Demand-based probe filter initialization after a low power condition

JP7592199B2Active Publication Date: 2024-11-29ADVANCED MICRO DEVICES INC
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
JP2023577540
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2021-06-24
Filing Date
2022-06-07
Publication Date
2024-11-29
Estimated Expiration
2042-06-07

Smart Images

  • Figure 0007592199000001
    Figure 0007592199000001
  • Figure 0007592199000002
    Figure 0007592199000002
  • Figure 0007592199000003
    Figure 0007592199000003
Patent Text Reader

Abstract

The data fabric routes requests among multiple requesters. The probe filter tracks a state of cached lines of memory in a probe filter coupled to the data fabric. In response to the data fabric leaving a non-operational power state while all requesters that are probe filter clients are in the non-operational power state, the power management controller delays a probe filter initialization state in which data for the cached lines is initialized after the non-operational power state.
Need to check novelty before this filing date? Find Prior Art

Description

[Background technology]

[0001] Computer systems utilize a variety of peripheral components for different input / output and communication functions. A system-on-chip (SOC) combines data processing circuitry, such as a central processing unit (CPU) core and a graphics processing unit (GPU), with peripheral controllers and memory interfaces on a single integrated circuit chip, and is well suited for portable, battery-powered operation. For example, a SOC may incorporate a display controller, an image signal processor (ISP), and other peripheral controllers on the SOC to enable the input and output of information to and from the computer system. In such large, complex SOCs, devices typically transfer data between resources, such as memory, by routing accesses through a large on-chip routing circuitry or "data fabric."

[0002] The diversity of circuits on a typical SOC presents a problem for implementing power saving modes. Many of the peripheral controllers operate in real time, are isochronous, and require a certain amount of bandwidth from the shared memory resource for a period of time. This requirement has prevented the implementation of chip-wide low power states. For example, when the CPU cores and GPU are idle, the peripheral controllers may still be active and receiving data that can be stored in buffers, but the data must be sent to memory before the buffers are full.

[0003] On the other hand, CPUs and GPUs tend to have periods of high activity separated by idle periods. If the idle periods are expected to be short, the data fabric, memory controller, and memory typically remain active and can therefore quickly service requests from the peripheral controller to transfer more data. [Brief description of the drawings]

[0004] [Figure 1] FIG. 1 is a block diagram of an Advanced Configuration and Power Interface (ACPI) model for computer system power states as known in the prior art. [Diagram 2] 1 is a block diagram of a data processing system according to some embodiments. [Diagram 3] 3 is a block diagram of the data fabric of FIG. 2 in accordance with some embodiments. [Figure 4] 5 is a flowchart illustrating a process for entering a retained low power state in accordance with an illustrative embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0005] In the following description, the use of the same reference numbers in different figures indicates similar or identical items. Unless otherwise noted, the word "coupled" and its related verb forms include both direct and indirect electrical connections by means known in the art, and unless otherwise noted, any description of a direct connection also refers to an alternative embodiment using a suitable form of indirect electrical connection.

[0006] The data processor includes a plurality of requesters, a plurality of responders, a data fabric, and a power management controller. The data fabric routes requests between the plurality of requesters and the plurality of responders. The data fabric tracks a state of cached lines of memory and includes a probe filter having an initialization state. The power management controller operates to delay the probe filter from entering the initialization state when the data fabric leaves a non-operational power state while all requesters that are probe filter clients are in a non-operational power state, and to allow the probe filter to enter the initialization state when at least one requester that is a probe filter client is in an operational power state.

[0007] The method includes routing requests between a plurality of requesters and a plurality of responders in a data fabric. A state of a cached line of memory is tracked in a probe filter coupled to the data fabric. The probe filter is placed in a non-operational power state in response to the data fabric entering a non-operational power state. A probe filter initialization state in which data for the cached line is initialized is delayed in response to the data fabric leaving its non-operational power state while all requesters that are probe filter clients are in the non-operational power state. In response to at least one requester that is a probe filter client leaving its non-operational power state, the probe filter enters its initialization state.

