02:01:56 fun little bug I just found in our regex code: echo 'à' | grep '[aà]' 02:02:01 does not match! 05:49:57 tsoome: so that issue with kcf/swrand not dealing with devices detaching is probably what's been messing me up w/ the tpm20 driver's rand support... which means I probably didn't do it wrong :) 06:10:22 jbk yep. it just means we have much to do:) 06:11:24 i've got to finish adding fm ereports in the various spots and then the TAB handling and then just try to tidy anything up that needs it, and it'll maybe be ready for review 06:12:27 hm, my test is running third day now.... for i in {1..1000000} will take a while to process... 06:13:17 building the tpm libraries (not required for the driver, just a already written consumer) is a bit of a pain 06:13:53 whatever they're doing with autoconf is failing on illumos 06:14:30 and as far as I can tell, somehow some macros are getting undefined, but trying to figure out where has been an experience only slightly more pleasant than sticking a fork in my eye 06:14:57 oh right, there is also this lex issue 06:15:17 i actually (for now) just wrote my own makefiles for everything 06:15:29 lex and yacc really. 06:15:30 because that was far less painful given that the code is already pretty portable 06:16:07 I think it is perfectly ok to have own makefile if it is connected to our build system 06:16:25 the only annoying bit is they assume all unix platforms's struct dirent includes a field to indicate if a file is a directory or not 06:17:18 i've got a workaround for that in my fork/branch for now, but it's an ugly (imo) #ifdef 08:03:26 jbk: physical host's serial console 15:01:14 i'll need to see if it helps any with SOL (though IIRC.. it's usually loader over SOL that has problems w/ losing characters.. have to type at approx 40-50 baud, etc.. though I don't think it's loader's fault and much more likely crappy UEFI firmware) 15:36:12 jbk: It can't do anything about sol pre-illumos. 15:41:30 out of interest has anyone looked at removing libgcc-unwind.map from gcc recently? 16:17:15 yeah.. i just can't remember if the losing characters was just at the loader stage, or if it carried over to after the os was booted 16:17:52 i just remember that because usually if I'm using SOL, it's because I need to muck with boot images or boot params 19:28:21 tsoome_: So for #14919 / #16126, I'm available to help out in any way if you need it. 21:30:18 god i so want to take a proverbal weed whacker to sd.c 21:30:39 like we still send READ(6) CDBs despite them being deprecated for probably longer than some people in here have been alive 21:30:52 similarly with other log pages, etc. 21:46:15 what do we do? shrink-to-fit to send the smallest CDB that the offset fits in? 21:46:34 yeah i believe so (at least for READ/WRITE) 21:48:20 also, even on a 4k device, we appear to send a 512 byte read to LBA 1 -- I suspect to read the label, but I'm still not sure yet how it's getting 512 since it seems like the blocksize should be set by then (this target gets very unhappy when we do that) 22:01:38 if we are reading label, then the request is coming from cmlb, but we should have had established device block size and capacity while attaching cmlb... 22:25:51 sommerfeld: Yes, we do shrink to fit 22:27:01 see also: https://www.illumos.org/issues/11952 22:27:03 → BUG 11952: large USB hard disks experience I/O failures (Closed) 22:45:47 tsoome_: yeah, i need to see why we're not (iSCSI is actually helpful here because it was easy to get the CDBs being sent and the response) 22:47:37 it looks like after calling cmlb_attach(), something else has to happen to update the initial blocksize -- cmlb keeps it's own private value of the device's block size that defaults to DEV_BSIZE 22:49:05 i haven't had a chance yet to dig into everything to see what the actual sequence has to be to cause that update 22:49:25 my guess is that somehow cmlb is issuing the write before that update is happening 22:50:47 btw, what is device-blksize property value for that disk? 22:54:31 the second argument for cmlb_attach() is pointer to struct cmlb_tg_ops, which does provide read/write and getinfo callbacks on device, that getinfo function should provide sector size too. 22:54:47 yeah, that's all correct