[PATCH v3 4/5] x86/mm: Initialize kernel TLB flush requests directly

[Chuyi Zhou]

flush_tlb_kernel_range() uses init_flush_tlb_info(), which records
initiating_cpu with smp_processor_id(). The kernel flush callbacks do
not use this field, but initializing it in a preemptible context would
trigger the CPU-pinning check.

Prepare for allowing preemption during the IPI completion wait by
initializing start and end directly in struct flush_tlb_info and using
the shared threshold predicate to select full or range flushes. Keep the
outer preemption guard unchanged.

Signed-off-by: Chuyi Zhou <zhouchuyi@xxxxxxxxxxxxx>
Link: https://lore.kernel.org/20260522104818.CbT5fyN8@xxxxxxxxxxxxx/
---
 arch/x86/mm/tlb.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/arch/x86/mm/tlb.c b/arch/x86/mm/tlb.c
index 0d03011c6825..2213f1a6f9e1 100644
--- a/arch/x86/mm/tlb.c
+++ b/arch/x86/mm/tlb.c
@@ -1504,13 +1504,14 @@ static void kernel_tlb_flush_range(struct flush_tlb_info *info)
 
 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
 {
-	struct flush_tlb_info info;
+	struct flush_tlb_info info = {
+		.start = start,
+		.end = end,
+	};
 
 	guard(preempt)();
-	init_flush_tlb_info(&info, NULL, start, end, PAGE_SHIFT, false,
-			    TLB_GENERATION_INVALID);
 
-	if (info.end == TLB_FLUSH_ALL)
+	if (tlb_range_exceeds_ceiling(start, end, PAGE_SHIFT))
 		flush_tlb_all();
 	else
 		kernel_tlb_flush_range(&info);