according to the illumos HCL, the LPe11000 and LPe12000 cards should work, but according to the driver it's unsupported?! Apr 4 22:27:41 gem emlxs: [ID 349649 kern.info] [ B.1C2C]emlxs1: ERROR: 101: Driver attach failed. (Unsupported adapter found. Id:35 Vendor id:0x10df Device id:0xfc40 SSDID:0xfc41 Model:LPe12000-S)

/* Oracle branded adapters are not supported in this driver */

oh, there was something

thanks for bringing that back up. would it help to flash a different firmware on it? I guess no?

you can try, I guess

I'll check if broadcom still has firmware for those cards...

I'd check with Carsten - maybe he can comment about background there

good idea

[illumos-gate] 16423 Import fletcher-4 algorithms from OpenZFS -- Andy Fiddaman <illumos⊙fn>

[illumos-gate] 11745 fmdump missing -i and -I parameter documentation in the man pages -- Keith M Wesolowski <wesolows⊙oc>

so after a bit of a journey, I've re-discovered that a set of our false positive smatch warnings are because of old-style flexible array member declarations. `struct foo { char foo[1]; }`. That, for instance, is why smatch things zfs_ctldir.c overflows strcpy()

seemingly smatch knows(?) that [0] is special, but takes [1] to mean [1]

what I don't know, is what the implications are of making them new-style

beyond -- I assume -- all the structs being a perhaps very shitty to discover byte shorter.

So, the old style unfortunately trips over the modern optimizer.

[0] is a GNU extension, so it knows that it's like C99 [].

ok, sorry, I thought [0] was the c99

The main implications are either GNU 89 or C 99, and the size implications.

So in non-public structures, I could go to C99 [] if you can.

and then audit for allocations that assume a -1

Yeah.

I dealt with some C99_FEATURES related bits for this for new public versions of this in the xregs procfs.

I suspect the rationale for what smatch does is that [0] could only possibly be useful for a flexible array, but [1] could really mean [1]..

Yeah, the rationale is fine, it's the fixing it.

the one to watch out for when auditing is where the extra byte is used for a null terminator. Or when it's using sizeof(struct foo) and it's actually a delta of alignof(something) rather than 1.

(sizeof(struct foo) rather than sizeof(foo.foo) that is)

comstar also has very annoying take on structures...