As noted in https://www.amazon.com/Linux-Device-.../dp/0596005903 the kernel virtual memory (top 1GB in IA-32 address space) is partitioned into kernel logical addresses and kernel virtual addresses. Kernel logical addresses are mapped directly which means that you can map virtual addresses to physical addresses by subtracting a certain (0xC0000000) value. For moving btwn mappings a pair of macros are defined: __pa() and __va().

My questions are: "What are the usecases for these macros?" and: "What is the benefit of direct mapping?"

I have heared that direct mapping allows using larger pages (e.g., 4MB) and consequently leads to more efficient translation. Are translations for the directly mapped region performed using page tables?

Unfortunately, this book published in 2005 no longer represents the actual implementations used within the Linux kernel today – twelve years later.

Physical addresses are those needed by DMA: a device controller card cannot perform memory-translation.

Virtual addresses, in both user-space and kernel-space, use the address-translation hardware. You must go through this translation process to determine where – and if – the memory resides in physical space.

It is erroneous to think that one translates from one to the other by means of "simple subtraction."

Today, the most up-to-date sources for device driver information today is the actual source-code itself. Go find a driver, one that is similar to the one you want, and study it carefully. Many drivers are "cabbaged" from other ones. In all cases it is critical that you are looking at current Linux source code, since fundamental changes (such as randomization of kernel-space addressing to reduce predictability) have occurred with every major-version release.

And this also bears emphasis: be sure that you are looking at the right CPU architecture branch. Device drivers work very close to the hardware. Macros simplify this coding considerably, but there still may be architecture-specific considerations throughout any driver(s).

1 members found this post helpful.

I have attached a picture from https://www.amazon.com/Professional-.../dp/0470343435. It seems that the kernel has a direct mapping to the physical RAM.

Attached Thumbnails

 

And it may have...NINE YEARS AGO when that second book was written.

Again; look at the source code and the documentation on kernel.org. THAT is going to be current and relevant.

3 members found this post helpful.

A memory, any memory stores data, obviously, but also has another key attribute, which is the memory address. Without the address, a memory becomes very inefficient.

To look at it from a human analogy, a library obviously has books (data), but it also has a shelving system like the Dewey Decimal Classification (address). This allows efficient means to retrieve the data. An example from our memory in the brain is the Method of loci or memory palace. Normally we just dump "memories" data into our brain, this method provides a visual addressing system for retrieving tose memories efficiently. The keyword is obviously efficiency here. the memory could be disorganized, and we could retrieve information our of it, example, a cluttered desk, or a pile of books. But we will have to employ other means.

The associativity is an attempt to map the main memory addresses to the cache memory that provides flexibility (to store any addressed location anywhere is cache) while providing an efficient means to retrieve the location later. Remember that a cache entry is supposed to be accessed again and again. So if your addressing scheme is inefficient, that inefficiency will occur with each access.

Regards,

I am talking about mapping of kernel virtual address to physical address of main memory. You are talking about mapping of memory address to cache blocks. I think that you have misunderstood my question.

No ... he did not misunderstand your question.

(And, I can say that "authoritatively.")