diff mbox series

[net-next,v3] net: mana: Implement get_ringparam/set_ringparam for mana

Message ID 1723805303-11432-1-git-send-email-shradhagupta@linux.microsoft.com (mailing list archive)
State Superseded
Delegated to: Netdev Maintainers
Headers show
Series [net-next,v3] net: mana: Implement get_ringparam/set_ringparam for mana | expand

Checks

Context Check Description
netdev/series_format success Single patches do not need cover letters
netdev/tree_selection success Clearly marked for net-next
netdev/ynl success Generated files up to date; no warnings/errors; no diff in generated;
netdev/fixes_present success Fixes tag not required for -next series
netdev/header_inline success No static functions without inline keyword in header files
netdev/build_32bit success Errors and warnings before: 16 this patch: 16
netdev/build_tools success Errors and warnings before: 0 this patch: 0
netdev/cc_maintainers success CCed 14 of 14 maintainers
netdev/build_clang success Errors and warnings before: 16 this patch: 16
netdev/verify_signedoff success Signed-off-by tag matches author and committer
netdev/deprecated_api success None detected
netdev/check_selftest success No net selftest shell script
netdev/verify_fixes success No Fixes tag
netdev/build_allmodconfig_warn success Errors and warnings before: 16 this patch: 16
netdev/checkpatch warning WARNING: line length of 100 exceeds 80 columns WARNING: line length of 83 exceeds 80 columns
netdev/build_clang_rust success No Rust files in patch. Skipping build
netdev/kdoc success Errors and warnings before: 0 this patch: 0
netdev/source_inline success Was 0 now: 0
netdev/contest success net-next-2024-08-16--21-00 (tests: 710)

Commit Message

Shradha Gupta Aug. 16, 2024, 10:48 a.m. UTC 
Currently the values of WQs for RX and TX queues for MANA devices
are hardcoded to default sizes.
Allow configuring these values for MANA devices as ringparam
configuration(get/set) through ethtool_ops.
Pre-allocate buffers at the beginning of set operation, to
prevent complete network loss in low-memory conditions.

Signed-off-by: Shradha Gupta <shradhagupta@linux.microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: Long Li <longli@microsoft.com>
---
 Changes in v3:
 * pre-allocate buffers before changing the queue sizes
 * rebased to latest net-next
 * Improved the commit message
---
 Changes in v2:
 * Removed unnecessary validations in mana_set_ringparam()
 * Fixed codespell error
 * Improved error message to indicate issue with the parameter
---
 drivers/net/ethernet/microsoft/mana/mana_en.c | 24 ++++---
 .../ethernet/microsoft/mana/mana_ethtool.c    | 71 +++++++++++++++++++
 include/net/mana/mana.h                       | 23 +++++-
 3 files changed, 105 insertions(+), 13 deletions(-)

Comments

Jakub Kicinski Aug. 17, 2024, 1:58 a.m. UTC | #1 
On Fri, 16 Aug 2024 03:48:23 -0700 Shradha Gupta wrote:
> +	old_tx = apc->tx_queue_size;
> +	old_rx = apc->rx_queue_size;
> +	new_tx = clamp_t(u32, ring->tx_pending, MIN_TX_BUFFERS_PER_QUEUE, MAX_TX_BUFFERS_PER_QUEUE);
> +	new_rx = clamp_t(u32, ring->rx_pending, MIN_RX_BUFFERS_PER_QUEUE, MAX_RX_BUFFERS_PER_QUEUE);

You can min(), the max side of clam is unnecessary. Core code won't let
user requests above max provided by "get" thru.

> +	if (!is_power_of_2(new_tx)) {
> +		netdev_err(ndev, "%s:Tx:%d not supported. Needs to be a power of 2\n",
> +			   __func__, new_tx);
> +		return -EINVAL;
> +	}

The power of 2 vs clamp is a bit odd.
On one hand you clamp the values to what's supported automatically.
On the other you hard reject values which are not power of 2.
Why not round them up?

IDK whether checking or auto-correction is better, but mixing the two
is odd.

> +	if (!is_power_of_2(new_rx)) {
> +		netdev_err(ndev, "%s:Rx:%d not supported. Needs to be a power of 2\n",
> +			   __func__, new_rx);

Instead of printing please use the extack passed in as an argument.
Shradha Gupta Aug. 19, 2024, 3:41 p.m. UTC | #2 
On Fri, Aug 16, 2024 at 06:58:05PM -0700, Jakub Kicinski wrote:
> On Fri, 16 Aug 2024 03:48:23 -0700 Shradha Gupta wrote:
> > +	old_tx = apc->tx_queue_size;
> > +	old_rx = apc->rx_queue_size;
> > +	new_tx = clamp_t(u32, ring->tx_pending, MIN_TX_BUFFERS_PER_QUEUE, MAX_TX_BUFFERS_PER_QUEUE);
> > +	new_rx = clamp_t(u32, ring->rx_pending, MIN_RX_BUFFERS_PER_QUEUE, MAX_RX_BUFFERS_PER_QUEUE);
> 
> You can min(), the max side of clam is unnecessary. Core code won't let
> user requests above max provided by "get" thru.
>
oh okay, got it. Will change this 
> > +	if (!is_power_of_2(new_tx)) {
> > +		netdev_err(ndev, "%s:Tx:%d not supported. Needs to be a power of 2\n",
> > +			   __func__, new_tx);
> > +		return -EINVAL;
> > +	}
> 
> The power of 2 vs clamp is a bit odd.
> On one hand you clamp the values to what's supported automatically.
> On the other you hard reject values which are not power of 2.
> Why not round them up?
> 
> IDK whether checking or auto-correction is better, but mixing the two
> is odd.
> 
That seems right. I will round up the value to nearest power of two, in
the range. Thanks
> > +	if (!is_power_of_2(new_rx)) {
> > +		netdev_err(ndev, "%s:Rx:%d not supported. Needs to be a power of 2\n",
> > +			   __func__, new_rx);
> 
> Instead of printing please use the extack passed in as an argument.
sure, working on it.
> -- 
> pw-bot: cr
diff mbox series

Patch

diff --git a/drivers/net/ethernet/microsoft/mana/mana_en.c b/drivers/net/ethernet/microsoft/mana/mana_en.c
index d2f07e179e86..4e3ade5926bc 100644
--- a/drivers/net/ethernet/microsoft/mana/mana_en.c
+++ b/drivers/net/ethernet/microsoft/mana/mana_en.c
@@ -511,7 +511,7 @@  static u16 mana_select_queue(struct net_device *ndev, struct sk_buff *skb,
 }
 
 /* Release pre-allocated RX buffers */
-static void mana_pre_dealloc_rxbufs(struct mana_port_context *mpc)
+void mana_pre_dealloc_rxbufs(struct mana_port_context *mpc)
 {
 	struct device *dev;
 	int i;
@@ -604,7 +604,7 @@  static void mana_get_rxbuf_cfg(int mtu, u32 *datasize, u32 *alloc_size,
 	*datasize = mtu + ETH_HLEN;
 }
 
-static int mana_pre_alloc_rxbufs(struct mana_port_context *mpc, int new_mtu)
+int mana_pre_alloc_rxbufs(struct mana_port_context *mpc, int new_mtu)
 {
 	struct device *dev;
 	struct page *page;
@@ -618,7 +618,7 @@  static int mana_pre_alloc_rxbufs(struct mana_port_context *mpc, int new_mtu)
 
 	dev = mpc->ac->gdma_dev->gdma_context->dev;
 
-	num_rxb = mpc->num_queues * RX_BUFFERS_PER_QUEUE;
+	num_rxb = mpc->num_queues * mpc->rx_queue_size;
 
 	WARN(mpc->rxbufs_pre, "mana rxbufs_pre exists\n");
 	mpc->rxbufs_pre = kmalloc_array(num_rxb, sizeof(void *), GFP_KERNEL);
@@ -1899,14 +1899,15 @@  static int mana_create_txq(struct mana_port_context *apc,
 		return -ENOMEM;
 
 	/*  The minimum size of the WQE is 32 bytes, hence
-	 *  MAX_SEND_BUFFERS_PER_QUEUE represents the maximum number of WQEs
+	 *  apc->tx_queue_size represents the maximum number of WQEs
 	 *  the SQ can store. This value is then used to size other queues
 	 *  to prevent overflow.
+	 *  Also note that the txq_size is always going to be MANA_PAGE_ALIGNED,
+	 *  as tx_queue_size is always a power of 2.
 	 */
-	txq_size = MAX_SEND_BUFFERS_PER_QUEUE * 32;
-	BUILD_BUG_ON(!MANA_PAGE_ALIGNED(txq_size));
+	txq_size = apc->tx_queue_size * 32;
 
-	cq_size = MAX_SEND_BUFFERS_PER_QUEUE * COMP_ENTRY_SIZE;
+	cq_size = apc->tx_queue_size * COMP_ENTRY_SIZE;
 	cq_size = MANA_PAGE_ALIGN(cq_size);
 
 	gc = gd->gdma_context;
@@ -2145,10 +2146,11 @@  static int mana_push_wqe(struct mana_rxq *rxq)
 
 static int mana_create_page_pool(struct mana_rxq *rxq, struct gdma_context *gc)
 {
+	struct mana_port_context *mpc = netdev_priv(rxq->ndev);
 	struct page_pool_params pprm = {};
 	int ret;
 
-	pprm.pool_size = RX_BUFFERS_PER_QUEUE;
+	pprm.pool_size = mpc->rx_queue_size;
 	pprm.nid = gc->numa_node;
 	pprm.napi = &rxq->rx_cq.napi;
 	pprm.netdev = rxq->ndev;
@@ -2180,13 +2182,13 @@  static struct mana_rxq *mana_create_rxq(struct mana_port_context *apc,
 
 	gc = gd->gdma_context;
 
-	rxq = kzalloc(struct_size(rxq, rx_oobs, RX_BUFFERS_PER_QUEUE),
+	rxq = kzalloc(struct_size(rxq, rx_oobs, apc->rx_queue_size),
 		      GFP_KERNEL);
 	if (!rxq)
 		return NULL;
 
 	rxq->ndev = ndev;
-	rxq->num_rx_buf = RX_BUFFERS_PER_QUEUE;
+	rxq->num_rx_buf = apc->rx_queue_size;
 	rxq->rxq_idx = rxq_idx;
 	rxq->rxobj = INVALID_MANA_HANDLE;
 
@@ -2734,6 +2736,8 @@  static int mana_probe_port(struct mana_context *ac, int port_idx,
 	apc->ndev = ndev;
 	apc->max_queues = gc->max_num_queues;
 	apc->num_queues = gc->max_num_queues;
+	apc->tx_queue_size = DEF_TX_BUFFERS_PER_QUEUE;
+	apc->rx_queue_size = DEF_RX_BUFFERS_PER_QUEUE;
 	apc->port_handle = INVALID_MANA_HANDLE;
 	apc->pf_filter_handle = INVALID_MANA_HANDLE;
 	apc->port_idx = port_idx;
diff --git a/drivers/net/ethernet/microsoft/mana/mana_ethtool.c b/drivers/net/ethernet/microsoft/mana/mana_ethtool.c
index 146d5db1792f..dff186811b25 100644
--- a/drivers/net/ethernet/microsoft/mana/mana_ethtool.c
+++ b/drivers/net/ethernet/microsoft/mana/mana_ethtool.c
@@ -369,6 +369,75 @@  static int mana_set_channels(struct net_device *ndev,
 	return err;
 }
 
