[PATCH v2 intel-net 0/6] ice: xsk: Rx processing fixes

STALE1766d

Revision v2 of 2 in this series.

14 messages, 3 authors, 2021-12-17 · open the first message on its own page

[PATCH v2 intel-net 0/6] ice: xsk: Rx processing fixes

From: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Date: 2021-12-13 15:31:30

Hi there,
it seems that previous [0] Rx fix was not enough and there are still
issues with AF_XDP Rx ZC support in ice driver. Elza reported that for
multiple XSK sockets configured on a single netdev, some of them were
becoming dead after a while. We have spotted more things that needed to
be addressed this time. More of information can be found in particular
commit messages.

v2 has a diff around only patch 2:
- use array_size() in memsets (Alexandr)
- remove unnecessary ternary operator from ice_alloc_rx_buf{, _zc}()
  (Alexandr)
- respect RCT in ice_construct_skb_zc() (Alexandr)
- fix kdoc issue (Anthony)

It also carries Alexandr's patch that was sent previously which was
overlapping with this set.

Thanks,
Maciej

[0]: https://lore.kernel.org/bpf/20211129231746.2767739-1-anthony.l.nguyen@intel.com/

Alexander Lobakin (1):
  ice: remove dead store on XSK hotpath

Maciej Fijalkowski (5):
  ice: xsk: return xsk buffers back to pool when cleaning the ring
  ice: xsk: allocate separate memory for XDP SW ring
  ice: xsk: do not clear status_error0 for ntu + nb_buffs descriptor
  ice: xsk: allow empty Rx descriptors on XSK ZC data path
  ice: xsk: fix cleaned_count setting

 drivers/net/ethernet/intel/ice/ice_base.c | 17 ++++++
 drivers/net/ethernet/intel/ice/ice_txrx.c | 19 ++++---
 drivers/net/ethernet/intel/ice/ice_txrx.h |  1 -
 drivers/net/ethernet/intel/ice/ice_xsk.c  | 66 +++++++++++------------
 4 files changed, 62 insertions(+), 41 deletions(-)

-- 
2.33.1

[PATCH v2 intel-net 2/6] ice: xsk: allocate separate memory for XDP SW ring

From: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Date: 2021-12-13 15:31:31

Currently, the zero-copy data path is reusing the memory region that was
initially allocated for an array of struct ice_rx_buf for its own
purposes. This is error prone as it is based on the ice_rx_buf struct
always being the same size or bigger than what the zero-copy path needs.
There can also be old values present in that array giving rise to errors
when the zero-copy path uses it.

Fix this by freeing the ice_rx_buf region and allocating a new array for
the zero-copy path that has the right length and is initialized to zero.

Fixes: 57f7f8b6bc0b ("ice: Use xdp_buf instead of rx_buf for xsk zero-copy")
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 drivers/net/ethernet/intel/ice/ice_base.c | 17 ++++++++++++
 drivers/net/ethernet/intel/ice/ice_txrx.c | 19 ++++++++-----
 drivers/net/ethernet/intel/ice/ice_xsk.c  | 33 ++++++++++++-----------
 3 files changed, 47 insertions(+), 22 deletions(-)

diff --git a/drivers/net/ethernet/intel/ice/ice_base.c b/drivers/net/ethernet/intel/ice/ice_base.c
index 1efc635cc0f5..fafe020e46ee 100644
--- a/drivers/net/ethernet/intel/ice/ice_base.c
+++ b/drivers/net/ethernet/intel/ice/ice_base.c

@@ -6,6 +6,18 @@
 #include "ice_lib.h"
 #include "ice_dcb_lib.h"
 
+static bool ice_alloc_rx_buf_zc(struct ice_rx_ring *rx_ring)
+{
+	rx_ring->xdp_buf = kcalloc(rx_ring->count, sizeof(*rx_ring->xdp_buf), GFP_KERNEL);
+	return !!rx_ring->xdp_buf;
+}
+
+static bool ice_alloc_rx_buf(struct ice_rx_ring *rx_ring)
+{
+	rx_ring->rx_buf = kcalloc(rx_ring->count, sizeof(*rx_ring->rx_buf), GFP_KERNEL);
+	return !!rx_ring->rx_buf;
+}
+
 /**
  * __ice_vsi_get_qs_contig - Assign a contiguous chunk of queues to VSI
  * @qs_cfg: gathered variables needed for PF->VSI queues assignment

@@ -492,8 +504,11 @@ int ice_vsi_cfg_rxq(struct ice_rx_ring *ring)
 			xdp_rxq_info_reg(&ring->xdp_rxq, ring->netdev,
 					 ring->q_index, ring->q_vector->napi.napi_id);
 
