[PATCH net] net: mtk: wed: add dma mask limitation and GFP_DMA32 for device with more than 4GB DRAM

[PATCH net] net: mtk: wed: add dma mask limitation and GFP_DMA32 for device with more than 4GB DRAM

[Lorenzo Bianconi]

From: Rex Lu <rex.lu@mediatek.com>

Limit tx/rx buffer address to 32-bit address space for board with more
than 4GB DRAM.

Tested-by: Daniel Pawlik <pawlik.dan@gmail.com>
Tested-by: Matteo Croce <teknoraver@meta.com>
Signed-off-by: Rex Lu <rex.lu@mediatek.com>
Co-developed-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
---
 drivers/net/ethernet/mediatek/mtk_wed.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/mediatek/mtk_wed.c b/drivers/net/ethernet/mediatek/mtk_wed.c
index 3dbb113b792cf00fb4f89ab20f7e7fa72ecac260..cc3f3380a9980a6525921205dc7adaf321a099ae 100644
--- a/drivers/net/ethernet/mediatek/mtk_wed.c
+++ b/drivers/net/ethernet/mediatek/mtk_wed.c
@@ -677,7 +677,7 @@ mtk_wed_tx_buffer_alloc(struct mtk_wed_device *dev)
 		void *buf;
 		int s;
 
-		page = __dev_alloc_page(GFP_KERNEL);
+		page = __dev_alloc_page(GFP_KERNEL | GFP_DMA32);
 		if (!page)
 			return -ENOMEM;
 
@@ -800,7 +800,7 @@ mtk_wed_hwrro_buffer_alloc(struct mtk_wed_device *dev)
 		struct page *page;
 		int s;
 
-		page = __dev_alloc_page(GFP_KERNEL);
+		page = __dev_alloc_page(GFP_KERNEL | GFP_DMA32);
 		if (!page)
 			return -ENOMEM;
 
@@ -2426,6 +2426,10 @@ mtk_wed_attach(struct mtk_wed_device *dev)
 	dev->version = hw->version;
 	dev->hw->pcie_base = mtk_wed_get_pcie_base(dev);
 
+	ret = dma_set_mask_and_coherent(hw->dev, DMA_BIT_MASK(32));
+	if (ret)
+		return ret;
+
 	if (hw->eth->dma_dev == hw->eth->dev &&
 	    of_dma_is_coherent(hw->eth->dev->of_node))
 		mtk_eth_set_dma_device(hw->eth, hw->dev);

---
base-commit: 07fdad3a93756b872da7b53647715c48d0f4a2d0
change-id: 20250930-wed-4g-ram-limitation-6490e7ed9d39

Best regards,
-- 
Lorenzo Bianconi <lorenzo@kernel.org>

Re: [PATCH net] net: mtk: wed: add dma mask limitation and GFP_DMA32 for device with more than 4GB DRAM

[Simon Horman]

On Wed, Oct 08, 2025 at 12:48:05PM +0200, Lorenzo Bianconi wrote:
> From: Rex Lu <rex.lu@mediatek.com>
> 
> Limit tx/rx buffer address to 32-bit address space for board with more
> than 4GB DRAM.
> 

Hi Lorenzo, Rex, all,

As a fix for net a Fixes tag should probably go here.
> Tested-by: Daniel Pawlik <pawlik.dan@gmail.com>
> Tested-by: Matteo Croce <teknoraver@meta.com>
> Signed-off-by: Rex Lu <rex.lu@mediatek.com>
> Co-developed-by: Lorenzo Bianconi <lorenzo@kernel.org>
> Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>

...

> @@ -2426,6 +2426,10 @@ mtk_wed_attach(struct mtk_wed_device *dev)
>  	dev->version = hw->version;
>  	dev->hw->pcie_base = mtk_wed_get_pcie_base(dev);
>  
> +	ret = dma_set_mask_and_coherent(hw->dev, DMA_BIT_MASK(32));
> +	if (ret)
> +		return ret;

I think 'goto out' is needed here to avoid leaking hw_lock.

> +
>  	if (hw->eth->dma_dev == hw->eth->dev &&
>  	    of_dma_is_coherent(hw->eth->dev->of_node))
>  		mtk_eth_set_dma_device(hw->eth, hw->dev);
> 

-- 
pw-bot: changes-requested

Re: [PATCH net] net: mtk: wed: add dma mask limitation and GFP_DMA32 for device with more than 4GB DRAM

[Lorenzo Bianconi]

[-- Attachment #1: Type: text/plain, Size: 1209 bytes --]

> On Wed, Oct 08, 2025 at 12:48:05PM +0200, Lorenzo Bianconi wrote:
> > From: Rex Lu <rex.lu@mediatek.com>
> > 
> > Limit tx/rx buffer address to 32-bit address space for board with more
> > than 4GB DRAM.
> > 
> 
> Hi Lorenzo, Rex, all,

Hi Simon,

> 
> As a fix for net a Fixes tag should probably go here.
> > Tested-by: Daniel Pawlik <pawlik.dan@gmail.com>
> > Tested-by: Matteo Croce <teknoraver@meta.com>
> > Signed-off-by: Rex Lu <rex.lu@mediatek.com>
> > Co-developed-by: Lorenzo Bianconi <lorenzo@kernel.org>
> > Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> 
> ...
> 
> > @@ -2426,6 +2426,10 @@ mtk_wed_attach(struct mtk_wed_device *dev)
> >  	dev->version = hw->version;
> >  	dev->hw->pcie_base = mtk_wed_get_pcie_base(dev);
> >  
> > +	ret = dma_set_mask_and_coherent(hw->dev, DMA_BIT_MASK(32));
> > +	if (ret)
> > +		return ret;
> 
> I think 'goto out' is needed here to avoid leaking hw_lock.

right, I will fix it in v2.

Regards,
Lorenzo

> 
> > +
> >  	if (hw->eth->dma_dev == hw->eth->dev &&
> >  	    of_dma_is_coherent(hw->eth->dev->of_node))
> >  		mtk_eth_set_dma_device(hw->eth, hw->dev);
> > 
> 
> -- 
> pw-bot: changes-requested

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]