Hi Tom,
On Mon, 10 Feb 2014 08:21:39 -0500, Tom Rini trini@ti.com wrote:
On Mon, Feb 10, 2014 at 10:24:47AM +0100, Albert ARIBAUD wrote:
Hi Tom,
On Tue, 4 Feb 2014 12:05:33 -0500, Tom Rini trini@ti.com wrote:
When we tell the compiler to optimize for ARMv7 it assumes a default of unaligned accesses being supported at the hardware level and can make use of this to perform what it deems as an optimization in any case, including allowing for data to become unaligned. We explicitly disallow this hardware feature so we must tell the compiler.
Cc: Albert ARIBAUD albert.u.boot@aribaud.net Cc: Mans Rullgard mans@mansr.com Signed-off-by: Tom Rini trini@ti.com
NAK -- the discrepancy between the compiler being told to allow native unaligned accesses while at the same time telling the hardware to trap them is conscious and voluntary. It was chosen to help detect unaligned accesses which are rarely necessary and can be explicitly performed by software on a case by case basis.
If and when a specific file requires unaligned access which cannot be made by some other mean than enabling -mno-unaligned-access, then this file should have it added, not the whole of U-Boot.
Right, I recall the discussion, and we chose wrong.
I am quite prepared to discuss whether we chose wrong or right, and to change my mind when the conditions are right, but I'll need more than such a short and simple statement. :)
We aren't being clever and making problems that would appear on armv5 and lower (or non-ARM never allows unaligned access platforms) problems to appear on more common armv7 platforms.
Agreed that we are making problems appear on ARMv7 which are not that much of an issue on ARMv7, but are a true issue on non-ARMv7 arches; that *is* the intent. We want to know as early as possible when some code which runs ok on unaligned-access-friendly platforms such as ARMv7 might cause trouble on unaligned-access-adverse platforms later, once it gets used there too.
We're telling the compiler it's OK to do one thing when it's not and then getting annoying problems such as the EFI partition one where the compiler looks at everything, says we can do $this and then fails at runtime because we lied to it. The whole splashguard set of options is another place where I believe we've shot ourselves in the foot, quite likely.
Ok, so the cause of this patch is not the apparent contradiction between the compiler and hardware setting per se; it is that the EFI code has issues which make it susceptible to crash on unaligned-access-adverse platforms.
This means the trap has worked as expected and has caught some code which does unaligned accesses. Let's analyze it: either we'll conclude it can and should be fixed through e.g. ad hoc unaligned access macros or we'll conclude it can't and we'll add -mno-unaligned-access to the files which can't work otherwise.
Amicalement,