Re: [PATCH 1/4] drivers/bus: Added Freescale Management Complex APIs

[Scott Wood <scottwood@freescale.com>]

On Tue, 2014-09-16 at 14:28 -0500, Kim Phillips wrote:
> On Mon, 15 Sep 2014 23:31:21 -0500
> Scott Wood <scottwood@freescale.com> wrote:
> 
> > On Mon, 2014-09-15 at 18:44 -0500, Kim Phillips wrote:
> > > On Thu, 11 Sep 2014 12:34:21 -0500
> > > "J. German Rivera" <German.Rivera@freescale.com> wrote:
> > > 
> > > > diff --git a/drivers/bus/fsl-mc/fsl_mc_sys.c b/drivers/bus/fsl-mc/fsl_mc_sys.c
> > > 
> > > > +/**
> > > > + * Map an MC portal in the kernel virtual address space
> > > > + */
> > > > +static int map_mc_portal(phys_addr_t mc_portal_phys_addr,
> > > > +			 uint32_t mc_portal_size,
> > > > +			 void __iomem **new_mc_portal_virt_addr)
> > > > +{
> > > > +	void __iomem *mc_portal_virt_addr = NULL;
> > > > +	struct resource *res = NULL;
> > > > +	int error = -EINVAL;
> > > > +
> > > > +	res =
> > > > +	    request_mem_region(mc_portal_phys_addr, mc_portal_size,
> > > > +			       "mc_portal");
> > > > +	if (res == NULL) {
> > > > +		pr_err("request_mem_region() failed for MC portal %#llx\n",
> > > > +		       mc_portal_phys_addr);
> > > > +		error = -EBUSY;
> > > > +		goto error;
> > > > +	}
> > > > +
> > > > +	mc_portal_virt_addr = ioremap_nocache(mc_portal_phys_addr,
> > > > +					      mc_portal_size);
> > > > +	if (mc_portal_virt_addr == NULL) {
> > > > +		pr_err("ioremap_nocache() failed for MC portal %#llx\n",
> > > > +		       mc_portal_phys_addr);
> > > > +		error = -EFAULT;
> > > > +		goto error;
> > > > +	}
> > > > +
> > > > +	*new_mc_portal_virt_addr = mc_portal_virt_addr;
> > > > +	return 0;
> > > > +error:
> > > > +	if (mc_portal_virt_addr != NULL)
> > > > +		iounmap(mc_portal_virt_addr);
> > > > +
> > > > +	if (res != NULL)
> > > > +		release_mem_region(mc_portal_phys_addr, mc_portal_size);
> > > > +
> > > > +	return error;
> > > > +}
> > > 
> > > unnecessary initializations, bad error codes (both should be
> > > -ENOMEM),
> > 
> > Why should the first one be -ENOMEM?  It's not allocating memory, but
> > rather reserving I/O space.
> 
> I was going with what most of the drivers are already doing, but I
> see EFAULT is 'Bad address', which, you're right, is probably more
> appropriate.

EFAULT is for userspace addresses that fault.

I don't see consistency on what other drivers use after
request_mem_region() -- some use ENOMEM, some ENODEV, some EBUSY... and
probably others.

I'd probably go with ENXIO or EBUSY.

-Scott