ISA DMA: Difference between revisions

From OSDev.wiki
Jump to navigation Jump to search
[unchecked revision][unchecked revision]
Content added Content deleted
m (DMA moved to ISA DMA: This article only describe ISA DMA, it do not describe any other kind of DMA)
mNo edit summary
 
(31 intermediate revisions by 16 users not shown)
Line 1: Line 1:
;The main points about ISA DMA are:
DMA, or Direct Memory Access, in today's PC architecture is in many ways like an appendix. It is used by the floppy disk controller, ISA sound cards, ISA network cards and parallel ports (if they support ECP mode). Whilst interrupt, keyboard and timer interface circuits have obvious and relevant uses, the DMA controller and it's programming interface are still well and truly stuck in the 1970's where they were first designed. Modern PCI controllers typically have their own "bus mastering", which is far better.
* ISA DMA is not the same thing as PCI Busmastering DMA;
* ISA DMA channels 1, 2 and 3 are available for 8 bit transfers to ISA peripherals;
* ISA DMA channels 5, 6 and 7 are available for 16 bit transfers to ISA peripherals;
* Transfers must not cross physical 64 KB boundaries and must never be bigger than 64 KB;
* Transfers must be physically contiguous, and can only target the lowest 16 MB of physical memory;
* ISA DMA is slow - theoretically 4.77 MB/second, but more like 400 KB/second due to ISA bus protocols;
* ISA DMA frees up CPU resources, but adds an extremely heavy load to the memory bus;
* Very few devices currently use ISA DMA -- only internal floppies, some embedded sound chips, some parallel ports, and some serial ports.


Notes:
The idea behind DMA is that you can set up a 'channel' with an address pointing into memory and the length of the data to be transferred. Once set up, you can tell the peripheral owning the 'channel' to do whatever it is supposed to do (e.g. read a sector). Then the CPU can go onto something else. Periodically when the data bus isn't being used by the CPU the DMA chip takes over and transfers data between the peripheral and memory without involving the CPU. When the transfer is complete (e.g. an entire sector has been sent to the floppy drive) the DMA chip will than signal that it is finished. The DMA chip can even signal if it has run out of data, allowing the CPU to program some new information and another transaction to proceed.
* Sound Blaster and Sound Blaster PRO only support 8 bit DMA;
DMA can improve the speed of a system quite a bit and was borrowed by Intel (who designed the DMA controller chip) from the old 1960's mainframes which had DMA channels for all devices (CPU's weren't all on a single chip and very slow back then).
* [[Sound Blaster 16]]+ supports both;
* [[Floppy disk controllers]] only support 8 bit DMA and are hardwired to use DMA Channel 2.


== There Is More Than One Kind of DMA on a PC ==
Of course all good ideas can have downsides and while Intel can't really be blamed for what is about to be described, IBM certainly can.
Modern PCI controllers always have their own 'Busmastering DMA', which is far better than ISA DMA. Even USB floppy drives send their DMA
data using PCI Busmastering, through the PCI USB controller.
PCI Busmasters can access memory with 32 bit addressing. Newer PCI cards are starting to support 64 bit addressing (although at the moment most
don't). Typically PCI cards use 'scatter-gather' bus mastering, where one page is used as a directory of data pages. This almost completely overcomes
the "physical memory only" limitation of all forms of DMA.




== ISA DMA Background ==
==DMA Genesis. Chapter 1, Verse 1==
'''ISA DMA''' (''Industry Standard Architecture Direct Memory Access''), like ISA itself, is an [https://en.wikipedia.org/wiki/Appendix_(anatomy) appendix] for modern PCs. It is used by the
In the beginning there was a PC, but the PC was slow. IBM looked down from the heavens and said "Slap on a DMA controller - that should speed it up." IBM's heart was in the right place, its collective brains were elsewhere as the DMA controller never met the needs of the system. The PC/AT standard contains 2 Intel 8237A DMA chips, connected as Master/Slave. The second chip is Master, and its first line (Channel 4) is used by the first chip, which is Slave. (This is unlike the interrupt controller, where the first chip is Master.) The 8237A was designed for the old 8080 8-bit processor and this is probably the main reason for so many DMA problems. 8088 and 8086 processors chosen by IBM for its PC were too advanced for the DMA controller.
internal floppy disk controller, ISA sound cards, ISA network cards, and parallel ports (if they support ECP mode). Whilst interrupt, keyboard
and timer interface circuits have obvious and relevant uses, the ISA DMA controller and its programming interface are still well and truly stuck in the
1970s where they were first designed.


The idea behind DMA is that you can set up a 'channel' with an address pointing into memory and the length of the data to be transferred. Once set up, the
Each 8237 DMA chip has 4 DMA channels. DMA0-DMA3 on the first chip and DMA4-DMA7 on the second. The first DMA controller is wired up for 8 bit transfers, while the second is wired up for 16 bit transfers.
CPU can tell the peripheral owning the channel to do whatever it is supposed to do (e.g. read a sector). Then the CPU can go do something else.
When the memory bus isn't being used by the CPU, the DMA chip takes over and transfers data between the peripheral and memory without involving the CPU. When
the transfer is complete (e.g. an entire sector has been sent to the floppy drive) the DMA chip then signals that it is finished. The DMA chip can even
signal if it has run out of data, allowing the system to locate the next block of data to transfer on the same DMA transaction.
DMA can improve the speed of a system quite a bit and was borrowed by Intel (who designed the DMA controller chip) from the old 1960s mainframes which had
DMA channels for all devices (CPUs weren't all on a single chip and very slow back then).

Of course all good ideas can have downsides and while Intel can't really be blamed for what is about to be described, IBM certainly can.


In the beginning there was a PC, but the PC was slow. IBM looked down from the heavens and said "Slap on a DMA controller -- that should speed it up." IBM's
DMA Channel 0 is unavailable as it was used for memory refresh, and remains reserved because of this (even though modern computers don't use it). DMA channel 4 cannot be used for peripherals because it is used for cascading the other DMA controller.
heart was in the right place; its collective brains were elsewhere as the DMA controller never met the needs of the system. The PC/AT standard contains 2
Intel 8237A DMA chips, connected as Master/Slave. The second chip is Master, and its first line (Channel 4) is used by the first chip, which is Slave. (This
is unlike the interrupt controller, where the first chip is Master.) The 8237A was designed for the old 8080 8-bit processor and this is probably the main
reason for so many DMA problems. The 8088 and 8086 processors chosen by IBM for its PC were too advanced for the DMA controller.


Previously it was mentioned that a DMA controller is able to signal completion and even ask for more information. Unfortunately this would make expansion
DMA Channels 1, 2 and 3 can only transfer 64 KB of information at a time (the internal registers are only 16 bit). In order to extend this, IBM added an EXTERNAL page register allowing access to 16 MB of memory (24 bits total). If a DMA transfer crosses a 64 KB boundary, the internal count register wraps around to zero, the external page register is NOT incremented. The DMA controller will happily provide the peripheral attached with whatever it finds at the new address. As a rule 8 bit DMA transfers must not cross 64 KB boundaries and must never be bigger than 64 KB.
slots too big, so IBM left all of the connections to the DMA chips off. The only time you know when a transfer is complete is for a peripheral to signal an
interrupt. This implies that all peripherals using an ISA DMA channel are limited to no more than 64 KB transfers for fear of upsetting the DMA controller.


Even with the PC/AT, IBM began bypassing the ISA DMA used in the PC/XT and used [[ATA PIO Mode]] for the hard disk. This
DMA Channels 5, 6 and 7 transfer data 16 bits at a time, by shifting addresses one bit to the right and clearing the lower bit. The addressing is exactly the same except you can now start on 128 KB boundaries and transfer up to 128 KB of information at a time before the DMA boldly transfers what has never been tranferred before....
was because of the 64 KB limitations outlined above and the fact that the 286 processor could perform 16 bit transactions at 6 MHz. Even the ISA bus could
run at a speed of up to 12 MHz, far faster than the 4.77 MHz the DMA controller was running at.


Expansion card designers were also upset with DMA's lack of capabilities, noticeably 'Hard-Card' hard disk expansion card manufacturers who depended on the
Previously it was mentioned that the DMA controller is able to signal completion and even ask for more information. Unfortunately this would make expansion slots too big, so IBM left all of the connections to the DMA chips off. The only time you know when a transfer is complete is for a peripheral to signal an interrupt. This implies that all peripherals using a DMA channel are limited to no more than 64/128 KB transfers for fear of upsetting the DMA controller.
speed of data transfer.


