diff options
| author | Rob Gardner | 2017-12-05 19:40:43 -0700 |
|---|---|---|
| committer | David S. Miller | 2018-01-22 08:17:16 -0800 |
| commit | dd0273284c7474100bcd331887443f0e4b1dcce8 (patch) | |
| tree | 2fe7bb4e2925efebad2ac1550f38f5dd88bdb7eb /Documentation/sparc | |
| parent | c2b5934ff505dc71247b2c7f5927c1e9b6b13c68 (diff) | |
sparc64: Oracle DAX driver
DAX is a coprocessor which resides on the SPARC M7 (DAX1) and M8
(DAX2) processor chips, and has direct access to the CPU's L3 caches
as well as physical memory. It can perform several operations on data
streams with various input and output formats. This driver provides a
transport mechanism and has limited knowledge of the various opcodes
and data formats. A user space library provides high level services
and translates these into low level commands which are then passed
into the driver and subsequently the hypervisor and the coprocessor.
The library is the recommended way for applications to use the
coprocessor, and the driver interface is not intended for general use.
Signed-off-by: Rob Gardner <rob.gardner@oracle.com>
Signed-off-by: Jonathan Helman <jonathan.helman@oracle.com>
Signed-off-by: Sanath Kumar <sanath099@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'Documentation/sparc')
| -rw-r--r-- | Documentation/sparc/oradax/dax-hv-api.txt | 1433 | ||||
| -rw-r--r-- | Documentation/sparc/oradax/oracle-dax.txt | 429 |
2 files changed, 1862 insertions, 0 deletions
diff --git a/Documentation/sparc/oradax/dax-hv-api.txt b/Documentation/sparc/oradax/dax-hv-api.txt new file mode 100644 index 000000000000..73e8d506cf64 --- /dev/null +++ b/Documentation/sparc/oradax/dax-hv-api.txt @@ -0,0 +1,1433 @@ +Excerpt from UltraSPARC Virtual Machine Specification +Compiled from version 3.0.20+15 +Publication date 2017-09-25 08:21 +Copyright © 2008, 2015 Oracle and/or its affiliates. All rights reserved. +Extracted via "pdftotext -f 547 -l 572 -layout sun4v_20170925.pdf" +Authors: + Charles Kunzman + Sam Glidden + Mark Cianchetti + + +Chapter 36. Coprocessor services + The following APIs provide access via the Hypervisor to hardware assisted data processing functionality. + These APIs may only be provided by certain platforms, and may not be available to all virtual machines + even on supported platforms. Restrictions on the use of these APIs may be imposed in order to support + live-migration and other system management activities. + +36.1. Data Analytics Accelerator + The Data Analytics Accelerator (DAX) functionality is a collection of hardware coprocessors that provide + high speed processoring of database-centric operations. The coprocessors may support one or more of + the following data query operations: search, extraction, compression, decompression, and translation. The + functionality offered may vary by virtual machine implementation. + + The DAX is a virtual device to sun4v guests, with supported data operations indicated by the virtual device + compatibilty property. Functionality is accessed through the submission of Command Control Blocks + (CCBs) via the ccb_submit API function. The operations are processed asynchronously, with the status + of the submitted operations reported through a Completion Area linked to each CCB. Each CCB has a + separate Completion Area and, unless execution order is specifically restricted through the use of serial- + conditional flags, the execution order of submitted CCBs is arbitrary. Likewise, the time to completion + for a given CCB is never guaranteed. + + Guest software may implement a software timeout on CCB operations, and if the timeout is exceeded, the + operation may be cancelled or killed via the ccb_kill API function. It is recommended for guest software + to implement a software timeout to account for certain RAS errors which may result in lost CCBs. It is + recommended such implementation use the ccb_info API function to check the status of a CCB prior to + killing it in order to determine if the CCB is still in queue, or may have been lost due to a RAS error. + + There is no fixed limit on the number of outstanding CCBs guest software may have queued in the virtual + machine, however, internal resource limitations within the virtual machine can cause CCB submissions + to be temporarily rejected with EWOULDBLOCK. In such cases, guests should continue to attempt + submissions until they succeed; waiting for an outstanding CCB to complete is not necessary, and would + not be a guarantee that a future submission would succeed. + + The availablility of DAX coprocessor command service is indicated by the presence of the DAX virtual + device node in the guest MD (Section 8.24.17, “Database Analytics Accelerators (DAX) virtual-device + node”). + +36.1.1. DAX Compatibility Property + The query functionality may vary based on the compatibility property of the virtual device: + +36.1.1.1. "ORCL,sun4v-dax" Device Compatibility + Available CCB commands: + + • No-op/Sync + + • Extract + + • Scan Value + + • Inverted Scan Value + + • Scan Range + + + 509 + Coprocessor services + + + • Inverted Scan Range + + • Translate + + • Inverted Translate + + • Select + + See Section 36.2.1, “Query CCB Command Formats” for the corresponding CCB input and output formats. + + Only version 0 CCBs are available. + +36.1.1.2. "ORCL,sun4v-dax-fc" Device Compatibility + "ORCL,sun4v-dax-fc" is compatible with the "ORCL,sun4v-dax" interface, and includes additional CCB + bit fields and controls. + +36.1.1.3. "ORCL,sun4v-dax2" Device Compatibility + Available CCB commands: + + • No-op/Sync + + • Extract + + • Scan Value + + • Inverted Scan Value + + • Scan Range + + • Inverted Scan Range + + • Translate + + • Inverted Translate + + • Select + + See Section 36.2.1, “Query CCB Command Formats” for the corresponding CCB input and output formats. + + Version 0 and 1 CCBs are available. Only version 0 CCBs may use Huffman encoded data, whereas only + version 1 CCBs may use OZIP. + +36.1.2. DAX Virtual Device Interrupts + The DAX virtual device has multiple interrupts associated with it which may be used by the guest if + desired. The number of device interrupts available to the guest is indicated in the virtual device node of the + guest MD (Section 8.24.17, “Database Analytics Accelerators (DAX) virtual-device node”). If the device + node indicates N interrupts available, the guest may use any value from 0 to N - 1 (inclusive) in a CCB + interrupt number field. Using values outside this range will result in the CCB being rejected for an invalid + field value. + + The interrupts may be bound and managed using the standard sun4v device interrupts API (Chapter 16, + Device interrupt services). Sysino interrupts are not available for DAX devices. + +36.2. Coprocessor Control Block (CCB) + CCBs are either 64 or 128 bytes long, depending on the operation type. The exact contents of the CCB + are command specific, but all CCBs contain at least one memory buffer address. All memory locations + + + 510 + Coprocessor services + + +referenced by a CCB must be pinned in memory until the CCB either completes execution or is killed +via the ccb_kill API call. Changes in virtual address mappings occurring after CCB submission are not +guaranteed to be visible, and as such all virtual address updates need to be synchronized with CCB +execution. + +All CCBs begin with a common 32-bit header. + +Table 36.1. CCB Header Format +Bits Field Description +[31:28] CCB version. For API version 2.0: set to 1 if CCB uses OZIP encoding; set to 0 if the CCB + uses Huffman encoding; otherwise either 0 or 1. For API version 1.0: always set to 0. +[27] When API version 2.0 is negotiated, this is the Pipeline Flag [512]. It is reserved in + API version 1.0 +[26] Long CCB flag [512] +[25] Conditional synchronization flag [512] +[24] Serial synchronization flag +[23:16] CCB operation code: + 0x00 No Operation (No-op) or Sync + 0x01 Extract + 0x02 Scan Value + 0x12 Inverted Scan Value + 0x03 Scan Range + 0x13 Inverted Scan Range + 0x04 Translate + 0x14 Inverted Translate + 0x05 Select +[15:13] Reserved +[12:11] Table address type + 0b'00 No address + 0b'01 Alternate context virtual address + 0b'10 Real address + 0b'11 Primary context virtual address +[10:8] Output/Destination address type + 0b'000 No address + 0b'001 Alternate context virtual address + 0b'010 Real address + 0b'011 Primary context virtual address + 0b'100 Reserved + 0b'101 Reserved + 0b'110 Reserved + 0b'111 Reserved +[7:5] Secondary source address type + + + 511 + Coprocessor services + + +Bits Field Description + 0b'000 No address + 0b'001 Alternate context virtual address + 0b'010 Real address + 0b'011 Primary context virtual address + 0b'100 Reserved + 0b'101 Reserved + 0b'110 Reserved + 0b'111 Reserved +[4:2] Primary source address type + 0b'000 No address + 0b'001 Alternate context virtual address + 0b'010 Real address + 0b'011 Primary context virtual address + 0b'100 Reserved + 0b'101 Reserved + 0b'110 Reserved + 0b'111 Reserved +[1:0] Completion area address type + 0b'00 No address + 0b'01 Alternate context virtual address + 0b'10 Real address + 0b'11 Primary context virtual address + +The Long CCB flag indicates whether the submitted CCB is 64 or 128 bytes long; value is 0 for 64 bytes +and 1 for 128 bytes. + +The Serial and Conditional flags allow simple relative ordering between CCBs. Any CCB with the Serial +flag set will execute sequentially relative to any previous CCB that is also marked as Serial in the same +CCB submission. CCBs without the Serial flag set execute independently, even if they are between CCBs +with the Serial flag set. CCBs marked solely with the Serial flag will execute upon the completion of the +previous Serial CCB, regardless of the completion status of that CCB. The Conditional flag allows CCBs +to conditionally execute based on the successful execution of the closest CCB marked with the Serial flag. +A CCB may only be conditional on exactly one CCB, however, a CCB may be marked both Conditional +and Serial to allow execution chaining. The flags do NOT allow fan-out chaining, where multiple CCBs +execute in parallel based on the completion of another CCB. + +The Pipeline flag is an optimization that directs the output of one CCB (the "source" CCB) directly to +the input of the next CCB (the "target" CCB). The target CCB thus does not need to read the input from +memory. The Pipeline flag is