On 08/05/2017 04:35 PM, Rob Clark wrote:
On Sat, Aug 5, 2017 at 10:28 AM, Mark Kettenis mark.kettenis@xs4all.nl wrote:
Date: Sat, 5 Aug 2017 16:01:51 +0200 (CEST) From: Mark Kettenis mark.kettenis@xs4all.nl
Unfortunately something in this patch series breaks things for me on a Banana Pi:
And according to git bisect:
4e3e748a50fc3f43e20c7ff407184596d7c9a589 is the first bad commit commit 4e3e748a50fc3f43e20c7ff407184596d7c9a589 Author: Peter Jones pjones@redhat.com Date: Wed Jun 21 16:39:02 2017 -0400
efi: add some more device path structures Signed-off-by: Peter Jones <pjones@redhat.com> Signed-off-by: Rob Clark <robdclark@gmail.com>hmm, odd.. it is only adding some #define's and structs that are not used until a later commit..
although it does also make 'struct efi_device_path_mac_addr' __packed, which it should have been before. Is this an armv7? I wonder if we have some troubles with unaligned accesses on armv7 that we don't have on aarch64 (or maybe compiler isn't turning access to device-path nodes into byte accesses if it can't do unaligned accesses. (The node in the device-path structure are byte-packed.)
http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.faqs/ka15414.htm...
<cite>On older processors, such as ARM9 family based processors, an unaligned load had to be synthesised in software. Typically by doing a series of small accesses, and combining the results. ... Unaligned access support must be enabled by setting the SCTLR.A bit in the system control coprocessor</cite>
Generally packing structures is not a good idea performance-wise. The sequence of fields should be carefully chosen to fill up to powers of two (2, 4 , 8).
Regards
Heinrich
addr2line the faulting address I guess should confirm that. If this is the issue, it's going to be a bit sad since we'll have to do a lot of copying back/forth of efi_device_path ptrs to aligned addresses :-/
BR, -R