]> git.kernelconcepts.de Git - karo-tx-linux.git/commitdiff
sfc: use new performance based event queue init
authorBert Kenward <bkenward@solarflare.com>
Thu, 11 Aug 2016 12:01:54 +0000 (13:01 +0100)
committerDavid S. Miller <davem@davemloft.net>
Sat, 13 Aug 2016 03:42:19 +0000 (20:42 -0700)
Rather than explicitly specifying flags we can now specify a desired
performance target to the firmware, ie higher throughput or lower latency.
For now we use the default "auto" configuration.

Signed-off-by: Bert Kenward <bkenward@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/sfc/ef10.c

index fd5d086f53a5672594975cd2179edee02c696b2e..e57fa2e8e7c80fe90365eb9c7d948939ee5d9f22 100644 (file)
@@ -2541,13 +2541,12 @@ fail:
 static int efx_ef10_ev_init(struct efx_channel *channel)
 {
        MCDI_DECLARE_BUF(inbuf,
-                        MC_CMD_INIT_EVQ_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
-                                               EFX_BUF_SIZE));
-       MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_OUT_LEN);
+                        MC_CMD_INIT_EVQ_V2_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
+                                                  EFX_BUF_SIZE));
+       MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_V2_OUT_LEN);
        size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
        struct efx_nic *efx = channel->efx;
        struct efx_ef10_nic_data *nic_data;
-       bool supports_rx_merge;
        size_t inlen, outlen;
        unsigned int enabled, implemented;
        dma_addr_t dma_addr;
@@ -2555,9 +2554,6 @@ static int efx_ef10_ev_init(struct efx_channel *channel)
        int i;
 
        nic_data = efx->nic_data;
-       supports_rx_merge =
-               !!(nic_data->datapath_caps &
-                  1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
 
        /* Fill event queue with all ones (i.e. empty events) */
        memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
@@ -2566,11 +2562,6 @@ static int efx_ef10_ev_init(struct efx_channel *channel)
        MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
        /* INIT_EVQ expects index in vector table, not absolute */
        MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
-       MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
-                             INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
-                             INIT_EVQ_IN_FLAG_RX_MERGE, 1,
-                             INIT_EVQ_IN_FLAG_TX_MERGE, 1,
-                             INIT_EVQ_IN_FLAG_CUT_THRU, !supports_rx_merge);
        MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
                       MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
        MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
@@ -2579,6 +2570,27 @@ static int efx_ef10_ev_init(struct efx_channel *channel)
                       MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
        MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
 
+       if (nic_data->datapath_caps2 &
+           1 << MC_CMD_GET_CAPABILITIES_V2_OUT_INIT_EVQ_V2_LBN) {
+               /* Use the new generic approach to specifying event queue
+                * configuration, requesting lower latency or higher throughput.
+                * The options that actually get used appear in the output.
+                */
+               MCDI_POPULATE_DWORD_2(inbuf, INIT_EVQ_V2_IN_FLAGS,
+                                     INIT_EVQ_V2_IN_FLAG_INTERRUPTING, 1,
+                                     INIT_EVQ_V2_IN_FLAG_TYPE,
+                                     MC_CMD_INIT_EVQ_V2_IN_FLAG_TYPE_AUTO);
+       } else {
+               bool cut_thru = !(nic_data->datapath_caps &
+                       1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
+
+               MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
+                                     INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
+                                     INIT_EVQ_IN_FLAG_RX_MERGE, 1,
+                                     INIT_EVQ_IN_FLAG_TX_MERGE, 1,
+                                     INIT_EVQ_IN_FLAG_CUT_THRU, cut_thru);
+       }
+
        dma_addr = channel->eventq.buf.dma_addr;
        for (i = 0; i < entries; ++i) {
                MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
@@ -2589,6 +2601,13 @@ static int efx_ef10_ev_init(struct efx_channel *channel)
 
        rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
                          outbuf, sizeof(outbuf), &outlen);
+
+       if (outlen >= MC_CMD_INIT_EVQ_V2_OUT_LEN)
+               netif_dbg(efx, drv, efx->net_dev,
+                         "Channel %d using event queue flags %08x\n",
+                         channel->channel,
+                         MCDI_DWORD(outbuf, INIT_EVQ_V2_OUT_FLAGS));
+
        /* IRQ return is ignored */
        if (channel->channel || rc)
                return rc;