advisory and may be dropped. + +Both the Pipeline and Serial bits must be set in the source CCB. The Conditional bit must be set in the +target CCB. Exactly one CCB must be made conditional on the source CCB; either 0 or 2 target CCBs +is invalid. However, Pipelines can be extended beyond two CCBs: the sequence would start with a CCB +with both the Pipeline and Serial bits set, proceed through CCBs with the Pipeline, Serial, and Conditional +bits set, and terminate at a CCB that has the Conditional bit set, but not the Pipeline bit. + + + 512 + Coprocessor services + + + The input of the target CCB must start within 64 bytes of the output of the source CCB or the pipeline flag + will be ignored. All CCBs in a pipeline must be submitted in the same call to ccb_submit. + + The various address type fields indicate how the various address values used in the CCB should be + interpreted by the virtual machine. Not all of the types specified are used by every CCB format. Types + which are not applicable to the given CCB command should be indicated as type 0 (No address). Virtual + addresses used in the CCB must have translation entries present in either the TLB or a configured TSB + for the submitting virtual processor. Virtual addresses which cannot be translated by the virtual machine + will result in the CCB submission being rejected, with the causal virtual address indicated. The CCB + may be resubmitted after inserting the translation, or the address may be translated by guest software and + resubmitted using the real address translation. + +36.2.1. Query CCB Command Formats +36.2.1.1. Supported Data Formats, Elements Sizes and Offsets + Data for query commands may be encoded in multiple possible formats. The data query commands use a + common set of values to indicate the encoding formats of the data being processed. Some encoding formats + require multiple data streams for processing, requiring the specification of both primary data formats (the + encoded data) and secondary data streams (meta-data for the encoded data). + +36.2.1.1.1. Primary Input Format + + The primary input format code is a 4-bit field when it is used. There are 10 primary input formats available. + The packed formats are not endian neutral. Code values not listed below are reserved. + + Code Format Description + 0x0 Fixed width byte packed Up to 16 bytes + 0x1 Fixed width bit packed Up to 15 bits (CCB version 0) or 23 bits (CCB version + 1); bits are read most significant bit to least significant bit + within a byte + 0x2 Variable width byte packed Data stream of lengths must be provided as a secondary + input + 0x4 Fixed width byte packed with run Up to 16 bytes; data stream of run lengths must be + length encoding provided as a secondary input + 0x5 Fixed width bit packed with run Up to 15 bits (CCB version 0) or 23 bits (CCB version + length encoding 1); bits are read most significant bit to least significant bit + within a byte; data stream of run lengths must be provided + as a secondary input + 0x8 Fixed width byte packed with Up to 16 bytes before the encoding; compressed stream + Huffman (CCB version 0) or bits are read most significant bit to least significant bit + OZIP (CCB version 1) encoding within a byte; pointer to the encoding table must be + provided + 0x9 Fixed width bit packed with Up to 15 bits (CCB version 0) or 23 bits (CCB version + Huffman (CCB version 0) or 1); compressed stream bits are read most significant bit to + OZIP (CCB version 1) encoding least significant bit within a byte; pointer to the encoding + table must be provided + 0xA Variable width byte packed with Up to 16 bytes before the encoding; compressed stream + Huffman (CCB version 0) or bits are read most significant bit to least significant bit + OZIP (CCB version 1) encoding within a byte; data stream of lengths must be provided as + a secondary input; pointer to the encoding table must be + provided + + + 513 + Coprocessor services + + + Code Format Description + 0xC Fixed width byte packed with Up to 16 bytes before the encoding; compressed stream + run length encoding, followed by bits are read most significant bit to least significant bit + Huffman (CCB version 0) or within a byte; data stream of run lengths must be provided + OZIP (CCB version 1) encoding as a secondary input; pointer to the encoding table must + be provided + 0xD Fixed width bit packed with Up to 15 bits (CCB version 0) or 23 bits(CCB version 1) + run length encoding, followed by before the encoding; compressed stream bits are read most + Huffman (CCB version 0) or significant bit to least significant bit within a byte; data + OZIP (CCB version 1) encoding stream of run lengths must be provided as a secondary + input; pointer to the encoding table must be provided + + If OZIP encoding is used, there must be no reserved bytes in the table. + +36.2.1.1.2. Primary Input Element Size + + For primary input data streams with fixed size elements, the element size must be indicated in the CCB + command. The size is encoded as the number of bits or bytes, minus one. The valid value range for this + field depends on the input format selected, as listed in the table above. + +36.2.1.1.3. Secondary Input Format + + For primary input data streams which require a secondary input stream, the secondary input stream is + always encoded in a fixed width, bit-packed format. The bits are read from most significant bit to least + significant bit within a byte. There are two encoding options for the secondary input stream data elements, + depending on whether the value of 0 is needed: + + Secondary Input Description + Format Code + 0 Element is stored as value minus 1 (0 evalutes to 1, 1 evalutes + to 2, etc) + 1 Element is stored as value + +36.2.1.1.4. Secondary Input Element Size + + Secondary input element size is encoded as a two bit field: + + Secondary Input Size Description + Code + 0x0 1 bit + 0x1 2 bits + 0x2 4 bits + 0x3 8 bits + +36.2.1.1.5. Input Element Offsets + + Bit-wise input data streams may have any alignment within the base addressed byte. The offset, specified + from most significant bit to least significant bit, is provided as a fixed 3 bit field for each input type. A + value of 0 indicates that the first input element begins at the most significant bit in the first byte, and a + value of 7 indicates it begins with the least significant bit. + + This field should be zero for any byte-wise primary input data streams. + + + 514 + Coprocessor services + + +36.2.1.1.6. Output Format + + Query commands support multiple sizes and encodings for output data streams. There are four possible + output encodings, and up to four supported element sizes per encoding. Not all output encodings are + supported for every command. The format is indicated by a 4-bit field in the CCB: + + Output Format Code Description + 0x0 Byte aligned, 1 byte elements + 0x1 Byte aligned, 2 byte elements + 0x2 Byte aligned, 4 byte elements + 0x3 Byte aligned, 8 byte elements + 0x4 16 byte aligned, 16 byte elements + 0x5 Reserved + 0x6 Reserved + 0x7 Reserved + 0x8 Packed vector of single bit elements + 0x9 Reserved + 0xA Reserved + 0xB Reserved + 0xC Reserved + 0xD 2 byte elements where each element is the index value of a bit, + from an bit vector, which was 1. + 0xE 4 byte elements where each element is the index value of a bit, + from an bit vector, which was 1. + 0xF Reserved + +36.2.1.1.7. Application Data Integrity (ADI) + + On platforms which support ADI, the ADI version number may be specified for each separate memory + access type used in the CCB command. ADI checking only occurs when reading data. When writing data, + the specified ADI version number overwrites any existing ADI value in memory. + + An ADI version value of 0 or 0xF indicates the ADI checking is disabled for that data access, even if it is + enabled in memory. By setting the appropriate flag in CCB_SUBMIT (Section 36.3.1, “ccb_submit”) it is + also an option to disable ADI checking for all inputs accessed via virtual address for all CCBs submitted + during that hypercall invocation. + + The ADI value is only guaranteed to be checked on the first 64 bytes of each data access. Mismatches on + subsequent data chunks may not be detected, so guest software should be careful to use page size checking + to protect against buffer overruns. + +36.2.1.1.8. Page size checking + + All data accesses used in CCB commands must be bounded within a single memory page. When addresses + are provided using a virtual address, the page size for checking is extracted from the TTE for that virtual + address. When using real addresses, the guest must supply the page size in the same field as the address + value. The page size must be one of the sizes supported by the underlying virtual machine. Using a value + that is not supported may result in the CCB submission being rejected or the generation of a CCB parsing + error in the completion area. + + + 515 + Coprocessor services + + +36.2.1.2. Extract command + + Converts an input vector in one format to an output vector in another format. All input format types are + supported. + + The only supported output format is a padded, byte-aligned output stream, using output codes 0x0 - 0x4. + When the specified output element size is larger than the extracted input element size, zeros are padded to + the extracted input element. First, if the decompressed input size is not a whole number of bytes, 0 bits are + padded to the most significant bit side till the next byte boundary. Next, if the output element size is larger + than the byte padded input element, bytes of value 0 are added based on the Padding Direction bit in the + CCB. If the output element size is smaller than the byte-padded input element size, the input element is + truncated by dropped from the least significant byte side until the selected output size is reached. + + The return value of the CCB completion area is invalid. The “number of elements processed” field in the + CCB completion area will be valid. + + The extract CCB is a 64-byte “short format” CCB. + + The extract CCB command format can be specified by the following packed C structure for a big-endian + machine: + + + struct extract_ccb { + uint32_t header; + uint32_t control; + uint64_t completion; + uint64_t primary_input; + uint64_t data_access_control; + uint64_t secondary_input; + uint64_t reserved; + uint64_t output; + uint64_t table; + }; + + + The exact field offsets, sizes, and composition are as follows: + + Offset Size Field Description + 0 4 CCB header (Table 36.1, “CCB Header Format”) + 4 4 Command control + Bits Field Description + [31:28] Primary Input Format (see Section 36.2.1.1.1, “Primary Input + Format”) + [27:23] Primary Input Element Size (see Section 36.2.1.1.2, “Primary + Input Element Size”) + [22:20] Primary Input Starting Offset (see Section 36.2.1.1.5, “Input + Element Offsets”) + [19] Secondary Input Format (see Section 36.2.1.1.3, “Secondary + Input Format”) + [18:16] Secondary Input Starting Offset (see Section 36.2.1.1.5, “Input + Element Offsets”) + + + 516 + Coprocessor services + + +Offset Size Field Description + Bits Field Description + [15:14] Secondary Input Element Size (see Section 36.2.1.1.4, + “Secondary Input Element Size” + [13:10] Output Format (see Section 36.2.1.1.6, “Output Format”) + [9] Padding Direction selector: A value of 1 causes padding bytes + to be added to the left side of output elements. A value of 0 + causes padding bytes to be added to the right side of output + elements. + [8:0] Reserved +8 8 Completion + Bits Field Description + [63:60] ADI version (see Section 36.2.1.1.7, “Application Data + Integrity (ADI)”) + [59] If set to 1, a virtual device interrupt will be generated using + the device interrupt number specified in the lower bits of this + completion word. If 0, the lower bits of this completion word + are ignored. + [58:6] Completion area address bits [58:6]. Address type is + determined by CCB header. + [5:0] Virtual device interrupt number for completion interrupt, if + enabled. +16 8 Primary Input + Bits Field Description + [63:60] ADI version (see Section 36.2.1.1.7, “Application Data + Integrity (ADI)”) + [59:56] If using real address, these bits should be filled in with the + page size code for the page boundary checking the guest wants + the virtual machine to use when accessing this data stream + (checking is only guaranteed to be performed when using API + version 1.1 and later). If using a virtual address, this field will + be used as as primary input address bits [59:56]. + [55:0] Primary input address bits [55:0]. Address type is determined + by CCB header. +24 8 Data Access Control + Bits Field Description + [63:62] Flow Control + Value Description + 0b'00 Disable flow control + 0b'01 Enable flow control (only valid with "ORCL,sun4v- + dax-fc" compatible virtual device variants) + 0b'10 Reserved + 0b'11 Reserved + [61:60] Reserved (API 1.0) + + + 517 + Coprocessor services + + +Offset Size Field Description + Bits Field Description + Pipeline target (API 2.0) + Value Description + 0b'00 Connect to primary input + 0b'01 Connect to secondary input + 0b'10 Reserved + 0b'11 Reserved + [59:40] Output buffer size given in units of 64 bytes, minus 1. Value of + 0 means 64 bytes, value of 1 means 128 bytes, etc. Buffer size is + only enforced if flow control is enabled in Flow Control field. + [39:32] Reserved + [31:30] Output Data Cache Allocation + Value Description + 0b'00 Do not allocate cache lines for output data stream. + 0b'01 Force cache lines for output data stream to be + allocated in the cache that is local to the submitting + virtual cpu. + 0b'10 Allocate cache lines for output data stream, but allow + existing cache lines associated with the data to remain + in their current cache instance. Any memory not + already in cache will be allocated in the cache local + to the submitting virtual cpu. + 0b'11 Reserved + [29:26] Reserved + [25:24] Primary Input Length Format + Value Description + 0b'00 Number of primary symbols + 0b'01 Number of primary bytes + 0b'10 Number of primary bits + 0b'11 Reserved + [23:0] Primary Input Length + Format Field Value + # of primary symbols Number of input elements to process, + minus 1. Command execution stops + once count is reached. + # of primary bytes Number of input bytes to process, + minus 1. Command execution stops + once count is reached. The count is + done before any decompression or + decoding. + # of primary bits Number of input bits to process, + minus 1. Command execution stops + + + + 518 + Coprocessor services + + + Offset Size Field Description + Bits Field Description + Format Field Value + once count is reached. The count is + done before any decompression or + decoding, and does not include any + bits skipped by the Primary Input + Offset field value of the command + control word. + 32 8 Secondary Input, if used by Primary Input Format. Same fields as Primary + Input. + 40 8 Reserved + 48 8 Output (same fields as Primary Input) + 56 8 Symbol Table (if used by Primary Input) + Bits Field Description + [63:60] ADI version (see Section 36.2.1.1.7, “Application Data + Integrity (ADI)”) + [59:56] If using real address, these bits should be filled in with the + page size code for the page boundary checking the guest wants + the virtual machine to use when accessing this data stream + (checking is only guaranteed to be performed when using API + version 1.1 and later). If using a virtual address, this field will + be used as as symbol table address bits [59:56]. + [55:4] Symbol table address bits [55:4]. Address type is determined + by CCB header. + [3:0] Symbol table version + Value Description + 0 Huffman encoding. Must use 64 byte aligned table + address. (Only available when using version 0 CCBs) + 1 OZIP encoding. Must use 16 byte aligned table + address. (Only available when using version 1 CCBs) + + +36.2.1.3. Scan commands + + The scan commands search a stream of input data elements for values which match the selection criteria. + All the input format types are supported. There are multiple formats for the scan commands, allowing the + scan to search for exact matches to one value, exact matches to either of two values, or any value within + a specified range. The specific type of scan is indicated by the command code in the CCB header. For the + scan range commands, the boundary conditions can be specified as greater-than-or-equal-to a value, less- + than-or-equal-to a value, or both by using two boundary values. + + There are two supported formats for the output stream: the bit vector and index array formats (codes 0x8, + 0xD, and 0xE). For the standard scan command using the bit vector output, for each input element there + exists one bit in the vector that is set if the input element matched the scan criteria, or clear if not. The + inverted scan command inverts the polarity of the bits in the output. The most significant bit of the first + byte of the output stream corresponds to the first element in the input stream. The standard index array + output format contains one array entry for each input element that matched the scan criteria. Each array + + + + 519 + Coprocessor services + + +entry is the index of an input element that matched the scan criteria. An inverted scan command produces +a similar array, but of all the input elements which did NOT match the scan criteria. + +The return value of the CCB completion area contains the number of input elements found which match +the scan criteria (or number that did not match for the inverted scans). The “number of elements processed” +field in the CCB completion area will be valid, indicating the number of input elements processed. + +These commands are 128-byte “long format” CCBs. + +The scan CCB command format can be specified by the following packed C structure for a big-endian +machine: + + + struct scan_ccb { + uint32_t header; + uint32_t control; + uint64_t completion; + uint64_t primary_input; + uint64_t data_access_control; + uint64_t secondary_input; + uint64_t match_criteria0; + uint64_t output; + uint64_t table; + uint64_t match_criteria1; + uint64_t match_criteria2; + uint64_t match_criteria3; + uint64_t reserved[5]; + }; + + +The exact field offsets, sizes, and composition are as follows: + +Offset Size Field Description +0 4 CCB header (Table 36.1, “CCB Header Format”) +4 4 Command control + Bits Field Description + [31:28] Primary Input Format (see Section 36.2.1.1.1, “Primary Input + Format”) + [27:23] Primary Input Element Size (see Section 36.2.1.1.2, “Primary + Input Element Size”) + [22:20] Primary Input Starting Offset (see Section 36.2.1.1.5, “Input + Element Offsets”) + [19] Secondary Input Format (see Section 36.2.1.1.3, “Secondary + Input Format”) + [18:16] Secondary Input Starting Offset (see Section 36.2.1.1.5, “Input + Element Offsets”) + [15:14] Secondary Input Element Size (see Section 36.2.1.1.4, + “Secondary Input Element Size” + [13:10] Output Format (see Section 36.2.1.1.6, “Output Format”) + [9:5] Operand size for first scan criteria value. In a scan value + operation, this is one of two potential extact match values. + In a scan range operation, this is the size of the upper range + + + 520 + Coprocessor services + + +Offset Size Field Description + Bits Field Description + boundary. The value of this field is the number of bytes in the + operand, minus 1. Values 0xF-0x1E are reserved. A value of + 0x1F indicates this operand is not in use for this scan operation. + [4:0] Operand size for second scan criteria value. In a scan value + operation, this is one of two potential extact match values. + In a scan range operation, this is the size of the lower range + boundary. The value of this field is the number of bytes in the + operand, minus 1. Values 0xF-0x1E are reserved. A value of + 0x1F indicates this operand is not in use for this scan operation. +8 8 Completion (same fields as Section 36.2.1.2, “Extract command”) +16 8 Primary Input (same fields as Section 36.2.1.2, “Extract command”) +24 8 Data Access Control (same fields as Section 36.2.1.2, “Extract command”) +32 8 Secondary Input, if used by Primary Input Format. Same fields as Primary + Input. +40 4 Most significant 4 bytes of first scan criteria operand. If first operand is less + than 4 bytes, the value is left-aligned to the lowest address bytes. +44 4 Most significant 4 bytes of second scan criteria operand. If second operand + is less than 4 bytes, the value is left-aligned to the lowest address bytes. +48 8 Output (same fields as Primary Input) +56 8 Symbol Table (if used by Primary Input). Same fields as Section 36.2.1.2, + “Extract command” +64 4 Next 4 most significant bytes of first scan criteria operand occuring after the + bytes specified