To get around the limitations of the 'on board' DMA controller, expansion card manufacturers began to put their own DMA controllers on their expansion
Finally, ***ALL*** DMA controllers run at 4Mhz. (Information taken from ISA Specification) ***NO EXCEPTIONS***.
cards. They functioned exactly the same way as the 'on board' DMA, 'stealing memory bus cycles' when the processor wasn't looking and thus improving the
Ignore that Ghz tag on your processor or the fact that even PCI runs at 33MHZ. DMA controllers are fixed at this rate. This includes EISA and PS/2 32-bit DMA controllers. The only differences between these and the stock ISA DMA controllers is a extra page register allowing for 64k transfers anywhere in 4GBytes of space and the ability to do 32-bit transfers, and these DMA controllers exists only on EISA and MCA systems, which are now obsolete and are not described here.
performance of the system as a whole. These "ISA Bus Masters" are still usually limited to the lower 16 MiB of memory, but do not have the 4.77 MHz issue.
This trend continued through the creation of the PCI bus, which eventually entirely replaced the ISA bus in PCs.


== Technical Details ==
So after reading all the above the main point are
*DMA channel 1, 2 and 3 are for 8 bit transfers
Each 8237 DMA chip has 4 DMA channels. DMA0-DMA3 on the first chip and DMA4-DMA7 on the second. The first DMA controller is wired up for 8 bit transfers,
while the second is wired up for 16 bit transfers. On some tutorials or other wiki articles, you will sometimes see the '''second''' DMA chip (channels 4 to 7) labeled as the "'''master'''" controller, and the first (channels 0 to 3) called the "slave". This is highly confusing, and these terms will not be used again, here.
*DMA channel 5, 6 and 7 are 16 bit transfers
*Transfers shouldn't cross 64 KB or 128 KB boundaries
*Transfers can only be from the lowest 16 MB of memory
*DMA is slow - theoretically 4 MB/second but more like 500 KB/second due to ISA bus protocols


DMA Channel 0 is unavailable as it was used for a short time for DRAM memory refresh, and remains reserved because of this (even though modern computers
Sound Blaster and Sound Blaster PRO only support 8 bit DMA
don't use it). DMA channel 4 cannot be used for peripherals because it is used for cascading the other DMA controller.
Sound Blaster 16+ supports both
Floppy disk controllers only support 8 bit DMA and usually use DMA Channel 2


The internal address registers of the DMA controller are only 16 bits. In order to extend this, IBM added one ''external'' "page register" byte per channel,
==Busmaster DMA==
allowing access to 16 MB of memory (24 bits total). If a DMA transfer crosses a 64 KB boundary, the internal address register wraps around to zero,
and the external page register is '''not''' incremented. The DMA controller will happily continue the transfer with whatever data it finds at
the new address.


ISA-based DMA controllers are specified to run at 4.77 MHz. '''No exceptions'''. If the "front-side" (memory) bus of a system runs at 133 MHz, it will be
The IBM/XT used DMA channel 3 for hard disk connections but instead of using DMA the IBM/AT used direct writing to the hard disk using the processor, this was because of the 128 KB limitations outlined above and the fact that the 286 processor could perform 16 bit transactions at 6 MHz. Even the ISA bus could run at a speed of up to 12 MHz, far faster than the 4.77 MHz the DMA controller was running at.
artificially slowed down to 4.77 MHz when transferring each ISA DMA byte/word. This includes EISA and PS/2 32-bit
controllers, even though these controllers have an extra page register (which allows a 4 GiB addressing space)
and the ability to do 32 bit transfers. (These DMA controllers exist only on EISA and MCA systems, which are now obsolete and are not further
described here.)


Expansion card designers were also upset with DMA's lack of capabilities, noticeably "Hard-Card" hard disk expansion card manufacturers who depended on the speed of data transfer.


== Programming Details ==
To get around the limitations of the "onboard" DMA controller, expansion card manufacturers began to put their own DMA controllers on their expansion cards. The controllers didn't have any of the design 'features' of the 8237A and could address all 16 MBytes of ISA addressable memory with the full 16 bit data bus width. They functioned exactly the same way as the "onboard" DMA, 'stealing bus cycles' when the processor wasn't looking and thus improving the performance of the system as a whole.


=== 16 bit issues ===
Busmaster DMA is probably easier to understand than normal DMA. Provided you know what registers are on your expansion card, you can send them the address and the amount of data to transfer. The expansion card will send you an interrupt upon completion.
The 16 bit channels (5, 6, and 7) have a special addressing scheme to handle the way they increment addresses. The internal registers increment by 1,
PCI Bus Masters can of course access memory with 32 bit addressing. Newer PCI cards are starting to support 64 bit addressing (although at the moment most don't). Typically PCI cards use "scatter-gather" bus mastering, where one page is used as a directory of data pages. This is makes them especially suitable for paging OSs (where contiguous pages are rare).
but the memory address needs to be incremented by 2 between each access. The solution is that the "start address" that the CPU stores in the DMA controller
needs to be shifted right by 1 bit. On each memory access, this internal 16 bit address value is incremented by one, then that value is taken and shifted
one bit to the left (clearing the lower bit) before being used as an address. The external "page register" addressing byte is then appended in the normal
way. It is important to note that the upper bit of the internal address is '''lost''' when it is shifted left -- it is not ORed into the page register byte.
This prevents DMA transfers bigger than 64 KB from working even though they should be technically possible for 16 bit channels, because the address register
will effectively wrap around to zero without incrementing the page register byte.


=== Physical Memory vs. Paging ===
ISA Bus Masters are still limited to the lower 16Mbytes of memory - this is an ISA limitation.
Paged memory mapping is exclusively controlled by the CPU. The whole point of DMA is to bypass the CPU. Therefore, no DMA can ever access any virtual
memory addresses. All DMA is always done on physical memory addresses only. ISA DMA has a 16 MB physical address limit.


Since DMA runs independently of the CPU, it is important that an OS allocates a block of contiguous physical memory for the DMA transfer in
a way that prevents that memory from being used for any other purpose, or swapped out, until the DMA transfer is complete.


Note: VM86 mode does not use ''physical'' addressing. The memory addresses are ''fake''. In VM86 mode the OS must emulate any DMA transactions on
==Programming the PC DMA controllers==
behalf of an application.


===Paging===
=== Buffer Size ===
A typical 1.44 MB floppy disk can easily transfer 36 sectors of data in a single transfer. This is only 18 KB. The biggest internal floppy with
worst-case formatting may be able to transfer 84 sectors at once. This is still only 42 KB. A Soundblaster card may run best with a 64 KB buffer. It
is never necessary to try to double-buffer ISA DMA transfers, because they are so slow anyway. There are at most 6 usable DMA channels, and it is not
necessary to allocate a full 64 KB to each of them. Putting all of this together, any OS should be easily able to allocate all the ISA DMA physical
memory that it needs from a 256 KB pool, or even only half of that.


=== The Flip-Flop ===
As this article assumes you are writing your own OS, the subject of memory management is very important unless you are writing a 286 based operating system. This is because as mentioned earlier, all DMA transactions must occur in the lower 16Mbytes of memory (Yes, I'm repeating myself - memorize - it is crucial)
Many devices on a PC (e.g. ATA disk drives) use 8 bit IO Ports to receive 16 bit values. This is done using a flip-flop. The device expects the low
To be more specific all DMA transactions must occur in PHYSICAL MEMORY. The DMA controller only functions with PHYSICAL memory, if you pass it a virtual address happily mapped into anywhere in memory, the DMA controller will happily ignore your carefully constructed page mapping and access whatever it decides to. VM86 mode does not use PHYSICAL addressing. The physical addresses are FAKE. In VM86 mode the OS must emulate any DMA transactions on behalf of an application, probably in a totally non-DMA way. EMM386 supplies virtual DMA functions to Windows 3.x, but these are basically the same, EMM386 "fakes" the DMA transaction. DMA has no knowledge of paging whatsoever and I should point out that this includes Bus Master DMA.
byte first. As soon as it receives a byte, the flip-flop changes state and then the device expects the high byte. When the high byte is received,
Finally (and this is a great laugh when it happens) don't forget to "lock" the pages in memory. Imagine the floppy drive trying to read data from memory and you swap that data out to the hard drive. The DMA chip doesn't know -
the flip-flop changes state again, and the device expects a new low byte. Usually, each 16 bit 'register' will have its own flip-flop, but the ISA
"Paging? Whats that then? Dunno. You've just mapped in the password file? No problem, I'll write that out to the floppy instead."
DMA controller has a problem regarding this.
DMA memory is like Memory Mapped I/O - do not move - do not page.
(I apologise for sounding really dictatorial, it's just that DMA is so easy to screw up. Lots' more things to screw up later! - just wait till we get to commands)


On a 8237 chip, there is '''only one''' flip-flop. And there are eight of the 16 bit registers. And there can be up to three device drivers all competing
One way to handle this situation is to manage memory below 16 MB using a "bitmap" so that contiguous pages that don't cross 64 KB or 128 KB boundaries can be found, and use free page stacks (or anything else) for memory above 16 MB.
for the use of that one flip-flop simultaneously.


