Skip to contentSkip to conversation

net:ethernet: mtk_wed: setup WDMA_RING_TX(0) for non-DBDC MT7986

PATCH Superseded Zhi-Jun You 2 participants 3 messages Jun 30 – Jul 9

Subsystems (4)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net:ethernet: mtk_wed: setup WDMA_RING_TX(0) for non-DBDC MT7986

WDMA_RING_TX(0) is required to set MTK_WED_WDMA_RING_TX for WED RX but on a non-DBDC MT7986 it is never setup because idx is 1.

Setting MTK_WED_WDMA_RING_TX with WDMA_RING_TX(1) is not feasible because WED still tries to send through WDMA_RING_TX(0). This is verified with register dump.

Fix this by calling mtk_wed_wdma_tx_ring_setup if WDMA_RING_TX(0) is not setup and guard it with mtk_wed_is_v2.

Fixes 4c5de09eb0d0 ("net: ethernet: mtk_wed: add configure wed wo support")
---
 drivers/net/ethernet/mediatek/mtk_wed.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/net/ethernet/mediatek/mtk_wed.c +4 -0
index 10d9beaae372..2420557f6c96 100644
@@ -2334,6 +2334,10 @@ mtk_wed_start(struct mtk_wed_device *dev, u32 irq_mask)
 		if (!dev->rx_wdma[i].desc)
 			mtk_wed_wdma_rx_ring_setup(dev, i, 16, false);
 
+	if (mtk_wed_is_v2(dev->hw))
+		if (!dev->tx_wdma[0].desc)
+			mtk_wed_wdma_tx_ring_setup(dev, 0, MTK_WED_WDMA_RING_SIZE, false);
+
 	if (dev->wlan.hw_rro) {
 		for (i = 0; i < MTK_WED_RX_PAGE_QUEUES; i++) {
 			u32 addr = MTK_WED_RRO_MSDU_PG_CTRL0(i) +
Signature · 1 line
2.47.3
maintainer
#1 lore raw

From: AI Reviewer <[email protected]>

This is an AI-generated review of your patch. The human sending this email has considered the AI review valid, or at least plausible. Full review at: https://sashiko.dev/
---
net:ethernet: mtk_wed: setup WDMA_RING_TX(0) for non-DBDC MT7986

This commit updates mtk_wed_start() to configure WDMA_RING_TX(0) on non-DBDC MT7986 devices. It adds a call to mtk_wed_wdma_tx_ring_setup() for v2 hardware if the descriptor is not yet set up.

drivers/net/ethernet/mediatek/mtk_wed.c mtk_wed_start(struct mtk_wed_device *dev, u32 irq_mask) · line 2335

Quoted 7 lines
drivers/net/ethernet/mediatek/mtk_wed.c +0 -0
index 10d9beaae3721..2420557f6c966 100644
@@ -2334,6 +2334,10 @@ mtk_wed_start(struct mtk_wed_device *dev, u32 irq_mask)
 		if (!dev->rx_wdma[i].desc)
 			mtk_wed_wdma_rx_ring_setup(dev, i, 16, false);

[Severity: High]
This isn't a bug introduced by this patch, but does this loop need to check the return value of mtk_wed_wdma_rx_ring_setup()?

If mtk_wed_wdma_rx_ring_setup() returns -ENOMEM, the hardware registers are not programmed, but mtk_wed_start() continues execution and enables DMA.

Could this lead to hardware DMA using an uninitialized ring base address?

drivers/net/ethernet/mediatek/mtk_wed.c mtk_wed_start(struct mtk_wed_device *dev, u32 irq_mask) · line 2339

Quoted 4 lines
drivers/net/ethernet/mediatek/mtk_wed.c +3 -0
+	if (mtk_wed_is_v2(dev->hw))
+		if (!dev->tx_wdma[0].desc)
+			mtk_wed_wdma_tx_ring_setup(dev, 0, MTK_WED_WDMA_RING_SIZE, false);

[Severity: High]
Are there similar error handling concerns with this new call to mtk_wed_wdma_tx_ring_setup()?

Like the RX path, if dma_alloc_coherent() fails inside this setup function, it returns -ENOMEM without programming the hardware registers.

Since mtk_wed_start() doesn't check this return value, it will proceed to enable hardware DMA, which might attempt to use an uninitialized DMA address.

[Severity: High]
Does the check for !dev->tx_wdma[0].desc prevent necessary hardware initialization after a warm reset?

When the WLAN driver triggers a hardware reset, the WED hardware registers are cleared, and mtk_wed_start() is called again.

Because dev->tx_wdma[0].desc was allocated during the initial start, the guard will evaluate to false, skipping the hardware register reprogramming for tx_wdma[0].

Will the hardware DMA end up using a zeroed ring base address after a reset?

