[U-Boot] [PATCH] arm: socfpga: Fix cache configuration

[Marek Vasut]

On Monday, November 09, 2015 at 12:42:24 PM, Stefan Roese wrote:
> Hi Marek,
> 
> On 09.11.2015 01:10, Marek Vasut wrote:
> > On Thursday, September 17, 2015 at 05:30:29 PM, Stefan Roese wrote:
> > 
> > Hi!
> > 
> >> By not defining CONFIG_SYS_ARM_CACHE_WRITEALLOC, the WRITEBACK cache
> >> policy is selected. This leads to much better performance on the
> >> SoCFPGA. A quick network test shows this:
> >> 
> >> Without this patch:
> >> => tftp 100000 big-40mb
> >> Speed: 1000, full duplex
> >> Using dwmac.ff702000 device
> >> TFTP from server 192.168.1.54; our IP address is 192.168.1.252
> >> Filename 'big-40mb'.
> >> Load address: 0x100000
> >> Loading:
> >> #################################################################
> >> 
> >>           ##############################################################
> >>           ###
> >>           #############################################################
> >>           ####
> >>           #############################################################
> >>           #### ##########################
> >>           2.5 MiB/s
> >> 
> >> With this patch:
> >> => tftp 100000 big-40mb
> >> Speed: 1000, full duplex
> >> Using dwmac.ff702000 device
> >> TFTP from server 192.168.1.54; our IP address is 192.168.1.252
> >> Filename 'big-40mb'.
> >> Load address: 0x100000
> >> Loading:
> >> #################################################################
> >> 
> >>           ##############################################################
> >>           ###
> >>           #############################################################
> >>           ####
> >>           #############################################################
> >>           #### ##########################
> >>           7.6 MiB/s
> >> 
> >> A performance improvement of factor ~3.
> >> 
> >> Signed-off-by: Stefan Roese <sr at denx.de>
> >> Cc: Dinh Nguyen <dinguyen at opensource.altera.com>
> >> Cc: Chin Liang See <clsee at altera.com>
> >> Cc: Pavel Machek <pavel at denx.de>
> >> Cc: Marek Vasut <marex at denx.de>
> >> ---
> >> 
> >>   include/configs/socfpga_common.h | 1 -
> >>   1 file changed, 1 deletion(-)
> >> 
> >> diff --git a/include/configs/socfpga_common.h
> >> b/include/configs/socfpga_common.h index 38ae763..a2811ba 100644
> >> --- a/include/configs/socfpga_common.h
> >> +++ b/include/configs/socfpga_common.h
> >> @@ -73,7 +73,6 @@
> >> 
> >>   /*
> >>   
> >>    * Cache
> >>    */
> >> 
> >> -#define CONFIG_SYS_ARM_CACHE_WRITEALLOC
> >> 
> >>   #define CONFIG_SYS_CACHELINE_SIZE 32
> >>   #define CONFIG_SYS_L2_PL310
> >>   #define CONFIG_SYS_PL310_BASE		SOCFPGA_MPUL2_ADDRESS
> > 
> > I hate to say it, but I am running into issues with this patch :-(
> 
> I'm sorry to hear this.
> 
> > I use a standard USB stick here and with this patch, I am getting the
> > following failure (with enabled and disabled cache):
> > 
> > => usb reset
> > resetting USB...
> > USB0:   Core Release: 2.93a
> > scanning bus 0 for devices... unable to get descriptor, error 0
> > usb_new_device: Cannot read configuration, skipping device 058f:6387
> > 1 USB Device(s) found
> > 
> >         scanning usb for storage devices... 0 Storage Device(s) found
> > 
> > => dcache off
> > => usb reset
> > resetting USB...
> > USB0:   Core Release: 2.93a
> > scanning bus 0 for devices... 2 USB Device(s) found
> > 
> >         scanning usb for storage devices... 1 Storage Device(s) found
> > 
> > If I revert this patch, my USB stick works as well.
> > 
> > I am also aware that Stefan mentions that without this patch, cache is
> > not enabled at all. On the other hand, I cannot find any obviously
> > faulty behavior in the dwc2 driver, it does
> > flush_dcache_range()/invalidate_dcache_range() in the right places.
> > 
> > Any ideas please ?
> 
> Perhaps its a timing related issue? As the code is executed faster
> with cache enabled. Just an idea - perhaps there is still some ugly
> code that doesn't do proper timer based loops / delays.

I doubt that's not the case. If I disable cache just around the hcdma bit
in the driver (disable before the flush_dcache_range() and enable after 
invalidate_dcache_range() in the driver), it still fails.

Best regards,
Marek Vasut