at offset 40, if needed by the operand size. If first operand + is less than 8 bytes, the valid bytes are left-aligned to the lowest address. +68 4 Next 4 most significant bytes of second scan criteria operand occuring after + the bytes specified at offset 44, if needed by the operand size. If second + operand is less than 8 bytes, the valid bytes are left-aligned to the lowest + address. +72 4 Next 4 most significant bytes of first scan criteria operand occuring after the + bytes specified at offset 64, if needed by the operand size. If first operand + is less than 12 bytes, the valid bytes are left-aligned to the lowest address. +76 4 Next 4 most significant bytes of second scan criteria operand occuring after + the bytes specified at offset 68, if needed by the operand size. If second + operand is less than 12 bytes, the valid bytes are left-aligned to the lowest + address. +80 4 Next 4 most significant bytes of first scan criteria operand occuring after the + bytes specified at offset 72, if needed by the operand size. If first operand + is less than 16 bytes, the valid bytes are left-aligned to the lowest address. +84 4 Next 4 most significant bytes of second scan criteria operand occuring after + the bytes specified at offset 76, if needed by the operand size. If second + operand is less than 16 bytes, the valid bytes are left-aligned to the lowest + address. + + + + + 521 + Coprocessor services + + +36.2.1.4. Translate commands + + The translate commands takes an input array of indicies, and a table of single bit values indexed by those + indicies, and outputs a bit vector or index array created by reading the tables bit value at each index in + the input array. The output should therefore contain exactly one bit per index in the input data stream, + when outputing as a bit vector. When outputing as an index array, the number of elements depends on the + values read in the bit table, but will always be less than, or equal to, the number of input elements. Only + a restricted subset of the possible input format types are supported. No variable width or Huffman/OZIP + encoded input streams are allowed. The primary input data element size must be 3 bytes or less. + + The maximum table index size allowed is 15 bits, however, larger input elements may be used to provide + additional processing of the output values. If 2 or 3 byte values are used, the least significant 15 bits are + used as an index into the bit table. The most significant 9 bits (when using 3-byte input elements) or single + bit (when using 2-byte input elements) are compared against a fixed 9-bit test value provided in the CCB. + If the values match, the value from the bit table is used as the output element value. If the values do not + match, the output data element value is forced to 0. + + In the inverted translate operation, the bit value read from bit table is inverted prior to its use. The additional + additional processing based on any additional non-index bits remains unchanged, and still forces the output + element value to 0 on a mismatch. The specific type of translate command is indicated by the command + code in the CCB header. + + There are two supported formats for the output stream: the bit vector and index array formats (codes 0x8, + 0xD, and 0xE). The index array format is an array of indicies of bits which would have been set if the + output format was a bit array. + + The return value of the CCB completion area contains the number of bits set in the output bit vector, + or number of elements in the output index array. The “number of elements processed” field in the CCB + completion area will be valid, indicating the number of input elements processed. + + These commands are 64-byte “short format” CCBs. + + The translate CCB command format can be specified by the following packed C structure for a big-endian + machine: + + + struct translate_ccb { + uint32_t header; + uint32_t control; + uint64_t completion; + uint64_t primary_input; + uint64_t data_access_control; + uint64_t secondary_input; + uint64_t reserved; + uint64_t output; + uint64_t table; + }; + + + The exact field offsets, sizes, and composition are as follows: + + + Offset Size Field Description + 0 4 CCB header (Table 36.1, “CCB Header Format”) + + + 522 + Coprocessor services + + +Offset Size Field Description +4 4 Command control + Bits Field Description + [31:28] Primary Input Format (see Section 36.2.1.1.1, “Primary Input + Format”) + [27:23] Primary Input Element Size (see Section 36.2.1.1.2, “Primary + Input Element Size”) + [22:20] Primary Input Starting Offset (see Section 36.2.1.1.5, “Input + Element Offsets”) + [19] Secondary Input Format (see Section 36.2.1.1.3, “Secondary + Input Format”) + [18:16] Secondary Input Starting Offset (see Section 36.2.1.1.5, “Input + Element Offsets”) + [15:14] Secondary Input Element Size (see Section 36.2.1.1.4, + “Secondary Input Element Size” + [13:10] Output Format (see Section 36.2.1.1.6, “Output Format”) + [9] Reserved + [8:0] Test value used for comparison against the most significant bits + in the input values, when using 2 or 3 byte input elements. +8 8 Completion (same fields as Section 36.2.1.2, “Extract command” +16 8 Primary Input (same fields as Section 36.2.1.2, “Extract command” +24 8 Data Access Control (same fields as Section 36.2.1.2, “Extract command”, + except Primary Input Length Format may not use the 0x0 value) +32 8 Secondary Input, if used by Primary Input Format. Same fields as Primary + Input. +40 8 Reserved +48 8 Output (same fields as Primary Input) +56 8 Bit Table + Bits Field Description + [63:60] ADI version (see Section 36.2.1.1.7, “Application Data + Integrity (ADI)”) + [59:56] If using real address, these bits should be filled in with the + page size code for the page boundary checking the guest wants + the virtual machine to use when accessing this data stream + (checking is only guaranteed to be performed when using API + version 1.1 and later). If using a virtual address, this field will + be used as as bit table address bits [59:56] + [55:4] Bit table address bits [55:4]. Address type is determined by + CCB header. Address must be 64-byte aligned (CCB version + 0) or 16-byte aligned (CCB version 1). + [3:0] Bit table version + Value Description + 0 4KB table size + 1 8KB table size + + + + 523 + Coprocessor services + + +36.2.1.5. Select command + The select command filters the primary input data stream by using a secondary input bit vector to determine + which input elements to include in the output. For each bit set at a given index N within the bit vector, + the Nth input element is included in the output. If the bit is not set, the element is not included. Only a + restricted subset of the possible input format types are supported. No variable width or run length encoded + input streams are allowed, since the secondary input stream is used for the filtering bit vector. + + The only supported output format is a padded, byte-aligned output stream. The stream follows the same + rules and restrictions as padded output stream described in Section 36.2.1.2, “Extract command”. + + The return value of the CCB completion area contains the number of bits set in the input bit vector. The + "number of elements processed" field in the CCB completion area will be valid, indicating the number + of input elements processed. + + The select CCB is a 64-byte “short format” CCB. + + The select CCB command format can be specified by the following packed C structure for a big-endian + machine: + + + struct select_ccb { + uint32_t header; + uint32_t control; + uint64_t completion; + uint64_t primary_input; + uint64_t data_access_control; + uint64_t secondary_input; + uint64_t reserved; + uint64_t output; + uint64_t table; + }; + + + The exact field offsets, sizes, and composition are as follows: + + Offset Size Field Description + 0 4 CCB header (Table 36.1, “CCB Header Format”) + 4 4 Command control + Bits Field Description + [31:28] Primary Input Format (see Section 36.2.1.1.1, “Primary Input + Format”) + [27:23] Primary Input Element Size (see Section 36.2.1.1.2, “Primary + Input Element Size”) + [22:20] Primary Input Starting Offset (see Section 36.2.1.1.5, “Input + Element Offsets”) + [19] Secondary Input Format (see Section 36.2.1.1.3, “Secondary + Input Format”) + [18:16] Secondary Input Starting Offset (see Section 36.2.1.1.5, “Input + Element Offsets”) + [15:14] Secondary Input Element Size (see Section 36.2.1.1.4, + “Secondary Input Element Size” + + + 524 + Coprocessor services + + + Offset Size Field Description + Bits Field Description + [13:10] Output Format (see Section 36.2.1.1.6, “Output Format”) + [9] Padding Direction selector: A value of 1 causes padding bytes + to be added to the left side of output elements. A value of 0 + causes padding bytes to be added to the right side of output + elements. + [8:0] Reserved + 8 8 Completion (same fields as Section 36.2.1.2, “Extract command” + 16 8 Primary Input (same fields as Section 36.2.1.2, “Extract command” + 24 8 Data Access Control (same fields as Section 36.2.1.2, “Extract command”) + 32 8 Secondary Bit Vector Input. Same fields as Primary Input. + 40 8 Reserved + 48 8 Output (same fields as Primary Input) + 56 8 Symbol Table (if used by Primary Input). Same fields as Section 36.2.1.2, + “Extract command” + +36.2.1.6. No-op and Sync commands + The no-op (no operation) command is a CCB which has no processing effect. The CCB, when processed + by the virtual machine, simply updates the completion area with its execution status. The CCB may have + the serial-conditional flags set in order to restrict when it executes. + + The sync command is a variant of the no-op command which with restricted execution timing. A sync + command CCB will only execute when all previous commands submitted in the same request have + completed. This is stronger than the conditional flag sequencing, which is only dependent on a single + previous serial CCB. While the relative ordering is guaranteed, virtual machine implementations with + shared hardware resources may cause the sync command to wait for longer than the minimum required + time. + + The return value of the CCB completion area is invalid for these CCBs. The “number of elements + processed” field is also invalid for these CCBs. + + These commands are 64-byte “short format” CCBs. + + The no-op CCB command format