The present disclosure relates generally to write hazard handling in information handling systems. More particularly, the present disclosure describes a system and method useful for write hazard handling implemented using memory attributes.
Certain pipelines may allow for reads and writes to be issued out of the programming sequence. A pipeline bypass of the write hazard for a previous programming read is a common practice to resolve this issue. However, shared processors and/or direct memory accesses (DMAs) and/or peripheral spaces may have sequential programming requirements that cannot be bypassed.
A write buffer at the end of the pipeline may allow for buffering writes to avoid additional pipeline stalls in the case of memory conflicts. However, additional write buffer hazard detection may be required to ensure reads inside the write buffer are bypassed. A write is held in the write buffer until an access is free, or until there is an overflow, but this may not occur for a shared processors and/or a direct memory accesses (DMA) and/or peripherals. For L2 data cache memory accesses, an L1 data cache memory management unit (MMU) may indicate whether memory is volatile or not, so that the memory management unit (MMU) may indicate whether the write is to be buffered. Integration of out-of-order (OOO) programming pipeline bypass and write buffer functionality remains problematic.
According to various illustrative embodiments, a system and method for write hazard handling using memory attributes are described. The various illustrative embodiments described may be used with a Harvard architecture memory subsystem but are not limited to use with such subsystems. A Harvard architecture memory subsystem is a memory subsystem in which program instructions are stored in separate memory from data, and in which each type of memory is accessed via a separate bus, allowing program instructions and data to be accessed in parallel. In one aspect, the method comprises pre-computing at least one memory management unit policy for at least one write request using at least one address that is at least one clock cycle before data. The method also comprises registering the at least one pre-computed memory management unit policy. The method also comprises using the at least one pre-computed memory management unit policy to control a pipeline stall of a pipeline to ensure that non-bufferable writes are pipeline-protected, ensuring that no non-bufferable locations are bypassed from within the pipeline and all subsequent non-bufferable reads will get data from a final destination. The method also comprises bypassing a read request only after a corresponding write request is updated in at least one write pending buffer. The method also comprises decoding the at least one write request with the at least one write request aligned to data. The method also comprises registering the at least one write request in the at least one write pending buffer. The method also comprises allowing arbitration logic in an arbiter to force the pipeline stall for a region that will have a write conflict. The method also comprises stalling read requests to protect against write hazards.
In various aspects, the method further comprises all memory write requests going through the at least one write pending buffer. In various aspects, in the method, pre-computing the at least one memory management unit policy for the at least one write request using the at least one address further comprises using a first write request pipeline stage of a central processing unit to perform the memory management unit computation based on only the at least one address that is available in the first write request pipeline stage. In various aspects, the method further comprises pre-computing the at least one memory management unit policy for the at least one write request using the at least one address further comprising using a memory management unit disposed within a memory controller.
In various aspects, the method further comprises registering the at least one pre-computed memory management unit policy in a memory interface disposed within a memory controller. In this method, bypassing the read request only after the corresponding write request is updated in the at least one write pending buffer further comprises bypassing the read request only after the corresponding write request is updated in the at least one write pending buffer disposed within the memory interface. This method further comprises all memory write requests going through the at least one write pending buffer. In this method, pre-computing the at least one memory management unit policy for the at least one write request using the at least one address further comprises using a first write request pipeline stage of a central processing unit to perform the memory management unit computation based on only the at least one address that is available in the first write request pipeline stage.
In various aspects, the method further comprises decoding the at least one write request with the at least one write request aligned to data further comprising using a decoder disposed within the memory controller. In various aspects, the method further comprises allowing the arbitration logic in the arbiter to force the pipeline stall for the region that will have the write conflict further comprising allowing the arbitration logic in the arbiter disposed in a memory controller to force the pipeline stall for the region that will have the write conflict.
In another aspect, a method for write hazard handling comprises determining at least one memory management unit policy by performing at least one memory management unit lookup when at least one write address is generated more than one pipeline stage in a pipeline prior to execution and write data availability. The method also comprises preventing pipeline bypass of an issued read request. The method also comprises preventing further read requests from issuing by stalling the pipeline. The method also comprises preventing placement of the write data in a write buffer. The method also comprises disabling write buffer hazard detection.
In various aspects, the method further comprises identifying at least one non-bufferable write access to an L2 data cache. This method further comprises checking the at least one write address with the at least one memory management unit policy to determine if the at least one non-bufferable write access to the L2 data cache needs to be protected. This method further comprises performing a read stall to prevent the further read requests from propagating. This method further comprises holding the issued read request using an external bus interface until the at least one non-bufferable write access to the L2 data cache has advanced and been released at the external bus interface. This method further comprises releasing the issued read request being released at the external bus interface and removing the stalling of the pipeline, once the at least one non-bufferable write access to the L2 data cache has advanced and been released at the external bus interface.
In various aspects, the method further comprises managing at least one non-bufferable write access to an L1 memory by not placing the write data into the write buffer based on the at least one memory management unit lookup. In various aspects, the method further comprises integrating write ordering and write posting into a single non-bufferable/non-cacheable region that can be used for peripheral access in regular memory space. In various aspects, the method further comprises allowing a direct memory pipeline to access at least one volatile peripheral with a simple memory management unit addition.
In yet another aspect, a system for write hazard handling is provided, the system comprising a memory controller. The system also comprises a memory management unit disposed in the memory controller, the memory management unit arranged to pre-compute at least one memory management unit policy for at least one write request using at least one address that is at least one clock cycle before data. The system also comprises a memory interface disposed in the memory controller, the memory interface arranged to register the at least one pre-computed memory management unit policy. The system also comprises the memory controller arranged to use the at least one pre-computed memory management unit policy to control a pipeline stall of a pipeline to ensure that non-bufferable writes are pipeline-protected, ensuring that no non-bufferable locations are bypassed from within the pipeline and all subsequent non-bufferable reads will get data from a final destination. The system also comprises at least one write pending buffer disposed in the memory interface, wherein a read request is bypassed only after a corresponding write request is updated in the at least one write pending buffer. The system also comprises a decoder disposed in the memory controller, the decoder arranged to decode the at least one write request with the at least one write request aligned to data. The system also comprises the at least one write pending buffer arranged to register the at least one write request in the at least one write pending buffer. The system also comprises an arbiter disposed in the memory controller, the arbiter allowing arbitration logic in the arbiter to force the pipeline stall for a region that will have a write conflict, wherein the memory controller is also arranged to stall read requests to protect against write hazards. The system also comprises a central processing unit coupled to the memory controller.
The following figures form part of the present specification and are included to further demonstrate certain aspects of the present claimed subject matter, and should not be used to limit or define the present claimed subject matter. The present claimed subject matter may be better understood by reference to one or more of these drawings in combination with the description of embodiments presented herein. Consequently, a more complete understanding of the present embodiments and further features and advantages thereof may be acquired by referring to the following description taken in conjunction with the accompanying drawings, in which the leftmost significant digit(s) in the reference numerals denote(s) the first figure in which the respective reference numerals appear, wherein:
It is to be noted, however, that the appended drawings illustrate only typical embodiments of the present claimed subject matter and are, therefore, not to be considered limiting of the scope of the present claimed subject matter, as the present claimed subject matter may admit to other equally effective embodiments.
Certain terms are used throughout the following description and claims to refer to particular system components and configurations. As one skilled in the art having the benefit of the present disclosure will appreciate, companies may refer to a component by different names. This document does not intend to distinguish between components that differ in name but not function. In the following discussion and in the claims, the terms “including” and “comprising” are used in an open-ended fashion, and, thus, should be interpreted to mean “including, but not limited to . . . ,” and so forth. Also, the term “couple” or “couples” is intended to mean either an indirect or direct electrical connection. Thus, if a first device couples to a second device, that connection may be through a direct electrical connection or though an indirect electrical connection via other devices and/or connections. Furthermore, the term “information” is intended to refer to any data, instructions, or control sequences that may be communicated between components of a device. For example, if information is sent between two components, data, instructions, control sequences, or any combination thereof may be sent between the two components.
Illustrative embodiments of the present claimed subject matter are described in detail below. In the interest of clarity, not all features of an actual implementation are described in this specification. It will of course be appreciated that in the development of any such actual embodiment, numerous implementation-specific decisions must be made to achieve the developers' specific goals, such as compliance with system-related and business-related constraints, which will vary from one implementation to another. Moreover, it will be appreciated that such a development effort might be complex and time-consuming, but would nevertheless be a routine undertaking for those of ordinary skill in the art having the benefit of the present disclosure.
