qca-nss-dp: add support for IPQ50xx

Add support for the Qualcomm IPQ50xx in the QCA NSS dataplane driver.
The QCA implementation uses depracated DMA api calls and a downstream SCM
call, so convert to proper Linux DMA and SCM api calls.
In addition, add fixed-link support to support SGMII which is used to
connect the internal IPQ50xx switch to an external switch (ex. QCA8337)

Co-developed-by: Ziyang Huang <hzyitc@outlook.com>
Signed-off-by: Ziyang Huang <hzyitc@outlook.com>
Signed-off-by: George Moussalem <george.moussalem@outlook.com>
Link: https://github.com/openwrt/openwrt/pull/17182
Signed-off-by: Robert Marko <robimarko@gmail.com>

diff --git a/package/kernel/qca-nss-dp/patches/0012-01-syn-gmac-use-standard-DMA-api.patch b/package/kernel/qca-nss-dp/patches/0012-01-syn-gmac-use-standard-DMA-api.patch
new file mode 100644 (file)
index 0000000..882bc8c
--- /dev/null
+++ b/package/kernel/qca-nss-dp/patches/0012-01-syn-gmac-use-standard-DMA-api.patch
@@ -0,0 +1,245 @@
+From 5ad8cf24897ff903112967a9662cb13ed4cbbf57 Mon Sep 17 00:00:00 2001
+From: Ziyang Huang <hzyitc@outlook.com>
+Date: Mon, 22 Apr 2024 21:47:58 +0800
+Subject: [PATCH] syn-gmac: use standard DMA api
+
+The SSDK uses several old DMA API calls based on raw assembly which have 
+been deprecated. So let's convert to and use the standard Linux DMA API 
+instead.
+
+Signed-off-by: Ziyang Huang <hzyitc@outlook.com>
+Signed-off-by: George Moussalem <george.moussalem@outlook.com>
+---
+ hal/dp_ops/syn_gmac_dp/syn_dp_cfg_rx.c | 14 ++++++--
+ hal/dp_ops/syn_gmac_dp/syn_dp_cfg_tx.c |  2 ++
+ hal/dp_ops/syn_gmac_dp/syn_dp_rx.c     | 47 +++++++++++++-------------
+ hal/dp_ops/syn_gmac_dp/syn_dp_tx.c     | 23 ++++---------
+ 4 files changed, 42 insertions(+), 44 deletions(-)
+
+diff --git a/hal/dp_ops/syn_gmac_dp/syn_dp_cfg_rx.c b/hal/dp_ops/syn_gmac_dp/syn_dp_cfg_rx.c
+index 8cbbcaaf..1c9006c7 100644
+--- a/hal/dp_ops/syn_gmac_dp/syn_dp_cfg_rx.c
++++ b/hal/dp_ops/syn_gmac_dp/syn_dp_cfg_rx.c
+@@ -26,6 +26,7 @@ static int syn_dp_cfg_rx_setup_desc_queue(struct syn_dp_info *dev_info)
+ {
+       struct syn_dp_info_rx *rx_info = &dev_info->dp_info_rx;
+       struct dma_desc_rx *first_desc = NULL;
++      dma_addr_t dma_addr;
+       struct net_device *netdev = rx_info->netdev;
+ 
+       netdev_dbg(netdev, "Total size of memory required for Rx Descriptors in Ring Mode = %u\n", (uint32_t)((sizeof(struct dma_desc_rx) * SYN_DP_RX_DESC_SIZE)));
+@@ -33,13 +34,15 @@ static int syn_dp_cfg_rx_setup_desc_queue(struct syn_dp_info *dev_info)
+       /*
+        * Allocate cacheable descriptors for Rx
+        */
+-      first_desc = kzalloc(sizeof(struct dma_desc_rx) * SYN_DP_RX_DESC_SIZE, GFP_KERNEL);
++      first_desc = dma_alloc_coherent(rx_info->dev,
++                                      sizeof(struct dma_desc_rx) * SYN_DP_RX_DESC_SIZE,
++                                      &dma_addr, GFP_KERNEL);
+       if (!first_desc) {
+               netdev_dbg(netdev, "Error in Rx Descriptor Memory allocation in Ring mode\n");
+               return -ENOMEM;
+       }
+ 
+-      dev_info->rx_desc_dma_addr = (dma_addr_t)virt_to_phys(first_desc);
