[PATCH net-next v3 0/6] tsnep: multi queue support and some other improvements

[PATCH net-next v3 0/6] tsnep: multi queue support and some other improvements

[Gerhard Engleder]

Add support for additional TX/RX queues along with RX flow classification
support.

Binding is extended to allow additional interrupts for additional TX/RX
queues. Also dma-coherent is allowed as minor improvement.

RX path optimisation is done by using page pool as preparations for future
XDP support.

v3:
- now with changes in cover letter

v2:
- use netdev_name() (Jakub Kicinski)
- use ENOENT if RX flow rule is not found (Jakub Kicinski)
- eliminate return code of tsnep_add_rule() (Jakub Kicinski)
- remove commit with lazy refill due to depletion problem (Jakub
  Kicinski)

Gerhard Engleder (6):
  dt-bindings: net: tsnep: Allow dma-coherent
  dt-bindings: net: tsnep: Allow additional interrupts
  tsnep: Move interrupt from device to queue
  tsnep: Support multiple TX/RX queue pairs
  tsnep: Add EtherType RX flow classification support
  tsnep: Use page pool for RX

 .../bindings/net/engleder,tsnep.yaml          |  39 +-
 drivers/net/ethernet/engleder/Kconfig         |   1 +
 drivers/net/ethernet/engleder/Makefile        |   2 +-
 drivers/net/ethernet/engleder/tsnep.h         |  47 ++-
 drivers/net/ethernet/engleder/tsnep_ethtool.c |  38 ++
 drivers/net/ethernet/engleder/tsnep_hw.h      |  13 +-
 drivers/net/ethernet/engleder/tsnep_main.c    | 356 +++++++++++++-----
 drivers/net/ethernet/engleder/tsnep_rxnfc.c   | 281 ++++++++++++++
 8 files changed, 662 insertions(+), 115 deletions(-)
 create mode 100644 drivers/net/ethernet/engleder/tsnep_rxnfc.c

-- 
2.30.2

[PATCH net-next v3 1/6] dt-bindings: net: tsnep: Allow dma-coherent

[Gerhard Engleder]

Fix the following dtbs_check error if dma-coherent is used:

...: 'dma-coherent' does not match any of the regexes: 'pinctrl-[0-9]+'
From schema: .../Documentation/devicetree/bindings/net/engleder,tsnep.yaml

Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
---
 Documentation/devicetree/bindings/net/engleder,tsnep.yaml | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/Documentation/devicetree/bindings/net/engleder,tsnep.yaml b/Documentation/devicetree/bindings/net/engleder,tsnep.yaml
index d0e1476e15b5..37e08ee744a8 100644
--- a/Documentation/devicetree/bindings/net/engleder,tsnep.yaml
+++ b/Documentation/devicetree/bindings/net/engleder,tsnep.yaml
@@ -22,6 +22,8 @@ properties:
   interrupts:
     maxItems: 1
 
+  dma-coherent: true
+
   local-mac-address: true
 
   mac-address: true
-- 
2.30.2

Re: [PATCH net-next v3 1/6] dt-bindings: net: tsnep: Allow dma-coherent

[Krzysztof Kozlowski]

On 23/09/2022 22:29, Gerhard Engleder wrote:
> Fix the following dtbs_check error if dma-coherent is used:
> 
> ...: 'dma-coherent' does not match any of the regexes: 'pinctrl-[0-9]+'
> From schema: .../Documentation/devicetree/bindings/net/engleder,tsnep.yaml

Skip last line - it's obvious. What instead you miss here - the
DTS/target which has this warning. I assume that some existing DTS uses
this property?

> 
> Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
Best regards,
Krzysztof

Re: [PATCH net-next v3 1/6] dt-bindings: net: tsnep: Allow dma-coherent

[Gerhard Engleder]

On 24.09.22 11:15, Krzysztof Kozlowski wrote:
> On 23/09/2022 22:29, Gerhard Engleder wrote:
>> Fix the following dtbs_check error if dma-coherent is used:
>>
>> ...: 'dma-coherent' does not match any of the regexes: 'pinctrl-[0-9]+'
>>  From schema: .../Documentation/devicetree/bindings/net/engleder,tsnep.yaml
> 
> Skip last line - it's obvious. What instead you miss here - the
> DTS/target which has this warning. I assume that some existing DTS uses
> this property?

I will skip that line.

The binding is for an FPGA based Ethernet MAC. I'm working with
an evaluation platform currently. The DTS for the evaluation platform
is mainline, but my derived DTS was not accepted mainline. So there is
no DTS. This is similar for other FPGA based devices.

>>
>> Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
> Best regards,
> Krzysztof

Thanks!

Gerhard

Re: [PATCH net-next v3 1/6] dt-bindings: net: tsnep: Allow dma-coherent

[Krzysztof Kozlowski]

On 24/09/2022 20:11, Gerhard Engleder wrote:
> On 24.09.22 11:15, Krzysztof Kozlowski wrote:
>> On 23/09/2022 22:29, Gerhard Engleder wrote:
>>> Fix the following dtbs_check error if dma-coherent is used:
>>>
>>> ...: 'dma-coherent' does not match any of the regexes: 'pinctrl-[0-9]+'
>>>  From schema: .../Documentation/devicetree/bindings/net/engleder,tsnep.yaml
>>
>> Skip last line - it's obvious. What instead you miss here - the
>> DTS/target which has this warning. I assume that some existing DTS uses
>> this property?
> 
> I will skip that line.
> 
> The binding is for an FPGA based Ethernet MAC. I'm working with
> an evaluation platform currently. The DTS for the evaluation platform
> is mainline, but my derived DTS was not accepted mainline. So there is
> no DTS. This is similar for other FPGA based devices.

If this is not coming from mainline, then there is no warning...  we are
not interested in warnings in out-of-tree code, because we are not
fixing them.

Best regards,
Krzysztof

Re: [PATCH net-next v3 1/6] dt-bindings: net: tsnep: Allow dma-coherent

[Gerhard Engleder]

On 25.09.22 09:41, Krzysztof Kozlowski wrote:
> On 24/09/2022 20:11, Gerhard Engleder wrote:
>> On 24.09.22 11:15, Krzysztof Kozlowski wrote:
>>> On 23/09/2022 22:29, Gerhard Engleder wrote:
>>>> Fix the following dtbs_check error if dma-coherent is used:
>>>>
>>>> ...: 'dma-coherent' does not match any of the regexes: 'pinctrl-[0-9]+'
>>>>   From schema: .../Documentation/devicetree/bindings/net/engleder,tsnep.yaml
>>>
>>> Skip last line - it's obvious. What instead you miss here - the
>>> DTS/target which has this warning. I assume that some existing DTS uses
>>> this property?
>>
>> I will skip that line.
>>
>> The binding is for an FPGA based Ethernet MAC. I'm working with
>> an evaluation platform currently. The DTS for the evaluation platform
>> is mainline, but my derived DTS was not accepted mainline. So there is
>> no DTS. This is similar for other FPGA based devices.
> 
> If this is not coming from mainline, then there is no warning...  we are
> not interested in warnings in out-of-tree code, because we are not
> fixing them.

Ok. So I would rewrite the description that it just allows dma-coherent
and remove the fix/warning stuff. Is that ok?

Thanks!

Gerhard

Re: [PATCH net-next v3 1/6] dt-bindings: net: tsnep: Allow dma-coherent

[Krzysztof Kozlowski]

On 25/09/2022 10:14, Gerhard Engleder wrote:
> On 25.09.22 09:41, Krzysztof Kozlowski wrote:
>> On 24/09/2022 20:11, Gerhard Engleder wrote:
>>> On 24.09.22 11:15, Krzysztof Kozlowski wrote:
>>>> On 23/09/2022 22:29, Gerhard Engleder wrote:
>>>>> Fix the following dtbs_check error if dma-coherent is used:
>>>>>
>>>>> ...: 'dma-coherent' does not match any of the regexes: 'pinctrl-[0-9]+'
>>>>>   From schema: .../Documentation/devicetree/bindings/net/engleder,tsnep.yaml
>>>>
>>>> Skip last line - it's obvious. What instead you miss here - the
>>>> DTS/target which has this warning. I assume that some existing DTS uses
>>>> this property?
>>>
>>> I will skip that line.
>>>
>>> The binding is for an FPGA based Ethernet MAC. I'm working with
>>> an evaluation platform currently. The DTS for the evaluation platform
>>> is mainline, but my derived DTS was not accepted mainline. So there is
>>> no DTS. This is similar for other FPGA based devices.
>>
>> If this is not coming from mainline, then there is no warning...  we are
>> not interested in warnings in out-of-tree code, because we are not
>> fixing them.
> 
> Ok. So I would rewrite the description that it just allows dma-coherent
> and remove the fix/warning stuff. Is that ok?

That would be okay, but please add answer to why you are making this change.