+		kfree(ring->rx_buf);
 		ring->xsk_pool = ice_xsk_pool(ring);
 		if (ring->xsk_pool) {
+			if (!ice_alloc_rx_buf_zc(ring))
+				return -ENOMEM;
 			xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
 
 			ring->rx_buf_len =

@@ -508,6 +523,8 @@ int ice_vsi_cfg_rxq(struct ice_rx_ring *ring)
 			dev_info(dev, "Registered XDP mem model MEM_TYPE_XSK_BUFF_POOL on Rx ring %d\n",
 				 ring->q_index);
 		} else {
+			if (!ice_alloc_rx_buf(ring))
+				return -ENOMEM;
 			if (!xdp_rxq_info_is_reg(&ring->xdp_rxq))
 				/* coverity[check_return] */
 				xdp_rxq_info_reg(&ring->xdp_rxq,

diff --git a/drivers/net/ethernet/intel/ice/ice_txrx.c b/drivers/net/ethernet/intel/ice/ice_txrx.c
index bc3ba19dc88f..dccf09eefc75 100644
--- a/drivers/net/ethernet/intel/ice/ice_txrx.c
+++ b/drivers/net/ethernet/intel/ice/ice_txrx.c

@@ -419,7 +419,10 @@ void ice_clean_rx_ring(struct ice_rx_ring *rx_ring)
 	}
 
 rx_skip_free:
-	memset(rx_ring->rx_buf, 0, sizeof(*rx_ring->rx_buf) * rx_ring->count);
+	if (rx_ring->xsk_pool)
+		memset(rx_ring->xdp_buf, 0, array_size(rx_ring->count, sizeof(*rx_ring->xdp_buf)));
+	else
+		memset(rx_ring->rx_buf, 0, array_size(rx_ring->count, sizeof(*rx_ring->rx_buf)));
 
 	/* Zero out the descriptor ring */
 	size = ALIGN(rx_ring->count * sizeof(union ice_32byte_rx_desc),

@@ -446,8 +449,13 @@ void ice_free_rx_ring(struct ice_rx_ring *rx_ring)
 		if (xdp_rxq_info_is_reg(&rx_ring->xdp_rxq))
 			xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 	rx_ring->xdp_prog = NULL;
-	devm_kfree(rx_ring->dev, rx_ring->rx_buf);
-	rx_ring->rx_buf = NULL;
+	if (rx_ring->xsk_pool) {
+		kfree(rx_ring->xdp_buf);
+		rx_ring->xdp_buf = NULL;
+	} else {
+		kfree(rx_ring->rx_buf);
+		rx_ring->rx_buf = NULL;
+	}
 
 	if (rx_ring->desc) {
 		size = ALIGN(rx_ring->count * sizeof(union ice_32byte_rx_desc),

@@ -475,8 +483,7 @@ int ice_setup_rx_ring(struct ice_rx_ring *rx_ring)
 	/* warn if we are about to overwrite the pointer */
 	WARN_ON(rx_ring->rx_buf);
 	rx_ring->rx_buf =
-		devm_kcalloc(dev, sizeof(*rx_ring->rx_buf), rx_ring->count,
-			     GFP_KERNEL);
+		kcalloc(rx_ring->count, sizeof(*rx_ring->rx_buf), GFP_KERNEL);
 	if (!rx_ring->rx_buf)
 		return -ENOMEM;

@@ -505,7 +512,7 @@ int ice_setup_rx_ring(struct ice_rx_ring *rx_ring)
 	return 0;
 
 err:
-	devm_kfree(dev, rx_ring->rx_buf);
+	kfree(rx_ring->rx_buf);
 	rx_ring->rx_buf = NULL;
 	return -ENOMEM;
 }

diff --git a/drivers/net/ethernet/intel/ice/ice_xsk.c b/drivers/net/ethernet/intel/ice/ice_xsk.c
index 8593717a755e..c124229d98fe 100644
--- a/drivers/net/ethernet/intel/ice/ice_xsk.c
+++ b/drivers/net/ethernet/intel/ice/ice_xsk.c

@@ -12,6 +12,11 @@
 #include "ice_txrx_lib.h"
 #include "ice_lib.h"
 
+static struct xdp_buff **ice_xdp_buf(struct ice_rx_ring *rx_ring, u32 idx)
+{
+	return &rx_ring->xdp_buf[idx];
+}
+
 /**
  * ice_qp_reset_stats - Resets all stats for rings of given index
  * @vsi: VSI that contains rings of interest

