On 24.03.21 13:39, Andreas Schwab wrote: > On Mär 24 2021, Heinrich Schuchardt wrote: > >> And foo is obviously "changed between the setjmp invocation and longjmp >> call". >> >> The current version of the patch is: >> https://patchwork.ozlabs.org/project/uboot/patch/[email protected]/ >> >> So I guess we have to declare env as volatile in setjmp() in this >> version of the patch because it is changed between the setjmp and >> longjmp invocations? > > Yes, I think so, or make it static.
The whole point of this variable during test is that it lives on the stack. So static is not what I am looking for. Actually adding volatile does not change the generated machine code because test_longjmp() is marked as noinline and therefore has its own stack frame separated from test_setjmp(). But adding volatile adds to the portability. So I will respin the series. Best regards Heinrich