Best regards,
Krzysztof

Re: [PATCH net-next v3 1/6] dt-bindings: net: tsnep: Allow dma-coherent

[Gerhard Engleder]

On 26.09.22 10:55, Krzysztof Kozlowski wrote:
> On 25/09/2022 10:14, Gerhard Engleder wrote:
>> On 25.09.22 09:41, Krzysztof Kozlowski wrote:
>>> On 24/09/2022 20:11, Gerhard Engleder wrote:
>>>> On 24.09.22 11:15, Krzysztof Kozlowski wrote:
>>>>> On 23/09/2022 22:29, Gerhard Engleder wrote:
>>>>>> Fix the following dtbs_check error if dma-coherent is used:
>>>>>>
>>>>>> ...: 'dma-coherent' does not match any of the regexes: 'pinctrl-[0-9]+'
>>>>>>    From schema: .../Documentation/devicetree/bindings/net/engleder,tsnep.yaml
>>>>>
>>>>> Skip last line - it's obvious. What instead you miss here - the
>>>>> DTS/target which has this warning. I assume that some existing DTS uses
>>>>> this property?
>>>>
>>>> I will skip that line.
>>>>
>>>> The binding is for an FPGA based Ethernet MAC. I'm working with
>>>> an evaluation platform currently. The DTS for the evaluation platform
>>>> is mainline, but my derived DTS was not accepted mainline. So there is
>>>> no DTS. This is similar for other FPGA based devices.
>>>
>>> If this is not coming from mainline, then there is no warning...  we are
>>> not interested in warnings in out-of-tree code, because we are not
>>> fixing them.
>>
>> Ok. So I would rewrite the description that it just allows dma-coherent
>> and remove the fix/warning stuff. Is that ok?
> 
> That would be okay, but please add answer to why you are making this change.

I already prepared it:

Within SoCs like ZynqMP, FPGA logic can be connected to different kinds
of AXI master ports. Also cache coherent AXI master ports are available.
The property "dma-coherent" is used to signal that DMA is cache
coherent.
Add "dma-coherent" property to allow the configuration of cache coherent
DMA.

Thanks!

Gerhard

Re: [PATCH net-next v3 1/6] dt-bindings: net: tsnep: Allow dma-coherent

[Krzysztof Kozlowski]

On 26/09/2022 19:46, Gerhard Engleder wrote:
>> That would be okay, but please add answer to why you are making this change.
> 
> I already prepared it:
> 
> Within SoCs like ZynqMP, FPGA logic can be connected to different kinds
> of AXI master ports. Also cache coherent AXI master ports are available.
> The property "dma-coherent" is used to signal that DMA is cache
> coherent.
> Add "dma-coherent" property to allow the configuration of cache coherent
> DMA.
> 

Sounds good, thanks.

Best regards,
Krzysztof

[PATCH net-next v3 2/6] dt-bindings: net: tsnep: Allow additional interrupts

[Gerhard Engleder]

Additional TX/RX queue pairs require dedicated interrupts. Extend
binding with additional interrupts.

Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
---
 .../bindings/net/engleder,tsnep.yaml          | 37 ++++++++++++++++++-
 1 file changed, 36 insertions(+), 1 deletion(-)

diff --git a/Documentation/devicetree/bindings/net/engleder,tsnep.yaml b/Documentation/devicetree/bindings/net/engleder,tsnep.yaml
index 37e08ee744a8..ce1f1bd413c2 100644
--- a/Documentation/devicetree/bindings/net/engleder,tsnep.yaml
+++ b/Documentation/devicetree/bindings/net/engleder,tsnep.yaml
@@ -20,7 +20,23 @@ properties:
     maxItems: 1
 
   interrupts:
-    maxItems: 1
+    minItems: 1
+    maxItems: 8
+
+  interrupt-names:
+    minItems: 1
+    maxItems: 8
+    items:
+      pattern: '^mac|txrx-[1-7]$'
+    description:
+      If more than one interrupt is available, then interrupts are
+      identified by their names.
+      "mac" is the main interrupt for basic MAC features and the first
+      TX/RX queue pair. If only a single interrupt is available, then
+      it is assumed that this interrupt is the "mac" interrupt.
+      "txrx-[1-7]" are the interrupts for additional TX/RX queue pairs.
+      These interrupt names shall start with index 1 and increment the
+      index by 1 with every further TX/RX queue pair.
 
   dma-coherent: true
 
@@ -78,4 +94,23 @@ examples:
                 };
             };
         };
+        tnsep1: ethernet@a0010000 {
+            compatible = "engleder,tsnep";
+            reg = <0x0 0xa0010000 0x0 0x10000>;
+            interrupts = <0 93 1>, <0 94 1>, <0 95 1>, <0 96 1>;
+            interrupt-names = "mac", "txrx-1", "txrx-2", "txrx-3";
+            interrupt-parent = <&gic>;
+            local-mac-address = [00 00 00 00 00 00];
+            phy-mode = "rgmii";
+            phy-handle = <&phy1>;
+            mdio {
+                #address-cells = <1>;
+                #size-cells = <0>;
+                suppress-preamble;
+                phy1: ethernet-phy@1 {
+                    reg = <1>;
+                    rxc-skew-ps = <1080>;
+                };
+            };
+        };
     };
-- 
2.30.2

Re: [PATCH net-next v3 2/6] dt-bindings: net: tsnep: Allow additional interrupts

[Krzysztof Kozlowski]

On 23/09/2022 22:29, Gerhard Engleder wrote:
> Additional TX/RX queue pairs require dedicated interrupts. Extend
> binding with additional interrupts.
> 
> Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
> ---
>  .../bindings/net/engleder,tsnep.yaml          | 37 ++++++++++++++++++-
>  1 file changed, 36 insertions(+), 1 deletion(-)
> 
> diff --git a/Documentation/devicetree/bindings/net/engleder,tsnep.yaml b/Documentation/devicetree/bindings/net/engleder,tsnep.yaml
> index 37e08ee744a8..ce1f1bd413c2 100644
> --- a/Documentation/devicetree/bindings/net/engleder,tsnep.yaml
> +++ b/Documentation/devicetree/bindings/net/engleder,tsnep.yaml
> @@ -20,7 +20,23 @@ properties:
>      maxItems: 1
>  
>    interrupts:
> -    maxItems: 1
> +    minItems: 1
> +    maxItems: 8
> +
> +  interrupt-names:
> +    minItems: 1
> +    maxItems: 8
> +    items:
> +      pattern: '^mac|txrx-[1-7]$'

No. The order of items must be fixed. Now you allow any combination,
which is exactly what we do not want.

> +    description:
> +      If more than one interrupt is available, then interrupts are
> +      identified by their names.

Not really. Interrupts are fixed, unless explicitly mentioned otherwise.

> +      "mac" is the main interrupt for basic MAC features and the first
> +      TX/RX queue pair. If only a single interrupt is available, then
> +      it is assumed that this interrupt is the "mac" interrupt.
> +      "txrx-[1-7]" are the interrupts for additional TX/RX queue pairs.
> +      These interrupt names shall start with index 1 and increment the
> +      index by 1 with every further TX/RX queue pair.

Skip last three sentences - they will become redundant after
implementing proper items.
>  
>    dma-coherent: true
>  
> @@ -78,4 +94,23 @@ examples:
>                  };
>              };
>          };

Missing line break.


Best regards,
Krzysztof

Re: [PATCH net-next v3 2/6] dt-bindings: net: tsnep: Allow additional interrupts

[Gerhard Engleder]

On 24.09.22 11:17, Krzysztof Kozlowski wrote:
> On 23/09/2022 22:29, Gerhard Engleder wrote:
>> Additional TX/RX queue pairs require dedicated interrupts. Extend
>> binding with additional interrupts.
>>
>> Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
>> ---
>>   .../bindings/net/engleder,tsnep.yaml          | 37 ++++++++++++++++++-
>>   1 file changed, 36 insertions(+), 1 deletion(-)
>>
>> diff --git a/Documentation/devicetree/bindings/net/engleder,tsnep.yaml b/Documentation/devicetree/bindings/net/engleder,tsnep.yaml
>> index 37e08ee744a8..ce1f1bd413c2 100644
>> --- a/Documentation/devicetree/bindings/net/engleder,tsnep.yaml
>> +++ b/Documentation/devicetree/bindings/net/engleder,tsnep.yaml
>> @@ -20,7 +20,23 @@ properties:
>>       maxItems: 1
>>   
>>     interrupts:
>> -    maxItems: 1
>> +    minItems: 1
>> +    maxItems: 8
>> +
>> +  interrupt-names:
>> +    minItems: 1
>> +    maxItems: 8
>> +    items:
>> +      pattern: '^mac|txrx-[1-7]$'
> 
> No. The order of items must be fixed. Now you allow any combination,
> which is exactly what we do not want.

Ok. I will do it like in 
https://elixir.bootlin.com/linux/latest/source/Documentation/devicetree/bindings/net/microchip,sparx5-switch.yaml#L58

>> +    description:
>> +      If more than one interrupt is available, then interrupts are
>> +      identified by their names.
> 
> Not really. Interrupts are fixed, unless explicitly mentioned otherwise.
> 
>> +      "mac" is the main interrupt for basic MAC features and the first
>> +      TX/RX queue pair. If only a single interrupt is available, then
>> +      it is assumed that this interrupt is the "mac" interrupt.
>> +      "txrx-[1-7]" are the interrupts for additional TX/RX queue pairs.
>> +      These interrupt names shall start with index 1 and increment the
>> +      index by 1 with every further TX/RX queue pair.
> 
> Skip last three sentences - they will become redundant after
> implementing proper items.