This creates two serious problems. One is "contention issues". The other is that it is difficult to be sure what state the flip-flop is currently in.
<pre>
The standard solution for dealing with the flip-flop state issue is to reset the flip-flop to "low byte" state every single time you want to use it, just
An example for getting it right:
so you can be certain it is in the proper state before sending bytes. There are only two solutions to "contention": either use a [[lock]], or allow only
(Only Psuedo Code for this, it is up to you to design your memory management)
one ISA DMA driver, so that contention is impossible.


=== Masking DRQ ===
When your OS starts it detects (or is told of) a device requiring DMA memory.
Setting up a DMA transfer always requires setting up both "ends" of the transfer. That is, whichever peripheral owns the DMA channel needs to be told
Map the lowest 64K of PHYSICAL memory elsewhere in memory and aim the DMA channel you are
to transfer a block of data via DMA. And the DMA controller needs to be told the memory address, the transfer length, perhaps a transfer "mode", and
setting up at address 0.
a transfer direction (read or write). So, one of these two things always needs to be done ''first'' -- and it is usually the peripheral that has a long
Set up the DMA channel for the correct type of transaction (more on this later).
latency time before being ready to transfer the first byte. If you set up the peripheral first, however, its first DMA request signal (often called
DRQ) may arrive while you are in the middle of setting up the DMA controller.


The answer is to mask DRQ for a particular channel while you are initializing the DMA controller. There are three ways of temporarily disabling channels
Next DMA Channel, map the next 64K of PHYSICAL memory somewhere, aim your DMA channel at
described below.
address 64K.
Set up the DMA channel for the correct type of transaction (more on this later).


=== Transfer Length ===
Complete building your pages stacks / bitmaps whatever and add any lower memory left over.
The value that gets stored into each Count Register is always the transfer length (either bytes or words) '''minus 1'''. If you forget to subtract
</pre>
the one, you will get an error on your transfer.


=== Interrupt on Completion ===
Apart from making sure that you have a copy somewhere of the information in the first 4K of memory and the Extended BIOS Data Area (these are for VM86 tasks if you decide to add BIOS/DOS/VESA/PCI support later), the above example will allow you to have a flat paged memory area from 0 to forever while still allowing you OS to use old DMA dependent peripherals. The device drivers work with CPU paged memory areas, while the DMA controller is happily working with low memory UNDER 16MBYTES.
Have the driver write data to the paging relocated area, send commands to the peripheral and the DMA chip will access PHYSICAL memory correctly when asked to do so.
As implemented in a PC, the DMA controller can not send interrupts. Hopefully, whichever peripheral "owns" each DMA channel will send the CPU an
interrupt when a transfer completes. However, some peripherals may '''not''' send an interrupt if a transfer fails with an error. As always, timeouts
are important.


=== The Registers ===
The master and slave DMA controllers are very similar, so (to save space) both of them have been combined into the following table. Please try
not to let this confuse you.
Note: for Address and Count Registers on channels 5 to 7, see [[#16 bit issues|16 bit issues]] above.


Each 8237A has 18 registers, addressed via the I/O Port bus:
===Registers===
Each 8237A has 5 registers addressed via I/O space


{| {{wikitable}}
{| {{wikitable}}
|-
|-
!Channels 0-3
!8237A No. 1
!Channels 4-7
!8237A No. 2
|
|
|
|
|
|-
|-
!IO Port
!Address
!IO Port
!Address
!Size
!Read or Write
!Read or Write
!Function
!Function
|-
|0x00
|0xC0
|Word
|W
|Start Address Register channel 0/4 (unusable)
|-
|0x01
|0xC2
|Word
|W
|Count Register channel 0/4 (unusable)
|-
|0x02
|0xC4
|Word
|W
|Start Address Register channel 1/5
|-
|0x03
|0xC6
|Word
|W
|Count Register channel 1/5
|-
|0x04
|0xC8
|Word
|W
|Start Address Register channel 2/6
|-
|0x05
|0xCA
|Word
|W
|Count Register channel 2/6
|-
|0x06
|0xCC
|Word
|W
|Start Address Register channel 3/7
|-
|0x07
|0xCE
|Word
|W
|Count Register channel 3/7
|-
|-
|0x08
|0x08
|0xD0
|0xD0
|Byte
|R
|R
|Status Register
|Status Register
Line 103: Line 217:
|0x08
|0x08
|0xD0
|0xD0
|Byte
|W
|W
|Command Register
|Command Register
Line 108: Line 223:
|0x09
|0x09
|0xD2
|0xD2
|Byte
|W
|W
|Request Register
|Request Register
|-
|-
|0x0A
|0x0A
|0xD4
|0XD4
|Byte
|W
|W
|DMA Channel Mask Register
|Single Channel Mask Register
|-
|-
|0x0B
|0x0B
|0xD6
|0xD6
|Byte
|W
|W
|DMA Mode Register
|Mode Register
|-
|-
|0x0C
|0x0C
|0xD8
|0XD8
|Byte
|R
|W
|Byte / Word Register
|Flip-Flop Reset Register
|-
|-
|0x0D
|0x0D
|0xDA
|0xDA
|Byte
|R
|R
|Intermediate Register
|Intermediate Register
|-
|-
|0x0D
|0x0F
|0xDA
|0XDE
|Byte
|W
|W
|DMA Mask Register
|Master Reset Register
|-
|0x0E
|0xDC
|Byte
|W
|Mask Reset Register
|-
|0x0F
|0xDE
|Byte
|RW
|MultiChannel Mask Register (reading is undocumented, but it works!)
|-
|-
|}
|}


Each DMA Channel has an External Page Address Register that is added to the upper 8 bits of the DMA transfer address
Each Channel also has an external R/W Page Address Register that contains the upper 8 bits of the 24 bit transfer memory address:


{| {{wikitable}}
{| {{wikitable}}
|-
|-
|0x87
|0x87
|DMA Channel 0 Page Address Register (bits A17 - A24)
|Channel 0 Page Address Register (unusable)
|-
|-
|0x83
|0x83
|DMA Channel 1 Page Address Register (bits A17 - A24)
|Channel 1 Page Address Register
|-
|0x81
|Channel 2 Page Address Register
|-
|-
|0x82
|0x82
|DMA Channel 2 Page Address Register (bits A17 - A24)
|Channel 3 Page Address Register
|-
|0x81
|DMA Channel 3 Page Address Register (bits A17 - A24)
|-
|-
|0x8F
|0x8F
|DMA Channel 4 Page Address Register (bits A17 - A24)
|Channel 4 Page Address Register (unusable)
|-
|-
|0x8B
|0x8B
|DMA Channel 5 Page Address Register (bits A17 - A24)
|Channel 5 Page Address Register
|-
|-
|0x89
|0x89
|DMA Channel 6 Page Address Register (bits A17 - A24)
|Channel 6 Page Address Register
|-
|-
|0x8A
|0x8A
|DMA Channel 7 Page Address Register (bits A17 - A24)
|Channel 7 Page Address Register
|-
|-
|}
|}


==== Useful Registers ====
(Bit Patterns from Indispensable PC Hardware Book - I'm commenting the useful parts i.e. the bits you can actually use, a lot of the 'features' of the 8237 do not function in the PC and the fact they do not work introduces new 'features'. Please assume that if anything isn't explicitly mentioned, it implicitly does not function. Messing with things not mentioned will usually result in screwing up of your computer. Don't take my word for it (I could be Kermit the Frog for all you know!) so feel free to ask other people and do whatever makes you feel good.)


;Status Registers 0x08 and 0xD0 (Read)
;Single Channel Mask Registers 0x0A and 0xD4 (Write)
{| {{wikitable}}
{| {{wikitable}}
|-
|-
Line 182: Line 315:
|Bit 0
|Bit 0
|-
|-
|
|REQ3
|
|REQ2
|
|REQ1
|
|REQ0
|TC3
|
|MASK_ON
|TC2
|SEL 1
|TC1
|SEL 0
|TC0
|}
|}




These registers are used to mask (or unmask) DRQ for a single channel only, on either the master or slave DMA chip.
REQ3-0 1 = DMA Request Pending
That is, if you do not want to figure out the mask states of all the other channels, you can mask/unmask DRQ for one channel at a time.
TC3-0 1 = Transfer Complete
Use the SEL 0 and 1 bits to select the channel, and the MASK_ON bit to set or clear masking for it.
This register isn't very important in light of the fact that the 8237 can't send an IRQ to tell you that it has finished. Usually there is no need to poll this register as again the peripheral owner will send an interrupt when a transaction has finished and polling would be a total waste of time.
Note that masking DMA channel 4 will mask 7, 6, 5 and 4 due to cascading.