+static void mana_get_ringparam(struct net_device *ndev,
+			       struct ethtool_ringparam *ring,
+			       struct kernel_ethtool_ringparam *kernel_ring,
+			       struct netlink_ext_ack *extack)
+{
+	struct mana_port_context *apc = netdev_priv(ndev);
+
+	ring->rx_pending = apc->rx_queue_size;
+	ring->tx_pending = apc->tx_queue_size;
+	ring->rx_max_pending = MAX_RX_BUFFERS_PER_QUEUE;
+	ring->tx_max_pending = MAX_TX_BUFFERS_PER_QUEUE;
+}
+
+static int mana_set_ringparam(struct net_device *ndev,
+			      struct ethtool_ringparam *ring,
+			      struct kernel_ethtool_ringparam *kernel_ring,
+			      struct netlink_ext_ack *extack)
+{
+	struct mana_port_context *apc = netdev_priv(ndev);
+	u32 new_tx, new_rx;
+	u32 old_tx, old_rx;
+	int err;
+
+	old_tx = apc->tx_queue_size;
+	old_rx = apc->rx_queue_size;
+	new_tx = clamp_t(u32, ring->tx_pending, MIN_TX_BUFFERS_PER_QUEUE, MAX_TX_BUFFERS_PER_QUEUE);
+	new_rx = clamp_t(u32, ring->rx_pending, MIN_RX_BUFFERS_PER_QUEUE, MAX_RX_BUFFERS_PER_QUEUE);
+
+	if (!is_power_of_2(new_tx)) {
+		netdev_err(ndev, "%s:Tx:%d not supported. Needs to be a power of 2\n",
+			   __func__, new_tx);
+		return -EINVAL;
+	}
+
+	if (!is_power_of_2(new_rx)) {
+		netdev_err(ndev, "%s:Rx:%d not supported. Needs to be a power of 2\n",
+			   __func__, new_rx);
+		return -EINVAL;
+	}
+
+	/* pre-allocating new buffers to prevent failures in mana_attach() later */
+	apc->rx_queue_size = new_rx;
+	err = mana_pre_alloc_rxbufs(apc, ndev->mtu);
+	apc->rx_queue_size = old_rx;
+	if (err) {
+		netdev_err(ndev, "Insufficient memory for new allocations\n");
+		return err;
+	}
+
+	err = mana_detach(ndev, false);
+	if (err) {
+		netdev_err(ndev, "mana_detach failed: %d\n", err);
+		goto out;
+	}
+
+	apc->tx_queue_size = new_tx;
+	apc->rx_queue_size = new_rx;
+
+	err = mana_attach(ndev);
+	if (err) {
+		netdev_err(ndev, "mana_attach failed: %d\n", err);
+		apc->tx_queue_size = old_tx;
+		apc->rx_queue_size = old_rx;
+	}
+out:
+	mana_pre_dealloc_rxbufs(apc);
+	return err;
+}
+
 const struct ethtool_ops mana_ethtool_ops = {
 	.get_ethtool_stats	= mana_get_ethtool_stats,
 	.get_sset_count		= mana_get_sset_count,
@@ -380,4 +449,6 @@  const struct ethtool_ops mana_ethtool_ops = {
 	.set_rxfh		= mana_set_rxfh,
 	.get_channels		= mana_get_channels,
 	.set_channels		= mana_set_channels,
+	.get_ringparam          = mana_get_ringparam,
+	.set_ringparam          = mana_set_ringparam,
 };
diff --git a/include/net/mana/mana.h b/include/net/mana/mana.h
index 6439fd8b437b..80a1e53471a6 100644
--- a/include/net/mana/mana.h
+++ b/include/net/mana/mana.h
@@ -38,9 +38,21 @@  enum TRI_STATE {
 
 #define COMP_ENTRY_SIZE 64
 
-#define RX_BUFFERS_PER_QUEUE 512
+/* This Max value for RX buffers is derived from __alloc_page()'s max page
+ * allocation calculation. It allows maximum 2^(MAX_ORDER -1) pages. RX buffer
+ * size beyond this value gets rejected by __alloc_page() call.
+ */
+#define MAX_RX_BUFFERS_PER_QUEUE 8192
+#define DEF_RX_BUFFERS_PER_QUEUE 512
+#define MIN_RX_BUFFERS_PER_QUEUE 128
 
-#define MAX_SEND_BUFFERS_PER_QUEUE 256
+/* This max value for TX buffers is derived as the maximum allocatable
+ * pages supported on host per guest through testing. TX buffer size beyond
+ * this value is rejected by the hardware.
+ */
+#define MAX_TX_BUFFERS_PER_QUEUE 16384
+#define DEF_TX_BUFFERS_PER_QUEUE 256
+#define MIN_TX_BUFFERS_PER_QUEUE 128
 
 #define EQ_SIZE (8 * MANA_PAGE_SIZE)
 
@@ -285,7 +297,7 @@  struct mana_recv_buf_oob {
 	void *buf_va;
 	bool from_pool; /* allocated from a page pool */
 
-	/* SGL of the buffer going to be sent has part of the work request. */
+	/* SGL of the buffer going to be sent as part of the work request. */
 	u32 num_sge;
 	struct gdma_sge sgl[MAX_RX_WQE_SGL_ENTRIES];
 
@@ -437,6 +449,9 @@  struct mana_port_context {
 	unsigned int max_queues;
 	unsigned int num_queues;
 
+	unsigned int rx_queue_size;
+	unsigned int tx_queue_size;
+
 	mana_handle_t port_handle;
 	mana_handle_t pf_filter_handle;
 
@@ -472,6 +487,8 @@  struct bpf_prog *mana_xdp_get(struct mana_port_context *apc);
 void mana_chn_setxdp(struct mana_port_context *apc, struct bpf_prog *prog);
 int mana_bpf(struct net_device *ndev, struct netdev_bpf *bpf);
 void mana_query_gf_stats(struct mana_port_context *apc);
+int mana_pre_alloc_rxbufs(struct mana_port_context *apc, int mtu);
+void mana_pre_dealloc_rxbufs(struct mana_port_context *apc);
 
 extern const struct ethtool_ops mana_ethtool_ops;