I will rework description for fixed order.

>>   
>>     dma-coherent: true
>>   
>> @@ -78,4 +94,23 @@ examples:
>>                   };
>>               };
>>           };
> 
> Missing line break.

I will add it.
> Best regards,
> Krzysztof

Thanks!

Gerhard

[PATCH net-next v3 3/6] tsnep: Move interrupt from device to queue

[Gerhard Engleder]

For multiple queues multiple interrupts shall be used. Therefore, rework
global interrupt to per queue interrupt.

Every interrupt name shall contain interface name and queue information.
To get a valid interface name, the interrupt request needs to by done
during open like in other drivers. Additionally, this allows the removal
of some initialisation checks in the interrupt handler.

Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
---
 drivers/net/ethernet/engleder/tsnep.h      |   4 +-
 drivers/net/ethernet/engleder/tsnep_main.c | 128 +++++++++++++++------
 2 files changed, 99 insertions(+), 33 deletions(-)

diff --git a/drivers/net/ethernet/engleder/tsnep.h b/drivers/net/ethernet/engleder/tsnep.h
index 147fe03ca979..149c9acbae9c 100644
--- a/drivers/net/ethernet/engleder/tsnep.h
+++ b/drivers/net/ethernet/engleder/tsnep.h
@@ -55,6 +55,7 @@ struct tsnep_tx_entry {
 struct tsnep_tx {
 	struct tsnep_adapter *adapter;
 	void __iomem *addr;
+	int queue_index;
 
 	void *page[TSNEP_RING_PAGE_COUNT];
 	dma_addr_t page_dma[TSNEP_RING_PAGE_COUNT];
@@ -105,12 +106,14 @@ struct tsnep_rx {
 
 struct tsnep_queue {
 	struct tsnep_adapter *adapter;
+	char name[IFNAMSIZ + 9];
 
 	struct tsnep_tx *tx;
 	struct tsnep_rx *rx;
 
 	struct napi_struct napi;
 
+	int irq;
 	u32 irq_mask;
 };
 
@@ -126,7 +129,6 @@ struct tsnep_adapter {
 	struct platform_device *pdev;
 	struct device *dmadev;
 	void __iomem *addr;
-	int irq;
 
 	bool gate_control;
 	/* gate control lock */
diff --git a/drivers/net/ethernet/engleder/tsnep_main.c b/drivers/net/ethernet/engleder/tsnep_main.c
index 19db8b1dddc4..c95328ef992b 100644
--- a/drivers/net/ethernet/engleder/tsnep_main.c
+++ b/drivers/net/ethernet/engleder/tsnep_main.c
@@ -60,22 +60,29 @@ static irqreturn_t tsnep_irq(int irq, void *arg)
 		iowrite32(active, adapter->addr + ECM_INT_ACKNOWLEDGE);
 
 	/* handle link interrupt */
-	if ((active & ECM_INT_LINK) != 0) {
-		if (adapter->netdev->phydev)
-			phy_mac_interrupt(adapter->netdev->phydev);
-	}
+	if ((active & ECM_INT_LINK) != 0)
+		phy_mac_interrupt(adapter->netdev->phydev);
 
 	/* handle TX/RX queue 0 interrupt */
 	if ((active & adapter->queue[0].irq_mask) != 0) {
-		if (adapter->netdev) {
-			tsnep_disable_irq(adapter, adapter->queue[0].irq_mask);
-			napi_schedule(&adapter->queue[0].napi);
-		}
+		tsnep_disable_irq(adapter, adapter->queue[0].irq_mask);
+		napi_schedule(&adapter->queue[0].napi);
 	}
 
 	return IRQ_HANDLED;
 }
 
+static irqreturn_t tsnep_irq_txrx(int irq, void *arg)
+{
+	struct tsnep_queue *queue = arg;
+
+	/* handle TX/RX queue interrupt */
+	tsnep_disable_irq(queue->adapter, queue->irq_mask);
+	napi_schedule(&queue->napi);
+
+	return IRQ_HANDLED;
+}
+
 static int tsnep_mdiobus_read(struct mii_bus *bus, int addr, int regnum)
 {
 	struct tsnep_adapter *adapter = bus->priv;
@@ -536,7 +543,7 @@ static bool tsnep_tx_poll(struct tsnep_tx *tx, int napi_budget)
 }
 
 static int tsnep_tx_open(struct tsnep_adapter *adapter, void __iomem *addr,
-			 struct tsnep_tx *tx)
+			 int queue_index, struct tsnep_tx *tx)
 {
 	dma_addr_t dma;
 	int retval;
@@ -544,6 +551,7 @@ static int tsnep_tx_open(struct tsnep_adapter *adapter, void __iomem *addr,
 	memset(tx, 0, sizeof(*tx));
 	tx->adapter = adapter;
 	tx->addr = addr;
+	tx->queue_index = queue_index;
 
 	retval = tsnep_tx_ring_init(tx);
 	if (retval)
@@ -854,6 +862,56 @@ static int tsnep_poll(struct napi_struct *napi, int budget)
 	return min(done, budget - 1);
 }
 
+static int tsnep_request_irq(struct tsnep_queue *queue, bool first)
+{
+	const char *name = netdev_name(queue->adapter->netdev);
+	irq_handler_t handler;
+	void *dev;
+	int retval;
+
+	if (first) {
+		sprintf(queue->name, "%s-mac", name);
+		handler = tsnep_irq;
+		dev = queue->adapter;
+	} else {
+		if (queue->tx && queue->rx)
+			sprintf(queue->name, "%s-txrx-%d", name,
+				queue->rx->queue_index);
+		else if (queue->tx)
+			sprintf(queue->name, "%s-tx-%d", name,
+				queue->tx->queue_index);
+		else
+			sprintf(queue->name, "%s-rx-%d", name,
+				queue->rx->queue_index);
+		handler = tsnep_irq_txrx;
+		dev = queue;
+	}
+
+	retval = request_irq(queue->irq, handler, 0, queue->name, dev);
+	if (retval) {
+		/* if name is empty, then interrupt won't be freed */
+		memset(queue->name, 0, sizeof(queue->name));
+	}
+
+	return retval;
+}
+
+static void tsnep_free_irq(struct tsnep_queue *queue, bool first)
+{
+	void *dev;
+
+	if (!strlen(queue->name))
+		return;
+
+	if (first)
+		dev = queue->adapter;
+	else
+		dev = queue;
+
+	free_irq(queue->irq, dev);
+	memset(queue->name, 0, sizeof(queue->name));
+}
+
 static int tsnep_netdev_open(struct net_device *netdev)
 {
 	struct tsnep_adapter *adapter = netdev_priv(netdev);
@@ -863,15 +921,11 @@ static int tsnep_netdev_open(struct net_device *netdev)
 	int rx_queue_index = 0;
 	int retval;
 
-	retval = tsnep_phy_open(adapter);
-	if (retval)
-		return retval;
-
 	for (i = 0; i < adapter->num_queues; i++) {
 		adapter->queue[i].adapter = adapter;
 		if (adapter->queue[i].tx) {
 			addr = adapter->addr + TSNEP_QUEUE(tx_queue_index);
-			retval = tsnep_tx_open(adapter, addr,
+			retval = tsnep_tx_open(adapter, addr, tx_queue_index,
 					       adapter->queue[i].tx);
 			if (retval)
 				goto failed;
@@ -886,6 +940,14 @@ static int tsnep_netdev_open(struct net_device *netdev)
 				goto failed;
 			rx_queue_index++;
 		}
+
+		retval = tsnep_request_irq(&adapter->queue[i], i == 0);
+		if (retval) {
+			netif_err(adapter, drv, adapter->netdev,
+				  "can't get assigned irq %d.\n",
+				  adapter->queue[i].irq);
+			goto failed;
+		}
 	}
 
 	retval = netif_set_real_num_tx_queues(adapter->netdev,
@@ -897,6 +959,11 @@ static int tsnep_netdev_open(struct net_device *netdev)
 	if (retval)
 		goto failed;
 
+	tsnep_enable_irq(adapter, ECM_INT_LINK);
+	retval = tsnep_phy_open(adapter);
+	if (retval)
+		goto phy_failed;
+
 	for (i = 0; i < adapter->num_queues; i++) {
 		netif_napi_add(adapter->netdev, &adapter->queue[i].napi,
 			       tsnep_poll, 64);
@@ -907,14 +974,18 @@ static int tsnep_netdev_open(struct net_device *netdev)
 
 	return 0;
 
+phy_failed:
+	tsnep_disable_irq(adapter, ECM_INT_LINK);
+	tsnep_phy_close(adapter);
 failed:
 	for (i = 0; i < adapter->num_queues; i++) {
+		tsnep_free_irq(&adapter->queue[i], i == 0);
+
 		if (adapter->queue[i].rx)
 			tsnep_rx_close(adapter->queue[i].rx);
 		if (adapter->queue[i].tx)
 			tsnep_tx_close(adapter->queue[i].tx);
 	}
-	tsnep_phy_close(adapter);
 	return retval;
 }
 