++      dev_info->rx_desc_dma_addr = dma_addr;
+       rx_info->rx_desc = first_desc;
+       syn_dp_gmac_rx_desc_init_ring(rx_info->rx_desc, SYN_DP_RX_DESC_SIZE);
+ 
+@@ -98,6 +101,10 @@ void syn_dp_cfg_rx_cleanup_rings(struct syn_dp_info *dev_info)
+       for (i = 0; i < rx_info->busy_rx_desc_cnt; i++) {
+               rx_skb_index = (rx_skb_index + i) & SYN_DP_RX_DESC_MAX_INDEX;
+               rxdesc = rx_info->rx_desc;
++
++              dma_unmap_single(rx_info->dev, rxdesc->buffer1,
++                               rxdesc->length, DMA_FROM_DEVICE);
++
+               skb = rx_info->rx_buf_pool[rx_skb_index].skb;
+               if (unlikely(skb != NULL)) {
+                       dev_kfree_skb_any(skb);
+@@ -105,7 +112,8 @@ void syn_dp_cfg_rx_cleanup_rings(struct syn_dp_info *dev_info)
+               }
+       }
+ 
+-      kfree(rx_info->rx_desc);
++      dma_free_coherent(rx_info->dev, (sizeof(struct dma_desc_rx) * SYN_DP_RX_DESC_SIZE),
++                        rx_info->rx_desc, dev_info->rx_desc_dma_addr);
+       rx_info->rx_desc = NULL;
+       dev_info->rx_desc_dma_addr = (dma_addr_t)0;
+ }
+diff --git a/hal/dp_ops/syn_gmac_dp/syn_dp_cfg_tx.c b/hal/dp_ops/syn_gmac_dp/syn_dp_cfg_tx.c
+index bf5e19a0..284e8880 100644
+--- a/hal/dp_ops/syn_gmac_dp/syn_dp_cfg_tx.c
++++ b/hal/dp_ops/syn_gmac_dp/syn_dp_cfg_tx.c
+@@ -91,6 +91,8 @@ void syn_dp_cfg_tx_cleanup_rings(struct syn_dp_info *dev_info)
+               tx_skb_index = syn_dp_tx_inc_index(tx_skb_index, i);
+               txdesc = tx_info->tx_desc;
+ 
++              dma_unmap_single(tx_info->dev, txdesc->buffer1, txdesc->length, DMA_TO_DEVICE);
++
+               skb = tx_info->tx_buf_pool[tx_skb_index].skb;
+               if (unlikely(skb != NULL)) {
+                       dev_kfree_skb_any(skb);
+diff --git a/hal/dp_ops/syn_gmac_dp/syn_dp_rx.c b/hal/dp_ops/syn_gmac_dp/syn_dp_rx.c
+index 1ddeb7d6..1798d4e7 100644
+--- a/hal/dp_ops/syn_gmac_dp/syn_dp_rx.c
++++ b/hal/dp_ops/syn_gmac_dp/syn_dp_rx.c
+@@ -73,16 +73,6 @@ static inline void syn_dp_rx_refill_one_desc(struct dma_desc_rx *rx_desc,
+  */
+ static inline void syn_dp_rx_inval_and_flush(struct syn_dp_info_rx *rx_info, uint32_t start, uint32_t end)
+ {
+-      /*
+-       * Batched flush and invalidation of the rx descriptors
+-       */
+-      if (end > start) {
+-              dmac_flush_range_no_dsb((void *)&rx_info->rx_desc[start], (void *)&rx_info->rx_desc[end] + sizeof(struct dma_desc_rx));
+-      } else {
+-              dmac_flush_range_no_dsb((void *)&rx_info->rx_desc[start], (void *)&rx_info->rx_desc[SYN_DP_RX_DESC_MAX_INDEX] + sizeof(struct dma_desc_rx));
+-              dmac_flush_range_no_dsb((void *)&rx_info->rx_desc[0], (void *)&rx_info->rx_desc[end] + sizeof(struct dma_desc_rx));
+-      }
+-
+       dsb(st);
+ }
+ 
+@@ -124,15 +114,19 @@ int syn_dp_rx_refill_page_mode(struct syn_dp_info_rx *rx_info)
+                       break;
+               }
+ 
++              skb_fill_page_desc(skb, 0, pg, 0, PAGE_SIZE);
++
+               /*
+                * Get virtual address of allocated page.
+                */
+               page_addr = page_address(pg);
+-              dma_addr = (dma_addr_t)virt_to_phys(page_addr);
+-
+-              skb_fill_page_desc(skb, 0, pg, 0, PAGE_SIZE);
++              dma_addr  = dma_map_page(rx_info->dev, pg, 0, PAGE_SIZE, DMA_FROM_DEVICE);
++              if (unlikely(dma_mapping_error(rx_info->dev, dma_addr))) {
++                      dev_kfree_skb(skb);
++                      netdev_dbg(netdev, "DMA mapping failed for empty buffer\n");
++                      break;
++              }
+ 
+-              dmac_inv_range_no_dsb(page_addr, (page_addr + PAGE_SIZE));
+               rx_refill_idx = rx_info->rx_refill_idx;
+               rx_desc = rx_info->rx_desc + rx_refill_idx;
+ 
+@@ -181,8 +175,15 @@ int syn_dp_rx_refill(struct syn_dp_info_rx *rx_info)
+ 
+               skb_reserve(skb, SYN_DP_SKB_HEADROOM + NET_IP_ALIGN);
+ 
+-              dma_addr = (dma_addr_t)virt_to_phys(skb->data);
+-              dmac_inv_range_no_dsb((void *)skb->data, (void *)(skb->data + inval_len));
++              dma_addr = dma_map_single(rx_info->dev, skb->data,
++                                        inval_len,
++                                        DMA_FROM_DEVICE);
++              if (unlikely(dma_mapping_error(rx_info->dev, dma_addr))) {
++                      dev_kfree_skb(skb);
++                      netdev_dbg(netdev, "DMA mapping failed for empty buffer\n");
++                      break;
++              }
++
+               rx_refill_idx = rx_info->rx_refill_idx;
+               rx_desc = rx_info->rx_desc + rx_refill_idx;
+ 
+@@ -407,12 +408,6 @@ int syn_dp_rx(struct syn_dp_info_rx *rx_info, int budget)
+        * this code is executing.
+        */
+       end = syn_dp_rx_inc_index(rx_info->rx_idx, busy);
+-      if (end > start) {
+-              dmac_inv_range_no_dsb((void *)&rx_info->rx_desc[start], (void *)&rx_info->rx_desc[end] + sizeof(struct dma_desc_rx));
+-      } else {
+-              dmac_inv_range_no_dsb((void *)&rx_info->rx_desc[start], (void *)&rx_info->rx_desc[SYN_DP_RX_DESC_MAX_INDEX] + sizeof(struct dma_desc_rx));
+-              dmac_inv_range_no_dsb((void *)&rx_info->rx_desc[0], (void *)&rx_info->rx_desc[end] + sizeof(struct dma_desc_rx));
+-      }
+ 
+       dsb(st);
+ 
+@@ -439,8 +434,12 @@ int syn_dp_rx(struct syn_dp_info_rx *rx_info, int budget)
+                * speculative prefetch by CPU may have occurred.
+                */
+               frame_length = syn_dp_gmac_get_rx_desc_frame_length(status);
+-              dmac_inv_range((void *)rx_buf->map_addr_virt,
+-                      (void *)(((uint8_t *)rx_buf->map_addr_virt) + frame_length));
++              if (likely(!rx_info->page_mode))
++                      dma_unmap_single(rx_info->dev, rx_desc->buffer1,
++                                       rx_info->alloc_buf_len, DMA_FROM_DEVICE);
++              else
++                      dma_unmap_page(rx_info->dev, rx_desc->buffer1,
++                                     PAGE_SIZE, DMA_FROM_DEVICE);
+               prefetch((void *)rx_buf->map_addr_virt);
+ 
+               rx_next_idx = syn_dp_rx_inc_index(rx_idx, 1);
+diff --git a/hal/dp_ops/syn_gmac_dp/syn_dp_tx.c b/hal/dp_ops/syn_gmac_dp/syn_dp_tx.c
+index c97e252b..6d4adb3f 100644
+--- a/hal/dp_ops/syn_gmac_dp/syn_dp_tx.c
++++ b/hal/dp_ops/syn_gmac_dp/syn_dp_tx.c
