[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20250306-3f25ee3d4686442bd215a0fa@orel>
Date: Thu, 6 Mar 2025 09:13:43 +0100
From: Andrew Jones <ajones@...tanamicro.com>
To: Charlie Jenkins <charlie@...osinc.com>
Cc: linux-riscv@...ts.infradead.org, linux-kernel@...r.kernel.org,
linux-doc@...r.kernel.org, paul.walmsley@...ive.com, palmer@...belt.com,
cleger@...osinc.com, alex@...ti.fr, Anup Patel <apatel@...tanamicro.com>,
corbet@....net
Subject: Re: [PATCH v3 0/8] riscv: Unaligned access speed probing fixes and
skipping
On Wed, Mar 05, 2025 at 02:48:58PM -0800, Charlie Jenkins wrote:
> On Tue, Mar 04, 2025 at 01:00:15PM +0100, Andrew Jones wrote:
> > The first six patches of this series are fixes and cleanups of the
> > unaligned access speed probing code. The next patch introduces a
> > kernel command line option that allows the probing to be skipped.
> > This command line option is a different approach than Jesse's [1].
> > [1] takes a cpu-list for a particular speed, supporting heterogeneous
> > platforms. With this approach, the kernel command line should only
> > be used for homogeneous platforms. [1] also only allowed 'fast' and
> > 'slow' to be selected. This parameter also supports 'unsupported',
> > which could be useful for testing code paths gated on that. The final
> > patch adds the documentation.
>
> Why constrain the command line option to homogeneous platforms?
Based on feedback at the last Plumber's, we've decided not to go out of
our way to support heterogeneous platforms unless they start to
materialize. With that in mind, and the fact that heterogeneous platforms
can use the probing mechanism instead of the command line, then I didn't
think the cpu-list support was worth it yet. Also, we can introduce
support for an optional [,<cpu-list>] attribute later, since the
definition of the parameters would stay the same for when the cpu-list
attribute is absent. Indeed, even if I was to introduce the cpu-list
support now, I would make it optional with the absence of it behaving
as this patch series implements.
Thanks,
drew
>
> - Charlie
>
> >
> > (I'd be happy to split the fixes from the new skip support if we want to
> > discuss the skip support independently, but I want to base on the fixes
> > and I'm not sure if patchwork supports Based-on: $MESSAGE_ID/$LORE_URL
> > or not at the moment, so I'm just posting together for now in order to
> > be able to check for my patchwork green lights!)
> >
> > [1] https://lore.kernel.org/linux-riscv/20240805173816.3722002-1-jesse@rivosinc.com/
> >
> > Thanks,
> > drew
> >
> > ---
> > v3:
> > - Fix compile when RISCV_PROBE_UNALIGNED_ACCESS is not selected
> >
> > v2:
> > - Change to command line option from table
> >
> >
> > Andrew Jones (8):
> > riscv: Annotate unaligned access init functions
> > riscv: Fix riscv_online_cpu_vec
> > riscv: Fix check_unaligned_access_all_cpus
> > riscv: Change check_unaligned_access_speed_all_cpus to void
> > riscv: Fix set up of cpu hotplug callbacks
> > riscv: Fix set up of vector cpu hotplug callback
> > riscv: Add parameter for skipping access speed tests
> > Documentation/kernel-parameters: Add riscv unaligned speed parameters
> >
> > .../admin-guide/kernel-parameters.txt | 16 ++
> > arch/riscv/include/asm/cpufeature.h | 4 +-
> > arch/riscv/kernel/traps_misaligned.c | 14 +-
> > arch/riscv/kernel/unaligned_access_speed.c | 237 +++++++++++-------
> > 4 files changed, 168 insertions(+), 103 deletions(-)
> >
> > --
> > 2.48.1
> >
Powered by blists - more mailing lists