On Sun, Jan 11, 2026 at 02:25:54AM -0500, Seungjin Bae wrote:

2026년 1월 10일 (토) PM 3:02, Greg Kroah-Hartman [off-list ref]님이
작성:

quoted

On Sat, Jan 10, 2026 at 02:22:33PM -0500, pip-izony wrote:

quoted

From: Seungjin Bae <redacted>

The `prepare_packet_read()` function calculates the number of packets
required for a transfer by dividing the transfer buffer length by the
maximum packet size (`td->maxpacket`). The `td->maxpacket` is
initialized in `r8a66597_make_td()` function based on the endpoint
descriptor. However, it does not validate whether `td->maxpacket` is
zero before using it in the `DIV_ROUND_UP` macro.

If a malicious USB device sends a descriptor with `wMaxPacketSize` set to
0, it triggers a divide-by-zero exception (kernel panic).

Same here, when can this happen, before a device is bound to the device
or afterward?

thanks,

greg k-h

This logic triggers after the driver is bound, specifically during the
packet preparation phase.

I checked `usb_parse_endpoint()` and confirmed that if `wMaxPacketSize`
is 0, it only prints a warning and continues the binding process.
It does not reject the device at that stage.

However, I missed that `usb_submit_urb()` explicitly checks if
`wMaxPacketSize` is 0 before calling the driver's enqueue function.
Since the core rejects such requests with -EMSGSIZE, this code path
is unreachable in practice.

That's great to verify, thanks for checking.

also, when writing a patch that says "this will fix a crash", please
verify that the tool that found this issue is actually correct and not
making things up.  We have a lot of bad AI generated bug reports these
days, so be careful as we are getting much more sensitive to this.

thanks,

greg k-h