Re: [PATCH v1 0/6] RISC-V -march handling improvements

[Marc Kleine-Budde <mkl@pengutronix.de>]

[-- Attachment #1.1: Type: text/plain, Size: 3399 bytes --]

On 01.04.2022 22:00:35, Palmer Dabbelt wrote:
> As pointed out recently [1], sparse is parsing -march on RISC-V in order
> to obtain the default set of preprocessor macros to use.  Back when this
> was written ISA string was a simple affair, but these days it's a lot
> more complicated.  It's going to be a big chunk of work to get a proper
> ISA string parser into sparse, but we can at least fix the breakages for
> the subset of legal ISA strings that Linux currently uses (and are
> breaking users).
> 
> This patch set does three things:
> 
> * Stops die()ing on unknown ISA strings, unless the user has passed
>   -Wsparse-error.  This prints a warning and guesses at the macros to
>   use, which is probably fine for Linux.
> * Cleans up some of the differences between GCC's -march parsing and
>   sparse's.  None of this should really matter for Linux, as GCC will
>   blow up on bad ISA strings, but it just seemed worth doing when I was
>   in there.
> * Adds support for the Zicsr and Zifencei extensions, which were
>   recently enabled.  With these the unknown ISA string warning goes away
>   for Linux builds.
> 
> They're all sort of independent (and happen in this order), but they're
> all touching the same code so I'm just sending it as a series.  It's my
> first time touching sparse.
> 
> I've poked around with the first patch on its own and it seems to
> largely work as expected: I'm still getting a bunch of sparse-related
> warnings when I turn on sparse in my builds, but at least I don't get an
> error (after updating to a binutils that supports the new arguments, so
> Linux detects them).  I tried CF="-Wsparse-error", which also behaves as
> expected (that trinary boolean tripped me up for a bit).
> 
> The first patch alone should be a sufficient band-aid for systems that
> are actively broken right now, the rest are cleanups -- these may be
> necessary to get the RISC-V port sparse-clean, but that's a WIP so there
> might be more.  I'm going to play around with that, but just looking at
> the volume of spew it's probably going to take a while.  I gave these
> patches a bit of testing one-by-one, but not nearly as much as the
> first.
> 
> I just spun up a sparse repo [2] at kernel.org, these are on the riscv
> branch if that helps for anyone.  I've also started messing around with
> parsing a few more of the multi-letter extensions, but there's so much
> coupling I got fed up -- it's on riscv-wip, but I definitely don't like
> that last patch.  I figured it's better to send out these bits, as they
> look solid to me and builds are broken.  The new stuff (B, K, and V) are
> all in GCC-12 anyway, so we have a bit of time before they're useful.
> 
> [1]: https://lore.kernel.org/linux-sparse/mhng-c280d48c-477d-4589-baee-255c774b5a51@palmer-mbp2014/T/#maef705f448e4a1f12d853c0d8bc756f037ce1ce0
> [2]: https://git.kernel.org/pub/scm/linux/kernel/git/palmer/sparse.git

Works without warnings on Debian testing, with gcc-riscv64-linux-gnu
4:11.2.0--1.

Tested-by: Marc Kleine-Budde <mkl@pengutronix.de>

regards,
Marc

-- 
Pengutronix e.K.                 | Marc Kleine-Budde           |
Embedded Linux                   | https://www.pengutronix.de  |
Vertretung West/Dortmund         | Phone: +49-231-2826-924     |
Amtsgericht Hildesheim, HRA 2686 | Fax:   +49-5121-206917-5555 |

[-- Attachment #1.2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

[-- Attachment #2: Type: text/plain, Size: 161 bytes --]

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv