Electrical computers and digital data processing systems: input/ – Input/output data processing – Input/output command process
Reexamination Certificate
1995-09-29
2001-05-01
Robertson, David L. (Department: 2187)
Electrical computers and digital data processing systems: input/
Input/output data processing
Input/output command process
C710S024000, C711S154000
Reexamination Certificate
active
06226695
ABSTRACT:
TECHNICAL FIELD
The present invention relates to information handling systems and, more particularly, to information handling systems including efficient means for moving data between storage and auxiliary processors without disrupting activity on a main processor bus.
BACKGROUND OF THE INVENTION
Field of the Invention
Most modern computer systems today use a concept of virtual memory wherein there is more memory available to the application programs than really exists in the machine (so-called real memory). This memory is called virtual because the operating system and hardware let the application think this memory is there, but in reality may not exist in physical memory accessible by the processor(s) but is allocated out on the system hard disk. The hardware and software translate virtual addresses by the program into addresses where the memory really is, either in real physical memory or somewhere out on the hard disk. It does this on a so-called page unit basis which is typically 4K bytes.
These translations are kept in the processor hardware in a translation lookaside buffer (TLB) because they are done constantly and need to be done rapidly. When a page is accessed by a processor and it is not in real memory, a page fault interrupt occurs and the software brings in the page from disk and maps it to a real page in memory. If there was no empty real memory space to put that page in from the disk, the software first selects a page to be copied to the disk freeing up space before replacing it with the page from the disk. This is called page swapping. In order to remove a real page from memory, the software changes the hardware translation buffers (TLBs) so that the old virtual addresses no longer map to their old real page location. This is called invalidating the TLB. If that virtual address is then referenced, the software will take a page fault and then know it is not in real memory and to look for it on the hard disk. When the new page is brought in from the disk, the TLB is then changed to map the new virtual address to that real page address in memory.
Today's computer systems also consist of one or more processors, each having a cache memory which contains a copy of recently used data from real memory to speed up execution. When a processor fetches or stores data to memory, the data is loaded or saved in its cache. A similar technique is used to save data back to memory when not recently used and to update a section of the cache with data currently being accessed by the processor(s). This is usually done entirely in hardware for speed.
When a processor is accessing cached data, it causes no external bus or memory activity and, therefore, is extremely efficient.
In these types of computer systems, several alternatives currently exist for moving data between memory (or a processor cache when data may be modified in a processor cache) and an I/O device. The first alternative is to have the processor issue loads and then stores directly to the devices using PIO (programmed I/O). The processor accesses memory (or cache) using a Load instruction into one of its internal registers. The hardware translates the virtual address using the TLB and gets the data from the real memory (cache) location. As noted above, a page fault will occur if the data is not presently in real memory, and the OS software will swap the data in and then the access will occur. Once the data is in the processor register, it is then written to the I/O device using a store to the I/O location. (The reverse procedure is used if the I/O device is the source of the data and the memory is the target.)
This method, although simple in programming terms, has the drawback of consuming any processor cycles since the processor is slowed by the speed of the I/O device, as well as consuming system bus and I/O bus bandwidth since there are no burst transfers available, and the transfers are limited to the processor operand sizes (words, double words, etc.). Transferring a 4K page of data in this manner would require a thousand such operations using the typical word size operand load and stores.
Another common alternative is to use Direct Memory Access (DMA) to transfer blocks of data from memory to I/O or vice versa. This has the advantage over the first alternative of saving many CPU cycles, using more efficient burst transfers and potentially not using the system bus bandwidth, if due to the system organization, the traffic can be kept off of the main system (processor/memory bus); however, there is still a large processor overhead involving the DMA setup, as will be explained below, and in handling the terminating interrupt, which again involves the OS kernel.
The DMA setup is complicated by the fact that when an application wishes to write or read some data from I/O from one of its virtual pages, the I/O DMA devices do not typically understand these virtual addresses and, second, where is the data, in memory or on the hard disk? As noted before, the OS software may have temporarily swapped an application's data page out to disk.
To set up a DMA transfer requires the processor to get the source (or target) memory address, translated from a virtual address to a real memory address, and then get the OS software to “pin” the real page in memory while the transfer is taking place. Both of these operations involve an OS kernel call which can be expensive in processor cycles. The “pinning” operation is for the real page manager to mark the real page unavailable to be paged out to disk and not be replaced by the OS software. If this were allowed, the I/O device could transfer data to an application other than the one requesting the transfer, with disastrous results.
For data intensive transfers such as graphics screen painting or multimedia device transfers, the CPU overhead or system bus bandwidth is the limiting factor.
Auxiliary processing functions such as graphics processing have generally been performed by relatively expensive graphics adapters or three dimensional pixel processing in prior art systems.
Recent prior art software developments have allowed graphics processing to be handled by the main processor complex in an information handling system. However, the result was not totally satisfactory since graphics pipeline processing does not adapt well to a normal CPU architecture.
Some newer systems have moved graphics processing functions into the memory controller to take advantage of the speed and bandwidth of the memory interface and to offload the processor as much as is practical.
SUMMARY OF THE INVENTION
Therefore, it is an object of the present invention to execute graphics processing functions in a graphics processor located in a controller attached to the processor bus including means for managing command blocks and data block movement requests in such a way as to minimize contention with normal processor or direct memory access traffic to a storage subsystem.
Accordingly, an information handling system which efficiently processes auxiliary functions such as graphics processing includes one or more processors, a high speed processor bus connecting the one or more processors, a memory controller for controlling memory and for controlling the auxiliary function processing, a memory subsystem, and an I/O bus having one or more I/O controllers with I/O devices connected thereto, further includes means in a processor for building a queue of command blocks related to the auxiliary function processor, writing one or more command blocks to memory, which may include a processor cache, writing a start address for each command block to an address FIFO associated with the auxiliary function processor, reading a command block queue start address by the auxiliary processor, recognizing the command block read, issuing a coherent read for queued data blocks with intervention enabled, flushing a copy of the queued data block from processor cache to memory subsystem if a copy of the queued data block is in processor cache, intercepting the flushed copy of the queued data block from cache to memory and ca
Kaiser John Michael
Maule Warren Edward
Victor David Wayne
Clark George E.
International Business Machines - Corporation
Leeuwen Leslie A. Van
Robertson David L.
LandOfFree
Information handling system including non-disruptive command... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Information handling system including non-disruptive command..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Information handling system including non-disruptive command... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2497321