All of lore.kernel.org
 help / color / mirror / Atom feed
From: Sebastian Frias <sebastian_frias@sigmadesigns.com>
To: Michal Hocko <mhocko@kernel.org>,
	Marc Gonzalez <marc_gonzalez@sigmadesigns.com>
Cc: linux-mm <linux-mm@kvack.org>, LKML <linux-kernel@vger.kernel.org>
Subject: Re: m(un)map kmalloc buffers to userspace
Date: Wed, 9 Dec 2015 15:53:22 +0100	[thread overview]
Message-ID: <56684062.9090505@sigmadesigns.com> (raw)
In-Reply-To: <20151209143207.GF30907@dhcp22.suse.cz>

On 12/09/2015 03:32 PM, Michal Hocko wrote:
> On Wed 09-12-15 15:07:50, Marc Gonzalez wrote:
>> On 09/12/2015 14:55, Michal Hocko wrote:
>>> On Tue 08-12-15 18:25:31, Sebastian Frias wrote:
>>>> Hi,
>>>>
>>>> We are porting a driver from Linux 3.4.39+ to 4.1.13+, CPU is Cortex-A9.
>>>>
>>>> The driver maps kmalloc'ed memory to user space.
>>>
>>> This sounds like a terrible idea to me. Why don't you simply use the
>>> page allocator directly? Try to imagine what would happen if you mmaped
>>> a kmalloc with a size which is not page aligned? mmaped memory uses
>>> whole page granularity.
>>
>> According to the source code, this kernel module calls
>>
>>    kmalloc(1 << 17, GFP_KERNEL | __GFP_REPEAT);
>
> So I guess you are mapping with 32pages granularity? If this is really
> needed for internal usage you can use highorder page and map its
> subpages directly.
>
>> I suppose kmalloc() would return page-aligned memory?
>
> I do not think there is any guarantee like that. AFAIK you only get
> guarantee for the natural word alignment. Slab allocator is allowed
> to use larger allocation and put its metadata or whatever before the
> returned pointer.
>

Thanks for your answer.
Do you have any suggestions regarding the rest of the questions? 
(copy/pasted below for convenience)

2) Now that VM_RESERVED was removed, is there another recommended flag 
to replace it for the purposes above?
3) Since it was working before, we suppose that something that was 
previously done by default on the kernel it is not done anymore, could 
that be a remap_pfn_range during mmap or kmalloc?
4) We tried using remap_pfn_range inside mmap and while it seems to 
work, we still get occasional crashes due to corrupted memory (in this 
case the behaviour is the same between 4.1 and 3.4 when using the same 
modified driver), are we missing something?

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

WARNING: multiple messages have this Message-ID (diff)
From: Sebastian Frias <sebastian_frias@sigmadesigns.com>
To: Michal Hocko <mhocko@kernel.org>,
	Marc Gonzalez <marc_gonzalez@sigmadesigns.com>
Cc: linux-mm <linux-mm@kvack.org>, LKML <linux-kernel@vger.kernel.org>
Subject: Re: m(un)map kmalloc buffers to userspace
Date: Wed, 9 Dec 2015 15:53:22 +0100	[thread overview]
Message-ID: <56684062.9090505@sigmadesigns.com> (raw)
In-Reply-To: <20151209143207.GF30907@dhcp22.suse.cz>

On 12/09/2015 03:32 PM, Michal Hocko wrote:
> On Wed 09-12-15 15:07:50, Marc Gonzalez wrote:
>> On 09/12/2015 14:55, Michal Hocko wrote:
>>> On Tue 08-12-15 18:25:31, Sebastian Frias wrote:
>>>> Hi,
>>>>
>>>> We are porting a driver from Linux 3.4.39+ to 4.1.13+, CPU is Cortex-A9.
>>>>
>>>> The driver maps kmalloc'ed memory to user space.
>>>
>>> This sounds like a terrible idea to me. Why don't you simply use the
>>> page allocator directly? Try to imagine what would happen if you mmaped
>>> a kmalloc with a size which is not page aligned? mmaped memory uses
>>> whole page granularity.
>>
>> According to the source code, this kernel module calls
>>
>>    kmalloc(1 << 17, GFP_KERNEL | __GFP_REPEAT);
>
> So I guess you are mapping with 32pages granularity? If this is really
> needed for internal usage you can use highorder page and map its
> subpages directly.
>
>> I suppose kmalloc() would return page-aligned memory?
>
> I do not think there is any guarantee like that. AFAIK you only get
> guarantee for the natural word alignment. Slab allocator is allowed
> to use larger allocation and put its metadata or whatever before the
> returned pointer.
>

Thanks for your answer.
Do you have any suggestions regarding the rest of the questions? 
(copy/pasted below for convenience)

2) Now that VM_RESERVED was removed, is there another recommended flag 
to replace it for the purposes above?
3) Since it was working before, we suppose that something that was 
previously done by default on the kernel it is not done anymore, could 
that be a remap_pfn_range during mmap or kmalloc?
4) We tried using remap_pfn_range inside mmap and while it seems to 
work, we still get occasional crashes due to corrupted memory (in this 
case the behaviour is the same between 4.1 and 3.4 when using the same 
modified driver), are we missing something?


  reply	other threads:[~2015-12-09 14:53 UTC|newest]

Thread overview: 26+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2015-12-08 17:25 m(un)map kmalloc buffers to userspace Sebastian Frias
2015-12-08 17:25 ` Sebastian Frias
2015-12-09 13:55 ` Michal Hocko
2015-12-09 13:55   ` Michal Hocko
2015-12-09 14:07   ` Marc Gonzalez
2015-12-09 14:07     ` Marc Gonzalez
2015-12-09 14:32     ` Michal Hocko
2015-12-09 14:32       ` Michal Hocko
2015-12-09 14:53       ` Sebastian Frias [this message]
2015-12-09 14:53         ` Sebastian Frias
2015-12-09 15:12         ` Michal Hocko
2015-12-09 15:12           ` Michal Hocko
2015-12-09 15:35           ` Sebastian Frias
2015-12-09 15:35             ` Sebastian Frias
2015-12-10 11:40             ` Michal Hocko
2015-12-10 11:40               ` Michal Hocko
2015-12-10 12:04               ` Richard Weinberger
2015-12-10 12:04                 ` Richard Weinberger
2015-12-10 13:37               ` Sebastian Frias
2015-12-10 13:37                 ` Sebastian Frias
2015-12-10 14:06                 ` Michal Hocko
2015-12-10 14:06                   ` Michal Hocko
2015-12-10 16:48                   ` Sebastian Frias
2015-12-10 16:48                     ` Sebastian Frias
2015-12-11  9:42                     ` Michal Hocko
2015-12-11  9:42                       ` Michal Hocko

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=56684062.9090505@sigmadesigns.com \
    --to=sebastian_frias@sigmadesigns.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --cc=marc_gonzalez@sigmadesigns.com \
    --cc=mhocko@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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.