Date:	Tue, 18 Mar 2014 11:00:58 +1100
From:	Stewart Smith <stewart@...ux.vnet.ibm.com>
To:	Greg KH <greg@...ah.com>,
	Benjamin Herrenschmidt <benh@...nel.crashing.org>
Cc:	Stephen Rothwell <sfr@...b.auug.org.au>,
	Mark Brown <broonie@...nel.org>, Tejun Heo <tj@...nel.org>,
	linux-next@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: linux-next: build failure after merge of the driver-core tree

Greg KH <greg@...ah.com> writes:
> On Tue, Mar 18, 2014 at 07:33:30AM +1100, Benjamin Herrenschmidt wrote:
>> On Mon, 2014-03-17 at 11:33 -0700, Greg KH wrote:
>> 
>> > There were only 3 (or 4), users of this api, and no new ones had been
>> > added in _years_, it's a very obscure thing, and odds are, it wouldn't
>> > ever be added again, especially as it was just removed entirely not
>> > being needed anymore.  And I'd argue, it's something that you shouldn't
>> > have even been doing in the first place, so why a new user of it was
>> > added now is quite strange to me.
>> 
>> Actually that's a good question. Is there a better way for us than to
>> use this API ? Essentially we use that because it's our understanding
>> that it is what is needed for a sysfs file that can remove its parent
>> directory from within a write() op.
>> 
>> We followed the driver "remove" implementation as an example.
>> 
>> The reason the opal elog and dump patches use it is that those two
>> patches add sysfs interface that represent the error logs (and platform
>> dumps but you can treat the latter as some kind of special case of error
>> logs) from the service processor in sysfs.
>> 
>> So for each log we create a dir (kobject) which we populate with a few
>> things like the log unique ID, raw data, etc.... and a file that can be
>> used to "ack" the log with the service processor.
>> 
>> The latter has the effect of removing it. This is done by the collection
>> daemon after it has confirmed that the error log has been stored to disk
>> and properly flushed.
>> 
>> Is there a better interface ? Can we implement for example unlink() on
>> the kobject itself ? IE. Do the ack by essentially having userspace rm
>> the directory ? :-)
>
> No you can't, sorry.
>
> And this seems like a huge abuse of sysfs, you better be using binary
> sysfs files for your log data...

We are. It's a binary format we get from the service processor that we
then parse in userspace.

Seeing as each of these log entries is a separate thing that can be
retrieved and acknowledged to the service processor, it seemed fitting
to have each of them exposed to userspace so that policy can be dictated
there (acking them removes them from FSP and we have use cases for
peeking at them)

If there's a better way, I'd be interested in hearing it - it appeared
to be relatively good fit for sysfs (no sysadmin overhead, busybox has
enough utility to semi-meaningfully poke at it).

> Do you have a pointer to where you document this sysfs api in
> Documentation/ABI/ ?

Yes:
- Documentation/ABI/stable/sysfs-firmware-opal-dump
- Documentation/ABI/stable/sysfs-firmware-opal-elog

>> Now regarding the practicals of sorting out our trees, Stephen suggested
>> that rather than doing anything on my side (heh, I like that !), you
>> should revert the last patch of the series, the one removing the old
>> API, in your tree.
>> 
>> It can then be re-applied later around rc1.
>
> I'll look to see if we can do that, let me dig it up out of my tree...

Thanks. The new API looks like it should work fine for us, I'll be sure
to covert my API usage and let you know.

--
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