@@ -923,20 +994,23 @@ static int tsnep_netdev_close(struct net_device *netdev)
 	struct tsnep_adapter *adapter = netdev_priv(netdev);
 	int i;
 
+	tsnep_disable_irq(adapter, ECM_INT_LINK);
+	tsnep_phy_close(adapter);
+
 	for (i = 0; i < adapter->num_queues; i++) {
 		tsnep_disable_irq(adapter, adapter->queue[i].irq_mask);
 
 		napi_disable(&adapter->queue[i].napi);
 		netif_napi_del(&adapter->queue[i].napi);
 
+		tsnep_free_irq(&adapter->queue[i], i == 0);
+
 		if (adapter->queue[i].rx)
 			tsnep_rx_close(adapter->queue[i].rx);
 		if (adapter->queue[i].tx)
 			tsnep_tx_close(adapter->queue[i].tx);
 	}
 
-	tsnep_phy_close(adapter);
-
 	return 0;
 }
 
@@ -1225,10 +1299,8 @@ static int tsnep_probe(struct platform_device *pdev)
 	adapter->addr = devm_ioremap_resource(&pdev->dev, io);
 	if (IS_ERR(adapter->addr))
 		return PTR_ERR(adapter->addr);
-	adapter->irq = platform_get_irq(pdev, 0);
 	netdev->mem_start = io->start;
 	netdev->mem_end = io->end;
-	netdev->irq = adapter->irq;
 
 	type = ioread32(adapter->addr + ECM_TYPE);
 	revision = (type & ECM_REVISION_MASK) >> ECM_REVISION_SHIFT;
@@ -1238,10 +1310,14 @@ static int tsnep_probe(struct platform_device *pdev)
 	adapter->num_tx_queues = TSNEP_QUEUES;
 	adapter->num_rx_queues = TSNEP_QUEUES;
 	adapter->num_queues = TSNEP_QUEUES;
+	adapter->queue[0].irq = platform_get_irq(pdev, 0);
 	adapter->queue[0].tx = &adapter->tx[0];
 	adapter->queue[0].rx = &adapter->rx[0];
 	adapter->queue[0].irq_mask = ECM_INT_TX_0 | ECM_INT_RX_0;
 
+	netdev->irq = adapter->queue[0].irq;
+	tsnep_disable_irq(adapter, ECM_INT_ALL);
+
 	retval = dma_set_mask_and_coherent(&adapter->pdev->dev,
 					   DMA_BIT_MASK(64));
 	if (retval) {
@@ -1249,19 +1325,9 @@ static int tsnep_probe(struct platform_device *pdev)
 		return retval;
 	}
 
-	tsnep_disable_irq(adapter, ECM_INT_ALL);
-	retval = devm_request_irq(&adapter->pdev->dev, adapter->irq, tsnep_irq,
-				  0, TSNEP, adapter);
-	if (retval != 0) {
-		dev_err(&adapter->pdev->dev, "can't get assigned irq %d.\n",
-			adapter->irq);
-		return retval;
-	}
-	tsnep_enable_irq(adapter, ECM_INT_LINK);
-
 	retval = tsnep_mac_init(adapter);
 	if (retval)
-		goto mac_init_failed;
+		return retval;
 
 	retval = tsnep_mdio_init(adapter);
 	if (retval)
@@ -1307,8 +1373,6 @@ static int tsnep_probe(struct platform_device *pdev)
 	if (adapter->mdiobus)
 		mdiobus_unregister(adapter->mdiobus);
 mdio_init_failed:
-mac_init_failed:
-	tsnep_disable_irq(adapter, ECM_INT_ALL);
 	return retval;
 }
 
-- 
2.30.2

[PATCH net-next v3 4/6] tsnep: Support multiple TX/RX queue pairs

[Gerhard Engleder]

Support additional TX/RX queue pairs if dedicated interrupt is
available. Interrupts are detected by name in device tree.

Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
---
 drivers/net/ethernet/engleder/tsnep.h      |  2 -
 drivers/net/ethernet/engleder/tsnep_hw.h   |  1 +
 drivers/net/ethernet/engleder/tsnep_main.c | 61 ++++++++++++++++++----
 3 files changed, 53 insertions(+), 11 deletions(-)

diff --git a/drivers/net/ethernet/engleder/tsnep.h b/drivers/net/ethernet/engleder/tsnep.h
index 149c9acbae9c..62a279bcb011 100644
--- a/drivers/net/ethernet/engleder/tsnep.h
+++ b/drivers/net/ethernet/engleder/tsnep.h
@@ -21,8 +21,6 @@
 #define TSNEP_RING_ENTRIES_PER_PAGE (PAGE_SIZE / TSNEP_DESC_SIZE)
 #define TSNEP_RING_PAGE_COUNT (TSNEP_RING_SIZE / TSNEP_RING_ENTRIES_PER_PAGE)
 
-#define TSNEP_QUEUES 1
-
 struct tsnep_gcl {
 	void __iomem *addr;
 
diff --git a/drivers/net/ethernet/engleder/tsnep_hw.h b/drivers/net/ethernet/engleder/tsnep_hw.h
index e03aaafab559..e6cc6fbaf0d7 100644
--- a/drivers/net/ethernet/engleder/tsnep_hw.h
+++ b/drivers/net/ethernet/engleder/tsnep_hw.h
@@ -34,6 +34,7 @@
 #define ECM_INT_LINK 0x00000020
 #define ECM_INT_TX_0 0x00000100
 #define ECM_INT_RX_0 0x00000200
+#define ECM_INT_TXRX_SHIFT 2
 #define ECM_INT_ALL 0x7FFFFFFF
 #define ECM_INT_DISABLE 0x80000000
 
diff --git a/drivers/net/ethernet/engleder/tsnep_main.c b/drivers/net/ethernet/engleder/tsnep_main.c
index c95328ef992b..bf088b91efb7 100644
--- a/drivers/net/ethernet/engleder/tsnep_main.c
+++ b/drivers/net/ethernet/engleder/tsnep_main.c
@@ -1265,6 +1265,52 @@ static int tsnep_phy_init(struct tsnep_adapter *adapter)
 	return 0;
 }
 
+static int tsnep_queue_init(struct tsnep_adapter *adapter, int queue_count)
+{
+	u32 irq_mask = ECM_INT_TX_0 | ECM_INT_RX_0;
+	char name[8];
+	int i;
+	int retval;
+
+	/* one TX/RX queue pair for netdev is mandatory */
+	if (platform_irq_count(adapter->pdev) == 1)
+		retval = platform_get_irq(adapter->pdev, 0);
+	else
+		retval = platform_get_irq_byname(adapter->pdev, "mac");
+	if (retval < 0)
+		return retval;
+	adapter->num_tx_queues = 1;
+	adapter->num_rx_queues = 1;
+	adapter->num_queues = 1;
+	adapter->queue[0].irq = retval;
+	adapter->queue[0].tx = &adapter->tx[0];
+	adapter->queue[0].rx = &adapter->rx[0];
+	adapter->queue[0].irq_mask = irq_mask;
+
+	adapter->netdev->irq = adapter->queue[0].irq;
+
+	/* add additional TX/RX queue pairs only if dedicated interrupt is
+	 * available
+	 */
+	for (i = 1; i < queue_count; i++) {
+		sprintf(name, "txrx-%d", i);
+		retval = platform_get_irq_byname_optional(adapter->pdev, name);
+		if (retval < 0)
+			break;
+
+		adapter->num_tx_queues++;
+		adapter->num_rx_queues++;
+		adapter->num_queues++;
+		adapter->queue[i].irq = retval;
+		adapter->queue[i].tx = &adapter->tx[i];
+		adapter->queue[i].rx = &adapter->rx[i];
+		adapter->queue[i].irq_mask =
+			irq_mask << (ECM_INT_TXRX_SHIFT * i);
+	}
+
+	return 0;
+}
+
 static int tsnep_probe(struct platform_device *pdev)
 {
 	struct tsnep_adapter *adapter;
@@ -1273,6 +1319,7 @@ static int tsnep_probe(struct platform_device *pdev)
 	u32 type;
 	int revision;
 	int version;
+	int queue_count;
 	int retval;
 
 	netdev = devm_alloc_etherdev_mqs(&pdev->dev,
@@ -1305,19 +1352,15 @@ static int tsnep_probe(struct platform_device *pdev)
 	type = ioread32(adapter->addr + ECM_TYPE);
 	revision = (type & ECM_REVISION_MASK) >> ECM_REVISION_SHIFT;
 	version = (type & ECM_VERSION_MASK) >> ECM_VERSION_SHIFT;
+	queue_count = (type & ECM_QUEUE_COUNT_MASK) >> ECM_QUEUE_COUNT_SHIFT;
 	adapter->gate_control = type & ECM_GATE_CONTROL;
 
-	adapter->num_tx_queues = TSNEP_QUEUES;
-	adapter->num_rx_queues = TSNEP_QUEUES;
-	adapter->num_queues = TSNEP_QUEUES;
-	adapter->queue[0].irq = platform_get_irq(pdev, 0);
-	adapter->queue[0].tx = &adapter->tx[0];
-	adapter->queue[0].rx = &adapter->rx[0];
-	adapter->queue[0].irq_mask = ECM_INT_TX_0 | ECM_INT_RX_0;
-
-	netdev->irq = adapter->queue[0].irq;
 	tsnep_disable_irq(adapter, ECM_INT_ALL);
 
+	retval = tsnep_queue_init(adapter, queue_count);
+	if (retval)
+		return retval;
+
 	retval = dma_set_mask_and_coherent(&adapter->pdev->dev,
 					   DMA_BIT_MASK(64));
 	if (retval) {
-- 
2.30.2

[PATCH net-next v3 5/6] tsnep: Add EtherType RX flow classification support

[Gerhard Engleder]

Received Ethernet frames are assigned to first RX queue per default.
Based on EtherType Ethernet frames can be assigned to other RX queues.
This enables processing of real-time Ethernet protocols on dedicated
RX queues.

