According to the AMD64 Architecture Programmer's Manual Volume 2 (system programming), a logical address is valid only if the bits 48-63 are all the same as bit 47:
5.3.1 Canonical Address Form
The AMD64 architecture requires implementations supporting fewer than the full 64-bit virtual address to ensure that those addresses are in canonical form. An address is in canonical form if the address bits from the most-significant implemented bit up to bit 63 are all ones or all zeros. If the addresses of all bytes in a virtual-memory reference are not in canonical form, the processor generates a general-protection exception (#GP) or a stack fault (#SS) as appropriate.
So it seems the only valid address ranges are 0x0000_0000_0000_0000 ~ 0x0000_7FFF_FFFF_FFFF and 0xFFFF_8000_0000_0000 ~ 0xFFFF_FFFF_FFFF_FFFF, that is, the lower 128 TiB and higher 128 TiB. However, according to MSDN, the addresses used by Windows x64 kernel don't seem to be the case.
In 64-bit Windows, the theoretical amount of virtual address space is 2^64 bytes (16 exabytes), but only a small portion of the 16-exabyte range is actually used. The 8-terabyte range from 0x000'00000000 through 0x7FF'FFFFFFFF is used for user space, and portions of the 248-terabyte range from 0xFFFF0800'00000000 through 0xFFFFFFFF'FFFFFFFF are used for system space.
So, how can Windows split the virtual address space into lower 8 TiB and higher 248 TiB, despite the hardware specification? I'd like to know why it doesn't cause any problems with the hardware that checks whether the addresses are canonical.
**UPDATE: ** Seems like Microsoft fixed this discrepancy in Windows 8.1. See https://www.facebook.com/codemachineinc/posts/567137303353192 for details.
You're right; current x86-64 hardware with 48-bit virtual address support requires that the high 16 bits be the sign-extension of the low 48 (i.e. bit 47 matches bits [63:48]). That means about half of the 0xFFFF0800'00000000 to 0xFFFFFFFF'FFFFFFFF range is non-canonical on current x86-64 hardware.
Windows is just describing how it carves up the full 64-bit virtual address space, not which parts of that are actually in use on current x86-64 hardware. It can of course only use the 128 TiB that is canonical, from 0xFFFF8000'00000000 to -1. (Note the position of the 8; there's no gap between it and the high 16 bytes that are all-ones, unlike in the theoretical Windows range.)
Top-end servers can be built with 6TiB of RAM or maybe even more. (Xeon Platinum Scalable Processors are apparently available with up to 1.5TiB per socket, and up to 8-way, e.g. the 8180M).
Intel has proposed an extension for larger physical and virtual addressing that adds another level of page tables, https://software.intel.com/sites/default/files/managed/2b/80/5-level_paging_white_paper.pdf, so OSes will hopefully not be stuck without enough virtual address space to map all the RAM (like in the bad old days of PAE on 32-bit-only systems) before we have systems that have more than 128TiB of physical RAM.
Related
There are the words in linux/Documentation/x86/x86_64/5level-paging.rst
Original x86-64 was limited by 4-level paging to 256 TiB of virtual address space and 64 TiB of physical address space.
I know that the limit of virtual address is 256TB because 2^48 = 256TB. But I don't know why its limit of physical is only 64TB.
Suppose we set the size of each page to 4k. Thus a linear address has 12 bits of offset, 9 bits indicate the index in each four level, which means 512 entries per level. A linear address can cover 512^4 pages, 512^4 * 4k = 256TB of space.
This is my understanding of the calculation of space limit. I'm wondering what's wrong with it.
The x86-64 ISA's physical address space limit is unchanged by PML5, remaining at 52-bit. Real CPUs implement some narrower number of physical address bits, saving bits in cache tags and TLB entries, among other places.
The 64 TiB limit is not imposed by x86-64 itself, but by the way Linux requires more virtual address space than physical for its own convenience and efficiency. See x86_64/mm.txt for the actual layout of Linux's virtual address space on x86-64 with PML4 paging, and note the 64 TB "direct mapping of all physical memory (page_offset_base)"
x86-64 Linux doesn't do HIGHMEM / LOWMEM
Linux can't actually use more than phys mem = 1/4 virtual address space, without nasty HIGHMEM / LOWMEM stuff like in the bad old days of 32-bit kernels on machines with more than 1 GiB of RAM (vm/highmem.html). (With a 3:1 user:kernel split of address space, letting user-space have 3GiB, but with the kernel having to map pages in/out of its own space if not accessing them via the current process's user-space addresses.)
