* [PATCH net-next 1/6] net: move HDS config from ethtool state
2025-01-17 19:48 [PATCH net-next 0/6] net: ethtool: fixes for HDS threshold Jakub Kicinski
@ 2025-01-17 19:48 ` Jakub Kicinski
2025-01-17 19:48 ` [PATCH net-next 2/6] net: provide pending ring configuration in net_device Jakub Kicinski
` (4 subsequent siblings)
5 siblings, 0 replies; 9+ messages in thread
From: Jakub Kicinski @ 2025-01-17 19:48 UTC (permalink / raw)
To: davem
Cc: netdev, edumazet, pabeni, andrew+netdev, horms, michael.chan,
pavan.chebbi, ap420073, Jakub Kicinski
Separate the HDS config from the ethtool state struct.
The HDS config contains just simple parameters, not state.
Having it as a separate struct will make it easier to clone / copy
and also long term potentially make it per-queue.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
include/linux/ethtool.h | 4 ----
include/linux/netdevice.h | 3 +++
include/net/netdev_queues.h | 10 ++++++++++
drivers/net/ethernet/broadcom/bnxt/bnxt.c | 4 ++--
drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c | 5 +++--
drivers/net/netdevsim/ethtool.c | 9 +++++----
drivers/net/netdevsim/netdev.c | 10 +++++-----
net/core/dev.c | 10 ++++++++--
net/core/devmem.c | 4 ++--
net/ethtool/rings.c | 8 +++++---
10 files changed, 43 insertions(+), 24 deletions(-)
diff --git a/include/linux/ethtool.h b/include/linux/ethtool.h
index e4136b0df892..ddae13cfd9ca 100644
--- a/include/linux/ethtool.h
+++ b/include/linux/ethtool.h
@@ -1164,16 +1164,12 @@ int ethtool_virtdev_set_link_ksettings(struct net_device *dev,
* @rss_ctx: XArray of custom RSS contexts
* @rss_lock: Protects entries in @rss_ctx. May be taken from
* within RTNL.
- * @hds_thresh: HDS Threshold value.
- * @hds_config: HDS value from userspace.
* @wol_enabled: Wake-on-LAN is enabled
* @module_fw_flash_in_progress: Module firmware flashing is in progress.
*/
struct ethtool_netdev_state {
struct xarray rss_ctx;
struct mutex rss_lock;
- u32 hds_thresh;
- u8 hds_config;
unsigned wol_enabled:1;
unsigned module_fw_flash_in_progress:1;
};
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 8308d9c75918..173a8b3a9eb2 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -63,6 +63,7 @@ struct dsa_port;
struct ip_tunnel_parm_kern;
struct macsec_context;
struct macsec_ops;
+struct netdev_config;
struct netdev_name_node;
struct sd_flow_limit;
struct sfp_bus;
@@ -2410,6 +2411,8 @@ struct net_device {
const struct udp_tunnel_nic_info *udp_tunnel_nic_info;
struct udp_tunnel_nic *udp_tunnel_nic;
+ /** @cfg: net_device queue-related configuration */
+ struct netdev_config *cfg;
struct ethtool_netdev_state *ethtool;
/* protected by rtnl_lock */
diff --git a/include/net/netdev_queues.h b/include/net/netdev_queues.h
index 5ca019d294ca..b02bb9f109d5 100644
--- a/include/net/netdev_queues.h
+++ b/include/net/netdev_queues.h
@@ -4,6 +4,16 @@
#include <linux/netdevice.h>
+/**
+ * struct netdev_config - queue-related configuration for a netdev
+ * @hds_thresh: HDS Threshold value.
+ * @hds_config: HDS value from userspace.
+ */
+struct netdev_config {
+ u32 hds_thresh;
+ u8 hds_config;
+};
+
/* See the netdev.yaml spec for definition of each statistic */
struct netdev_queue_stats_rx {
u64 bytes;
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 748c9b1ea701..0998b20578b4 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -4610,7 +4610,7 @@ void bnxt_set_tpa_flags(struct bnxt *bp)
static void bnxt_init_ring_params(struct bnxt *bp)
{
bp->rx_copybreak = BNXT_DEFAULT_RX_COPYBREAK;
- bp->dev->ethtool->hds_thresh = BNXT_DEFAULT_RX_COPYBREAK;
+ bp->dev->cfg->hds_thresh = BNXT_DEFAULT_RX_COPYBREAK;
}
/* bp->rx_ring_size, bp->tx_ring_size, dev->mtu, BNXT_FLAG_{G|L}RO flags must
@@ -6585,7 +6585,7 @@ static void bnxt_hwrm_update_rss_hash_cfg(struct bnxt *bp)
static int bnxt_hwrm_vnic_set_hds(struct bnxt *bp, struct bnxt_vnic_info *vnic)
{
- u16 hds_thresh = (u16)bp->dev->ethtool->hds_thresh;
+ u16 hds_thresh = (u16)bp->dev->cfg->hds_thresh;
struct hwrm_vnic_plcmodes_cfg_input *req;
int rc;
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c
index 65a20931c579..0a6d47d4d66b 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c
@@ -24,6 +24,7 @@
#include <linux/ptp_clock_kernel.h>
#include <linux/net_tstamp.h>
#include <linux/timecounter.h>
+#include <net/netdev_queues.h>
#include <net/netlink.h>
#include "bnxt_hsi.h"
#include "bnxt.h"
@@ -834,7 +835,7 @@ static void bnxt_get_ringparam(struct net_device *dev,
ering->rx_jumbo_pending = bp->rx_agg_ring_size;
ering->tx_pending = bp->tx_ring_size;
- kernel_ering->hds_thresh = dev->ethtool->hds_thresh;
+ kernel_ering->hds_thresh = dev->cfg->hds_thresh;
kernel_ering->hds_thresh_max = BNXT_HDS_THRESHOLD_MAX;
}
@@ -852,7 +853,7 @@ static int bnxt_set_ringparam(struct net_device *dev,
(ering->tx_pending < BNXT_MIN_TX_DESC_CNT))
return -EINVAL;
- hds_config_mod = tcp_data_split != dev->ethtool->hds_config;
+ hds_config_mod = tcp_data_split != dev->cfg->hds_config;
if (tcp_data_split == ETHTOOL_TCP_DATA_SPLIT_DISABLED && hds_config_mod)
return -EINVAL;
diff --git a/drivers/net/netdevsim/ethtool.c b/drivers/net/netdevsim/ethtool.c
index 12163635b759..189793debdb7 100644
--- a/drivers/net/netdevsim/ethtool.c
+++ b/drivers/net/netdevsim/ethtool.c
@@ -3,6 +3,7 @@
#include <linux/debugfs.h>
#include <linux/random.h>
+#include <net/netdev_queues.h>
#include "netdevsim.h"
@@ -71,8 +72,8 @@ static void nsim_get_ringparam(struct net_device *dev,
struct netdevsim *ns = netdev_priv(dev);
memcpy(ring, &ns->ethtool.ring, sizeof(ns->ethtool.ring));
- kernel_ring->tcp_data_split = dev->ethtool->hds_config;
- kernel_ring->hds_thresh = dev->ethtool->hds_thresh;
+ kernel_ring->tcp_data_split = dev->cfg->hds_config;
+ kernel_ring->hds_thresh = dev->cfg->hds_thresh;
kernel_ring->hds_thresh_max = NSIM_HDS_THRESHOLD_MAX;
if (kernel_ring->tcp_data_split == ETHTOOL_TCP_DATA_SPLIT_UNKNOWN)
@@ -190,8 +191,8 @@ static void nsim_ethtool_ring_init(struct netdevsim *ns)
ns->ethtool.ring.rx_mini_max_pending = 4096;
ns->ethtool.ring.tx_max_pending = 4096;
- ns->netdev->ethtool->hds_config = ETHTOOL_TCP_DATA_SPLIT_UNKNOWN;
- ns->netdev->ethtool->hds_thresh = 0;