[0008] The data fabric circuit routes requests between a plurality of requesters and a plurality of responders. The data fabric circuit tracks a state of cached lines of memory and includes a probe filter having an initialization state. The data fabric is operable, in response to a command from a power management controller, to delay the probe filter initialization state when the data fabric leaves the non-operational power state when each of a first group of requesters acting as probe filter clients is in a non-operational power state.

[0009] 1 is a block diagram of an Advanced Configuration and Power Interface (ACPI) model for computer system power states 100, as known in the prior art. The ACPI model 100 is part of the ACPI specification developed by various computer system, software and component vendors to establish an industry standard interface for device configuration and power management of both devices and entire computer systems.

[0010] The ACPI model 100 defines a hierarchy of states for various devices within a computer system organized around a set of global system states including a G0 (S0) state 110, known as a "work" state, a G1 state 120, known as a "sleep" state, a G2 (S5) state 130, known as a "soft off" state, and a G3 state 140, known as a "mechanical off" state. In addition, the ACPI model 100 includes a state 150, known as a "legacy" state, that accommodates existing basic input / output system (BIOS) power management interfaces that correspond to basic input / output system (BIOS) operation before an ACPI-compatible operating system is loaded. The ACPI model 100 also defines a set of central processing unit (CPU) states 160 and a set of device states or "D" states 170 for devices such as modems, hard disk drives (HDDs), and compact disc read only memory (CDROMs).

[0011] The ACPI model 100 is well known and most aspects will not be described further, however those aspects relating to the power states of the central processing unit are relevant to understanding the present disclosure and will be described here.

[0012] In the G0 (S0) state 110, peripheral devices can dynamically change their power states and the user can select various performance and power characteristics of the system through a user interface so that software systems can optimize the computer for desired performance or battery life. In this state, the computer system responds in real time to external events. From the G0 (S0) state, the computer system can transition to the G1 state 120, which includes various power saving sleep states or "S states", under the control of system firmware such as the BIOS, and return to a working state in response to a wake event.

[0013] In the G0(S0) state 110, the CPU can operate in various states including a "C0" state, where the processor executes instructions, and various low power or "idle" states labeled "C1" through "Cn". In the C0 state, the CPU can operate in various performance states or "P-states", where the operating frequency can be increased or decreased to support the needs of currently running application programs using dynamic voltage and frequency scaling (DVFS). In the idle states C1 through Cn, the CPU is idle and does not execute instructions, but has various combinations of power consumption and wake-up latency, where lower power consumption is traded off for longer wake-up latency. Thus, for example, the C1 state has the highest power consumption but the shortest wake-up latency, while the Cn state has the lowest power consumption but the longest wake-up latency.

[0014] Using ACPI Model 100, each particular computer system may support different combinations of P-states and C-states appropriate to the application environment. Achieving a better tradeoff between reduced power consumption and low latency in ways not specified by ACPI Model 100 remains a desirable goal.

[0015] 2 is a block diagram of a data processing system 200, according to some embodiments. Data processing system 200 includes a data processor in the form of a SOC 210 and external memory in the form of low-power Double Data Rate synchronous dynamic random-access memory (LPDDRx SDRAM) 273 and 283. Many other components of an actual data processing system would typically be present, but are not relevant to understanding the present disclosure and are not shown in FIG. 2 for ease of illustration.

[0016] SOC 210 generally includes a system management unit (SMU) 211, a system management network unit (SMN) 212, a central processing (CPU) core complex 220, labeled "CCX," a graphics controller 230, labeled "GFX," a real-time client subsystem 240, a memory / client subsystem 250, a data fabric 260, memory channels 270 and 280, and a Peripheral Component Interface Express (PCIe) subsystem 290. As will be appreciated by those skilled in the art, SOC 210 may not have all of these elements present in all embodiments and may further have additional elements included therein.

[0017] SMU 211 is bidirectionally connected to the major components in SOC 210 via SMN 212. SMN 212 forms the control fabric for SOC 210. SMU 211 is a local controller that controls the operation of resources on SOC 210 and synchronizes communication between them. SMU 211 manages the power-up sequencing of the various processors on SOC 210 and controls multiple off-chip devices via reset, enable, and other signals. SMU 211 includes one or more clock sources (not shown), such as a phase locked loop (PLL), to provide clock signals to each of the components of SOC 210. SMU 211 also manages power for the various processors and other functional blocks and may receive measured power consumption values ​​from the CPU cores in CPU core complex 220 and graphics controller 230 to determine the appropriate P-state.

[0018] CPU core complex 220 includes a set of CPU cores, each of which is bidirectionally connected to SMU 211 via SMN 212. Each CPU core may be a single core that shares only a last-level cache with the other CPU cores, or may be combined with some, but not all, of the other cores in a cluster.

[0019] Graphics controller 230 is bidirectionally connected to SMU 211 via SMN 212. Graphics controller 230 is a high-performance graphics processing unit capable of performing graphics operations such as vertex processing, fragment processing, shading, and texture blending in a highly integrated and parallel manner. Graphics controller 230 requires periodic access to external memory to perform its operations. In the embodiment shown in FIG. 2, graphics controller 230 shares a common memory subsystem with the CPU cores in CPU core complex 220, an architecture known as a unified memory architecture. Because SOC 210 includes both a CPU and a GPU, it is also referred to as an accelerated processing unit (APU).

[0020] The real-time client subsystem 240 includes a set of real-time clients, such as representative real-time clients 242 and 243, and a memory management hub 241, labeled "MM HUB." Each real-time client is bidirectionally connected to the SMU 211 via the SMN 212, and to the memory management hub 241. A real-time client may be any type of peripheral controller that requires the periodic movement of data, such as an image signal processor (ISP), an audio coder-decoder (CODEC), a display controller that renders and rasterizes objects generated by the graphics controller 230 for display on a monitor, etc.

[0021] Memory / client subsystem 250 includes a set of memory elements or peripheral controllers, such as representative memory / client devices 252 and 253, and a system and input / output hub, labeled "SYSHUB / IOHUB," 251. Each memory / client device is bidirectionally connected to SMU 211 via SMN 212, and bidirectionally connected to system and input / output hub 251. A memory / client device is a circuit that stores or requires access to data on a non-periodic basis, such as non-volatile memory, static random-access memory (SRAM), external disk controllers such as a Serial Advanced Technology Attachment (SATA) interface controller, a universal serial bus (USB) controller, a system management hub, etc.

[0022] Data fabric 260 is an interconnect that controls the flow of traffic in SOC 210. Data fabric 260 is bidirectionally connected to SMU 211 via SMN 212, and bidirectionally connected to CPU core complex 220, graphics controller 230, memory management hub 241, and system and input / output hub 251. Data fabric 260 includes a crossbar switch for routing memory-mapped access requests and responses between any of the various devices in SOC 210. The data fabric includes a system memory map and buffers for each virtual connection defined by the basic input / output system (BIOS) to determine the destination of memory accesses based on the system configuration.

[0023] Memory channels 270 and 280 are circuits that control the transfer of data between external memories 273 and 283. Memory channel 270 is formed by memory controller 271 and physical interface circuit 272, labeled "PHY", connected to external memory 273. Memory controller 271 is bidirectionally connected to SMU 211 via SMN 212, and has an upstream port and a downstream port bidirectionally connected to data fabric 260. Physical interface circuit 272 has an upstream port bidirectionally connected to memory controller 271, and a downstream port bidirectionally connected to external memory 273. Similarly, memory channel 280 is formed by memory controller 281 and physical interface circuit 282 connected to external memory 283. Memory controller 281 is bidirectionally connected to SMU 211 via SMN 212, and has an upstream port and a downstream port bidirectionally connected to data fabric 260. The physical interface circuit 282 has an upstream port bidirectionally connected to the memory controller 281 and a downstream port bidirectionally connected to the external memory 283 .

[0024] Peripheral Component Interface Express (PCIe) subsystem 290 includes PCIe controller 291 and PCIe physical interface circuit 292. PCIe controller 291 is bidirectionally connected to SMU 211 via SMN 212, and has an upstream port bidirectionally connected to the system and input / output hub 251, and a downstream port. PCIe physical interface circuit 292 has an upstream port bidirectionally connected to PCIe controller 291, and a downstream port bidirectionally connected to a PCIe fabric, not shown in FIG. 2. The PCIe controller can form a PCIe root complex of a PCIe system for connection to a PCIe network, including PCIe switches, routers, and devices.