@@ -372,7 +377,7 @@ bool ice_alloc_rx_bufs_zc(struct ice_rx_ring *rx_ring, u16 count)
 	dma_addr_t dma;
 
 	rx_desc = ICE_RX_DESC(rx_ring, ntu);
-	xdp = &rx_ring->xdp_buf[ntu];
+	xdp = ice_xdp_buf(rx_ring, ntu);
 
 	nb_buffs = min_t(u16, count, rx_ring->count - ntu);
 	nb_buffs = xsk_buff_alloc_batch(rx_ring->xsk_pool, xdp, nb_buffs);

@@ -419,19 +424,18 @@ static void ice_bump_ntc(struct ice_rx_ring *rx_ring)
 /**
  * ice_construct_skb_zc - Create an sk_buff from zero-copy buffer
  * @rx_ring: Rx ring
- * @xdp_arr: Pointer to the SW ring of xdp_buff pointers
+ * @xdp: Pointer to XDP buffer
  *
  * This function allocates a new skb from a zero-copy Rx buffer.
  *
  * Returns the skb on success, NULL on failure.
  */
 static struct sk_buff *
-ice_construct_skb_zc(struct ice_rx_ring *rx_ring, struct xdp_buff **xdp_arr)
+ice_construct_skb_zc(struct ice_rx_ring *rx_ring, struct xdp_buff *xdp)
 {
-	struct xdp_buff *xdp = *xdp_arr;
+	unsigned int datasize_hard = xdp->data_end - xdp->data_hard_start;
 	unsigned int metasize = xdp->data - xdp->data_meta;
 	unsigned int datasize = xdp->data_end - xdp->data;
-	unsigned int datasize_hard = xdp->data_end - xdp->data_hard_start;
 	struct sk_buff *skb;
 
 	skb = __napi_alloc_skb(&rx_ring->q_vector->napi, datasize_hard,

@@ -445,7 +449,6 @@ ice_construct_skb_zc(struct ice_rx_ring *rx_ring, struct xdp_buff **xdp_arr)
 		skb_metadata_set(skb, metasize);
 
 	xsk_buff_free(xdp);
-	*xdp_arr = NULL;
 	return skb;
 }

@@ -522,7 +525,7 @@ int ice_clean_rx_irq_zc(struct ice_rx_ring *rx_ring, int budget)
 	while (likely(total_rx_packets < (unsigned int)budget)) {
 		union ice_32b_rx_flex_desc *rx_desc;
 		unsigned int size, xdp_res = 0;
-		struct xdp_buff **xdp;
+		struct xdp_buff *xdp;
 		struct sk_buff *skb;
 		u16 stat_err_bits;
 		u16 vlan_tag = 0;

@@ -545,18 +548,17 @@ int ice_clean_rx_irq_zc(struct ice_rx_ring *rx_ring, int budget)
 		if (!size)
 			break;
 
-		xdp = &rx_ring->xdp_buf[rx_ring->next_to_clean];
-		xsk_buff_set_size(*xdp, size);
-		xsk_buff_dma_sync_for_cpu(*xdp, rx_ring->xsk_pool);
+		xdp = *ice_xdp_buf(rx_ring, rx_ring->next_to_clean);
+		xsk_buff_set_size(xdp, size);
+		xsk_buff_dma_sync_for_cpu(xdp, rx_ring->xsk_pool);
 
-		xdp_res = ice_run_xdp_zc(rx_ring, *xdp, xdp_prog, xdp_ring);
+		xdp_res = ice_run_xdp_zc(rx_ring, xdp, xdp_prog, xdp_ring);
 		if (xdp_res) {
 			if (xdp_res & (ICE_XDP_TX | ICE_XDP_REDIR))
 				xdp_xmit |= xdp_res;
 			else
-				xsk_buff_free(*xdp);
+				xsk_buff_free(xdp);
 
-			*xdp = NULL;
 			total_rx_bytes += size;
 			total_rx_packets++;
 			cleaned_count++;

@@ -816,10 +818,9 @@ void ice_xsk_clean_rx_ring(struct ice_rx_ring *rx_ring)
 	u16 ntu = rx_ring->next_to_use;
 
 	for ( ; ntc != ntu; ntc = (ntc + 1) & count_mask) {
-		struct xdp_buff **xdp = &rx_ring->xdp_buf[ntc];
+		struct xdp_buff *xdp = *ice_xdp_buf(rx_ring, ntc);
 
-		xsk_buff_free(*xdp);
-		*xdp = NULL;
+		xsk_buff_free(xdp);
 	}
 }

-- 
2.33.1

[PATCH v2 intel-net 1/6] ice: xsk: return xsk buffers back to pool when cleaning the ring

From: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Date: 2021-12-13 15:31:31

Currently we only NULL the xdp_buff pointer in the internal SW ring but
we never give it back to the xsk buffer pool. This means that buffers
can be leaked out of the buff pool and never be used again.