+ ns->netdev->cfg->hds_config = ETHTOOL_TCP_DATA_SPLIT_UNKNOWN;
+ ns->netdev->cfg->hds_thresh = 0;
}
void nsim_ethtool_init(struct netdevsim *ns)
diff --git a/drivers/net/netdevsim/netdev.c b/drivers/net/netdevsim/netdev.c
index f92b05ccdca9..42f247cbdcee 100644
--- a/drivers/net/netdevsim/netdev.c
+++ b/drivers/net/netdevsim/netdev.c
@@ -55,10 +55,10 @@ static int nsim_forward_skb(struct net_device *dev, struct sk_buff *skb,
static netdev_tx_t nsim_start_xmit(struct sk_buff *skb, struct net_device *dev)
{
struct netdevsim *ns = netdev_priv(dev);
- struct ethtool_netdev_state *ethtool;
struct net_device *peer_dev;
unsigned int len = skb->len;
struct netdevsim *peer_ns;
+ struct netdev_config *cfg;
struct nsim_rq *rq;
int rxq;
@@ -76,11 +76,11 @@ static netdev_tx_t nsim_start_xmit(struct sk_buff *skb, struct net_device *dev)
rxq = rxq % peer_dev->num_rx_queues;
rq = peer_ns->rq[rxq];
- ethtool = peer_dev->ethtool;
+ cfg = peer_dev->cfg;
if (skb_is_nonlinear(skb) &&
- (ethtool->hds_config != ETHTOOL_TCP_DATA_SPLIT_ENABLED ||
- (ethtool->hds_config == ETHTOOL_TCP_DATA_SPLIT_ENABLED &&
- ethtool->hds_thresh > len)))
+ (cfg->hds_config != ETHTOOL_TCP_DATA_SPLIT_ENABLED ||
+ (cfg->hds_config == ETHTOOL_TCP_DATA_SPLIT_ENABLED &&
+ cfg->hds_thresh > len)))
skb_linearize(skb);
skb_tx_timestamp(skb);
diff --git a/net/core/dev.c b/net/core/dev.c
index fe5f5855593d..5bcf44e3bc6c 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -106,6 +106,7 @@
#include <net/dst.h>
#include <net/dst_metadata.h>
#include <net/gro.h>
+#include <net/netdev_queues.h>
#include <net/pkt_sched.h>
#include <net/pkt_cls.h>
#include <net/checksum.h>
@@ -9716,7 +9717,7 @@ int dev_xdp_propagate(struct net_device *dev, struct netdev_bpf *bpf)
if (!dev->netdev_ops->ndo_bpf)
return -EOPNOTSUPP;
- if (dev->ethtool->hds_config == ETHTOOL_TCP_DATA_SPLIT_ENABLED &&
+ if (dev->cfg->hds_config == ETHTOOL_TCP_DATA_SPLIT_ENABLED &&
bpf->command == XDP_SETUP_PROG &&
bpf->prog && !bpf->prog->aux->xdp_has_frags) {
NL_SET_ERR_MSG(bpf->extack,
@@ -9761,7 +9762,7 @@ static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
struct netdev_bpf xdp;
int err;
- if (dev->ethtool->hds_config == ETHTOOL_TCP_DATA_SPLIT_ENABLED &&
+ if (dev->cfg->hds_config == ETHTOOL_TCP_DATA_SPLIT_ENABLED &&
prog && !prog->aux->xdp_has_frags) {
NL_SET_ERR_MSG(extack, "unable to install XDP to device using tcp-data-split");
return -EBUSY;
@@ -11539,6 +11540,10 @@ struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
if (!dev->ethtool)
goto free_all;
+ dev->cfg = kzalloc(sizeof(*dev->cfg), GFP_KERNEL_ACCOUNT);
+ if (!dev->cfg)
+ goto free_all;
+
napi_config_sz = array_size(maxqs, sizeof(*dev->napi_config));
dev->napi_config = kvzalloc(napi_config_sz, GFP_KERNEL_ACCOUNT);
if (!dev->napi_config)
@@ -11595,6 +11600,7 @@ void free_netdev(struct net_device *dev)
mutex_destroy(&dev->lock);
+ kfree(dev->cfg);
kfree(dev->ethtool);
netif_free_tx_queues(dev);
netif_free_rx_queues(dev);
diff --git a/net/core/devmem.c b/net/core/devmem.c
index c971b8aceac8..3bba3f018df0 100644
--- a/net/core/devmem.c
+++ b/net/core/devmem.c
@@ -141,12 +141,12 @@ int net_devmem_bind_dmabuf_to_queue(struct net_device *dev, u32 rxq_idx,