[0025] In operation, SOC 210 integrates on a single chip a complex combination of computing and storage devices, including CPU core complex 220 and graphics controller 230. Most of these controllers are well known and will not be described further. SOC 210 also implements various system monitoring and power saving functions, including P-state elevation and throttling, and entry and exit into C-states, under the control of SMU 211 and / or operating system software.

[0026] For example, SOC 210 controls CPU core complex 220 and graphics controller 230 to enter and exit various idle states based on idle periods during which no instructions are being executed. As is known, when CPU core complex 220 and graphics controller 230 all enter an idle state, other components of SOC 210 can also be placed in appropriate idle states to save power. In one example, CPU core complex 220 supports two idle states, including a shallow idle state (C1) and a deep idle state (C6) state. The "package C6" (PC6) state is a state in which all of the CCX cores are in the C6 state and all caches, including the level 3 cache in the CCX, are flushed. According to ACPI mode, the C1 state is the lightest C state, which may simply involve clock gating and reduced power supply voltage, allowing SOC 210 to not only remain in that state but also return to an active state with very low latency. On the other hand, the C6 state is a very deep C state, which reduces power consumption by gating the clock signal and power supply voltage, but also has a very long latency to exit. The expected residence time in the C6 state may be on the order of, say, tens of microseconds, so that net power savings are still provided despite the longer entry and exit latencies.

[0027] In some known systems, other components of SOC 210 may also be placed into various idle states to correspond to the idle states of the data processing elements. In one example, when each CPU and GPU enters the C1 state, the data fabric and memory system remain fully operational but consume significant power when they are not generally needed. However, when each CPU and GPU enters the C6 state, data fabric 260 is also placed into a corresponding idle power state involving power and clock gating. Before entering this idle power state, the data fabric places external memory into a self-refresh state and the memory controller and PHY into a similar low power state.

[0028] 3 is a block diagram of data fabric 260 of FIG. 2 according to some embodiments. Data fabric 260 includes a set of master ports 310, a set of slave ports 320, a crossbar router 330, and a power controller circuit 350.

[0029] Master ports 310 include a port for each device capable of generating memory access requests, including representative coherent master ports 311 and 312, and representative master ports 313 and 314. Each of ports 311, 312, 313, 314 is adapted to be bidirectionally connected to a corresponding requesting device, as shown in FIG. 2, and has a bidirectional downstream port. Each master port has an internal FIFO buffer for temporarily storing accesses. Coherent master ports 311 and 312 connect in the upstream direction to coherent requesters, such as CCX 220, GFX 230, or other CPU or GPU requesters, in various embodiments. Master ports 313 and 314 connect in the upstream direction through MM hub 241 and IOHUB 251, respectively, to primarily non-coherent requesters, such as display controllers or multimedia controllers.

[0030] Slave ports 320 include a port for each device capable of responding to memory access requests, including representative non-coherent slave port 321 and representative coherent slave ports 322 and 323. Each of slave ports 321, 322, 323 has a bidirectional upstream port and is adapted to be bidirectionally connected to a corresponding responding device as shown in FIG. 2. Each slave port has an internal FIFO buffer for temporarily storing accesses. Each of coherent slave ports 322 and 323 is bidirectionally connected to a respective probe filter 360 to track the state of cached lines of memory. Each coherent slave port includes a coherent slave controller for the port.

[0031] Crossbar router 330 has a bidirectional connection to the downstream bidirectional connection of master ports 310 and a bidirectional connection to the upstream bidirectional connection of slave ports 320. Crossbar router 330 is the heart of data fabric 260, routing accesses between master ports and corresponding slave ports, thereby creating virtual connections between masters and slaves based on the destination address of the requester.

[0032] Power controller circuit 350 includes a local C-state controller 351, a clock gating circuit 352, a power gating circuit 353, and a probe filter clock gate 345. Local C-state controller 351 has a first input for receiving a signal labeled "SOC STATE", a first output connected to a probe filter power gate 355, and second and third outputs. Clock gating circuit 352 is connected to a probe filter power gate 355, and a clock ... CLK ", a control input connected to a second output of local C-state controller 351, and an output for providing a selectively gated clock signal to data fabric 260. Power gating circuit 353 has an input for receiving a data fabric clock signal labeled "V DD", a control input connected to the third output of local C-state controller 351, and an output for providing a selectively gated power supply voltage to data fabric 260.

[0033] Probe filter power gate 355 has an input (not shown) that receives a power supply voltage from power gating circuit 353, and an output connected to probe filter 360. Probe filter power gate 355 selectively operates under the control of C-state controller 351 to gate the power supply to probe filter 360 to provide a low power state as will be further described with respect to FIG.

[0034] In general, data fabric 260 routes memory access requests and responses between requesters (CPU cores, GPUs, real-time client controllers, etc.) and responders (external memory through a memory controller and PHY, on-chip memory, etc.). The local C-state controller activates both clock gate 352 and power gate 353 in the C-state. In addition, it activates fabric clock signal F when entering the C-state. CLK 3. Additional steps not shown in FIG. 3 may be taken, such as disabling the phase-locked loop that generates

[0035] In this embodiment, C-state controller 351 and probe-filter power gate 355 include power state control circuitry operative to provide a probe-filter low power state and a probe-filter initialization state. When exiting the probe-filter low power state, the power state control circuitry is operative to delay the probe-filter from entering the probe-filter initialization state if the data fabric leaves the non-operational power state while all first requestors are in a non-operational power state, as described further below, and to allow the probe-filter to enter the probe-filter initialization state if at least one first requestor is in an operational power state.

[0036] 4 is a flow chart 400 illustrating a process for entering and exiting a low power state in a probe filter, according to an exemplary embodiment. The illustrated process is suitable for implementation with the data fabric 260 under the control of the local C-state controller 351, or other suitable data fabric circuitry including a probe filter clock gating capability and suitable power state controller power state control circuitry similar to that of the local C-state controller 351. In this embodiment, the process services requests from CCX clients and non-CCX clients, although in other embodiments these two groups of clients are generalized as probe filter clients and non-probe filter clients, respectively. Probe filter clients are cacheable coherent clients whose cached lines are tracked by the probe filter, such that a request by another coherent client that misses in the probe filter does not need to send a probe to the cacheable coherent probe filter client. All other clients are non-probe filter clients, some of which may be both cacheable and coherent, in which case cached lines are not tracked in the probe filter, and therefore any request by another coherent client must unconditionally send a probe to the cacheable coherent non-probe filter clients mentioned above, and therefore no filtering takes place.

[0037] The process begins at block 402, where CCX 220 and GFX 230 become idle, the CCX cache is flushed, and a low power state is entered, such as the PC6 state described above. In response, at block 404, local C-state controller 351 determines that the data fabric should enter a low power state. The low power state for the data fabric includes clock gating 352, which gates the clock signal to data fabric 260, and power gating 353, which removes power to data fabric 260. This determination can be made by a local power state controller for the data fabric, such as local C-state controller 351, or a system power state controller, such as SMU 211 (FIG. 2).

[0038] At block 406, the process determines that the data fabric should leave the low power state. Exiting the low power state may occur for a variety of reasons. For example, the low power state may be exited to service a direct memory access (DMA) request by a requester on the data fabric.

[0039] In block 408, the process determines whether any CCX clients, or in other embodiments, any probe-filter clients, have exited their low power states. If not, the process proceeds to block 410, where the probe-filter initialization state is delayed to wait for the CCX clients to exit their low power states. The determination in block 408 is made in this embodiment by the local C-state controller 351 based on information in the SOC STATE signal. Once the CCX clients have exited their low power states in block 408, the process proceeds to block 412.

[0040] At block 412, the probe filter enters an initialization state in parallel with the initialization of the CCX. (Although the CCX is described here, the same process behavior is adopted for embodiments having a requester that is a single CPU core rather than a core complex design.) During the probe filter initialization state, the probe filter's data, including one or more arrays of memory location status data, is initialized to change the array contents from the random or unpredictable data present after the memory is powered on to an initial state. Typically, the initial state reflects an "invalid" indication for each tracked line in the probe filter, reflecting that cache state tracking data for the associated memory line in the probe filter has not yet been received. The desired data is written into each array entry to track the respective memory line in the probe filter.

[0041] As shown in block 412, probe-filter initialization is performed in parallel with certain steps in the initialization process of the CCX. For example, to initialize from a low power state, the CCX performs certain steps such as repair, fuse propagation, and microcode initialization to restore the state from which the CCX state was exited. These CCX initialization steps, like the probe-filter initialization, have design-specific latencies. Thus, in some embodiments, the probe-filter initialization is completed before the CCX initialization, while in other embodiments, the CCX initialization is completed first. Thus, in some embodiments, it is possible for the CCX to start generating coherent and cacheable requests to the coherent slave controller before the probe filter is initialized to service those requests. Additionally, CCX requests made during CCX initialization, such as state restore requests, may include cacheable but non-coherent requests that can be fulfilled (executed) without accessing the probe filter. Other requesters may also make requests to the coherent slave controller during the probe-filter initialization state. Thus, the coherent slave controller includes the ability to service such requests in a suitable manner during the probe-filter initialization state, as shown in blocks 414-421.

[0042] Blocks 414-421 show the response of the coherent slave controller to various types of requests when the requests occur during the probe filter initialization state.

[0043] If a cacheable coherent request is received from a CCX client or a probe-filter client at block 414, the process at block 415 blocks such a request until the probe-filter initialization is complete. This function of the coherent slave controller addresses the scenario where the CCX initialization is completed before the probe-filter initialization is completed.

[0044] In block 416, if a request is made that is dependent on a previous request that was blocked in block 415, the process in block 417 also blocks this request until the dependent request is completed. Such requests include requests made by a CCX requestor, a GFX requestor, or a real-time client requestor to the same address as the blocked request. This function allows the coherent slave controller to service such requests and maintain coherency.

[0045] If a coherent request is received from a non-CCX client requester, such as a GFX requester or a real-time client requester, in block 418, the request is fulfilled (executed) and treated as a cache miss by the coherent slave controller receiving the request in block 419. Note that address collisions are handled according to block 416, regardless of whether they are coherent requests or not. This functionality allows the coherent slave controller to fulfill (execute) certain requests without waiting for probe filter initialization to complete.

[0046] If a non-cacheable coherent request a is received from a CCX client in block 420, the request is fulfilled (executed) in block 421 and treated as a cache miss by the coherent slave controller receiving the request. Again, such requests are blocked in 417 if they depend on a previous blocked request.

[0047] If a noncoherent request is received from any requestor in block 422, the request is fulfilled (executed) as would normally be done by a coherent slave controller receiving the request in block 423. Again, such requests are blocked in 417 if they depend on a previous blocked request.

[0048] Non-cacheable or non-coherent requests generally do not require a probe filter and are normally processed by the coherent slave controller during the probe filter initialization state unless a dependency is found for the request in block 416 .

[0049] In block 424, the probe filter initialization is complete when all data arrays in the probe filter have been initialized with valid data, exiting the probe filter initialization state. The coherent slave controller then begins generating processing probe filter lookups in block 426.

[0050] Any portion of, such as SOC 210 or data fabric 260 of FIG. 2, may be described or represented by a computer-accessible data structure in the form of a database or other data structure that can be read by a program and used directly or indirectly to manufacture an integrated circuit. For example, the data structure may be a behavioral or register-transfer level (RTL) description of the hardware functionality in a high-level design language (HDL) such as Verilog or VHDL. The description may be read by a synthesis tool capable of synthesizing the description to generate a netlist that includes a list of gates from a synthesis library. The netlist includes a set of gates that also represent the functionality of the hardware that comprises the integrated circuit. The netlist may then be placed and routed to generate a data set that describes the geometric shapes to be applied to a mask. The mask may then be used in various semiconductor manufacturing processes to manufacture the integrated circuit. Alternatively, the database on the computer-accessible storage medium may be a netlist (with or without a synthesis library) or a data set, or Graphic Data System (GDS) II data, as desired.

