Processor Programming (Continued)
3.11.7 SMM Memory Space
Hardware interrupts, INTRs and NMIs, may be serviced
during an SMI service routine. If interrupts are to be ser-
viced while executing in the SMM memory space, the
SMM memory space must be within the address range of
0 to 1 MB to guarantee proper return to the SMI service
routine after handling the interrupt.
SMM memory space is defined by specifying the base
address and size of the SMM memory space in the SMAR
register. The base address must be a multiple of the SMM
memory space size. For example, a 32 KB SMM memory
space must be located at a 32 KB address boundary. The
memory space size can range from 4 KB to 32 MB. Execu-
tion of the interrupt begins at the base of the SMM memory
space.
INTRs are automatically disabled when entering SMM
since the IF flag (EFLAGS register, bit 9) is set to its reset
value. Once in SMM, the INTR can be enabled by setting
the IF flag. An NMI event in SMM can be enabled by set-
ting NMI_EN high in the CCR3 register (Index C3h[1]). If
NMI is not enabled while in SMM, the CPU latches one
NMI event and services the interrupt after NMI has been
enabled or after exiting SMM through the RSM instruction.
The processor is always in real mode in SMM, but it may
exit to either real or protected mode depending on its
state when SMM was initiated. The IDT (Interrupt Descrip-
tor Table) indicates which state it will exit to.
SMM memory space accesses are always cacheable,
which allows SMM routines to run faster.
3.11.8 SMI Generation
Virtualization software depends on processor-specific
hardware to generate SMI interrupts for each memory or
I/O access to the device being implemented. The GXm
processor implements SMI generation for VGA accesses.
Memory write operations in regions A0000h to AFFFFh,
B0000h to B7FFFh, and B8000h to BFFFFh generate an
SMI.
Within the SMI service routine, protected mode may be
entered and exited as required, and real or protected
mode device drivers may be called.
Memory reads are not trapped by the GXm processor.
The GXm processor traps I/O addresses for VGA in the
following regions: 3B0h to 3BFh, 3C0h to 3CFh, and 3D0h
to 3DFh. Memory-write trapping is performed during
instruction decode in the processor core. I/O read and
write trapping is implemented in the Internal Bus Interface
Unit of the GXm processor.
To exit the SMI service routine, a Resume (RSM) instruc-
tion, rather than an IRET, is executed. The RSM instruc-
tion causes the GXm processor core to restore the CPU
state using the SMM header information and resume exe-
cution at the interrupted point. If the full CPU state was
saved by the programmer, the stored values should be
reloaded before executing the RSM instruction using the
MOV, RSDC, RSLDT and RSTS instructions.
The SMI-generation hardware requires two additional
configuration registers to control and mask SMI interrupts
in the VGA memory space: VGACTL and VGAM. The
VGACTL register has a control bit for each address range
shown above. The VGAM register has 32 bits that can
selectively disable 2 KB regions within the VGA memory.
The VGAM applies only to the A0000h-to-AFFFFh region.
If this region is not enabled in VGA_CTL, then the con-
tents of VGAM is ignored. The purpose of VGAM is to pre-
vent SMI from occurring when non-displayed VGA
memory is accessed. This is an enhancement which
improves performance for double-buffered applications.
The format of each register is shown in Chapter 4 of this
document.
3.11.9.1 SMI Nesting
The SMI mechanism supports nesting of SMI interrupts
through the SMI handler, the SMI_NEST bit in CCR4[6]
(Index E8h), and the Nested SMI Status bit (bit N in the
SMM header, see Table on page 80). Nesting is an impor-
tant capability in allowing high-priority events, such as
audio virtualization, to interrupt lower-priority SMI code for
VGA virtualization or power management. SMI_NEST
controls whether SMI interrupts can occur during SMM.
SMI handlers can optionally set SMI_NEST high to allow
higher-priority SMI interrupts while handling the current
event.
3.11.9 SMI Service Routine Execution
The SMI handler is responsible for managing the SMI
header data for nested SMI interrupts. The SMI header
must be saved before SMI_NEST is set high, and
SMI_NEST must be cleared and its header information
restored before an RSM instruction is executed.
Upon entry into SMM, after the SMM header has been
saved, the CR0, EFLAGS, and DR7 registers are set to
their reset values. The Code Segment (CS) register is
loaded with the base, as defined by the SMAR register,
and a limit of 4 GBytes. The SMI service routine then
begins execution at the SMM base address in real mode.
The Nested SMI Status bit has been added to the SMM
header to show whether the current SMI is nested. The
processor sets Nested SMI Status high if the processor
was in SMM when the SMI was taken. The processor
uses Nested SMI Status on exit to determine whether the
processor should stay in SMM.
The programmer must save the value of any registers that
may be changed by the SMI service routine. For data
accesses immediately after entering the SMI service rou-
tine, the programmer must use CS as a segment override.
I/O port access is possible during the routine but care
must be taken to save registers modified by the I/O
instructions. Before using a segment register, the register
and the register’s descriptor cache contents should be saved
using the SVDC instruction.
When SMI nesting is disabled, the processor holds off
external SMI interrupts until the currently executing SMM
code exits. When SMI nesting is enabled, the processor
can proceed with the SMI. The SMI handler will guarantee
Revision 3.1
83
www.national.com
NSC [ National Semiconductor ]
