Re: [PATCH v9 19/27] dmaengine: dw-edma: Use non-atomic io-64 methods

[Bjorn Helgaas]

On Sun, Jan 22, 2023 at 02:03:59AM +0300, Serge Semin wrote:
> On Fri, Jan 20, 2023 at 06:54:01PM -0600, Bjorn Helgaas wrote:
> > On Fri, Jan 13, 2023 at 08:14:01PM +0300, Serge Semin wrote:
> > > Instead of splitting the 64-bits IOs up into two 32-bits ones it's
> > > possible to use the already available non-atomic readq/writeq methods
> > > implemented exactly for such cases. They are defined in the dedicated
> > > header files io-64-nonatomic-lo-hi.h/io-64-nonatomic-hi-lo.h. So in case
> > > if the 64-bits readq/writeq methods are unavailable on some platforms at
> > > consideration, the corresponding drivers can have any of these headers
> > > included and stop locally re-implementing the 64-bits IO accessors taking
> > > into account the non-atomic nature of the included methods. Let's do that
> > > in the DW eDMA driver too. Note by doing so we can discard the
> > > CONFIG_64BIT config ifdefs from the code.
> > > 
> > > Signed-off-by: Serge Semin <Sergey.Semin@xxxxxxxxxxxxxxxxxxxx>
> > > Reviewed-by: Manivannan Sadhasivam <manivannan.sadhasivam@xxxxxxxxxx>
> > > Tested-by: Manivannan Sadhasivam <manivannan.sadhasivam@xxxxxxxxxx>
> > > Acked-by: Vinod Koul <vkoul@xxxxxxxxxx>
> > > ---
> > >  drivers/dma/dw-edma/dw-edma-v0-core.c | 55 +++++++++------------------
> > >  1 file changed, 18 insertions(+), 37 deletions(-)
> > > 
> > > diff --git a/drivers/dma/dw-edma/dw-edma-v0-core.c b/drivers/dma/dw-edma/dw-edma-v0-core.c
> > > index 66f296daac5a..51a34b43434c 100644
> > > --- a/drivers/dma/dw-edma/dw-edma-v0-core.c
> > > +++ b/drivers/dma/dw-edma/dw-edma-v0-core.c
> > > @@ -8,6 +8,8 @@
> > >  
> > >  #include <linux/bitfield.h>
> > >  
> > > +#include <linux/io-64-nonatomic-lo-hi.h>
> > > +
> > >  #include "dw-edma-core.h"
> > >  #include "dw-edma-v0-core.h"
> > >  #include "dw-edma-v0-regs.h"
> > > @@ -53,8 +55,6 @@ static inline struct dw_edma_v0_regs __iomem *__dw_regs(struct dw_edma *dw)
> > >  		SET_32(dw, rd_##name, value);		\
> > >  	} while (0)
> > >  
> > > -#ifdef CONFIG_64BIT
> > > -
> > >  #define SET_64(dw, name, value)				\
> > >  	writeq(value, &(__dw_regs(dw)->name))
> > >  
> > > @@ -80,8 +80,6 @@ static inline struct dw_edma_v0_regs __iomem *__dw_regs(struct dw_edma *dw)
> > >  		SET_64(dw, rd_##name, value);		\
> > >  	} while (0)
> > >  
> > > -#endif /* CONFIG_64BIT */
> > 
> 
> > Great to get rid of these #ifdefs!
> > 
> > Am I missing something?  It looks like SET_64 is used only by
> > SET_RW_64 and SET_BOTH_64, and neither of *them is used at all.
> > 
> > Similarly for GET_64 and GET_RW_64.
> > 
> > So maybe we could get rid of everything inside the #ifdefs as well?
> 
> Even though these macros are indeed unused in the driver they are
> still a part of the DW eDMA CSRs access interface. In particular they
> are supposed to be used to access the 64-bit registers declared in the
> dw_edma_v0_regs, dw_edma_v0_unroll and dw_edma_v0_ch_regs structures.
> So until the interface is converted to a more preferable direct MMIO
> usage without any packed-structures I'd rather leave these macros
> be.
> ...

> As I said above I'd rather leave the 64-bit accessors be until the
> packed structure-based interface is removed from the driver.

I wouldn't bother polishing something that's unused since it can't be
tested, it's easy to resurrect from the history if/when it becomes
necessary, and it makes it much harder to connect the commit log with
the code change.  But this isn't a drivers/pci change, so I'm fine
with it since Vinod acked it.

I guess the point is that when !CONFIG_64BIT, there was no writeq() so
we used SET_LL_32 twice.  linux/io-64-nonatomic-lo-hi.h provides that
writeq() implementation, so we can define and use SET_LL_64 for that
case.

Bjorn