linux-pci.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Yijing Wang <wangyijing@huawei.com>
To: Bjorn Helgaas <bhelgaas@google.com>, Yinghai Lu <yinghai@kernel.org>
Cc: PCI <linux-pci@vger.kernel.org>, "Herui (Ray)" <Ray.herui@huawei.com>
Subject: Re: Assign mem resource fail after remove and rescan
Date: Thu, 25 Jun 2015 09:59:43 +0800	[thread overview]
Message-ID: <558B608F.9060003@huawei.com> (raw)
In-Reply-To: <20150505190729.GA24643@google.com>

>> Subject: [PATCH] PCI: Optimize bus mem sizing to small size
>>
>> Current code try to get min_align as possible and use that to
>> align final size.
>>
>> That could cause generate wrong align/size or too big size in some case.
>>
>>  when we have align/size: 16M/64M
>>  min_align/size0 will be 8M/64M, that is wrong, align must be 16M.
>>
>>  when we have align/size: 1M/1M, 64M/64M,
>>  min_align/size0 will be 32M/96M, that is way too big for sum size 65M.
>>
>> That will cuase allocation fails.
>>
>> The patch introduce max_align/size0_max, and size0_max is just
>> sum of all children resource.
>>
>> Prefer small size instead of small align. Only use min_align when
>> two size is the same.
>>
>> The new size will only need to be aligned to bus window alignment.
>>
>> Signed-off-by: Yinghai Lu <yinghai@kernel.org>
> 
> I'm sorry, I can't make any sense out of this.  I can't understand
> what you're trying to say.
> 
> A specific simple example might help.
> 
> Is this a regression?  Should it be marked for stable?  If so, how far
> back?
> 
> It doesn't apply on v4.1-rc2 (I would apply it manually if I could
> understand the changelog and comments, but you might as well refresh it at
> the same time as you rewrite those).
> 
> Please include a link to the problem report.
> 
> Please include the patch inline in the message.  It is a significant hassle
> for me to deal with attachments, and you are the only major contributor who
> uses them.


Hi Yinghai, do you have some updates for this patch ?
I open a bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=100451

I hope this fix could be merged in upstream.

Thanks!
Yijing.

> 
> Bjorn
> 
>> ---
>>  drivers/pci/setup-bus.c |   50 +++++++++++++++++++++++++++++++++++++++++++++---
>>  1 file changed, 47 insertions(+), 3 deletions(-)
>>
>> Index: linux-2.6/drivers/pci/setup-bus.c
>> ===================================================================
>> --- linux-2.6.orig/drivers/pci/setup-bus.c
>> +++ linux-2.6/drivers/pci/setup-bus.c
>> @@ -882,12 +882,14 @@ static void pbus_size_io(struct pci_bus
>>  	}
>>  
>>  	size0 = calculate_iosize(size, min_size, size1,
>> -			resource_size(b_res), min_align);
>> +				 resource_size(b_res),
>> +				 window_alignment(bus, IORESOURCE_IO));
>>  	if (children_add_size > add_size)
>>  		add_size = children_add_size;
>>  	size1 = (!realloc_head || (realloc_head && !add_size)) ? size0 :
>>  		calculate_iosize(size, min_size, add_size + size1,
>> -			resource_size(b_res), min_align);
>> +				 resource_size(b_res),
>> +				 window_alignment(bus, IORESOURCE_IO));
>>  	if (!size0 && !size1) {
>>  		if (b_res->start || b_res->end)
>>  			dev_info(&bus->self->dev, "disabling bridge window %pR to %pR (unused)\n",
>> @@ -962,6 +964,8 @@ static int pbus_size_mem(struct pci_bus
>>  	struct resource *b_res = find_free_bus_resource(bus,
>>  					mask | IORESOURCE_PREFETCH, type);
>>  	resource_size_t children_add_size = 0;
>> +	resource_size_t max_align = 0, size0_max;
>> +	int count = 0;
>>  
>>  	if (!b_res)
>>  		return -ENOSPC;
>> @@ -1016,19 +1020,59 @@ static int pbus_size_mem(struct pci_bus
>>  			if (order > max_order)
>>  				max_order = order;
>>  
>> +			count++;
>> +			if (align > max_align)
>> +				max_align = align;
>> +
>>  			if (realloc_head)
>>  				children_add_size += get_res_add_size(realloc_head, r);
>>  		}
>>  	}
>>  
>> +	/*
>> +	 * New rule: Prefer to small size instead of small align,
>> +	 * when we have align/size: 1M/1M, 2M/2M,
>> +	 *  min_align/size0: 1M/3M, max_align/size0_max: 2M/3M
>> +	 *  pick 1M/3M.
>> +	 * when we have align/size: 1M/1M, 64M/64M,
>> +	 *  min_align/size0: 32M/96M, max_align/size0_max: 64M/65M
>> +	 *  pick 64M/65M.
>> +	 * when we have align/size: 1M/1M, 16M/64M,
>> +	 *  min_align/size0: 8M/72M, max_align/size0_max: 16M/65M
>> +	 *  pick 16M/65M.
>> +	 * when we have align/size: 32M/64M, 128M/512M
>> +	 *  min_align/size0: 64M/576M, max_align/size0_max: 128M/576M
>> +	 *  pick 64M/576M.
>> +	 * when we have align/size: 16M/32M, 128M/512M
>> +	 *  min_align/size0: 64M/576M, max_align/size0_max: 128M/554M
>> +	 *  pick 128M/554M.
>> +	 * when we have align/size: 16M/64M
>> +	 *  min_align/size0: 8M/64M, max_align/size0_max: 16M/64M
>> +	 *  have to use 16M/64M.
>> +	 */
>>  	min_align = calculate_mem_align(aligns, max_order);
>>  	min_align = max(min_align, window_alignment(bus, b_res->flags));
>> +	max_align = max(max_align, window_alignment(bus, b_res->flags));
>> +	if (count == 1)
>> +		min_align = max_align;
>> +
>>  	size0 = calculate_memsize(size, min_size, 0, resource_size(b_res), min_align);
>> +	size0_max = calculate_memsize(size, min_size, 0, resource_size(b_res),
>> +					window_alignment(bus, b_res->flags));
>> +
>> +	if (size0_max < size0) {
>> +		size0 = size0_max;
>> +		min_align = max_align;
>> +		max_align--; /* to use small align for size1 calculation */
>> +	}
>> +
>>  	if (children_add_size > add_size)
>>  		add_size = children_add_size;
>>  	size1 = (!realloc_head || (realloc_head && !add_size)) ? size0 :
>>  		calculate_memsize(size, min_size, add_size,
>> -				resource_size(b_res), min_align);
>> +				resource_size(b_res),
>> +				min_align <= max_align ? min_align :
>> +					window_alignment(bus, b_res->flags));
>>  	if (!size0 && !size1) {
>>  		if (b_res->start || b_res->end)
>>  			dev_info(&bus->self->dev, "disabling bridge window %pR to %pR (unused)\n",
> 
> 
> .
> 


-- 
Thanks!
Yijing


  reply	other threads:[~2015-06-25  1:59 UTC|newest]

Thread overview: 11+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2015-03-28 10:02 Assign mem resource fail after remove and rescan Yijing Wang
2015-03-29  6:18 ` Yinghai Lu
2015-03-30  4:05   ` Yijing Wang
2015-03-31  6:38   ` Yijing Wang
2015-04-01 22:21     ` Yinghai Lu
2015-04-02  8:35       ` Yijing Wang
2015-04-09 11:12       ` Yijing Wang
2015-05-05 19:07       ` Bjorn Helgaas
2015-06-25  1:59         ` Yijing Wang [this message]
2015-06-25  7:05           ` Yinghai Lu
2015-06-25  7:25             ` Yijing Wang

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=558B608F.9060003@huawei.com \
    --to=wangyijing@huawei.com \
    --cc=Ray.herui@huawei.com \
    --cc=bhelgaas@google.com \
    --cc=linux-pci@vger.kernel.org \
    --cc=yinghai@kernel.org \
    /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).