Re: [PATCH] usb: cdns3: Do not access memory after free

[Siddharth Vadapalli <s-vadapalli@ti.com>]

On Thu, Aug 28, 2025 at 05:46:56PM +0200, Marek Vasut wrote:
> On 8/24/25 10:02 AM, Siddharth Vadapalli wrote:
> > On Sat, Aug 23, 2025 at 02:21:18PM +0200, Marek Vasut wrote:
> > > On 8/23/25 4:07 AM, Siddharth Vadapalli wrote:
> > > 
> > > Hi,
> > > 
> > > > > > > > I was planning to test this patch but the change being made is only
> > > > > > > > applicable to Controller Versions:
> > > > > > > > 	#define DEV_VER_NXP_V1		0x00024502
> > > > > > > > 	#define DEV_VER_TI_V1		0x00024509
> > > > > > > > and not to:
> > > > > > > > 	#define DEV_VER_V2		0x0002450C
> > > > > > > > 	#define DEV_VER_V3		0x0002450d
> > > > > > > > 
> > > > > > > > Since I don't have an SoC and a Board with DEV_VER_TI_V1, I cannot test
> > > > > > > > it. However, the change looks correct to me.
> > > > > > > > 
> > > > > > > > Reviewed-by: Siddharth Vadapalli <s-vadapalli@ti.com>
> > > > > > > The change does indeed look correct.
> > > > > > > 
> > > > > > > Do you know who might still have that board and could test ? (and which
> > > > > > > board/soc is that) ?
> > > > > > 
> > > > > > None of the boards that I have worked with have a DEV_VER_TI_V1 version
> > > > > > of the controller. I also tried to use the Linux device-tree to check if
> > > > > > I could identify the SoC/board but I was unable to do so.
> > > > > Do you know which SoC is V2 and V3 ?
> > > > 
> > > > I spent more time on this and found out that J721E SR 1.0 has the
> > > > controller with DEV_VER_TI_V1 version but other revisions of J721E as
> > > > well as all of the following SoCs have DEV_VER_V3 version of the
> > > > controller:
> > > > AM64, AM68, AM69, J7200, J721S2, J722S, J742S2 and J784S4.
> > > > 
> > > > I will try to find an SR 1.0 J721E SoC and test the patch on it and
> > > > share the results here.
> > > This is awesome, thank you !
> > 
> > I was able to get an SR 1.0 J721E SoC and also the J721E
> > Common-Processor-Board for testing the patch.
> > 
> > Enabling debug info in the cdns3/gadget.c driver, I see:
> > 	cdns-usb3-peripheral usb@6000000: Device Controller version: 00024509
> > 	cdns-usb3-peripheral usb@6000000: USB Capabilities:: 09203324
> > 	cdns-usb3-peripheral usb@6000000: On-Chip memory cnfiguration: 00000c34
> > which confirms the Controller Version.
> > 
> > However, the code changed by the current patch is only affecting the
> > execution of code associated with Workaround 2 which is described in
> > detail in the driver. I am summarizing it here for your reference:
> > 
> > Issue:
> > 	Controller for OUT endpoints has shared on-chip buffers for all
> > 	incoming packets. The buffer acts as a FIFO, due to which,
> > 	missing a DMA descriptor for one packet will block subsequent
> > 	transfers/packets meant for other Endpoints that were queued.
> > 
> > Workaround:
> > 	If the Endpoint Status register indicates a Descriptor Miss,
> > 	rearm the DMA transfer to complete the missed transfer.
> > 
> > In order to test the patch, I used USBACM for STDIO/STDOUT to check if I
> > could trigger the descriptor miss workaround. The command I ran was:
> > setenv stdio usbacm; setenv stdout usbacm;
> > /dev/ttyACM0 showed up on my PC and I was also able to access the U-Boot
> > prompt via ACM0. While it is functional, I didn't see the code
> > associated with the workaround being triggered. Reviewing the driver
> > again, I identified that the workaround is being disabled very early on
> > within "cdns3_wa2_gadget_ep_queue()" with the comment stating:
> > 
> > 	 * If transfer was queued before DESCMISS appear than we
> > 	 * can disable handling of DESCMISS interrupt. Driver assumes that it
> > 	 * can disable special treatment for this endpoint.
> > 
> > Given the above, it doesn't seem easy to recreate the issue since I
> > would have to trigger a descriptor miss event prior to the very first
> > USB transfer for the Endpoint. Please let me know if you have any
> > suggestions for speeding up testing.
> Can you maybe simply force-enable the workaround code and see if that itself
> works, without crashing ? If yes, then I would say let's apply this.

I did try it out when I wasn't able to trigger the workaround and I
didn't see a crash. It seemed to work without issues but I didn't
mention it since I felt that this wasn't the right way to test it.
Nevertheless, I can confirm that I did not see a crash when forcing the
workaround code to execute.

Regards,
Siddharth.