can be specified by the following packed C structure for a big-endian + machine: + + + struct nop_ccb { + uint32_t header; + uint32_t control; + uint64_t completion; + uint64_t reserved[6]; + }; + + + The exact field offsets, sizes, and composition are as follows: + + Offset Size Field Description + 0 4 CCB header (Table 36.1, “CCB Header Format”) + + + 525 + Coprocessor services + + + Offset Size Field Description + 4 4 Command control + Bits Field Description + [31] If set, this CCB functions as a Sync command. If clear, this + CCB functions as a No-op command. + [30:0] Reserved + 8 8 Completion (same fields as Section 36.2.1.2, “Extract command” + 16 46 Reserved + +36.2.2. CCB Completion Area + All CCB commands use a common 128-byte Completion Area format, which can be specified by the + following packed C structure for a big-endian machine: + + + struct completion_area { + uint8_t status_flag; + uint8_t error_note; + uint8_t rsvd0[2]; + uint32_t error_values; + uint32_t output_size; + uint32_t rsvd1; + uint64_t run_time; + uint64_t run_stats; + uint32_t elements; + uint8_t rsvd2[20]; + uint64_t return_value; + uint64_t extra_return_value[8]; + }; + + + The Completion Area must be a 128-byte aligned memory location. The exact layout can be described + using byte offsets and sizes relative to the memory base: + + Offset Size Field Description + 0 1 CCB execution status + 0x0 Command not yet completed + 0x1 Command ran and succeeded + 0x2 Command ran and failed (partial results may be been + produced) + 0x3 Command ran and was killed (partial execution may + have occurred) + 0x4 Command was not run + 0x5-0xF Reserved + 1 1 Error reason code + 0x0 Reserved + 0x1 Buffer overflow + + + 526 + Coprocessor services + + +Offset Size Field Description + 0x2 CCB decoding error + 0x3 Page overflow + 0x4-0x6 Reserved + 0x7 Command was killed + 0x8 Command execution timeout + 0x9 ADI miscompare error + 0xA Data format error + 0xB-0xD Reserved + 0xE Unexpected hardware error (Do not retry) + 0xF Unexpected hardware error (Retry is ok) + 0x10-0x7F Reserved + 0x80 Partial Symbol Warning + 0x81-0xFF Reserved +2 2 Reserved +4 4 If a partial symbol warning was generated, this field contains the number + of remaining bits which were not decoded. +8 4 Number of bytes of output produced +12 4 Reserved +16 8 Runtime of command (unspecified time units) +24 8 Reserved +32 4 Number of elements processed +36 20 Reserved +56 8 Return value +64 64 Extended return value + +The CCB completion area should be treated as read-only by guest software. The CCB execution status +byte will be cleared by the Hypervisor to reflect the pending execution status when the CCB is submitted +successfully. All other fields are considered invalid upon CCB submission until the CCB execution status +byte becomes non-zero. + +CCBs which complete with status 0x2 or 0x3 may produce partial results and/or side effects due to partial +execution of the CCB command. Some valid data may be accessible depending on the fault type, however, +it is recommended that guest software treat the destination buffer as being in an unknown state. If a CCB +completes with a status byte of 0x2, the error reason code byte can be read to determine what corrective +action should be taken. + +A buffer overflow indicates that the results of the operation exceeded the size of the output buffer indicated +in the CCB. The operation can be retried by resubmitting the CCB with a larger output buffer. + +A CCB decoding error indicates that the CCB contained some invalid field values. It may be also be +triggered if the CCB output is directed at a non-existent secondary input and the pipelining hint is followed. + +A page overflow error indicates that the operation required accessing a memory location beyond the page +size associated with a given address. No data will have been read or written past the page boundary, but +partial results may have been written to the destination buffer. The CCB can be resubmitted with a larger +page size memory allocation to complete the operation. + + + 527 + Coprocessor services + + + In the case of pipelined CCBs, a page overflow error will be triggered if the output from the pipeline source + CCB ends before the input of the pipeline target CCB. Page boundaries are ignored when the pipeline + hint is followed. + + Command kill indicates that the CCB execution was halted or prevented by use of the ccb_kill API call. + + Command timeout indicates that the CCB execution began, but did not complete within a pre-determined + limit set by the virtual machine. The command may have produced some or no output. The CCB may be + resubmitted with no alterations. + + ADI miscompare indicates that the memory buffer version specified in the CCB did not match the value + in memory when accessed by the virtual machine. Guest software should not attempt to resubmit the CCB + without determining the cause of the version mismatch. + + A data format error indicates that the input data stream did not follow the specified data input formatting + selected in the CCB. + + Some CCBs which encounter hardware errors may be resubmitted without change. Persistent hardware + errors may result in multiple failures until RAS software can identify and isolate the faulty component. + + The output size field indicates the number of bytes of valid output in the destination buffer. This field is + not valid for all possible CCB commands. + + The runtime field indicates the execution time of the CCB command once it leaves the internal virtual + machine queue. The time units are fixed, but unspecified, allowing only relative timing comparisons + by guest software. The time units may also vary by hardware platform, and should not be construed to + represent any absolute time value. + + Some data query commands process data in units of elements. If applicable to the command, the number of + elements processed is indicated in the listed field. This field is not valid for all possible CCB commands. + + The return value and extended return value fields are output locations for commands which do not use + a destination output buffer, or have secondary return results. The field is not valid for all possible CCB + commands. + +36.3. Hypervisor API Functions +36.3.1. ccb_submit + trap# FAST_TRAP + function# CCB_SUBMIT + arg0 address + arg1 length + arg2 flags + arg3 reserved + ret0 status + ret1 length + ret2 status data + ret3 reserved + + Submit one or more coprocessor control blocks (CCBs) for evaluation and processing by the virtual + machine. The CCBs are passed in a linear array indicated by address. length indicates the size of + the array in bytes. + + + 528 + Coprocessor services + + +The address should be aligned to the size indicated by length, rounded up to the nearest power of +two. Virtual machines implementations may reject submissions which do not adhere to that alignment. +length must be a multiple of 64 bytes. If length is zero, the maximum supported array length will be +returned as length in ret1. In all other cases, the length value in ret1 will reflect the number of bytes +successfully consumed from the input CCB array. + + Implementation note + Virtual machines should never reject submissions based on the alignment of address if the + entire array is contained within a single memory page of the smallest page size supported by the + virtual machine. + +A guest may choose to submit addresses used in this API function, including the CCB array address, +as either a real or virtual addresses, with the type of each address indicated in flags. Virtual addresses +must be present in either the TLB or an active TSB to be processed. The translation context for virtual +addresses is determined by a combination of CCB contents and the flags argument. + +The flags argument is divided into multiple fields defined as follows: + + +Bits Field Description +[63:16] Reserved +[15] Disable ADI for VA reads (in API 2.0) + Reserved (in API 1.0) +[14] Virtual addresses within CCBs are translated in privileged context +[13:12] Alternate translation context for virtual addresses within CCBs: + 0b'00 CCBs requesting alternate context are rejected + 0b'01 Reserved + 0b'10 CCBs requesting alternate context use secondary context + 0b'11 CCBs requesting alternate context use nucleus context +[11:9] Reserved +[8] Queue info flag +[7] All-or-nothing flag +[6] If address is a virtual address, treat its translation context as privileged +[5:4] Address type of address: + 0b'00 Real address + 0b'01 Virtual address in primary context + 0b'10 Virtual address in secondary context + 0b'11 Virtual address in nucleus context +[3:2] Reserved +[1:0] CCB command type: + 0b'00 Reserved + 0b'01 Reserved + 0b'10 Query command + 0b'11 Reserved + + + + 529 + Coprocessor services + + + The CCB submission type and address type for the CCB array must be provided in the flags argument. + All other fields are optional values which change the default behavior of the CCB processing. + + When set to one, the "Disable ADI for VA reads" bit will turn off ADI checking when using a virtual + address to load data. ADI checking will still be done when loading real-addressed memory. This bit is only + available when using major version 2 of the coprocessor API group; at major version 1 it is reserved. For + more information about using ADI and DAX, see Section 36.2.1.1.7, “Application Data Integrity (ADI)”. + + By default, all virtual addresses are treated as user addresses. If the virtual address translations are + privileged, they must be marked as such in the appropriate flags field. The virtual addresses used within + the submitted CCBs must all be translated with the same privilege level. + + By default, all virtual addresses used within the submitted CCBs are translated using the primary context + active at the time of the submission. The address type field within a CCB allows each address to request + translation in an alternate address context. The address context used when the alternate address context is + requested is selected in the flags argument. + + The all-or-nothing flag specifies whether the virtual machine should allow partial submissions of the + input CCB array. When using CCBs with serial-conditional flags, it is strongly recommended to use + the all-or-nothing flag to avoid broken conditional chains. Using long CCB chains on a machine under + high