On Thu, Feb 28, 2008 at 5:10 AM, Steven Stallion <[EMAIL PROTECTED]> wrote:
> All,
>
> As a part of the GLDv3 conversion work on dnet, I updated quite a bit of
> the STREAMS initialization code. After doing a bit of testing this
> evening, it looks like dnet is having an issue in _info (the _init
> breakpoint was never reached).
>
> My gut is telling me I botched the STREAMS init which is causing
> mod_info (and subsequently _info) to return a non 0 condition.
>
> Essentially, I changed this older block of STREAMS init:
>
> 293 /* Standard Streams initialization */
> 294 static struct module_info minfo = {
> 295 DNETIDNUM, "dnet", 0, INFPSZ, DNETHIWAT, DNETLOWAT
> 296 };
> 297
> 298 static struct qinit rinit = { /* read queues */
> 299 NULL, gld_rsrv, gld_open, gld_close, NULL, &minfo, NULL
> 300 };
> 301
> 302 static struct qinit winit = { /* write queues */
> 303 gld_wput, gld_wsrv, NULL, NULL, NULL, &minfo, NULL
> 304 };
> 305
> 306 static struct streamtab dnetinfo = {&rinit, &winit, NULL, NULL};
> ...
> 310
> 311 static struct cb_ops cb_dnetops = {
> 312 nulldev, /* cb_open */
> 313 nulldev, /* cb_close */
> 314 nodev, /* cb_strategy */
> 315 nodev, /* cb_print */
> 316 nodev, /* cb_dump */
> 317 nodev, /* cb_read */
> 318 nodev, /* cb_write */
> 319 nodev, /* cb_ioctl */
> 320 nodev, /* cb_devmap */
> 321 nodev, /* cb_mmap */
> 322 nodev, /* cb_segmap */
> 323 nochpoll, /* cb_chpoll */
> 324 ddi_prop_op, /* cb_prop_op */
> 325 &dnetinfo, /* cb_stream */
> 326 (int)(D_MP | D_HOTPLUG) /* cb_flag */
> 327 };
> 328
> 329 static struct dev_ops dnetops = {
> 330 DEVO_REV, /* devo_rev */
> 331 0, /* devo_refcnt */
> 332 gld_getinfo, /* devo_getinfo */
> 333 nulldev, /* devo_identify */
> 334 dnetprobe, /* devo_probe */
> 335 dnetattach, /* devo_attach */
> 336 dnetdetach, /* devo_detach */
> 337 nodev, /* devo_reset */
> 338 &cb_dnetops, /* devo_cb_ops */
> 339 (struct bus_ops *)NULL /* devo_bus_ops */
> 340 };
>
> To (similar to afe):
>
> 293 DDI_DEFINE_STREAM_OPS(dnet_ops, nulldev, dnet_probe, dnet_attach,
> dnet_detach,
> 294 nodev, NULL, D_MP, NULL);
>
Steve,
What catches my eye is the 7th argument to the DDI_DEFINE_STREAM_OPS() macro
It should be a xx_getinfo() entry point. Since it is a network driver
and it can get away
without proper getinfo() it should be nodev() or ddi_no_info(). I am
really not sure what
happens if NULL is passed there.
Well, I just checked the kernel source, and it appears that there is
at least one point
where getinfo() provided by the driver is called without checking it
is non-NULL beforehand.
Since it is called from uts/common/os/driver.c::bdev_strategy() I
highly doubt it has something
to do with your problem. Still leaving it as NULL bothers me.
Back to your problem. Have you tried to dtrace the driver loading
process ? We'd see
what was called and in which sequence.
>
> Looking back, cutting 293 - 306 seems wrong, but unfortunately I do not
> have enough experience with STREAMS to say why, or if there is a more
> appropriate route to take to remove the gld_getinfo reference in dnetops.
>
> Does anyone have any advice?
>
> TIA,
>
> Steve
>
> --
> Yet magic and hierarchy
> arise from the same source,
> and this source has a null pointer.
>
> Reference the NULL within NULL,
> it is the gateway to all wizardry.
> _______________________________________________
> driver-discuss mailing list
> [email protected]
> http://mail.opensolaris.org/mailman/listinfo/driver-discuss
>
--
Regards,
Cyril
_______________________________________________
driver-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/driver-discuss
