[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <aJOHObGgfzxIDzHW@shell.armlinux.org.uk>
Date: Wed, 6 Aug 2025 17:47:53 +0100
From: "Russell King (Oracle)" <linux@...linux.org.uk>
To: Andrew Lunn <andrew@...n.ch>
Cc: Xu Yang <xu.yang_2@....com>, hkallweit1@...il.com,
o.rempel@...gutronix.de, pabeni@...hat.com, netdev@...r.kernel.org,
imx@...ts.linux.dev, linux-kernel@...r.kernel.org
Subject: Re: [RESEND] net: phy: fix NULL pointer dereference in
phy_polling_mode()
On Wed, Aug 06, 2025 at 05:01:22PM +0200, Andrew Lunn wrote:
> > > > Reproduce step is simple:
> > > >
> > > > 1. connect an USB to Ethernet device to USB port, I'm using "D-Link Corp.
> > > > DUB-E100 Fast Ethernet Adapter".
>
> static const struct driver_info dlink_dub_e100_info = {
> .description = "DLink DUB-E100 USB Ethernet",
> .bind = ax88172_bind,
> .status = asix_status,
> .link_reset = ax88172_link_reset,
> .reset = ax88172_link_reset,
> .flags = FLAG_ETHER | FLAG_LINK_INTR,
> .data = 0x009f9d9f,
> };
>
> {
> // DLink DUB-E100
> USB_DEVICE (0x2001, 0x1a00),
> .driver_info = (unsigned long) &dlink_dub_e100_info,
> }, {
>
> Is this the device you have?
>
> > > > 2. the asix driver (drivers/net/usb/asix_devices.c) will bind to this USB
> > > > device.
> > > >
> > > > root@...95evk:~# lsusb -t
> > > > /: Bus 001.Port 001: Dev 001, Class=root_hub, Driver=ci_hdrc/1p, 480M
> > > > |__ Port 001: Dev 003, If 0, Class=Vendor Specific Class, Driver=asix, 480M
> > > >
> > > > 3. then the driver will create many mdio devices.
> > > >
> > > > root@...95evk:/sys/bus/mdio_bus# ls -d devices/usb*
> > > > devices/usb-001:005:00 devices/usb-001:005:04 devices/usb-001:005:08 devices/usb-001:005:0c devices/usb-001:005:10 devices/usb-001:005:14 devices/usb-001:005:18 devices/usb-001:005:1c
> > > > devices/usb-001:005:01 devices/usb-001:005:05 devices/usb-001:005:09 devices/usb-001:005:0d devices/usb-001:005:11 devices/usb-001:005:15 devices/usb-001:005:19 devices/usb-001:005:1d
> > > > devices/usb-001:005:02 devices/usb-001:005:06 devices/usb-001:005:0a devices/usb-001:005:0e devices/usb-001:005:12 devices/usb-001:005:16 devices/usb-001:005:1a devices/usb-001:005:1e
> > > > devices/usb-001:005:03 devices/usb-001:005:07 devices/usb-001:005:0b devices/usb-001:005:0f devices/usb-001:005:13 devices/usb-001:005:17 devices/usb-001:005:1b devices/usb-001:005:1f
> > >
> > > This looks broken - please check what
> > > /sys/bus/mdio_bus/devices/usb*/phy_id contains.
> >
> > root@...95evk:~# cat /sys/bus/mdio_bus/devices/usb*/phy_id
> > 0x00000000
> > 0x00000000
> > 0x00000000
> > 0x02430c54
> > 0x0c540c54
> > 0x0c540c54
> > 0x0c540c54
> > 0x0c540c54
>
> This suggests which version of the asix device has broken MDIO bus
> access.
>
> The first three 0x00000000 are odd. If there is no device at an
> address you expect to read 0xffffffff. phylib will ignore 0xffffffff
> and not create a device. 0x00000000 suggests something actually is on
> the bus, and is responding to reads of registers 2 and 3, but
> returning 0x0000 is not expected.
>
> And then 0x02430c54 for all other addresses suggests the device is not
> correctly handling the bus address, and is mapping the address
> parameter to a single bus address.
Notice that the following return the PHY 3 register 3 value, so
I suspect for anything that isn't PHY 3, it just returns whatever
data was last read from PHY 3. This makes it an incredibly buggy
USB device.
Looking at usbnet_read_cmd(), the above can be the only explanation,
as usbnet_read_cmd() memcpy()'s the data into &res, so the value
in the kmalloc()'d buf (which likely be poisoned on free, or if not
unlikely to reallocate the same memory - that needs to be verified)
must be coming from firmware on the device itself.
asix_read_cmd() will catch a short read, and usbnet_read_cmd() will
catch a zero-length read as invalid.
So, my conclusion is... broken firmware on this device.
--
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!
Powered by blists - more mailing lists