diff mbox series

[RFC,1/2,stable-6.1] dmaengine: at_hdmac: get next dma transfer from the right list

Message ID 15c92c2f-71e7-f4fd-b90b-412ab53e5a25@pcs.com (mailing list archive)
State Changes Requested
Headers show
Series [RFC,1/2,stable-6.1] dmaengine: at_hdmac: get next dma transfer from the right list | expand

Commit Message

Thomas Pfaff Nov. 14, 2023, 12:22 p.m. UTC
From: Thomas Pfaff <tpfaff@pcs.com>

In kernel 6.1, atc_advance_work and atc_handle_error are checking for the 
next dma transfer inside active list, but the descriptor is taken from the 
queue instead.

Signed-off-by: Thomas Pfaff <tpfaff@pcs.com>
---

Comments

Vinod Koul Nov. 24, 2023, 12:53 p.m. UTC | #1
On 14-11-23, 13:22, Thomas Pfaff wrote:
> From: Thomas Pfaff <tpfaff@pcs.com>
> 
> In kernel 6.1, atc_advance_work and atc_handle_error are checking for the 
> next dma transfer inside active list, but the descriptor is taken from the 
> queue instead.

Sorry that is not how this works. Please send the patch for mainline and
add a stable tag to the patches. They will be backported to stable
kernels

Also, your patch threading is broken, they appear disjoint and not as a
series

Thanks

> 
> Signed-off-by: Thomas Pfaff <tpfaff@pcs.com>
> ---
> diff --git a/drivers/dma/at_hdmac.c b/drivers/dma/at_hdmac.c
> index 858bd64f1313..68c1bfbefc5c 100644
> --- a/drivers/dma/at_hdmac.c
> +++ b/drivers/dma/at_hdmac.c
> @@ -490,6 +490,27 @@ atc_chain_complete(struct at_dma_chan *atchan, struct at_desc *desc)
>  	}
>  }
>  
> +/**
> + * atc_start_next - start next pending transaction if any
> + * @atchan: channel where the transaction ended
> + *
> + * Called with atchan->lock held
> + */
> +static void atc_start_next(struct at_dma_chan *atchan)
> +{
> +	struct at_desc *desc = NULL;
> +
> +	if (!list_empty(&atchan->active_list))
> +		desc = atc_first_active(atchan);
> +	else if (!list_empty(&atchan->queue)) {
> +		desc = atc_first_queued(atchan);
> +		list_move_tail(&desc->desc_node, &atchan->active_list);
> +	}
> +
> +	if (desc)
> +		atc_dostart(atchan, desc);
> +}
> +
>  /**
>   * atc_advance_work - at the end of a transaction, move forward
>   * @atchan: channel where the transaction ended
> @@ -513,11 +534,7 @@ static void atc_advance_work(struct at_dma_chan *atchan)
>  
>  	/* advance work */
>  	spin_lock_irqsave(&atchan->lock, flags);
> -	if (!list_empty(&atchan->active_list)) {
> -		desc = atc_first_queued(atchan);
> -		list_move_tail(&desc->desc_node, &atchan->active_list);
> -		atc_dostart(atchan, desc);
> -	}
> +	atc_start_next(atchan);
>  	spin_unlock_irqrestore(&atchan->lock, flags);
>  }
>  
> @@ -529,7 +546,6 @@ static void atc_advance_work(struct at_dma_chan *atchan)
>  static void atc_handle_error(struct at_dma_chan *atchan)
>  {
>  	struct at_desc *bad_desc;
> -	struct at_desc *desc;
>  	struct at_desc *child;
>  	unsigned long flags;
>  
> @@ -543,11 +559,7 @@ static void atc_handle_error(struct at_dma_chan *atchan)
>  	list_del_init(&bad_desc->desc_node);
>  
>  	/* Try to restart the controller */
> -	if (!list_empty(&atchan->active_list)) {
> -		desc = atc_first_queued(atchan);
> -		list_move_tail(&desc->desc_node, &atchan->active_list);
> -		atc_dostart(atchan, desc);
> -	}
> +	atc_start_next(atchan);
>  
>  	/*
>  	 * KERN_CRITICAL may seem harsh, but since this only happens
>
Thomas Pfaff Nov. 24, 2023, 1:16 p.m. UTC | #2
Thank you for your answer.

I was marking the patches as RFC and for stable 6.1 only.
Between 6.1 and mainline, at_hdmac was ported to use virt-dma, there is 
no possibility for my patches to go to mainline and being backported.
Most likely they are no longer needed in recent kernels.

Kind regards,
Thomas

On Fri, 24 Nov 2023, Vinod Koul wrote:

