[PATCH 21.11] net/nfp: fix Tx descriptor free logic of NFD3

[PATCH 21.11] net/nfp: fix Tx descriptor free logic of NFD3

[Chaoyong He]

[ upstream commit e97738919c2315e07c2e98b6a9cc3912c335364a ]

In the Tx descriptor free logic of nfd3, the former logic might force
cast a negative number into a very big unsigned number, and which will
cause potential problem in the xmit loop.

The xmit loop will continue in the place where it should break, and will
overwrite the Tx descriptor which is not free to use by the PMD.

Fixes: 74a640dac864 ("net/nfp: avoid modulo operations for handling ring wrapping")

Signed-off-by: Chaoyong He <chaoyong.he@corigine.com>
---
 drivers/net/nfp/nfp_rxtx.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/drivers/net/nfp/nfp_rxtx.c b/drivers/net/nfp/nfp_rxtx.c
index 639c1c925b..4fa608d417 100644
--- a/drivers/net/nfp/nfp_rxtx.c
+++ b/drivers/net/nfp/nfp_rxtx.c
@@ -796,10 +796,14 @@ nfp_net_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 static inline
 uint32_t nfp_free_tx_desc(struct nfp_net_txq *txq)
 {
+	uint32_t free_desc;
+
 	if (txq->wr_p >= txq->rd_p)
-		return txq->tx_count - (txq->wr_p - txq->rd_p) - 8;
+		free_desc = txq->tx_count - (txq->wr_p - txq->rd_p);
 	else
-		return txq->rd_p - txq->wr_p - 8;
+		free_desc = txq->rd_p - txq->wr_p - 8;
+
+	return (free_desc > 8) ? (free_desc - 8) : 0;
 }
 
 /*
-- 
2.39.1

Re: [PATCH 21.11] net/nfp: fix Tx descriptor free logic of NFD3

[Kevin Traynor]

On 08/11/2023 02:19, Chaoyong He wrote:
> [ upstream commit e97738919c2315e07c2e98b6a9cc3912c335364a ]
> 
> In the Tx descriptor free logic of nfd3, the former logic might force
> cast a negative number into a very big unsigned number, and which will
> cause potential problem in the xmit loop.
> 
> The xmit loop will continue in the place where it should break, and will
> overwrite the Tx descriptor which is not free to use by the PMD.
> 
> Fixes: 74a640dac864 ("net/nfp: avoid modulo operations for handling ring wrapping")
> 
> Signed-off-by: Chaoyong He<chaoyong.he@corigine.com>
> ---
>   drivers/net/nfp/nfp_rxtx.c | 8 ++++++--
>   1 file changed, 6 insertions(+), 2 deletions(-)

Applied to 21.11 branch. Thanks for backporting.

[PATCH 21.11] net/nfp: fix offloading flows

[Chaoyong He]

[ upstream commit 925c27ec8de92efe69f4cb56e3fc0e413354c739 ]

The symbol '_abi_flower_extra_features' is in IMEM for NFP4000,
but in EMU_CACHE for NFP3800 because which does not have IMEM.

The original logic can't read symbol from EMU_CACHE, so the probe
process will fail when we try to offload flows use NFP3800.

Modify the related data structure and logics to support read symbol
from EMU_CACHE.

Fixes: c7e9729da6b5 ("net/nfp: support CPP")
Cc: stable@dpdk.org

Signed-off-by: Chaoyong He <chaoyong.he@corigine.com>
---
 drivers/net/nfp/nfp_ethdev.c          |   4 +-
 drivers/net/nfp/nfpcore/nfp_cpp.h     |   7 +-
 drivers/net/nfp/nfpcore/nfp_cppcore.c |  52 +++++++++---
 drivers/net/nfp/nfpcore/nfp_rtsym.c   | 117 ++++++++++++++++++++++++--
 4 files changed, 158 insertions(+), 22 deletions(-)

