Re: [PATCH 03/13] filemap: align the index to mapping_min_order in the page cache

From: Kent Overstreet <kent.overstreet@linux.dev>
To: "Pankaj Raghav (Samsung)" <kernel@pankajraghav.com>
Cc: Matthew Wilcox <willy@infradead.org>,
	linux-xfs@vger.kernel.org,  linux-fsdevel@vger.kernel.org,
	linux-kernel@vger.kernel.org, david@fromorbit.com,
	 chandan.babu@oracle.com, akpm@linux-foundation.org,
	mcgrof@kernel.org, ziy@nvidia.com,  hare@suse.de,
	djwong@kernel.org, gost.dev@samsung.com, linux-mm@kvack.org,
	 Pankaj Raghav <p.raghav@samsung.com>
Subject: Re: [PATCH 03/13] filemap: align the index to mapping_min_order in the page cache
Date: Tue, 27 Feb 2024 12:02:45 -0500	[thread overview]
Message-ID: <na2k4nnvkseh2yh27eqkbfyouf7vnerd6i7pt4z7f7xsjsm6pu@ry5tvdcr2ggw> (raw)
In-Reply-To: <4zpsfvy3e4hkc4avvjjr34rgo7ggpd6hpflptmiauvxwm3dpvk@5wulihwpwbyp>

On Tue, Feb 27, 2024 at 05:55:35PM +0100, Pankaj Raghav (Samsung) wrote:
> > > > 
> > > > you guys are both wrong, just use rounddown()
> > > 
> > > Umm, what do you mean just use rounddown? rounddown to ...?
> > > 
> > > We need to get index that are in PAGE units but aligned to min_order
> > > pages.
> > > 
> > > The original patch did this:
> > > 
> > > index = mapping_align_start_index(mapping, iocb->ki_pos >> PAGE_SHIFT);
> > > 
> > > Which is essentially a rounddown operation (probably this is what you
> > > are suggesting?).
> > > 
> > > So what willy is proposing will do the same. To me, what I proposed is
> > > less complicated but to willy it is the other way around.
> > 
> > Ok, I just found the code for mapping_align_start_index() - it is just a
> > round_down().
> > 
> > Never mind; patch looks fine (aside from perhaps some quibbling over
> > whether the round_down()) should be done before calling readahead or
> > within readahead; I think that might have been more what willy was
> > keying in on)
> 
> Yeah, exactly.
> 
> I have one question while I have you here. 
> 
> When we have this support in the page cache, do you think bcachefs can make
> use of this support to enable bs > ps in bcachefs as it already makes use 
> of large folios? 

Yes, of course.

> Do you think it is just a simple mapping_set_large_folios ->
> mapping_set_folio_min_order(.., block_size order) or it requires more
> effort?

I think that's all that would be required. There's very little in the
way of references to PAGE_SIZE in bcachefs.