;Command Registers 0x08 and 0xD0 (Write)
;MultiChannel Mask Registers 0x0F and 0xDE (Read and Write)
{| {{wikitable}}
{| {{wikitable}}
|-
|-
Line 210: Line 344:
|Bit 0
|Bit 0
|-
|-
|
|DACKP
|
|DRQP
|
|EXTW
|
|PRIO
|MASK3
|COMP
|MASK2
|COND
|MASK1
|ADHE
|MASK0
|MMT
|}
|}


Setting the appropriate bits to 0 or 1 allows you to unmask or mask (respectively) DRQ for those channels. Using this register means that your driver needs
This register really shows how incompatible the 8237 is with the PC hardware.
to know the desired mask states of ''all'' the channels at that moment. There are several ways to do this, but one is simply to read this register, first.
Let's start with EXTW and COMP. These increase the speed of DMA transfer by 25% be removing one of the clock cycles. Does it work? No.
Note that masking DMA channel 4 will mask 7, 6, 5 and 4 due to cascading.
PRIO. When zeroed, this allows DMA priorities to be rotated allowing freedom and liberty for all peripherals that share the data bus. Does it work? No.
MMT and ADHE. Did you know that the IBM PC could do memory to memory transfers since 1981? Thats right, hardware sprites, hardware frame buffering from one location to another. Does it work? No.
COND. Hooray the only bit in the control register that does something useful. Setting this bit disables the DMA controller. This lets you set up multiple DMA channels without masking each and every channel.


;Request Registers 0x09 and 0xD2 (Write)
:Used for memory to memory transfers and setting up priority rotation- absolutely useless.

;DMA Channel Mask Registers 0x0A and 0xD4 (Write)


;DMA Mode Registers 0x0B and 0xD6 (Write)
{| {{wikitable}}
{| {{wikitable}}
|-
|-
Line 242: Line 371:
|Bit 0
|Bit 0
|-
|-
|MOD1
|
|MOD0
|
|DOWN
|
|AUTO
|
|TRA1
|
|TRA0
|STCL
|SEL1
|SEL1
|SEL0
|SEL 0
|}
|}


Setting this register is a little tricky as it depends highly on the peripheral you are programming the DMA controller for. However, the
To mask a channel for programming / whatever set STCL to 1 and SEL0 and SEL1 to select the channel to be masked
driver for the peripheral is the entity that needs to set this register, and it should know what mode the peripheral needs.
sending 0x4(100b) would mask DMA Channel 0 of the appropriate 8237.
sending 0x0(000b) would unmask DMA Channel 0 of the appropriate 8237.
To unmask a channel set STCL to 0
This register only allows you to mask / unmask a '''SINGLE''' DMA channel if you need to mask multiple DMA channels you need to use -


* '''SEL0''' and '''SEL1''' select the channel you want to change;
;DMA Mask Registers 0x0F and 0xDE (Write)
* '''TRA0''' and '''TRA1''' selects the transfer type;
** 0b00 runs a self test of the controller;
** 0b01 Peripheral is writing to memory;
** 0b10 Peripheral is reading from memory;
** 0b11 invalid.
* '''AUTO''': When this bit is set, after a transfer has completed the channel resets itself to the address and count values you programmed into it. This is great for floppy transfers. Read in a track - the values set themselves up for reading again immediately. For writing you'd only need to alter the transfer mode - not the addresses. Some expansion cards do not support auto-init DMA such as Sound Blaster 1.x. These devices will crash if used with auto-init DMA. Sound Blaster 2.0 and later do support auto-init DMA.
* '''DOWN''': Reverses the memory order of the data, when set. Memory is accessed from high addresses down to low addresses (the address is decremented between each transfer).
* '''MOD0''' and '''MOD1''': This is where some problems can arise based on the peripheral the DMA controller is attached to. The DMA controller has several modes:
** 0b00 = Transfer on Demand;
** 0b01 = Single DMA Transfer;
** 0b10 = Block DMA Transfer;
** 0b11 = Cascade Mode (use to cascade another DMA controller).

Single transfer mode is good for peripherals than cannot cache a lot of data at once.
Non-82077AA Floppy controllers, Sound Blaster, and Sound Blaster Pro should use Single Transfer DMA Mode.

Block transfer mode is good for peripherals that can buffer entire blocks of information. An example of this is a hard disk controller board.

Demand transfer mode is good for peripherals that start and stop intermittently such as a tape drive. The drive can read a whole load of information for as
long as it can and the suspend the transfer to move to another section of the tape. Newer floppy controllers also work well with demand transfer because
they have FIFO buffers to store information being read and written (but you need to set up the FIFO properly). The peripheral controls the flow, and as
the information flow is uninterrupted, performance can be gained. CPUs in these later computers generally have caches and can continue working
uninterrupted during a demand DMA transfer. Older computers will slow down as their CPUs wait for the memory bus to become available.

;Flip-Flop Reset Registers 0x0C and 0xD8 (Write)
;Master Reset Registers 0x0D and 0xDA (Write)
;Mask Reset Registers 0x0E and 0xDC (Write)

Send any value to the Reset registers to activate them.
Master Reset sets Flip-Flop low, clears Status, and sets all Mask bits ON.
Mask Reset sets all Mask bits OFF.

The following statement from the previous wiki article needs to be verified on real hardware, because it is likely to be wrong: "The Reset Flip-Flop command
must be sent ''before any'' 16 bit transaction. The flip-flop ''does not reset'' after the DMA controller has received the second byte."


==== The Other Registers ====

;Status Registers 0x08 and 0xD0 (Read)
{| {{wikitable}}
{| {{wikitable}}
|-
|-
Line 270: Line 435:
|Bit 0
|Bit 0
|-
|-
|REQ3
|
|REQ2
|
|REQ1
|
|REQ0
|
|TC3
|STC3
|TC2
|STC2
|TC1
|STC1
|TC0
|STC0
|}
|}


* REQ3-0: When set: DMA Request Pending.
Setting the appropriate bits to 0 or 1 allows you to mask or unmask those channels.
* TC3-0: When set: Transfer Complete.
DMA channels 3, 2, 1, 0 or 7, 6, 5, 4. Note that masking DMA channel 4 will mask 7, 6, 5 and 4
* Reading this register will clear the TC bits.
due to cascading.

This register isn't very important in light of the fact that the 8237 can't send an IRQ to tell you that it has finished. Usually there is no need to poll
this register as the peripheral (at the other end of the DMA) will send an interrupt when a transaction has completed.



;DMA Mode Registers 0x0B and 0xD6 (Write)
;Command Registers 0x08 and 0xD0 (Write)
{| {{wikitable}}
{| {{wikitable}}
|-
|-
Line 296: Line 465:
|Bit 0
|Bit 0
|-
|-
|DACKP
|MOD1
|DRQP
|MOD0
|EXTW
|IDEC
|PRIO
|AUTO
|COMP
|TRA1
|COND
|TRA0
|ADHE
|SEL1
|MMT
|SEL0
|}
|}


This register is tricky as it depends highly on the peripheral you are programming the DMA controller for.
This register really shows how incompatible the 8237 is with the PC hardware.
* Let's start with EXTW and COMP. These increase the speed of DMA transfer by 25% be removing one of the clock cycles. Does it work? No.
* PRIO. When zeroed, this allows DMA priorities to be rotated allowing freedom and liberty for all peripherals that share the data bus. Does it work? No.
* MMT and ADHE. Did you know that the IBM PC could do memory to memory transfers since 1981? That's right, hardware sprites, hardware frame buffering from one location to another. Does it work? No.
* COND. Hooray the only bit in the control register that does something useful. Setting this bit disables the DMA controller. This is one way to set up multiple DMA channels without masking each and every channel.