Add missing xsk_buff_free() call to the routine that is supposed to
clean the entries that are left in the ring so that these buffers in the
umem can be used by other sockets.

Also, only go through the space that is actually left to be cleaned
instead of a whole ring.

Fixes: 2d4238f55697 ("ice: Add support for AF_XDP")
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 drivers/net/ethernet/intel/ice/ice_xsk.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/intel/ice/ice_xsk.c b/drivers/net/ethernet/intel/ice/ice_xsk.c
index bb9a80847298..8593717a755e 100644
--- a/drivers/net/ethernet/intel/ice/ice_xsk.c
+++ b/drivers/net/ethernet/intel/ice/ice_xsk.c

@@ -811,14 +811,14 @@ bool ice_xsk_any_rx_ring_ena(struct ice_vsi *vsi)
  */
 void ice_xsk_clean_rx_ring(struct ice_rx_ring *rx_ring)
 {
-	u16 i;
-
-	for (i = 0; i < rx_ring->count; i++) {
-		struct xdp_buff **xdp = &rx_ring->xdp_buf[i];
+	u16 count_mask = rx_ring->count - 1;
+	u16 ntc = rx_ring->next_to_clean;
+	u16 ntu = rx_ring->next_to_use;
 
-		if (!xdp)
-			continue;
+	for ( ; ntc != ntu; ntc = (ntc + 1) & count_mask) {
+		struct xdp_buff **xdp = &rx_ring->xdp_buf[ntc];
 
+		xsk_buff_free(*xdp);
 		*xdp = NULL;
 	}
 }

-- 
2.33.1

[PATCH v2 intel-net 3/6] ice: remove dead store on XSK hotpath

From: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Date: 2021-12-13 15:31:35

From: Alexander Lobakin <redacted>

The 'if (ntu == rx_ring->count)' block in ice_alloc_rx_buffers_zc()
was previously residing in the loop, but after introducing the
batched interface it is used only to wrap-around the NTU descriptor,
thus no more need to assign 'xdp'.

Fixes: db804cfc21e9 ("ice: Use the xsk batched rx allocation interface")
Signed-off-by: Alexander Lobakin <redacted>
Acked-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 drivers/net/ethernet/intel/ice/ice_xsk.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/drivers/net/ethernet/intel/ice/ice_xsk.c b/drivers/net/ethernet/intel/ice/ice_xsk.c
index c124229d98fe..27f5f64dcbd6 100644
--- a/drivers/net/ethernet/intel/ice/ice_xsk.c
+++ b/drivers/net/ethernet/intel/ice/ice_xsk.c

@@ -397,7 +397,6 @@ bool ice_alloc_rx_bufs_zc(struct ice_rx_ring *rx_ring, u16 count)
 	ntu += nb_buffs;
 	if (ntu == rx_ring->count) {
 		rx_desc = ICE_RX_DESC(rx_ring, 0);
-		xdp = rx_ring->xdp_buf;
 		ntu = 0;
 	}

-- 
2.33.1

[PATCH v2 intel-net 4/6] ice: xsk: do not clear status_error0 for ntu + nb_buffs descriptor

From: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Date: 2021-12-13 15:31:38

The descriptor that ntu is pointing at when we exit
ice_alloc_rx_bufs_zc() should not have its corresponding DD bit cleared
as descriptor is not allocated in there and it is not valid for HW
usage.

The allocation routine at the entry will fill the descriptor that ntu
points to after it was set to ntu + nb_buffs on previous call.

Even the spec says:
"The tail pointer should be set to one descriptor beyond the last empty
descriptor in host descriptor ring."

Therefore, step away from clearing the status_error0 on ntu + nb_buffs
descriptor.

Fixes: db804cfc21e9 ("ice: Use the xsk batched rx allocation interface")
Reported-by: Elza Mathew <redacted>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 drivers/net/ethernet/intel/ice/ice_xsk.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

diff --git a/drivers/net/ethernet/intel/ice/ice_xsk.c b/drivers/net/ethernet/intel/ice/ice_xsk.c
index 27f5f64dcbd6..ffa9a160766a 100644
--- a/drivers/net/ethernet/intel/ice/ice_xsk.c
+++ b/drivers/net/ethernet/intel/ice/ice_xsk.c

