Re: [PATCH] arm64: PCI: Enable SMC conduit

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, Jan 26, 2021 at 10:46:04AM -0600, Jeremy Linton wrote:
> On 1/22/21 1:48 PM, Will Deacon wrote:
> > This isn't like the usual fragmentation problems, where firmware swoops in
> > to save the day; CPU onlining, spectre mitigations, early entropy etc. All
> > of these problems exist because there isn't a standard method to implement
> > them outside of firmware, and so adding a layer of abstraction there makes
> > sense.
> 
> There are a lot of parallels with PSCI here because there were existing
> standards for cpu online.

I don't recall anything that I would consider a standard at the time.

> > But PCIe is already a standard!
> 
> And it says that ECAM is optional, particularly if there are
> firmware/platform standardized ways of accessing the config space.

Nice loophole; I haven't checked.

> > We shouldn't paper over hardware designers' inability to follow a ~20 year
> > old standard by hiding it behind another standard that is hot off the press.
> > Seriously.
> 
> No disagreement, but its been more than half a decade and there are some
> high (millions!) volume parts, that still don't have kernel support.

Ok.

> > There is not a scrap of evidence to suggest that the firmware
> > implementations will be any better, but they will certainly be harder to
> > debug and maintain.  I have significant reservations about Arm's interest in
> > maintaining the spec as both more errata appear and the PCIe spec evolves
> > (after all, this is outside of SBSA, no?). The whole thing stinks of "if all
> > you have is a hammer, then everything looks like a nail". But this isn't the
> > sort of problem that is solved with yet another spec -- instead, how about
> > encouraging vendors to read the specs that already exist?
> 
> PSCI, isn't a good example of a firmware interface that works?

Not sure what you're getting at here.

> > > The SMC is an olive branch and just to make sure it is crystal clear
> > > there won't be room for adding quirks if the implementation turns out
> > > to be broken, if a line in the sand is what we want here it is.
> > 
> > I appreciate the sentiment, but you're not solving the problem here. You're
> > moving it somewhere else. Somewhere where you don't have to deal with it
> > (and I honestly can't blame you for that), but also somewhere where you
> > _can't_ necessarily deal with it. The inevitable outcome is an endless
> > succession of crappy, non-compliant machines which only appear to operate
> > correctly with particularly kernel/firmware combinations. Imagine trying to
> > use something like that?
> > 
> > The approach championed here actively discourages vendors from building
> > spec-compliant hardware and reduces our ability to work around problems
> > on such hardware at the same time.
> > 
> > So I won't be applying these patches, sorry.
> 
> Does that mean its open season for ECAM quirks, and we can expect them to
> start being merged now?

That's not for me to say.

Will



[Index of Archives]     [DMA Engine]     [Linux Coverity]     [Linux USB]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [Greybus]

  Powered by Linux