Timer callback in UEFI application is hanging in AMI BIOS
Posted: Thu Aug 24, 2023 6:05 am
Hello. While porting UEFI application from AARCH64 to X64 i ran into following problem: application was hanging after enabling periodic timer (timer is used in application for I/O polling). Timer callback finishes and nothing executes afterwards.
During debugging i tried to find error in code but failed to do so. Now i have very small helloworld example with following properties:
Here is the whole source code of crashing application. As you can see, it only tries to overwrite two bytes in static arrays in timer callback, however application hangs after first execution of callback. Removing any part of code (using only 1 array, inlining functions) prevents application from hanging. What was expected: spam of "Hello World", "timer begin", "timer end" messages. This is what happens in QEMU and on AARCH64 device. What was received: few "Hello World" messages, then one iteration of timer callback, after that application hangs. By default EDK2 uses "-Os" optimization which requires more complex code to crash, so this functions are explicitly have "-O0" optimization. Is there a severe bug in modern AMI BIOS implementations or there is something wrong with my code?
During debugging i tried to find error in code but failed to do so. Now i have very small helloworld example with following properties:
- It was built in EDK2 according to their instructions (Ubuntu 20 PC with GCC 9.4.0);
It works perfectly on AARCH64 device;
It works perfectly in X64 QEMU;
It hangs on any actual X64 computer (i tested Gigabyte AM2, MSI LGA1200, Asus LGA1700 motherboards)
Here is the whole source code of crashing application. As you can see, it only tries to overwrite two bytes in static arrays in timer callback, however application hangs after first execution of callback. Removing any part of code (using only 1 array, inlining functions) prevents application from hanging. What was expected: spam of "Hello World", "timer begin", "timer end" messages. This is what happens in QEMU and on AARCH64 device. What was received: few "Hello World" messages, then one iteration of timer callback, after that application hangs. By default EDK2 uses "-Os" optimization which requires more complex code to crash, so this functions are explicitly have "-O0" optimization. Is there a severe bug in modern AMI BIOS implementations or there is something wrong with my code?
Code: Select all
#include <Library/UefiApplicationEntryPoint.h>
#include <Library/UefiLib.h>
#include <Library/DebugLib.h>
#include <Library/UefiBootServicesTableLib.h>
static volatile UINT8 test_buf1[200];
static volatile UINT8 test_buf2[200];
void* timer;
__attribute((optimize("O0")))
void assign(volatile UINT8* ptr, int ind)
{
ptr[ind] = 10;
}
__attribute((optimize("O0")))
void test_func(void)
{
assign(test_buf1, 0);
assign(test_buf2, 0);
}
__attribute((optimize("O0")))
void test_interrupt(EFI_EVENT Event, VOID *Context)
{
DEBUG ((EFI_D_ERROR, "timer begin\r\n"));
test_func();
DEBUG ((EFI_D_ERROR, "timer end\r\n"));
}
EFI_STATUS
EFIAPI
UefiMain (
IN EFI_HANDLE ImageHandle,
IN EFI_SYSTEM_TABLE *SystemTable
)
{
EFI_STATUS Status;
Print(L"Hello World \n");
Status = gBS->CreateEvent( EVT_TIMER | EVT_NOTIFY_SIGNAL, TPL_CALLBACK, (EFI_EVENT_NOTIFY)test_interrupt, NULL, &timer);
if (EFI_ERROR(Status)) {
return -1;
}
Status = gBS->SetTimer( timer, TimerPeriodic, EFI_TIMER_PERIOD_MILLISECONDS (16));
if (EFI_ERROR(Status)) {
return -1;
}
while(1)
Print(L"Hello World \n");
return EFI_SUCCESS;
}