@@ -395,13 +395,9 @@ bool ice_alloc_rx_bufs_zc(struct ice_rx_ring *rx_ring, u16 count)
 	}
 
 	ntu += nb_buffs;
-	if (ntu == rx_ring->count) {
-		rx_desc = ICE_RX_DESC(rx_ring, 0);
+	if (ntu == rx_ring->count)
 		ntu = 0;
-	}
 
-	/* clear the status bits for the next_to_use descriptor */
-	rx_desc->wb.status_error0 = 0;
 	ice_release_rx_desc(rx_ring, ntu);
 
 	return count == nb_buffs;

-- 
2.33.1

[PATCH v2 intel-net 5/6] ice: xsk: allow empty Rx descriptors on XSK ZC data path

From: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Date: 2021-12-13 15:31:41

Commit ac6f733a7bd5 ("ice: allow empty Rx descriptors") stated that ice
HW can produce empty descriptors that are valid and they should be
processed.

Add this support to xsk ZC path to avoid potential processing problems.

Fixes: 2d4238f55697 ("ice: Add support for AF_XDP")
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 drivers/net/ethernet/intel/ice/ice_xsk.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/intel/ice/ice_xsk.c b/drivers/net/ethernet/intel/ice/ice_xsk.c
index ffa9a160766a..c1491dc0675d 100644
--- a/drivers/net/ethernet/intel/ice/ice_xsk.c
+++ b/drivers/net/ethernet/intel/ice/ice_xsk.c

@@ -538,12 +538,18 @@ int ice_clean_rx_irq_zc(struct ice_rx_ring *rx_ring, int budget)
 		 */
 		dma_rmb();
 
+		xdp = *ice_xdp_buf(rx_ring, rx_ring->next_to_clean);
+
 		size = le16_to_cpu(rx_desc->wb.pkt_len) &
 				   ICE_RX_FLX_DESC_PKT_LEN_M;
-		if (!size)
-			break;
+		if (!size) {
+			xdp->data = NULL;
+			xdp->data_end = NULL;
+			xdp->data_hard_start = NULL;
+			xdp->data_meta = NULL;
+			goto construct_skb;
+		}
 
-		xdp = *ice_xdp_buf(rx_ring, rx_ring->next_to_clean);
 		xsk_buff_set_size(xdp, size);
 		xsk_buff_dma_sync_for_cpu(xdp, rx_ring->xsk_pool);

@@ -561,7 +567,7 @@ int ice_clean_rx_irq_zc(struct ice_rx_ring *rx_ring, int budget)
 			ice_bump_ntc(rx_ring);
 			continue;
 		}
-
+construct_skb:
 		/* XDP_PASS path */
 		skb = ice_construct_skb_zc(rx_ring, xdp);
 		if (!skb) {

-- 
2.33.1

[PATCH v2 intel-net 6/6] ice: xsk: fix cleaned_count setting

From: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Date: 2021-12-13 15:31:41

Currently cleaned_count is initialized to ICE_DESC_UNUSED(rx_ring) and
later on during the Rx processing it is incremented per each frame that
driver consumed. This can result in excessive buffers requested from xsk
pool based on that value.

To address this, just drop cleaned_count and pass
ICE_DESC_UNUSED(rx_ring) directly as a function argument to
ice_alloc_rx_bufs_zc(). Idea is to ask for buffers as many as consumed.

Let us also call ice_alloc_rx_bufs_zc unconditionally at the end of
ice_clean_rx_irq_zc. This has been changed in that way for corresponding
ice_clean_rx_irq, but not here.

Fixes: 2d4238f55697 ("ice: Add support for AF_XDP")
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 drivers/net/ethernet/intel/ice/ice_txrx.h | 1 -
 drivers/net/ethernet/intel/ice/ice_xsk.c  | 6 +-----
 2 files changed, 1 insertion(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/intel/ice/ice_txrx.h b/drivers/net/ethernet/intel/ice/ice_txrx.h
index c56dd1749903..b7b3bd4816f0 100644
--- a/drivers/net/ethernet/intel/ice/ice_txrx.h
+++ b/drivers/net/ethernet/intel/ice/ice_txrx.h

@@ -24,7 +24,6 @@
 #define ICE_MAX_DATA_PER_TXD_ALIGNED \
 	(~(ICE_MAX_READ_REQ_SIZE - 1) & ICE_MAX_DATA_PER_TXD)
 
-#define ICE_RX_BUF_WRITE	16	/* Must be power of 2 */
 #define ICE_MAX_TXQ_PER_TXQG	128
 
 /* Attempt to maximize the headroom available for incoming frames. We use a 2K

diff --git a/drivers/net/ethernet/intel/ice/ice_xsk.c b/drivers/net/ethernet/intel/ice/ice_xsk.c
index c1491dc0675d..c895351b25e0 100644
--- a/drivers/net/ethernet/intel/ice/ice_xsk.c
+++ b/drivers/net/ethernet/intel/ice/ice_xsk.c

