lkml.org 
[lkml]   [2018]   [Mar]   [27]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Subject[PATCH net-next] net: mvpp2: Use relaxed I/O in data path
Date
From: Yan Markman <ymarkman@marvell.com>

Use relaxed I/O on the hot path. This achieves significant performance
improvements. On a 10G link, this makes a basic iperf TCP test go from
an average of 4.5 Gbits/sec to about 9.40 Gbits/sec.

Signed-off-by: Yan Markman <ymarkman@marvell.com>
[Maxime: Commit message, cosmetic changes]
Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
---
David, this patch should not conflict with other the pending PPv2 series I sent
earlier ("[PATCH net-next 0/2] net: mvpp2: Remove unnecessary dynamic allocs")

drivers/net/ethernet/marvell/mvpp2.c | 43 +++++++++++++++++++++++++-----------
1 file changed, 30 insertions(+), 13 deletions(-)

diff --git a/drivers/net/ethernet/marvell/mvpp2.c b/drivers/net/ethernet/marvell/mvpp2.c
index f8bc3d4a39ff..5a68e00e9c5b 100644
--- a/drivers/net/ethernet/marvell/mvpp2.c
+++ b/drivers/net/ethernet/marvell/mvpp2.c
@@ -1359,6 +1359,10 @@ static u32 mvpp2_read(struct mvpp2 *priv, u32 offset)
return readl(priv->swth_base[0] + offset);
}

+static u32 mvpp2_read_relaxed(struct mvpp2 *priv, u32 offset)
+{
+ return readl_relaxed(priv->swth_base[0] + offset);
+}
/* These accessors should be used to access:
*
* - per-CPU registers, where each CPU has its own copy of the
@@ -1407,6 +1411,18 @@ static u32 mvpp2_percpu_read(struct mvpp2 *priv, int cpu,
return readl(priv->swth_base[cpu] + offset);
}

+static void mvpp2_percpu_write_relaxed(struct mvpp2 *priv, int cpu,
+ u32 offset, u32 data)
+{
+ writel_relaxed(data, priv->swth_base[cpu] + offset);
+}
+
+static u32 mvpp2_percpu_read_relaxed(struct mvpp2 *priv, int cpu,
+ u32 offset)
+{
+ return readl_relaxed(priv->swth_base[cpu] + offset);
+}
+
static dma_addr_t mvpp2_txdesc_dma_addr_get(struct mvpp2_port *port,
struct mvpp2_tx_desc *tx_desc)
{
@@ -4488,8 +4504,8 @@ static inline void mvpp2_bm_pool_put(struct mvpp2_port *port, int pool,
<< MVPP22_BM_ADDR_HIGH_VIRT_RLS_SHIFT) &
MVPP22_BM_ADDR_HIGH_VIRT_RLS_MASK;

- mvpp2_percpu_write(port->priv, cpu,
- MVPP22_BM_ADDR_HIGH_RLS_REG, val);
+ mvpp2_percpu_write_relaxed(port->priv, cpu,
+ MVPP22_BM_ADDR_HIGH_RLS_REG, val);
}

/* MVPP2_BM_VIRT_RLS_REG is not interpreted by HW, and simply
@@ -4497,10 +4513,10 @@ static inline void mvpp2_bm_pool_put(struct mvpp2_port *port, int pool,
* descriptor. Instead of storing the virtual address, we
* store the physical address
*/
- mvpp2_percpu_write(port->priv, cpu,
- MVPP2_BM_VIRT_RLS_REG, buf_phys_addr);
- mvpp2_percpu_write(port->priv, cpu,
- MVPP2_BM_PHY_RLS_REG(pool), buf_dma_addr);
+ mvpp2_percpu_write_relaxed(port->priv, cpu,
+ MVPP2_BM_VIRT_RLS_REG, buf_phys_addr);
+ mvpp2_percpu_write_relaxed(port->priv, cpu,
+ MVPP2_BM_PHY_RLS_REG(pool), buf_dma_addr);

put_cpu();
}
@@ -5592,7 +5608,8 @@ static int mvpp2_aggr_desc_num_check(struct mvpp2 *priv,
if ((aggr_txq->count + num) > MVPP2_AGGR_TXQ_SIZE) {
/* Update number of occupied aggregated Tx descriptors */
int cpu = smp_processor_id();
- u32 val = mvpp2_read(priv, MVPP2_AGGR_TXQ_STATUS_REG(cpu));
+ u32 val = mvpp2_read_relaxed(priv,
+ MVPP2_AGGR_TXQ_STATUS_REG(cpu));

aggr_txq->count = val & MVPP2_AGGR_TXQ_PENDING_MASK;
}
@@ -5616,9 +5633,9 @@ static int mvpp2_txq_alloc_reserved_desc(struct mvpp2 *priv,
int cpu = smp_processor_id();

val = (txq->id << MVPP2_TXQ_RSVD_REQ_Q_OFFSET) | num;
- mvpp2_percpu_write(priv, cpu, MVPP2_TXQ_RSVD_REQ_REG, val);
+ mvpp2_percpu_write_relaxed(priv, cpu, MVPP2_TXQ_RSVD_REQ_REG, val);

- val = mvpp2_percpu_read(priv, cpu, MVPP2_TXQ_RSVD_RSLT_REG);
+ val = mvpp2_percpu_read_relaxed(priv, cpu, MVPP2_TXQ_RSVD_RSLT_REG);

return val & MVPP2_TXQ_RSVD_RSLT_MASK;
}
@@ -5723,8 +5740,8 @@ static inline int mvpp2_txq_sent_desc_proc(struct mvpp2_port *port,
u32 val;

/* Reading status reg resets transmitted descriptor counter */
- val = mvpp2_percpu_read(port->priv, smp_processor_id(),
- MVPP2_TXQ_SENT_REG(txq->id));
+ val = mvpp2_percpu_read_relaxed(port->priv, smp_processor_id(),
+ MVPP2_TXQ_SENT_REG(txq->id));

return (val & MVPP2_TRANSMITTED_COUNT_MASK) >>
MVPP2_TRANSMITTED_COUNT_OFFSET;
@@ -7090,8 +7107,8 @@ static int mvpp2_poll(struct napi_struct *napi, int budget)
*
* Each CPU has its own Rx/Tx cause register
*/
- cause_rx_tx = mvpp2_percpu_read(port->priv, qv->sw_thread_id,
- MVPP2_ISR_RX_TX_CAUSE_REG(port->id));
+ cause_rx_tx = mvpp2_percpu_read_relaxed(port->priv, qv->sw_thread_id,
+ MVPP2_ISR_RX_TX_CAUSE_REG(port->id));

cause_misc = cause_rx_tx & MVPP2_CAUSE_MISC_SUM_MASK;
if (cause_misc) {
--
2.11.0
\
 
 \ /
  Last update: 2018-03-27 16:50    [W:0.075 / U:0.592 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site