return -ERANGE;
}
- if (dev->ethtool->hds_config != ETHTOOL_TCP_DATA_SPLIT_ENABLED) {
+ if (dev->cfg->hds_config != ETHTOOL_TCP_DATA_SPLIT_ENABLED) {
NL_SET_ERR_MSG(extack, "tcp-data-split is disabled");
return -EINVAL;
}
- if (dev->ethtool->hds_thresh) {
+ if (dev->cfg->hds_thresh) {
NL_SET_ERR_MSG(extack, "hds-thresh is not zero");
return -EINVAL;
}
diff --git a/net/ethtool/rings.c b/net/ethtool/rings.c
index d8cd4e4d7762..7a3c2a2dff12 100644
--- a/net/ethtool/rings.c
+++ b/net/ethtool/rings.c
@@ -1,5 +1,7 @@
// SPDX-License-Identifier: GPL-2.0-only
+#include <net/netdev_queues.h>
+
#include "netlink.h"
#include "common.h"
@@ -219,7 +221,7 @@ ethnl_set_rings(struct ethnl_req_info *req_info, struct genl_info *info)
dev->ethtool_ops->get_ringparam(dev, &ringparam,
&kernel_ringparam, info->extack);
- kernel_ringparam.tcp_data_split = dev->ethtool->hds_config;
+ kernel_ringparam.tcp_data_split = dev->cfg->hds_config;
ethnl_update_u32(&ringparam.rx_pending, tb[ETHTOOL_A_RINGS_RX], &mod);
ethnl_update_u32(&ringparam.rx_mini_pending,
@@ -295,8 +297,8 @@ ethnl_set_rings(struct ethnl_req_info *req_info, struct genl_info *info)
ret = dev->ethtool_ops->set_ringparam(dev, &ringparam,
&kernel_ringparam, info->extack);
if (!ret) {
- dev->ethtool->hds_config = kernel_ringparam.tcp_data_split;
- dev->ethtool->hds_thresh = kernel_ringparam.hds_thresh;
+ dev->cfg->hds_config = kernel_ringparam.tcp_data_split;
+ dev->cfg->hds_thresh = kernel_ringparam.hds_thresh;
}
return ret < 0 ? ret : 1;
--
2.48.1
^ permalink raw reply related [flat|nested] 9+ messages in thread* [PATCH net-next 2/6] net: provide pending ring configuration in net_device
2025-01-17 19:48 [PATCH net-next 0/6] net: ethtool: fixes for HDS threshold Jakub Kicinski
2025-01-17 19:48 ` [PATCH net-next 1/6] net: move HDS config from ethtool state Jakub Kicinski
@ 2025-01-17 19:48 ` Jakub Kicinski
2025-01-17 22:07 ` Michael Chan
2025-01-17 19:48 ` [PATCH net-next 3/6] eth: bnxt: apply hds_thrs settings correctly Jakub Kicinski
` (3 subsequent siblings)
5 siblings, 1 reply; 9+ messages in thread
From: Jakub Kicinski @ 2025-01-17 19:48 UTC (permalink / raw)
To: davem
Cc: netdev, edumazet, pabeni, andrew+netdev, horms, michael.chan,
pavan.chebbi, ap420073, Jakub Kicinski
Record the pending configuration in net_device struct.
ethtool core duplicates the current config and the specific
handlers (for now just ringparam) can modify it.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
include/linux/netdevice.h | 6 ++++++
net/core/dev.c | 2 ++
net/ethtool/netlink.c | 22 +++++++++++++++++++---
net/ethtool/rings.c | 8 +++-----
4 files changed, 30 insertions(+), 8 deletions(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 173a8b3a9eb2..8da4c61f97b9 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -2413,6 +2413,12 @@ struct net_device {
/** @cfg: net_device queue-related configuration */
struct netdev_config *cfg;
+ /**
+ * @cfg_pending: same as @cfg but when device is being actively
+ * reconfigured includes any changes to the configuration
+ * requested by the user, but which may or may not be rejected.
+ */
+ struct netdev_config *cfg_pending;
struct ethtool_netdev_state *ethtool;
/* protected by rtnl_lock */
diff --git a/net/core/dev.c b/net/core/dev.c
index 5bcf44e3bc6c..1daedb0f8aad 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -11543,6 +11543,7 @@ struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
dev->cfg = kzalloc(sizeof(*dev->cfg), GFP_KERNEL_ACCOUNT);
if (!dev->cfg)
goto free_all;
+ dev->cfg_pending = dev->cfg;
napi_config_sz = array_size(maxqs, sizeof(*dev->napi_config));
dev->napi_config = kvzalloc(napi_config_sz, GFP_KERNEL_ACCOUNT);
@@ -11600,6 +11601,7 @@ void free_netdev(struct net_device *dev)
mutex_destroy(&dev->lock);
+ WARN_ON(dev->cfg != dev->cfg_pending);
kfree(dev->cfg);
kfree(dev->ethtool);
netif_free_tx_queues(dev);
diff --git a/net/ethtool/netlink.c b/net/ethtool/netlink.c
index 849c98e637c6..0d47e69dcda0 100644
--- a/net/ethtool/netlink.c
+++ b/net/ethtool/netlink.c
@@ -1,5 +1,6 @@
// SPDX-License-Identifier: GPL-2.0-only
+#include <net/netdev_queues.h>
#include <net/sock.h>
#include <linux/ethtool_netlink.h>
#include <linux/phy_link_topology.h>
@@ -667,6 +668,7 @@ static int ethnl_default_set_doit(struct sk_buff *skb, struct genl_info *info)
const struct ethnl_request_ops *ops;
struct ethnl_req_info req_info = {};
const u8 cmd = info->genlhdr->cmd;
+ struct net_device *dev;
int ret;
ops = ethnl_default_requests[cmd];
@@ -688,21 +690,35 @@ static int ethnl_default_set_doit(struct sk_buff *skb, struct genl_info *info)
goto out_dev;
}
+ dev = req_info.dev;
+
+ dev->cfg_pending = kmemdup(dev->cfg, sizeof(*dev->cfg),
+ GFP_KERNEL_ACCOUNT);
+ if (!dev->cfg_pending) {
+ ret = -ENOMEM;
+ goto out_tie_cfg;
+ }
+
rtnl_lock();
- ret = ethnl_ops_begin(req_info.dev);
+ ret = ethnl_ops_begin(dev);
if (ret < 0)
goto out_rtnl;
ret = ops->set(&req_info, info);
if (ret <= 0)
goto out_ops;
- ethtool_notify(req_info.dev, ops->set_ntf_cmd, NULL);
+ ethtool_notify(dev, ops->set_ntf_cmd, NULL);
ret = 0;
out_ops:
- ethnl_ops_complete(req_info.dev);
+ ethnl_ops_complete(dev);
out_rtnl:
rtnl_unlock();
+ if (ret >= 0) /* success! */
+ swap(dev->cfg, dev->cfg_pending);
+ kfree(dev->cfg_pending);
+out_tie_cfg:
+ dev->cfg_pending = dev->cfg;
out_dev:
ethnl_parse_header_dev_put(&req_info);
return ret;
diff --git a/net/ethtool/rings.c b/net/ethtool/rings.c
index 7a3c2a2dff12..5e8ba81fbb3e 100644
--- a/net/ethtool/rings.c
+++ b/net/ethtool/rings.c
@@ -294,13 +294,11 @@ ethnl_set_rings(struct ethnl_req_info *req_info, struct genl_info *info)
return -EINVAL;
}
+ dev->cfg_pending->hds_config = kernel_ringparam.tcp_data_split;
+ dev->cfg_pending->hds_thresh = kernel_ringparam.hds_thresh;
+
ret = dev->ethtool_ops->set_ringparam(dev, &ringparam,
&kernel_ringparam, info->extack);
- if (!ret) {
- dev->cfg->hds_config = kernel_ringparam.tcp_data_split;
- dev->cfg->hds_thresh = kernel_ringparam.hds_thresh;
- }
-
return ret < 0 ? ret : 1;
}
--
2.48.1
^ permalink raw reply related [flat|nested] 9+ messages in thread* Re: [PATCH net-next 2/6] net: provide pending ring configuration in net_device
2025-01-17 19:48 ` [PATCH net-next 2/6] net: provide pending ring configuration in net_device Jakub Kicinski
@ 2025-01-17 22:07 ` Michael Chan
2025-01-17 22:18 ` Jakub Kicinski
0 siblings, 1 reply; 9+ messages in thread
From: Michael Chan @ 2025-01-17 22:07 UTC (permalink / raw)
To: Jakub Kicinski
Cc: davem, netdev, edumazet, pabeni, andrew+netdev, horms,
pavan.chebbi, ap420073
[-- Attachment #1: Type: text/plain, Size: 1783 bytes --]
On Fri, Jan 17, 2025 at 11:48 AM Jakub Kicinski <kuba@kernel.org> wrote:
>
> Record the pending configuration in net_device struct.
> ethtool core duplicates the current config and the specific
> handlers (for now just ringparam) can modify it.
>
> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
> @@ -688,21 +690,35 @@ static int ethnl_default_set_doit(struct sk_buff *skb, struct genl_info *info)
> goto out_dev;
> }
>
> + dev = req_info.dev;
> +
> + dev->cfg_pending = kmemdup(dev->cfg, sizeof(*dev->cfg),
> + GFP_KERNEL_ACCOUNT);
> + if (!dev->cfg_pending) {
> + ret = -ENOMEM;
> + goto out_tie_cfg;
> + }
> +
> rtnl_lock();
> - ret = ethnl_ops_begin(req_info.dev);
> + ret = ethnl_ops_begin(dev);
> if (ret < 0)
> goto out_rtnl;
>
> ret = ops->set(&req_info, info);
> if (ret <= 0)
> goto out_ops;
> - ethtool_notify(req_info.dev, ops->set_ntf_cmd, NULL);
> + ethtool_notify(dev, ops->set_ntf_cmd, NULL);
>
> ret = 0;
> out_ops:
> - ethnl_ops_complete(req_info.dev);
> + ethnl_ops_complete(dev);
> out_rtnl:
> rtnl_unlock();
> + if (ret >= 0) /* success! */
> + swap(dev->cfg, dev->cfg_pending);
> + kfree(dev->cfg_pending);
> +out_tie_cfg:
> + dev->cfg_pending = dev->cfg;
If I understand this correctly, cfg_pending can momentarily point to
the old cfg and freed memory before pointing to the new cfg outside of
rtnl_lock. Shouldn't it be inside the rtnl_lock?
In the bnxt patch, we now look at cfg_pending so it must always point
to the correct cfg.
[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4209 bytes --]
^ permalink raw reply [flat|nested] 9+ messages in thread* Re: [PATCH net-next 2/6] net: provide pending ring configuration in net_device
2025-01-17 22:07 ` Michael Chan
@ 2025-01-17 22:18 ` Jakub Kicinski
0 siblings, 0 replies; 9+ messages in thread
From: Jakub Kicinski @ 2025-01-17 22:18 UTC (permalink / raw)
To: Michael Chan
Cc: davem, netdev, edumazet, pabeni, andrew+netdev, horms,
pavan.chebbi, ap420073
On Fri, 17 Jan 2025 14:07:03 -0800 Michael Chan wrote:
> > +out_tie_cfg:
> > + dev->cfg_pending = dev->cfg;
>
> If I understand this correctly, cfg_pending can momentarily point to
> the old cfg and freed memory before pointing to the new cfg outside of
> rtnl_lock. Shouldn't it be inside the rtnl_lock?
>
> In the bnxt patch, we now look at cfg_pending so it must always point
> to the correct cfg.
Good catch, I moved it out to avoid an allocation under rtnl_lock..
--
pw-bot: cr
^ permalink raw reply [flat|nested] 9+ messages in thread
* [PATCH net-next 3/6] eth: bnxt: apply hds_thrs settings correctly
2025-01-17 19:48 [PATCH net-next 0/6] net: ethtool: fixes for HDS threshold Jakub Kicinski
2025-01-17 19:48 ` [PATCH net-next 1/6] net: move HDS config from ethtool state Jakub Kicinski
2025-01-17 19:48 ` [PATCH net-next 2/6] net: provide pending ring configuration in net_device Jakub Kicinski
@ 2025-01-17 19:48 ` Jakub Kicinski
2025-01-17 19:48 ` [PATCH net-next 4/6] net: ethtool: populate the default HDS params in the core Jakub Kicinski
` (2 subsequent siblings)
5 siblings, 0 replies; 9+ messages in thread
From: Jakub Kicinski @ 2025-01-17 19:48 UTC (permalink / raw)
To: davem
Cc: netdev, edumazet, pabeni, andrew+netdev, horms, michael.chan,
pavan.chebbi, ap420073, Jakub Kicinski
Use the pending config for hds_thrs. Core will only update the "current"
one after we return success. Without this change 2 reconfigs would be
required for the setting to reach the device.
Fixes: 6b43673a25c3 ("bnxt_en: add support for hds-thresh ethtool command")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
drivers/net/ethernet/broadcom/bnxt/bnxt.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 0998b20578b4..2eeed4c11b64 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -6585,7 +6585,7 @@ static void bnxt_hwrm_update_rss_hash_cfg(struct bnxt *bp)
static int bnxt_hwrm_vnic_set_hds(struct bnxt *bp, struct bnxt_vnic_info *vnic)
{
- u16 hds_thresh = (u16)bp->dev->cfg->hds_thresh;
+ u16 hds_thresh = (u16)bp->dev->cfg_pending->hds_thresh;
struct hwrm_vnic_plcmodes_cfg_input *req;
int rc;
--
2.48.1
^ permalink raw reply related [flat|nested] 9+ messages in thread* [PATCH net-next 4/6] net: ethtool: populate the default HDS params in the core
2025-01-17 19:48 [PATCH net-next 0/6] net: ethtool: fixes for HDS threshold Jakub Kicinski
` (2 preceding siblings ...)
2025-01-17 19:48 ` [PATCH net-next 3/6] eth: bnxt: apply hds_thrs settings correctly Jakub Kicinski
@ 2025-01-17 19:48 ` Jakub Kicinski
2025-01-17 19:48 ` [PATCH net-next 5/6] eth: bnxt: allocate enough buffer space to meet HDS threshold Jakub Kicinski
2025-01-17 19:48 ` [PATCH net-next 6/6] eth: bnxt: update header sizing defaults Jakub Kicinski
5 siblings, 0 replies; 9+ messages in thread
From: Jakub Kicinski @ 2025-01-17 19:48 UTC (permalink / raw)
To: davem
Cc: netdev, edumazet, pabeni, andrew+netdev, horms, michael.chan,
pavan.chebbi, ap420073, Jakub Kicinski
The core has the current HDS config, it can pre-populate the values
for the drivers. While at it, remove the zero-setting in netdevsim.
Zero are the default values since the config is zalloc'ed.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c | 1 -
drivers/net/netdevsim/ethtool.c | 5 -----
net/ethtool/rings.c | 4 ++++
3 files changed, 4 insertions(+), 6 deletions(-)
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c
index 0a6d47d4d66b..9c5820839514 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c
@@ -835,7 +835,6 @@ static void bnxt_get_ringparam(struct net_device *dev,
ering->rx_jumbo_pending = bp->rx_agg_ring_size;
ering->tx_pending = bp->tx_ring_size;
- kernel_ering->hds_thresh = dev->cfg->hds_thresh;
kernel_ering->hds_thresh_max = BNXT_HDS_THRESHOLD_MAX;
}
diff --git a/drivers/net/netdevsim/ethtool.c b/drivers/net/netdevsim/ethtool.c
index 189793debdb7..3b23f3d3ca2b 100644
--- a/drivers/net/netdevsim/ethtool.c
+++ b/drivers/net/netdevsim/ethtool.c
@@ -72,8 +72,6 @@ static void nsim_get_ringparam(struct net_device *dev,
struct netdevsim *ns = netdev_priv(dev);
memcpy(ring, &ns->ethtool.ring, sizeof(ns->ethtool.ring));
- kernel_ring->tcp_data_split = dev->cfg->hds_config;
- kernel_ring->hds_thresh = dev->cfg->hds_thresh;
kernel_ring->hds_thresh_max = NSIM_HDS_THRESHOLD_MAX;
if (kernel_ring->tcp_data_split == ETHTOOL_TCP_DATA_SPLIT_UNKNOWN)
@@ -190,9 +188,6 @@ static void nsim_ethtool_ring_init(struct netdevsim *ns)
ns->ethtool.ring.rx_jumbo_max_pending = 4096;
ns->ethtool.ring.rx_mini_max_pending = 4096;
ns->ethtool.ring.tx_max_pending = 4096;
-
- ns->netdev->cfg->hds_config = ETHTOOL_TCP_DATA_SPLIT_UNKNOWN;
- ns->netdev->cfg->hds_thresh = 0;
}
void nsim_ethtool_init(struct netdevsim *ns)
diff --git a/net/ethtool/rings.c b/net/ethtool/rings.c
index 5e8ba81fbb3e..7839bfd1ac6a 100644
--- a/net/ethtool/rings.c
+++ b/net/ethtool/rings.c
@@ -39,6 +39,10 @@ static int rings_prepare_data(const struct ethnl_req_info *req_base,
ret = ethnl_ops_begin(dev);
if (ret < 0)
return ret;
+
+ data->kernel_ringparam.tcp_data_split = dev->cfg->hds_config;
+ data->kernel_ringparam.hds_thresh = dev->cfg->hds_thresh;
+
dev->ethtool_ops->get_ringparam(dev, &data->ringparam,
&data->kernel_ringparam, info->extack);
ethnl_ops_complete(dev);
--
2.48.1
^ permalink raw reply related [flat|nested] 9+ messages in thread* [PATCH net-next 5/6] eth: bnxt: allocate enough buffer space to meet HDS threshold
2025-01-17 19:48 [PATCH net-next 0/6] net: ethtool: fixes for HDS threshold Jakub Kicinski
` (3 preceding siblings ...)
2025-01-17 19:48 ` [PATCH net-next 4/6] net: ethtool: populate the default HDS params in the core Jakub Kicinski
@ 2025-01-17 19:48 ` Jakub Kicinski
2025-01-17 19:48 ` [PATCH net-next 6/6] eth: bnxt: update header sizing defaults Jakub Kicinski
5 siblings, 0 replies; 9+ messages in thread
From: Jakub Kicinski @ 2025-01-17 19:48 UTC (permalink / raw)
To: davem
Cc: netdev, edumazet, pabeni, andrew+netdev, horms, michael.chan,
pavan.chebbi, ap420073, Jakub Kicinski
Now that we can configure HDS threshold separately from the rx_copybreak
HDS threshold may be higher than rx_copybreak.
We need to make sure that we have enough space for the headers.
Fixes: 6b43673a25c3 ("bnxt_en: add support for hds-thresh ethtool command")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
drivers/net/ethernet/broadcom/bnxt/bnxt.c | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 2eeed4c11b64..19e723493c4e 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -4671,9 +4671,10 @@ void bnxt_set_ring_params(struct bnxt *bp)
ALIGN(max(NET_SKB_PAD, XDP_PACKET_HEADROOM), 8) -
SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
} else {
- rx_size = SKB_DATA_ALIGN(max(BNXT_DEFAULT_RX_COPYBREAK,
- bp->rx_copybreak) +
- NET_IP_ALIGN);
+ rx_size = max3(BNXT_DEFAULT_RX_COPYBREAK,
+ bp->rx_copybreak,
+ bp->dev->cfg_pending->hds_thresh);
+ rx_size = SKB_DATA_ALIGN(rx_size + NET_IP_ALIGN);
rx_space = rx_size + NET_SKB_PAD +
SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
}
--
2.48.1
^ permalink raw reply related [flat|nested] 9+ messages in thread* [PATCH net-next 6/6] eth: bnxt: update header sizing defaults
2025-01-17 19:48 [PATCH net-next 0/6] net: ethtool: fixes for HDS threshold Jakub Kicinski
` (4 preceding siblings ...)
2025-01-17 19:48 ` [PATCH net-next 5/6] eth: bnxt: allocate enough buffer space to meet HDS threshold Jakub Kicinski
@ 2025-01-17 19:48 ` Jakub Kicinski
5 siblings, 0 replies; 9+ messages in thread
From: Jakub Kicinski @ 2025-01-17 19:48 UTC (permalink / raw)
To: davem
Cc: netdev, edumazet, pabeni, andrew+netdev, horms, michael.chan,
pavan.chebbi, ap420073, Jakub Kicinski
300-400B RPC requests are fairly common. With the current default
of 256B HDS threshold bnxt ends up splitting those, lowering PCIe
bandwidth efficiency and increasing the number of memory allocation.
Increase the HDS threshold to fit 4 buffers in a 4k page.
This works out to 640B as the threshold on a typical kernel confing.
This change increases the performance for a microbenchmark which
receives 400B RPCs and sends empty responses by 4.5%.
Admittedly this is just a single benchmark, but 256B works out to
just 6 (so 2 more) packets per head page, because shinfo size
dominates the headers.
Now that we use page pool for the header pages I was also tempted
to default rx_copybreak to 0, but in synthetic testing the copybreak
size doesn't seem to make much difference.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
drivers/net/ethernet/broadcom/bnxt/bnxt.c | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 19e723493c4e..589a1008601c 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -4609,8 +4609,13 @@ void bnxt_set_tpa_flags(struct bnxt *bp)
static void bnxt_init_ring_params(struct bnxt *bp)
{
+ unsigned int rx_size;
+
bp->rx_copybreak = BNXT_DEFAULT_RX_COPYBREAK;
- bp->dev->cfg->hds_thresh = BNXT_DEFAULT_RX_COPYBREAK;
+ /* Try to fit 4 chunks into a 4k page */
+ rx_size = SZ_1K -
+ NET_SKB_PAD - SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
+ bp->dev->cfg->hds_thresh = max(BNXT_DEFAULT_RX_COPYBREAK, rx_size);
}
/* bp->rx_ring_size, bp->tx_ring_size, dev->mtu, BNXT_FLAG_{G|L}RO flags must
--
2.48.1
^ permalink raw reply related [flat|nested] 9+ messages in thread