diff --git a/drivers/net/nfp/nfp_ethdev.c b/drivers/net/nfp/nfp_ethdev.c
index 748bacff07..37593fd216 100644
--- a/drivers/net/nfp/nfp_ethdev.c
+++ b/drivers/net/nfp/nfp_ethdev.c
@@ -800,6 +800,7 @@ static int nfp_pf_init(struct rte_pci_device *pci_dev)
 {
 	struct nfp_pf_dev *pf_dev = NULL;
 	struct nfp_cpp *cpp;
+	uint32_t cpp_id;
 	struct nfp_hwinfo *hwinfo;
 	struct nfp_rtsym_table *sym_tbl;
 	struct nfp_eth_table *nfp_eth_table = NULL;
@@ -904,7 +905,8 @@ static int nfp_pf_init(struct rte_pci_device *pci_dev)
 	PMD_INIT_LOG(DEBUG, "ctrl bar: %p", pf_dev->ctrl_bar);
 
 	/* configure access to tx/rx vNIC BARs */
-	pf_dev->hw_queues = nfp_cpp_map_area(pf_dev->cpp, 0, 0,
+	cpp_id = NFP_CPP_ISLAND_ID(0, NFP_CPP_ACTION_RW, 0, 0);
+	pf_dev->hw_queues = nfp_cpp_map_area(pf_dev->cpp, cpp_id,
 					      NFP_PCIE_QUEUE(0),
 					      NFP_QCP_QUEUE_AREA_SZ,
 					      &pf_dev->hwqueues_area);
diff --git a/drivers/net/nfp/nfpcore/nfp_cpp.h b/drivers/net/nfp/nfpcore/nfp_cpp.h
index 720d3989e6..61d36047e1 100644
--- a/drivers/net/nfp/nfpcore/nfp_cpp.h
+++ b/drivers/net/nfp/nfpcore/nfp_cpp.h
@@ -34,6 +34,9 @@ struct nfp_cpp {
 	 */
 	uint32_t imb_cat_table[16];
 
+	/* MU access type bit offset */
+	uint32_t mu_locality_lsb;
+
 	int driver_lock_needed;
 };
 
@@ -363,7 +366,7 @@ struct nfp_cpp_area *nfp_cpp_area_alloc_acquire(struct nfp_cpp *cpp,
  */
 void nfp_cpp_area_release_free(struct nfp_cpp_area *area);
 
-uint8_t *nfp_cpp_map_area(struct nfp_cpp *cpp, int domain, int target,
+uint8_t *nfp_cpp_map_area(struct nfp_cpp *cpp, uint32_t cpp_id,
 			   uint64_t addr, unsigned long size,
 			   struct nfp_cpp_area **area);
 /*
@@ -778,4 +781,6 @@ int nfp_cpp_mutex_unlock(struct nfp_cpp_mutex *mutex);
  */
 int nfp_cpp_mutex_trylock(struct nfp_cpp_mutex *mutex);
 
+uint32_t nfp_cpp_mu_locality_lsb(struct nfp_cpp *cpp);
+
 #endif /* !__NFP_CPP_H__ */
diff --git a/drivers/net/nfp/nfpcore/nfp_cppcore.c b/drivers/net/nfp/nfpcore/nfp_cppcore.c
index 37799af558..13d979633d 100644
--- a/drivers/net/nfp/nfpcore/nfp_cppcore.c
+++ b/drivers/net/nfp/nfpcore/nfp_cppcore.c
@@ -118,6 +118,36 @@ nfp_cpp_area_name(struct nfp_cpp_area *cpp_area)
 	return cpp_area->name;
 }
 
