On 01/08/2013 12:14 PM, R Sricharan wrote: (..) > We had this problem of speculative aborts in the kernel uncompress code > as well, which maps all of 4GB address space. It was solved by setting > the non-DRAM region as non-executable(XN) and with client permissions > to the domain in the DACR register. > > This way speculative prefetches are avoided not only to the page 0, > but also to other read sensitive I/O regions. > > I have created a similar patch in u-boot and posted a RFC now. > I was using your first patch [1] and rest from me. (..) > > Please let me know your take on that.
Hi Sricharan, Your solution to this issue looks more elegant to me than my unmapping page 0 completely. I tested your patches and they work for me on both GP (without security) and EMU (with security) OMAP5 ES1.0 devices. I'll keep them, thanks :) You can add my 'Tested-by' if you want: Tested-by: Vincent Stehlé <v-ste...@ti.com> Best regards, V. _______________________________________________ U-Boot mailing list U-Boot@lists.denx.de http://lists.denx.de/mailman/listinfo/u-boot
