Hi Johannes & Jukka, Thank you for looking into this, I just wanted to mention something, so it does not get over looked, if it is still appropriate:
In the past (9 or so years ago) there was an issue with floating point that required the stack to be 8 byte aligned. It was too long ago, so I do not remember the details, j (There is a note in the release notes, "ARM EABI: Fix stack alignment required for passing floating point values.") But I do remember that printf printed garbage until the STACK_ALIGN_xxxx stuff was done. Regards, David -----Original Message----- From: Schock, Johannes - NIVUS GmbH [mailto:johannes.sch...@nivus.com] Sent: Friday, September 25, 2020 5:49 AM To: dev@nuttx.apache.org Subject: RE: Stack size alignment on arm > 1. Set the stack size as aligned already in > arch/arm/src/common/arm_createstack.c: > > - tcb->adj_stack_size = stack_size; > + tcb->adj_stack_size = STACK_ALIGN_DOWN(stack_size); > > 2. fix the do_stackcheck in a way that it doesn't start at (unaligned) > top, but at the first aligned stack address? > > 3. just leave nuttx as is, and go fix all the stack sizes in px4? > > > Initally I did the (1.) which seems to fix my problem. I had a look, and you're right. I've missed that arm_stack_color and do_check are now working differently concerning calculation of top (stackbase) address. If you don't mind (and since it is my mistake) I will provide a fix that weekend. Perhaps 32bit downalign instead of 32bit upalign the arm_stack_color stackbase would also be a solution as well, since TLS size is also 32bit aligned. This would keep the requested size in tcb->adj_stack_size. But I have to verify with arm_usestack.c Johannes
