[U-Boot] JFFS2-Summary support for u-boot
Hi *,
please feel free to correct me if I tell sh**: As far as I was told, bad JFFS2 boot performance of u-boot is result of not supporting summary. (Because there is no JFFS2 but only JFFS support?)
So my question: Has anybody done or is planning in next time (~weeks) to add summary support for u-boot JFFS2? We need a file system on 32MB NOR flash with wear leveling and acceptable boot time. Ubi is currently not supported for NOR, JFFS2 slow, ... :-(
TIA Kind Regards, M.Lawnick
On Mon, 2008-09-29 at 08:45 +0000, Michael Lawnick wrote:
Hi *,
please feel free to correct me if I tell sh**: As far as I was told, bad JFFS2 boot performance of u-boot is result of not supporting summary. (Because there is no JFFS2 but only JFFS support?)
It is not missing SUMMARY, it is excessive scanning of empty flash blocks. Look at jffs2/scan.c, it only scan the first 4KB or before declaring the block empty.
Jocke
Joakim Tjernlund <joakim.tjernlund <at> transmode.se> writes:
On Mon, 2008-09-29 at 08:45 +0000, Michael Lawnick wrote:
Hi *,
please feel free to correct me if I tell sh**: As far as I was told, bad JFFS2 boot performance of u-boot is result of not supporting summary. (Because there is no JFFS2 but only JFFS support?)
It is not missing SUMMARY, it is excessive scanning of empty flash blocks. Look at jffs2/scan.c, it only scan the first 4KB or before declaring the block empty.
Ok, did I get you correct: The performance difference between u-boot and linux is based on skipping senseless scans of empty blocks? I've taken a little look into scan.c The comment "The code lacks more or less any comment, and is still arcane and difficult to read in places." doesn't make me happy.
Has anybody tried to (and succeeded in) speed up u-boot?
Michael
On Mon, 2008-09-29 at 11:34 +0000, Michael Lawnick wrote:
Joakim Tjernlund <joakim.tjernlund <at> transmode.se> writes:
On Mon, 2008-09-29 at 08:45 +0000, Michael Lawnick wrote:
Hi *,
please feel free to correct me if I tell sh**: As far as I was told, bad JFFS2 boot performance of u-boot is result of not supporting summary. (Because there is no JFFS2 but only JFFS support?)
It is not missing SUMMARY, it is excessive scanning of empty flash blocks. Look at jffs2/scan.c, it only scan the first 4KB or before declaring the block empty.
Ok, did I get you correct: The performance difference between u-boot and linux is based on skipping senseless scans of empty blocks? I've taken a little look into scan.c The comment "The code lacks more or less any comment, and is still arcane and difficult to read in places." doesn't make me happy.
Has anybody tried to (and succeeded in) speed up u-boot?
Yes, there was an attempt some time ago(months or a year, dunno) where someone had improved scanning dramatically, patch was posted to the u-boot list. Can't remember who though.
Jocke
Joakim Tjernlund <joakim.tjernlund <at> transmode.se> writes:
Has anybody tried to (and succeeded in) speed up u-boot?
Yes, there was an attempt some time ago(months or a year, dunno) where someone had improved scanning dramatically, patch was posted to the u-boot list. Can't remember who though.
Seems to be the thread 'jffs2 fsload - SOLVED.' from 12 Jan 03:14 Dan Merillat. As I've looke into the patch, it seems not to have found its way into u-boot. Any(body any) idea why?
Michael
On Mon, 2008-09-29 at 12:49 +0000, Michael Lawnick wrote:
Joakim Tjernlund <joakim.tjernlund <at> transmode.se> writes:
Has anybody tried to (and succeeded in) speed up u-boot?
Yes, there was an attempt some time ago(months or a year, dunno) where someone had improved scanning dramatically, patch was posted to the u-boot list. Can't remember who though.
Seems to be the thread 'jffs2 fsload - SOLVED.' from 12 Jan 03:14 Dan Merillat. As I've looke into the patch, it seems not to have found its way into u-boot.
Can't remember, possibly it was a hack.
Jocke
Hi Michael,
Joakim Tjernlund <joakim.tjernlund <at> transmode.se> writes:
Can't remember, possibly it was a hack.
Seems so. I tried to put it in my 1.3.3, but 'ls' never returns :-(
So I will wait for Detlev Zundel.
Actually someone doing the development for DENX will post the code, but it will not be me ;)
The code should hit the mailing list any day now. The speed improvements are pretty drastic, although dependant on the actual file system. Improvements by factors of 4 have been measured however...
Cheers Detlev
Hi Michael,
Joakim Tjernlund <joakim.tjernlund <at> transmode.se> writes:
On Mon, 2008-09-29 at 08:45 +0000, Michael Lawnick wrote:
Hi *,
please feel free to correct me if I tell sh**: As far as I was told, bad JFFS2 boot performance of u-boot is result of not supporting summary. (Because there is no JFFS2 but only JFFS support?)
It is not missing SUMMARY, it is excessive scanning of empty flash blocks. Look at jffs2/scan.c, it only scan the first 4KB or before declaring the block empty.
Ok, did I get you correct: The performance difference between u-boot and linux is based on skipping senseless scans of empty blocks? I've taken a little look into scan.c The comment "The code lacks more or less any comment, and is still arcane and difficult to read in places." doesn't make me happy.
Has anybody tried to (and succeeded in) speed up u-boot?
We are currently working on such a speedup bringing U-Boot up to par with Linux. I am confident that we can post our results in the next week.
Cheers Detlev
PS: Wow, message-mode in Emacs wants a confirmation sending an email to you as your address looks bogus ;)
Detlev Zundel schrieb:
Has anybody tried to (and succeeded in) speed up u-boot?
We are currently working on such a speedup bringing U-Boot up to par with Linux. I am confident that we can post our results in the next week.
Sounds promising!
PS: Wow, message-mode in Emacs wants a confirmation sending an email to you as your address looks bogus ;)
Address is valid and in use. This nospam_ kept away spam for some time. Seems some filters sorted it out, others removed it.
Michael
participants (4)
-
Detlev Zundel -
Joakim Tjernlund -
Michael Lawnick -
Michael Lawnick