[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20190415054949.GA9279@infradead.org>
Date: Sun, 14 Apr 2019 22:49:49 -0700
From: Christoph Hellwig <hch@...radead.org>
To: Anup Patel <Anup.Patel@....com>
Cc: Palmer Dabbelt <palmer@...ive.com>,
Albert Ou <aou@...s.berkeley.edu>,
Atish Patra <Atish.Patra@....com>,
Paul Walmsley <paul.walmsley@...ive.com>,
Christoph Hellwig <hch@...radead.org>,
"linux-riscv@...ts.infradead.org" <linux-riscv@...ts.infradead.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v2 1/3] RISC-V: Add separate asm/encoding.h for spec
related defines
On Sat, Apr 13, 2019 at 03:38:35PM +0000, Anup Patel wrote:
> It's better to have all RISC-V spec related defines in one place
> so this patch adds separate asm/encoding.h for such defines which
> can be included in assembly as well as C code.
As per the discussion of version 1 on Saturday I disagree with this
move. There is both a philosophical and a practical reason for that:
a) in RISC-V CSR access is really nicely split out into a separate
number space, just accessed through a few special instructions.
It has no overlap with the rest of instruction encoding
b) <asm/csr.h> is pulled into just about every file build in the kernel
though <asm/irqflags.h>, <asm/switch_to.h> and <asm/thread_info.h>
So even if you want to later add a new <asm/encoding.h> for instruction
encoding details later, I'd rather prefer to keep CSR access separate
from it.
Powered by blists - more mailing lists