Skip to contentSkip to conversation

eth: bnxt: use the new RSS API

PATCH In review Jakub Kicinski 6 participants 30 messages Jul 2, 2024 – Jul 5, 2024

Subsystems (4)
lore
Share & tools
lore

Jakub Kicinski

author maintainer
#0 lore raw

cover eth: bnxt: use the new RSS API

Convert bnxt from using the set_rxfh API to separate create/modify/remove callbacks.

Two small extensions to the core APIs are necessary:
- the ability to discard contexts if for some catastrophic reasons device can no longer provide them;
- the ability to reserve space in the context for RSS table growth.

The driver is adjusted to store indirection tables on u32 to make it easier to use core structs directly.

With that out of the way the conversion is fairly straightforward.

Jakub Kicinski (11):
net: ethtool: let drivers remove lost RSS contexts
net: ethtool: let drivers declare max size of RSS indir table and key
eth: bnxt: allow deleting RSS contexts when the device is down
eth: bnxt: move from .set_rxfh to .create_rxfh_context and friends
eth: bnxt: remove rss_ctx_bmap
eth: bnxt: depend on core cleaning up RSS contexts
eth: bnxt: use context priv for struct bnxt_rss_ctx
eth: bnxt: use the RSS context XArray instead of the local list
eth: bnxt: bump the entry size in indir tables to u32
eth: bnxt: use the indir table from ethtool context
eth: bnxt: pad out the correct indirection table

 drivers/net/ethernet/broadcom/bnxt/bnxt.c     | 115 +++++--------
 drivers/net/ethernet/broadcom/bnxt/bnxt.h     |  14 +-
 .../net/ethernet/broadcom/bnxt/bnxt_ethtool.c | 159 ++++++++++--------
 include/linux/ethtool.h                       |  25 +--
 net/ethtool/ioctl.c                           |  46 +++--
 net/ethtool/rss.c                             |  14 ++
 6 files changed, 201 insertions(+), 172 deletions(-)
Signature · 1 line
2.45.2
author maintainer
#1 lore raw

PATCH 1/11 net: ethtool: let drivers remove lost RSS contexts

RSS contexts may get lost from a device, in various extreme circumstances. Specifically if the firmware leaks resources and resets, or crashes and either recovers in partially working state or the crash causes a different FW version to run - creating the context again may fail.

Drivers should do their absolute best to prevent this from happening. When it does, however, telling user that a context exists, when it can't possibly be used any more is counter productive. Add a helper for drivers to discard contexts. Print an error, in the future netlink notification will also be sent.

---
 include/linux/ethtool.h |  2 ++
 net/ethtool/rss.c       | 14 ++++++++++++++
 2 files changed, 16 insertions(+)
include/linux/ethtool.h +2 -0
index f74bb0cf8ed1..3ce5be0d168a 100644
@@ -210,6 +210,8 @@ static inline size_t ethtool_rxfh_context_size(u32 indir_size, u32 key_size,
 	return struct_size_t(struct ethtool_rxfh_context, data, flex_len);
 }
 
+void ethtool_rxfh_context_lost(struct net_device *dev, u32 context_id);
+
 /* declare a link mode bitmap */
 #define __ETHTOOL_DECLARE_LINK_MODE_MASK(name)		\
 	DECLARE_BITMAP(name, __ETHTOOL_LINK_MODE_MASK_NBITS)
net/ethtool/rss.c +14 -0
index 71679137eff2..e2e5bab56a6b 100644
@@ -159,3 +159,17 @@ const struct ethnl_request_ops ethnl_rss_request_ops = {
 	.fill_reply		= rss_fill_reply,
 	.cleanup_data		= rss_cleanup_data,
 };
+
+void ethtool_rxfh_context_lost(struct net_device *dev, u32 context_id)
+{
+	struct ethtool_rxfh_context *ctx;
+
+	WARN_ONCE(!rtnl_is_locked() &&
+		  !lockdep_is_held_type(&dev->ethtool->rss_lock, -1),
+		  "RSS context lock assertion failed\n");
+
+	netdev_err(dev, "device error, RSS context %d lost\n", context_id);
+	ctx = xa_erase(&dev->ethtool->rss_ctx, context_id);
+	kfree(ctx);
+}
+EXPORT_SYMBOL(ethtool_rxfh_context_lost);
Signature · 1 line
2.45.2
#2 lore raw

Re: [PATCH net-next 01/11] net: ethtool: let drivers remove lost RSS contexts

Quoted 4 lines · Jakub Kicinski

RSS contexts may get lost from a device, in various extreme circumstances. Specifically if the firmware leaks resources and resets, or crashes and either recovers in partially working state or the crash causes a different FW version to run - creating the context again may fail.

So, I deliberately *didn't* do this, on the grounds that if the user
fixed things by updating FW and resetting again, their contexts could get restored. I suppose big users like Meta will have orchestration doing all that work anyway so it doesn't matter.

Quoted 5 lines

Drivers should do their absolute best to prevent this from happening. When it does, however, telling user that a context exists, when it can't possibly be used any more is counter productive. Add a helper for drivers to discard contexts. Print an error, in the future netlink notification will also be sent.

Possibility of a netlink notification makes the idea of a broken flag
a bit more workable imho. But it's up to you which way to go.

author maintainer
#3 lore raw

Re: [PATCH net-next 01/11] net: ethtool: let drivers remove lost RSS contexts

Quoted 10 lines · On Wed, 3 Jul 2024 12:08:36 +0100 Edward Cree wrote:
Quoted 4 lines · Jakub Kicinski

RSS contexts may get lost from a device, in various extreme circumstances. Specifically if the firmware leaks resources and resets, or crashes and either recovers in partially working state or the crash causes a different FW version to run - creating the context again may fail.

So, I deliberately *didn't* do this, on the grounds that if the user
fixed things by updating FW and resetting again, their contexts could get restored. I suppose big users like Meta will have orchestration doing all that work anyway so it doesn't matter.

"We" don't reset FW while workload is running. I'm speculating why bnxt may lose the contexts. From my perspective if contexts get lost the machine should get taken out of production and at least power cycled.

Quoted 8 lines
Quoted 5 lines

Drivers should do their absolute best to prevent this from happening. When it does, however, telling user that a context exists, when it can't possibly be used any more is counter productive. Add a helper for drivers to discard contexts. Print an error, in the future netlink notification will also be sent.

Possibility of a netlink notification makes the idea of a broken flag
a bit more workable imho. But it's up to you which way to go.

Oh, have we talked about this? Now that you mention the broken flag I recall talking about devlink health reporter.. a while back.

I don't have a preference on how we deal with the lost contexts. The more obvious we make it to orchestration that the machine is broken the better. Can you point me to the discussion / describe the broken flag?

#4 lore raw

Re: [PATCH net-next 01/11] net: ethtool: let drivers remove lost RSS contexts

Quoted 11 lines · Jakub Kicinski
Quoted 2 lines · On Wed, 3 Jul 2024 12:08:36 +0100 Edward Cree wrote:

Possibility of a netlink notification makes the idea of a broken flag
a bit more workable imho. But it's up to you which way to go.

Oh, have we talked about this? Now that you mention the broken flag I recall talking about devlink health reporter.. a while back.

I don't have a preference on how we deal with the lost contexts. The more obvious we make it to orchestration that the machine is broken the better. Can you point me to the discussion / describe the broken flag?

We discussed it briefly on v4 back in October [1][2]. Idea was
there'd be a flag in struct ethtool_rxfh_context with meaning of "this context is not present in hw owing to reinsertion failure after a device reset", reported in ethtool -x and perhaps also devlink health. Driver could set this flag (which would trigger a netlink notification) but could also clear it if a second reset (or some runtime configuration change) triggers another round of reinsertion which succeeds this time.
Fwiw I don't have a strong preference either — like I say, you do
what you think is best.

-ed

[1]: https://lore.kernel.org/all/[email protected]/T/#ma8fce7df8b65601009551839d9d102c49e79803a
[2]: https://lore.kernel.org/all/[email protected]/T/#m073cb990982d89e35d78edf364389de62256664b

author maintainer
#5 lore raw

PATCH 2/11 net: ethtool: let driver declare max size of RSS indir table and key

In reply to Jakub Kicinski

Some drivers (bnxt but I think also mlx5 from ML discussions) change the size of the indirection table depending on the number of Rx rings. Decouple the max table size from the size of the currently used table, so that we can reserve space in the context for table growth.

Static members in ethtool_ops are good enough for now, we can add callbacks to read the max size more dynamically if someone needs that.

---
 include/linux/ethtool.h | 20 +++++++-----------
 net/ethtool/ioctl.c     | 46 ++++++++++++++++++++++++++++++++---------
 2 files changed, 44 insertions(+), 22 deletions(-)
include/linux/ethtool.h +8 -12
index 3ce5be0d168a..dc8ed93097c3 100644
@@ -173,6 +173,7 @@ static inline u32 ethtool_rxfh_indir_default(u32 index, u32 n_rx_rings)
 struct ethtool_rxfh_context {
 	u32 indir_size;
 	u32 key_size;
+	u32 key_off;
 	u16 priv_size;
 	u8 hfunc;
 	u8 input_xfrm;
@@ -196,18 +197,7 @@ static inline u32 *ethtool_rxfh_context_indir(struct ethtool_rxfh_context *ctx)
 
 static inline u8 *ethtool_rxfh_context_key(struct ethtool_rxfh_context *ctx)
 {
-	return (u8 *)(ethtool_rxfh_context_indir(ctx) + ctx->indir_size);
-}
-
-static inline size_t ethtool_rxfh_context_size(u32 indir_size, u32 key_size,
-					       u16 priv_size)
-{
-	size_t indir_bytes = array_size(indir_size, sizeof(u32));
-	size_t flex_len;
-
-	flex_len = size_add(size_add(indir_bytes, key_size),
-			    ALIGN(priv_size, sizeof(u32)));
-	return struct_size_t(struct ethtool_rxfh_context, data, flex_len);
+	return &ctx->data[ctx->key_off];
 }
 
 void ethtool_rxfh_context_lost(struct net_device *dev, u32 context_id);
@@ -723,6 +713,10 @@ struct ethtool_rxfh_param {
  *	contexts.
  * @cap_rss_sym_xor_supported: indicates if the driver supports symmetric-xor
  *	RSS.
+ * @rxfh_indir_space: max size of RSS indirection tables, if indirection table
+ *	size as returned by @get_rxfh_indir_size may change during lifetime
+ *	of the device. Leave as 0 if the table size is constant.
+ * @rxfh_key_space: same as @rxfh_indir_space, but for the key.
  * @rxfh_priv_size: size of the driver private data area the core should
  *	allocate for an RSS context (in &struct ethtool_rxfh_context).
  * @rxfh_max_context_id: maximum (exclusive) supported RSS context ID.  If this
@@ -940,6 +934,8 @@ struct ethtool_ops {
 	u32     cap_link_lanes_supported:1;
 	u32     cap_rss_ctx_supported:1;
 	u32	cap_rss_sym_xor_supported:1;
+	u32	rxfh_indir_space;
+	u16	rxfh_key_space;
 	u16	rxfh_priv_size;
 	u32	rxfh_max_context_id;
 	u32	supported_coalesce_params;
net/ethtool/ioctl.c +36 -10

62 lines not on this page · raw message

Signature · 1 line
2.45.2
reviewer
#6 lore raw

Re: [PATCH net-next 02/11] net: ethtool: let driver declare max size of RSS indir table and key

Quoted 20 lines · Jakub Kicinski

Some drivers (bnxt but I think also mlx5 from ML discussions) change the size of the indirection table depending on the number of Rx rings. Decouple the max table size from the size of the currently used table, so that we can reserve space in the context for table growth.

Static members in ethtool_ops are good enough for now, we can add callbacks to read the max size more dynamically if someone needs that.

---
 include/linux/ethtool.h | 20 +++++++-----------
 net/ethtool/ioctl.c     | 46 ++++++++++++++++++++++++++++++++---------
 2 files changed, 44 insertions(+), 22 deletions(-)
include/linux/ethtool.h +0 -0
index 3ce5be0d168a..dc8ed93097c3 100644

include/linux/ethtool.h static inline u32 ethtool_rxfh_indir_default(u32 index, u32 n_rx_rings) · line 179

Quoted 7 lines
include/linux/ethtool.h +1 -0
@@ -173,6 +173,7 @@ static inline u32 ethtool_rxfh_indir_default(u32 index, u32 n_rx_rings)
 struct ethtool_rxfh_context {
 	u32 indir_size;
 	u32 key_size;
+	u32 key_off;
 	u16 priv_size;
 	u8 hfunc;
 	u8 input_xfrm;

nit: key_off should have an entry in the kernel doc for
struct ethtool_rxfh_context.

...

author maintainer
#7 lore raw

PATCH 2/11 net: ethtool: let drivers declare max size of RSS indir table and key

In reply to Jakub Kicinski

Some drivers (bnxt but I think also mlx5 from ML discussions) change the size of the indirection table depending on the number of Rx rings. Decouple the max table size from the size of the currently used table, so that we can reserve space in the context for table growth.

Static members in ethtool_ops are good enough for now, we can add callbacks to read the max size more dynamically if someone needs that.

---
 include/linux/ethtool.h | 20 +++++++-----------
 net/ethtool/ioctl.c     | 46 ++++++++++++++++++++++++++++++++---------
 2 files changed, 44 insertions(+), 22 deletions(-)
include/linux/ethtool.h +8 -12
index 3ce5be0d168a..dc8ed93097c3 100644
@@ -173,6 +173,7 @@ static inline u32 ethtool_rxfh_indir_default(u32 index, u32 n_rx_rings)
 struct ethtool_rxfh_context {
 	u32 indir_size;
 	u32 key_size;
+	u32 key_off;
 	u16 priv_size;
 	u8 hfunc;
 	u8 input_xfrm;
@@ -196,18 +197,7 @@ static inline u32 *ethtool_rxfh_context_indir(struct ethtool_rxfh_context *ctx)
 
 static inline u8 *ethtool_rxfh_context_key(struct ethtool_rxfh_context *ctx)
 {
-	return (u8 *)(ethtool_rxfh_context_indir(ctx) + ctx->indir_size);
-}
-
-static inline size_t ethtool_rxfh_context_size(u32 indir_size, u32 key_size,
-					       u16 priv_size)
-{
-	size_t indir_bytes = array_size(indir_size, sizeof(u32));
-	size_t flex_len;
-
-	flex_len = size_add(size_add(indir_bytes, key_size),
-			    ALIGN(priv_size, sizeof(u32)));
-	return struct_size_t(struct ethtool_rxfh_context, data, flex_len);
+	return &ctx->data[ctx->key_off];
 }
 
 void ethtool_rxfh_context_lost(struct net_device *dev, u32 context_id);
@@ -723,6 +713,10 @@ struct ethtool_rxfh_param {
  *	contexts.
  * @cap_rss_sym_xor_supported: indicates if the driver supports symmetric-xor
  *	RSS.
+ * @rxfh_indir_space: max size of RSS indirection tables, if indirection table
+ *	size as returned by @get_rxfh_indir_size may change during lifetime
+ *	of the device. Leave as 0 if the table size is constant.
+ * @rxfh_key_space: same as @rxfh_indir_space, but for the key.
  * @rxfh_priv_size: size of the driver private data area the core should
  *	allocate for an RSS context (in &struct ethtool_rxfh_context).
  * @rxfh_max_context_id: maximum (exclusive) supported RSS context ID.  If this
@@ -940,6 +934,8 @@ struct ethtool_ops {
 	u32     cap_link_lanes_supported:1;
 	u32     cap_rss_ctx_supported:1;
 	u32	cap_rss_sym_xor_supported:1;
+	u32	rxfh_indir_space;
+	u16	rxfh_key_space;
 	u16	rxfh_priv_size;
 	u32	rxfh_max_context_id;
 	u32	supported_coalesce_params;
net/ethtool/ioctl.c +36 -10

62 lines not on this page · raw message

Signature · 1 line
2.45.2
#8 lore raw

Re: [PATCH net-next 02/11] net: ethtool: let drivers declare max size of RSS indir table and key

Quoted 20 lines · Jakub Kicinski

Some drivers (bnxt but I think also mlx5 from ML discussions) change the size of the indirection table depending on the number of Rx rings. Decouple the max table size from the size of the currently used table, so that we can reserve space in the context for table growth.

Static members in ethtool_ops are good enough for now, we can add callbacks to read the max size more dynamically if someone needs that.

---
 include/linux/ethtool.h | 20 +++++++-----------
 net/ethtool/ioctl.c     | 46 ++++++++++++++++++++++++++++++++---------
 2 files changed, 44 insertions(+), 22 deletions(-)
include/linux/ethtool.h +0 -0
index 3ce5be0d168a..dc8ed93097c3 100644

include/linux/ethtool.h static inline u32 ethtool_rxfh_indir_default(u32 index, u32 n_rx_rings) · line 176

Quoted 4 lines
include/linux/ethtool.h +1 -0
@@ -173,6 +173,7 @@ static inline u32 ethtool_rxfh_indir_default(u32 index, u32 n_rx_rings)
 struct ethtool_rxfh_context {
 	u32 indir_size;
 	u32 key_size;
+	u32 key_off;

kdoc needed.
Other than that LGTM.

author maintainer
#9 lore raw

PATCH 3/11 eth: bnxt: allow deleting RSS contexts when the device is down

In reply to Jakub Kicinski

Contexts get deleted from FW when the device is down, but they are kept in SW and re-added back on open. bnxt_set_rxfh_context() apparently does not want to deal with complexity of dealing with both the device down and device up cases. This is perhaps acceptable for creating new contexts, but not being able to delete contexts makes core-driven cleanups messy. Specifically with the new RSS API core will try to delete contexts automatically after bringing the device down.

Support the delete-while-down case. Skip the FW logic and delete just the driver state.

---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c         | 10 ++++++----
 drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c |  6 ++++--
 2 files changed, 10 insertions(+), 6 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +6 -4
index 6fc34ccb86e3..ab6dae416532 100644
@@ -10195,10 +10195,12 @@ void bnxt_del_one_rss_ctx(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx,
 	struct bnxt_ntuple_filter *ntp_fltr;
 	int i;
 
-	bnxt_hwrm_vnic_free_one(bp, &rss_ctx->vnic);
-	for (i = 0; i < BNXT_MAX_CTX_PER_VNIC; i++) {
-		if (vnic->fw_rss_cos_lb_ctx[i] != INVALID_HW_RING_ID)
-			bnxt_hwrm_vnic_ctx_free_one(bp, vnic, i);
+	if (netif_running(bp->dev)) {
+		bnxt_hwrm_vnic_free_one(bp, &rss_ctx->vnic);
+		for (i = 0; i < BNXT_MAX_CTX_PER_VNIC; i++) {
+			if (vnic->fw_rss_cos_lb_ctx[i] != INVALID_HW_RING_ID)
+				bnxt_hwrm_vnic_ctx_free_one(bp, vnic, i);
+		}
 	}
 	if (!all)
 		return;
drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c +4 -2
index bf157f6cc042..0a7524cba5c3 100644
@@ -1868,6 +1868,7 @@ static int bnxt_set_rxfh_context(struct bnxt *bp,
 	struct bnxt_rss_ctx *rss_ctx;
 	struct bnxt_vnic_info *vnic;
 	bool modify = false;
+	bool delete;
 	int bit_id;
 	int rc;
 
@@ -1876,7 +1877,8 @@ static int bnxt_set_rxfh_context(struct bnxt *bp,
 		return -EOPNOTSUPP;
 	}
 
-	if (!netif_running(bp->dev)) {
+	delete = *rss_context != ETH_RXFH_CONTEXT_ALLOC && rxfh->rss_delete;
+	if (!netif_running(bp->dev) && !delete) {
 		NL_SET_ERR_MSG_MOD(extack, "Unable to set RSS contexts when interface is down");
 		return -EAGAIN;
 	}
@@ -1888,7 +1890,7 @@ static int bnxt_set_rxfh_context(struct bnxt *bp,
 					       *rss_context);
 			return -EINVAL;
 		}
-		if (*rss_context && rxfh->rss_delete) {
+		if (delete) {
 			bnxt_del_one_rss_ctx(bp, rss_ctx, true);
 			return 0;
 		}
Signature · 1 line
2.45.2
author maintainer
#10 lore raw

PATCH 4/11 eth: bnxt: move from .set_rxfh to .create_rxfh_context and friends

In reply to Jakub Kicinski

Use the new ethtool ops for RSS context management. The conversion is pretty straightforward cut / paste of the right chunks of the combined handler. Main change is that we let the core pick the IDs (bitmap will be removed separately for ease of review), so we need to tell the core when we lose a context.
Since the new API passes rxfh as const, change bnxt_modify_rss() to also take const.

