[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CA+=Fv5QwEmgoHj=N-A-Ur1=XJMhP0QSoaWfdiw22RGvrGbgFdg@mail.gmail.com>
Date: Tue, 29 Jul 2025 10:58:08 +0200
From: Magnus Lindholm <linmag7@...il.com>
To: Christoph Hellwig <hch@...radead.org>
Cc: linux-kernel@...r.kernel.org, linux-scsi@...r.kernel.org,
linux-alpha@...r.kernel.org, martin.petersen@...cle.com,
James.Bottomley@...senpartnership.com, macro@...am.me.uk
Subject: Re: [PATCH 1/1] scsi: qla1280: Make 64-bit DMA addressing a Kconfig option
On Tue, Jul 29, 2025 at 9:32 AM Christoph Hellwig <hch@...radead.org> wrote:
>
> Compile time options like this are annoying as they require
> distributions to pick a default. Why not at least a module option
> even if it just a workaround?
I see your point, I've been looking at how the driver for the NCR53C8XX
SYM53C8XX family of PCI-SCSI controllers does things. The driver
has a SCSI_SYM53C8XX_DMA_ADDRESSING_MOD compile time
option for setting DMA parameters.
The qla1280 driver has different code for 32-bit and 64-bit mode of
the driver, implying that 32 or 64 bit modes need to be selected at
compile time.
However, it would work just using the 64-bit version
of the driver but, while defaulting to a 64-bit DMA bit mask, making
it possible to limit the DMA bit mask to 32-bit via a kernel module
option. This would make it possible to avoid using the
"monster window" for DMA transfers on tsunami based Alphas.
If the driver is built in 32-bit mode, this module option would simply
have no effect.
Powered by blists - more mailing lists