+#define NFP_IMB_TGTADDRESSMODECFG_MODE_of(_x)       (((_x) >> 13) & 0x7)
+#define NFP_IMB_TGTADDRESSMODECFG_ADDRMODE          RTE_BIT32(12)
+
+static int
+nfp_cpp_set_mu_locality_lsb(struct nfp_cpp *cpp)
+{
+	int ret;
+	int mode;
+	int addr40;
+	uint32_t imbcppat;
+
+	imbcppat = cpp->imb_cat_table[NFP_CPP_TARGET_MU];
+	mode = NFP_IMB_TGTADDRESSMODECFG_MODE_of(imbcppat);
+	addr40 = imbcppat & NFP_IMB_TGTADDRESSMODECFG_ADDRMODE;
+
+	ret = nfp_cppat_mu_locality_lsb(mode, addr40);
+	if (ret < 0)
+		return ret;
+
+	cpp->mu_locality_lsb = ret;
+
+	return 0;
+}
+
+uint32_t
+nfp_cpp_mu_locality_lsb(struct nfp_cpp *cpp)
+{
+	return cpp->mu_locality_lsb;
+}
+
 /*
  * nfp_cpp_area_alloc - allocate a new CPP area
  * @cpp:    CPP handle
@@ -142,10 +172,6 @@ nfp_cpp_area_alloc_with_name(struct nfp_cpp *cpp, uint32_t dest,
 	if (!cpp)
 		return NULL;
 
-	/* CPP bus uses only a 40-bit address */
-	if ((address + size) > (1ULL << 40))
-		return NFP_ERRPTR(EFAULT);
-
 	/* Remap from cpp_island to cpp_target */
 	err = nfp_target_cpp(dest, tmp64, &dest, &tmp64, cpp->imb_cat_table);
 	if (err < 0)
@@ -588,6 +614,12 @@ nfp_cpp_alloc(struct rte_pci_device *dev, int driver_lock_needed)
 		}
 	}
 
+	err = nfp_cpp_set_mu_locality_lsb(cpp);
+	if (err < 0) {
+		free(cpp);
+		return NULL;
+	}
+
 	return cpp;
 }
 
@@ -819,8 +851,7 @@ __nfp_cpp_model_autodetect(struct nfp_cpp *cpp, uint32_t *model)
 /*
  * nfp_cpp_map_area() - Helper function to map an area
  * @cpp:    NFP CPP handler
- * @domain: CPP domain
- * @target: CPP target
+ * @cpp_id: CPP ID
  * @addr:   CPP address
  * @size:   Size of the area
  * @area:   Area handle (output)
@@ -831,16 +862,13 @@ __nfp_cpp_model_autodetect(struct nfp_cpp *cpp, uint32_t *model)
  * Return: Pointer to memory mapped area or ERR_PTR
  */
 uint8_t *
-nfp_cpp_map_area(struct nfp_cpp *cpp, int domain, int target, uint64_t addr,
+nfp_cpp_map_area(struct nfp_cpp *cpp, uint32_t cpp_id, uint64_t addr,
 		 unsigned long size, struct nfp_cpp_area **area)
 {
 	uint8_t *res;
-	uint32_t dest;
-
-	dest = NFP_CPP_ISLAND_ID(target, NFP_CPP_ACTION_RW, 0, domain);
 
-	*area = nfp_cpp_area_alloc_acquire(cpp, dest, addr, size);
-	if (!*area)
+	*area = nfp_cpp_area_alloc_acquire(cpp, cpp_id, addr, size);
+	if (*area == NULL)
 		goto err_eio;
 
 	res = nfp_cpp_area_iomem(*area);
diff --git a/drivers/net/nfp/nfpcore/nfp_rtsym.c b/drivers/net/nfp/nfpcore/nfp_rtsym.c
index 2feca2ed81..2c6c639367 100644
--- a/drivers/net/nfp/nfpcore/nfp_rtsym.c
+++ b/drivers/net/nfp/nfpcore/nfp_rtsym.c
@@ -232,6 +232,102 @@ nfp_rtsym_lookup(struct nfp_rtsym_table *rtbl, const char *name)
 	return NULL;
 }
 
