On 12/17/18 9:43 AM, Alex Bennée wrote: >> Use "register" TBFLAG_ANY to indicate shared state between >> A32 and A64, and "registers" TBFLAG_A32 & TBFLAG_A64 for >> fields that are specific to the given cpu state. >> >> Move ARM_TBFLAG_BE to shared state, instead of its current >> placement within "Bit usage when in AArch32 state". > > This looks like a worthwhile clean-up but really we need to defer this > calculation. It's in the top 2 hot functions on my build load stressor: > > 5.74% qemu-system-aarch64 [.] helper_lookup_tb_ptr > 5.05% qemu-system-aarch64 [.] get_phys_addr_lpae > 4.39% qemu-system-aarch64 [.] cpu_get_tb_cpu_state > 3.73% qemu-system-aarch64 [.] tcg_gen_code > 3.65% qemu-system-aarch64 [.] qht_lookup_custom
Yes, I know. We need to put on our thinking caps and come up with a scheme that minimizes the chances of missing an update to any (partially) pre-computed value. That has been Peter's (legitimate) resistance to such a scheme so far. There are a *lot* of variables that go into computing tb->flags. In the meantime this patch just makes the task of adding bits less ugly. r~