> On 14-11-23, 13:22, Thomas Pfaff wrote:
> > From: Thomas Pfaff <tpfaff@pcs.com>
> > 
> > In kernel 6.1, atc_advance_work and atc_handle_error are checking for the 
> > next dma transfer inside active list, but the descriptor is taken from the 
> > queue instead.
> 
> Sorry that is not how this works. Please send the patch for mainline and
> add a stable tag to the patches. They will be backported to stable
> kernels
> 
> Also, your patch threading is broken, they appear disjoint and not as a
> series
> 
> Thanks
> 
> > 
> > Signed-off-by: Thomas Pfaff <tpfaff@pcs.com>
> > ---
> > diff --git a/drivers/dma/at_hdmac.c b/drivers/dma/at_hdmac.c
> > index 858bd64f1313..68c1bfbefc5c 100644
> > --- a/drivers/dma/at_hdmac.c
> > +++ b/drivers/dma/at_hdmac.c
> > @@ -490,6 +490,27 @@ atc_chain_complete(struct at_dma_chan *atchan, struct at_desc *desc)
> >  	}
> >  }
> >  
> > +/**
> > + * atc_start_next - start next pending transaction if any
> > + * @atchan: channel where the transaction ended
> > + *
> > + * Called with atchan->lock held
> > + */
> > +static void atc_start_next(struct at_dma_chan *atchan)
> > +{
> > +	struct at_desc *desc = NULL;
> > +
> > +	if (!list_empty(&atchan->active_list))
> > +		desc = atc_first_active(atchan);
> > +	else if (!list_empty(&atchan->queue)) {
> > +		desc = atc_first_queued(atchan);
> > +		list_move_tail(&desc->desc_node, &atchan->active_list);
> > +	}
> > +
> > +	if (desc)
> > +		atc_dostart(atchan, desc);
> > +}
> > +
> >  /**
> >   * atc_advance_work - at the end of a transaction, move forward
> >   * @atchan: channel where the transaction ended
> > @@ -513,11 +534,7 @@ static void atc_advance_work(struct at_dma_chan *atchan)
> >  
> >  	/* advance work */
> >  	spin_lock_irqsave(&atchan->lock, flags);
> > -	if (!list_empty(&atchan->active_list)) {
> > -		desc = atc_first_queued(atchan);
> > -		list_move_tail(&desc->desc_node, &atchan->active_list);
> > -		atc_dostart(atchan, desc);
> > -	}
> > +	atc_start_next(atchan);
> >  	spin_unlock_irqrestore(&atchan->lock, flags);
> >  }
> >  
> > @@ -529,7 +546,6 @@ static void atc_advance_work(struct at_dma_chan *atchan)
> >  static void atc_handle_error(struct at_dma_chan *atchan)
> >  {
> >  	struct at_desc *bad_desc;
> > -	struct at_desc *desc;
> >  	struct at_desc *child;
> >  	unsigned long flags;
> >  
> > @@ -543,11 +559,7 @@ static void atc_handle_error(struct at_dma_chan *atchan)
> >  	list_del_init(&bad_desc->desc_node);
> >  
> >  	/* Try to restart the controller */
> > -	if (!list_empty(&atchan->active_list)) {
> > -		desc = atc_first_queued(atchan);
> > -		list_move_tail(&desc->desc_node, &atchan->active_list);
> > -		atc_dostart(atchan, desc);
> > -	}
> > +	atc_start_next(atchan);
> >  
> >  	/*
> >  	 * KERN_CRITICAL may seem harsh, but since this only happens
> > 
> 
> -- 
> ~Vinod
> 
>
diff mbox series

Patch

diff --git a/drivers/dma/at_hdmac.c b/drivers/dma/at_hdmac.c
index 858bd64f1313..68c1bfbefc5c 100644
--- a/drivers/dma/at_hdmac.c
+++ b/drivers/dma/at_hdmac.c
@@ -490,6 +490,27 @@  atc_chain_complete(struct at_dma_chan *atchan, struct at_desc *desc)
 	}
 }
 
+/**
+ * atc_start_next - start next pending transaction if any
+ * @atchan: channel where the transaction ended
+ *
+ * Called with atchan->lock held
+ */
+static void atc_start_next(struct at_dma_chan *atchan)
+{
+	struct at_desc *desc = NULL;
+
+	if (!list_empty(&atchan->active_list))
+		desc = atc_first_active(atchan);
+	else if (!list_empty(&atchan->queue)) {
+		desc = atc_first_queued(atchan);
+		list_move_tail(&desc->desc_node, &atchan->active_list);
+	}
+
+	if (desc)
+		atc_dostart(atchan, desc);
+}
+
 /**
  * atc_advance_work - at the end of a transaction, move forward
  * @atchan: channel where the transaction ended
@@ -513,11 +534,7 @@  static void atc_advance_work(struct at_dma_chan *atchan)
 
 	/* advance work */
 	spin_lock_irqsave(&atchan->lock, flags);
-	if (!list_empty(&atchan->active_list)) {
-		desc = atc_first_queued(atchan);
-		list_move_tail(&desc->desc_node, &atchan->active_list);
-		atc_dostart(atchan, desc);
-	}
+	atc_start_next(atchan);
 	spin_unlock_irqrestore(&atchan->lock, flags);
 }
 
@@ -529,7 +546,6 @@  static void atc_advance_work(struct at_dma_chan *atchan)
 static void atc_handle_error(struct at_dma_chan *atchan)
 {
 	struct at_desc *bad_desc;
-	struct at_desc *desc;
 	struct at_desc *child;
 	unsigned long flags;
 
@@ -543,11 +559,7 @@  static void atc_handle_error(struct at_dma_chan *atchan)
 	list_del_init(&bad_desc->desc_node);
 
 	/* Try to restart the controller */
-	if (!list_empty(&atchan->active_list)) {
-		desc = atc_first_queued(atchan);
-		list_move_tail(&desc->desc_node, &atchan->active_list);
-		atc_dostart(atchan, desc);
-	}
+	atc_start_next(atchan);
 
 	/*
 	 * KERN_CRITICAL may seem harsh, but since this only happens