linux-kernel - Re: [PATCH 6/6] pci, acpi: Share ACPI PCI config space accessors.

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20141121180825.GD6578@google.com>
Date:	Fri, 21 Nov 2014 11:08:25 -0700
From:	Bjorn Helgaas <bhelgaas@...gle.com>
To:	Arnd Bergmann <arnd@...db.de>
Cc:	linux-arm-kernel@...ts.infradead.org,
	Myron Stowe <myron.stowe@...il.com>, rjw@...ysocki.net,
	lorenzo.pieralisi@....com, linaro-acpi@...ts.linaro.org,
	linux-pci <linux-pci@...r.kernel.org>, catalin.marinas@....com,
	x86 <x86@...nel.org>, will.deacon@....com,
	LKML <linux-kernel@...r.kernel.org>,
	Tomasz Nowicki <tomasz.nowicki@...aro.org>,
	"linux-acpi@...r.kernel.org" <linux-acpi@...r.kernel.org>,
	Ingo Molnar <mingo@...hat.com>, hanjun.guo@...aro.org,
	"H. Peter Anvin" <hpa@...or.com>, wangyijing@...wei.com,
	Liviu.Dudau@....com, Thomas Gleixner <tglx@...utronix.de>
Subject: Re: [PATCH 6/6] pci, acpi: Share ACPI PCI config space accessors.

On Fri, Nov 21, 2014 at 01:24:52PM +0100, Arnd Bergmann wrote:
> On Thursday 20 November 2014 21:00:17 Myron Stowe wrote:
> > On Thu, Nov 20, 2014 at 3:26 PM, Bjorn Helgaas <bhelgaas@...gle.com> wrote:
> > > On Wed, Nov 19, 2014 at 05:19:20PM +0100, Arnd Bergmann wrote:
> > >> On Wednesday 19 November 2014 17:04:51 Tomasz Nowicki wrote:
> > >> > +/*
> > >> > + * raw_pci_read/write - ACPI PCI config space accessors.
> > >> > + *
> > >> > + * ACPI spec defines MMCFG as the way we can access PCI config space,
> > >> > + * so let MMCFG be default (__weak).
> > >> > + *
> > >> > + * If platform needs more fancy stuff, should provides its own implementation.
> > >> > + */
> > >> > +int __weak raw_pci_read(unsigned int domain, unsigned int bus,
> > >> > +                       unsigned int devfn, int reg, int len, u32 *val)
> > >> > +{
> > >> > +       return pci_mmcfg_read(domain, bus, devfn, reg, len, val);
> > >> > +}
> > >> > +
> > >> > +int __weak raw_pci_write(unsigned int domain, unsigned int bus,
> > >> > +                        unsigned int devfn, int reg, int len, u32 val)
> > >> > +{
> > >> > +       return pci_mmcfg_write(domain, bus, devfn, reg, len, val);
> > >> > +}
> > >> > +
> > >> >
> > >>
> > >> I think it would be better to avoid __weak functions here, as they tend
> > >> to be hard to follow when trying to understand the code.
> > >
> > > That's interesting.  I would have said exactly the opposite -- I think the
> > > extra Kconfiggery is harder to follow than weak/strong functions 
> > >
> > > But consistency is better than my personal opinion.  Is there a consensus
> > > that we should use the Kconfig strategy instead of __weak?
> > 
> > I too find weak/strong functions easier to follow than "Kconfiggery" (nice term
> > invention there).
> 
> I don't think there is a universal consensus, but the majority of
> maintainers seems to avoid them for the same reasons that I think
> __weak is problematic.
> 
> We have some uses of __weak in the core kernel, but there is
> basically none in drivers outside of PCI, and the most common
> uses are all providing an empty __weak function that can be
> overridden with a function that actually does something, unlike
> the code above.

One thing I like better about __weak (when used correctly) is that you have
exactly one declaration, and the role of each definition (weak default
implementation or strong override) is obvious from looking at it.

In your #ifdef example, the extern declaration and the inline definition
are never compiled together, so you have to repeat the signature and the
compiler doesn't enforce that they match.  So you end up with the extern
and the inline in one file, a #define in an arch header file or Kconfig,
and an arch definition in a third file.

But it's certainly true that everybody knows how #ifdef works, and the fact
that __weak on a declaration affects all in-scope definitions is definitely
a land mine (multiple weak definitions with no strong one is a disaster).

> My pragmatic approach so far has been to advocate __weak for
> drivers/pci patches but discourage it elsewhere when I review
> patches, in order to maintain consistency. I also think it
> would be nice to change the way that PCI handles architecture
> specific overrides in the process of unifying the host bridge
> handling.
> 
> I wouldn't use Kconfig symbols in most cases though. My preferred
> choice would be to turn a lot of the __weak symbols into function
> pointers within a per-hostbridge structure. As an example, we could
> replace pcibios_add_device() with a pointer in pci_host_bridge->ops
> that gets set by all the architectures and host drivers that currently
> override it, and replace the one caller with
> 
> 	if (pci_host_bridge->ops->add_device)
> 		pci_host_bridge->ops->add_device(dev);

I definitely agree with this part, but I think it's orthogonal to the
__weak question.  In this case, we'd like to support multiple host bridges,
each with a different flavor of add_device().  We can't do that at all with
either __weak or #ifdef.

Bjorn
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/