diff mbox series

[V2] ethernet: stmmac: support driver work for DTs without child queue node

Message ID 20220429004605.1010751-1-jay.xu@rock-chips.com (mailing list archive)
State New
Headers show
Series [V2] ethernet: stmmac: support driver work for DTs without child queue node | expand

Commit Message

jay.xu@rock-chips.com April 29, 2022, 12:46 a.m. UTC
The driver use the value of property 'snps,rx-queues-to-use' to loop
same numbers child nodes as queues, such as:

    gmac {
        rx-queues-config {
            snps,rx-queues-to-use = <1>;
            queue0 {
                // nothing need here.
	    };
	};
    };

Since a patch for dtc from rockchip will delete all node without any
properties or child node, the queue0 node will be deleted, that caused
the driver fail to probe:

    rk_gmac-dwmac: probe of ffa80000.ethernet failed with error -22

This patch try to support driver work well for DTs without setting for
the child queue nodes and then have none child queue nodes.

Signed-off-by: Jianqun Xu <jay.xu@rock-chips.com>
---
v2:
 - change subject and update commit comment, sugguested by Kicinski

 drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Andrew Lunn April 29, 2022, 12:56 a.m. UTC | #1
On Fri, Apr 29, 2022 at 08:46:05AM +0800, Jianqun Xu wrote:
> The driver use the value of property 'snps,rx-queues-to-use' to loop
> same numbers child nodes as queues, such as:
> 
>     gmac {
>         rx-queues-config {
>             snps,rx-queues-to-use = <1>;
>             queue0 {
>                 // nothing need here.
> 	    };
> 	};
>     };
> 
> Since a patch for dtc from rockchip will delete all node without any
> properties or child node, the queue0 node will be deleted, that caused
> the driver fail to probe:

Is this the in tree dtc? Do you have a commit hash for it? That should
probably be used as a Fixes: tag. Or that change to dtc needs
reverting because it breaks stuff.

	  Andrew
Jakub Kicinski April 29, 2022, 1:40 a.m. UTC | #2
On Fri, 29 Apr 2022 08:46:05 +0800 Jianqun Xu wrote:
> The driver use the value of property 'snps,rx-queues-to-use' to loop
> same numbers child nodes as queues, such as:
> 
>     gmac {
>         rx-queues-config {
>             snps,rx-queues-to-use = <1>;
>             queue0 {
>                 // nothing need here.
> 	    };
> 	};
>     };

I think you mean tx, not rx, given the code.

>  
>  		queue++;
>  	}
> -	if (queue != plat->tx_queues_to_use) {
> +	if (queue != plat->tx_queues_to_use && of_get_child_count(tx_node)) {
>  		ret = -EINVAL;
>  		dev_err(&pdev->dev, "Not all TX queues were configured\n");
>  		goto out;

Also what about the init to defaults I asked about?
jay.xu@rock-chips.com April 29, 2022, 1:54 a.m. UTC | #3
Hi, andrew

--------------
jay.xu@rock-chips.com
>On Fri, Apr 29, 2022 at 08:46:05AM +0800, Jianqun Xu wrote:
>> The driver use the value of property 'snps,rx-queues-to-use' to loop
>> same numbers child nodes as queues, such as:
>>
>>     gmac {
>>         rx-queues-config {
>>             snps,rx-queues-to-use = <1>;
>>             queue0 {
>>                 // nothing need here.
>>     };
>> };
>>     };
>>
>> Since a patch for dtc from rockchip will delete all node without any
>> properties or child node, the queue0 node will be deleted, that caused
>> the driver fail to probe:
>
>Is this the in tree dtc? Do you have a commit hash for it? That should
>probably be used as a Fixes: tag. Or that change to dtc needs
>reverting because it breaks stuff.
> 
The patch is a hack patch for some products and have not in tree dtc, I said that to
explain a possible case how things happed, it's only a case of no child queue node.