;Request Registers 0x09 and 0xD2 (Write)
*SEL0 and SEL1 selects the channel you want to change.
Used for memory to memory transfers and setting up priority rotation -- absolutely useless.
*TRA0 and TRA1 selects the transfer type.
**00 runs a self test of the controller.
**01 DMA Channel is for writing to memory
**10 DMA Channel is for reading from memory
**11 invalid


;"Intermediate" Registers 0x0D and 0xDA (Read)
AUTO. When this bit is set, after a transfer has completed the channel resets itself to the values you programmed into it.
Never implemented on PCs. Useless.
This is great for floppy transfers. Read in a track - the values set themselves up for reading again immediately. For writing you'd only need to alter the transfer mode - not the addresses. Some expansion cards do not support auto-initialized DMA such as Soundblaster and Soundblaster Pro. These devices will crash if used with auto-initialised DMA
Soundblaster 16 + do support Autoinitialisation and 16 bit DMA.
IDEC. Increment/Decrement. I guess if you wanted to write big endian data to a floppy drive you could. Big endian processor based computers wouldn't be using an 8237.
MOD0 and MOD1. This is where some problems can arise based on the peripheral the DMA controller is attached to.


== Examples ==
<pre>
The DMA controller has several modes:
00 = Transfer on Demand
01 = Single DMA Transfer
10 = Block DMA Transfer
11 = Cascade Mode (use to cascade another DMA controller)
</pre>


=== Floppy Disk DMA Initialization ===
Single transfer mode is good for peripherals than cannot cache a lot of data at once. A good example is early floppy controllers that have only a very small buffer. The DMA controller and CPU can share the data bus happily.
Sound Blaster and Sound Blaster Pro use software or Single Transfer DMA Mode.


You need only implement 1 to 3 tiny routines to perform a DMA transfer. This example is the Floppy Drive controller (probably the most common followed by SoundBlaster).
Block transfer mode is good for peripherals that can buffer entire blocks of information. A good example of this is an ESDI or SCSI controller without Bus Master DMA.


Note: the following code is not optimal, because there is an OUT to the same IO port twice (in two places). This causes an extra delay on the IO Port
Demand transfer mode is good for peripherals that start and stop intermittently such as a tape drive. The drive can read a whole load of information for as long as it can and the suspend the transfer to move to another section of the tape. Newer floppy controllers also work well with demand transfer becasue they have FIFO buffers to store information being read and written. The peripheral controls the flow, and as the information flow is uninterrupted, performance can be gained. CPU's in these later computers generally have caches and can continue working uninterrupted during a demand DMA transfer. Older computers will slow down as their CPU's wait for the data bus to become available.
bus. Real code should separate the two "out 0x4" and "out 0x5" calls with an "out" to some other port.


<syntaxhighlight lang="asm">
initialize_floppy_DMA:
; set DMA channel 2 to transfer data from 0x1000 - 0x33ff in memory
; paging must map this _physical_ memory elsewhere and _pin_ it from paging to disk!
; set the counter to 0x23ff, the length of a track on a 1.44 MiB floppy - 1 (assuming 512 byte sectors)
; transfer length = counter + 1
out 0x0a, 0x06 ; mask DMA channel 2 and 0 (assuming 0 is already masked)
out 0x0c, 0xFF ; reset the master flip-flop
out 0x04, 0 ; address to 0 (low byte)
out 0x04, 0x10 ; address to 0x10 (high byte)
out 0x0c, 0xFF ; reset the master flip-flop (again!!!)
out 0x05, 0xFF ; count to 0x23ff (low byte)
out 0x05, 0x23 ; count to 0x23ff (high byte),
out 0x81, 0 ; external page register to 0 for total address of 00 10 00
out 0x0a, 0x02 ; unmask DMA channel 2
ret
</syntaxhighlight>


Once you have set up your start address and transfer length you do not need to touch it again, if you are using autoinit. Once reading or writing
In addition to the above registers the 8237 has three command registers.
is selected, you don't need to change that, either. To ''change'' selecting reading or writing you use the mode register.


<syntaxhighlight lang="asm">
The first command register is 0xd8 which is called the 'reset flip-flop' register.
The DMA controller is an 8-bit chip. In order to write 16-bit values into an address register or count register you must write two 8 bit values appropriate register address (to follow).
Imagine however that you only write one 8 bit value and the the next time try to write a 16 bit value. The DMA chip will be expecting the high byte, not the low byte. Clearing the flip-flop guarantees getting the address into the DMA controller in the right order.
To reset the flip-flop write anything to the register :-

;ASM
out 0xd8, 0xFF

;C
<pre>
void reset_flipflop_DMA()
{
outb(0xdb, 0xFF);
}
</pre>

Another dictate - The reset flip-flop command must be sent BEFORE ANY 16 bit transaction. The flip-flop DOES NOT RESET after the DMA controller has received the second byte. Easily missed, can cause many problems.

The second command register is the 'master reset and clear'.
Writing anything to this register (0x0d) resets all count, address and mask values ready for you to program.

;ASM
out 0x0d, 0xFF
;reinitialize all DMA channels now

;C
<pre>
void hard_reset_DMA()
{
outb(0x0d, 0xFF);
}
</pre>

The third command register is the 'clear mask register'.
Writing anything to this register (0xdc) will clear the mask register (wow!) releasing all DMA channels to accept DMA requests.

;ASM
out 0xdc, 0xFF
; reinitialize all DMA channels now

;C
<pre>
void unmask_all_DMA()
{
outb(0xdc, 0xFF);
}
</pre>

==Address and Count Registers==

Each DMA chip has 4 16-bit address registers, 4 16-bit count registers and 4 *external* 8-bit page registers.
As previously mentioned a DMA controller CANNOT address or transfer more than 65536 bytes or words of information at a time.
The page registers can allow access to the lower 16 MBytes of memory but are NOT incremented when crossing a 64k boundary. The easiest way is to make sure the address you supply in the address register is ZERO and that any addressing that you program the DMA controller to do is done using the external page register. It seems stupid, but unless you know that your transfer will not be over 64k it is the safest.
Example page register settings:

<pre>
Page register Address (Physical - '''NO VIRTUAL ADDRESSES ALLOWED BY DMA''')
0 0
1 0x10000 (64k)
2 0x20000 (128k)
. .
. .
. .
. .
etc. etc.
</pre>

Make sure that you map the PHYSICAL ADDRESS of your DMA buffer somewhere in memory if using paging.
This buffer can THEN be referred to by it's virtual memory address. Mark the memory as non pagable.
Even though it's not memory mapped I/O, it is, except that isn't. Think of it as borrowing memory from the system because the expansion card manufacturer was too stingy to supply adequate buffering memory on their card.

OK the I/O port addresses -

<pre>
DMA 1 DMA 2 Register name
0x00 0xc0 Address Register channel 0/4
0x01 0xc1 Count Register channel 0/4
0x02 0xc2 Address Register channel 1/5
0x03 0xc3 Count Register channel 1/5
0x04 0xc4 Address Register channel 2/6
0x05 0xc5 Count Register channel 2/6
0x06 0xc6 Address Register channel 3/7
0x07 0xc7 Count Register channel 3/7

PAGE REGISTERS
0x87 channel 0 0x8f channel 4 // useless due to cascading
0x83 channel 1 0x8b channel 5
0x81 channel 2 0x89 channel 6
0x82 channel 3 0x8a channel 8
</pre>

For the pedantic among you, the page and address registers are actually used (not in modern PC's). These registers hold the next DRAM memory address to be refreshed. The reason it is useless today is because the registers can only address 16 MB of memory with these registers, and the CPU needs to be involved to 'up' the page register per 64 KB, which it can't because the DMA controller has no ability to inform the CPU it needs to be updated, because IBM forgot to connect the pins of the DMA controller to interrupts because they cost-cut on building the original PC motherboard! (Please don't feed that sentence through a grammar checker.)

'' Please notice that the number of bytes transferred is one more than the count programmed for the channel. So programming count=0 will transfer one byte, while programming count=0xffff will transfer full 64k. ''

==[[Floppy Driver|Floppy Disk]] DMA Programming==

After all that ranting, the next section may come as something of a let down. Firstly - you need implement
only 3 routines to perform a DMA transfer. The example I am using is the Floppy Drive controller (probably the most common followed Sound Blaster).

So putting it all together - let's set up a DMA buffer for channel 2 (the floppy disk).

<pre>
initialise_floppy_DMA:
; set DMA channel 2 to transfer data from 0 - 64k in memory
; paging must map this PHYSICAL memory elsewhere and LOCK it from paging to disk!
; set the counter to 0x23ff, the length of a track on a 1.44Mbyte floppy (assuming 512 byte sectors)
; the length of transfer is programmed count+1 (check datasheet if you want)
out 0x0a, 0x06 ; mask DMA channel 2
out 0xd8, 0xFF ; reset the master flip-flop
out 0x04, 0 ; address to 0 (low byte)
out 0x04, 0 ; address to 0 (high byte)
out 0xd8, 0xFF ; reset the master flip-flop (again!!!)
out 0x05, 0xFF ; count to 0x23ff (low byte)
out 0x05, 0x23 ; count to 0x23ff (high byte),
out 0x81, 0 ; external page register to 0 for total address of 00 00 00
out 0x0a, 0x02 ; unmask DMA channel 2
ret
</pre>

Please note: If you wanted to be stingy about memory, you can use any of the memory above the amount of the maximum transfer allowed (0x2400 in this case)
Once you have set up your addressing and amount of transfer you do not need to touch it again, to select reading or writing you use the mode register.

