[PATCH] usb: dwc3: Always deassert xilinx resets

[Sean Anderson]

On 1/12/26 19:49, Thinh Nguyen wrote:
> On Fri, Jan 09, 2026, Sean Anderson wrote:
>> On 1/9/26 01:01, Pandey, Radhey Shyam wrote:
>> > [AMD Official Use Only - AMD Internal Distribution Only]
>> > 
>> >> -----Original Message-----
>> >> From: Thinh Nguyen <Thinh.Nguyen at synopsys.com>
>> >> Sent: Friday, January 9, 2026 6:19 AM
>> >> To: Sean Anderson <sean.anderson at linux.dev>
>> >> Cc: Thinh Nguyen <Thinh.Nguyen at synopsys.com>; open list:DESIGNWARE
>> >> USB3 DRD IP DRIVER <linux-usb at vger.kernel.org>; Frager, Neal
>> >> <neal.frager at amd.com>; Simek, Michal <michal.simek at amd.com>; open list
>> >> <linux-kernel at vger.kernel.org>; moderated list:ARM/ZYNQ ARCHITECTURE
>> >> <linux-arm-kernel at lists.infradead.org>; Philipp Zabel <p.zabel at pengutronix.de>;
>> >> Pandey, Radhey Shyam <radhey.shyam.pandey at amd.com>; Greg Kroah-Hartman
>> >> <gregkh at linuxfoundation.org>
>> >> Subject: Re: [PATCH] usb: dwc3: Always deassert xilinx resets
>> >>
>> >> On Tue, Jan 06, 2026, Sean Anderson wrote:
>> >> > If we don't have a usb3 phy we don't need to assert the core resets.
>> >> > Deassert them even if we didn't assert them to support booting when
>> >> > the bootloader never released the core from reset.
>> > Is it a customized bootloader ? i.e it assert reset but don't deassert.
>> 
>> No. Most peripheral resets are asserted on PoR. So if the bootloader
>> doesn't deassert them then Linux has to.
>> 
>> My goal is to make init_serdes() in psu_init_gpl.c optional and do all
>> serdes initialization in the phy driver (and in the consumer drivers). I
>> have this working for DP/PCIe. I'm working on SATA, and I don't think
>> USB/SGMII need much special. This gives the following advantages:
>> 
>> - On some boards (mine) the reference clocks may not be configured in
>>   SPL/FSBL. So ILL calibration will fail (and take a long time to do so)
>>   unless we defer initialization to U-Boot/Linux where the phy driver
>>   can request the clocks.
>> - If PCIe/SATA are not used in U-Boot, ILL calibration can be deferred
>>   until Linux when it can be done it parallel with other initialization.
>> - We will have flexibility to switch between different serdes
>>   configurations at runtime. For example, this could allow the
>>   bootloader to fixup the devicetree to support PCIe and SATA M.2
>>   drives, depending on what the user has plugged in.
>> 
>> > I think ideally core /APB reset should be done independent on
>> > MAC 2.0/3.0 configuration.
>> 
>> I agree, but I think the existing code does this optimization to reduce
>> boot time when the bootloader has already initialized USB. I have
>> preserved that in this patch.
>> 
> 
> I think all this info is useful. Can we include it in the change log?

OK, how about:

I am working on moving serdes initialization to the phy (and consumer)
drivers to improve flexibility and boot times (depending on
configuration). Currently, core resets are released in the bootloader by
init_serdes() in psu_init_gpl.c. In order to remove init_serdes, we
need to handle the case where the bootloader never released the core
resets. If we don't have a usb3 phy we don't need to assert the core
resets, but deassert them anyway to handle this case.

We could assert all resets every boot, but I believe the existing
procedure is an optimization to reduce boot time when the bootloader has
already initialized USB. So this patch preserves the separate code
paths.