---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c     |   1 +
 .../net/ethernet/broadcom/bnxt/bnxt_ethtool.c | 119 +++++++++++-------
 2 files changed, 75 insertions(+), 45 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +1 -0
index ab6dae416532..3dc43c263b91 100644
@@ -10242,6 +10242,7 @@ static void bnxt_hwrm_realloc_rss_ctx_vnic(struct bnxt *bp)
 			netdev_err(bp->dev, "Failed to restore RSS ctx %d\n",
 				   rss_ctx->index);
 			bnxt_del_one_rss_ctx(bp, rss_ctx, true);
+			ethtool_rxfh_context_lost(bp->dev, rss_ctx->index);
 		}
 	}
 }
drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c +74 -45

185 lines not on this page · raw message

Signature · 1 line
2.45.2
#11 lore raw

Re: [PATCH net-next 04/11] eth: bnxt: move from .set_rxfh to .create_rxfh_context and friends

Quoted 9 lines · Jakub Kicinski

Use the new ethtool ops for RSS context management. The conversion is pretty straightforward cut / paste of the right chunks of the combined handler. Main change is that we let the core pick the IDs (bitmap will be removed separately for ease of review), so we need to tell the core when we lose a context.
Since the new API passes rxfh as const, change bnxt_modify_rss() to also take const.

...

drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c void bnxt_ethtool_free(struct bnxt *bp) · line 5299

Quoted 5 lines
drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c +1 -0
@@ -5271,6 +5296,7 @@ void bnxt_ethtool_free(struct bnxt *bp)
 const struct ethtool_ops bnxt_ethtool_ops = {
 	.cap_link_lanes_supported	= 1,
 	.cap_rss_ctx_supported		= 1,
+	.rxfh_max_context_id		= BNXT_MAX_ETH_RSS_CTX,

According to Pavan [1], this limit only existed for the sake of the
SW side (presumably the rss_ctx_bmap), so probably it can be removed in patch #5.
The higher FW limit Pavan mentions appears to be on number rather
than index; at least I can't see anything in the driver feeding the user-facing context ID to the device. But I don't know whether FW has any opportunity to say ENOMEM, or whether the driver needs to validate against the hardware limit itself. Hopefully Pavan (CCed) can elaborate.

-ed

[1] https://lore.kernel.org/netdev/CALs4sv2dyy3uy+Xznm41M3uOkv1TSoGMwVBL5Cwzv=_E=+L_4A@mail.gmail.com/

maintainer
#12 lore raw

Re: [PATCH net-next 04/11] eth: bnxt: move from .set_rxfh to .create_rxfh_context and friends

Quoted 12 lines · Edward Cree
Quoted 9 lines · Jakub Kicinski

Use the new ethtool ops for RSS context management. The conversion is pretty straightforward cut / paste of the right chunks of the combined handler. Main change is that we let the core pick the IDs (bitmap will be removed separately for ease of review), so we need to tell the core when we lose a context.
Since the new API passes rxfh as const, change bnxt_modify_rss() to also take const.

...

Quoted 12 lines

drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c void bnxt_ethtool_free(struct bnxt *bp) · line 5299

Quoted 5 lines
drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c +1 -0
@@ -5271,6 +5296,7 @@ void bnxt_ethtool_free(struct bnxt *bp)
 const struct ethtool_ops bnxt_ethtool_ops = {
      .cap_link_lanes_supported       = 1,
      .cap_rss_ctx_supported          = 1,
+     .rxfh_max_context_id            = BNXT_MAX_ETH_RSS_CTX,

According to Pavan [1], this limit only existed for the sake of the
SW side (presumably the rss_ctx_bmap), so probably it can be removed in patch #5.
The higher FW limit Pavan mentions appears to be on number rather
than index; at least I can't see anything in the driver feeding the user-facing context ID to the device. But I don't know whether FW

Hi Ed, you are right. It's a SW side number and the device has no connection with the ID.

Quoted 3 lines

has any opportunity to say ENOMEM, or whether the driver needs to validate against the hardware limit itself. Hopefully Pavan (CCed) can elaborate.

Because the driver is not aware of the hardware limit, and the limit is dynamic, we can rely on FW to know if the resource request we made was honored (there is no direct ENOMEM mechanism)
The driver already does this when we make a runtime check for resources using bnxt_rfs_capable() when an RSS ctx is being created. But for this version of the driver, I would prefer to keep a limit because we have some FW improvements coming in, in the area of resource management.
Though removing the limit may not break anything, I'd prefer to have it removed once a FW with improvements (indicated by a query flag/caps) is available.
Michael may also add his thoughts on this.

Quoted 4 lines

-ed

[1] https://lore.kernel.org/netdev/CALs4sv2dyy3uy+Xznm41M3uOkv1TSoGMwVBL5Cwzv=_E=+L_4A@mail.gmail.com/

author maintainer
#13 lore raw

Re: [PATCH net-next 04/11] eth: bnxt: move from .set_rxfh to .create_rxfh_context and friends

Quoted 4 lines · On Wed, 3 Jul 2024 18:19:18 +0530 Pavan Chebbi wrote:
Quoted 3 lines

has any opportunity to say ENOMEM, or whether the driver needs to validate against the hardware limit itself. Hopefully Pavan (CCed) can elaborate.

Quoted 11 lines

Because the driver is not aware of the hardware limit, and the limit is dynamic, we can rely on FW to know if the resource request we made was honored (there is no direct ENOMEM mechanism)
The driver already does this when we make a runtime check for resources using bnxt_rfs_capable() when an RSS ctx is being created. But for this version of the driver, I would prefer to keep a limit because we have some FW improvements coming in, in the area of resource management.
Though removing the limit may not break anything, I'd prefer to have it removed once a FW with improvements (indicated by a query flag/caps) is available.

I like keeping the limit too, FWIW, it'd be great to give this info to user space in due course to let it make more informed decisions.

#14 lore raw

Re: [PATCH net-next 04/11] eth: bnxt: move from .set_rxfh to .create_rxfh_context and friends

In reply to Jakub Kicinski
kernel test robot noticed the following build errors:

Hi Jakub,

kernel test robot noticed the following build errors:

[auto build test ERROR on net-next/main]

url:    https://github.com/intel-lab-lkp/linux/commits/Jakub-Kicinski/net-ethtool-let-drivers-remove-lost-RSS-contexts/20240703-163816
base:   net-next/main
patch link:    https://lore.kernel.org/r/20240702234757.4188344-6-kuba%40kernel.org
patch subject: [PATCH net-next 04/11] eth: bnxt: move from .set_rxfh to .create_rxfh_context and friends
config: i386-randconfig-014-20240704 (https://download.01.org/0day-ci/archive/20240704/[email protected]/config)
compiler: clang version 18.1.5 (https://github.com/llvm/llvm-project 617a15a9eac96088ae5e9134248d8236e34b91b1)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240704/[email protected]/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

All errors (new ones prefixed by >>, old ones prefixed by <<):

WARNING: modpost: missing MODULE_DESCRIPTION() in lib/test_dynamic_debug.o
WARNING: modpost: missing MODULE_DESCRIPTION() in lib/test_printf.o
WARNING: modpost: missing MODULE_DESCRIPTION() in lib/test_bitmap.o
WARNING: modpost: missing MODULE_DESCRIPTION() in lib/test_uuid.o
WARNING: modpost: missing MODULE_DESCRIPTION() in lib/test_xarray.o
WARNING: modpost: missing MODULE_DESCRIPTION() in lib/test_blackhole_dev.o
WARNING: modpost: missing MODULE_DESCRIPTION() in lib/test_meminit.o
WARNING: modpost: missing MODULE_DESCRIPTION() in lib/test_free_pages.o
WARNING: modpost: missing MODULE_DESCRIPTION() in lib/test_objpool.o
WARNING: modpost: missing MODULE_DESCRIPTION() in lib/ts_kmp.o
WARNING: modpost: missing MODULE_DESCRIPTION() in lib/ts_bm.o
WARNING: modpost: missing MODULE_DESCRIPTION() in lib/ts_fsm.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/pinctrl/pinctrl-mcp23s08_i2c.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/pinctrl/pinctrl-mcp23s08.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/video/backlight/rt4831-backlight.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/regulator/da9121-regulator.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/regulator/max20411-regulator.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/regulator/tps6286x-regulator.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/tty/n_hdlc.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/tty/n_gsm.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/tty/ttynull.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/char/agp/amd-k7-agp.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/char/agp/amd64-agp.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/char/agp/sis-agp.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/base/regmap/regmap-i2c.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/base/regmap/regmap-spmi.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/block/brd.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/block/ublk_drv.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/mfd/arizona.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/mfd/pcf50633-gpio.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/mfd/rt4831.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/nvdimm/nd_pmem.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/nvdimm/nd_btt.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/dax/dax.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/usb/class/usbtmc.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/usb/misc/isight_firmware.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/usb/misc/yurex.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/input/matrix-keymap.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/i2c/busses/i2c-ccgx-ucsi.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/i2c/busses/i2c-ali1563.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/i2c/busses/i2c-pxa.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/media/i2c/uda1342.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/media/rc/rc-core.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/media/usb/dvb-usb/dvb-usb-dibusb-common.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/media/usb/dvb-usb/dvb-usb-dibusb-mc-common.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/media/usb/go7007/go7007.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/media/dvb-frontends/au8522_decoder.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/media/dvb-frontends/mb86a16.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/media/v4l2-core/v4l2-async.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/media/v4l2-core/v4l2-fwnode.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/media/radio/si470x/radio-si470x-common.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hwmon/corsair-cpro.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hwmon/mr75203.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/watchdog/simatic-ipc-wdt.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/cpufreq/cpufreq-dt-platdev.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/leds/flash/leds-rt4505.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/leds/simple/simatic-ipc-leds.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/leds/simple/simatic-ipc-leds-gpio-core.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/leds/simple/simatic-ipc-leds-gpio-f7188x.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/firmware/google/coreboot_table.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/firmware/google/memconsole.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/firmware/google/memconsole-coreboot.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/firmware/google/vpd-sysfs.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/crypto/atmel-sha204a.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-a4tech.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-aureal.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-cypress.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-elecom.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-evision.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-ezkey.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-gyration.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-kensington.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-lenovo.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-logitech-dj.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-magicmouse.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-megaworld.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-ntrig.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-pl.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-primax.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-redragon.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-retrode.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-samsung.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-topseed.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/hid/hid-viewsonic.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/platform/x86/siemens/simatic-ipc.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/devfreq/governor_performance.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/nvmem/nvmem_u-boot-env.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/mtd/chips/cfi_util.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/mtd/chips/cfi_cmdset_0020.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/mtd/maps/map_funcs.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/spmi/hisi-spmi-controller.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/uio/uio.o
WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/uio/uio_cif.o
WARNING: modpost: missing MODULE_DESCRIPTION() in sound/soc/amd/yc/snd-soc-acp6x-mach.o
WARNING: modpost: missing MODULE_DESCRIPTION() in sound/soc/amd/renoir/snd-acp3x-rn.o
WARNING: modpost: missing MODULE_DESCRIPTION() in samples/vfio-mdev/mtty.o
WARNING: modpost: missing MODULE_DESCRIPTION() in samples/vfio-mdev/mdpy.o
WARNING: modpost: missing MODULE_DESCRIPTION() in samples/configfs/configfs_sample.o
WARNING: modpost: missing MODULE_DESCRIPTION() in samples/kprobes/kprobe_example.o
WARNING: modpost: missing MODULE_DESCRIPTION() in samples/kprobes/kretprobe_example.o

>> ERROR: modpost: "ethtool_rxfh_context_lost" [drivers/net/ethernet/broadcom/bnxt/bnxt_en.ko] undefined!
Signature · 2 lines
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
#15 lore raw

Re: [PATCH net-next 04/11] eth: bnxt: move from .set_rxfh to .create_rxfh_context and friends

In reply to Jakub Kicinski
kernel test robot noticed the following build errors:

Hi Jakub,

kernel test robot noticed the following build errors:

[auto build test ERROR on net-next/main]

url:    https://github.com/intel-lab-lkp/linux/commits/Jakub-Kicinski/net-ethtool-let-drivers-remove-lost-RSS-contexts/20240703-163816
base:   net-next/main
patch link:    https://lore.kernel.org/r/20240702234757.4188344-6-kuba%40kernel.org
patch subject: [PATCH net-next 04/11] eth: bnxt: move from .set_rxfh to .create_rxfh_context and friends
config: csky-randconfig-r063-20240704 (https://download.01.org/0day-ci/archive/20240705/[email protected]/config)
compiler: csky-linux-gcc (GCC) 13.2.0
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240705/[email protected]/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

All errors (new ones prefixed by >>):

csky-linux-ld: drivers/net/ethernet/broadcom/bnxt/bnxt.o: in function `__bnxt_open_nic':

>> bnxt.c:(.text+0x11e32): undefined reference to `ethtool_rxfh_context_lost'
>> csky-linux-ld: bnxt.c:(.text+0x11f0c): undefined reference to `ethtool_rxfh_context_lost'
Signature · 2 lines
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
author maintainer
#16 lore raw

PATCH 5/11 eth: bnxt: remove rss_ctx_bmap

In reply to Jakub Kicinski

Core will allocate IDs for the driver, from the range [1, BNXT_MAX_ETH_RSS_CTX], no need to track the allocations.

---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c         | 13 ++-----------
 drivers/net/ethernet/broadcom/bnxt/bnxt.h         |  2 --
 drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c |  5 -----
 3 files changed, 2 insertions(+), 18 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +2 -11
index 3dc43c263b91..02aeba4b5df5 100644
@@ -10224,7 +10224,6 @@ void bnxt_del_one_rss_ctx(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx,
 	kfree(rss_ctx->rss_indir_tbl);
 	list_del(&rss_ctx->list);
 	bp->num_rss_ctx--;
-	clear_bit(rss_ctx->index, bp->rss_ctx_bmap);
 	kfree(rss_ctx);
 }
 
@@ -10266,20 +10265,12 @@ void bnxt_clear_rss_ctxs(struct bnxt *bp, bool all)
 
 	list_for_each_entry_safe(rss_ctx, tmp, &bp->rss_ctx_list, list)
 		bnxt_del_one_rss_ctx(bp, rss_ctx, all);
-
-	if (all)
-		bitmap_free(bp->rss_ctx_bmap);
 }
 
 static void bnxt_init_multi_rss_ctx(struct bnxt *bp)
 {
-	bp->rss_ctx_bmap = bitmap_zalloc(BNXT_RSS_CTX_BMAP_LEN, GFP_KERNEL);
-	if (bp->rss_ctx_bmap) {
-		/* burn index 0 since we cannot have context 0 */
-		__set_bit(0, bp->rss_ctx_bmap);
-		INIT_LIST_HEAD(&bp->rss_ctx_list);
-		bp->rss_cap |= BNXT_RSS_CAP_MULTI_RSS_CTX;
-	}
+	INIT_LIST_HEAD(&bp->rss_ctx_list);
+	bp->rss_cap |= BNXT_RSS_CAP_MULTI_RSS_CTX;
 }
 
 /* Allow PF, trusted VFs and VFs with default VLAN to be in promiscuous mode */
drivers/net/ethernet/broadcom/bnxt/bnxt.h +0 -2
index e46bd11e52b0..f4365a840e3a 100644
@@ -1298,7 +1298,6 @@ struct bnxt_rss_ctx {
 };
 
 #define BNXT_MAX_ETH_RSS_CTX	32
-#define BNXT_RSS_CTX_BMAP_LEN	(BNXT_MAX_ETH_RSS_CTX + 1)
 #define BNXT_VNIC_ID_INVALID	0xffffffff
 
 struct bnxt_hw_rings {
@@ -2332,7 +2331,6 @@ struct bnxt {
 	struct bnxt_ring_grp_info	*grp_info;
 	struct bnxt_vnic_info	*vnic_info;
 	struct list_head	rss_ctx_list;
-	unsigned long		*rss_ctx_bmap;
 	u32			num_rss_ctx;
 	int			nr_vnics;
 	u16			*rss_indir_tbl;
drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c +0 -5
index a7f71ebca2fe..48f8e14685a1 100644
@@ -1896,11 +1896,6 @@ static int bnxt_create_rxfh_context(struct net_device *dev,
 		return -EINVAL;
 	}
 
-	if (test_and_set_bit(rxfh->rss_context, bp->rss_ctx_bmap)) {
-		NL_SET_ERR_MSG_MOD(extack, "Context ID conflict");
-		return -EINVAL;
-	}
-
 	if (!bnxt_rfs_capable(bp, true)) {
 		NL_SET_ERR_MSG_MOD(extack, "Out hardware resources");
 		return -ENOMEM;
Signature · 1 line
2.45.2
author maintainer
#17 lore raw

PATCH 6/11 eth: bnxt: depend on core cleaning up RSS contexts

In reply to Jakub Kicinski

New RSS context API removes old contexts on netdev unregister. No need to wipe them manually.

---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c         | 11 ++++-------
 drivers/net/ethernet/broadcom/bnxt/bnxt.h         |  2 +-
 drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c |  2 +-
 3 files changed, 6 insertions(+), 9 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +4 -7
index 02aeba4b5df5..b6915261c15d 100644
@@ -10259,12 +10259,12 @@ struct bnxt_rss_ctx *bnxt_alloc_rss_ctx(struct bnxt *bp)
 	return rss_ctx;
 }
 
-void bnxt_clear_rss_ctxs(struct bnxt *bp, bool all)
+void bnxt_clear_rss_ctxs(struct bnxt *bp)
 {
 	struct bnxt_rss_ctx *rss_ctx, *tmp;
 
 	list_for_each_entry_safe(rss_ctx, tmp, &bp->rss_ctx_list, list)
-		bnxt_del_one_rss_ctx(bp, rss_ctx, all);
+		bnxt_del_one_rss_ctx(bp, rss_ctx, false);
 }
 
 static void bnxt_init_multi_rss_ctx(struct bnxt *bp)
@@ -12316,7 +12316,7 @@ static void __bnxt_close_nic(struct bnxt *bp, bool irq_re_init,
 		msleep(20);
 
 	if (BNXT_SUPPORTS_MULTI_RSS_CTX(bp))
-		bnxt_clear_rss_ctxs(bp, false);
+		bnxt_clear_rss_ctxs(bp);
 	/* Flush rings and disable interrupts */
 	bnxt_shutdown_nic(bp, irq_re_init);
 
@@ -15227,8 +15227,7 @@ static void bnxt_remove_one(struct pci_dev *pdev)
 
 	bnxt_free_l2_filters(bp, true);
 	bnxt_free_ntp_fltrs(bp, true);
-	if (BNXT_SUPPORTS_MULTI_RSS_CTX(bp))
-		bnxt_clear_rss_ctxs(bp, true);
+	WARN_ON(bp->num_rss_ctx);
 	clear_bit(BNXT_STATE_IN_FW_RESET, &bp->state);
 	/* Flush any pending tasks */
 	cancel_work_sync(&bp->sp_task);
@@ -15879,8 +15878,6 @@ static int bnxt_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 	bnxt_clear_int_mode(bp);
 
 init_err_pci_clean:
-	if (BNXT_SUPPORTS_MULTI_RSS_CTX(bp))
-		bnxt_clear_rss_ctxs(bp, true);
 	bnxt_hwrm_func_drv_unrgtr(bp);
 	bnxt_free_hwrm_resources(bp);
 	bnxt_hwmon_uninit(bp);
drivers/net/ethernet/broadcom/bnxt/bnxt.h +1 -1
index f4365a840e3a..04c4ff7b9052 100644
@@ -2846,7 +2846,7 @@ int __bnxt_setup_vnic_p5(struct bnxt *bp, struct bnxt_vnic_info *vnic);
 void bnxt_del_one_rss_ctx(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx,
 			  bool all);
 struct bnxt_rss_ctx *bnxt_alloc_rss_ctx(struct bnxt *bp);
-void bnxt_clear_rss_ctxs(struct bnxt *bp, bool all);
+void bnxt_clear_rss_ctxs(struct bnxt *bp);
 int bnxt_open_nic(struct bnxt *, bool, bool);
 int bnxt_half_open_nic(struct bnxt *bp);
 void bnxt_half_close_nic(struct bnxt *bp);
drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c +1 -1
index 48f8e14685a1..397aedad3d4f 100644
@@ -970,7 +970,7 @@ static int bnxt_set_channels(struct net_device *dev,
 
 	bnxt_clear_usr_fltrs(bp, true);
 	if (BNXT_SUPPORTS_MULTI_RSS_CTX(bp))
-		bnxt_clear_rss_ctxs(bp, false);
+		bnxt_clear_rss_ctxs(bp);
 	if (netif_running(dev)) {
 		if (BNXT_PF(bp)) {
 			/* TODO CHIMP_FW: Send message to all VF's
Signature · 1 line
2.45.2
author maintainer
#18 lore raw

PATCH 7/11 eth: bnxt: use context priv for struct bnxt_rss_ctx

In reply to Jakub Kicinski

Core can allocate space for per-context driver-private data, use it for struct bnxt_rss_ctx. Inline bnxt_alloc_rss_ctx() at this point, most of the init (as in the actions bnxt_del_one_rss_ctx() will undo) is open coded in bnxt_create_rxfh_context(), anyway.