+@@ -104,9 +104,7 @@ static inline struct dma_desc_tx *syn_dp_tx_process_nr_frags(struct syn_dp_info_
+               BUG_ON(!length);
+ #endif
+ 
+-              dma_addr = (dma_addr_t)virt_to_phys(frag_addr);
+-
+-              dmac_clean_range_no_dsb(frag_addr, frag_addr + length);
++              dma_addr = dma_map_single(tx_info->dev, frag_addr, length, DMA_TO_DEVICE);
+ 
+               *total_length += length;
+               tx_desc = syn_dp_tx_set_desc_sg(tx_info, dma_addr, length, DESC_OWN_BY_DMA);
+@@ -150,8 +148,7 @@ int syn_dp_tx_nr_frags(struct syn_dp_info_tx *tx_info, struct sk_buff *skb)
+       /*
+        * Flush the dma for non-paged skb data
+        */
+-      dma_addr = (dma_addr_t)virt_to_phys(skb->data);
+-      dmac_clean_range_no_dsb((void *)skb->data, (void *)(skb->data + length));
++      dma_addr = dma_map_single(tx_info->dev, skb->data, length, DMA_TO_DEVICE);
+ 
+       total_len = length;
+ 
+@@ -256,12 +253,7 @@ int syn_dp_tx_frag_list(struct syn_dp_info_tx *tx_info, struct sk_buff *skb)
+               return NETDEV_TX_BUSY;
+       }
+ 
+-      dma_addr = (dma_addr_t)virt_to_phys(skb->data);
+-
+-      /*
+-       * Flush the data area of the head skb
+-       */
+-      dmac_clean_range_no_dsb((void *)skb->data, (void *)(skb->data + length));
++      dma_addr = dma_map_single(tx_info->dev, skb->data, length, DMA_TO_DEVICE);
+ 
+       total_len = length;
+ 
+@@ -290,9 +282,7 @@ int syn_dp_tx_frag_list(struct syn_dp_info_tx *tx_info, struct sk_buff *skb)
+               BUG_ON(!length);
+ #endif
+ 
+-              dma_addr = (dma_addr_t)virt_to_phys(iter_skb->data);
+-
+-              dmac_clean_range_no_dsb((void *)iter_skb->data, (void *)(iter_skb->data + length));
++              dma_addr = dma_map_single(tx_info->dev, iter_skb->data, length, DMA_TO_DEVICE);
+ 
+               total_len += length;
+ 
+@@ -445,6 +435,7 @@ int syn_dp_tx_complete(struct syn_dp_info_tx *tx_info, int budget)
+                       break;
+               }
+ 
++              dma_unmap_single(tx_info->dev, desc->buffer1, desc->length, DMA_TO_DEVICE);
+ 
+               if (likely(status & DESC_TX_LAST)) {
+                       tx_skb_index = syn_dp_tx_comp_index_get(tx_info);
+@@ -571,9 +562,7 @@ int syn_dp_tx(struct syn_dp_info_tx *tx_info, struct sk_buff *skb)
+               return NETDEV_TX_BUSY;
+       }
+ 
+-      dma_addr = (dma_addr_t)virt_to_phys(skb->data);
+-
+-      dmac_clean_range_no_dsb((void *)skb->data, (void *)(skb->data + skb->len));
++      dma_addr = dma_map_single(tx_info->dev, skb->data, skb->len, DMA_TO_DEVICE);
+ 
+       /*
+        * Queue packet to the GMAC rings
+-- 
+2.40.1
+

diff --git a/package/kernel/qca-nss-dp/patches/0012-02-ipq50xx-use-corrent-scm-function-to-write-tcsr.patch b/package/kernel/qca-nss-dp/patches/0012-02-ipq50xx-use-corrent-scm-function-to-write-tcsr.patch
new file mode 100644 (file)
index 0000000..10a8eef
--- /dev/null
+++ b/package/kernel/qca-nss-dp/patches/0012-02-ipq50xx-use-corrent-scm-function-to-write-tcsr.patch
@@ -0,0 +1,46 @@
+From ba430b1a512dc1972807a1dd5a8d31a78ac572ff Mon Sep 17 00:00:00 2001
+From: Ziyang Huang <hzyitc@outlook.com>
+Date: Mon, 22 Apr 2024 21:49:18 +0800
+Subject: [PATCH] ipq50xx: use corrent scm function to write tcsr
+
+QCA leverages its own API function to write to the TCSR register space, 
+not available upstream. As such, convert to the upstream qcom SCM IO 
+write function.
