Re: [RFC PATCH] docs/memory-barriers.txt: Rewrite "KERNEL I/O BARRIER EFFECTS" section

[Will Deacon]

Hi Thomas,

On Tue, Feb 19, 2019 at 11:27:47AM +0100, Thomas Petazzoni wrote:
> On Mon, 18 Feb 2019 21:37:25 +0100
> Arnd Bergmann <arnd@xxxxxxxx> wrote:
> 
> > > > I would say we should strengthen the behavior of outX() where possible.
> > > > I don't know if arm64 actually has a way of doing that, my understanding
> > > > earlier was that the AXI bus was already posted, so there is not much
> > > > you can do here to define __io_paw() in a way that will prevent posted
> > > > writes.  
> > >
> > > If we could map I/O space using different page table attributes (probably by
> > > hacking pci_remap_iospace() ?) then we could disable the
> > > early-write-acknowledge hint and implement __io_paw() as a completion
> > > barrier, although it would be at the mercy of the system as to whether or
> > > not that requires a response from the RC.  
> > 
> > Ah, it seems we actually do that on 32-bit ARM, at least on one platform,
> > see 6a02734d420f ("ARM: mvebu: map PCI I/O regions strongly ordered")
> > and prior commits.
> 
> Yes, some Marvell Armada 32-bit platforms have an errata that require
> the PCI MEM and PCI I/O regions to be mapped strongly ordered.
> 
> BTW, this requirement prevents us from using the pci_remap_iospace()
> API from drivers/pci, because it assumes page attributes of
> pgprot_device(PAGE_KERNEL). That's why we're still using the
> ARM-specific pci_ioremap_io() function.

Ah, I think I vaguely remember that. It was to avoid a hardware deadlock,
right? In which case, I'd rather consider this use of strongly-ordered
memory an exceptional case as opposed to a general property of I/O mappings.

> > > I would still prefer to document the weaker semantics as the portable
> > > interface, unless there are portable drivers relying on this today (which
> > > would imply that it's widely supported by other architectures).  
> > 
> > I don't know of any portable driver that actually relies on it, but
> > that's mainly because there are very few portable drivers that
> > use inb()/outb() in the first place. How many of those require
> > the non-posted behavior I don't know
> > 
> > Adding Thomas, Gregory and Russell to Cc, as they were involved
> > in the discussion that led to the 32-bit change, maybe they are
> > aware of a specific example.
> 
> I'm just arriving in the middle of this thread, and I'm not sure to
> understand what is the question. If the question is whether PCI I/O is
> really used in practice, then I've never seen it be used with Marvell
> platforms (but I'm also not aware of all PCIe devices people are
> using). I personally have a hacked-up version of the e1000e driver
> that intentionally does some PCI I/O accesses, that I use as a way to
> validate that PCI I/O support is minimally working, but that's it.

It was actually even more subtle than that! The question is whether outX()
is relied upon to be non-posted in portable drivers, because at the moment
it's typicall posted for arm/arm64 systems, with the exception of the Armada
erratum above.

Cheers,

Will