---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c     | 14 -----------
 drivers/net/ethernet/broadcom/bnxt/bnxt.h     |  1 -
 .../net/ethernet/broadcom/bnxt/bnxt_ethtool.c | 24 +++++++------------
 3 files changed, 9 insertions(+), 30 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +0 -14
index b6915261c15d..39876feae1a4 100644
@@ -10224,7 +10224,6 @@ void bnxt_del_one_rss_ctx(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx,
 	kfree(rss_ctx->rss_indir_tbl);
 	list_del(&rss_ctx->list);
 	bp->num_rss_ctx--;
-	kfree(rss_ctx);
 }
 
 static void bnxt_hwrm_realloc_rss_ctx_vnic(struct bnxt *bp)
@@ -10246,19 +10245,6 @@ static void bnxt_hwrm_realloc_rss_ctx_vnic(struct bnxt *bp)
 	}
 }
 
-struct bnxt_rss_ctx *bnxt_alloc_rss_ctx(struct bnxt *bp)
-{
-	struct bnxt_rss_ctx *rss_ctx = NULL;
-
-	rss_ctx = kzalloc(sizeof(*rss_ctx), GFP_KERNEL);
-	if (rss_ctx) {
-		rss_ctx->vnic.rss_ctx = rss_ctx;
-		list_add_tail(&rss_ctx->list, &bp->rss_ctx_list);
-		bp->num_rss_ctx++;
-	}
-	return rss_ctx;
-}
-
 void bnxt_clear_rss_ctxs(struct bnxt *bp)
 {
 	struct bnxt_rss_ctx *rss_ctx, *tmp;
drivers/net/ethernet/broadcom/bnxt/bnxt.h +0 -1
index 04c4ff7b9052..21c3296cf6d9 100644
@@ -2845,7 +2845,6 @@ int bnxt_hwrm_vnic_rss_cfg_p5(struct bnxt *bp, struct bnxt_vnic_info *vnic);
 int __bnxt_setup_vnic_p5(struct bnxt *bp, struct bnxt_vnic_info *vnic);
 void bnxt_del_one_rss_ctx(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx,
 			  bool all);
-struct bnxt_rss_ctx *bnxt_alloc_rss_ctx(struct bnxt *bp);
 void bnxt_clear_rss_ctxs(struct bnxt *bp);
 int bnxt_open_nic(struct bnxt *, bool, bool);
 int bnxt_half_open_nic(struct bnxt *bp);
drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c +9 -15
index 397aedad3d4f..2e6e060e2b44 100644
@@ -1901,11 +1901,13 @@ static int bnxt_create_rxfh_context(struct net_device *dev,
 		return -ENOMEM;
 	}
 
-	rss_ctx = bnxt_alloc_rss_ctx(bp);
-	if (!rss_ctx)
-		return -ENOMEM;
+	rss_ctx = ethtool_rxfh_context_priv(ctx);
+
+	list_add_tail(&rss_ctx->list, &bp->rss_ctx_list);
+	bp->num_rss_ctx++;
 
 	vnic = &rss_ctx->vnic;
+	vnic->rss_ctx = rss_ctx;
 	vnic->flags |= BNXT_VNIC_RSSCTX_FLAG;
 	vnic->vnic_id = BNXT_VNIC_ID_INVALID;
 	rc = bnxt_alloc_rss_ctx_rss_table(bp, rss_ctx);
@@ -1958,12 +1960,7 @@ static int bnxt_modify_rxfh_context(struct net_device *dev,
 	if (rc)
 		return rc;
 
-	rss_ctx = bnxt_get_rss_ctx_from_index(bp, rxfh->rss_context);
-	if (!rss_ctx) {
-		NL_SET_ERR_MSG_FMT_MOD(extack, "RSS context %u not found",
-				       rxfh->rss_context);
-		return -EINVAL;
-	}
+	rss_ctx = ethtool_rxfh_context_priv(ctx);
 
 	bnxt_modify_rss(bp, rss_ctx, rxfh);
 
@@ -1978,12 +1975,7 @@ static int bnxt_remove_rxfh_context(struct net_device *dev,
 	struct bnxt *bp = netdev_priv(dev);
 	struct bnxt_rss_ctx *rss_ctx;
 
-	rss_ctx = bnxt_get_rss_ctx_from_index(bp, rss_context);
-	if (!rss_ctx) {
-		NL_SET_ERR_MSG_FMT_MOD(extack, "RSS context %u not found",
-				       rss_context);
-		return -EINVAL;
-	}
+	rss_ctx = ethtool_rxfh_context_priv(ctx);
 
 	bnxt_del_one_rss_ctx(bp, rss_ctx, true);
 	return 0;
@@ -5292,6 +5284,8 @@ const struct ethtool_ops bnxt_ethtool_ops = {
 	.cap_link_lanes_supported	= 1,
 	.cap_rss_ctx_supported		= 1,
 	.rxfh_max_context_id		= BNXT_MAX_ETH_RSS_CTX,
+	.rxfh_indir_space		= BNXT_MAX_RSS_TABLE_ENTRIES_P5,
+	.rxfh_priv_size			= sizeof(struct bnxt_rss_ctx),
 	.supported_coalesce_params = ETHTOOL_COALESCE_USECS |
 				     ETHTOOL_COALESCE_MAX_FRAMES |
 				     ETHTOOL_COALESCE_USECS_IRQ |
Signature · 1 line
2.45.2
author maintainer
#19 lore raw

PATCH 8/11 eth: bnxt: use the RSS context XArray instead of the local list

In reply to Jakub Kicinski

Core already maintains all RSS contexts in an XArray, no need to keep a second list in the driver.

---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c     | 41 ++++++++++---------
 drivers/net/ethernet/broadcom/bnxt/bnxt.h     |  2 -
 .../net/ethernet/broadcom/bnxt/bnxt_ethtool.c | 11 +++--
 3 files changed, 26 insertions(+), 28 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +21 -20
index 39876feae1a4..6b8966d3ecb6 100644
@@ -5970,17 +5970,20 @@ bnxt_cfg_rfs_ring_tbl_idx(struct bnxt *bp,
 			  struct hwrm_cfa_ntuple_filter_alloc_input *req,
 			  struct bnxt_ntuple_filter *fltr)
 {
-	struct bnxt_rss_ctx *rss_ctx, *tmp;
 	u16 rxq = fltr->base.rxq;
 
 	if (fltr->base.flags & BNXT_ACT_RSS_CTX) {
-		list_for_each_entry_safe(rss_ctx, tmp, &bp->rss_ctx_list, list) {
-			if (rss_ctx->index == fltr->base.fw_vnic_id) {
-				struct bnxt_vnic_info *vnic = &rss_ctx->vnic;
+		struct ethtool_rxfh_context *ctx;
+		struct bnxt_rss_ctx *rss_ctx;
+		struct bnxt_vnic_info *vnic;
 
-				req->dst_id = cpu_to_le16(vnic->fw_vnic_id);
-				break;
-			}
+		ctx = xa_load(&bp->dev->ethtool->rss_ctx,
+			      fltr->base.fw_vnic_id);
+		if (ctx) {
+			rss_ctx = ethtool_rxfh_context_priv(ctx);
+			vnic = &rss_ctx->vnic;
+
+			req->dst_id = cpu_to_le16(vnic->fw_vnic_id);
 		}
 		return;
 	}
@@ -10222,16 +10225,17 @@ void bnxt_del_one_rss_ctx(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx,
 				  vnic->rss_table,
 				  vnic->rss_table_dma_addr);
 	kfree(rss_ctx->rss_indir_tbl);
-	list_del(&rss_ctx->list);
 	bp->num_rss_ctx--;
 }
 
 static void bnxt_hwrm_realloc_rss_ctx_vnic(struct bnxt *bp)
 {
 	bool set_tpa = !!(bp->flags & BNXT_FLAG_TPA);
-	struct bnxt_rss_ctx *rss_ctx, *tmp;
+	struct ethtool_rxfh_context *ctx;
+	unsigned long context;
 
-	list_for_each_entry_safe(rss_ctx, tmp, &bp->rss_ctx_list, list) {
+	xa_for_each(&bp->dev->ethtool->rss_ctx, context, ctx) {
+		struct bnxt_rss_ctx *rss_ctx = ethtool_rxfh_context_priv(ctx);
 		struct bnxt_vnic_info *vnic = &rss_ctx->vnic;
 
 		if (bnxt_hwrm_vnic_alloc(bp, vnic, 0, bp->rx_nr_rings) ||
@@ -10247,16 +10251,14 @@ static void bnxt_hwrm_realloc_rss_ctx_vnic(struct bnxt *bp)
 
 void bnxt_clear_rss_ctxs(struct bnxt *bp)
 {
-	struct bnxt_rss_ctx *rss_ctx, *tmp;
+	struct ethtool_rxfh_context *ctx;
+	unsigned long context;
+
+	xa_for_each(&bp->dev->ethtool->rss_ctx, context, ctx) {
+		struct bnxt_rss_ctx *rss_ctx = ethtool_rxfh_context_priv(ctx);
 