<pre>
prepare_for_floppy_DMA_write:
prepare_for_floppy_DMA_write:
out 0x0a, 0x06 ; mask DMA channel 2
out 0x0a, 0x06 ; mask DMA channel 2 and 0 (assuming 0 is already masked)
out 0x0b, 0x5A ; 01011010
out 0x0b, 0x5A ; 01011010
; single transfer, address increment, autoinit, write, channel2)
; single transfer, address increment, autoinit, write, channel2)
out 0x0a, 0x02 ; unmask DMA channel 2
out 0x0a, 0x02 ; unmask DMA channel 2
ret
ret


prepare_for_floppy_DMA_read:
prepare_for_floppy_DMA_read:
out 0x0a, 0x06 ; mask DMA channel 2
out 0x0a, 0x06 ; mask DMA channel 2 and 0 (assuming 0 is already masked)
out 0x0b, 0x56 ; 01010110
out 0x0b, 0x56 ; 01010110
; single transfer, address increment, autoinit, read, channel2)
; single transfer, address increment, autoinit, read, channel2)
out 0x0a, 0x02 ; unmask DMA channel 2
out 0x0a, 0x02 ; unmask DMA channel 2
ret
ret
</syntaxhighlight>
</pre>


Some hardware, as well as VirtualPC do not support autoinit. You may want to set the Mode registers to 0x4A and 0x46 in the above routines, instead.
The above uses single transfer only for compatibility, but during the init of you floppy driver if
you detect a different floppy controller (using the init command), demand transfer should be used to
reduce overhead. In this case for a read the


The above routines use single transfer mode for compatibility, but during the initialization of your floppy driver if you detect an "advanced" floppy
out 0x0b, 0x56
controller (using the Version command), "demand transfer" should be used to reduce overhead.


== References ==


=== Articles ===
would change to
* [[Floppy Disk Controller]]

* [[Sound Blaster 16]]
out 0x0b, 0x16.


;To use -

During the initialization of your floppy driver


:Map enough physical memory to virtual memory for the size of transfer you are going to do
:(NO MORE THAN 64K, ALL TRANSFERS TO BE DONE IN 64K CHUNKS)
:(MEMORY MUST BEGIN ON A 64K BOUNDARY FROM BETWEEN 0 and 16MBytes)
:Lock the pages allocated so youd don't get page faults during reads or writes
:Call initialise_floppy_DMA altered with suitable values


;For a write
:Call prepare_for_floppy_DMA_write
:Copy 18 sectors worth of information to your DMA buffer
:Send a write command to the floppy controller for 18 sectors (say track 1)
:The floppy will now write 18 sectors of information!
:Copy 18 sectors worth of information to your DMA buffer
:Send a write command to the floppy controller for 18 sectors (say track 2)

Note that there is no need to fiddle with the DMA addressing - it automatically resets itself to 0x0 ready for another transaction.

;For a read
:Call prepare_for_floppy_DMA_read
:Send a read command to the floppy controller for 18 sectors (say track 1)
:Multiply the sector you wanted by 512 and use that as an offset into the DMA buffer.

:On the next read you can see if the track you want is in the buffer already and go straight to it.(This is called track buffering.)

The good thing about this setup is that it is simple and fast. There is no need to set up a track caching scheme and associated code (apart from checking if the current track number is in the cache).

I wouldn't advise trying to improve the speed of floppy access any further, the floppy disk is so slow compared to the rest of the computer another 1 millisecond of your time is going to be unnoticable. Perhaps the only real improvement that could be made is to read in track 0 on the first read of a floppy disk and buffer the FAT (File Allocation Table) or Inode information. This would only be useful if you were never going to write to the floppy, as the first append to a file will force the OS to update the FAT in memory, copy the altered track 0 to the DMA buffer and perform a floppy write.
Putting it simply, there's no real way to improve the write-performance of a floppy disk unless you are writing entire tracks of information at once and aiming for the fastest floppy formatting time record!

==In Conclusion==

As I've (constantly) pointed out the DMA chip has lot's of useful features. The main point in programming the chip is remembering the features that '''DON'T''' work. Regardless of the problems, the less the DMA chip is used the better because of it's negative impact on system performance (4.77Mhz, 8-bit data transfer). USB floppy disk drives are now available and standard interface floppy disk drives are only an option on some systems. The future will only make this more so.

== References ==


=== External Links ===
* [http://www.stud.fh-hannover.de/~heineman/extern/231466.pdf Intel 8237A datasheet]
* [http://www.intel-assembler.it/PORTALE/4/231466_8237A_DMA.pdf Intel 8237A datasheet]
*[http://bos.asmhackers.net/docs/dma/docs/ http://bos.asmhackers.net/docs/dma/docs/]


[[Category:Storage]]
[[Category:Storage]]
[[de:DMA]]

Latest revision as of 16:39, 25 June 2024

The main points about ISA DMA are
  • ISA DMA is not the same thing as PCI Busmastering DMA;
  • ISA DMA channels 1, 2 and 3 are available for 8 bit transfers to ISA peripherals;
  • ISA DMA channels 5, 6 and 7 are available for 16 bit transfers to ISA peripherals;
  • Transfers must not cross physical 64 KB boundaries and must never be bigger than 64 KB;
  • Transfers must be physically contiguous, and can only target the lowest 16 MB of physical memory;
  • ISA DMA is slow - theoretically 4.77 MB/second, but more like 400 KB/second due to ISA bus protocols;
  • ISA DMA frees up CPU resources, but adds an extremely heavy load to the memory bus;
  • Very few devices currently use ISA DMA -- only internal floppies, some embedded sound chips, some parallel ports, and some serial ports.

Notes:

There Is More Than One Kind of DMA on a PC

Modern PCI controllers always have their own 'Busmastering DMA', which is far better than ISA DMA. Even USB floppy drives send their DMA data using PCI Busmastering, through the PCI USB controller. PCI Busmasters can access memory with 32 bit addressing. Newer PCI cards are starting to support 64 bit addressing (although at the moment most don't). Typically PCI cards use 'scatter-gather' bus mastering, where one page is used as a directory of data pages. This almost completely overcomes the "physical memory only" limitation of all forms of DMA.


ISA DMA Background

ISA DMA (Industry Standard Architecture Direct Memory Access), like ISA itself, is an appendix for modern PCs. It is used by the internal floppy disk controller, ISA sound cards, ISA network cards, and parallel ports (if they support ECP mode). Whilst interrupt, keyboard and timer interface circuits have obvious and relevant uses, the ISA DMA controller and its programming interface are still well and truly stuck in the 1970s where they were first designed.

The idea behind DMA is that you can set up a 'channel' with an address pointing into memory and the length of the data to be transferred. Once set up, the CPU can tell the peripheral owning the channel to do whatever it is supposed to do (e.g. read a sector). Then the CPU can go do something else. When the memory bus isn't being used by the CPU, the DMA chip takes over and transfers data between the peripheral and memory without involving the CPU. When the transfer is complete (e.g. an entire sector has been sent to the floppy drive) the DMA chip then signals that it is finished. The DMA chip can even signal if it has run out of data, allowing the system to locate the next block of data to transfer on the same DMA transaction. DMA can improve the speed of a system quite a bit and was borrowed by Intel (who designed the DMA controller chip) from the old 1960s mainframes which had DMA channels for all devices (CPUs weren't all on a single chip and very slow back then).

Of course all good ideas can have downsides and while Intel can't really be blamed for what is about to be described, IBM certainly can.

In the beginning there was a PC, but the PC was slow. IBM looked down from the heavens and said "Slap on a DMA controller -- that should speed it up." IBM's heart was in the right place; its collective brains were elsewhere as the DMA controller never met the needs of the system. The PC/AT standard contains 2 Intel 8237A DMA chips, connected as Master/Slave. The second chip is Master, and its first line (Channel 4) is used by the first chip, which is Slave. (This is unlike the interrupt controller, where the first chip is Master.) The 8237A was designed for the old 8080 8-bit processor and this is probably the main reason for so many DMA problems. The 8088 and 8086 processors chosen by IBM for its PC were too advanced for the DMA controller.

Previously it was mentioned that a DMA controller is able to signal completion and even ask for more information. Unfortunately this would make expansion slots too big, so IBM left all of the connections to the DMA chips off. The only time you know when a transfer is complete is for a peripheral to signal an interrupt. This implies that all peripherals using an ISA DMA channel are limited to no more than 64 KB transfers for fear of upsetting the DMA controller.

Even with the PC/AT, IBM began bypassing the ISA DMA used in the PC/XT and used ATA PIO Mode for the hard disk. This was because of the 64 KB limitations outlined above and the fact that the 286 processor could perform 16 bit transactions at 6 MHz. Even the ISA bus could run at a speed of up to 12 MHz, far faster than the 4.77 MHz the DMA controller was running at.

Expansion card designers were also upset with DMA's lack of capabilities, noticeably 'Hard-Card' hard disk expansion card manufacturers who depended on the speed of data transfer.

To get around the limitations of the 'on board' DMA controller, expansion card manufacturers began to put their own DMA controllers on their expansion cards. They functioned exactly the same way as the 'on board' DMA, 'stealing memory bus cycles' when the processor wasn't looking and thus improving the performance of the system as a whole. These "ISA Bus Masters" are still usually limited to the lower 16 MiB of memory, but do not have the 4.77 MHz issue. This trend continued through the creation of the PCI bus, which eventually entirely replaced the ISA bus in PCs.

Technical Details

Each 8237 DMA chip has 4 DMA channels. DMA0-DMA3 on the first chip and DMA4-DMA7 on the second. The first DMA controller is wired up for 8 bit transfers, while the second is wired up for 16 bit transfers. On some tutorials or other wiki articles, you will sometimes see the second DMA chip (channels 4 to 7) labeled as the "master" controller, and the first (channels 0 to 3) called the "slave". This is highly confusing, and these terms will not be used again, here.

DMA Channel 0 is unavailable as it was used for a short time for DRAM memory refresh, and remains reserved because of this (even though modern computers don't use it). DMA channel 4 cannot be used for peripherals because it is used for cascading the other DMA controller.

The internal address registers of the DMA controller are only 16 bits. In order to extend this, IBM added one external "page register" byte per channel, allowing access to 16 MB of memory (24 bits total). If a DMA transfer crosses a 64 KB boundary, the internal address register wraps around to zero, and the external page register is not incremented. The DMA controller will happily continue the transfer with whatever data it finds at the new address.

ISA-based DMA controllers are specified to run at 4.77 MHz. No exceptions. If the "front-side" (memory) bus of a system runs at 133 MHz, it will be artificially slowed down to 4.77 MHz when transferring each ISA DMA byte/word. This includes EISA and PS/2 32-bit controllers, even though these controllers have an extra page register (which allows a 4 GiB addressing space) and the ability to do 32 bit transfers. (These DMA controllers exist only on EISA and MCA systems, which are now obsolete and are not further described here.)


Programming Details

16 bit issues

The 16 bit channels (5, 6, and 7) have a special addressing scheme to handle the way they increment addresses. The internal registers increment by 1, but the memory address needs to be incremented by 2 between each access. The solution is that the "start address" that the CPU stores in the DMA controller needs to be shifted right by 1 bit. On each memory access, this internal 16 bit address value is incremented by one, then that value is taken and shifted one bit to the left (clearing the lower bit) before being used as an address. The external "page register" addressing byte is then appended in the normal way. It is important to note that the upper bit of the internal address is lost when it is shifted left -- it is not ORed into the page register byte. This prevents DMA transfers bigger than 64 KB from working even though they should be technically possible for 16 bit channels, because the address register will effectively wrap around to zero without incrementing the page register byte.

Physical Memory vs. Paging

Paged memory mapping is exclusively controlled by the CPU. The whole point of DMA is to bypass the CPU. Therefore, no DMA can ever access any virtual memory addresses. All DMA is always done on physical memory addresses only. ISA DMA has a 16 MB physical address limit.

Since DMA runs independently of the CPU, it is important that an OS allocates a block of contiguous physical memory for the DMA transfer in a way that prevents that memory from being used for any other purpose, or swapped out, until the DMA transfer is complete.

Note: VM86 mode does not use physical addressing. The memory addresses are fake. In VM86 mode the OS must emulate any DMA transactions on behalf of an application.

Buffer Size

A typical 1.44 MB floppy disk can easily transfer 36 sectors of data in a single transfer. This is only 18 KB. The biggest internal floppy with worst-case formatting may be able to transfer 84 sectors at once. This is still only 42 KB. A Soundblaster card may run best with a 64 KB buffer. It is never necessary to try to double-buffer ISA DMA transfers, because they are so slow anyway. There are at most 6 usable DMA channels, and it is not necessary to allocate a full 64 KB to each of them. Putting all of this together, any OS should be easily able to allocate all the ISA DMA physical memory that it needs from a 256 KB pool, or even only half of that.

The Flip-Flop

Many devices on a PC (e.g. ATA disk drives) use 8 bit IO Ports to receive 16 bit values. This is done using a flip-flop. The device expects the low byte first. As soon as it receives a byte, the flip-flop changes state and then the device expects the high byte. When the high byte is received, the flip-flop changes state again, and the device expects a new low byte. Usually, each 16 bit 'register' will have its own flip-flop, but the ISA DMA controller has a problem regarding this.

On a 8237 chip, there is only one flip-flop. And there are eight of the 16 bit registers. And there can be up to three device drivers all competing for the use of that one flip-flop simultaneously.

This creates two serious problems. One is "contention issues". The other is that it is difficult to be sure what state the flip-flop is currently in. The standard solution for dealing with the flip-flop state issue is to reset the flip-flop to "low byte" state every single time you want to use it, just so you can be certain it is in the proper state before sending bytes. There are only two solutions to "contention": either use a lock, or allow only one ISA DMA driver, so that contention is impossible.

Masking DRQ

Setting up a DMA transfer always requires setting up both "ends" of the transfer. That is, whichever peripheral owns the DMA channel needs to be told to transfer a block of data via DMA. And the DMA controller needs to be told the memory address, the transfer length, perhaps a transfer "mode", and a transfer direction (read or write). So, one of these two things always needs to be done first -- and it is usually the peripheral that has a long latency time before being ready to transfer the first byte. If you set up the peripheral first, however, its first DMA request signal (often called DRQ) may arrive while you are in the middle of setting up the DMA controller.

The answer is to mask DRQ for a particular channel while you are initializing the DMA controller. There are three ways of temporarily disabling channels described below.

Transfer Length

The value that gets stored into each Count Register is always the transfer length (either bytes or words) minus 1. If you forget to subtract the one, you will get an error on your transfer.

Interrupt on Completion

As implemented in a PC, the DMA controller can not send interrupts. Hopefully, whichever peripheral "owns" each DMA channel will send the CPU an interrupt when a transfer completes. However, some peripherals may not send an interrupt if a transfer fails with an error. As always, timeouts are important.

The Registers

The master and slave DMA controllers are very similar, so (to save space) both of them have been combined into the following table. Please try not to let this confuse you. Note: for Address and Count Registers on channels 5 to 7, see 16 bit issues above.

Each 8237A has 18 registers, addressed via the I/O Port bus:

Channels 0-3 Channels 4-7
IO Port IO Port Size Read or Write Function
0x00 0xC0 Word W Start Address Register channel 0/4 (unusable)
0x01 0xC2 Word W Count Register channel 0/4 (unusable)
0x02 0xC4 Word W Start Address Register channel 1/5
0x03 0xC6 Word W Count Register channel 1/5
0x04 0xC8 Word W Start Address Register channel 2/6
0x05 0xCA Word W Count Register channel 2/6
0x06 0xCC Word W Start Address Register channel 3/7
0x07 0xCE Word W Count Register channel 3/7
0x08 0xD0 Byte R Status Register
0x08 0xD0 Byte W Command Register
0x09 0xD2 Byte W Request Register
0x0A 0xD4 Byte W Single Channel Mask Register
0x0B 0xD6 Byte W Mode Register
0x0C 0xD8 Byte W Flip-Flop Reset Register
0x0D 0xDA Byte R Intermediate Register
0x0D 0xDA Byte W Master Reset Register
0x0E 0xDC Byte W Mask Reset Register
0x0F 0xDE Byte RW MultiChannel Mask Register (reading is undocumented, but it works!)