Add RX flow classification interface for EtherType based RX queue
assignment.

Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
---
 drivers/net/ethernet/engleder/Makefile        |   2 +-
 drivers/net/ethernet/engleder/tsnep.h         |  36 +++
 drivers/net/ethernet/engleder/tsnep_ethtool.c |  38 +++
 drivers/net/ethernet/engleder/tsnep_hw.h      |  12 +-
 drivers/net/ethernet/engleder/tsnep_main.c    |  11 +
 drivers/net/ethernet/engleder/tsnep_rxnfc.c   | 281 ++++++++++++++++++
 6 files changed, 375 insertions(+), 5 deletions(-)
 create mode 100644 drivers/net/ethernet/engleder/tsnep_rxnfc.c

diff --git a/drivers/net/ethernet/engleder/Makefile b/drivers/net/ethernet/engleder/Makefile
index cce2191cb889..b6e3b16623de 100644
--- a/drivers/net/ethernet/engleder/Makefile
+++ b/drivers/net/ethernet/engleder/Makefile
@@ -6,5 +6,5 @@
 obj-$(CONFIG_TSNEP) += tsnep.o
 
 tsnep-objs := tsnep_main.o tsnep_ethtool.o tsnep_ptp.o tsnep_tc.o \
-	      $(tsnep-y)
+	      tsnep_rxnfc.o $(tsnep-y)
 tsnep-$(CONFIG_TSNEP_SELFTESTS) += tsnep_selftests.o
diff --git a/drivers/net/ethernet/engleder/tsnep.h b/drivers/net/ethernet/engleder/tsnep.h
index 62a279bcb011..2ca34ae9b55a 100644
--- a/drivers/net/ethernet/engleder/tsnep.h
+++ b/drivers/net/ethernet/engleder/tsnep.h
@@ -37,6 +37,24 @@ struct tsnep_gcl {
 	bool change;
 };
 
+enum tsnep_rxnfc_filter_type {
+	TSNEP_RXNFC_ETHER_TYPE,
+};
+
+struct tsnep_rxnfc_filter {
+	enum tsnep_rxnfc_filter_type type;
+	union {
+		u16 ether_type;
+	};
+};
+
+struct tsnep_rxnfc_rule {
+	struct list_head list;
+	struct tsnep_rxnfc_filter filter;
+	int queue_index;
+	int location;
+};
+
 struct tsnep_tx_entry {
 	struct tsnep_tx_desc *desc;
 	struct tsnep_tx_desc_wb *desc_wb;
@@ -141,6 +159,12 @@ struct tsnep_adapter {
 	/* ptp clock lock */
 	spinlock_t ptp_lock;
 
+	/* RX flow classification rules lock */
+	struct mutex rxnfc_lock;
+	struct list_head rxnfc_rules;
+	int rxnfc_count;
+	int rxnfc_max;
+
 	int num_tx_queues;
 	struct tsnep_tx tx[TSNEP_MAX_QUEUES];
 	int num_rx_queues;
@@ -161,6 +185,18 @@ void tsnep_tc_cleanup(struct tsnep_adapter *adapter);
 int tsnep_tc_setup(struct net_device *netdev, enum tc_setup_type type,
 		   void *type_data);
 
+int tsnep_rxnfc_init(struct tsnep_adapter *adapter);
+void tsnep_rxnfc_cleanup(struct tsnep_adapter *adapter);
+int tsnep_rxnfc_get_rule(struct tsnep_adapter *adapter,
+			 struct ethtool_rxnfc *cmd);
+int tsnep_rxnfc_get_all(struct tsnep_adapter *adapter,
+			struct ethtool_rxnfc *cmd,
+			u32 *rule_locs);
+int tsnep_rxnfc_add_rule(struct tsnep_adapter *adapter,
+			 struct ethtool_rxnfc *cmd);
+int tsnep_rxnfc_del_rule(struct tsnep_adapter *adapter,
+			 struct ethtool_rxnfc *cmd);
+
 #if IS_ENABLED(CONFIG_TSNEP_SELFTESTS)
 int tsnep_ethtool_get_test_count(void);
 void tsnep_ethtool_get_test_strings(u8 *data);
diff --git a/drivers/net/ethernet/engleder/tsnep_ethtool.c b/drivers/net/ethernet/engleder/tsnep_ethtool.c
index e6760dc68ddd..b9c4c45db052 100644
--- a/drivers/net/ethernet/engleder/tsnep_ethtool.c
+++ b/drivers/net/ethernet/engleder/tsnep_ethtool.c
@@ -250,6 +250,42 @@ static int tsnep_ethtool_get_sset_count(struct net_device *netdev, int sset)
 	}
 }
 
+static int tsnep_ethtool_get_rxnfc(struct net_device *dev,
+				   struct ethtool_rxnfc *cmd, u32 *rule_locs)
+{
+	struct tsnep_adapter *adapter = netdev_priv(dev);
+
+	switch (cmd->cmd) {
+	case ETHTOOL_GRXRINGS:
+		cmd->data = adapter->num_rx_queues;
+		return 0;
+	case ETHTOOL_GRXCLSRLCNT:
+		cmd->rule_cnt = adapter->rxnfc_count;
+		return 0;
+	case ETHTOOL_GRXCLSRULE:
+		return tsnep_rxnfc_get_rule(adapter, cmd);
+	case ETHTOOL_GRXCLSRLALL:
+		return tsnep_rxnfc_get_all(adapter, cmd, rule_locs);
+	default:
+		return -EOPNOTSUPP;
+	}
+}
+
+static int tsnep_ethtool_set_rxnfc(struct net_device *dev,
+				   struct ethtool_rxnfc *cmd)
+{
+	struct tsnep_adapter *adapter = netdev_priv(dev);
+
+	switch (cmd->cmd) {
+	case ETHTOOL_SRXCLSRLINS:
+		return tsnep_rxnfc_add_rule(adapter, cmd);
+	case ETHTOOL_SRXCLSRLDEL:
+		return tsnep_rxnfc_del_rule(adapter, cmd);
+	default:
+		return -EOPNOTSUPP;
+	}
+}
+
 static int tsnep_ethtool_get_ts_info(struct net_device *dev,
 				     struct ethtool_ts_info *info)
 {
@@ -287,6 +323,8 @@ const struct ethtool_ops tsnep_ethtool_ops = {
 	.get_strings = tsnep_ethtool_get_strings,
 	.get_ethtool_stats = tsnep_ethtool_get_ethtool_stats,
 	.get_sset_count = tsnep_ethtool_get_sset_count,
+	.get_rxnfc = tsnep_ethtool_get_rxnfc,
+	.set_rxnfc = tsnep_ethtool_set_rxnfc,
 	.get_ts_info = tsnep_ethtool_get_ts_info,
 	.get_link_ksettings = phy_ethtool_get_link_ksettings,
 	.set_link_ksettings = phy_ethtool_set_link_ksettings,
diff --git a/drivers/net/ethernet/engleder/tsnep_hw.h b/drivers/net/ethernet/engleder/tsnep_hw.h
index e6cc6fbaf0d7..315dada75323 100644
--- a/drivers/net/ethernet/engleder/tsnep_hw.h
+++ b/drivers/net/ethernet/engleder/tsnep_hw.h
@@ -122,10 +122,6 @@
 #define TSNEP_RX_STATISTIC_BUFFER_TOO_SMALL 0x0191
 #define TSNEP_RX_STATISTIC_FIFO_OVERFLOW 0x0192
 #define TSNEP_RX_STATISTIC_INVALID_FRAME 0x0193
-#define TSNEP_RX_ASSIGN 0x01A0
-#define TSNEP_RX_ASSIGN_ETHER_TYPE_ACTIVE 0x00000001
-#define TSNEP_RX_ASSIGN_ETHER_TYPE_MASK 0xFFFF0000
-#define TSNEP_RX_ASSIGN_ETHER_TYPE_SHIFT 16
 #define TSNEP_MAC_ADDRESS_LOW 0x0800
 #define TSNEP_MAC_ADDRESS_HIGH 0x0804
 #define TSNEP_RX_FILTER 0x0806
@@ -152,6 +148,14 @@
 #define TSNEP_GCL_A 0x2000
 #define TSNEP_GCL_B 0x2800
 #define TSNEP_GCL_SIZE SZ_2K
+#define TSNEP_RX_ASSIGN 0x0840
+#define TSNEP_RX_ASSIGN_ACTIVE 0x00000001
+#define TSNEP_RX_ASSIGN_QUEUE_MASK 0x00000006
+#define TSNEP_RX_ASSIGN_QUEUE_SHIFT 1
+#define TSNEP_RX_ASSIGN_OFFSET 1
+#define TSNEP_RX_ASSIGN_ETHER_TYPE 0x0880
+#define TSNEP_RX_ASSIGN_ETHER_TYPE_OFFSET 2
+#define TSNEP_RX_ASSIGN_ETHER_TYPE_COUNT 2
 