-	list_for_each_entry_safe(rss_ctx, tmp, &bp->rss_ctx_list, list)
 		bnxt_del_one_rss_ctx(bp, rss_ctx, false);
-}
-
-static void bnxt_init_multi_rss_ctx(struct bnxt *bp)
-{
-	INIT_LIST_HEAD(&bp->rss_ctx_list);
-	bp->rss_cap |= BNXT_RSS_CAP_MULTI_RSS_CTX;
+	}
 }
 
 /* Allow PF, trusted VFs and VFs with default VLAN to be in promiscuous mode */
@@ -15840,8 +15842,7 @@ static int bnxt_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 	INIT_LIST_HEAD(&bp->usr_fltr_list);
 
 	if (BNXT_SUPPORTS_NTUPLE_VNIC(bp))
-		bnxt_init_multi_rss_ctx(bp);
-
+		bp->rss_cap |= BNXT_RSS_CAP_MULTI_RSS_CTX;
 
 	rc = register_netdev(dev);
 	if (rc)
drivers/net/ethernet/broadcom/bnxt/bnxt.h +0 -2
index 21c3296cf6d9..be40e0513777 100644
@@ -1291,7 +1291,6 @@ struct bnxt_vnic_info {
 };
 
 struct bnxt_rss_ctx {
-	struct list_head list;
 	struct bnxt_vnic_info vnic;
 	u16	*rss_indir_tbl;
 	u8	index;
@@ -2330,7 +2329,6 @@ struct bnxt {
 	/* grp_info indexed by completion ring index */
 	struct bnxt_ring_grp_info	*grp_info;
 	struct bnxt_vnic_info	*vnic_info;
-	struct list_head	rss_ctx_list;
 	u32			num_rss_ctx;
 	int			nr_vnics;
 	u16			*rss_indir_tbl;
drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c +5 -6
index 2e6e060e2b44..74765583405b 100644
@@ -1210,12 +1210,12 @@ static int bnxt_grxclsrule(struct bnxt *bp, struct ethtool_rxnfc *cmd)
 static struct bnxt_rss_ctx *bnxt_get_rss_ctx_from_index(struct bnxt *bp,
 							u32 index)
 {
-	struct bnxt_rss_ctx *rss_ctx, *tmp;
+	struct ethtool_rxfh_context *ctx;
 
-	list_for_each_entry_safe(rss_ctx, tmp, &bp->rss_ctx_list, list)
-		if (rss_ctx->index == index)
-			return rss_ctx;
-	return NULL;
+	ctx = xa_load(&bp->dev->ethtool->rss_ctx, index);
+	if (!ctx)
+		return NULL;
+	return ethtool_rxfh_context_priv(ctx);
 }
 
 static int bnxt_alloc_rss_ctx_rss_table(struct bnxt *bp,
@@ -1903,7 +1903,6 @@ static int bnxt_create_rxfh_context(struct net_device *dev,
 
 	rss_ctx = ethtool_rxfh_context_priv(ctx);
 
-	list_add_tail(&rss_ctx->list, &bp->rss_ctx_list);
 	bp->num_rss_ctx++;
 
 	vnic = &rss_ctx->vnic;
Signature · 1 line
2.45.2
author maintainer
#20 lore raw

PATCH 9/11 eth: bnxt: bump the entry size in indir tables to u32

In reply to Jakub Kicinski

Ethtool core stores indirection table with u32 entries, "just to be safe". Switch the type in the driver, so that it's easier to swap local tables for the core ones.

---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c         | 8 ++++----
 drivers/net/ethernet/broadcom/bnxt/bnxt.h         | 4 ++--
 drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c | 6 +++---
 3 files changed, 9 insertions(+), 9 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +4 -4
index 6b8966d3ecb6..4176459921b5 100644
@@ -6225,7 +6225,7 @@ static u16 bnxt_cp_ring_for_tx(struct bnxt *bp, struct bnxt_tx_ring_info *txr)
 int bnxt_alloc_rss_indir_tbl(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx)
 {
 	int entries;
-	u16 *tbl;
+	u32 *tbl;
 
 	if (bp->flags & BNXT_FLAG_CHIP_P5_PLUS)
 		entries = BNXT_MAX_RSS_TABLE_ENTRIES_P5;
@@ -6248,7 +6248,7 @@ int bnxt_alloc_rss_indir_tbl(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx)
 void bnxt_set_dflt_rss_indir_tbl(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx)
 {
 	u16 max_rings, max_entries, pad, i;
-	u16 *rss_indir_tbl;
+	u32 *rss_indir_tbl;
 
 	if (!bp->rx_nr_rings)
 		return;
@@ -6269,12 +6269,12 @@ void bnxt_set_dflt_rss_indir_tbl(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx)
 
 	pad = bp->rss_indir_tbl_entries - max_entries;
 	if (pad)
-		memset(&rss_indir_tbl[i], 0, pad * sizeof(u16));
+		memset(&rss_indir_tbl[i], 0, pad * sizeof(u32));
 }
 
 static u16 bnxt_get_max_rss_ring(struct bnxt *bp)
 {
-	u16 i, tbl_size, max_ring = 0;
+	u32 i, tbl_size, max_ring = 0;
 
 	if (!bp->rss_indir_tbl)
 		return 0;
drivers/net/ethernet/broadcom/bnxt/bnxt.h +2 -2
index be40e0513777..1a33824a32a8 100644
@@ -1292,7 +1292,7 @@ struct bnxt_vnic_info {
 
 struct bnxt_rss_ctx {
 	struct bnxt_vnic_info vnic;
-	u16	*rss_indir_tbl;
+	u32	*rss_indir_tbl;
 	u8	index;
 };
 
@@ -2331,7 +2331,7 @@ struct bnxt {
 	struct bnxt_vnic_info	*vnic_info;
 	u32			num_rss_ctx;
 	int			nr_vnics;
-	u16			*rss_indir_tbl;
+	u32			*rss_indir_tbl;
 	u16			rss_indir_tbl_entries;
 	u32			rss_hash_cfg;
 	u32			rss_hash_delta;
drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c +3 -3
index 74765583405b..e5f687d4a455 100644
@@ -1804,7 +1804,7 @@ static int bnxt_get_rxfh(struct net_device *dev,
 	u32 rss_context = rxfh->rss_context;
 	struct bnxt_rss_ctx *rss_ctx = NULL;
 	struct bnxt *bp = netdev_priv(dev);
-	u16 *indir_tbl = bp->rss_indir_tbl;
+	u32 *indir_tbl = bp->rss_indir_tbl;
 	struct bnxt_vnic_info *vnic;
 	u32 i, tbl_size;
 
@@ -1848,7 +1848,7 @@ static void bnxt_modify_rss(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx,
 	}
 	if (rxfh->indir) {
 		u32 i, pad, tbl_size = bnxt_get_rxfh_indir_size(bp->dev);
-		u16 *indir_tbl = bp->rss_indir_tbl;
+		u32 *indir_tbl = bp->rss_indir_tbl;
 
 		if (rss_ctx)
 			indir_tbl = rss_ctx->rss_indir_tbl;
@@ -1856,7 +1856,7 @@ static void bnxt_modify_rss(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx,
 			indir_tbl[i] = rxfh->indir[i];
 		pad = bp->rss_indir_tbl_entries - tbl_size;
 		if (pad)
-			memset(&bp->rss_indir_tbl[i], 0, pad * sizeof(u16));
+			memset(&bp->rss_indir_tbl[i], 0, pad * sizeof(u32));
 	}
 }
 
Signature · 1 line
2.45.2
Reviewed-by on v1 9/11
#21 lore raw

Re: [PATCH net-next 09/11] eth: bnxt: bump the entry size in indir tables to u32

Quoted 16 lines · Jakub Kicinski

Ethtool core stores indirection table with u32 entries, "just to be safe". Switch the type in the driver, so that it's easier to swap local tables for the core ones.

---
  drivers/net/ethernet/broadcom/bnxt/bnxt.c         | 8 ++++----
  drivers/net/ethernet/broadcom/bnxt/bnxt.h         | 4 ++--
  drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c | 6 +++---
  3 files changed, 9 insertions(+), 9 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +0 -0
index 6b8966d3ecb6..4176459921b5 100644

drivers/net/ethernet/broadcom/bnxt/bnxt.c static u16 bnxt_cp_ring_for_tx(struct bnxt *bp, struct bnxt_tx_ring_info *txr) · line 6231

Quoted 8 lines
drivers/net/ethernet/broadcom/bnxt/bnxt.c +1 -1
@@ -6225,7 +6225,7 @@ static u16 bnxt_cp_ring_for_tx(struct bnxt *bp, struct bnxt_tx_ring_info *txr)
  int bnxt_alloc_rss_indir_tbl(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx)
  {
  	int entries;
-	u16 *tbl;
+	u32 *tbl;
 
  	if (bp->flags & BNXT_FLAG_CHIP_P5_PLUS)
  		entries = BNXT_MAX_RSS_TABLE_ENTRIES_P5;

it's a shame that default git context is so small, a few lines below we have tbl alloc:
tbl = kmalloc_array(entries, sizeof(*bp->rss_indir_tbl), GFP_KERNEL);

Quoted 10 lines
drivers/net/ethernet/broadcom/bnxt/bnxt.c +1 -1
@@ -6248,7 +6248,7 @@ int bnxt_alloc_rss_indir_tbl(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx)
  void bnxt_set_dflt_rss_indir_tbl(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx)
  {
  	u16 max_rings, max_entries, pad, i;
-	u16 *rss_indir_tbl;
+	u32 *rss_indir_tbl;
 
  	if (!bp->rx_nr_rings)
  		return;

drivers/net/ethernet/broadcom/bnxt/bnxt.c void bnxt_set_dflt_rss_indir_tbl(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx) · line 6273

Quoted 6 lines
drivers/net/ethernet/broadcom/bnxt/bnxt.c +1 -1
@@ -6269,12 +6269,12 @@ void bnxt_set_dflt_rss_indir_tbl(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx)
 
  	pad = bp->rss_indir_tbl_entries - max_entries;
  	if (pad)
-		memset(&rss_indir_tbl[i], 0, pad * sizeof(u16));
+		memset(&rss_indir_tbl[i], 0, pad * sizeof(u32));
  }

without the above allocation line (perhaps you could mention necessary bit of that in commit msg in case of v2) this patch could not be proved correct :)

with due diligence done:

OTOH, I assume we need this in-driver table only to keep it over the up→down→up cycle. Could we just keep it as inactive in the core? (And xa_mark() it as inactive to avoid reporting to the user or any other actions that we want to avoid)

PS. patch2 is duplicated for me

author maintainer
#22 lore raw

Re: [PATCH net-next 09/11] eth: bnxt: bump the entry size in indir tables to u32

Quoted 4 lines · On Wed, 3 Jul 2024 12:51:58 +0200 Przemek Kitszel wrote:

OTOH, I assume we need this in-driver table only to keep it over the up→down→up cycle. Could we just keep it as inactive in the core? (And xa_mark() it as inactive to avoid reporting to the user or any other actions that we want to avoid)

Do you mean keep the table for the default context / context 0 in the core as well?

