linuxppc-dev.lists.ozlabs.org archive mirror
 help / color / mirror / Atom feed
From: Anton Vorontsov <cbouatmailru@gmail.com>
To: Grant Likely <grant.likely@secretlab.ca>
Cc: sfr@canb.auug.org.au, monstr@monstr.eu,
	devicetree-discuss@lists.ozlabs.org,
	microblaze-uclinux@itee.uq.edu.au, jeremy.kerr@canonical.com,
	linuxppc-dev@lists.ozlabs.org
Subject: Re: [PATCH 6/6] of/device: populate platform_device (of_device) resource table on allocation
Date: Tue, 8 Jun 2010 20:46:45 +0400	[thread overview]
Message-ID: <20100608164645.GA15216@oksana.dev.rtsoft.ru> (raw)
In-Reply-To: <AANLkTinZT4z93SWxjce5oRZzmaQHtVnO96zsXO9OX2UZ@mail.gmail.com>

On Tue, Jun 08, 2010 at 10:02:49AM -0600, Grant Likely wrote:
> On Tue, Jun 8, 2010 at 9:57 AM, Anton Vorontsov <cbouatmailru@gmail.com> wrote:
> > On Tue, Jun 08, 2010 at 08:26:43AM -0600, Grant Likely wrote:
> > [...]
> >> +     dev = kzalloc(sizeof(*dev) + (sizeof(struct resource) * i), GFP_KERNEL);
> >>       if (!dev)
> >>               return NULL;
> >> -
> >>       dev->dev.of_node = of_node_get(np);
> >>       dev->dev.dma_mask = &dev->archdata.dma_mask;
> >>       dev->dev.parent = parent;
> >>       dev->dev.release = of_release_dev;
> >>
> >> +     /* Populate the resource table */
> >> +     if (num_irq || num_reg) {
> >> +             dev->resource = (void*)&dev[1];
> >
> > This is ugly. Why not allocate the memory specifically for
> > dev->resource? Is this because you plan to get rid of
> > of_release_dev(), and the generic release_dev() won't
> > know if it should free the dev->resource? There must
> > be a better way to handle this.
> 
> Allocating in one big block means less potential memory fragmentation,
> and the kernel can free it all at once.

Are there any numbers of saved amount of memory so that we
could compare?

The "less memory fragmentation" is indeed potential, but
introduction of obscure code is going on now at this precise
moment.

> This is a common pattern.

This can't be true because it produces ugly casts and fragile
code all over the place -- which is exactly what everybody
tries to avoid in the kernel.

Such a pattern is common when a driver allocates e.g. tx and rx
buffers (of the same type) together, and then split the buffer
into two pointers.

But I heard of no such pattern for 'struct device + struct
resources' allocation without even some kind of _priv struct,
which is surely something new, and ugly.

If you really want to avoid (an unproven) memory fragmentation,
you could do:

struct of_device_with_resources {
	struct device dev;
	struct resource resourses[0];
};

This at least will get rid of the casts and incomprehensible
"dev->resource = (void*)&dev[1];" construction.

-- 
Anton Vorontsov
email: cbouatmailru@gmail.com
irc://irc.freenode.net/bd2

  reply	other threads:[~2010-06-08 16:46 UTC|newest]

Thread overview: 25+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2010-06-08 14:26 [PATCH 0/6] OF device code merges and improvements Grant Likely
2010-06-08 14:26 ` [PATCH 1/6] of: Use full node name in resource structures Grant Likely
2010-06-08 14:26 ` [PATCH 2/6] of/device: merge of_device_uevent Grant Likely
2010-06-08 14:26 ` [PATCH 3/6] of: Modify of_device_get_modalias to be passed struct device Grant Likely
2010-06-08 14:26 ` [PATCH 4/6] of/device: Merge of_platform_bus_probe() Grant Likely
2010-06-08 14:26 ` [PATCH 5/6] of: Merge of_device_alloc Grant Likely
2010-06-08 14:26 ` [PATCH 6/6] of/device: populate platform_device (of_device) resource table on allocation Grant Likely
2010-06-08 15:57   ` Anton Vorontsov
2010-06-08 16:02     ` Grant Likely
2010-06-08 16:46       ` Anton Vorontsov [this message]
2010-06-08 18:41         ` Grant Likely
2010-06-08 19:48           ` Anton Vorontsov
2010-06-10  6:17             ` Benjamin Herrenschmidt
2010-06-10 14:18               ` Grant Likely
2010-06-10 15:13                 ` M. Warner Losh
2010-06-10 15:47                   ` Anton Vorontsov
2010-06-10 16:01                     ` M. Warner Losh
2010-06-10 16:52                       ` Anton Vorontsov
2010-06-10 17:09                         ` Mitch Bradley
2010-06-10 17:20                           ` Grant Likely
2010-06-10 17:09                         ` M. Warner Losh
2010-06-11  1:14                       ` Benjamin Herrenschmidt
2010-06-10 16:30                     ` Grant Likely
2010-06-10 17:10                       ` Anton Vorontsov
2010-06-10 17:21                         ` Grant Likely

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20100608164645.GA15216@oksana.dev.rtsoft.ru \
    --to=cbouatmailru@gmail.com \
    --cc=devicetree-discuss@lists.ozlabs.org \
    --cc=grant.likely@secretlab.ca \
    --cc=jeremy.kerr@canonical.com \
    --cc=linuxppc-dev@lists.ozlabs.org \
    --cc=microblaze-uclinux@itee.uq.edu.au \
    --cc=monstr@monstr.eu \
    --cc=sfr@canb.auug.org.au \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).