goes with PR #5090 at https://github.com/MicrosoftDocs/cpp-docs/pull/5090/files

Author: TylerMSFT

docs/build/arm64-windows-abi-conventions.md

@@ -1,7 +1,7 @@
 ---
 description: "Learn more about: Overview of ARM64 ABI conventions"
 title: "Overview of ARM64 ABI conventions"
-ms.date: "03/27/2019"
+ms.date: 03/25/2025
 ---
 # Overview of ARM64 ABI conventions
 
@@ -118,7 +118,15 @@ Like AArch32, the AArch64 specification provides three system-controlled "thread
 
 ## Floating-point exceptions
 
-Support for IEEE floating-point exceptions on AArch64 systems is optional. This can be verified by writing a value that enables exceptions to the `FPCR` register and then reading it back. The bits corresponding to supported exceptions will remain set, while the bits corresponding to unsupported exceptions will be reset by the CPU.
+Most ARM hardware doesn't support IEEE floating-point exceptions. You can determine if an ARM CPU supports them by writing a value that enables exceptions to the FPCR register and then reading it back. If the CPU supports floating-point exceptions, the bits corresponding to supported exceptions will remain set, while the bits corresponding to unsupported exceptions will be reset by the CPU.
+
+For ARM CPUs that do support IEEE floating-point exceptions, the behavior on Windows is as follows:
+
+- **Windows ARM ABI**: The Windows kernel silently catches these exceptions and implicitly disables them in the FPSCR register. This approach ensures normalized behavior across processor variants, preventing unexpected exceptions when code developed on a platform without exception support runs on a platform with support.
+
+- **Windows ARM64 ABI**: For processor variants that support hardware floating-point exceptions, the Windows kernel catches these exceptions and disables them in the FPCR register. This ensures consistent behavior across different processor variants.
+
+- The [`_set_controlfp`](/cpp/c-runtime-library/reference/controlfp-s) function on ARM platforms correctly changes the FPCR register when unmasking floating-point exceptions. However, instead of raising an unmasked exception, Windows resets the FPCR register to its defaults every time an FP exception is about to be raised.
 
 ## Parameter passing