#23 lore raw

Re: [PATCH net-next 09/11] eth: bnxt: bump the entry size in indir tables to u32

Quoted 9 lines · Jakub Kicinski
Quoted 4 lines · On Wed, 3 Jul 2024 12:51:58 +0200 Przemek Kitszel wrote:

OTOH, I assume we need this in-driver table only to keep it over the up→down→up cycle. Could we just keep it as inactive in the core? (And xa_mark() it as inactive to avoid reporting to the user or any other actions that we want to avoid)

Do you mean keep the table for the default context / context 0 in the core as well?

it was not obvious for me that this is only about the default one, but that would yield more uniformity and cleanups I guess

author maintainer
#24 lore raw

Re: [PATCH net-next 09/11] eth: bnxt: bump the entry size in indir tables to u32

Quoted 7 lines · On Wed, 3 Jul 2024 16:02:38 +0200 Przemek Kitszel wrote:
Quoted 6 lines · Jakub Kicinski
Quoted 4 lines · On Wed, 3 Jul 2024 12:51:58 +0200 Przemek Kitszel wrote:

OTOH, I assume we need this in-driver table only to keep it over the up→down→up cycle. Could we just keep it as inactive in the core? (And xa_mark() it as inactive to avoid reporting to the user or any other actions that we want to avoid)

Quoted 5 lines
Quoted 2 lines

Do you mean keep the table for the default context / context 0 in the core as well?

it was not obvious for me that this is only about the default one, but that would yield more uniformity and cleanups I guess

I'll add a note to the commit message. Initially I was hoping to add context 0, too, but there's a number of gaps. We will hopefully converge to that over time.

author maintainer
#25 lore raw

PATCH 10/11 eth: bnxt: use the indir table from ethtool context

In reply to Jakub Kicinski

Instead of allocating a separate indir table in the vnic use the one already present in the RSS context allocated by the core. This doesn't save much LoC but we won't have to worry about syncing the local version back to the core, once core learns how to dump contexts.

Add ethtool_rxfh_priv_context() for converting from priv pointer to the context. The cast is a bit ugly (understatement) and some driver paths make carrying the context pointer in addition to driver priv pointer quite tedious.

---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c     | 27 +++++++--------
 drivers/net/ethernet/broadcom/bnxt/bnxt.h     |  5 ++-
 .../net/ethernet/broadcom/bnxt/bnxt_ethtool.c | 34 +++++++++----------
 include/linux/ethtool.h                       |  5 +++
 4 files changed, 35 insertions(+), 36 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +12 -15
index 4176459921b5..8ee57b07ffb6 100644
@@ -6222,10 +6222,9 @@ static u16 bnxt_cp_ring_for_tx(struct bnxt *bp, struct bnxt_tx_ring_info *txr)
 		return bnxt_cp_ring_from_grp(bp, &txr->tx_ring_struct);
 }
 
-int bnxt_alloc_rss_indir_tbl(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx)
+static int bnxt_alloc_rss_indir_tbl(struct bnxt *bp)
 {
 	int entries;
-	u32 *tbl;
 
 	if (bp->flags & BNXT_FLAG_CHIP_P5_PLUS)
 		entries = BNXT_MAX_RSS_TABLE_ENTRIES_P5;
@@ -6233,19 +6232,16 @@ int bnxt_alloc_rss_indir_tbl(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx)
 		entries = HW_HASH_INDEX_SIZE;
 
 	bp->rss_indir_tbl_entries = entries;
-	tbl = kmalloc_array(entries, sizeof(*bp->rss_indir_tbl), GFP_KERNEL);
-	if (!tbl)
+	bp->rss_indir_tbl =
+		kmalloc_array(entries, sizeof(*bp->rss_indir_tbl), GFP_KERNEL);
+	if (!bp->rss_indir_tbl)
 		return -ENOMEM;
 
-	if (rss_ctx)
-		rss_ctx->rss_indir_tbl = tbl;
-	else
-		bp->rss_indir_tbl = tbl;
-
 	return 0;
 }
 
-void bnxt_set_dflt_rss_indir_tbl(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx)
+void bnxt_set_dflt_rss_indir_tbl(struct bnxt *bp,
+				 struct ethtool_rxfh_context *ctx)
 {
 	u16 max_rings, max_entries, pad, i;
 	u32 *rss_indir_tbl;
@@ -6259,8 +6255,8 @@ void bnxt_set_dflt_rss_indir_tbl(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx)
 		max_rings = bp->rx_nr_rings;
 
 	max_entries = bnxt_get_rxfh_indir_size(bp->dev);
-	if (rss_ctx)
-		rss_indir_tbl = &rss_ctx->rss_indir_tbl[0];
+	if (ctx)
+		rss_indir_tbl = ethtool_rxfh_context_indir(ctx);
 	else
 		rss_indir_tbl = &bp->rss_indir_tbl[0];
 
@@ -6315,10 +6311,12 @@ static void bnxt_fill_hw_rss_tbl_p5(struct bnxt *bp,
 				    struct bnxt_vnic_info *vnic)
 {
 	__le16 *ring_tbl = vnic->rss_table;
+	struct ethtool_rxfh_context *ctx;
 	struct bnxt_rx_ring_info *rxr;
 	u16 tbl_size, i;
 
 	tbl_size = bnxt_get_rxfh_indir_size(bp->dev);
+	ctx = ethtool_rxfh_priv_context(vnic->rss_ctx);
 
 	for (i = 0; i < tbl_size; i++) {
 		u16 ring_id, j;
@@ -6326,7 +6324,7 @@ static void bnxt_fill_hw_rss_tbl_p5(struct bnxt *bp,
 		if (vnic->flags & BNXT_VNIC_NTUPLE_FLAG)
 			j = ethtool_rxfh_indir_default(i, bp->rx_nr_rings);
 		else if (vnic->flags & BNXT_VNIC_RSSCTX_FLAG)
-			j = vnic->rss_ctx->rss_indir_tbl[i];
+			j = ethtool_rxfh_context_indir(ctx)[i];
 		else
 			j = bp->rss_indir_tbl[i];
 		rxr = &bp->rx_ring[j];
@@ -10224,7 +10222,6 @@ void bnxt_del_one_rss_ctx(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx,
 		dma_free_coherent(&bp->pdev->dev, vnic->rss_table_size,
 				  vnic->rss_table,
 				  vnic->rss_table_dma_addr);
-	kfree(rss_ctx->rss_indir_tbl);
 	bp->num_rss_ctx--;
 }
 
@@ -15685,7 +15682,7 @@ static int bnxt_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 			bp->flags |= BNXT_FLAG_CHIP_P7;
 	}
 
-	rc = bnxt_alloc_rss_indir_tbl(bp, NULL);
+	rc = bnxt_alloc_rss_indir_tbl(bp);
 	if (rc)
 		goto init_err_pci_clean;
 
drivers/net/ethernet/broadcom/bnxt/bnxt.h +2 -3
index 1a33824a32a8..a4420db55de9 100644
@@ -1292,7 +1292,6 @@ struct bnxt_vnic_info {
 
 struct bnxt_rss_ctx {
 	struct bnxt_vnic_info vnic;
-	u32	*rss_indir_tbl;
 	u8	index;
 };
 
@@ -2808,8 +2807,8 @@ int bnxt_hwrm_cfa_ntuple_filter_alloc(struct bnxt *bp,
 int bnxt_hwrm_vnic_set_tpa(struct bnxt *bp, struct bnxt_vnic_info *vnic,
 			   u32 tpa_flags);
 void bnxt_fill_ipv6_mask(__be32 mask[4]);
-int bnxt_alloc_rss_indir_tbl(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx);
-void bnxt_set_dflt_rss_indir_tbl(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx);
+void bnxt_set_dflt_rss_indir_tbl(struct bnxt *bp,
+				 struct ethtool_rxfh_context *ctx);
 int bnxt_get_nr_rss_ctxs(struct bnxt *bp, int rx_rings);
 int bnxt_hwrm_vnic_cfg(struct bnxt *bp, struct bnxt_vnic_info *vnic);
 int bnxt_hwrm_vnic_alloc(struct bnxt *bp, struct bnxt_vnic_info *vnic,
drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c +16 -18
index e5f687d4a455..13e9b3b26f09 100644
@@ -1218,11 +1218,10 @@ static struct bnxt_rss_ctx *bnxt_get_rss_ctx_from_index(struct bnxt *bp,
 	return ethtool_rxfh_context_priv(ctx);
 }
 