+
+Signed-off-by: Ziyang Huang <hzyitc@outlook.com>
+Signed-off-by: George Moussalem <george.moussalem@outlook.com>
+---
+ hal/soc_ops/ipq50xx/nss_ipq50xx.c | 9 ++-------
+ 1 file changed, 2 insertions(+), 7 deletions(-)
+
+diff --git a/hal/soc_ops/ipq50xx/nss_ipq50xx.c b/hal/soc_ops/ipq50xx/nss_ipq50xx.c
+index 3e4491c0..e56de1cc 100644
+--- a/hal/soc_ops/ipq50xx/nss_ipq50xx.c
++++ b/hal/soc_ops/ipq50xx/nss_ipq50xx.c
+@@ -18,7 +18,7 @@
+ 
+ #include <linux/of.h>
+ #include <linux/ioport.h>
+-#include <linux/qcom_scm.h>
++#include <linux/firmware/qcom/qcom_scm.h>
+ #include "nss_dp_hal.h"
+ 
+ /*
+@@ -78,13 +78,8 @@ static void nss_dp_hal_tcsr_set(void)
+        * If TZ is not enabled, we can write to the register directly.
+        */
+       if (qcom_scm_is_available()) {
+-#if (LINUX_VERSION_CODE < KERNEL_VERSION(5, 4, 0))
+-              err = qcom_scm_tcsr_reg_write((tcsr_base + TCSR_GMAC_AXI_CACHE_OVERRIDE_OFFSET),
++              err = qcom_scm_io_writel((tcsr_base + TCSR_GMAC_AXI_CACHE_OVERRIDE_OFFSET),
+                                               TCSR_GMAC_AXI_CACHE_OVERRIDE_VALUE);
+-#else
+-              err = qti_scm_tcsr_reg_write((tcsr_base + TCSR_GMAC_AXI_CACHE_OVERRIDE_OFFSET),
+-                                              TCSR_GMAC_AXI_CACHE_OVERRIDE_VALUE);
+-#endif
+               if (err) {
+                       pr_err("%s: SCM TCSR write error: %d\n", __func__, err);
+               }
+-- 
+2.40.1
+

diff --git a/package/kernel/qca-nss-dp/patches/0013-nss_dp_main-support-fixed-link.patch b/package/kernel/qca-nss-dp/patches/0013-nss_dp_main-support-fixed-link.patch
new file mode 100644 (file)
index 0000000..6680468
--- /dev/null
+++ b/package/kernel/qca-nss-dp/patches/0013-nss_dp_main-support-fixed-link.patch
@@ -0,0 +1,49 @@
+From 309a1a330ccaa103a7648e944d97a0032116b338 Mon Sep 17 00:00:00 2001
+From: Ziyang Huang <hzyitc@outlook.com>
+Date: Mon, 22 Apr 2024 21:50:39 +0800
+Subject: [PATCH] nss_dp_main: support fixed-link
+
+Add support for "fixed link" to support ethernet MACs which are not 
+connected to an MDIO managed PHY. For example, IPQ5018 has two internal 
+MACs, one connected to a GE PHY and the other connected to a Uniphy to 
+support connections to an external switch over a fixed link. As such, 
+check for a fixed link in absence of a phy-handle and return an error 
+when no phy-handle and fixed link is found.
+
+Signed-off-by: Ziyang Huang <hzyitc@outlook.com>
+Signed-off-by: George Moussalem <george.moussalem@outlook.com>
+---
+ nss_dp_main.c | 15 ++++++++++++---
+ 1 file changed, 12 insertions(+), 3 deletions(-)
+
+diff --git a/nss_dp_main.c b/nss_dp_main.c
+index 9a09edd5..204063bf 100644
+--- a/nss_dp_main.c
++++ b/nss_dp_main.c
+@@ -619,11 +619,20 @@ static int32_t nss_dp_of_get_pdata(struct device_node *np,
+       }
+ 
+       dp_priv->phy_node = of_parse_phandle(np, "phy-handle", 0);
+-      if (!dp_priv->phy_node) {
+-              pr_err("%s: error parsing phy-handle\n", np->name);
+-              return -EFAULT;
++      if(!dp_priv->phy_node) {
++              if(of_phy_is_fixed_link(np)) {
++                      int ret = of_phy_register_fixed_link(np);
++                      if(ret < 0) {
++                              pr_err("%s: fail to register fixed-link: %d\n", np->name, ret);
++                              return -EFAULT;
++                      }
++              }
++              dp_priv->phy_node = of_node_get(np);
+       }
+ 
++      if(!dp_priv->phy_node)
++              pr_err("%s: no phy-handle or fixed-link found\n", np->name);
++
+       if (of_property_read_u32(np, "qcom,mactype", &hal_pdata->mactype)) {
+               pr_err("%s: error reading mactype\n", np->name);
+               return -EFAULT;
+-- 
+2.40.1
+