[0051] Although specific embodiments have been described, various modifications to these embodiments will be apparent to those skilled in the art. The various techniques used in conjunction with C-states for the data fabric disclosed herein may be used independently or in conjunction with other techniques. Additionally, different techniques and circuits for detecting idle traffic states through the data fabric may be used.

[0052] Therefore, the appended claims are intended to cover all modifications of the disclosed embodiments that fall within the scope of the disclosed embodiments.

Claims

1. 1. A data processor comprising: Multiple requesters; Multiple responders, a data fabric for routing requests between the plurality of requesters and the plurality of responders; a power management controller; The data fabric comprises: a probe filter for tracking a state of a cached line of memory, the probe filter having an initialization state; The power management controller includes: and operable to delay the probe filter from entering the initialization state when the data fabric leaves a non-operational power state to execute a request from a requester that is not a probe filter client while all requesters that are probe filter clients are in a non-operational power state, and to allow the probe filter to enter the initialization state when at least one requester that is a probe filter client is in an operational power state. Data processor.

2. The data fabric comprises: A crossbar router; one or more coherent slave controllers coupled to the crossbar router and operable, during the initialization state, to block cacheable coherent requests from requesters that are probe filter clients and to fulfill requests from requesters that are not probe filter clients.

2. The data processor of claim 1.

3. the coherent slave controller is operable during the initialization state to receive blocked cacheable coherent requests that depend on the blocked cacheable coherent requests; 3. The data processor of claim 2.

4. the coherent slave controller is operable during the initialization state to classify and fulfill coherent requests from requesters that are not probe filter clients as cache misses.

3. The data processor of claim 2.

5. the power management controller causes the probe filter to enter its initialization state in response to at least one of the requesters that is a probe filter client leaving its non-operational power state and entering an initialization state.

2. The data processor of claim 1.

6. The requester, which is a probe filter client, comprises a central processing unit (CPU) core.

2. The data processor of claim 1.

7. a requester that is not a probe filter client, the requester including one or more of a graphics processing unit (GPU) core and an input / output (I / O) interface controller; 7. The data processor of claim 6.

8. 1. A method comprising: Routing requests between a plurality of requesters and a plurality of responders in a data fabric; tracking a state of a cached line of memory in a probe filter coupled to the data fabric; placing the probe filter in a non-operational power state in response to the data fabric entering a non-operational power state; responsive to the data fabric leaving a non-operational power state while all requesters that are probe-filter clients are in the non-operational power state, executing requests from requesters that are not probe-filter clients and delaying a probe-filter initialization state in which data for cached lines is initialized after the non-operational power state; and causing the probe filter to enter its initialization state in response to at least one of the requestors, which is a probe filter client, leaving its non-operational power state. method.

9. blocking cacheable coherent requests from the requesters that are probe filter clients on the data fabric during the probe filter initialization state.

9. The method of claim 8.

10. blocking requests that depend on blocked cacheable coherent requests on the data fabric during the probe filter initialization state.

9. The method of claim 8.

11. during the initialization state, fulfilling as a cache miss any coherent request from a requester on the data fabric that is not a probe filter client.

9. The method of claim 8.

12. the requester, which is not a probe filter client, comprises one or more of a graphics processing unit (GPU) core and an input / output (I / O) interface controller; 12. The method of claim 11.

13. during the probe-filter initialization state, executing a non-cacheable coherent request from the requester that is a probe-filter client as a cache miss.

9. The method of claim 8.

14. The requester, which is a probe filter client, comprises a central processing unit (CPU) core.

9. The method of claim 8.

Citation Information

Patent Citations

  • Coherent Interconnect Power Reduction Using Hardware-Controlled Partitioned Snoop Directories

    JP2019534501A

  • System and method for power management

    US20140195830A1

  • Dynamic memory power management

    US20190265778A1

  • Information processing apparatus and scheduling method

    WO2013027261A1