-static int bnxt_alloc_rss_ctx_rss_table(struct bnxt *bp,
-					struct bnxt_rss_ctx *rss_ctx)
+static int bnxt_alloc_vnic_rss_table(struct bnxt *bp,
+				     struct bnxt_vnic_info *vnic)
 {
 	int size = L1_CACHE_ALIGN(BNXT_MAX_RSS_TABLE_SIZE_P5);
-	struct bnxt_vnic_info *vnic = &rss_ctx->vnic;
 
 	vnic->rss_table_size = size + HW_HASH_KEY_SIZE;
 	vnic->rss_table = dma_alloc_coherent(&bp->pdev->dev,
@@ -1801,7 +1800,6 @@ static u32 bnxt_get_rxfh_key_size(struct net_device *dev)
 static int bnxt_get_rxfh(struct net_device *dev,
 			 struct ethtool_rxfh_param *rxfh)
 {
-	u32 rss_context = rxfh->rss_context;
 	struct bnxt_rss_ctx *rss_ctx = NULL;
 	struct bnxt *bp = netdev_priv(dev);
 	u32 *indir_tbl = bp->rss_indir_tbl;
@@ -1815,10 +1813,13 @@ static int bnxt_get_rxfh(struct net_device *dev,
 
 	vnic = &bp->vnic_info[BNXT_VNIC_DEFAULT];
 	if (rxfh->rss_context) {
-		rss_ctx = bnxt_get_rss_ctx_from_index(bp, rss_context);
-		if (!rss_ctx)
+		struct ethtool_rxfh_context *ctx;
+
+		ctx = xa_load(&bp->dev->ethtool->rss_ctx, rxfh->rss_context);
+		if (!ctx)
 			return -EINVAL;
-		indir_tbl = rss_ctx->rss_indir_tbl;
+		indir_tbl = ethtool_rxfh_context_indir(ctx);
+		rss_ctx = ethtool_rxfh_context_priv(ctx);
 		vnic = &rss_ctx->vnic;
 	}
 
@@ -1834,7 +1835,8 @@ static int bnxt_get_rxfh(struct net_device *dev,
 	return 0;
 }
 
-static void bnxt_modify_rss(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx,
+static void bnxt_modify_rss(struct bnxt *bp, struct ethtool_rxfh_context *ctx,
+			    struct bnxt_rss_ctx *rss_ctx,
 			    const struct ethtool_rxfh_param *rxfh)
 {
 	if (rxfh->key) {
@@ -1851,7 +1853,7 @@ static void bnxt_modify_rss(struct bnxt *bp, struct bnxt_rss_ctx *rss_ctx,
 		u32 *indir_tbl = bp->rss_indir_tbl;
 
 		if (rss_ctx)
-			indir_tbl = rss_ctx->rss_indir_tbl;
+			indir_tbl = ethtool_rxfh_context_indir(ctx);
 		for (i = 0; i < tbl_size; i++)
 			indir_tbl[i] = rxfh->indir[i];
 		pad = bp->rss_indir_tbl_entries - tbl_size;
@@ -1909,15 +1911,11 @@ static int bnxt_create_rxfh_context(struct net_device *dev,
 	vnic->rss_ctx = rss_ctx;
 	vnic->flags |= BNXT_VNIC_RSSCTX_FLAG;
 	vnic->vnic_id = BNXT_VNIC_ID_INVALID;
-	rc = bnxt_alloc_rss_ctx_rss_table(bp, rss_ctx);
+	rc = bnxt_alloc_vnic_rss_table(bp, vnic);
 	if (rc)
 		goto out;
 
-	rc = bnxt_alloc_rss_indir_tbl(bp, rss_ctx);
-	if (rc)
-		goto out;
-
-	bnxt_set_dflt_rss_indir_tbl(bp, rss_ctx);
+	bnxt_set_dflt_rss_indir_tbl(bp, ctx);
 	memcpy(vnic->rss_hash_key, bp->rss_hash_key, HW_HASH_KEY_SIZE);
 
 	rc = bnxt_hwrm_vnic_alloc(bp, vnic, 0, bp->rx_nr_rings);
@@ -1931,7 +1929,7 @@ static int bnxt_create_rxfh_context(struct net_device *dev,
 		NL_SET_ERR_MSG_MOD(extack, "Unable to setup TPA");
 		goto out;
 	}
-	bnxt_modify_rss(bp, rss_ctx, rxfh);
+	bnxt_modify_rss(bp, ctx, rss_ctx, rxfh);
 
 	rc = __bnxt_setup_vnic_p5(bp, vnic);
 	if (rc) {
@@ -1961,7 +1959,7 @@ static int bnxt_modify_rxfh_context(struct net_device *dev,
 
 	rss_ctx = ethtool_rxfh_context_priv(ctx);
 
-	bnxt_modify_rss(bp, rss_ctx, rxfh);
+	bnxt_modify_rss(bp, ctx, rss_ctx, rxfh);
 
 	return bnxt_hwrm_vnic_rss_cfg_p5(bp, &rss_ctx->vnic);
 }
@@ -1990,7 +1988,7 @@ static int bnxt_set_rxfh(struct net_device *dev,
 	if (rxfh->hfunc && rxfh->hfunc != ETH_RSS_HASH_TOP)
 		return -EOPNOTSUPP;
 
-	bnxt_modify_rss(bp, NULL, rxfh);
+	bnxt_modify_rss(bp, NULL, NULL, rxfh);
 
 	bnxt_clear_usr_fltrs(bp, false);
 	if (netif_running(bp->dev)) {
include/linux/ethtool.h +5 -0
index dc8ed93097c3..5b28407d0619 100644
@@ -190,6 +190,11 @@ static inline void *ethtool_rxfh_context_priv(struct ethtool_rxfh_context *ctx)
 	return ctx->data;
 }
 
+static inline struct ethtool_rxfh_context *ethtool_rxfh_priv_context(void *priv)
+{
+	return container_of((u8(*)[])priv, struct ethtool_rxfh_context, data);
+}
+
 static inline u32 *ethtool_rxfh_context_indir(struct ethtool_rxfh_context *ctx)
 {
 	return (u32 *)(ctx->data + ALIGN(ctx->priv_size, sizeof(u32)));
Signature · 1 line
2.45.2
#26 lore raw

Re: [PATCH net-next 10/11] eth: bnxt: use the indir table from ethtool context

Quoted 8 lines · Jakub Kicinski

Instead of allocating a separate indir table in the vnic use the one already present in the RSS context allocated by the core. This doesn't save much LoC but we won't have to worry about syncing the local version back to the core, once core learns how to dump contexts.

Add ethtool_rxfh_priv_context() for converting from priv pointer to the context. The cast is a bit ugly (understatement) and some

could we wrap flex u8 data[] array of struct ethtool_rxfh_context in an union with void *priv?, then no cast will be needed.

but I still would prefer to don't abuse ownership over in-core data, and just keep it as inactive there, at the very end this could be moved to devm_ to avoid leaks

Quoted 20 lines

driver paths make carrying the context pointer in addition to driver priv pointer quite tedious.

---

include/linux/ethtool.h +5 -0
@@ -190,6 +190,11 @@ static inline void *ethtool_rxfh_context_priv(struct ethtool_rxfh_context *ctx)
  	return ctx->data;
  }
 
+static inline struct ethtool_rxfh_context *ethtool_rxfh_priv_context(void *priv)
+{
+	return container_of((u8(*)[])priv, struct ethtool_rxfh_context, data);
+}
+
  static inline u32 *ethtool_rxfh_context_indir(struct ethtool_rxfh_context *ctx)
  {
  	return (u32 *)(ctx->data + ALIGN(ctx->priv_size, sizeof(u32)));
#27 lore raw

Re: [PATCH net-next 10/11] eth: bnxt: use the indir table from ethtool context

In reply to Jakub Kicinski
Quoted 12 lines · Jakub Kicinski

Instead of allocating a separate indir table in the vnic use the one already present in the RSS context allocated by the core. This doesn't save much LoC but we won't have to worry about syncing the local version back to the core, once core learns how to dump contexts.

Add ethtool_rxfh_priv_context() for converting from priv pointer to the context. The cast is a bit ugly (understatement) and some driver paths make carrying the context pointer in addition to driver priv pointer quite tedious.

...

drivers/net/ethernet/broadcom/bnxt/bnxt.c static void bnxt_fill_hw_rss_tbl_p5(struct bnxt *bp, · line 6319

Quoted 10 lines
drivers/net/ethernet/broadcom/bnxt/bnxt.c +2 -0
@@ -6315,10 +6311,12 @@ static void bnxt_fill_hw_rss_tbl_p5(struct bnxt *bp,
 				    struct bnxt_vnic_info *vnic)
 {
 	__le16 *ring_tbl = vnic->rss_table;
+	struct ethtool_rxfh_context *ctx;
 	struct bnxt_rx_ring_info *rxr;
 	u16 tbl_size, i;
 
 	tbl_size = bnxt_get_rxfh_indir_size(bp->dev);
+	ctx = ethtool_rxfh_priv_context(vnic->rss_ctx);

Not super familiar with this driver or why this need arises, but
would it be simpler to just store ctx in vnic instead of priv?

author maintainer
#28 lore raw

Re: [PATCH net-next 10/11] eth: bnxt: use the indir table from ethtool context

Quoted 5 lines · On Wed, 3 Jul 2024 12:39:09 +0100 Edward Cree wrote:

drivers/net/ethernet/broadcom/bnxt/bnxt.c static void bnxt_fill_hw_rss_tbl_p5(struct bnxt *bp, · line 6319

Quoted 2 lines
drivers/net/ethernet/broadcom/bnxt/bnxt.c +1 -0
 	tbl_size = bnxt_get_rxfh_indir_size(bp->dev);
+	ctx = ethtool_rxfh_priv_context(vnic->rss_ctx);  

Not super familiar with this driver or why this need arises, but
would it be simpler to just store ctx in vnic instead of priv?

Yup, I think that should work!

author maintainer
#29 lore raw

PATCH 11/11 eth: bnxt: pad out the correct indirection table

In reply to Jakub Kicinski

bnxt allocates tables of max size, and changes the used size based on number of active rings. The unused entries get padded out with zeros. bnxt_modify_rss() seems to always pad out the table of the main / default RSS context, instead of the table of the modified context.

I haven't observed any behavior change due to this patch, so I don't think it's a fix. Not entirely sure what role the padding plays, 0 is a valid queue ID.

---
 drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c +1 -1
index 13e9b3b26f09..1d587846c394 100644
@@ -1858,7 +1858,7 @@ static void bnxt_modify_rss(struct bnxt *bp, struct ethtool_rxfh_context *ctx,
 			indir_tbl[i] = rxfh->indir[i];
 		pad = bp->rss_indir_tbl_entries - tbl_size;
 		if (pad)
-			memset(&bp->rss_indir_tbl[i], 0, pad * sizeof(u32));
+			memset(&indir_tbl[i], 0, pad * sizeof(u32));
 	}
 }
 
Signature · 1 line
2.45.2