Linus's rant about 32-bit PAE expands on why it's nice for an OS to have enough virtual address space to keep everything mapped, with the usual assertion that people who don't agree with him are morons. :P I tend to agree with him on this, that there are obvious efficiency advantages and that PAE is a huge PITA for the kernel. Probably even moreso on an SMP system.
If anyone had proposed a patch to add highmem support for x86-64 to allow using more than 64 TiB of physical memory with the existing PML4 format, I'd expect Linus would tell them 1995 called and wants its bad idea back. He wouldn't consider merging such a patch unless much RAM became common for servers, but hardware vendors still hadn't provided an extension for wider virtual addresses.
Fortunately that didn't happen: probably no CPU has supported wider than 46-bit phys addrs without supporting PML5. Vendors know that supporting more RAM than mainline Linux can use wouldn't be a selling point. But as the doc said, commercial systems were getting up to a max capacity of 64 TiB.
x86-64's page-table format has room for 52-bit physical addresses
The x86-64 page-table format itself has always had that much room: Why in x86-64 the virtual address are 4 bits shorter than physical (48 bits vs. 52 long)? has diagrams from AMD's manuals. Of course early CPUs had narrower physical addresses so you couldn't for example have a PCIe device put its device memory way up high in physical address space.
Your calculation has nothing to do with physical address limits, which is set by the number of bits in each page-table entry that can be used for that.
In x86-64 (and PAE), the page table format reserves bits up to bit #51 for use as physical-address bits, so OSes must zero them for forward compatibility with future CPUs. The low 12 bits are used for other things, but the physical address is formed by zeroing out the bits other than the phys-address bits in the PTE, so those low 12 bits become the low zero bits in an aligned physical-page address.
x86 terminology note: logical addresses are seg:off, and segment_base + offset gives you a linear address. With paging enabled (as required in long mode), linear addresses are virtual, and are what's used as a search key for the page tables (effectively a radix tree cached by the TLB).
Your calculation is just correctly reiterating the 256 TiB size of virtual address space, based on 4-level page tables with 4k pages. That's how much memory can be simultaneously mapped with PML4.
A physical page has to be the same size as a virtual page, and in x86-64 yes that's 4 KiB. (Or 2M largepage or 1G hugepage).
Fun fact: the x86-64 page-table-entry format is the same as PAE, so modern CPUs can also access large amounts of memory 32-bit mode. But of course not map it all at once. It's probably not a coincidence that AMD chose to use an existing well-designed format when making AMD64, so their CPUs would only need two different modes for hardware page-table walker: legacy x86 with 4-byte PTEs (10 bits per level) and PAE/AMD64 with 8-byte PTEs (9 bits per level).
I'm having some hard time understanding PAE. I know it creates a 3rd level of indirection via the PDPT, so that the address translation goes from CR3 -> PDPT(4 entries) -> PD(512 entries) -> PT (512 entries) -> PAGE (4096). But the address is still 32 bits, how do you get 36 bit addresses from this scheme? I'd appreciate an example. How does adding another table "increases" the address space?
PAE changes nothing about 32-bit virtual addresses, only the size of physical address they're mapped to. (Which sucks a lot, nowhere near enough virtual address space to map all those physical pages at once. Linus Torvalds wrote a nice rant about PAE: https://cl4ssic4l.wordpress.com/2011/05/24/linus-torvalds-about-pae/ originally posted on https://www.realworldtech.com/forum/?threadid=76912&curpostid=76973 / https://www.realworldtech.com/forum/?threadid=76912&curpostid=76980)
It also widens a PTE (Page Table Entry) from 4 bytes to 8 bytes, which means 2 levels aren't enough anymore; that's where the small extra level comes to translate the top 2 bits of virtual addresses via those 4 entries.
36-bit only happened to be the supported physical address size in the first generation of CPUs that implemented PAE, Pentium Pro There is no inherent 36-bit limit to PAE.
x86-64 adopted the PTE format, which has room for up to 52-bit physical addresses. Current x86-64 CPUs support the same physical address-size in legacy mode with PAE as they do in 64-bit mode. (As reported by CPUID). That limit is a design choice that saves bits in cache tags, TLB entries, store-buffer entries, etc. and in comparators involved with them. It's normally chosen to be more than the amount of RAM that a real system could actually use, given the commercially available DIMM sizes and number of memory controllers even in multi-socket systems, and still leave room for some I/O address space.
x86-64 came soon after PAE, or soon enough for desktop use to be relevant, so it's a common misconception that PAE is only 36 bits. (Because 64-bit mode is a vastly better way to address more memory, allowing a single process to use more than 2G or 3G depending on user/kernel split.)
My cpuinfo file says that my processor has address sizes as 39 bits physical, and 48 bits virtual. This has got me into some confusion.
Mine is a 64 bit machine. From what I understand, this is the word size of my machine. That is, it will fetch data from memory in chunks of 8 bytes. Also, a 64 bit machine means that the CPU can address 2^64 byte addressable locations, which is a lot. So, manufacturers cut-down some of these lines.
Here are the questions:
If the CPU only generates logical addresses, then what is the need for having 39 bits physical address size.
When we say that the CPU can access 2^64 bytes, do we mean Physical memory or the virtual memory?
I read somewhere that a 64 bit machine has size of its registers as 64 bits, and a 32 bit machine has 32 bit registers. Is this the case?
I think I have confused myself terribly, and need some help. Any other information on this would be appreciated. Thanks!
I can see why people are puzzled considering the number of academic questions posed on this board that suggest there is some mathematical relationship among address sizes.
The processor word size, physical address size, logical address size, and bus size are all independent to some degrees.
If the CPU only generates logical addresses, then what is the need for having 39 bits physical address size.
The CPU translates logical addresses to physical addresses.
When we say that the CPU can access 2^64 bytes, do we mean Physical memory or the virtual memory?
I could be either.
I read somewhere that a 64 bit machine has size of its registers as 64 bits, and a 32 bit machine has 32 bit registers. Is this the case?
Generally this is true for general registers but special purpose registers may be a different size (e.g., floating point, control registers)
There have been many occasions when a processor does not use all available bits for the generation of addresses.
In ancient times, the old MC68000 had 32 bit registers but only a 24 bit address bus.
For the i5 consider that a 64 bit address would control a mind boggling memory space of 17,179,869,184 gigabytes. A stupendously huge number even compared to the storage at Google or the NSA or the planet Earth.
The i5 designers, trim this insane number down to a more manageable 512 gigabytes of physical address space and 262,144 gigabytes of virtual address space.
I'm a bit confused about bit architectures. I just cant find a good article that answers my questions, so I figured I'd ask SO.
Question 1:
When speaking of a 16 bit architecture, does it mean each ram address is 16 bits long? So if I create an int (32 bit) in C++ the variable would take up 2 addresses?
Question 2:
in a 16 bit architecture there are only 2^16 (65536) amount of addresses inside the RAM. Why can't they add more? Is this because 16 bit can't represent a higher value and therefore can't reference to adresses above 65535?
When speaking of a 16 bit architecture, does it mean each ram address is 16 bits long? So if I create an int (32 bit) in C++ the variable would take up 2 addresses?
You'd have to ask whoever was speaking of a 16-bit architecture what they meant by it. They could mean addresses are 16-bits long. They could mean general-purpose CPU registers are 16-bits long. They could mean something else. But there's no way we could know what some hypothetical person might mean. There is no universal definition of what makes something a "16-bit architecture".
For example, the 8032 is an 8-bit architecture with 8-bit general purpose registers. But it has a 16-bit pointer register that can be used to address 65,536 bytes of storage.
Regardless of bitness, almost all systems use byte addresses. So a 32-bit variable will take up 4 addresses on a machine of any bitness.
in a 16 bit architecture there are only 2^16 (65536) amount of addresses inside the RAM. Why can't they add more? Is this because 16 bit can't represent a higher value and therefore can't reference to adresses above 65535?
With 16-bits, there are only 65,536 possible ways those bits can be set. So a 16-bit register has 65,536 possible values.
Yes. Note, though that int on 16-bit architectures is usually just 16 bits wide.
Also note that it doesn't make sense to say that a variable "takes up" two addresses. The correct thing to say is that a 32-bit variable is as wide as two pointers on a 16-bit platform.
It will still occupy four bytes of space, no matter what architecture.
Yes; that's exactly what 16-bit addresses mean.
Note that each of these addresses points to a single byte of memory.
Depends on your definitions of 8-bit and 16-bit architecture.
The 6502 was considered an 8-bit CPU, because it operated on 8-bit values (the register size), yet had 16-bit addresses.
The 68000 was considered a 16-bit CPU, yet had 32-bit registers and addresses.
With x86, it is generally the address size that defines the architecture.
Also, '64-bit' CPUs don't always have a full 64-bit external address bus. They might internally handle addresses of that size, so the virtual address space can be large, but it doesn't mean they can have that much external memory.
Example From Wikipedia - All internal registers, as well as internal and external data buses, were 16 bits wide, firmly establishing the "16-bit microprocessor" identity of the 8086. A 20-bit external address bus gave a 1 MB physical address space (2^20 = 1,048,576). This address space was addressed by means of internal 'segmentation'. The data bus was multiplexed with the address bus in order to fit a standard 40-pin dual in-line package. 16-bit I/O addresses meant 64 KB of separate I/O space (2^16 = 65,536). The maximum linear address space was limited to 64 KB, simply because internal registers were only 16 bits wide. Programming over 64 KB boundaries involved adjusting segment registers (see below) and remained so until the 80386 introduced wider (32 bits) registers (and more advanced memory management hardware).
So you can see that there are no fixed rules that a 16 bit architecture will have 16 address lines only. Don't mix up two things, though it's intuitive to believe so.
I heard that the 8086 has 16-bit registers which allow it to only address 64K of memory. Yet it is still able to address 1MB of memory which would require 20-bit registers. It does this by using another register to to hold another 16 bits, and then adds the value in the 16-bit registers to the value in this other register to be able to generate numbers which can address up to 1MB of memory. Is that right?
Why is it done this way? It seems that there are 32-bit registers, which is more than sufficient to address 1MB of memory.
Actually this has nothing to do with number of registers. Its the size of the register which matters. A 16 bit register can hold up to 2^16 values so it can address 64K bytes of memory.
To address 1M, you need 20 bits (2^20 = 1M), so you need to use another register for the the additional 4 bits.
The segment registers in an 8086 are also sixteen bits wide. However, the segment number is shifted left by four bits before being added to the base address. This gives you the 20 bits.
the 8088 (and by extension, 8086) is instruction compatible with its ancestor, the 8008, including the way it uses its registers and handles memory addressing. the 8008 was a purely 16 bit architecture, which really couldn't address more than 64K of ram. At the time the 8008 was created, that was adequate for most of its intended uses, but by the time the 8088 was being designed, it was clear that more was needed.
Instead of making a new way for addressing more ram, intel chose to keep the 8088 as similar as possible to the 8008, and that included using 16 bit addressing. To allow newer programs to take advantage of more ram, intel devised a scheme using some additional registers that were not present on the 8008 that would be combined with the normal registers. these "segment" registers would not affect programs that were targeted at the 8008; they just wouldn't use those extra registers, and would only 'see' 16 addres bits, the 64k of ram. Applications targeting the newer 8088 could instead 'see' 20 address bits, which gave them access to 1MB of ram
I heard that the 8086 has 16 registers which allow it to only address 64K of memory. Yet it is still able to address 1MB of memory which would require 20 registers.
You're misunderstanding the number of registers and the registers' width. 8086 has eight 16-bit "general purpose" registers (that can be used for addressing) along with four segment registers. 16-bit addressing means that it can only support 216 B = 64 KB of memory. By getting 4 more bits from the segment registers we'll have 20 bits that can be used to address a total of 24*64KB = 1MB of memory
Why is it done this way? It seems that there are 32 registers, which is more than sufficient to address 1MB of memory.
As said, the 8086 doesn't have 32 registers. Even x86-64 nowadays don't have 32 general purpose registers. And the number of registers isn't relevant to how much memory a machine can address. Only the address bus width determines the amount of addressable memory
At the time of 8086, memory is extremely expensive and 640 KB is an enormous amount that people didn't think that would be reached in the near future. Even with a lot of money one may not be able to get that large amount of RAM. So there's no need to use the full 32-bit address
Besides, it's not easy to produce a 32-bit CPU with the contemporary technology. Even 64-bit CPUs today aren't designed to use all 64-bit address lines
Why can't OS use entire 64-bits for addressing? Why only the 48-bits?
Why do x86-64 systems have only a 48 bit virtual address space?
It'll takes more wires, registers, silicons... and much more human effort to design, debug... a CPU with wider address space. With the limited transistor size of the technology in the 70s-80s that may not even come into reality.
8086 doesn't have any 32-bit integer registers; that came years later in 386 which had a much higher transistor budget.
8086's segmentation design made sense for a 16-bit-only CPU that wanted to be able to use 20-bit linear addresses.
Segment registers could have been only 8-bit or something with a larger shift, but apparently there are some advantages to fine-grained segmentation where a segment start address can be any 16-byte aligned linear address. (A linear address is computed from (seg << 4) + off.)