From sacadmin Fri Sep 5 20:26:33 2008 Received: from sunmail2sca.sfbay.sun.com (sunmail2sca.SFBay.Sun.COM [129.145.155.234]) by sac.sfbay.sun.com (8.13.8+Sun/8.13.8) with ESMTP id m863QXI6006170 for ; Fri, 5 Sep 2008 20:26:33 -0700 (PDT) Received: from brm-avmta-1.central.sun.com (brm-avmta-1.Central.Sun.COM [129.147.4.11]) by sunmail2sca.sfbay.sun.com (8.13.7+Sun/8.13.7/ENSMAIL,v2.2) with ESMTP id m863QWBE022464 for <@sunmail2sca.sfbay.sun.com:fwarc@sun.com>; Fri, 5 Sep 2008 20:26:33 -0700 (PDT) Received: from pmxchannel-daemon.brm-avmta-1.central.sun.com by brm-avmta-1.central.sun.com (Sun Java System Messaging Server 6.2-3.04 (built Jul 15 2005)) id <0K6R00F05888P300@brm-avmta-1.central.sun.com> for fwarc@sun.com (ORCPT fwarc@sun.com); Fri, 05 Sep 2008 21:26:32 -0600 (MDT) Received: from sca-es-mail-1.sun.com ([192.18.43.132]) by brm-avmta-1.central.sun.com (Sun Java System Messaging Server 6.2-3.04 (built Jul 15 2005)) with ESMTP id <0K6R0098L8878K40@brm-avmta-1.central.sun.com> for fwarc@sun.com (ORCPT fwarc@sun.com); Fri, 05 Sep 2008 21:26:31 -0600 (MDT) Received: from fe-sfbay-10.sun.com ([192.18.43.129]) by sca-es-mail-1.sun.com (8.13.7+Sun/8.12.9) with ESMTP id m863QVPj010860 for ; Fri, 05 Sep 2008 20:26:31 -0700 (PDT) Received: from conversion-daemon.fe-sfbay-10.sun.com by fe-sfbay-10.sun.com (Sun Java System Messaging Server 6.2-8.04 (built Feb 28 2007)) id <0K6R0050186A8300@fe-sfbay-10.sun.com> (original mail from Hitendra.Zhangada@Sun.COM) for fwarc@sun.com (ORCPT fwarc@sun.com); Fri, 05 Sep 2008 20:26:31 -0700 (PDT) Received: from [129.150.36.231] by fe-sfbay-10.sun.com (Sun Java System Messaging Server 6.2-8.04 (built Feb 28 2007)) with ESMTPSA id <0K6R000WA885ZU10@fe-sfbay-10.sun.com> for fwarc@sun.com (ORCPT fwarc@sun.com); Fri, 05 Sep 2008 20:26:31 -0700 (PDT) Date: Fri, 05 Sep 2008 20:26:28 -0700 From: Hitendra Zhangada Subject: Converting to Fast-track : 2008/399 - Rock Performance Counters/Samplers Hypervisor API Sender: Hitendra.Zhangada@sun.com To: Firmware ARC Cc: Shesha Sreenivasamurthy , Prashanth Sreenivasa Message-id: <48C1F864.8050803@sun.com> MIME-version: 1.0 Content-type: multipart/mixed; boundary="Boundary_(ID_8gj8j8JFKaAipt3A455Lsw)" X-PMX-Version: 5.4.1.325704 User-Agent: Thunderbird 2.0.0.16 (Windows/20080708) Status: RO Content-Length: 16020 This is a multi-part message in MIME format. --Boundary_(ID_8gj8j8JFKaAipt3A455Lsw) Content-type: text/plain; format=flowed; charset=ISO-8859-1 Content-transfer-encoding: 7BIT I am converting this case to fast-track case for Shesha Sreenivasamurthy. This case describes ROCK performance counters and samplers along with their associated Hypervisor APIs. See details in the attached specification. This case will be approved for minor/micro/patch OS binding and minor/micro binding for the firmware. The case timer is set to expire on September 12, 2008. Thanks. -- Hitendra Zhangada ============================================= SPS Common SW Features Engineering Software Group, Sun Microsystems, Inc. Work Ph# (858) 625 3757, Ext. x53757 SUN Internal homepage http://esp.west/~hitu --Boundary_(ID_8gj8j8JFKaAipt3A455Lsw) Content-type: text/plain; name=perf.spec Content-transfer-encoding: 7BIT Content-disposition: inline; filename=perf.spec 1. Introduction: This document provides description of performance counters and samplers along with their associated Hypervisor API. This specification provides description of family of functions to initialize, release, get, set, configure, stop, get overflows for both counters and samplers. There are a total of 13 APIs in this document, 7 APIs for counters and 6 for samplers. All these new APIs are a part of new API group API_GROUP_RKPERF (0x206) 2. References: Rock PRM can be found