Open soby-mathew opened 6 years ago
An issue has been reported that sometimes on resume after system suspend, the Userspace sometimes crashes due to invalid TLB entries. To rule out any firmware issue, I have had a look at this problem. The ARM impl CPUs guarantee that after reset, the TLB entries are invalidated after reset. Hence no explicit TLB invalidations are performed in firmware on resume. But we see this seqeunce in rockchip platform port here: https://github.com/ARM-software/arm-trusted-firmware/blob/5ff6da948710361dec294f4b5106978501531caf/plat/rockchip/rk3328/drivers/pmu/pmu.c#L595
@TonyXie06 Could you let us know the reason for this TLBI for EL3 before system suspend ? Is there a chance that the TLBI needs to be done for all ELs and not just EL3 ?
@rkchrome , any comments regarding this ?
An issue has been reported that sometimes on resume after system suspend, the Userspace sometimes crashes due to invalid TLB entries. To rule out any firmware issue, I have had a look at this problem. The ARM impl CPUs guarantee that after reset, the TLB entries are invalidated after reset. Hence no explicit TLB invalidations are performed in firmware on resume. But we see this seqeunce in rockchip platform port here: https://github.com/ARM-software/arm-trusted-firmware/blob/5ff6da948710361dec294f4b5106978501531caf/plat/rockchip/rk3328/drivers/pmu/pmu.c#L595
@TonyXie06 Could you let us know the reason for this TLBI for EL3 before system suspend ? Is there a chance that the TLBI needs to be done for all ELs and not just EL3 ?