@@ -505,7 +505,6 @@ ice_run_xdp_zc(struct ice_rx_ring *rx_ring, struct xdp_buff *xdp,
 int ice_clean_rx_irq_zc(struct ice_rx_ring *rx_ring, int budget)
 {
 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
-	u16 cleaned_count = ICE_DESC_UNUSED(rx_ring);
 	struct ice_tx_ring *xdp_ring;
 	unsigned int xdp_xmit = 0;
 	struct bpf_prog *xdp_prog;

@@ -562,7 +561,6 @@ int ice_clean_rx_irq_zc(struct ice_rx_ring *rx_ring, int budget)
 
 			total_rx_bytes += size;
 			total_rx_packets++;
-			cleaned_count++;
 
 			ice_bump_ntc(rx_ring);
 			continue;

@@ -575,7 +573,6 @@ int ice_clean_rx_irq_zc(struct ice_rx_ring *rx_ring, int budget)
 			break;
 		}
 
-		cleaned_count++;
 		ice_bump_ntc(rx_ring);
 
 		if (eth_skb_pad(skb)) {

@@ -597,8 +594,7 @@ int ice_clean_rx_irq_zc(struct ice_rx_ring *rx_ring, int budget)
 		ice_receive_skb(rx_ring, skb, vlan_tag);
 	}
 
-	if (cleaned_count >= ICE_RX_BUF_WRITE)
-		failure = !ice_alloc_rx_bufs_zc(rx_ring, cleaned_count);
+	failure = !ice_alloc_rx_bufs_zc(rx_ring, ICE_DESC_UNUSED(rx_ring));
 
 	ice_finalize_xdp_rx(xdp_ring, xdp_xmit);
 	ice_update_rx_ring_stats(rx_ring, total_rx_packets, total_rx_bytes);

-- 
2.33.1

Re: [Intel-wired-lan] [PATCH v2 intel-net 0/6] ice: xsk: Rx processing fixes

From: Magnus Karlsson <hidden>
Date: 2021-12-14 14:46:32

On Mon, Dec 13, 2021 at 4:31 PM Maciej Fijalkowski
[off-list ref] wrote:

Hi there,
it seems that previous [0] Rx fix was not enough and there are still
issues with AF_XDP Rx ZC support in ice driver. Elza reported that for
multiple XSK sockets configured on a single netdev, some of them were
becoming dead after a while. We have spotted more things that needed to
be addressed this time. More of information can be found in particular
commit messages.

v2 has a diff around only patch 2:
- use array_size() in memsets (Alexandr)
- remove unnecessary ternary operator from ice_alloc_rx_buf{, _zc}()
  (Alexandr)
- respect RCT in ice_construct_skb_zc() (Alexandr)
- fix kdoc issue (Anthony)

It also carries Alexandr's patch that was sent previously which was
overlapping with this set.

Thanks,
Maciej

[0]: https://lore.kernel.org/bpf/20211129231746.2767739-1-anthony.l.nguyen@intel.com/

Thank you so much for all these fixes Maciej and Alexandr!

BTW, ice zero-copy support in bpf and bpf-next does not work at all
without this patch set, so need to get this in as soon as possible.
net and net-next might not work either, but have not tried.

Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>

Alexander Lobakin (1):
  ice: remove dead store on XSK hotpath

Maciej Fijalkowski (5):
  ice: xsk: return xsk buffers back to pool when cleaning the ring
  ice: xsk: allocate separate memory for XDP SW ring
  ice: xsk: do not clear status_error0 for ntu + nb_buffs descriptor
  ice: xsk: allow empty Rx descriptors on XSK ZC data path
  ice: xsk: fix cleaned_count setting

 drivers/net/ethernet/intel/ice/ice_base.c | 17 ++++++
 drivers/net/ethernet/intel/ice/ice_txrx.c | 19 ++++---
 drivers/net/ethernet/intel/ice/ice_txrx.h |  1 -
 drivers/net/ethernet/intel/ice/ice_xsk.c  | 66 +++++++++++------------
 4 files changed, 62 insertions(+), 41 deletions(-)

--
2.33.1

_______________________________________________
Intel-wired-lan mailing list
Intel-wired-lan@osuosl.org
https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

RE: [Intel-wired-lan] [PATCH v2 intel-net 6/6] ice: xsk: fix cleaned_count setting