coprocessor load may make this impractical, however, and require submitting without the flag. + When submitting serial-conditional CCBs without the all-or-nothing flag, guest software must manually + implement the serial-conditional behavior at any point where the chain was not submitted in a single API + call, and resubmission of the remaining CCBs should clear any conditional flag that might be set in the + first remaining CCB. Failure to do so will produce indeterminate CCB execution status and ordering. + + When the all-or-nothing flag is not specified, callers should check the value of length in ret1 to determine + how many CCBs from the array were successfully submitted. Any remaining CCBs can be resubmitted + without modifications. + + The value of length in ret1 is also valid when the API call returns an error, and callers should always + check its value to determine which CCBs in the array were already processed. This will additionally + identify which CCB encountered the processing error, and was not submitted successfully. + + If the queue info flag is used during submission, and at least one CCB was successfully submitted, the + length value in ret1 will be a multi-field value defined as follows: + Bits Field Description + [63:48] DAX unit instance identifier + [47:32] DAX queue instance identifier + [31:16] Reserved + [15:0] Number of CCB bytes successfully submitted + + The value of status data depends on the status value. See error status code descriptions for details. + The value is undefined for status values that do not specifically list a value for the status data. + + The API has a reserved input and output register which will be used in subsequent minor versions of this + API function. Guest software implementations should treat that register as voltile across the function call + in order to maintain forward compatibility. + +36.3.1.1. Errors + EOK One or more CCBs have been accepted and enqueued in the virtual machine + and no errors were been encountered during submission. Some submitted + CCBs may not have been enqueued due to internal virtual machine limitations, + and may be resubmitted without changes. + + + 530 + Coprocessor services + + +EWOULDBLOCK An internal resource conflict within the virtual machine has prevented it from + being able to complete the CCB submissions sufficiently quickly, requiring + it to abandon processing before it was complete. Some CCBs may have been + successfully enqueued prior to the block, and all remaining CCBs may be + resubmitted without changes. +EBADALIGN CCB array is not on a 64-byte boundary, or the array length is not a multiple + of 64 bytes. +ENORADDR A real address used either for the CCB array, or within one of the submitted + CCBs, is not valid for the guest. Some CCBs may have been enqueued prior + to the error being detected. +ENOMAP A virtual address used either for the CCB array, or within one of the submitted + CCBs, could not be translated by the virtual machine using either the TLB + or TSB contents. The submission may be retried after adding the required + mapping, or by converting the virtual address into a real address. Due to the + shared nature of address translation resources, there is no theoretical limit on + the number of times the translation may fail, and it is recommended all guests + implement some real address based backup. The virtual address which failed + translation is returned as status data in ret2. Some CCBs may have been + enqueued prior to the error being detected. +EINVAL The virtual machine detected an invalid CCB during submission, or invalid + input arguments, such as bad flag values. Note that not all invalid CCB values + will be detected during submission, and some may be reported as errors in the + completion area instead. Some CCBs may have been enqueued prior to the + error being detected. This error may be returned if the CCB version is invalid. +ETOOMANY The request was submitted with the all-or-nothing flag set, and the array size is + greater than the virtual machine can support in a single request. The maximum + supported size for the current virtual machine can be queried by submitting a + request with a zero length array, as described above. +ENOACCESS The guest does not have permission to submit CCBs, or an address used in a + CCBs lacks sufficient permissions to perform the required operation (no write + permission on the destination buffer address, for example). A virtual address + which fails permission checking is returned as status data in ret2. Some + CCBs may have been enqueued prior to the error being detected. +EUNAVAILABLE The requested CCB operation could not be performed at this time. The + restricted operation availability may apply only to the first unsuccessfully + submitted CCB, or may apply to a larger scope. The status should not be + interpreted as permanent, and the guest should attempt to submit CCBs in + the future which had previously been unable to be performed. The status + data provides additional information about scope of the retricted availability + as follows: + Value Description + 0 Processing for the exact CCB instance submitted was unavailable, + and it is recommended the guest emulate the operation. The + guest should continue to submit all other CCBs, and assume no + restrictions beyond this exact CCB instance. + 1 Processing is unavailable for all CCBs using the requested opcode, + and it is recommended the guest emulate the operation. The + guest should continue to submit all other CCBs that use different + opcodes, but can expect continued rejections of CCBs using the + same opcode in the near future. + + + 531 + Coprocessor services + + + Value Description + 2 Processing is unavailable for all CCBs using the requested CCB + version, and it is recommended the guest emulate the operation. + The guest should continue to submit all other CCBs that use + different CCB versions, but can expect continued rejections of + CCBs using the same CCB version in the near future. + 3 Processing is unavailable for all CCBs on the submitting vcpu, + and it is recommended the guest emulate the operation or resubmit + the CCB on a different vcpu. The guest should continue to submit + CCBs on all other vcpus but can expect continued rejections of all + CCBs on this vcpu in the near future. + 4 Processing is unavailable for all CCBs, and it is recommended + the guest emulate the operation. The guest should expect all CCB + submissions to be similarly rejected in the near future. + + +36.3.2. ccb_info + + trap# FAST_TRAP + function# CCB_INFO + arg0 address + ret0 status + ret1 CCB state + ret2 position + ret3 dax + ret4 queue + + Requests status information on a previously submitted CCB. The previously submitted CCB is identified + by the 64-byte aligned real address of the CCBs completion area. + + A CCB can be in one of 4 states: + + + State Value Description + COMPLETED 0 The CCB has been fetched and executed, and is no longer active in + the virtual machine. + ENQUEUED 1 The requested CCB is current in a queue awaiting execution. + INPROGRESS 2 The CCB has been fetched and is currently being executed. It may still + be possible to stop the execution using the ccb_kill hypercall. + NOTFOUND 3 The CCB could not be located in the virtual machine, and does not + appear to have been executed. This may occur if the CCB was lost + due to a hardware error, or the CCB may not have been successfully + submitted to the virtual machine in the first place. + + Implementation note + Some platforms may not be able to report CCBs that are currently being processed, and therefore + guest software should invoke the ccb_kill hypercall prior to assuming the request CCB will never + be executed because it was in the NOTFOUND state. + + + 532 + Coprocessor services + + + The position return value is only valid when the state is ENQUEUED. The value returned is the number + of other CCBs ahead of the requested CCB, to provide a relative estimate of when the CCB may execute. + + The dax return value is only valid when the state is ENQUEUED. The value returned is the DAX unit + instance indentifier for the DAX unit processing the queue where the requested CCB is located. The value + matches the value that would have been, or was, returned by ccb_submit using the queue info flag. + + The queue return value is only valid when the state is ENQUEUED. The value returned is the DAX + queue instance indentifier for the DAX unit processing the queue where the requested CCB is located. The + value matches the value that would have been, or was, returned by ccb_submit using the queue info flag. + +36.3.2.1. Errors + + EOK The request was proccessed and the CCB state is valid. + EBADALIGN address is not on a 64-byte aligned. + ENORADDR The real address provided for address is not valid. + EINVAL The CCB completion area contents are not valid. + EWOULDBLOCK Internal resource contraints prevented the CCB state from being queried at this + time. The guest should retry the request. + ENOACCESS The guest does not have permission to access the coprocessor virtual device + functionality. + +36.3.3. ccb_kill + + trap# FAST_TRAP + function# CCB_KILL + arg0 address + ret0 status + ret1 result + + Request to stop execution of a previously submitted CCB. The previously submitted CCB is identified by + the 64-byte aligned real address of the CCBs completion area. + + The kill attempt can produce one of several values in the result return value, reflecting the CCB state + and actions taken by the Hypervisor: + + Result Value Description + COMPLETED 0 The CCB has been fetched and executed, and is no longer active in + the virtual machine. It could not be killed and no action was taken. + DEQUEUED 1 The requested CCB was still enqueued when the kill request was + submitted, and has been removed from the queue. Since the CCB + never began execution, no memory modifications were produced by + it, and the completion area will never be updated. The same CCB may + be submitted again, if desired, with no modifications required. + KILLED 2 The CCB had been fetched and was being executed when the kill + request was submitted. The CCB execution was stopped, and the CCB + is no longer active in the virtual machine. The CCB completion area + will reflect the killed status, with the subsequent implications that + partial results may have been produced. Partial