Software Environments - Heterogeneous Multicore Processor Technologies for Embedded Systems - page 161

Linux

Physical page
allocation

(1)

HIGHMEM area
allocation

Memory access

(2) (5)

Architecture independent
Architecture dependent

Update PTE Resolve TLB Miss

(3) (6)

(4)

PTE

Software
Hardware

TLB

Fig. 5.6 Calling sequence of HIGHMEM memory access

does not have the I/O space, so we do not need to worry about the I/O space. The
address of the memory-mapped I/O space must not be changed regardless of the
existence or nonexistence of HIGHMEM, in order to keep the driver's source
code compatible. However, if I/O devices use DMA, which accesses the physical
address space directly, it is necessary to modify DMA's physical address in the
driver's source code.
5. Application Layer

By implementing the above design, application programs can run without any
modi fi cations.
In summary, we have designed a new HIGHMEM framework according to
the following policy:
a. Place the HIGHMEM area in the virtual address space.
b. Leave the memory-mapped I/O area as it is.
c. Leave the user area as it is.

5.1.3.2 Implementation

The calling sequence of physical page allocation to use the HIGHMEM area is
shown in Fig. 5.6 . This function returns the physical page address. In step (1), the
physical page allocator calls the HIGHMEM area allocator to access the HIGHMEM
area. In step (2), the HIGHMEM area allocator allocates the virtual address for a
HIGHMEM page and calls the page table entry (PTE) updater. In step (3), the PTE
updater updates the PTE. In step (5), a TLB miss occurs when Linux accesses a
virtual address, and the address is not registered in the TLB. At this time, the TLB