Re: [PATCH RFC] pci: Make return value of pcie_capability_read*() consistent

From: Bjorn Helgaas <helgaas@kernel.org>
To: Yicong Yang <yangyicong@hisilicon.com>
Cc: Bolarinwa Olayemi Saheed <refactormyself@gmail.com>,
	bjorn@helgaas.com, skhan@linuxfoundation.org,
	linux-pci@vger.kernel.org
Subject: Re: [PATCH RFC] pci: Make return value of pcie_capability_read*() consistent
Date: Thu, 23 Apr 2020 17:38:19 -0500	[thread overview]
Message-ID: <20200423223819.GA248903@google.com> (raw)
In-Reply-To: <80f70efc-0ff6-a53f-9a86-d52e053e6a69@hisilicon.com>

On Thu, Apr 23, 2020 at 07:55:17PM +0800, Yicong Yang wrote:
> On 2020/4/19 14:51, Bolarinwa Olayemi Saheed wrote:

> > diff --git a/drivers/pci/access.c b/drivers/pci/access.c
> > index 79c4a2ef269a..451f2b8b2b3c 100644
> > --- a/drivers/pci/access.c
> > +++ b/drivers/pci/access.c
> > @@ -409,7 +409,7 @@ int pcie_capability_read_word(struct pci_dev *dev, int pos, u16 *val)
> 
> Maybe provide some comments for the function, to notify the outside
> users to do the error code conversion.

A short function comment about the set of possible return values
wouldn't hurt.  We don't have those for pci_read_config_word() and
friends, and there are several pcie_capability_*() functions.  I don't
think it's worth repeating the comment for every function, so maybe we
could just extend the existing comment at pcie_capability_read_word().

> BTW, pci_{read, write}_config_*() may also have the issues that
> export the private err code outside. You may want to solve these in
> a series along with this patch.

If you see a specific issue, please point it out.

I looked at pci_read_config_word(), and it can return
PCIBIOS_DEVICE_NOT_FOUND, PCIBIOS_BAD_REGISTER_NUMBER, or the return
value from bus->ops->read().

I looked at all the users of PCIBIOS_*.  There's really no interesting
use of any of them except by pcibios_err_to_errno() and
xen_pcibios_err_to_errno(), so I'm not sure it's even worth keeping
them.

But I think it's probably more work to excise all of them than it is
to simply make pci_read_config_word() and pcie_capability_read_word()
return the same set of error values.  So I think we should do this
first.

> >  	*val = 0;
> >  	if (pos & 1)
> > -		return -EINVAL;
> > +		return PCIBIOS_BAD_REGISTER_NUMBER;
> >  
> >  	if (pcie_capability_reg_implemented(dev, pos)) {
> >  		ret = pci_read_config_word(dev, pci_pcie_cap(dev) + pos, val);
> > @@ -444,7 +444,7 @@ int pcie_capability_read_dword(struct pci_dev *dev, int pos, u32 *val)
> >  
> >  	*val = 0;
> >  	if (pos & 3)
> > -		return -EINVAL;
> > +		return PCIBIOS_BAD_REGISTER_NUMBER;
> >  
> >  	if (pcie_capability_reg_implemented(dev, pos)) {
> >  		ret = pci_read_config_dword(dev, pci_pcie_cap(dev) + pos, val);
> 