results may include full + + + 533 + Coprocessor services + + + Result Value Description + command execution if the command was stopped just prior to writing + to the completion area. + NOTFOUND 3 The CCB could not be located in the virtual machine, and does not + appear to have been executed. This may occur if the CCB was lost + due to a hardware error, or the CCB may not have been successfully + submitted to the virtual machine in the first place. CCBs in the state + are guaranteed to never execute in the future unless resubmitted. + +36.3.3.1. Interactions with Pipelined CCBs + + If the pipeline target CCB is killed but the pipeline source CCB was skipped, the completion area of the + target CCB may contain status (4,0) "Command was skipped" instead of (3,7) "Command was killed". + + If the pipeline source CCB is killed, the pipeline target CCB's completion status may read (1,0) "Success". + This does not mean the target CCB was processed; since the source CCB was killed, there was no + meaningful output on which the target CCB could operate. + +36.3.3.2. Errors + + EOK The request was proccessed and the result is valid. + EBADALIGN address is not on a 64-byte aligned. + ENORADDR The real address provided for address is not valid. + EINVAL The CCB completion area contents are not valid. + EWOULDBLOCK Internal resource contraints prevented the CCB from being killed at this time. + The guest should retry the request. + ENOACCESS The guest does not have permission to access the coprocessor virtual device + functionality. + +36.3.4. dax_info + trap# FAST_TRAP + function# DAX_INFO + ret0 status + ret1 Number of enabled DAX units + ret2 Number of disabled DAX units + + Returns the number of DAX units that are enabled for the calling guest to submit CCBs. The number of + DAX units that are disabled for the calling guest are also returned. A disabled DAX unit would have been + available for CCB submission to the calling guest had it not been offlined. + +36.3.4.1. Errors + + EOK The request was proccessed and the number of enabled/disabled DAX units + are valid. + + + + + 534 + diff --git a/Documentation/sparc/oradax/oracle-dax.txt b/Documentation/sparc/oradax/oracle-dax.txt new file mode 100644 index 000000000000..9d53ac93286f --- /dev/null +++ b/Documentation/sparc/oradax/oracle-dax.txt @@ -0,0 +1,429 @@ +Oracle Data Analytics Accelerator (DAX) +--------------------------------------- + +DAX is a coprocessor which resides on the SPARC M7 (DAX1) and M8 +(DAX2) processor chips, and has direct access to the CPU's L3 caches +as well as physical memory. It can perform several operations on data +streams with various input and output formats. A driver provides a +transport mechanism and has limited knowledge of the various opcodes +and data formats. A user space library provides high level services +and translates these into low level commands which are then passed +into the driver and subsequently the Hypervisor and the coprocessor. +The library is the recommended way for applications to use the +coprocessor, and the driver interface is not intended for general use. +This document describes the general flow of the driver, its +structures, and its programmatic interface. It also provides example +code sufficient to write user or kernel applications that use DAX +functionality. + +The user library is open source and available at: + https://oss.oracle.com/git/gitweb.cgi?p=libdax.git + +The Hypervisor interface to the coprocessor is described in detail in +the accompanying document, dax-hv-api.txt, which is a plain text +excerpt of the (Oracle internal) "UltraSPARC Virtual Machine +Specification" version 3.0.20+15, dated 2017-09-25. + + +High Level Overview +------------------- + +A coprocessor request is described by a Command Control Block +(CCB). The CCB contains an opcode and various parameters. The opcode +specifies what operation is to be done, and the parameters specify +options, flags, sizes, and addresses. The CCB (or an array of CCBs) +is passed to the Hypervisor, which handles queueing and scheduling of +requests to the available coprocessor execution units. A status code +returned indicates if the request was submitted successfully or if +there was an error. One of the addresses given in each CCB is a +pointer to a "completion area", which is a 128 byte memory block that +is written by the coprocessor to provide execution status. No +interrupt is generated upon completion; the completion area must be +polled by software to find out when a transaction has finished, but +the M7 and later processors provide a mechanism to pause the virtual +processor until the completion status has been updated by the +coprocessor. This is done using the monitored load and mwait +instructions, which are described in more detail later. The DAX +coprocessor was designed so that after a request is submitted, the +kernel is no longer involved in the processing of it. The polling is +done at the user level, which results in almost zero latency between +completion of a request and resumption of execution of the requesting +thread. + + +Addressing Memory +----------------- + +The kernel does not have access to physical memory in the Sun4v +architecture, as there is an additional level of memory virtualization +present. This intermediate level is called "real" memory, and the +kernel treats this as if it were physical. The Hypervisor handles the +translations between real memory and physical so that each logical +domain (LDOM) can have a partition of physical memory that is isolated +from that of other LDOMs. When the kernel sets up a virtual mapping, +it specifies a virtual address and the real address to which it should +be mapped. + +The DAX coprocessor can only operate on physical memory, so before a +request can be fed to the coprocessor, all the addresses in a CCB must +be converted into physical addresses. The kernel cannot do this since +it has no visibility into physical addresses. So a CCB may contain +either the virtual or real addresses of the buffers or a combination +of them. An "address type" field is available for each address that +may be given in the CCB. In all cases, the Hypervisor will translate +all the addresses to physical before dispatching to hardware. Address +translations are performed using the context of the process initiating +the request. + + +The Driver API +-------------- + +An application makes requests to the driver via the write() system +call, and gets results (if any) via read(). The completion areas are +made accessible via mmap(), and are read-only for the application. + +The request may either be an immediate command or an array of CCBs to +be submitted to the hardware. + +Each open instance of the device is exclusive to the thread that +opened it, and must be used by that thread for all subsequent +operations. The driver open function creates a new context for the +thread and initializes it for use. This context contains pointers and +values used internally by the driver to keep track of submitted +requests. The completion area buffer is also allocated, and this is +large enough to contain the completion areas for many concurrent +requests. When the device is closed, any outstanding transactions are +flushed and the context is cleaned up. + +On a DAX1 system (M7), the device will be called "oradax1", while on a +DAX2 system (M8) it will be "oradax2". If an application requires one +or the other, it should simply attempt to open the appropriate +device. Only one of the devices will exist on any given system, so the +name can be used to determine what the platform supports. + +The immediate commands are CCB_DEQUEUE, CCB_KILL, and CCB_INFO. For +all of these, success is indicated by a return value from write() +equal to the number of bytes given in the call. Otherwise -1 is +returned and errno is set. + +CCB_DEQUEUE + +Tells the driver to clean up resources associated with past +requests. Since no interrupt is generated upon the completion of a +request, the driver must be told when it may reclaim resources. No +further status information is returned, so the user should not +subsequently call read(). + +CCB_KILL + +Kills a CCB during execution. The CCB is guaranteed to not continue +executing once this call returns successfully. On success, read() must +be called to retrieve the result of the action. + +CCB_INFO + +Retrieves information about a currently executing CCB. Note that some +Hypervisors might return 'notfound' when the CCB is in 'inprogress' +state. To ensure a CCB in the 'notfound' state will never be executed, +CCB_KILL must be invoked on that CCB. Upon success, read() must be +called to retrieve the details of the action. + +Submission of an array of CCBs for execution + +A write() whose length is a multiple of the CCB size is treated as a +submit operation. The file offset is treated as the index of the +completion area to use, and may be set via lseek() or using the +pwrite() system call. If -1 is returned then errno is set to indicate +the error. Otherwise, the return value is the length of the array that +was actually accepted by the coprocessor. If the accepted length is +equal to the requested length, then the submission was completely +successful and there is no further status needed; hence, the user +should not subsequently call read(). Partial acceptance of the CCB +array is indicated by a return value less than the requested length, +and read() must be called to retrieve further status information. The +status will reflect the error caused by the first CCB that was not +accepted, and status_data will provide additional data in some cases. + +MMAP + +The mmap() function provides access to the completion area allocated +in the driver. Note that the completion area is not writeable by the +user process, and the mmap call must not specify PROT_WRITE. + + +Completion of a Request +----------------------- + +The first byte in each completion area is the command status which is +updated by the coprocessor hardware. Software may take advantage of +new M7/M8 processor capabilities to efficiently poll this status byte. +First, a "monitored load" is achieved via a Load from Alternate Space +(ldxa, lduba, etc.) with ASI 0x84 (ASI_MONITOR_PRIMARY). Second, a +"monitored wait" is achieved via the mwait instruction (a write to +%asr28). This instruction is like pause in that it suspends execution +of the