Re: [PATCH v5 1/3] usb: serial: Add MaxLinear/Exar USB to Serial driver

From: Mauro Carvalho Chehab
Date: Thu Feb 25 2021 - 13:07:33 EST


Em Thu, 25 Feb 2021 18:58:20 +0100
Mauro Carvalho Chehab <mchehab+huawei@xxxxxxxxxx> escreveu:

> Em Mon, 22 Feb 2021 16:47:36 +0100
> Johan Hovold <johan@xxxxxxxxxx> escreveu:
>
> > On Mon, Feb 22, 2021 at 04:27:34PM +0100, Mauro Carvalho Chehab wrote:
> > > Hi Johan,
> > >
> > > Em Tue, 26 Jan 2021 17:26:36 +0100
> > > Johan Hovold <johan@xxxxxxxxxx> escreveu:
> > >
> > > > On Tue, Jan 26, 2021 at 09:16:04PM +0530, Manivannan Sadhasivam wrote:
> > > > > On Thu, Jan 21, 2021 at 11:19:24AM +0100, Johan Hovold wrote:
> > > > > > On Sun, Nov 22, 2020 at 10:38:20PM +0530, Manivannan Sadhasivam wrote:
> > > > > > > Add support for MaxLinear/Exar USB to Serial converters. This driver
> > > > > > > only supports XR21V141X series but it can be extended to other series
> > > > > > > from Exar as well in future.
> >
> > > I'm now facing an issue with this driver. I have here two different
> > > boards with those USB UART from MaxLinear/Exar.
> > >
> > > The first one is identical to Mani's one:
> > > USB_DEVICE(0x04e2, 0x1411)
> > > The second one is a different version of it:
> > > USB_DEVICE(0x04e2, 0x1424)
> > >
> > > By looking at the final driver merged at linux-next, it sounds that
> > > somewhere during the review of this series, it lost the priv struct,
> > > and the xr_probe function. It also lost support for all MaxLinear/Exar
> > > devices, except for just one model (04e2:1411).
> > >
> > > The original submission:
> > >
> > > https://lore.kernel.org/linux-usb/20180404070634.nhspvmxcjwfgjkcv@advantechmxl-desktop
> > >
> > > And the manufacturer's Linux driver on their website:
> > >
> > > https://www.maxlinear.com/support/design-tools/software-drivers
> > >
> > > Had support for other 12 different models of the MaxLinear/Exar USB
> > > UART.
> >
> > IIRC Manivannan only had access to one of these models and his original
> > submission (based on the patch you link to above) didn't include support
> > for the others. And keeping the type abstraction didn't make sense for
> > just one model.
> >
> > > Those are grouped into 5 different major types:
> > >
> > > + init_xr2280x_reg_map();
> > > + init_xr21b142x_reg_map();
> > > + init_xr21b1411_reg_map();
> > > + init_xr21v141x_reg_map();
> > > +
> > > + if ((xrusb->DeviceProduct & 0xfff0) == 0x1400)
> > > + memcpy(&(xrusb->reg_map), &xr2280x_reg_map,
> > > + sizeof(struct reg_addr_map));
> > > + else if ((xrusb->DeviceProduct & 0xFFF0) == 0x1420)
> > > + memcpy(&(xrusb->reg_map), &xr21b142x_reg_map,
> > > + sizeof(struct reg_addr_map));
> > > + else if (xrusb->DeviceProduct == 0x1411)
> > > + memcpy(&(xrusb->reg_map), &xr21b1411_reg_map,
> > > + sizeof(struct reg_addr_map));
> > > + else if ((xrusb->DeviceProduct & 0xfff0) == 0x1410)
> > > + memcpy(&(xrusb->reg_map), &xr21v141x_reg_map,
> > > + sizeof(struct reg_addr_map));
> > > + else
> > > + rv = -1;
> > >
> > > Note: Please don't be confused by "reg_map" name. This has nothing
> > > to do with Linux regmap API ;-)
> > >
> > > What happens is that different USB IDs have different values for
> > > each register. So, for instance, the UART enable register is set to
> > > either one of the following values, depending on the value of
> > > udev->descriptor.idProduct:
> > >
> > > xr21b140x_reg_map.uart_enable_addr = 0x00;
> > > xr21b1411_reg_map.uart_enable_addr = 0xc00;
> > > xr21v141x_reg_map.uart_enable_addr = 0x03;
> > > xr21b142x_reg_map.uart_enable_addr = 0x00;
> > >
> > > There are other values that depend on the probing time detection,
> > > based on other USB descriptors. Those set several fields at the
> > > priv data that would allow to properly map the registers.
> > >
> > > Also, there are 4 models that support multiple channels. On those,
> > > there are one pair of register get/set for each channel.
> > >
> > > -
> > >
> > > In summary, while supporting just 04e2:1411 there's no need for
> > > a private struct, in order to properly support the other models,
> > > some autodetection is needed. The best way of doing that is to
> > > re-add the .probe method and adding a priv struct.
