[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <YPBkbcmNzFKipXfC@casper.infradead.org>
Date: Thu, 15 Jul 2021 17:38:05 +0100
From: Matthew Wilcox <willy@...radead.org>
To: Helge Deller <deller@....de>
Cc: linux-arch@...r.kernel.org, linux-parisc@...r.kernel.org,
linux-mips@...r.kernel.org, linux-arm-kernel@...ts.infradead.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH 0/3] Make PMD_ORDER generically available
On Thu, Jul 15, 2021 at 06:28:42PM +0200, Helge Deller wrote:
> On 7/15/21 3:46 PM, Matthew Wilcox (Oracle) wrote:
> > These three architectures each define PMD_ORDER to mean "the order of
> > an allocation for a PMD table", but logically PMD_ORDER should be the
> > order of a PMD allocation, ie (PMD_SHIFT - PAGE_SHIFT) as DAX defines it.
>
> Some architectures do have PGD_ORDER, PUD_ORDER and PTE_ORDER as well.
> If you rename PMD_ORDER, IMHO the others should be renamed too.
>
> Why not simply rename "PMD_ORDER" in fs/dax.c to e.g.
> #define DAX_PMD_SHIFT (PMD_SHIFT - PAGE_SHIFT)
> and use that inside the dax filesystem code?
Because that's wrong. PMD order is clearly the order of a PMD page,
not the order of a PMD table.
I'm sure the others should also be renamed, but this is the one which
is causing me pain today.
Powered by blists - more mailing lists