virtual processor for the given number of nanoseconds, but in +addition will terminate early when one of several events occur. If the +block of data containing the monitored location is modified, then the +mwait terminates. This causes software to resume execution immediately +(without a context switch or kernel to user transition) after a +transaction completes. Thus the latency between transaction completion +and resumption of execution may be just a few nanoseconds. + + +Application Life Cycle of a DAX Submission +------------------------------------------ + + - open dax device + - call mmap() to get the completion area address + - allocate a CCB and fill in the opcode, flags, parameters, addresses, etc. + - submit CCB via write() or pwrite() + - go into a loop executing monitored load + monitored wait and + terminate when the command status indicates the request is complete + (CCB_KILL or CCB_INFO may be used any time as necessary) + - perform a CCB_DEQUEUE + - call munmap() for completion area + - close the dax device + + +Memory Constraints +------------------ + +The DAX hardware operates only on physical addresses. Therefore, it is +not aware of virtual memory mappings and the discontiguities that may +exist in the physical memory that a virtual buffer maps to. There is +no I/O TLB or any scatter/gather mechanism. All buffers, whether input +or output, must reside in a physically contiguous region of memory. + +The Hypervisor translates all addresses within a CCB to physical +before handing off the CCB to DAX. The Hypervisor determines the +virtual page size for each virtual address given, and uses this to +program a size limit for each address. This prevents the coprocessor +from reading or writing beyond the bound of the virtual page, even +though it is accessing physical memory directly. A simpler way of +saying this is that a DAX operation will never "cross" a virtual page +boundary. If an 8k virtual page is used, then the data is strictly +limited to 8k. If a user's buffer is larger than 8k, then a larger +page size must be used, or the transaction size will be truncated to +8k. + +Huge pages. A user may allocate huge pages using standard interfaces. +Memory buffers residing on huge pages may be used to achieve much +larger DAX transaction sizes, but the rules must still be followed, +and no transaction will cross a page boundary, even a huge page. A +major caveat is that Linux on Sparc presents 8Mb as one of the huge +page sizes. Sparc does not actually provide a 8Mb hardware page size, +and this size is synthesized by pasting together two 4Mb pages. The +reasons for this are historical, and it creates an issue because only +half of this 8Mb page can actually be used for any given buffer in a +DAX request, and it must be either the first half or the second half; +it cannot be a 4Mb chunk in the middle, since that crosses a +(hardware) page boundary. Note that this entire issue may be hidden by +higher level libraries. + + +CCB Structure +------------- +A CCB is an array of 8 64-bit words. Several of these words provide +command opcodes, parameters, flags, etc., and the rest are addresses +for the completion area, output buffer, and various inputs: + + struct ccb { + u64 control; + u64 completion; + u64 input0; + u64 access; + u64 input1; + u64 op_data; + u64 output; + u64 table; + }; + +See libdax/common/sys/dax1/dax1_ccb.h for a detailed description of +each of these fields, and see dax-hv-api.txt for a complete description +of the Hypervisor API available to the guest OS (ie, Linux kernel). + +The first word (control) is examined by the driver for the following: + - CCB version, which must be consistent with hardware version + - Opcode, which must be one of the documented allowable commands + - Address types, which must be set to "virtual" for all the addresses + given by the user, thereby ensuring that the application can + only access memory that it owns + + +Example Code +------------ + +The DAX is accessible to both user and kernel code. The kernel code +can make hypercalls directly while the user code must use wrappers +provided by the driver. The setup of the CCB is nearly identical for +both; the only difference is in preparation of the completion area. An +example of user code is given now, with kernel code afterwards. + +In order to program using the driver API, the file +arch/sparc/include/uapi/asm/oradax.h must be included. + +First, the proper device must be opened. For M7 it will be +/dev/oradax1 and for M8 it will be /dev/oradax2. The simplest +procedure is to attempt to open both, as only one will succeed: + + fd = open("/dev/oradax1", O_RDWR); + if (fd < 0) + fd = open("/dev/oradax2", O_RDWR); + if (fd < 0) + /* No DAX found */ + +Next, the completion area must be mapped: + + completion_area = mmap(NULL, DAX_MMAP_LEN, PROT_READ, MAP_SHARED, fd, 0); + +All input and output buffers must be fully contained in one hardware +page, since as explained above, the DAX is strictly constrained by +virtual page boundaries. In addition, the output buffer must be +64-byte aligned and its size must be a multiple of 64 bytes because +the coprocessor writes in units of cache lines. + +This example demonstrates the DAX Scan command, which takes as input a +vector and a match value, and produces a bitmap as the output. For +each input element that matches the value, the corresponding bit is +set in the output. + +In this example, the input vector consists of a series of single bits, +and the match value is 0. So each 0 bit in the input will produce a 1 +in the output, and vice versa, which produces an output bitmap which +is the input bitmap inverted. + +For details of all the parameters and bits used in this CCB, please +refer to section 36.2.1.3 of the DAX Hypervisor API document, which +describes the Scan command in detail. + + ccb->control = /* Table 36.1, CCB Header Format */ + (2L << 48) /* command = Scan Value */ + | (3L << 40) /* output address type = primary virtual */ + | (3L << 34) /* primary input address type = primary virtual */ + /* Section 36.2.1, Query CCB Command Formats */ + | (1 << 28) /* 36.2.1.1.1 primary input format = fixed width bit packed */ + | (0 << 23) /* 36.2.1.1.2 primary input element size = 0 (1 bit) */ + | (8 << 10) /* 36.2.1.1.6 output format = bit vector */ + | (0 << 5) /* 36.2.1.3 First scan criteria size = 0 (1 byte) */ + | (31 << 0); /* 36.2.1.3 Disable second scan criteria */ + + ccb->completion = 0; /* Completion area address, to be filled in by driver */ + + ccb->input0 = (unsigned long) input; /* primary input address */ + + ccb->access = /* Section 36.2.1.2, Data Access Control */ + (2 << 24) /* Primary input length format = bits */ + | (nbits - 1); /* number of bits in primary input stream, minus 1 */ + + ccb->input1 = 0; /* secondary input address, unused */ + + ccb->op_data = 0; /* scan criteria (value to be matched) */ + + ccb->output = (unsigned long) output; /* output address */ + + ccb->table = 0; /* table address, unused */ + +The CCB submission is a write() or pwrite() system call to the +driver. If the call fails, then a read() must be used to retrieve the +status: + + if (pwrite(fd, ccb, 64, 0) != 64) { + struct ccb_exec_result status; + read(fd, &status, sizeof(status)); + /* bail out */ + } + +After a successful submission of the CCB, the completion area may be +polled to determine when the DAX is finished. Detailed information on +the contents of the completion area can be found in section 36.2.2 of +the DAX HV API document. + + while (1) { + /* Monitored Load */ + __asm__ __volatile__("lduba [%1] 0x84, %0\n" + : "=r" (status) + : "r" (completion_area)); + + if (status) /* 0 indicates command in progress */ + break; + + /* MWAIT */ + __asm__ __volatile__("wr %%g0, 1000, %%asr28\n" ::); /* 1000 ns */ + } + +A completion area status of 1 indicates successful completion of the +CCB and validity of the output bitmap, which may be used immediately. +All other non-zero values indicate error conditions which are +described in section 36.2.2. + + if (completion_area[0] != 1) { /* section 36.2.2, 1 = command ran and succeeded */ + /* completion_area[0] contains the completion status */ + /* completion_area[1] contains an error code, see 36.2.2 */ + } + +After the completion area has been processed, the driver must be +notified that it can release any resources associated with the +request. This is done via the dequeue operation: + + struct dax_command cmd; + cmd.command = CCB_DEQUEUE; + if (write(fd, &cmd, sizeof(cmd)) != sizeof(cmd)) { + /* bail out */ + } + +Finally, normal program cleanup should be done, i.e., unmapping +completion area, closing the dax device, freeing memory etc. + +[Kernel example] + +The only difference in using the DAX in kernel code is the treatment +of the completion area. Unlike user applications which mmap the +completion area allocated by the driver, kernel code must allocate its +own memory to use for the completion area, and this address and its +type must be given in the CCB: + + ccb->control |= /* Table 36.1, CCB Header Format */ + (3L << 32); /* completion area address type = primary virtual */ + + ccb->completion = (unsigned long) completion_area; /* Completion area address */ + +The dax submit hypercall is made directly. The flags used in the +ccb_submit call are documented in the DAX HV API in section 36.3.1. + +#include <asm/hypervisor.h> + + hv_rv = sun4v_ccb_submit((unsigned long)ccb, 64, + HV_CCB_QUERY_CMD | + HV_CCB_ARG0_PRIVILEGED | HV_CCB_ARG0_TYPE_PRIMARY | + HV_CCB_VA_PRIVILEGED, + 0, &bytes_accepted, &status_data); + + if (hv_rv != HV_EOK) { + /* hv_rv is an error code, status_data contains */ + /* potential additional status, see 36.3.1.1 */ + } + +After the submission, the completion area polling code is identical to +that in user land: + + while (1) { + /* Monitored Load */ + __asm__ __volatile__("lduba [%1] 0x84, %0\n" + : "=r" (status) + : "r" (completion_area)); + + if (status) /* 0 indicates command in progress */ + break; + + /* MWAIT */ + __asm__ __volatile__("wr %%g0, 1000, %%asr28\n" ::); /* 1000 ns */ + } + + if (completion_area[0] != 1) { /* section 36.2.2, 1 = command ran and succeeded */ + /* completion_area[0] contains the completion status */ + /* completion_area[1] contains an error code, see 36.2.2 */ + } + +The output bitmap is ready for consumption immediately after the +completion status indicates success. |