In my mind, it's hard to understand the one property (snps,rx-queues-to-use) does
have a limit for another nodes or properties, if it must work well, the dtc is better to
do a abort, like #size-cells ?

My personally question from a insigth of DTs but gmac, if it's not a good question,
I can simply add a status = "okay" for the empty node and drop this patch.

>	  Andrew
Andrew Lunn April 29, 2022, 12:14 p.m. UTC | #4
Adding device tree list.

This is mostly a device tree issue, so the device tree Maintainers
should be part of the discussion.

> >On Fri, Apr 29, 2022 at 08:46:05AM +0800, Jianqun Xu wrote:
> >> The driver use the value of property 'snps,rx-queues-to-use' to loop
> >> same numbers child nodes as queues, such as:
> >>
> >>     gmac {
> >>         rx-queues-config {
> >>             snps,rx-queues-to-use = <1>;
> >>             queue0 {
> >>                 // nothing need here.
> >>     };
> >> };
> >>     };
> >>
> >> Since a patch for dtc from rockchip will delete all node without any
> >> properties or child node, the queue0 node will be deleted, that caused
> >> the driver fail to probe:
> >
> >Is this the in tree dtc? Do you have a commit hash for it? That should
> >probably be used as a Fixes: tag. Or that change to dtc needs
> >reverting because it breaks stuff.
> > 
> The patch is a hack patch for some products and have not in tree dtc, I said that to
> explain a possible case how things happed, it's only a case of no child queue node.

So this has nothing to do with the kernel dtc, or the upstream
dtc. This is only a 'vendor crap' dtc which has been hacked?

Why should mainline care? Is there anything in the DT standard which
says the compiler can optimise out empty properties?

     Andrew
jay.xu@rock-chips.com May 1, 2022, 1:40 a.m. UTC | #5
Hi Andrew

--------------
jay.xu@rock-chips.com
>Adding device tree list.
>
>This is mostly a device tree issue, so the device tree Maintainers
>should be part of the discussion.
>
>> >On Fri, Apr 29, 2022 at 08:46:05AM +0800, Jianqun Xu wrote:
>> >> The driver use the value of property 'snps,rx-queues-to-use' to loop
>> >> same numbers child nodes as queues, such as:
>> >>
>> >>     gmac {
>> >>         rx-queues-config {
>> >>             snps,rx-queues-to-use = <1>;
>> >>             queue0 {
>> >>                 // nothing need here.
>> >>     };
>> >> };
>> >>     };
>> >>
>> >> Since a patch for dtc from rockchip will delete all node without any
>> >> properties or child node, the queue0 node will be deleted, that caused
>> >> the driver fail to probe:
>> >
>> >Is this the in tree dtc? Do you have a commit hash for it? That should
>> >probably be used as a Fixes: tag. Or that change to dtc needs
>> >reverting because it breaks stuff.
>> >
>> The patch is a hack patch for some products and have not in tree dtc, I said that to
>> explain a possible case how things happed, it's only a case of no child queue node.
>
>So this has nothing to do with the kernel dtc, or the upstream
>dtc. This is only a 'vendor crap' dtc which has been hacked?
>
>Why should mainline care? Is there anything in the DT standard which
>says the compiler can optimise out empty properties? 

Yes, it't nothing about what kind of dtc, does about what rule for properties, now the one property
request another child node must exist and number of child node must equal to the value of the property,
the 
#address-cells = <1>;
#size-cells = <1>;
does this but they are in special key words, the gmac has no likely key word.

>
>     Andrew
diff mbox series

Patch

diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c
index 2d8c095f3856..4f01a41c485c 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c
@@ -279,7 +279,7 @@  static int stmmac_mtl_setup(struct platform_device *pdev,
 
 		queue++;
 	}
-	if (queue != plat->tx_queues_to_use) {
+	if (queue != plat->tx_queues_to_use && of_get_child_count(tx_node)) {
 		ret = -EINVAL;
 		dev_err(&pdev->dev, "Not all TX queues were configured\n");
 		goto out;