 /* tsnep gate control list operation */
 struct tsnep_gcl_operation {
diff --git a/drivers/net/ethernet/engleder/tsnep_main.c b/drivers/net/ethernet/engleder/tsnep_main.c
index bf088b91efb7..a6b81f32d76b 100644
--- a/drivers/net/ethernet/engleder/tsnep_main.c
+++ b/drivers/net/ethernet/engleder/tsnep_main.c
@@ -1341,6 +1341,8 @@ static int tsnep_probe(struct platform_device *pdev)
 	netdev->max_mtu = TSNEP_MAX_FRAME_SIZE;
 
 	mutex_init(&adapter->gate_control_lock);
+	mutex_init(&adapter->rxnfc_lock);
+	INIT_LIST_HEAD(&adapter->rxnfc_rules);
 
 	io = platform_get_resource(pdev, IORESOURCE_MEM, 0);
 	adapter->addr = devm_ioremap_resource(&pdev->dev, io);
@@ -1354,6 +1356,7 @@ static int tsnep_probe(struct platform_device *pdev)
 	version = (type & ECM_VERSION_MASK) >> ECM_VERSION_SHIFT;
 	queue_count = (type & ECM_QUEUE_COUNT_MASK) >> ECM_QUEUE_COUNT_SHIFT;
 	adapter->gate_control = type & ECM_GATE_CONTROL;
+	adapter->rxnfc_max = TSNEP_RX_ASSIGN_ETHER_TYPE_COUNT;
 
 	tsnep_disable_irq(adapter, ECM_INT_ALL);
 
@@ -1388,6 +1391,10 @@ static int tsnep_probe(struct platform_device *pdev)
 	if (retval)
 		goto tc_init_failed;
 
+	retval = tsnep_rxnfc_init(adapter);
+	if (retval)
+		goto rxnfc_init_failed;
+
 	netdev->netdev_ops = &tsnep_netdev_ops;
 	netdev->ethtool_ops = &tsnep_ethtool_ops;
 	netdev->features = NETIF_F_SG;
@@ -1408,6 +1415,8 @@ static int tsnep_probe(struct platform_device *pdev)
 	return 0;
 
 register_failed:
+	tsnep_rxnfc_cleanup(adapter);
+rxnfc_init_failed:
 	tsnep_tc_cleanup(adapter);
 tc_init_failed:
 	tsnep_ptp_cleanup(adapter);
@@ -1425,6 +1434,8 @@ static int tsnep_remove(struct platform_device *pdev)
 
 	unregister_netdev(adapter->netdev);
 
+	tsnep_rxnfc_cleanup(adapter);
+
 	tsnep_tc_cleanup(adapter);
 
 	tsnep_ptp_cleanup(adapter);
diff --git a/drivers/net/ethernet/engleder/tsnep_rxnfc.c b/drivers/net/ethernet/engleder/tsnep_rxnfc.c
new file mode 100644
index 000000000000..ae3022548e92
--- /dev/null
+++ b/drivers/net/ethernet/engleder/tsnep_rxnfc.c
@@ -0,0 +1,281 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (C) 2022 Gerhard Engleder <gerhard@engleder-embedded.com> */
+
+#include "tsnep.h"
+
+#define ETHER_TYPE_FULL_MASK ((__force __be16)~0)
+
+static void tsnep_enable_rule(struct tsnep_adapter *adapter,
+			      struct tsnep_rxnfc_rule *rule)
+{
+	u8 rx_assign;
+	void __iomem *addr;
+
+	rx_assign = TSNEP_RX_ASSIGN_ACTIVE;
+	rx_assign |= (rule->queue_index << TSNEP_RX_ASSIGN_QUEUE_SHIFT) &
+		     TSNEP_RX_ASSIGN_QUEUE_MASK;
+
+	addr = adapter->addr + TSNEP_RX_ASSIGN_ETHER_TYPE +
+	       TSNEP_RX_ASSIGN_ETHER_TYPE_OFFSET * rule->location;
+	iowrite16(rule->filter.ether_type, addr);
+
+	/* enable rule after all settings are done */
+	addr = adapter->addr + TSNEP_RX_ASSIGN +
+	       TSNEP_RX_ASSIGN_OFFSET * rule->location;
+	iowrite8(rx_assign, addr);
+}
+
+static void tsnep_disable_rule(struct tsnep_adapter *adapter,
+			       struct tsnep_rxnfc_rule *rule)
+{
+	void __iomem *addr;
+
+	addr = adapter->addr + TSNEP_RX_ASSIGN +
+	       TSNEP_RX_ASSIGN_OFFSET * rule->location;
+	iowrite8(0, addr);
+}
+
+static struct tsnep_rxnfc_rule *tsnep_get_rule(struct tsnep_adapter *adapter,
+					       int location)
+{
+	struct tsnep_rxnfc_rule *rule;
+
+	list_for_each_entry(rule, &adapter->rxnfc_rules, list) {
+		if (rule->location == location)
+			return rule;
+		if (rule->location > location)
+			break;
+	}
+
+	return NULL;
+}
+
+static void tsnep_add_rule(struct tsnep_adapter *adapter,
+			   struct tsnep_rxnfc_rule *rule)
+{
+	struct tsnep_rxnfc_rule *pred, *cur;
+
+	tsnep_enable_rule(adapter, rule);
+
+	pred = NULL;
+	list_for_each_entry(cur, &adapter->rxnfc_rules, list) {
+		if (cur->location >= rule->location)
+			break;
+		pred = cur;
+	}
+
+	list_add(&rule->list, pred ? &pred->list : &adapter->rxnfc_rules);
+	adapter->rxnfc_count++;
+}
+
+static void tsnep_delete_rule(struct tsnep_adapter *adapter,
+			      struct tsnep_rxnfc_rule *rule)
+{
+	tsnep_disable_rule(adapter, rule);
+
+	list_del(&rule->list);
+	adapter->rxnfc_count--;
+
+	kfree(rule);
+}
+
+static void tsnep_flush_rules(struct tsnep_adapter *adapter)
+{
+	struct tsnep_rxnfc_rule *rule, *tmp;
+
+	mutex_lock(&adapter->rxnfc_lock);
+
+	list_for_each_entry_safe(rule, tmp, &adapter->rxnfc_rules, list)
+		tsnep_delete_rule(adapter, rule);
+
+	mutex_unlock(&adapter->rxnfc_lock);
+}
+
+int tsnep_rxnfc_get_rule(struct tsnep_adapter *adapter,
+			 struct ethtool_rxnfc *cmd)
+{
+	struct ethtool_rx_flow_spec *fsp = &cmd->fs;
+	struct tsnep_rxnfc_rule *rule = NULL;
+
+	cmd->data = adapter->rxnfc_max;
+
+	mutex_lock(&adapter->rxnfc_lock);
+
+	rule = tsnep_get_rule(adapter, fsp->location);
+	if (!rule) {
+		mutex_unlock(&adapter->rxnfc_lock);
+
+		return -ENOENT;
+	}
+
+	fsp->flow_type = ETHER_FLOW;
+	fsp->ring_cookie = rule->queue_index;
+
+	if (rule->filter.type == TSNEP_RXNFC_ETHER_TYPE) {
+		fsp->h_u.ether_spec.h_proto = htons(rule->filter.ether_type);
+		fsp->m_u.ether_spec.h_proto = ETHER_TYPE_FULL_MASK;
+	}
+
+	mutex_unlock(&adapter->rxnfc_lock);
+
+	return 0;
+}
+
+int tsnep_rxnfc_get_all(struct tsnep_adapter *adapter,
+			struct ethtool_rxnfc *cmd,
+			u32 *rule_locs)
+{
+	struct tsnep_rxnfc_rule *rule;
+	int count = 0;
+
+	cmd->data = adapter->rxnfc_max;
+
+	mutex_lock(&adapter->rxnfc_lock);
+
+	list_for_each_entry(rule, &adapter->rxnfc_rules, list) {
+		if (count == cmd->rule_cnt) {
+			mutex_unlock(&adapter->rxnfc_lock);
+
+			return -EMSGSIZE;
+		}
+
+		rule_locs[count] = rule->location;
+		count++;
+	}
+
+	mutex_unlock(&adapter->rxnfc_lock);
+
+	cmd->rule_cnt = count;
+
+	return 0;
+}
+
+static void tsnep_rxnfc_init_rule(struct tsnep_rxnfc_rule *rule,
+				  const struct ethtool_rx_flow_spec *fsp)
+{
+	INIT_LIST_HEAD(&rule->list);
+
+	rule->queue_index = fsp->ring_cookie;
+	rule->location = fsp->location;
+
+	rule->filter.type = TSNEP_RXNFC_ETHER_TYPE;
+	rule->filter.ether_type = ntohs(fsp->h_u.ether_spec.h_proto);
+}
+
+static int tsnep_rxnfc_check_rule(struct tsnep_adapter *adapter,
+				  struct tsnep_rxnfc_rule *rule)
+{
+	struct net_device *dev = adapter->netdev;
+	struct tsnep_rxnfc_rule *tmp;
+
+	list_for_each_entry(tmp, &adapter->rxnfc_rules, list) {
+		if (!memcmp(&rule->filter, &tmp->filter, sizeof(rule->filter)) &&
+		    tmp->location != rule->location) {
+			netdev_dbg(dev, "rule already exists\n");
+
+			return -EEXIST;
+		}
+	}
+
+	return 0;
+}
+
+int tsnep_rxnfc_add_rule(struct tsnep_adapter *adapter,
+			 struct ethtool_rxnfc *cmd)
+{
+	struct net_device *netdev = adapter->netdev;
+	struct ethtool_rx_flow_spec *fsp =
+		(struct ethtool_rx_flow_spec *)&cmd->fs;
+	struct tsnep_rxnfc_rule *rule, *old_rule;
+	int retval;
+
+	/* only EtherType is supported */
+	if (fsp->flow_type != ETHER_FLOW ||
+	    !is_zero_ether_addr(fsp->m_u.ether_spec.h_dest) ||
+	    !is_zero_ether_addr(fsp->m_u.ether_spec.h_source) ||
+	    fsp->m_u.ether_spec.h_proto != ETHER_TYPE_FULL_MASK) {
+		netdev_dbg(netdev, "only ethernet protocol is supported\n");
+
+		return -EOPNOTSUPP;
+	}
+
+	if (fsp->ring_cookie >
+	    (TSNEP_RX_ASSIGN_QUEUE_MASK >> TSNEP_RX_ASSIGN_QUEUE_SHIFT)) {
+		netdev_dbg(netdev, "invalid action\n");
+
+		return -EINVAL;
+	}
+
+	if (fsp->location >= adapter->rxnfc_max) {
+		netdev_dbg(netdev, "invalid location\n");
+
+		return -EINVAL;
+	}
+
+	rule = kzalloc(sizeof(*rule), GFP_KERNEL);
+	if (!rule)
+		return -ENOMEM;
+
+	tsnep_rxnfc_init_rule(rule, fsp);
+
+	mutex_lock(&adapter->rxnfc_lock);
+
+	retval = tsnep_rxnfc_check_rule(adapter, rule);
+	if (retval)
+		goto failed;
+
+	old_rule = tsnep_get_rule(adapter, fsp->location);
+	if (old_rule)
+		tsnep_delete_rule(adapter, old_rule);
+
+	tsnep_add_rule(adapter, rule);
+
+	mutex_unlock(&adapter->rxnfc_lock);
+
+	return 0;
+
+failed:
+	mutex_unlock(&adapter->rxnfc_lock);
+	kfree(rule);
+	return retval;
+}
+
+int tsnep_rxnfc_del_rule(struct tsnep_adapter *adapter,
+			 struct ethtool_rxnfc *cmd)
+{
+	struct ethtool_rx_flow_spec *fsp =
+		(struct ethtool_rx_flow_spec *)&cmd->fs;
+	struct tsnep_rxnfc_rule *rule;
+
+	mutex_lock(&adapter->rxnfc_lock);
+
+	rule = tsnep_get_rule(adapter, fsp->location);
+	if (!rule) {
+		mutex_unlock(&adapter->rxnfc_lock);
+
+		return -ENOENT;
+	}
+
+	tsnep_delete_rule(adapter, rule);
+
+	mutex_unlock(&adapter->rxnfc_lock);
+
+	return 0;
+}
+
+int tsnep_rxnfc_init(struct tsnep_adapter *adapter)
+{
+	int i;
+
+	/* disable all rules */
+	for (i = 0; i < adapter->rxnfc_max;
+	     i += sizeof(u32) / TSNEP_RX_ASSIGN_OFFSET)
+		iowrite32(0, adapter->addr + TSNEP_RX_ASSIGN + i);
+
+	return 0;
+}
+
+void tsnep_rxnfc_cleanup(struct tsnep_adapter *adapter)
+{
+	tsnep_flush_rules(adapter);
+}
-- 
2.30.2

