[PATCH v1 1/3] usb: dwc2: reset AHB hclk domain before init

Yunzhi Li lyz at rock-chips.com
Thu Aug 13 20:28:56 PDT 2015

在 2015/8/14 8:09, John Youn 写道:
> On 8/11/2015 12:57 AM, Yunzhi Li wrote:
>> We initiate dwc2 usb controller in BIOS, when kernel driver
>> start-up we should reset AHB hclk domain to reset all AHB
>> interface registers to default. Without this the FIFO value
>> setting might be incorrect because calculating FIFO size need the
>> power-on value of GRXFSIZ/GNPTXFSIZ/HPTXFSIZ registers.
>> This patch could avoid warnning massage like in rk3288 platform:
>> [    2.074764] dwc2 ff580000.usb: 256 invalid for
>> host_perio_tx_fifo_size. Check HW configuration.
>> ......
> I didn't receive the other two patches in this series so I was
> confused about where the "ahb_reset" was coming from when I
> replied to your other patch.
> I see you changed the name and documented the DT so never mind.
> Another thing is that there probably shouldn't be a debug
> message on the IS_ERR condition since that is the common case
> and of no interest to other platforms.
> The other two resets you added aren't used by the driver
> anywhere right? Maybe those should be left out until they are.
> John
Hi John ,

   Here is the other two patches :

   ahb_reset is hreset_n signal of dwc2 IP. Our rk3288 SoC implement 
connect this signal to a special
register in clock ang reset unit (CRU) module, set this register will 
reset dwc2 control and status registers(CSR)
to default value. You could find more info in <<DesignWare Cores USB 2.0 
Hi Speed On-TheGo (OTG) Databook 3.10a>>
4.4.1 System Clock and Reset Signals.

   Our problem is that dwc2_get_hwparams() reads fifo size registers and 
reguards it as the power-on reset value,
then dwc2_set_param_host_perio_tx_fifo_size() will check this value and 
make sure the new fifo size value is no bigger
than the power-on reset value. But we init and set these fifo registers 
in BIOS, so here hw->xxx_fifo_size is not the
real power-on reset vaule. So we hope to reset CSR before 

I have another ideal: we might use GRSTCTL.CSftRst instead of hreset_n 
to reset dwc2 CSR.

More information about the Linux-rockchip mailing list