RE: [PATCH] dma: zynqmp_dma: Initialize descriptor list after freeing during reset

From: Harini Katakam
Date: Mon May 04 2020 - 01:33:30 EST


Hi Vinod,

> -----Original Message-----
> From: Vinod Koul [mailto:vkoul@xxxxxxxxxx]
> Sent: Monday, May 4, 2020 10:46 AM
> To: RafaÅ Hibner <rafal.hibner@xxxxxxxxxxxx>
> Cc: Appana Durga Kedareswara Rao <appanad@xxxxxxxxxx>; Radhey Shyam
> Pandey <radheys@xxxxxxxxxx>; Harini Katakam <harinik@xxxxxxxxxx>; Dan
> Williams <dan.j.williams@xxxxxxxxx>; Michal Simek <michals@xxxxxxxxxx>; open
> list:DMA GENERIC OFFLOAD ENGINE SUBSYSTEM
> <dmaengine@xxxxxxxxxxxxxxx>; moderated list:ARM/ZYNQ ARCHITECTURE
> <linux-arm-kernel@xxxxxxxxxxxxxxxxxxx>; open list <linux-
> kernel@xxxxxxxxxxxxxxx>
> Subject: Re: [PATCH] dma: zynqmp_dma: Initialize descriptor list after freeing
> during reset
>
> On 02-05-20, 15:00, RafaÅ Hibner wrote:
> > Hello Vinod,
> >
> > On 02.05.2020 14:32, Vinod Koul wrote:
> > > Would it not be better to use list_del_init() where we delete it
> > > rather than do the init here?
> > >
> >
> > It is not a problem of list element itself not being initialized.
> > The problem is that during fault conditions (zynqmp_dma_reset) all
> > elements are moved to free list. List head however is not reinitialized.
> >
> > In normal flow elements are removed by list_del and resubmitted to
> > free list with zynqmp_dma_free_descriptor.
> >
> > static void zynqmp_dma_chan_desc_cleanup(struct zynqmp_dma_chan *chan)
> > {
> > ÂÂÂ ...
> > ÂÂÂ list_for_each_entry_safe(desc, next, &chan->done_list, node) {
> > ÂÂÂÂÂÂÂ ...
> > ÂÂÂÂÂÂÂ list_del(&desc->node);
> > ÂÂÂÂÂÂÂ ...
> > ÂÂÂÂÂÂÂ zynqmp_dma_free_descriptor(chan, desc);
> > ÂÂÂ }
> > }
> >
> > The zynqmp_dma_free_descriptor does not delete elements from the list
> > by itself.
> > I am not he author of this driver so I fixed it by doing non intrusive
> > changes.
> >
> > Anyways, I do not see how using list_del_init would fix the bug.
>
> Looking at this, i think it would make sense to do list_splice_init() before we
> send the list to be freed.
>
> Radhey/Appana are cced, they should test this.

I tested this patch (I took over from Appana Kedar).
Once this discussion concludes and Rafal sends a v2, I'll test again and add my tag.

Regards,
Harini