From: Bhandare, KiranX <hidden>
Date: 2021-12-17 15:23:52

-----Original Message-----
From: Intel-wired-lan <redacted> On Behalf Of
Maciej Fijalkowski
Sent: Monday, December 13, 2021 9:01 PM
To: intel-wired-lan@lists.osuosl.org
Cc: Mathew, Elza <redacted>; netdev@vger.kernel.org;
kuba@kernel.org; bpf@vger.kernel.org; davem@davemloft.net; Karlsson,
Magnus [off-list ref]
Subject: [Intel-wired-lan] [PATCH v2 intel-net 6/6] ice: xsk: fix cleaned_count
setting

Currently cleaned_count is initialized to ICE_DESC_UNUSED(rx_ring) and later
on during the Rx processing it is incremented per each frame that driver
consumed. This can result in excessive buffers requested from xsk pool
based on that value.

To address this, just drop cleaned_count and pass
ICE_DESC_UNUSED(rx_ring) directly as a function argument to
ice_alloc_rx_bufs_zc(). Idea is to ask for buffers as many as consumed.

Let us also call ice_alloc_rx_bufs_zc unconditionally at the end of
ice_clean_rx_irq_zc. This has been changed in that way for corresponding
ice_clean_rx_irq, but not here.

Fixes: 2d4238f55697 ("ice: Add support for AF_XDP")
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 drivers/net/ethernet/intel/ice/ice_txrx.h | 1 -
drivers/net/ethernet/intel/ice/ice_xsk.c  | 6 +-----
 2 files changed, 1 insertion(+), 6 deletions(-)

Tested-by: Kiran Bhandare <redacted>  A Contingent Worker at Intel

RE: [Intel-wired-lan] [PATCH v2 intel-net 5/6] ice: xsk: allow empty Rx descriptors on XSK ZC data path

From: Bhandare, KiranX <hidden>
Date: 2021-12-17 15:24:01

-----Original Message-----
From: Intel-wired-lan <redacted> On Behalf Of
Maciej Fijalkowski
Sent: Monday, December 13, 2021 9:01 PM
To: intel-wired-lan@lists.osuosl.org
Cc: Mathew, Elza <redacted>; netdev@vger.kernel.org;
kuba@kernel.org; bpf@vger.kernel.org; davem@davemloft.net; Karlsson,
Magnus [off-list ref]
Subject: [Intel-wired-lan] [PATCH v2 intel-net 5/6] ice: xsk: allow empty Rx
descriptors on XSK ZC data path

Commit ac6f733a7bd5 ("ice: allow empty Rx descriptors") stated that ice HW
can produce empty descriptors that are valid and they should be processed.

Add this support to xsk ZC path to avoid potential processing problems.

Fixes: 2d4238f55697 ("ice: Add support for AF_XDP")
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 drivers/net/ethernet/intel/ice/ice_xsk.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)

Tested-by: Kiran Bhandare <redacted>  A Contingent Worker at Intel

RE: [Intel-wired-lan] [PATCH v2 intel-net 4/6] ice: xsk: do not clear status_error0 for ntu + nb_buffs descriptor

From: Bhandare, KiranX <hidden>
Date: 2021-12-17 15:24:04

-----Original Message-----
From: Intel-wired-lan <redacted> On Behalf Of
Maciej Fijalkowski
Sent: Monday, December 13, 2021 9:01 PM
To: intel-wired-lan@lists.osuosl.org
Cc: Mathew, Elza <redacted>; netdev@vger.kernel.org;
kuba@kernel.org; bpf@vger.kernel.org; davem@davemloft.net; Karlsson,
Magnus [off-list ref]
Subject: [Intel-wired-lan] [PATCH v2 intel-net 4/6] ice: xsk: do not clear
status_error0 for ntu + nb_buffs descriptor

The descriptor that ntu is pointing at when we exit
ice_alloc_rx_bufs_zc() should not have its corresponding DD bit cleared as
descriptor is not allocated in there and it is not valid for HW usage.

The allocation routine at the entry will fill the descriptor that ntu points to
after it was set to ntu + nb_buffs on previous call.

Even the spec says:
"The tail pointer should be set to one descriptor beyond the last empty
descriptor in host descriptor ring."

Therefore, step away from clearing the status_error0 on ntu + nb_buffs
descriptor.