[PATCH net-next v3 6/6] tsnep: Use page pool for RX

[Gerhard Engleder]

Use page pool for RX buffer handling. Makes RX path more efficient and
is required prework for future XDP support.

Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
---
 drivers/net/ethernet/engleder/Kconfig      |   1 +
 drivers/net/ethernet/engleder/tsnep.h      |   5 +-
 drivers/net/ethernet/engleder/tsnep_main.c | 162 ++++++++++++---------
 3 files changed, 100 insertions(+), 68 deletions(-)

diff --git a/drivers/net/ethernet/engleder/Kconfig b/drivers/net/ethernet/engleder/Kconfig
index f4e2b1102d8f..3df6bf476ae7 100644
--- a/drivers/net/ethernet/engleder/Kconfig
+++ b/drivers/net/ethernet/engleder/Kconfig
@@ -21,6 +21,7 @@ config TSNEP
 	depends on HAS_IOMEM && HAS_DMA
 	depends on PTP_1588_CLOCK_OPTIONAL
 	select PHYLIB
+	select PAGE_POOL
 	help
 	  Support for the Engleder TSN endpoint Ethernet MAC IP Core.
 
diff --git a/drivers/net/ethernet/engleder/tsnep.h b/drivers/net/ethernet/engleder/tsnep.h
index 2ca34ae9b55a..09a723b827c7 100644
--- a/drivers/net/ethernet/engleder/tsnep.h
+++ b/drivers/net/ethernet/engleder/tsnep.h
@@ -96,9 +96,9 @@ struct tsnep_rx_entry {
 
 	u32 properties;
 
-	struct sk_buff *skb;
+	struct page *page;
 	size_t len;
-	DEFINE_DMA_UNMAP_ADDR(dma);
+	dma_addr_t dma;
 };
 
 struct tsnep_rx {
@@ -113,6 +113,7 @@ struct tsnep_rx {
 	int read;
 	u32 owner_counter;
 	int increment_owner_counter;
+	struct page_pool *page_pool;
 
 	u32 packets;
 	u32 bytes;
diff --git a/drivers/net/ethernet/engleder/tsnep_main.c b/drivers/net/ethernet/engleder/tsnep_main.c
index a6b81f32d76b..8a93d0aa7faa 100644
--- a/drivers/net/ethernet/engleder/tsnep_main.c
+++ b/drivers/net/ethernet/engleder/tsnep_main.c
@@ -27,10 +27,10 @@
 #include <linux/phy.h>
 #include <linux/iopoll.h>
 
-#define RX_SKB_LENGTH (round_up(TSNEP_RX_INLINE_METADATA_SIZE + ETH_HLEN + \
-				TSNEP_MAX_FRAME_SIZE + ETH_FCS_LEN, 4))
-#define RX_SKB_RESERVE ((16 - TSNEP_RX_INLINE_METADATA_SIZE) + NET_IP_ALIGN)
-#define RX_SKB_ALLOC_LENGTH (RX_SKB_RESERVE + RX_SKB_LENGTH)
+#define TSNEP_SKB_PAD (NET_SKB_PAD + NET_IP_ALIGN)
+#define TSNEP_HEADROOM ALIGN(TSNEP_SKB_PAD, 4)
+#define TSNEP_MAX_RX_BUF_SIZE (PAGE_SIZE - TSNEP_HEADROOM - \
+			       SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
 
 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
 #define DMA_ADDR_HIGH(dma_addr) ((u32)(((dma_addr) >> 32) & 0xFFFFFFFF))
@@ -587,14 +587,15 @@ static void tsnep_rx_ring_cleanup(struct tsnep_rx *rx)
 
 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
 		entry = &rx->entry[i];
-		if (dma_unmap_addr(entry, dma))
-			dma_unmap_single(dmadev, dma_unmap_addr(entry, dma),
-					 dma_unmap_len(entry, len),
-					 DMA_FROM_DEVICE);
-		if (entry->skb)
-			dev_kfree_skb(entry->skb);
+		if (entry->page)
+			page_pool_put_full_page(rx->page_pool, entry->page,
+						false);
+		entry->page = NULL;
 	}
 
+	if (rx->page_pool)
+		page_pool_destroy(rx->page_pool);
+
 	memset(rx->entry, 0, sizeof(rx->entry));
 
 	for (i = 0; i < TSNEP_RING_PAGE_COUNT; i++) {
@@ -607,31 +608,19 @@ static void tsnep_rx_ring_cleanup(struct tsnep_rx *rx)
 	}
 }
 
