *pdev_dev;
unsigned int mbox;
......
......@@ -1327,6 +1327,8 @@ static char stats_strings[][ETH_GSTRING_LEN] = {
"VLANinsertions ",
"GROpackets ",
"GROmerged ",
"WriteCoalSuccess ",
"WriteCoalFail ",
};
static int get_sset_count(struct net_device *dev, int sset)
......@@ -1422,11 +1424,25 @@ static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
{
struct port_info *pi = netdev_priv(dev);
struct adapter *adapter = pi->adapter;
u32 val1, val2;
t4_get_port_stats(adapter, pi->tx_chan, (struct port_stats *)data);
data += sizeof(struct port_stats) / sizeof(u64);
collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
data += sizeof(struct queue_port_stats) / sizeof(u64);
if (!is_t4(adapter->chip)) {
t4_write_reg(adapter, SGE_STAT_CFG, STATSOURCE_T5(7));
val1 = t4_read_reg(adapter, SGE_STAT_TOTAL);
val2 = t4_read_reg(adapter, SGE_STAT_MATCH);
*data = val1 - val2;
data++;
*data = val2;
data++;
} else {
memset(data, 0, 2 * sizeof(u64));
*data += 2;
}
}
/*
......@@ -5337,10 +5353,11 @@ static void free_some_resources(struct adapter *adapter)
#define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
#define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
#define SEGMENT_SIZE 128
static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
{
int func, i, err;
int func, i, err, s_qpp, qpp, num_seg;
struct port_info *pi;
bool highdma = false;
struct adapter *adapter = NULL;
......@@ -5420,7 +5437,34 @@ static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
err = t4_prep_adapter(adapter);
if (err)
goto out_unmap_bar;
goto out_unmap_bar0;
if (!is_t4(adapter->chip)) {
s_qpp = QUEUESPERPAGEPF1 * adapter->fn;
qpp = 1 << QUEUESPERPAGEPF0_GET(t4_read_reg(adapter,
SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp);
num_seg = PAGE_SIZE / SEGMENT_SIZE;
/* Each segment size is 128B. Write coalescing is enabled only
* when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
* queue is less no of segments that can be accommodated in
* a page size.
*/
if (qpp > num_seg) {
dev_err(&pdev->dev,
"Incorrect number of egress queues per page\n");
err = -EINVAL;
goto out_unmap_bar0;
}
adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
pci_resource_len(pdev, 2));
if (!adapter->bar2) {
dev_err(&pdev->dev, "cannot map device bar2 region\n");
err = -ENOMEM;
goto out_unmap_bar0;
}
}
setup_memwin(adapter);
err = adap_init0(adapter);
setup_memwin_rdma(adapter);
......@@ -5552,6 +5596,9 @@ static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
out_free_dev:
free_some_resources(adapter);
out_unmap_bar:
if (!is_t4(adapter->chip))
iounmap(adapter->bar2);
out_unmap_bar0:
iounmap(adapter->regs);
out_free_adapter:
kfree(adapter);
......@@ -5602,6 +5649,8 @@ static void remove_one(struct pci_dev *pdev)
free_some_resources(adapter);
iounmap(adapter->regs);
if (!is_t4(adapter->chip))
iounmap(adapter->bar2);
kfree(adapter);
pci_disable_pcie_error_reporting(pdev);
pci_disable_device(pdev);
......
......@@ -816,6 +816,22 @@ static void write_sgl(const struct sk_buff *skb, struct sge_txq *q,
*end = 0;
}
/* This function copies 64 byte coalesced work request to
* memory mapped BAR2 space(user space writes).
* For coalesced WR SGE, fetches data from the FIFO instead of from Host.
*/
static void cxgb_pio_copy(u64 __iomem *dst, u64 *src)
{
int count = 8;
while (count) {
writeq(*src, dst);
src++;
dst++;
count--;
}
}
/**
* ring_tx_db - check and potentially ring a Tx queue's doorbell
* @adap: the adapter
......@@ -826,11 +842,25 @@ static void write_sgl(const struct sk_buff *skb, struct sge_txq *q,
*/
static inline void ring_tx_db(struct adapter *adap, struct sge_txq *q, int n)
{
unsigned int *wr, index;
wmb(); /* write descriptors before telling HW */
spin_lock(&q->db_lock);
if (!q->db_disabled) {
t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
QID(q->cntxt_id) | PIDX(n));
if (is_t4(adap->chip)) {
t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
QID(q->cntxt_id) | PIDX(n));
} else {
if (n == 1) {
index = q->pidx ? (q->pidx - 1) : (q->size - 1);
wr = (unsigned int *)&q->desc[index];
cxgb_pio_copy((u64 __iomem *)
(adap->bar2 + q->udb + 64),
(u64 *)wr);
} else
writel(n, adap->bar2 + q->udb + 8);
wmb();
}
}
q->db_pidx = q->pidx;
spin_unlock(&q->db_lock);
......@@ -2151,11 +2181,27 @@ int t4_sge_alloc_rxq(struct adapter *adap, struct sge_rspq *iq, bool fwevtq,
static void init_txq(struct adapter *adap, struct sge_txq *q, unsigned int id)
{
q->cntxt_id = id;
if (!is_t4(adap->chip)) {
unsigned int s_qpp;
unsigned short udb_density;
unsigned long qpshift;
int page;
s_qpp = QUEUESPERPAGEPF1 * adap->fn;
udb_density = 1 << QUEUESPERPAGEPF0_GET((t4_read_reg(adap,
SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp));
qpshift = PAGE_SHIFT - ilog2(udb_density);
q->udb = q->cntxt_id << qpshift;
q->udb &= PAGE_MASK;
page = q->udb / PAGE_SIZE;
q->udb += (q->cntxt_id - (page * udb_density)) * 128;
}
q->in_use = 0;
q->cidx = q->pidx = 0;
q->stops = q->restarts = 0;
q->stat = (void *)&q->desc[q->size];
q->cntxt_id = id;
spin_lock_init(&q->db_lock);
adap->sge.egr_map[id - adap->sge.egr_start] = q;
}
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment