mm/kasan: prevent recursive report overflow

[Jackapyun]

Summary

This PR addresses a critical stack overflow issue in the KASAN (Kernel Address Sanitizer) error reporting mechanism when the report handler recursively triggers additional KASAN checks.

Problem

When kasan_report() is executing and calls _alert() for error reporting, the logging function may trigger additional memory access operations that can fail KASAN checks. This causes recursive re-entry into the KASAN report handler, leading to stack overflow and suppressing the original error report.

Solution

The fix implements a dual-approach strategy:

1. Call kasan_stop() before reporting: Halts all KASAN checks when entering the panic path, preventing recursive checks during error message printing
2. Reuse dump_only flag efficiently: Consolidates the panic/non-panic decision logic into a single flag, ensuring consistent behavior without additional stack usage

Key Changes

• Extract dump_only flag calculation at the beginning of kasan_report()
• Call kasan_stop() only when panic reporting is needed (before _alert())
• Reuse the flag for both panic decision and stack dump logic

---

Impact

Stability

• ✅ Eliminates stack overflow risk from recursive KASAN reports
• ✅ Preserves original error reporting despite recursive triggers
• ✅ Reduces stack memory usage in error paths

Compatibility

• ✅ No breaking API changes
• ✅ No impact on non-KASAN builds
• ✅ Maintains existing panic/dump behavior based on configuration

Code Quality

• ✅ Improves code maintainability by eliminating duplicate logic
• ✅ Reduces code duplication in panic decision handling
• ✅ Clear separation of concerns in error handling flow