On 12/20/18 18:31, Ard Biesheuvel wrote: > Print the minimal 'exception occurred' message to the console instead > of straight to the serial port if the console is available. This makes > such messages visible on systems where the console is graphical and > the serial is not connected. > > Contributed-under: TianoCore Contribution Agreement 1.1 > Signed-off-by: Ard Biesheuvel <[email protected]> > --- > ArmPkg/Library/DefaultExceptionHandlerLib/AArch64/DefaultExceptionHandler.c > | 16 +++++++++++++--- > ArmPkg/Library/DefaultExceptionHandlerLib/Arm/DefaultExceptionHandler.c > | 7 ++++++- > ArmPkg/Library/DefaultExceptionHandlerLib/DefaultExceptionHandlerLib.inf > | 1 + > 3 files changed, 20 insertions(+), 4 deletions(-) > > diff --git > a/ArmPkg/Library/DefaultExceptionHandlerLib/AArch64/DefaultExceptionHandler.c > b/ArmPkg/Library/DefaultExceptionHandlerLib/AArch64/DefaultExceptionHandler.c > index 1024bf48c63d..1aaf3c88f21e 100644 > --- > a/ArmPkg/Library/DefaultExceptionHandlerLib/AArch64/DefaultExceptionHandler.c > +++ > b/ArmPkg/Library/DefaultExceptionHandlerLib/AArch64/DefaultExceptionHandler.c > @@ -22,6 +22,7 @@ > #include <Library/PrintLib.h> > #include <Library/ArmDisassemblerLib.h> > #include <Library/SerialPortLib.h> > +#include <Library/UefiBootServicesTableLib.h> > > #include <Guid/DebugImageInfoTable.h> > #include <Protocol/DebugSupport.h> > @@ -159,14 +160,23 @@ DefaultExceptionHandler ( > INT32 Offset; > > if (mRecursiveException) { > - CharCount = AsciiSPrint (Buffer, sizeof (Buffer),"\nRecursive exception > occurred while dumping the CPU state\n"); > - SerialPortWrite ((UINT8 *) Buffer, CharCount); > + STATIC CHAR8 CONST Message[] = "\nRecursive exception occurred while > dumping the CPU state\n"; > + > + if (gST->ConOut != NULL) { > + AsciiPrint (Message); > + } else { > + SerialPortWrite ((UINT8 *)Message, AsciiStrLen (Message)); > + } > CpuDeadLoop (); > } > mRecursiveException = TRUE; > > CharCount = AsciiSPrint (Buffer,sizeof (Buffer),"\n\n%a Exception at > 0x%016lx\n", gExceptionTypeString[ExceptionType], > SystemContext.SystemContextAArch64->ELR); > - SerialPortWrite ((UINT8 *) Buffer, CharCount); > + if (gST->ConOut != NULL) { > + AsciiPrint (Buffer); > + } else { > + SerialPortWrite ((UINT8 *)Buffer, CharCount); > + } > > DEBUG_CODE_BEGIN (); > CHAR8 *Pdb, *PrevPdb; > diff --git > a/ArmPkg/Library/DefaultExceptionHandlerLib/Arm/DefaultExceptionHandler.c > b/ArmPkg/Library/DefaultExceptionHandlerLib/Arm/DefaultExceptionHandler.c > index 0b9da031b47d..9159b579da6f 100644 > --- a/ArmPkg/Library/DefaultExceptionHandlerLib/Arm/DefaultExceptionHandler.c > +++ b/ArmPkg/Library/DefaultExceptionHandlerLib/Arm/DefaultExceptionHandler.c > @@ -21,6 +21,7 @@ > #include <Library/PrintLib.h> > #include <Library/ArmDisassemblerLib.h> > #include <Library/SerialPortLib.h> > +#include <Library/UefiBootServicesTableLib.h> > > #include <Guid/DebugImageInfoTable.h> > > @@ -194,7 +195,11 @@ DefaultExceptionHandler ( > > CharCount = AsciiSPrint (Buffer,sizeof (Buffer),"\n%a Exception PC at > 0x%08x CPSR 0x%08x ", > gExceptionTypeString[ExceptionType], > SystemContext.SystemContextArm->PC, SystemContext.SystemContextArm->CPSR); > - SerialPortWrite ((UINT8 *) Buffer, CharCount); > + if (gST->ConOut != NULL) { > + AsciiPrint (Buffer); > + } else { > + SerialPortWrite ((UINT8 *)Buffer, CharCount); > + } > > DEBUG_CODE_BEGIN (); > CHAR8 *Pdb; > diff --git > a/ArmPkg/Library/DefaultExceptionHandlerLib/DefaultExceptionHandlerLib.inf > b/ArmPkg/Library/DefaultExceptionHandlerLib/DefaultExceptionHandlerLib.inf > index 7609f82d89a1..6bc48714c9dc 100644 > --- a/ArmPkg/Library/DefaultExceptionHandlerLib/DefaultExceptionHandlerLib.inf > +++ b/ArmPkg/Library/DefaultExceptionHandlerLib/DefaultExceptionHandlerLib.inf > @@ -42,6 +42,7 @@ > PeCoffGetEntryPointLib > ArmDisassemblerLib > SerialPortLib > + UefiBootServicesTableLib > > [Guids] > gEfiDebugImageInfoTableGuid >
I feel that invoking such high level functionality from an exception handler is risky, but I do understand this is a last resort / best effort action, and if it *happens* to work on systems with no serial, then it's already a win. However, I'd suggest improving the robustness by preserving the serial write first, and then performing the (optional) console write in addition, second. I guess it can lead to the message appearing twice on serial (if the console is available, and it happens to be multiplexed to the serial port as well), but I think that's a better compromise than possibly losing the message altogether. I don't feel strongly about it either way, I just thought this worth raising. Thanks, Laszlo _______________________________________________ edk2-devel mailing list [email protected] https://lists.01.org/mailman/listinfo/edk2-devel