In various illustrative embodiments, as shown in
The system 100 may also comprise at least one write pending buffer 140 disposed in the memory interface 130, wherein a read request is bypassed only after a corresponding write request is updated in the one or more write pending buffers 140. In various illustrative embodiments, as shown in
In various illustrative embodiments, all memory write requests may go through the at least one write pending buffer 140, 240. In various illustrative embodiments, pre-computing the at least one memory management unit (MMU) policy for the at least one write request using the at least one address, as shown at 510, may further comprise using a first write request pipeline stage of a central processing unit 170. The first write request pipeline stage of a central processing unit 170 may be used to perform the at least one memory management unit (MMU) policy pre-computation based on only the at least one address that is available in the first write request pipeline stage. In various illustrative embodiments, pre-computing the at least one memory management unit (MMU) policy for the at least one write request using the at least one address, as shown at 510, may further comprise using a memory management unit (MMU) 120 disposed within a memory controller 110.
In various illustrative embodiments, registering the at least one pre-computed memory management unit (MMU) policy, as shown at 520, may further comprise registering the at least one pre-computed memory management unit (MMU) policy in a memory interface 130 disposed within a memory controller 110. In these illustrative embodiments, bypassing the read request only after the corresponding write request is updated in the at least one write pending buffer 140, 240, as shown at 540, may further comprise bypassing the read request only after the corresponding write request is updated in the at least one write pending buffer 140, 240 disposed within the memory interface 130. In these illustrative embodiments, all memory write requests may go through the at least one write pending buffer 140, 240. In these illustrative embodiments, pre-computing the at least one memory management unit (MMU) policy for the at least one write request using the at least one address, as shown at 510, may further comprise using a first write request pipeline stage of a central processing unit 170 to perform the at least one memory management unit (MMU) policy pre-computation based on only the at least one address that is available in the first write request pipeline stage.
In various illustrative embodiments, decoding the at least one write request with the at least one write request aligned to data, as shown at 550, may further comprise using a decoder 150 disposed within the memory controller 110. In various illustrative embodiments, allowing the arbitration logic in the arbiter 160 to force the pipeline stall for the region that will have the write conflict, as shown at 570, may further comprise allowing the arbitration logic in the arbiter 160 disposed in a memory controller 110 to force the pipeline stall for the region that will have the write conflict.
In various illustrative embodiments such as those illustrated in
In various illustrative embodiments such as those illustrated in
In accordance with the present disclosure, an apparatus, system, and method useful for write hazard handling are disclosed. In various aspects, an apparatus in accordance with the present disclosure may comprise means for write hazard handling and means for enabling the means for write hazard handling, both the means for write hazard handling and the means for enabling the means for write hazard handling covering corresponding structures and/or materials described herein and equivalents thereof.
In various other aspects, a system in accordance with the present disclosure may comprise means for write hazard handling, means for enabling the means for write hazard handling, and means for using the means for write hazard handling, all of the means for write hazard handling, the means for enabling the means for write hazard handling, and the means for using the means for write hazard handling covering corresponding structures and/or materials described herein and equivalents thereof. In yet various other aspects, a method in accordance with the present disclosure may comprise steps for write hazard handling and steps for enabling the steps for write hazard handling, both the steps for write hazard handling and the steps for enabling the steps for write hazard handling covering corresponding acts described herein and equivalents thereof.
The particular embodiments disclosed above are illustrative only, as the present claimed subject matter may be modified and practiced in different but equivalent manners apparent to those skilled in the art having the benefit of the teachings herein. Furthermore, no limitations are intended to the details of construction or design herein shown, other than as described in the claims below. It is therefore evident that the particular illustrative embodiments disclosed above may be altered or modified and all such variations are considered within the scope and spirit of the present claimed subject matter. In particular, every range of values (of the form, “from about a to about b,” or, equivalently, “from approximately a to b,” or, equivalently, “from approximately a-b”) disclosed herein is to be understood as referring to the power set (the set of all subsets) of the respective range of values, in the sense of Georg Cantor. Accordingly, the protection sought herein is as set forth in the claims below.