On Sat, May 28, 2022 at 11:08:48AM -0700, Linus Torvalds wrote: > This smells like a compiler bug triggered by "there's a 16-bit member > field in that gtf2 structure, and despite it being packed and aligned > to 1, we somehow still align the size to 2". It's an age old thing, it's no compiler bug, and it's completely compliant with the C standards. Implementations are permitted by the C standard to pad structures and unions how they see fit - and some do if it makes sense for performance. The mistake is that people forget this detail, and they expect structs and unions to be laid out a certain way - because it doesn't matter to the same extent on x86. However, as older ARM CPUs could not do unaligned loads, ensuring that things were naturally aligned made complete sense, even if it meant that people who assume the world is x86 got tripped up - the only way around that would be to make every load very expensive. It's not "align to size of 2" in OABI, it tends to be align to a multiple of 4, because the underlying architecture is 32-bit. -- RMK's Patch system: https://www.armlinux.org.uk/developer/patches/ FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!
