On Sun, Aug 6, 2017 at 10:01 PM, Shawn Guo <shawnguo@xxxxxxxxxx> wrote: > From: Shawn Guo <shawn.guo@xxxxxxxxxx> > > The current usb device g-tx-fifo-size setting in DT causes two problems > for kernel driver. > > 1. On hi6220, there are 15 tx_fifo dedicated for all EPs except EP0, > while DT only provides tx_fifo settings for 6 EPs. It results in the > following annoying complaints from kernel. > > [ 4.451623] dwc2 f72c0000.usb: dwc2_check_param_tx_fifo_sizes: Invalid parameter g_tx_fifo_size[7]=0 > [ 4.461303] dwc2 f72c0000.usb: dwc2_check_param_tx_fifo_sizes: Invalid parameter g_tx_fifo_size[8]=0 > [ 4.470969] dwc2 f72c0000.usb: dwc2_check_param_tx_fifo_sizes: Invalid parameter g_tx_fifo_size[9]=0 > [ 4.480632] dwc2 f72c0000.usb: dwc2_check_param_tx_fifo_sizes: Invalid parameter g_tx_fifo_size[10]=0 > [ 4.490385] dwc2 f72c0000.usb: dwc2_check_param_tx_fifo_sizes: Invalid parameter g_tx_fifo_size[11]=0 > [ 4.500140] dwc2 f72c0000.usb: dwc2_check_param_tx_fifo_sizes: Invalid parameter g_tx_fifo_size[12]=0 > [ 4.509892] dwc2 f72c0000.usb: dwc2_check_param_tx_fifo_sizes: Invalid parameter g_tx_fifo_size[13]=0 > [ 4.519646] dwc2 f72c0000.usb: dwc2_check_param_tx_fifo_sizes: Invalid parameter g_tx_fifo_size[14]=0 > [ 4.529399] dwc2 f72c0000.usb: dwc2_check_param_tx_fifo_sizes: Invalid parameter g_tx_fifo_size[15]=0 > [ 4.539244] dwc2 f72c0000.usb: EPs: 16, dedicated fifos, 1920 entries in SPRAM > > Besides of that, the total 1920 fifo entries isn't fully utilized. > Endpoint Info Control block consumes 128 entries, g-rx-fifo-size > is 512, and g-np-tx-fifo-size is 128. So the fifi entries available > for tx_fifo is: 1920 - 128 - 512 - 128 = 1152. Considering that > the minimal valid tx_fifo size for each EP is 16, it should be > reasonable to allocate 1152 entries as: 128 x 8 + 16 x 7 = 1136 (only > 16 entries unused). With this new setting, we can get more EPs to > use while removing the above warning messages in the meantime. > > 2. Another consequence of above invalid g_tx_fifo_size parameter is that > kernel driver will use values read from hardware register as the > fall-back. The value is 2048 for each EP fifo. That's obviously > invalid either, because even fifo entries for one EP exceeds the > total entries 1920. That's why we see the following fat warning from > function dwc2_hsotg_init_fifo(). The new g-tx-fifo-size settings > help to remove the warning as well. Nice! This has been bothering me for awhile, and your fix seems to resolve it well! I can now remove one of the hack patches I've been carrying. Tested-by: John Stultz <john.stultz@xxxxxxxxxx> (if its useful: Acked-by: John Stultz <john.stultz@xxxxxxxxxx>) Wei: Can we be sure to get this queued for 4.14? thanks -john -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html