2.6: drivers/md/dm-io.c partially copies bio.c
Hi,
drivers/md/dm-io.c copies functionality from bio.c .
Is there a specific reason why you don't simply use the functionality bio.c provides?
cu
Adrian
Signature · 4 lines
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon SeedHi Adrian,
Quoted 4 lines · Adrian Bunk
drivers/md/dm-io.ccopies functionality frombio.c.Is there a specific reason why you don't simply use the functionality
bio.cprovides?
Can you give some specific examples of the functionality you think is duplicated? Meanwhile, I'll take a look and see if I can explain any code overlaps.
Signature · 3 lines
Kevin Corry [email protected] http://evms.sourceforge.net/
Quoted 11 lines · Kevin Corry
Hi Adrian,
Quoted 4 lines · Adrian Bunk
drivers/md/dm-io.ccopies functionality frombio.c.Is there a specific reason why you don't simply use the functionality
bio.cprovides?Can you give some specific examples of the functionality you think is duplicated? Meanwhile, I'll take a look and see if I can explain any code overlaps.
Ah come on Kevin, a 2 second glance shows lots of uneccesary duplication. Basically only the concept of the bio_set is not duplicated in the first many lines, you even set up matching slabs.
How was that ever accepted for merging?
Signature · 1 line
Jens Axboe
Quoted 13 lines · Jens Axboe
Quoted 11 lines · Kevin Corry
Hi Adrian,
Quoted 4 lines · Adrian Bunk
drivers/md/dm-io.ccopies functionality frombio.c.Is there a specific reason why you don't simply use the functionality
bio.cprovides?Can you give some specific examples of the functionality you think is duplicated? Meanwhile, I'll take a look and see if I can explain any code overlaps.
Quoted 5 lines
Ah come on Kevin, a 2 second glance shows lots of uneccesary duplication. Basically only the concept of the
bio_setis not duplicated in the first many lines, you even set up matching slabs.How was that ever accepted for merging?
If I recall correctly (and it's been a while since I've looked at that code), the bio_set was added because a few DM modules like to initiate their own I/O requests (things like snapshots and DM's kcopyd daemon), and we felt it was better to allow these modules to each create their own mempools to allocate bios from, rather than allocate from the single kernel-wide bio pool used by the filesystem layer.
Strictly speaking (and as you mentioned), the slabs in the bio_set are unnecessary, and they could use the global bio_slab. But there's probably some argument to be made for having separate bio mempools for these modules.
Actually, I also seem to recall discussions with Joe Thornber from quite awhile ago about trying to move this bio_set functionality into fs/bio.c, to allow other device drivers to create their own private bio pools. If you think something like this would be desireable, I can try to look into the specifics again. If you think that having the single kernel-wide bio pool is sufficient for all filesystems and device-drivers (you certainly understand the trade-offs better than I do), then I can look into removing the necessary code from dm-io.c
Signature · 3 lines
Kevin Corry [email protected] http://evms.sourceforge.net/
Quoted 18 lines · Kevin Corry
Quoted 17 lines · Jens Axboe
Quoted 11 lines · Kevin Corry
Hi Adrian,
Quoted 4 lines · Adrian Bunk
drivers/md/dm-io.ccopies functionality frombio.c.Is there a specific reason why you don't simply use the functionality
bio.cprovides?Can you give some specific examples of the functionality you think is duplicated? Meanwhile, I'll take a look and see if I can explain any code overlaps.
Ah come on Kevin, a 2 second glance shows lots of uneccesary duplication. Basically only the concept of the
bio_setis not duplicated in the first many lines, you even set up matching slabs.
Quoted 8 lines
Quoted 1 line
How was that ever accepted for merging?
If I recall correctly (and it's been a while since I've looked at that code), the
bio_setwas added because a few DM modules like to initiate their own I/O requests (things like snapshots and DM's kcopyd daemon), and we felt it was better to allow these modules to each create their own mempools to allocate bios from, rather than allocate from the single kernel-wide bio pool used by the filesystem layer.
That is fully correct and also something that eg the bouncing code needs to be 100% deadlock free.
Quoted 4 lines
Strictly speaking (and as you mentioned), the slabs in the
bio_setare unnecessary, and they could use the globalbio_slab. But there's probably some argument to be made for having separate bio mempools for these modules.
The mempools are needed, the duplicated slabs are not. But that's just a small part of it, basically the whole allocation and index stuff is 100% duplicated from bio.c. Even bio_init().
Quoted 8 lines
Actually, I also seem to recall discussions with Joe Thornber from quite awhile ago about trying to move this
bio_setfunctionality intofs/bio.c, to allow other device drivers to create their own private bio pools. If you think something like this would be desireable, I can try to look into the specifics again. If you think that having the single kernel-wide bio pool is sufficient for all filesystems and device-drivers (you certainly understand the trade-offs better than I do), then I can look into removing the necessary code fromdm-io.c
An abstraction for this would be nice, as there are two users of it then (dm-io and highmem.c). So if your team would do that, I would sure help you review and integrate it.
That's not what I'm arguing against, it's the (almost) wholesale duplication that's a bit silly.
Signature · 1 line
Jens Axboe
Quoted 29 lines · Jens Axboe
Quoted 6 lines · Kevin Corry
If I recall correctly (and it's been a while since I've looked at that code), the
bio_setwas added because a few DM modules like to initiate their own I/O requests (things like snapshots and DM's kcopyd daemon), and we felt it was better to allow these modules to each create their own mempools to allocate bios from, rather than allocate from the single kernel-wide bio pool used by the filesystem layer.That is fully correct and also something that eg the bouncing code needs to be 100% deadlock free.
Quoted 4 lines
Strictly speaking (and as you mentioned), the slabs in the
bio_setare unnecessary, and they could use the globalbio_slab. But there's probably some argument to be made for having separate bio mempools for these modules.The mempools are needed, the duplicated slabs are not. But that's just a small part of it, basically the whole allocation and index stuff is 100% duplicated from
bio.c. Evenbio_init().Quoted 8 lines
Actually, I also seem to recall discussions with Joe Thornber from quite awhile ago about trying to move this
bio_setfunctionality intofs/bio.c, to allow other device drivers to create their own private bio pools. If you think something like this would be desireable, I can try to look into the specifics again. If you think that having the single kernel-wide bio pool is sufficient for all filesystems and device-drivers (you certainly understand the trade-offs better than I do), then I can look into removing the necessary code fromdm-io.c
Quoted 6 lines
An abstraction for this would be nice, as there are two users of it then (dm-io and
highmem.c). So if your team would do that, I would sure help you review and integrate it.That's not what I'm arguing against, it's the (almost) wholesale duplication that's a bit silly.
I'll be happy to look into whether this duplication can be removed, and also whether the bio_set stuff could be moved out to bio.c or somewhere else common to the other device-drivers and filesystems. However, I've got a lot of stuff going on right now, so I won't really have time to get to this until after the start of the year (might be able to start looking into it over the holidays if I get everything else done in the next couple weeks). If/When I come up with some patches, I'll let you know.
Signature · 3 lines
Kevin Corry [email protected] http://evms.sourceforge.net/
Quoted 2 lines · Kevin Corry
Actually, I also seem to recall discussions with Joe Thornber from quite awhile ago about trying to move this
bio_setfunctionality intofs/bio.c,
Indeed - see his comments in dm-io.c:
/*-----------------------------------------------------------------
* Bio set, move this to bio.c *---------------------------------------------------------------*/
...
/*-----------------------------------------------------------------
* dm-io proper *---------------------------------------------------------------*/
Alasdair
Signature · 1 line
[email protected]