+static uint64_t
+nfp_rtsym_size(const struct nfp_rtsym *sym)
+{
+	switch (sym->type) {
+	case NFP_RTSYM_TYPE_NONE:
+		return 0;
+	case NFP_RTSYM_TYPE_OBJECT:    /* Fall through */
+	case NFP_RTSYM_TYPE_FUNCTION:
+		return sym->size;
+	case NFP_RTSYM_TYPE_ABS:
+		return sizeof(uint64_t);
+	default:
+		return 0;
+	}
+}
+
+static int
+nfp_rtsym_to_dest(struct nfp_cpp *cpp,
+		const struct nfp_rtsym *sym,
+		uint8_t action,
+		uint8_t token,
+		uint64_t offset,
+		uint32_t *cpp_id,
+		uint64_t *addr)
+{
+	if (sym->type != NFP_RTSYM_TYPE_OBJECT)
+		return -EINVAL;
+
+	*addr = sym->addr + offset;
+
+	if (sym->target >= 0) {
+		*cpp_id = NFP_CPP_ISLAND_ID(sym->target, action, token, sym->domain);
+	} else if (sym->target == NFP_RTSYM_TARGET_EMU_CACHE) {
+		int locality_off = nfp_cpp_mu_locality_lsb(cpp);
+
+		*addr &= ~(NFP_MU_ADDR_ACCESS_TYPE_MASK << locality_off);
+		*addr |= NFP_MU_ADDR_ACCESS_TYPE_DIRECT << locality_off;
+
+		*cpp_id = NFP_CPP_ISLAND_ID(NFP_CPP_TARGET_MU, action, token,
+				sym->domain);
+	} else {
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int
+nfp_rtsym_readl(struct nfp_cpp *cpp,
+		const struct nfp_rtsym *sym,
+		uint8_t action,
+		uint8_t token,
+		uint64_t offset,
+		uint32_t *value)
+{
+	int ret;
+	uint64_t addr;
+	uint32_t cpp_id;
+
+	if (offset + 4 > nfp_rtsym_size(sym))
+		return -ENXIO;
+
+	ret = nfp_rtsym_to_dest(cpp, sym, action, token, offset, &cpp_id, &addr);
+	if (ret != 0)
+		return ret;
+
+	return nfp_cpp_readl(cpp, cpp_id, addr, value);
+}
+
+static int
+nfp_rtsym_readq(struct nfp_cpp *cpp,
+		const struct nfp_rtsym *sym,
+		uint8_t action,
+		uint8_t token,
+		uint64_t offset,
+		uint64_t *value)
+{
+	int ret;
+	uint64_t addr;
+	uint32_t cpp_id;
+
+	if (offset + 8 > nfp_rtsym_size(sym))
+		return -ENXIO;
+
+	if (sym->type == NFP_RTSYM_TYPE_ABS) {
+		*value = sym->addr;
+		return 0;
+	}
+
+	ret = nfp_rtsym_to_dest(cpp, sym, action, token, offset, &cpp_id, &addr);
+	if (ret != 0)
+		return ret;
+
+	return nfp_cpp_readq(cpp, cpp_id, addr, value);
+}
+
 /*
  * nfp_rtsym_read_le() - Read a simple unsigned scalar value from symbol
  * @rtbl:	NFP RTsym table
@@ -248,7 +344,7 @@ uint64_t
 nfp_rtsym_read_le(struct nfp_rtsym_table *rtbl, const char *name, int *error)
 {
 	const struct nfp_rtsym *sym;
-	uint32_t val32, id;
+	uint32_t val32;
 	uint64_t val;
 	int err;
 
@@ -258,19 +354,17 @@ nfp_rtsym_read_le(struct nfp_rtsym_table *rtbl, const char *name, int *error)
 		goto exit;
 	}
 
