Daniel Corbe wrote:
> I found my problem, it was with the way I was declaring modldrv, I had
> the last struct member set to NULL instead of a valid and initialized
> dev_ops struct.
>
> So now the driver loads without crashing the kernel.
>
> # add_drv -m '* 0644 root sys' testdriver
> devfsadm: driver failed to attach: testdriver
> Warning: Driver (testdriver) successfully added to system but failed
> to attach
>
> It looks as if my attach procedure isn't even being called.
>
>
Your driver is for a pseudo device? If so, you need a driver.conf
file. Otherwise, the system
loads your driver, decides there are no devices that the driver drives,
and then unloads it.
See pseudo(4) and driver.conf(4) for a description of the driver.conf
file (yours should be
something like testdriver.conf. Also, you'll need to rem_drv before
add_drv again.
max
> /* Attach and detatch a device */
>
> int testdriver_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
>
> {
>
> int result = DDI_FAILURE;
>
> cmn_err(CE_WARN, "ATTACHING!!");
>
> ddi_create_minor_node(dip, "testdriver", S_IFCHR, 1, NULL, 0);
>
> result = DDI_SUCCESS;
>
> return(result);
>
> }
>
>
> and these are the relevant structs
>
>
>
> /* Character device specific stuff */
>
> static struct cb_ops cb_testdriver_ops =
>
> {
>
> nodev,
> /* cb_open */
>
> nodev,
> /* cb_close */
>
> nodev,
> /* cb_strategy */
>
> nodev,
> /* cb_print */
>
> nodev,
> /* cb_dump */
>
> nodev,
> /* cb_read */
>
> nodev,
> /* cb_write */
>
> nodev,
> /* cb_ioctl */
>
> nodev,
> /* cb_devmap */
>
> nodev,
> /* cb_mmap */
>
> nodev,
> /* cb_segmap */
>
> nochpoll,
> /* cb_chpoll */
>
> ddi_prop_op, /* cb_prop_op */
>
> &testdriver_strtab, /* cb_stream */
>
> (D_NEW | D_MP | D_MTPERMOD) /* cb_flag */
>
> };
>
>
> static struct dev_ops testdriver_ops =
>
> {
>
> DEVO_REV, /* devo_rev */
>
> 0 , /* devo_refcnt */
>
> nodev,
> /* devo_getinfo */
>
> nodev,
> /* devo_identify */
>
> nodev,
> /* devo_probe */
>
> testdriver_attach, /* devo_attach */
>
> testdriver_detach, /* devo_detach */
>
> nodev,
> /* devo_reset */
>
> &cb_testdriver_ops, /* devo_cb_ops */
>
> (struct bus_ops *)NULL /* devo_bus_ops */
>
> };
>
>
> /* Module specific (kernel linkage for the kernel*/
>
> static struct modldrv modldrv =
>
> {
>
> &mod_driverops,
>
> "testdriver" ,
>
> &testdriver_ops
>
> };
>
>
> static struct modlinkage modlinkage =
>
> {
>
> MODREV_1,
>
> &modldrv,
>
> NULL
>
> };
>
>
>
> SO what am I doing wrong now?
>
>
> -Daniel
>
>
>
>
> On 9/11/07, [EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>*
> <[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]> > wrote:
>
> Garrett D'Amore wrote:
> > On Tue, 2007-09-11 at 17:47 +0200, [EMAIL PROTECTED]
> <mailto:[EMAIL PROTECTED]> wrote:
> >
> >> Garrett D'Amore wrote:
> >>
> >>> On Tue, 2007-09-11 at 10:22 -0400, Daniel Corbe wrote:
> >>>
> >>>
> >>>> You guys are extremely patient with answering these
> questions, thank
> >>>> you.
> >>>>
> >>>>
> >>>> I've run into another issue.
> >>>>
> >>>>
> >>>> Running add_drv causes my system to reboot. I assume it's
> something
> >>>> I'm doing wrong. I figured it had something to do with
> either the
> >>>> attach or getinfo procedures in my driver, and in fact the
> getinfo
> >>>> procedure is using an uninitialized pointer.
> >>>>
> >>>>
> >>> I don't think getinfo() is the problem... its more likely to be
> >>> attach(). (getinfo() is normally only called as part of Dynamic
> >>> Reconfiguration operation, IIRC.)
> >>>
> >>> -- Garrett
> >>>
> >>>
> >> Actually, getinfo() is called twice on open(2). Try
> >>
> >> # dtrace -n 'pts_devinfo:entry{stack();}'
> >>
> >> and then open a new terminal window. The pts_devinfo() routine
> is the
> >> getinfo for the pseudo terminal slave
> >> driver (also a streams driver).
> >>
> >
> > Is that true for all drivers, or just clone devices? I confess
> I've not
> > looked at getinfo() much lately.. most of the NIC drivers these days
> > don't need a getinfo(), since they use qassociate() to provide the
> > mapping between dev_info_t and streams.
> >
> > -- Garrett
> >
> Yes. True for all drivers that use cb_ops. See getinfo(9e) and
> ddi_no_info(9f). And
> I thought pts was cloned, but glancing at the source code, I guess
> not!?
>
> max
>
>
>
>
>
_______________________________________________
driver-discuss mailing list
driver-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/driver-discuss