> > >
> > > As I dunno why this was dropped in the first place, I'm wondering
> > > if it would be ok to re-introduce them.
> >
> > Sure. It was just not needed if we were only going to support one model.
> >
> > > To be clear: my main focus here is just to avoid needing to use
> > > Windows in order to use the serial console of the hardware with
> > > the 0x1424 variant ;-)
> > >
> > > I can't test the driver with the other hardware, but, IMHO, instead
> > > of adding a hack to support 0x1424, the better (but more painful)
> > > would be to re-add the auto-detection part and support for the
> > > other models.
> >
> > Sounds good to me.
>
> While testing the xr_serial (as currently merged), I opted to apply
> the patches on the top of vanilla Kernel 5.11 - as it sounds too risky
> to use linux-next so early on a new development cycle :-)
>
> There, I'm getting an OOPS:
>
> [ 30.261291] BUG: kernel NULL pointer dereference, address: 00000000000000a8
> [ 30.261375] #PF: supervisor write access in kernel mode
> [ 30.261438] #PF: error_code(0x0002) - not-present page
> [ 30.261500] PGD 0 P4D 0
> [ 30.261539] Oops: 0002 [#1] SMP PTI
> [ 30.261586] CPU: 2 PID: 686 Comm: kworker/2:3 Not tainted 5.11.0+ #14
> [ 30.261666] Hardware name: /NUC5i7RYB, BIOS RYBDWi35.86A.0380.2019.0517.1530 05/17/2019
> [ 30.261757] Workqueue: usb_hub_wq hub_event
> [ 30.261816] RIP: 0010:mutex_lock+0x1e/0x40
> [ 30.261875] Code: c3 66 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 55 48 89 e5 41 54 49 89 fc e8 8d dd ff ff 31 c0 65 48 8b 14 25 c0 7b 01 00 <f0> 49 0f b1 14 24 75 04 41 5c 5d c3 4c 89 e7 e8 ae ff ff ff 41 5c
> [ 30.262076] RSP: 0018:ffffb937c0767b70 EFLAGS: 00010246
> [ 30.262140] RAX: 0000000000000000 RBX: ffff95e71ef75430 RCX: 0000000000000027
> [ 30.262223] RDX: ffff95e70597b000 RSI: 00000000ffffdfff RDI: 00000000000000a8
> [ 30.262305] RBP: ffffb937c0767b78 R08: ffff95ea76d18ac0 R09: ffffb937c0767948
> [ 30.262387] R10: 0000000000000001 R11: 0000000000000001 R12: 00000000000000a8
> [ 30.262469] R13: 0000000000000000 R14: ffff95e71ef75400 R15: 0000000000000000
> [ 30.262551] FS: 0000000000000000(0000) GS:ffff95ea76d00000(0000) knlGS:0000000000000000
> [ 30.262645] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [ 30.262713] CR2: 00000000000000a8 CR3: 000000042be0a002 CR4: 00000000003706e0
> [ 30.262796] Call Trace:
> [ 30.262832] usb_serial_disconnect+0x33/0x140
> [ 30.262897] usb_unbind_interface+0x8c/0x260
> [ 30.262957] device_release_driver_internal+0x103/0x1d0
> [ 30.263026] device_release_driver+0x12/0x20
> [ 30.263083] bus_remove_device+0xe1/0x150
> [ 30.263140] device_del+0x192/0x3f0
> [ 30.263188] ? usb_remove_ep_devs+0x1f/0x30
> [ 30.263244] usb_disable_device+0x95/0x1c0
> [ 30.263300] usb_disconnect+0xc0/0x270
> [ 30.263350] hub_event+0xa2e/0x1620
>
> After adding this hack:
>
> <snip>
> --- a/drivers/usb/serial/usb-serial.c
> +++ b/drivers/usb/serial/usb-serial.c
> @@ -1081,6 +1081,11 @@ static void usb_serial_disconnect(struct usb_interface *interface)
> struct usb_serial_port *port;
> struct tty_struct *tty;
>
> + if (!serial) {
> + dev_err(dev, "%s: Serial pointer is NULL!!!\n", __func__);
> + return;
> + }
> +
> usb_serial_console_disconnect(serial);
>
> mutex_lock(&serial->disc_mutex);
> </snip>
>
> It works fine:
>
> [ 283.005625] xr_serial 2-1:1.1: xr_serial converter detected
> [ 283.005868] usb 2-1: xr_serial converter now attached to ttyUSB0
> [ 283.007284] printk: console [ttyUSB0] enabled
> [ 284.444419] usb 2-1: USB disconnect, device number 5
> [ 284.444520] xr_serial 2-1:1.0: usb_serial_disconnect: Serial pointer is NULL!!!
> [ 284.444894] printk: console [ttyUSB0] disabled
> [ 284.445091] xr_serial ttyUSB0: xr_serial converter now disconnected from ttyUSB0
> [ 284.445141] xr_disconnect
> [ 284.445156] xr_serial 2-1:1.1: device disconnected
>
> I'm not sure if the bug is at xr_serial or if it is inside usb-serial.c.
>
> Any ideas?

Answering myself, as those devices may have two different interfaces
(one for control and another one for data), I suspect that the
driver needs to manually call usb_set_intfdata() after detecting the
data interface.

Thanks,
Mauro