-static int tsnep_rx_alloc_and_map_skb(struct tsnep_rx *rx,
-				      struct tsnep_rx_entry *entry)
+static int tsnep_rx_alloc_buffer(struct tsnep_rx *rx,
+				 struct tsnep_rx_entry *entry)
 {
-	struct device *dmadev = rx->adapter->dmadev;
-	struct sk_buff *skb;
-	dma_addr_t dma;
+	struct page *page;
 
-	skb = __netdev_alloc_skb(rx->adapter->netdev, RX_SKB_ALLOC_LENGTH,
-				 GFP_ATOMIC | GFP_DMA);
-	if (!skb)
+	page = page_pool_dev_alloc_pages(rx->page_pool);
+	if (unlikely(!page))
 		return -ENOMEM;
 
-	skb_reserve(skb, RX_SKB_RESERVE);
-
-	dma = dma_map_single(dmadev, skb->data, RX_SKB_LENGTH,
-			     DMA_FROM_DEVICE);
-	if (dma_mapping_error(dmadev, dma)) {
-		dev_kfree_skb(skb);
-		return -ENOMEM;
-	}
-
-	entry->skb = skb;
-	entry->len = RX_SKB_LENGTH;
-	dma_unmap_addr_set(entry, dma, dma);
-	entry->desc->rx = __cpu_to_le64(dma);
+	entry->page = page;
+	entry->len = TSNEP_MAX_RX_BUF_SIZE;
+	entry->dma = page_pool_get_dma_addr(entry->page);
+	entry->desc->rx = __cpu_to_le64(entry->dma + TSNEP_SKB_PAD);
 
 	return 0;
 }
@@ -640,6 +629,7 @@ static int tsnep_rx_ring_init(struct tsnep_rx *rx)
 {
 	struct device *dmadev = rx->adapter->dmadev;
 	struct tsnep_rx_entry *entry;
+	struct page_pool_params pp_params = { 0 };
 	struct tsnep_rx_entry *next_entry;
 	int i, j;
 	int retval;
@@ -661,12 +651,28 @@ static int tsnep_rx_ring_init(struct tsnep_rx *rx)
 			entry->desc_dma = rx->page_dma[i] + TSNEP_DESC_SIZE * j;
 		}
 	}
+
+	pp_params.flags = PP_FLAG_DMA_MAP | PP_FLAG_DMA_SYNC_DEV;
+	pp_params.order = 0;
+	pp_params.pool_size = TSNEP_RING_SIZE;
+	pp_params.nid = dev_to_node(dmadev);
+	pp_params.dev = dmadev;
+	pp_params.dma_dir = DMA_FROM_DEVICE;
+	pp_params.max_len = TSNEP_MAX_RX_BUF_SIZE;
+	pp_params.offset = TSNEP_SKB_PAD;
+	rx->page_pool = page_pool_create(&pp_params);
+	if (IS_ERR(rx->page_pool)) {
+		retval = PTR_ERR(rx->page_pool);
+		rx->page_pool = NULL;
+		goto failed;
+	}
+
 	for (i = 0; i < TSNEP_RING_SIZE; i++) {
 		entry = &rx->entry[i];
 		next_entry = &rx->entry[(i + 1) % TSNEP_RING_SIZE];
 		entry->desc->next = __cpu_to_le64(next_entry->desc_dma);
 
-		retval = tsnep_rx_alloc_and_map_skb(rx, entry);
+		retval = tsnep_rx_alloc_buffer(rx, entry);
 		if (retval)
 			goto failed;
 	}
@@ -682,7 +688,7 @@ static void tsnep_rx_activate(struct tsnep_rx *rx, int index)
 {
 	struct tsnep_rx_entry *entry = &rx->entry[index];
 
-	/* RX_SKB_LENGTH is a multiple of 4 */
+	/* TSNEP_MAX_RX_BUF_SIZE is a multiple of 4 */
 	entry->properties = entry->len & TSNEP_DESC_LENGTH_MASK;
 	entry->properties |= TSNEP_DESC_INTERRUPT_FLAG;
 	if (index == rx->increment_owner_counter) {
@@ -705,19 +711,52 @@ static void tsnep_rx_activate(struct tsnep_rx *rx, int index)
 	entry->desc->properties = __cpu_to_le32(entry->properties);
 }
 
+static struct sk_buff *tsnep_build_skb(struct tsnep_rx *rx, struct page *page,
+				       int length)
+{
+	struct sk_buff *skb;
+
+	skb = napi_build_skb(page_address(page), PAGE_SIZE);
+	if (unlikely(!skb))
+		return NULL;
+
+	/* update pointers within the skb to store the data */
+	skb_reserve(skb, TSNEP_SKB_PAD + TSNEP_RX_INLINE_METADATA_SIZE);
+	__skb_put(skb, length - TSNEP_RX_INLINE_METADATA_SIZE - ETH_FCS_LEN);
+
+	if (rx->adapter->hwtstamp_config.rx_filter == HWTSTAMP_FILTER_ALL) {
+		struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
+		struct tsnep_rx_inline *rx_inline =
+			(struct tsnep_rx_inline *)(page_address(page) +
+						   TSNEP_SKB_PAD);
+
+		skb_shinfo(skb)->tx_flags |=
+			SKBTX_HW_TSTAMP_NETDEV;
+		memset(hwtstamps, 0, sizeof(*hwtstamps));
+		hwtstamps->netdev_data = rx_inline;
+	}
+
+	skb_record_rx_queue(skb, rx->queue_index);
+	skb->protocol = eth_type_trans(skb, rx->adapter->netdev);
+
+	return skb;
+}
+
 static int tsnep_rx_poll(struct tsnep_rx *rx, struct napi_struct *napi,
 			 int budget)
 {
 	struct device *dmadev = rx->adapter->dmadev;
 	int done = 0;
+	enum dma_data_direction dma_dir;
 	struct tsnep_rx_entry *entry;
+	struct page *page;
 	struct sk_buff *skb;
-	size_t len;
-	dma_addr_t dma;
 	int length;
 	bool enable = false;
 	int retval;
 
+	dma_dir = page_pool_get_dma_dir(rx->page_pool);
+
 	while (likely(done < budget)) {
 		entry = &rx->entry[rx->read];
 		if ((__le32_to_cpu(entry->desc_wb->properties) &
@@ -730,43 +769,34 @@ static int tsnep_rx_poll(struct tsnep_rx *rx, struct napi_struct *napi,
 		 */
 		dma_rmb();
 
-		skb = entry->skb;
-		len = dma_unmap_len(entry, len);
-		dma = dma_unmap_addr(entry, dma);
+		prefetch(page_address(entry->page) + TSNEP_SKB_PAD);
+		length = __le32_to_cpu(entry->desc_wb->properties) &
+			 TSNEP_DESC_LENGTH_MASK;
+		dma_sync_single_range_for_cpu(dmadev, entry->dma, TSNEP_SKB_PAD,
+					      length, dma_dir);
+		page = entry->page;
 
 		/* forward skb only if allocation is successful, otherwise
-		 * skb is reused and frame dropped
+		 * page is reused and frame dropped
 		 */
-		retval = tsnep_rx_alloc_and_map_skb(rx, entry);
+		retval = tsnep_rx_alloc_buffer(rx, entry);
 		if (!retval) {
-			dma_unmap_single(dmadev, dma, len, DMA_FROM_DEVICE);
-
-			length = __le32_to_cpu(entry->desc_wb->properties) &
-				 TSNEP_DESC_LENGTH_MASK;
-			skb_put(skb, length - ETH_FCS_LEN);
-			if (rx->adapter->hwtstamp_config.rx_filter ==
-			    HWTSTAMP_FILTER_ALL) {
-				struct skb_shared_hwtstamps *hwtstamps =
-					skb_hwtstamps(skb);
-				struct tsnep_rx_inline *rx_inline =
-					(struct tsnep_rx_inline *)skb->data;
-
-				skb_shinfo(skb)->tx_flags |=
-					SKBTX_HW_TSTAMP_NETDEV;
-				memset(hwtstamps, 0, sizeof(*hwtstamps));
-				hwtstamps->netdev_data = rx_inline;
-			}
-			skb_pull(skb, TSNEP_RX_INLINE_METADATA_SIZE);
-			skb_record_rx_queue(skb, rx->queue_index);
-			skb->protocol = eth_type_trans(skb,
-						       rx->adapter->netdev);
+			skb = tsnep_build_skb(rx, page, length);
+			if (skb) {
+				page_pool_release_page(rx->page_pool, page);
+
+				rx->packets++;
+				rx->bytes += length -
+					     TSNEP_RX_INLINE_METADATA_SIZE;
+				if (skb->pkt_type == PACKET_MULTICAST)
+					rx->multicast++;
 
-			rx->packets++;
-			rx->bytes += length - TSNEP_RX_INLINE_METADATA_SIZE;
-			if (skb->pkt_type == PACKET_MULTICAST)
-				rx->multicast++;
+				napi_gro_receive(napi, skb);
+			} else {
+				page_pool_recycle_direct(rx->page_pool, page);
 
-			napi_gro_receive(napi, skb);
+				rx->dropped++;
+			}
 			done++;
 		} else {
 			rx->dropped++;
-- 
2.30.2