**** BEGIN LOGGING AT Thu Jan 09 02:59:58 2014
Jan 09 09:15:17 <ant_work> lumag_: I'm totally confused...the logs seems showing that cmdset_0001 code does write (both word and buffered) on an erase suspended block
Jan 09 09:15:48 <ant_work> strangely, with the word-write program the fs is not corrupted on umount...
Jan 09 09:16:08 <ant_work> anyway, smthg is bad....read/program can only be done in blocks "other than that which is suspended"
Jan 09 09:51:01 <lumag_> ant_work, gm
Jan 09 09:51:35 <ant_work> hello
Jan 09 09:51:36 <lumag_> Now I can play your game and say: 'Oh, you know about the thing more that I can understand ATM' :)
Jan 09 09:51:40 <bluelightning> morning ant_work, lumag_
Jan 09 09:51:45 <lumag_> hello bluelightning
Jan 09 09:51:52 <ant_work> lumag_: against my will :/
Jan 09 09:53:03 <ant_work> what I don't wont to learn (right now) is about mutex/lock/unlock/schedule
Jan 09 09:53:25 <ant_work> but the code is very similar in word and buffer write functions
Jan 09 09:53:53 <ant_work> still, the problem appears with 2 concurrent threads
Jan 09 09:54:15 <ant_work> bluelightning: gm
Jan 09 09:55:12 <ant_work> bluelightning: I've asked apelete to watch for linux-yocto_3.13 to merge the qi kernel (he says there are >30 patches right now)
Jan 09 09:55:14 <lumag_> Well. mutexes and spinlocks are ease - no counters, nothing fancy. Just either go all the way till unlock, or wait till the road is free.
Jan 09 09:56:04 <lumag_> ant_work, The major difference is that mutexes are like an motel - if the road is busy you sleep there. Spinlocks are like traffic jam. You just sit there, check your watches and damn everybody around ;)
Jan 09 09:56:06 <bluelightning> ant_work: ok, cool
Jan 09 09:56:51 <ant_work> lumag_: I see, how does that play with timeouts?
Jan 09 09:56:55 <lumag_> The major thing about schedule is that you must not schedule_*() while holding spinlock.
Jan 09 09:57:24 <lumag_> First it is not permitted, second it can (or even 100% cases will) become a nearly 0 delay instead of what you asked for.
Jan 09 09:57:31 <ant_work> hm...
Jan 09 09:57:44 <lumag_> CONFIG_ATOMIC_SLEEP (or something like that) can help to check if you have such condition.
Jan 09 09:58:22 <lumag_> And btw: on uniprocessor machines w/o preempt spinlocks are essentially void (optimized out).
Jan 09 09:59:02 <lumag_> spinlock_irq/_irqsave  will just disable IRQs and that's all.
Jan 09 09:59:20 <ant_work> it appears to me that the ubi thread in bg comes abruptly on the scene and puts the other thread to sleep
Jan 09 09:59:37 <ant_work> that other thread suspends/resume like a crazy...
Jan 09 10:07:40 <ant_work> lumag_: I've put on github the patches I've used for debug-log and asked linusw for logs
Jan 09 10:08:34 <ant_work> it's hard to believe vanilla kernel boots MTD on h3600, collie and hx4700 need patches
Jan 09 10:08:45 <ant_work> still, he has no CF/SD
Jan 09 10:29:25 <ant_work> lumag_: looking again at the block address, it could be maybe a bug about the block size x2
Jan 09 10:30:53 <ant_work> 19 32K-WORD 098000H - 09FFFFH
Jan 09 10:31:23 <ant_work> 20 32K-WORD 0A0000H - 0A7FFFF
Jan 09 10:31:25 <ant_work> etc
Jan 09 10:31:43 <ant_work> I see
Jan 09 10:31:45 <ant_work> PID 289 Entering inval_cache_and_wait: chip->state=7 cmd_adr=0x009e1840
Jan 09 10:31:54 <ant_work> PID 286 Entering chip_ready: mode=0 adr=0x00aa0000
Jan 09 10:32:25 <ant_work> so PID 289 is in block 19
Jan 09 10:32:46 <ant_work> and PID 286 in block 21
Jan 09 10:33:06 <ant_work> all fine up to here
Jan 09 10:33:54 <ant_work> but then I see
Jan 09 10:33:55 <ant_work> PID 289 Entering inval_cache_and_wait: chip->state=8 cmd_adr=0x009c18c0
Jan 09 10:34:01 <ant_work> PID 286 Entering chip_ready: mode=7 adr=0x00aa0040
Jan 09 10:38:53 <ant_work> no, that's ok
Jan 09 11:15:09 <ant_work> maybe I see it...
Jan 09 11:15:36 <ant_work> we have to program a single word (a 0)
Jan 09 11:15:47 <ant_work> seems buffer write has wrong nr. of bytes
Jan 09 11:16:06 <ant_work> inline_map_write e800e8 ofs aa0000
Jan 09 11:16:14 <ant_work> inline_map_write 0 ofs aa0000
Jan 09 11:16:14 <ant_work> inline_map_write 0 ofs aa0000
Jan 09 11:16:24 <ant_work> inline_map_write d000d0 ofs aa0000
Jan 09 11:17:19 <ant_work> the same word is written like that
Jan 09 11:17:20 <ant_work> inline_map_write 400040 ofs c60000
Jan 09 11:17:21 <ant_work> inline_map_write 0 ofs c60000
Jan 09 11:23:02 <ant_work> nope
Jan 09 11:23:04 <ant_work> That
Jan 09 11:23:04 <ant_work> is, when the number of [N] is 1 word, write (00H); if [N]
Jan 09 11:23:04 <ant_work> is 16 words, write (0FH).
Jan 09 11:23:09 <ant_work> :/
**** ENDING LOGGING AT Fri Jan 10 02:59:59 2014