On 10/5/26 11:57 PM, Amit Machhiwal wrote:
Hi Shrikanth, Thanks for the review. Please find my responses inline below.
I don't like this sprinkling of preempt disable/enable. Is there not a way to embedd this in try_lock_hpte/unlock_hpte?I understand the scattering of preempt_disable()/preempt_enable() looks ugly. But there are two blockers for that approach: 1. The real-mode hcall dispatch table (hcall_real_table in book3s_hv_rmhandlers.S) calls kvmppc_h_enter(), kvmppc_h_remove(), kvmppc_h_bulk_remove() and others in book3s_hv_rm_mmu.c, which all call try_lock_hpte() from real mode. Adding preempt_disable() inside try_lock_hpte() would affect those real-mode callers, which brings us to your second point.
Fair enough. Reviewed-by: Shrikanth Hegde <[email protected]>
