Re: [U-Boot] [PATCH] keystone2: use appropriate HD field for destination port

[Vitaly Andrianov]

On 07/08/2015 01:05 PM, Ivan Khoronzhuk wrote:
Vitaly,

On 08.07.15 20:05, Vitaly Andrianov wrote:


On 07/08/2015 12:38 PM, Ivan Khoronzhuk wrote:
Hi, Vitaly

I suppose it's better to decide in upper driver how to use swinfo field.
Like in drivers/net/keystone_net.c

The keystone navigator supposed to be used as a tool for communicating
between different IPs, and each of them decide how to use swinfo fields.
It's protocol specific information and should be known only in sending
parts. What if tomorrow you will decide to send some packet to PA?, you
will rewrite this function again?

It's not the place for such kind information.

Even more, this is the h/w specific decision and no need to check this
for each sent packet. You better statically assign how to use this field
depending on h/w revision, using #if.

On 08.07.15 18:45, Vitaly Andrianov wrote:
K2L and L2E have different from K2HK EthSS version, which uses tag_info
field for destination slave port. This commit adds the dest_port_info
field
to the struct pktdma_cfg, to configure which HD filed tag_info or
pkt_info
shall be used to configure descriptor.

Before that commit the swinfo[2] was used for that purpose. Even if
that
worked on K2HK devices, the correct field for K2HK is the pkt_info.

The netcp_send() configure appropriate HD info field depending on the
direct_info of the currently using netcp.

Signed-off-by: Vitaly Andrianov <vita...@ti.com>
Acked-by: Murali Karicheri <m-kariche...@ti.com>
---
  arch/arm/include/asm/ti-common/keystone_nav.h |  9 ++++++++-
  drivers/dma/keystone_nav.c                    | 12 ++++++++++--
  drivers/net/keystone_net.c                    |  3 +--
  3 files changed, 19 insertions(+), 5 deletions(-)

diff --git a/arch/arm/include/asm/ti-common/keystone_nav.h
b/arch/arm/include/asm/ti-common/keystone_nav.h
index 696d8c6..5a0e391 100644
--- a/arch/arm/include/asm/ti-common/keystone_nav.h
+++ b/arch/arm/include/asm/ti-common/keystone_nav.h
@@ -152,6 +152,11 @@ struct rx_flow_regs {
      u32    thresh[3];
  };

+enum dest_port_info {
+    PKT_INFO,
+    TAG_INFO
+};
+
  struct pktdma_cfg {
      struct global_ctl_regs    *global;
      struct tx_chan_regs    *tx_ch;
@@ -167,6 +172,7 @@ struct pktdma_cfg {
      u32            tx_snd_q;

      u32            rx_flow; /* flow that is used for RX */
+    enum dest_port_info     dest_port_info;/* HD fiels for dest port
bits */
  };

  extern struct pktdma_cfg netcp_pktdma;
@@ -184,7 +190,8 @@ struct rx_buff_desc {

  int ksnav_close(struct pktdma_cfg *pktdma);
  int ksnav_init(struct pktdma_cfg *pktdma, struct rx_buff_desc
*rx_buffers);
-int ksnav_send(struct pktdma_cfg *pktdma, u32 *pkt, int num_bytes,
u32 swinfo2);
+int ksnav_send(struct pktdma_cfg *pktdma, u32 *pkt, int num_bytes,
+           u32 dest_port);
  void *ksnav_recv(struct pktdma_cfg *pktdma, u32 **pkt, int
*num_bytes);
  void ksnav_release_rxhd(struct pktdma_cfg *pktdma, void *hd);

diff --git a/drivers/dma/keystone_nav.c b/drivers/dma/keystone_nav.c
index dfca75a..64b1cee 100644
--- a/drivers/dma/keystone_nav.c
+++ b/drivers/dma/keystone_nav.c
@@ -278,7 +278,8 @@ int ksnav_close(struct pktdma_cfg *pktdma)
      return QM_OK;
  }

-int ksnav_send(struct pktdma_cfg *pktdma, u32 *pkt, int num_bytes,
u32 swinfo2)
+int ksnav_send(struct pktdma_cfg *pktdma, u32 *pkt, int num_bytes,
+           u32 dest_port)
  {
      struct qm_host_desc *hd;

@@ -286,8 +287,15 @@ int ksnav_send(struct pktdma_cfg *pktdma, u32
*pkt, int num_bytes, u32 swinfo2)
      if (hd == NULL)
          return QM_ERR;

+    dest_port &= 0xf;
      hd->desc_info    = num_bytes;
-    hd->swinfo[2]    = swinfo2;
+    if (pktdma->dest_port_info == PKT_INFO) {
+        hd->packet_info    = qm_cfg->qpool_num | (dest_port << 16);
+    } else {
+        hd->packet_info = qm_cfg->qpool_num;
+        hd->tag_info = dest_port;
+    }
+
      hd->packet_info = qm_cfg->qpool_num;

      qm_buff_push(hd, pktdma->tx_snd_q, pkt, num_bytes);
diff --git a/drivers/net/keystone_net.c b/drivers/net/keystone_net.c
index 0c5fdee..e2adb67 100644
--- a/drivers/net/keystone_net.c
+++ b/drivers/net/keystone_net.c
@@ -381,8 +381,7 @@ int32_t cpmac_drv_send(u32 *buffer, int num_bytes,
int slave_port_num)
      if (num_bytes < EMAC_MIN_ETHERNET_PKT_SIZE)
          num_bytes = EMAC_MIN_ETHERNET_PKT_SIZE;

-    return ksnav_send(&netcp_pktdma, buffer,
-              num_bytes, (slave_port_num) << 16);
+    return ksnav_send(&netcp_pktdma, buffer, num_bytes,
slave_port_num);
  }

  /* Eth device open */


Hi Ivan,

I agree with you. And probably we will need to implement your proposal
in future commits. This commit is to fix the bug, which is in existing
driver.

Thanks,
Vitaly

Sorry, I supposed that first msg was not sent.
It's better to fix it in drivers/net/keystone_net.c


Ivan,

I don't understand your proposal. The network driver doesn't know 
anything about CPPI DMA engine internals. That is the navigator's job to 
fill appropriate packet BD fields and it only takes care about HW 
version differences.

Thanks,
Vitaly
_______________________________________________
U-Boot mailing list
U-Boot@lists.denx.de
http://lists.denx.de/mailman/listinfo/u-boot