Fixes: db804cfc21e9 ("ice: Use the xsk batched rx allocation interface")
Reported-by: Elza Mathew <redacted>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 drivers/net/ethernet/intel/ice/ice_xsk.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

Tested-by: Kiran Bhandare <redacted>  A Contingent Worker at Intel

RE: [Intel-wired-lan] [PATCH v2 intel-net 3/6] ice: remove dead store on XSK hotpath

From: Bhandare, KiranX <hidden>
Date: 2021-12-17 15:24:08

-----Original Message-----
From: Intel-wired-lan <redacted> On Behalf Of
Maciej Fijalkowski
Sent: Monday, December 13, 2021 9:01 PM
To: intel-wired-lan@lists.osuosl.org
Cc: Mathew, Elza <redacted>; netdev@vger.kernel.org;
kuba@kernel.org; bpf@vger.kernel.org; davem@davemloft.net; Karlsson,
Magnus [off-list ref]
Subject: [Intel-wired-lan] [PATCH v2 intel-net 3/6] ice: remove dead store on
XSK hotpath

From: Alexander Lobakin <redacted>

The 'if (ntu == rx_ring->count)' block in ice_alloc_rx_buffers_zc() was
previously residing in the loop, but after introducing the batched interface it
is used only to wrap-around the NTU descriptor, thus no more need to assign
'xdp'.

Fixes: db804cfc21e9 ("ice: Use the xsk batched rx allocation interface")
Signed-off-by: Alexander Lobakin <redacted>
Acked-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 drivers/net/ethernet/intel/ice/ice_xsk.c | 1 -
 1 file changed, 1 deletion(-)

Tested-by: Kiran Bhandare <redacted>  A Contingent Worker at Intel

RE: [Intel-wired-lan] [PATCH v2 intel-net 2/6] ice: xsk: allocate separate memory for XDP SW ring

From: Bhandare, KiranX <hidden>
Date: 2021-12-17 15:24:09

-----Original Message-----
From: Intel-wired-lan <redacted> On Behalf Of
Maciej Fijalkowski
Sent: Monday, December 13, 2021 9:01 PM
To: intel-wired-lan@lists.osuosl.org
Cc: Mathew, Elza <redacted>; netdev@vger.kernel.org;
kuba@kernel.org; bpf@vger.kernel.org; davem@davemloft.net; Karlsson,
Magnus [off-list ref]
Subject: [Intel-wired-lan] [PATCH v2 intel-net 2/6] ice: xsk: allocate separate
memory for XDP SW ring

Currently, the zero-copy data path is reusing the memory region that was
initially allocated for an array of struct ice_rx_buf for its own purposes. This
is error prone as it is based on the ice_rx_buf struct always being the same
size or bigger than what the zero-copy path needs.
There can also be old values present in that array giving rise to errors when
the zero-copy path uses it.

Fix this by freeing the ice_rx_buf region and allocating a new array for the
zero-copy path that has the right length and is initialized to zero.

Fixes: 57f7f8b6bc0b ("ice: Use xdp_buf instead of rx_buf for xsk zero-copy")
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 drivers/net/ethernet/intel/ice/ice_base.c | 17 ++++++++++++
drivers/net/ethernet/intel/ice/ice_txrx.c | 19 ++++++++-----
drivers/net/ethernet/intel/ice/ice_xsk.c  | 33 ++++++++++++-----------
 3 files changed, 47 insertions(+), 22 deletions(-)

Tested-by: Kiran Bhandare <redacted>  A Contingent Worker at Intel

RE: [Intel-wired-lan] [PATCH v2 intel-net 1/6] ice: xsk: return xsk buffers back to pool when cleaning the ring

From: Bhandare, KiranX <hidden>
Date: 2021-12-17 15:24:14

-----Original Message-----
From: Intel-wired-lan <redacted> On Behalf Of
Maciej Fijalkowski
Sent: Monday, December 13, 2021 9:01 PM
To: intel-wired-lan@lists.osuosl.org
Cc: Mathew, Elza <redacted>; netdev@vger.kernel.org;
kuba@kernel.org; bpf@vger.kernel.org; davem@davemloft.net; Karlsson,
Magnus [off-list ref]
Subject: [Intel-wired-lan] [PATCH v2 intel-net 1/6] ice: xsk: return xsk buffers
back to pool when cleaning the ring

Currently we only NULL the xdp_buff pointer in the internal SW ring but we
never give it back to the xsk buffer pool. This means that buffers can be
leaked out of the buff pool and never be used again.

Add missing xsk_buff_free() call to the routine that is supposed to clean the
entries that are left in the ring so that these buffers in the umem can be used
by other sockets.

Also, only go through the space that is actually left to be cleaned instead of a
whole ring.

Fixes: 2d4238f55697 ("ice: Add support for AF_XDP")
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 drivers/net/ethernet/intel/ice/ice_xsk.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

Tested-by: Kiran Bhandare <redacted>  A Contingent Worker at Intel

`h`	back out one level
`j`	next message in thread
`k`	previous message in thread
`l`	drill in
`Esc`	close help / fold thread tree
`?`	toggle this help