Each Channel also has an external R/W Page Address Register that contains the upper 8 bits of the 24 bit transfer memory address:

0x87 Channel 0 Page Address Register (unusable)
0x83 Channel 1 Page Address Register
0x81 Channel 2 Page Address Register
0x82 Channel 3 Page Address Register
0x8F Channel 4 Page Address Register (unusable)
0x8B Channel 5 Page Address Register
0x89 Channel 6 Page Address Register
0x8A Channel 7 Page Address Register

Useful Registers

Single Channel Mask Registers 0x0A and 0xD4 (Write)
Bit 7 Bit 6 Bit 5 Bit 4 Bit 3 Bit 2 Bit 1 Bit 0
MASK_ON SEL 1 SEL 0


These registers are used to mask (or unmask) DRQ for a single channel only, on either the master or slave DMA chip. That is, if you do not want to figure out the mask states of all the other channels, you can mask/unmask DRQ for one channel at a time. Use the SEL 0 and 1 bits to select the channel, and the MASK_ON bit to set or clear masking for it. Note that masking DMA channel 4 will mask 7, 6, 5 and 4 due to cascading.


MultiChannel Mask Registers 0x0F and 0xDE (Read and Write)
Bit 7 Bit 6 Bit 5 Bit 4 Bit 3 Bit 2 Bit 1 Bit 0
MASK3 MASK2 MASK1 MASK0

Setting the appropriate bits to 0 or 1 allows you to unmask or mask (respectively) DRQ for those channels. Using this register means that your driver needs to know the desired mask states of all the channels at that moment. There are several ways to do this, but one is simply to read this register, first. Note that masking DMA channel 4 will mask 7, 6, 5 and 4 due to cascading.


DMA Mode Registers 0x0B and 0xD6 (Write)
Bit 7 Bit 6 Bit 5 Bit 4 Bit 3 Bit 2 Bit 1 Bit 0
MOD1 MOD0 DOWN AUTO TRA1 TRA0 SEL1 SEL0

Setting this register is a little tricky as it depends highly on the peripheral you are programming the DMA controller for. However, the driver for the peripheral is the entity that needs to set this register, and it should know what mode the peripheral needs.

  • SEL0 and SEL1 select the channel you want to change;
  • TRA0 and TRA1 selects the transfer type;
    • 0b00 runs a self test of the controller;
    • 0b01 Peripheral is writing to memory;
    • 0b10 Peripheral is reading from memory;
    • 0b11 invalid.
  • AUTO: When this bit is set, after a transfer has completed the channel resets itself to the address and count values you programmed into it. This is great for floppy transfers. Read in a track - the values set themselves up for reading again immediately. For writing you'd only need to alter the transfer mode - not the addresses. Some expansion cards do not support auto-init DMA such as Sound Blaster 1.x. These devices will crash if used with auto-init DMA. Sound Blaster 2.0 and later do support auto-init DMA.
  • DOWN: Reverses the memory order of the data, when set. Memory is accessed from high addresses down to low addresses (the address is decremented between each transfer).
  • MOD0 and MOD1: This is where some problems can arise based on the peripheral the DMA controller is attached to. The DMA controller has several modes:
    • 0b00 = Transfer on Demand;
    • 0b01 = Single DMA Transfer;
    • 0b10 = Block DMA Transfer;
    • 0b11 = Cascade Mode (use to cascade another DMA controller).

Single transfer mode is good for peripherals than cannot cache a lot of data at once. Non-82077AA Floppy controllers, Sound Blaster, and Sound Blaster Pro should use Single Transfer DMA Mode.

Block transfer mode is good for peripherals that can buffer entire blocks of information. An example of this is a hard disk controller board.

Demand transfer mode is good for peripherals that start and stop intermittently such as a tape drive. The drive can read a whole load of information for as long as it can and the suspend the transfer to move to another section of the tape. Newer floppy controllers also work well with demand transfer because they have FIFO buffers to store information being read and written (but you need to set up the FIFO properly). The peripheral controls the flow, and as the information flow is uninterrupted, performance can be gained. CPUs in these later computers generally have caches and can continue working uninterrupted during a demand DMA transfer. Older computers will slow down as their CPUs wait for the memory bus to become available.

Flip-Flop Reset Registers 0x0C and 0xD8 (Write)
Master Reset Registers 0x0D and 0xDA (Write)
Mask Reset Registers 0x0E and 0xDC (Write)

Send any value to the Reset registers to activate them. Master Reset sets Flip-Flop low, clears Status, and sets all Mask bits ON. Mask Reset sets all Mask bits OFF.

The following statement from the previous wiki article needs to be verified on real hardware, because it is likely to be wrong: "The Reset Flip-Flop command must be sent before any 16 bit transaction. The flip-flop does not reset after the DMA controller has received the second byte."


The Other Registers

Status Registers 0x08 and 0xD0 (Read)
Bit 7 Bit 6 Bit 5 Bit 4 Bit 3 Bit 2 Bit 1 Bit 0
REQ3 REQ2 REQ1 REQ0 TC3 TC2 TC1 TC0
  • REQ3-0: When set: DMA Request Pending.
  • TC3-0: When set: Transfer Complete.
  • Reading this register will clear the TC bits.

This register isn't very important in light of the fact that the 8237 can't send an IRQ to tell you that it has finished. Usually there is no need to poll this register as the peripheral (at the other end of the DMA) will send an interrupt when a transaction has completed.


Command Registers 0x08 and 0xD0 (Write)
Bit 7 Bit 6 Bit 5 Bit 4 Bit 3 Bit 2 Bit 1 Bit 0
DACKP DRQP EXTW PRIO COMP COND ADHE MMT

This register really shows how incompatible the 8237 is with the PC hardware.

  • Let's start with EXTW and COMP. These increase the speed of DMA transfer by 25% be removing one of the clock cycles. Does it work? No.
  • PRIO. When zeroed, this allows DMA priorities to be rotated allowing freedom and liberty for all peripherals that share the data bus. Does it work? No.
  • MMT and ADHE. Did you know that the IBM PC could do memory to memory transfers since 1981? That's right, hardware sprites, hardware frame buffering from one location to another. Does it work? No.
  • COND. Hooray the only bit in the control register that does something useful. Setting this bit disables the DMA controller. This is one way to set up multiple DMA channels without masking each and every channel.
Request Registers 0x09 and 0xD2 (Write)

Used for memory to memory transfers and setting up priority rotation -- absolutely useless.

"Intermediate" Registers 0x0D and 0xDA (Read)

Never implemented on PCs. Useless.

Examples

Floppy Disk DMA Initialization

You need only implement 1 to 3 tiny routines to perform a DMA transfer. This example is the Floppy Drive controller (probably the most common followed by SoundBlaster).

Note: the following code is not optimal, because there is an OUT to the same IO port twice (in two places). This causes an extra delay on the IO Port bus. Real code should separate the two "out 0x4" and "out 0x5" calls with an "out" to some other port.

initialize_floppy_DMA:
; set DMA channel 2 to transfer data from 0x1000 - 0x33ff in memory
; paging must map this _physical_ memory elsewhere and _pin_ it from paging to disk!
; set the counter to 0x23ff, the length of a track on a 1.44 MiB floppy - 1 (assuming 512 byte sectors)
; transfer length = counter + 1
    out 0x0a, 0x06      ; mask DMA channel 2 and 0 (assuming 0 is already masked)
    out 0x0c, 0xFF      ; reset the master flip-flop
    out 0x04, 0         ; address to 0 (low byte)
    out 0x04, 0x10      ; address to 0x10 (high byte)
    out 0x0c, 0xFF      ; reset the master flip-flop (again!!!)
    out 0x05, 0xFF      ; count to 0x23ff (low byte)
    out 0x05, 0x23      ; count to 0x23ff (high byte),
    out 0x81, 0         ; external page register to 0 for total address of 00 10 00
    out 0x0a, 0x02      ; unmask DMA channel 2
    ret

Once you have set up your start address and transfer length you do not need to touch it again, if you are using autoinit. Once reading or writing is selected, you don't need to change that, either. To change selecting reading or writing you use the mode register.

prepare_for_floppy_DMA_write:
    out 0x0a, 0x06      ; mask DMA channel 2 and 0 (assuming 0 is already masked)
    out 0x0b, 0x5A      ; 01011010
                        ; single transfer, address increment, autoinit, write, channel2)
    out 0x0a, 0x02      ; unmask DMA channel 2
    ret

prepare_for_floppy_DMA_read:
    out 0x0a, 0x06      ; mask DMA channel 2 and 0 (assuming 0 is already masked)
    out 0x0b, 0x56      ; 01010110
                        ; single transfer, address increment, autoinit, read, channel2)
    out 0x0a, 0x02      ; unmask DMA channel 2
    ret

Some hardware, as well as VirtualPC do not support autoinit. You may want to set the Mode registers to 0x4A and 0x46 in the above routines, instead.

The above routines use single transfer mode for compatibility, but during the initialization of your floppy driver if you detect an "advanced" floppy controller (using the Version command), "demand transfer" should be used to reduce overhead.

References

Articles

External Links