On Mon, Sep 07, 2020 at 08:57:39AM -0400, Tom Rini wrote: > On Mon, Sep 07, 2020 at 11:52:35AM +0200, Edgar E. Iglesias wrote: > > On Fri, Sep 04, 2020 at 12:43:57PM -0600, Stephen Warren wrote: > > > On 9/4/20 3:07 AM, Edgar E. Iglesias wrote: > > > > From: "Edgar E. Iglesias" <edgar.igles...@xilinx.com> > > > > > > > > PIE requires a 4K aligned load address. If this is not met, trap > > > > the startup sequence in a WFI loop rather than running into obscure > > > > failures. > > > > > > > diff --git a/arch/arm/cpu/armv8/start.S b/arch/arm/cpu/armv8/start.S > > > > #if CONFIG_POSITION_INDEPENDENT > > > > + /* Verify that we're 4K aligned. */ > > > > > > Similar to the comment on the previous patch: I believe the code that > > > implements this check should be outside the #if check, since it's always > > > needed. > > > > But a check for non-PIE would have to be stricter, wouldn't it? > > I.e the load address needs to exactly match the link-time address. > > > > Perhaps we should add the non-PIE check in a separate patch (if at all)? > > If we can catch a bad configuration at link time in the non-PIE case (as > said in another part of this thread I believe) then we should, yes, > thanks!
The non-PIE configuration is expected to be loaded at a specific address. The actual load address cannot be checked at link-time (since it's up to the user at run-time) but given the assumption of a specific load-address, 4K alignment can be enforced at link-time. It really comes down to adding reasonable run-time checks for errors that users may reasonably struggle with. For PIE, checking for 4K aligment is reasonable because it's an easy enough misstake to make (since you've got a binary that's was supposed to handle relocation). For non-PIE, checking for the exact address at run-time is a little bit more border-line IMO but I guess also somewhat reasonable. There are still plenty of cases we can't catch though (loaded at odd addreses, non-RAM address-ranges etc etc). Cheers, Edgar