The osdblk driver hsa seen literally no change except for global
API changes since it was merged, and I've never heard of a user.
Is this something we want to keep around? I just noticed it because
it's the only non-SCSI user of some of the old block layer tagging
functions.
Drop it, if there are no users.
It was simulating an OSD, but there did not seem to be much beyond
academic interest, and some interest in the use of aforementioned
block functions as a useful exercise in block driver writing.
On Tue, Oct 21, 2014 at 8:06 AM, Christoph Hellwig <[email protected]> wrote:
> The osdblk driver hsa seen literally no change except for global
> API changes since it was merged, and I've never heard of a user.
>
> Is this something we want to keep around? I just noticed it because
> it's the only non-SCSI user of some of the old block layer tagging
> functions.
On 10/21/2014 03:21 PM, Jeff Garzik wrote:
> On Tue, Oct 21, 2014 at 8:06 AM, Christoph Hellwig <[email protected]> wrote:
>> The osdblk driver hsa seen literally no change except for global
>> API changes since it was merged, and I've never heard of a user.
>>
>> Is this something we want to keep around? I just noticed it because
>> it's the only non-SCSI user of some of the old block layer tagging
>> functions.
>
> Drop it, if there are no users.
>
> It was simulating an OSD, but there did not seem to be much beyond
> academic interest, and some interest in the use of aforementioned
> block functions as a useful exercise in block driver writing.
>
>
It is used in testing, its a good way to compare ext4 over objects
vs exofs over objects, kind of compare apples with apples.
(Also at the get go it was a good test that found some bugs in the
bring up of the osd target)
If there is an old tagging API need removing its fine to just drop
the support, and/or do the minimal change possible. What was the
tag used for anyway?
If it bothers you very much then please remove it, send a patch
and I'll ACK it.
[Just that I thought we don't do that. I can show you a dozen
ISA scsi LLD controllers which are a complete pain in the neck
with a all subsection core support. There is not a single system
on the planet that has this even possible, yet we do not remove
them nor the ISA bus crap. osdblk at least is a SW only exercise
that can be very much valid in any system.
]
Your call sir, I'm with you on this
Thanks
Boaz
On Tue, Oct 21, 2014 at 8:47 AM, Boaz Harrosh <[email protected]> wrote:
> [Just that I thought we don't do that. I can show you a dozen
> ISA scsi LLD controllers which are a complete pain in the neck
> with a all subsection core support. There is not a single system
> on the planet that has this even possible, yet we do not remove
> them nor the ISA bus crap. osdblk at least is a SW only exercise
> that can be very much valid in any system.
+1 also a fair point.
On Tue, Oct 21, 2014 at 03:47:36PM +0300, Boaz Harrosh wrote:
> It is used in testing, its a good way to compare ext4 over objects
> vs exofs over objects, kind of compare apples with apples.
> (Also at the get go it was a good test that found some bugs in the
> bring up of the osd target)
>
> If there is an old tagging API need removing its fine to just drop
> the support, and/or do the minimal change possible. What was the
> tag used for anyway?
>
> If it bothers you very much then please remove it, send a patch
> and I'll ACK it.
I'm mostly just trying to figure out if anyone cares. Would you be
willing to sign up as a maintainer, and maybe even look into a blk-mq
conversion? I'd be happy to help to answer questions and review it.
> [Just that I thought we don't do that. I can show you a dozen
> ISA scsi LLD controllers which are a complete pain in the neck
> with a all subsection core support. There is not a single system
> on the planet that has this even possible, yet we do not remove
> them nor the ISA bus crap. osdblk at least is a SW only exercise
> that can be very much valid in any system.
> ]
Not sure what exactly you mean, but I'm open to any serious proposal to
get rid of not actually usable cruft in old SCSI HBA drivers. Feel
free to send a proposal to linux-scsi - I was planning to come up with
one of my own, but never got to it.
On 10/21/2014 04:01 PM, Christoph Hellwig wrote:
<>
>
> I'm mostly just trying to figure out if anyone cares. Would you be
> willing to sign up as a maintainer, and maybe even look into a blk-mq
> conversion? I'd be happy to help to answer questions and review it.
>
I thought I was kind of am ;-) .
Good point I will put it on my list of things to do. Meanwhile can
I help you with the immediate problem of the tagging? could you do
something most brutal that even removes the tag support, just to
get you going?
And speaking of osdblk. Should it be its own full blown driver
or should it be under dm ? I know that at some LSF people said
it is better placed there? Specially since for a long time I wanted
to base it over ORE and not directly libosd so to support
RAID0/1/5/6 under the one simple block device.
>
> Not sure what exactly you mean, but I'm open to any serious proposal to
> get rid of not actually usable cruft in old SCSI HBA drivers. Feel
> free to send a proposal to linux-scsi - I was planning to come up with
> one of my own, but never got to it.
>
Yes exactly that!
I wish I could have the time, perhaps one day I will. There is so much
more cleaning I can see.
Thanks Christoph
Boaz