Re: [PATCH v9 03/11] dmaengine: qcom: bam_dma: implement support for BAM locking

[Vinod Koul]

On 16-12-25, 16:00, Bartosz Golaszewski wrote:
> On Tue, Dec 16, 2025 at 2:00 PM Vinod Koul <vkoul@xxxxxxxxxx> wrote:
> >
> > On 28-11-25, 12:44, Bartosz Golaszewski wrote:
> > > From: Bartosz Golaszewski <bartosz.golaszewski@xxxxxxxxxx>
> > >
> > > Use metadata operations in DMA descriptors to allow BAM users to pass
> > > additional information to the engine. To that end: define a new
> > > structure - struct bam_desc_metadata - as a medium and define two new
> > > commands: for locking and unlocking the BAM respectively. Handle the
> > > locking in the .attach() callback.
> > >
> > > Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@xxxxxxxxxx>
> > > ---
> > >  drivers/dma/qcom/bam_dma.c       | 59 +++++++++++++++++++++++++++++++++++++++-
> > >  include/linux/dma/qcom_bam_dma.h | 12 ++++++++
> > >  2 files changed, 70 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/dma/qcom/bam_dma.c b/drivers/dma/qcom/bam_dma.c
> > > index c9ae1fffe44d79c5eb59b8bbf7f147a8fa3aa0bd..d1dc80b29818897b333cd223ec7306a169cc51fd 100644
> > > --- a/drivers/dma/qcom/bam_dma.c
> > > +++ b/drivers/dma/qcom/bam_dma.c
> > > @@ -30,6 +30,7 @@
> > >  #include <linux/module.h>
> > >  #include <linux/interrupt.h>
> > >  #include <linux/dma-mapping.h>
> > > +#include <linux/dma/qcom_bam_dma.h>
> > >  #include <linux/scatterlist.h>
> > >  #include <linux/device.h>
> > >  #include <linux/platform_device.h>
> > > @@ -391,6 +392,8 @@ struct bam_chan {
> > >       struct list_head desc_list;
> > >
> > >       struct list_head node;
> > > +
> > > +     bool bam_locked;
> > >  };
> > >
> > >  static inline struct bam_chan *to_bam_chan(struct dma_chan *common)
> > > @@ -655,6 +658,53 @@ static int bam_slave_config(struct dma_chan *chan,
> > >       return 0;
> > >  }
> > >
> > > +static int bam_metadata_attach(struct dma_async_tx_descriptor *desc, void *data, size_t len)
> > > +{
> > > +     struct virt_dma_desc *vd = container_of(desc, struct virt_dma_desc, tx);
> > > +     struct bam_async_desc *async_desc = container_of(vd, struct bam_async_desc,  vd);
> > > +     struct bam_desc_hw *hw_desc = async_desc->desc;
> > > +     struct bam_desc_metadata *metadata = data;
> > > +     struct bam_chan *bchan = to_bam_chan(metadata->chan);
> > > +     struct bam_device *bdev = bchan->bdev;
> > > +
> > > +     if (!data)
> > > +             return -EINVAL;
> > > +
> > > +     if (metadata->op == BAM_META_CMD_LOCK || metadata->op == BAM_META_CMD_UNLOCK) {
> > > +             if (!bdev->dev_data->bam_pipe_lock)
> > > +                     return -EOPNOTSUPP;
> > > +
> > > +             /* Expecting a dummy write when locking, only one descriptor allowed. */
> > > +             if (async_desc->num_desc != 1)
> > > +                     return -EINVAL;
> > > +     }
> > > +
> > > +     switch (metadata->op) {
> > > +     case BAM_META_CMD_LOCK:
> > > +             if (bchan->bam_locked)
> > > +                     return -EBUSY;
> > > +
> > > +             hw_desc->flags |= DESC_FLAG_LOCK;
> >
> > Why does this flag imply for the hardware.

s/Why/What !
> 
> Please rephrase, I don't get what you mean.

I am trying to understand what the flag refers to and why do you need
this.. What is the problem that lock tries to solve

> >
> > I do not like the interface designed here. This is overloading. Can we
> > look at doing something better here.
> >
> 
> It used to be a generic flag in dmaengine visible for all users.
> Dmitry argued that it's too Qualcomm-specific for a generic flag and
> suggested using the metadata to hide the communication between the QCE
> and BAM drivers. I'm open to other suggestions but it has to be a bit
> more specific than "do something better". :)
> 
> Bartosz

-- 
~Vinod