ARMv8 fp register corruption in 4.14 PREEMPT_RT series
Hi Tom, I can confirm the ARMv8 fp register corruption reported here:
https://marc.info/?l=linux-rt-users&m=153930857708206&w=2
still exists in 4.14.106-rt56. I was able to reproduce this on both A57 and Carmel ARMv8 CPUs using the float_bessel subtest of LTP. This fix was to backport the following patch from 4.18.y-rt to 4.14-rt:
3afdfe419c33c1655d0763e070350aad44416594
arm64: fpsimd: use preemp_disable in addition to local_bh_disable()
Are you able to pick up this patch into 4.14-rt? I'm sure Anup and I are not the only ones who are hitting this.
-Allen
Hi Allen,
Quoted 9 lines · Allen Martin
Hi Tom, I can confirm the ARMv8 fp register corruption reported here:
https://marc.info/?l=linux-rt-users&m=153930857708206&w=2
still exists in 4.14.106-rt56. I was able to reproduce this on both A57 and Carmel ARMv8 CPUs using the
float_besselsubtest of LTP. This
fix was to backport the following patch from 4.18.y-rt to 4.14-rt:
Quoted 6 lines
3afdfe419c33c1655d0763e070350aad44416594
arm64: fpsimd: usepreemp_disablein addition tolocal_bh_disable()Are you able to pick up this patch into 4.14-rt? I'm sure Anup and I are not the only ones who are hitting this.
Yes, I was just about to start a new 4.14 update and will pull this one in. Thanks for letting me know.
Tom
Quoted 1 line
-Allen