-	id = NFP_CPP_ISLAND_ID(sym->target, NFP_CPP_ACTION_RW, 0, sym->domain);
-
 #ifdef DEBUG
 	printf("Reading symbol %s with size %" PRIu64 " at %" PRIx64 "\n",
 		name, sym->size, sym->addr);
 #endif
 	switch (sym->size) {
 	case 4:
-		err = nfp_cpp_readl(rtbl->cpp, id, sym->addr, &val32);
+		err = nfp_rtsym_readl(rtbl->cpp, sym, NFP_CPP_ACTION_RW, 0, 0, &val32);
 		val = val32;
 		break;
 	case 8:
-		err = nfp_cpp_readq(rtbl->cpp, id, sym->addr, &val);
+		err = nfp_rtsym_readq(rtbl->cpp, sym, NFP_CPP_ACTION_RW, 0, 0, &val);
 		break;
 	default:
 		printf("rtsym '%s' unsupported size: %" PRId64 "\n",
@@ -295,8 +389,11 @@ uint8_t *
 nfp_rtsym_map(struct nfp_rtsym_table *rtbl, const char *name,
 	      unsigned int min_size, struct nfp_cpp_area **area)
 {
-	const struct nfp_rtsym *sym;
+	int ret;
 	uint8_t *mem;
+	uint64_t addr;
+	uint32_t cpp_id;
+	const struct nfp_rtsym *sym;
 
 #ifdef DEBUG
 	printf("mapping symbol %s\n", name);
@@ -307,14 +404,18 @@ nfp_rtsym_map(struct nfp_rtsym_table *rtbl, const char *name,
 		return NULL;
 	}
 
+	ret = nfp_rtsym_to_dest(rtbl->cpp, sym, NFP_CPP_ACTION_RW, 0, 0,
+			&cpp_id, &addr);
+	if (ret != 0)
+		return NULL;
+
 	if (sym->size < min_size) {
 		printf("Symbol %s too small (%" PRIu64 " < %u)\n", name,
 			sym->size, min_size);
 		return NULL;
 	}
 
-	mem = nfp_cpp_map_area(rtbl->cpp, sym->domain, sym->target, sym->addr,
-			       sym->size, area);
+	mem = nfp_cpp_map_area(rtbl->cpp, cpp_id, addr, sym->size, area);
 	if (!mem) {
 		printf("Failed to map symbol %s\n", name);
 		return NULL;
-- 
2.39.1

[PATCH 21.11] net/nfp: fix Tx descriptor free logic of NFD3

[Chaoyong He]

[upstream commit e97738919c2315e07c2e98b6a9cc3912c335364a ]

In the Tx descriptor free logic of nfd3, the former logic might force
cast a negative number into a very big unsigned number, and which will
cause potential problem in the xmit loop.

The xmit loop will continue in the place where it should break, and will
overwrite the Tx descriptor which is not free to use by the PMD.

Fixes: 74a640dac864 ("net/nfp: avoid modulo operations for handling ring wrapping")
Cc: stable@dpdk.org

Signed-off-by: Chaoyong He <chaoyong.he@corigine.com>
---
 drivers/net/nfp/nfp_rxtx.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/drivers/net/nfp/nfp_rxtx.c b/drivers/net/nfp/nfp_rxtx.c
index 639c1c925b..3a20e33217 100644
--- a/drivers/net/nfp/nfp_rxtx.c
+++ b/drivers/net/nfp/nfp_rxtx.c
@@ -796,10 +796,14 @@ nfp_net_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 static inline
 uint32_t nfp_free_tx_desc(struct nfp_net_txq *txq)
 {
+	uint32_t free_desc;
+
 	if (txq->wr_p >= txq->rd_p)
-		return txq->tx_count - (txq->wr_p - txq->rd_p) - 8;
+		free_desc = txq->tx_count - (txq->wr_p - txq->rd_p);
 	else
-		return txq->rd_p - txq->wr_p - 8;
+		free_desc = txq->rd_p - txq->wr_p;
+
+	return (free_desc > 8) ? (free_desc - 8) : 0;
 }
 
 /*
-- 
2.39.1