Quoted 4 lines
drivers/net/ethernet/mediatek/mtk_wed.c +1 -0
+
 	if (dev->wlan.hw_rro) {
 		for (i = 0; i < MTK_WED_RX_PAGE_QUEUES; i++) {
 			u32 addr = MTK_WED_RRO_MSDU_PG_CTRL0(i) +
Signature · 1 line
This is an AI-generated review.
#2 lore raw
Quoted 25 lines · Paolo Abeni

From: AI Reviewer <[email protected]>

This is an AI-generated review of your patch. The human sending this email has considered the AI review valid, or at least plausible. Full review at: https://sashiko.dev/
---
net:ethernet: mtk_wed: setup WDMA_RING_TX(0) for non-DBDC MT7986

This commit updates mtk_wed_start() to configure WDMA_RING_TX(0) on non-DBDC MT7986 devices. It adds a call to mtk_wed_wdma_tx_ring_setup() for v2 hardware if the descriptor is not yet set up.

drivers/net/ethernet/mediatek/mtk_wed.c mtk_wed_start(struct mtk_wed_device *dev, u32 irq_mask) · line 2335

Quoted 7 lines
drivers/net/ethernet/mediatek/mtk_wed.c +0 -0
index 10d9beaae3721..2420557f6c966 100644
@@ -2334,6 +2334,10 @@ mtk_wed_start(struct mtk_wed_device *dev, u32 irq_mask)
              if (!dev->rx_wdma[i].desc)
                      mtk_wed_wdma_rx_ring_setup(dev, i, 16, false);

[Severity: High]
This isn't a bug introduced by this patch, but does this loop need to check the return value of mtk_wed_wdma_rx_ring_setup()?

Quoted 4 lines

If mtk_wed_wdma_rx_ring_setup() returns -ENOMEM, the hardware registers are not programmed, but mtk_wed_start() continues execution and enables DMA.

Could this lead to hardware DMA using an uninitialized ring base address?

Hi,

I am not sure how to proceed with this one. mtk_wed_start is a void function and both mt7915 and mt7996's dma function that calls it don't have return value handling either.

Quoted 20 lines

drivers/net/ethernet/mediatek/mtk_wed.c mtk_wed_start(struct mtk_wed_device *dev, u32 irq_mask) · line 2339

Quoted 5 lines
drivers/net/ethernet/mediatek/mtk_wed.c +3 -0
+     if (mtk_wed_is_v2(dev->hw))
+             if (!dev->tx_wdma[0].desc)
+                     mtk_wed_wdma_tx_ring_setup(dev, 0, MTK_WED_WDMA_RING_SIZE, false);

[Severity: High]
Are there similar error handling concerns with this new call to mtk_wed_wdma_tx_ring_setup()?

Like the RX path, if dma_alloc_coherent() fails inside this setup function, it returns -ENOMEM without programming the hardware registers.

Since mtk_wed_start() doesn't check this return value, it will proceed to enable hardware DMA, which might attempt to use an uninitialized DMA address.

[Severity: High]
Does the check for !dev->tx_wdma[0].desc prevent necessary hardware initialization after a warm reset?

Quoted 8 lines

When the WLAN driver triggers a hardware reset, the WED hardware registers are cleared, and mtk_wed_start() is called again.

Because dev->tx_wdma[0].desc was allocated during the initial start, the guard will evaluate to false, skipping the hardware register reprogramming for tx_wdma[0].

Will the hardware DMA end up using a zeroed ring base address after a reset?

During warm reset, MTK_WDMA_RING_TX(i) + MTK_WED_RING_OFS_CPU_IDX is set to 0. Other parts are left untouched.
After the warm reset, mtk_wed_rx_ring_setup will be called with reset=true, which also calls
mtk_wed_wdma_tx_ring_setup with reset=true.

reset=true will skip ring allocation and call an extra mtk_wed_ring_reset but for non-DBDC mt7986, this path won't happen for the tx_wdma[0].

Usually rx_ring and tx_wdma are paired together but this is a special case that rx_ring[1] and
tx_wdma[1] are allocated but tx_wdma[0] is still required. Maybe something like this would work?
Runs only if wed is v2 and rx ring 0 is not allocated, reset only when tx_wdma[0] was already allocated.

if (mtk_wed_is_v2 && !dev->rx_ring[0].desc)
mtk_wed_wdma_tx_ring_setup(dev, 0, MTK_WED_WDMA_RING_SIZE,
!!dev->tx_wdma[0].desc);

Best regards,
Zhi-Jun

Quoted 8 lines

drivers/net/ethernet/mediatek/mtk_wed.c mtk_wed_start(struct mtk_wed_device *dev, u32 irq_mask) · line 2343

Quoted 5 lines
drivers/net/ethernet/mediatek/mtk_wed.c +1 -0
+
      if (dev->wlan.hw_rro) {
              for (i = 0; i < MTK_WED_RX_PAGE_QUEUES; i++) {
                      u32 addr = MTK_WED_RRO_MSDU_PG_CTRL0(i) +

--
This is an AI-generated review.