Date:	Mon, 11 Jul 2011 09:54:42 -0700
From:	Sergiu Iordache <sergiu@...omium.org>
To:	Marco Stornelli <marco.stornelli@...il.com>
Cc:	Greg KH <greg@...ah.com>,
	Andrew Morton <akpm@...ux-foundation.org>,
	"Ahmed S. Darwish" <darwish.07@...il.com>,
	Artem Bityutskiy <Artem.Bityutskiy@...ia.com>,
	Kyungmin Park <kyungmin.park@...sung.com>,
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH v3 3/3] char drivers: ramoops debugfs entry

On Thu, Jul 7, 2011 at 11:43 PM, Marco Stornelli
<marco.stornelli@...il.com> wrote:
> 2011/7/8 Greg KH <greg@...ah.com>:
>> On Thu, Jul 07, 2011 at 04:27:27PM -0700, Andrew Morton wrote:
>>> On Thu, 7 Jul 2011 16:16:43 -0700
>>> Sergiu Iordache <sergiu@...gle.com> wrote:
>>>
>>> > Ramoops currently dumps the log of a panic/oops in a memory area which
>>> > is known not to be overwritten on restart (for example 1MB starting at
>>> > 15MB). The way it works is by dividing the memory area in records of a
>>> > set size (fixed at 4K before my patches, configurable after) and by
>>> > dumping a record there for each oops/panic. The problem is that right
>>> > now you have to access that memory area through other means, such as
>>> > /dev/mem, which is not always possible.
>>> >
>>> > What my patch did was to add a debugfs entry which returns a valid
>>> > record each time (a single dump done by ramoops). The first call
>>> > returns the first dump. The first call after the last valid dump
>>> > returns an empty buffer. .
>>>
>>> Please fully describe this "record" in the v2 patch changelog.  We'll
>>> want to review it for endianness, 32/64-bit compat issues,
>>> maintainability, extensibility, etc.
>>>
>>> > After it has returned nothing, the next
>>> > calls return records from the start again.
>>>
>>> That sounds a bit weird.  One would expect it to keep returning zero,
>>> requiring userspace to lseek or close/open.
>>>
>>> > The validity of a dump is
>>> > checked by looking after the header. Any comments on this approach are
>>> > welcome.
>>> >
>>> > Changing the entry from debugfs to sysfs wouldn't be a problem. If
>>> > sysfs is a valid solution I'll come with a patch that updates the
>>> > documentation as well along with the sysfs entry.
>>>
>>> sysfs sounds OK to me.  Then again, sysfs is supposed to be
>>> one-value-per-file, so using it would be naughty.
>>>
>>> I dunno, I'd be inclined to abuse the sysfs rule and hope that nobody
>>> notices rather than create a fake char device.  But there's certainly
>>> plenty of precedent for the fake char driver.
>>
>> No, please don't abuse sysfs that way.
>>
>> Use debugfs or a char device node.
>>
>> thanks,
>>
>> greg k-h
>>
>
> I agree with Greg. I asked to not break the existent way to read data
> via /dev/mem because for me it's the right way to do this thing.
> However to do an easy *debug* a debugfs entry can be useful. IMHO, a
> "production" script/application that use debugfs instead of /dev/mem
> in this case is simply broken because the debugfs can't be like a
> system call or other kernel interaction mechanism. Debugfs should be
> used only for debug.
>
> Marco

Any consensus/decision on how to go on with this patch idea?

The options that I see right now are:
- keep access through /dev/mem only (but access to /dev/mem is
sometimes restricted);
- keep the debugfs entry as well(as in the patch);
- remove the debugfs entry and add a char driver to access the memory
using read and seek operations.

+ the rejected(?) options from before

Sergiu
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists