On 30.06.16 09:21, Ziyuan Xu wrote:
Hi,
This short series add the fastboot support for rk3288, and I have tested on firefly-rk3288 board.
I encounter an odd issue while debugging usb-otg ep2out-bulk data transfer. The buffer was always zero which mapped to DMA buffer, even though driver call invalidate_dcache_range() to make sure the cpu read memroy data directly. IMHO, dcache was abnormal after something initialization code. Everything work fine without CONFIG_EFI_LOADER. Moreover, invalidate_icache_all cause the issue. I'm not able to figure out why and explain it. But I think there is no need to enable EFI applications on rk3288 platform.
I disagree. Rk3288 (and especially the firefly) is actually a very interesting platform for EFI applications like grub2.
If a full icache invalidate causes breakage, maybe in the non-efi case you're just getting lucky because you code that got overwritten is still in the cache.
Please keep debugging to figure out what is causing the invalid data to go into your icache. Maybe just disable icache altogether for verification and run checksums on the text section after every dma?
Alex