[PATCH net] dmaengine: ti: k3-udma-glue: fix k3_udma_glue_tx_get_irq() error checking

[PATCH net] dmaengine: ti: k3-udma-glue: fix k3_udma_glue_tx_get_irq() error checking

[Dan Carpenter]

The problem is that xudma_pktdma_tflow_get_irq() returns zero on error
and k3_ringacc_get_ring_irq_num() returns negatives.  This complicates
the error handling.  Change it to always return negative errors.

Both callers have other bugs as well.  The am65_cpsw_nuss_init_tx_chns()
function doesn't preserve the error code but instead returns success.
In prueth_init_tx_chns() there is a signedness bug since "tx_chn->irq"
is unsigned so negative errors are not handled correctly.

Fixes: 93a76530316a ("net: ethernet: ti: introduce am65x/j721e gigabit eth subsystem driver")
Fixes: 5b65781d06ea ("dmaengine: ti: k3-udma-glue: Add support for K3 PKTDMA")
Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
---
The k3_udma_glue_tx_get_irq() function is in dmaengine, but this bug
affects networking so I think the fix should go through networking.

 drivers/dma/ti/k3-udma-glue.c                | 3 +++
 drivers/net/ethernet/ti/am65-cpsw-nuss.c     | 3 ++-
 drivers/net/ethernet/ti/icssg/icssg_prueth.c | 6 +++---
 3 files changed, 8 insertions(+), 4 deletions(-)

diff --git a/drivers/dma/ti/k3-udma-glue.c b/drivers/dma/ti/k3-udma-glue.c
index 789193ed0386..c278d5facf7d 100644
--- a/drivers/dma/ti/k3-udma-glue.c
+++ b/drivers/dma/ti/k3-udma-glue.c
@@ -558,6 +558,9 @@ int k3_udma_glue_tx_get_irq(struct k3_udma_glue_tx_channel *tx_chn)
 		tx_chn->virq = k3_ringacc_get_ring_irq_num(tx_chn->ringtxcq);
 	}
 
+	if (!tx_chn->virq)
+		return -ENXIO;
+
 	return tx_chn->virq;
 }
 EXPORT_SYMBOL_GPL(k3_udma_glue_tx_get_irq);
diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.c b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
index bea6fc0f324c..24120605502f 100644
--- a/drivers/net/ethernet/ti/am65-cpsw-nuss.c
+++ b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
@@ -1747,9 +1747,10 @@ static int am65_cpsw_nuss_init_tx_chns(struct am65_cpsw_common *common)
 		}
 
 		tx_chn->irq = k3_udma_glue_tx_get_irq(tx_chn->tx_chn);
-		if (tx_chn->irq <= 0) {
+		if (tx_chn->irq < 0) {
 			dev_err(dev, "Failed to get tx dma irq %d\n",
 				tx_chn->irq);
+			ret = tx_chn->irq;
 			goto err;
 		}
 
diff --git a/drivers/net/ethernet/ti/icssg/icssg_prueth.c b/drivers/net/ethernet/ti/icssg/icssg_prueth.c
index 47b941fb0198..98b061a3f3c1 100644
--- a/drivers/net/ethernet/ti/icssg/icssg_prueth.c
+++ b/drivers/net/ethernet/ti/icssg/icssg_prueth.c
@@ -316,12 +316,12 @@ static int prueth_init_tx_chns(struct prueth_emac *emac)
 			goto fail;
 		}
 
-		tx_chn->irq = k3_udma_glue_tx_get_irq(tx_chn->tx_chn);
-		if (tx_chn->irq <= 0) {
-			ret = -EINVAL;
+		ret = k3_udma_glue_tx_get_irq(tx_chn->tx_chn);
+		if (ret < 0) {
 			netdev_err(ndev, "failed to get tx irq\n");
 			goto fail;
 		}
+		tx_chn->irq = ret;
 
 		snprintf(tx_chn->name, sizeof(tx_chn->name), "%s-tx%d",
 			 dev_name(dev), tx_chn->id);
-- 
2.39.2

Re: [PATCH net] dmaengine: ti: k3-udma-glue: fix k3_udma_glue_tx_get_irq() error checking

[Simon Horman]

On Wed, Sep 13, 2023 at 02:05:31PM +0300, Dan Carpenter wrote:
> The problem is that xudma_pktdma_tflow_get_irq() returns zero on error
> and k3_ringacc_get_ring_irq_num() returns negatives.  This complicates
> the error handling.  Change it to always return negative errors.
> 
> Both callers have other bugs as well.  The am65_cpsw_nuss_init_tx_chns()
> function doesn't preserve the error code but instead returns success.
> In prueth_init_tx_chns() there is a signedness bug since "tx_chn->irq"
> is unsigned so negative errors are not handled correctly.

Hi Dan,

I understand that the problems are related, but there are several of them.
Could they be handled in separate patches (applied in a specific order) ?
I suspect this would aid backporting, and, moreover, I think it is nice
to try to work on a one-fix-per-patch basis.

The above notwithstanding, I do agree with the correctness of your changes.

> Fixes: 93a76530316a ("net: ethernet: ti: introduce am65x/j721e gigabit eth subsystem driver")
> Fixes: 5b65781d06ea ("dmaengine: ti: k3-udma-glue: Add support for K3 PKTDMA")
> Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
> ---
> The k3_udma_glue_tx_get_irq() function is in dmaengine, but this bug
> affects networking so I think the fix should go through networking.

...

Re: [PATCH net] dmaengine: ti: k3-udma-glue: fix k3_udma_glue_tx_get_irq() error checking

[Dan Carpenter]

On Fri, Sep 15, 2023 at 08:33:24AM +0200, Simon Horman wrote:
> On Wed, Sep 13, 2023 at 02:05:31PM +0300, Dan Carpenter wrote:
> > The problem is that xudma_pktdma_tflow_get_irq() returns zero on error
> > and k3_ringacc_get_ring_irq_num() returns negatives.  This complicates
> > the error handling.  Change it to always return negative errors.
> > 
> > Both callers have other bugs as well.  The am65_cpsw_nuss_init_tx_chns()
> > function doesn't preserve the error code but instead returns success.
> > In prueth_init_tx_chns() there is a signedness bug since "tx_chn->irq"
> > is unsigned so negative errors are not handled correctly.
> 
> Hi Dan,
> 
> I understand that the problems are related, but there are several of them.
> Could they be handled in separate patches (applied in a specific order) ?
> I suspect this would aid backporting, and, moreover, I think it is nice
> to try to work on a one-fix-per-patch basis.
> 
> The above notwithstanding, I do agree with the correctness of your changes.
> 

Sure.  Let me write it like:

patch 1: fix first caller
patch 2: fix second caller
patch 3: re-write both callers to cleaner API

And we can push everything to net because it's nice to have one version
of the API instead of a version for net and a different version in
net-next.  Or we could apply patch 3 to only net-next.

regards,
dan carpenter