On Mon, Feb 10, 2020 at 08:44:31PM +0530, rishi gupta wrote:
> Tried dev_groups approach, doesn't fit here. Please see inline.
>
> On Fri, Jan 10, 2020 at 12:50 PM Greg KH <[email protected]> wrote:
> >
> > On Thu, Jan 09, 2020 at 02:59:59PM +0530, rishi gupta wrote:
> > > > > +/* UART frame structure definitions */
> > > > > +#define VS_CRTSCTS 0x0001
> > > > > +#define VS_XON 0x0002
> > > > > +#define VS_NONE 0X0004
> > > > > +#define VS_DATA_5 0X0008
> > > > > +#define VS_DATA_6 0X0010
> > > > > +#define VS_DATA_7 0X0020
> > > > > +#define VS_DATA_8 0X0040
> > > >
> > > > Why the "X"?
> > > Sorry I did not understand, do you mean why VS_XON.
> >
> > No, I mean why the "0X0040" instead of "0x0040" like all other hex
> > digits in your list of defines.
> >
> > > > > +static int vs_alloc_reg_one_dev(int oidx, int pidx, int rtsmap,
> > > > > + int dtrmap, int dtropn)
> > > > > +{
> > > > > + int ret;
> > > > > + struct vs_dev *vsdev;
> > > > > + struct device *dev;
> > > > > +
> > > > > + /* Allocate and init virtual tty device private data */
> > > > > + vsdev = kcalloc(1, sizeof(struct vs_dev), GFP_KERNEL);
> > > > > + if (!vsdev)
> > > > > + return -ENOMEM;
> > > > > +
> > > > > + vsdev->own_tty = NULL;
> > > > > + vsdev->peer_tty = NULL;
> > > > > + vsdev->own_index = oidx;
> > > > > + vsdev->peer_index = pidx;
> > > > > + vsdev->rts_mappings = rtsmap;
> > > > > + vsdev->dtr_mappings = dtrmap;
> > > > > + vsdev->set_odtr_at_open = dtropn;
> > > > > + vsdev->msr_reg = 0;
> > > > > + vsdev->mcr_reg = 0;
> > > > > + vsdev->waiting_msr_chg = 0;
> > > > > + vsdev->tx_paused = 0;
> > > > > + vsdev->faulty_cable = 0;
> > > > > + mutex_init(&vsdev->lock);
> > > > > +
> > > > > + /* Register with tty core with specific minor number */
> > > > > + dev = tty_register_device(ttyvs_driver, oidx, NULL);
> > > > > + if (!dev) {
> > > > > + ret = -ENOMEM;
> > > > > + goto fail;
> > > > > + }
> > > > > +
> > > > > + vsdev->device = dev;
> > > > > + dev_set_drvdata(dev, vsdev);
> > > > > +
> > > > > + /* Create custom sysfs files for this device for events */
> > > > > + ret = sysfs_create_group(&dev->kobj, &vs_info_attr_grp);
> > > >
> > > > Please no. You just raced with userspace and lost (i.e. userspace has
> > > > no idea these files are present.)
> > > >
> > > > Please use the correct apis for this, if you _REALLY_ want special sysfs
> > > > files for a tty device.
> > > Any specific API would you like to suggest. I am unable to progress on
> > > how to address this one.
> >
> > Now that you have moved things to configfs, maybe you do not need the
> > sysfs files anymore?
> >
> > Ah your "control" sysfs files, ok, you need to set the driver's
> > dev_groups variable to point to your sysfs attributes, and then the
> > driver core will properly set up these files.
> >
> > hope this helps,
> >
> > greg k-h
>
> Everything done except using dev_groups approach (full driver after
> all changes https://github.com/test209/t/blob/master/ttyvs.c#L1957).
>
> Currently to emulate parity error (or any event), user writes to a
> device specific node (0 is device number):
> echo "2" > /sys/devices/virtual/tty/ttyvs0/event
>
> With dev_groups, sysfs is created (1) for driver not for devices

Huh? It's there for devices.

> (2) for platform devices only

No, should work for any device type, as the logic is in the driver core.

> Due to (1), parsing based approach will be needed, for ex (0 is device number);
> echo "0-2" > /sys/devices/platform/ttyvs-card@0/event
> or
> echo "0-parity" > /sys/devices/platform/ttyvs-card@0/event

No, the 0- should not be needed, it should be a device-specific file.

> Due to (2), event file will not exist on desktop systems as there will
> be no device tree node; no platform device.

I don't understand, this file belongs in the tty device that you have
created, not in any other device.

> Original problem was user space doesn't know when
> "/sys/devices/virtual/tty/ttyvs0/event" will exist.

And if you set the devices group up properly, the sysfs file will be
created when the device is created.

> User space gets a uevent when a device is registered with tty core.
> Application must access only after this.

Yes, but you will race in creation of your file with userspace unless
you tell the core to do this properly.

> Is this okay in case of this particular driver.

No.

thanks,

greg k-h