at: https://systemsweb.sfbay.sun.com/rock/team/docs/index.html 3. Performance Counters and Samplers description: In earlier processors, a config register was set with an event of interest and the count was read off a different register after a specific duration of time. In Rock that is not the case. Lot of different registers has to be touched before the actual value can be read. In addition to counters Rock implements samplers that enables collection of statistical data. Software can express interest in certain types of events and when such an event occurs, the software is informed by the hardware by means of a trap. Software may have to then read multiple registers to get a detailed report of the event. Counters and samplers are shared by multiple strands. A strand cannot use a counter/sampler without initialization. In most of the cases a combination of counter and sampler is used to measure events. Therefore, a strand will have exclusive access to both counters and samplers or none of them. To facilitate the correctness and ease of operation a family of APIs are provided. Combination of counters and samplers enables the guest to count various events. 4: Performance Event Trap handler: At the core of performance counter framework lies the performance event trap and its handler. When a counter/sampler overflows/ underflows, a performance event trap is generated. There will be one trap handling strand per chip. The trap handler determines what is the event and which strand is responsible for the event. The handler sends a hyper-privileged cross call to the responsible strand for further processing of the event. The strand restarts counting and sampling before it informs the guest about that event. In case of samplers, handler will put all the sample information requested by the guest in a ring-buffer whose start address and size is provided by the guest during initializing the counter. Hypervisor performs address validation before ring-buffer. Ring-buffer is allocated by the guest. Hypervisor is the producer of events into the ring-buffer. It validates the addresses before touching any address and updates the tail pointer. Guest acts as the consumer and updates the head once it consumes the samples in the ring-buffer. 5. Interfaces: In all the interfaces, counter type is a combination of two values. Lower 32 bits {31..0} contains the counter type and upper 32 bits {63..32} contains the source type. a. Valid counter/sampler types: RK_PERF_CYC (0x100) RK_PERF_INSTR (0x200) RK_PERF_L2 (0x400) RK_PERF_MMU (0x800) RK_PERF_SPEC (0x1000) RK_PERF_YANK (0x2000) RK_PERF_SIBLK (0x4000) RK_PERF_LVLK (0x8000) b.Valid source types: RK_PERF_SRC_STRAND (0x1) RK_PERF_SRC_STRANDS (0x2) RK_PERF_SRC_SIU (0x4) RK_PERF_SRC_MMU (0x8) Source types are used to control the source of the events. When interested in only one strand's event, RK_PERF_SRC_STRAND is specified. When interested in event from all the strands in a chip RK_PERF_SRC_STRANDS is specified. Similarly for L2 counters/samplers there can be two additional source of events, SIU and MMU. They can be specified using RK_PERF_SRC_SIU & RK_PERF_SRC_MMU respectively. All the interfaces belong to: Group#: API_GROUP_RKPERF (0x206) Major#: 1 Minor#: 0 5.1: COUNTERS: ` 5.1.1: Trap#: FAST_TRAP Function: RK_PERF_COUNT_INIT (0x108) arg0: Counter Type ret0: EOK: Success EINVAL: Invalid Counter type ENOACCESS: Counter is in use by a strand in a different guest EWOULDBLOCK: Counter is in use by a strand in the same guest ENOCPU: Source or destination CPU-ID is not a part of current guest This interface provides exclusive access of the counter to the strand on success and initializes different configuration registers. 5.1.2: Trap#: FAST_TRAP Function: RK_PERF_COUNT_RELEASE (0x109) arg0: Counter Type ret0: EOK: Success EINVAL: Invalid Counter type Counter un-initialized ENOACCESS: Current strand is not the owner of the counter This interfaces stops any counting activity and removes exclusive access to the counter. It also clears off the counter and associated configuration values. 5.1.3: Trap#: FAST_TRAP Function: RK_PERF_COUNT_SET (0x10A) arg0: Counter Type arg1: Counter Value ret0: EOK: Success EINVAL: Invalid Counter type Counter out of range Counter un-initialized ENOACCESS: Current strand is not the owner of the counter This API can be used to preset a counter to a specific value. This is usually used to stop a particular counter on one CPU & restart counting on a different CPU from where the first CPU had stopped. At the start of initial counting, it is set to zero. 5.1.4: Trap#: FAST_TRAP Function: RK_PERF_COUNT_GET (0x10B) arg0: Counter Type ret0: EOK: Success EINVAL: Invalid Counter type Counter un-initialized ENOACCESS: Current strand is not the owner of the counter ret1: Counter Value This returns the value of the counter. In case of RK_PERF_LVLK, RK_PERF_SIBLK and RK_PERF_YANK, there is a start of day counter that records all the events. To determine the count for a specific duration, the performance counter framework takes a checkpoint of current count when the counter is started, and returns the difference of current value and the checkpoint as the actual count for that duration. 5.1.5: Trap#: FAST_TRAP Function: RK_PERF_COUNT_START (0x10C) arg0: Counter Type arg1: Configuration value ret0: EOK: Success EINVAL: Invalid Counter type Counter un-initialized ENOACCESS: Current strand is not the owner of the counter This API can be used to start a counter. The configuration value (arg1) to start a counter depends on the counter type (arg0). Fields specified in arg1 may belong to different hardware registers. Hypervisor takes care of setting appropriate bits in appropriate registers. Different configuration values for different counters are explained below. Format of arg1 when arg0 is RK_PERF_CYC or RK_PERF_INSTR: 63 |62 3| 2 | 1 | 0 | +-------|------|------|------|-------+ | TOE | RSVD | User | Priv | Hpriv | +-------|------|------|------|-------+ Format of arg1 when arg0 is RK_PERF_L2 or RK_PERF_MMU: 63 |62 7|6 0 +-------|------|-----+ | TOE | RSVD | TXN | +-------|------|-----+ TXN corresponds to ASI_PERF_L2_CNT_FILTER.TXN for RK_PERF_L2 and ASI_PERF_MMU_CNT_FILTER.TXN for RK_PERF_MMU Format of arg1 when arg0 is RK_PERF_YANK, RK_PERF_SIBLK or RK_PERF_LVLK: 63 |62 0 +-------|------| | TOE | RSVD | +-------|------+ TOE bit corresponds to ASI_CC_ERER.YANK/SIBLK/LVLK 5.1.6: Trap#: FAST_TRAP Function: RK_PERF_COUNT_OVERFLOW (0x10D) arg0: Counter Type ret0: EOK: Success EINVAL: Invalid Counter type Counter un-initialized ENOACCESS: Current strand is not the owner of the counter ret1: Num of times counter has overflowed since last call to this function Hypervisor performance counter framework will record number of times a counter overflowed. This API is used by the guest to retrieve that value and as a side effect, the old value is reset to zero after it is read and before returning to the guest. 5.1.7: Trap#: FAST_TRAP Function: RK_PERF_COUNT_STOP (0x10E) arg0: Counter Type ret0: EOK: Success EINVAL: Invalid Counter type Counter un-initialized ENOACCESS: Current strand is not the owner of the counter This will effectively stop counting. This is achieved by different techniques depending on the counter type. In case of cycle and instruction counters, counters are stopped by zeroing the priv bits, meaning, filtering out user, sys and hpriv cycles/instructions, which effectively stops counting. In case of L2 and MMU counters, it is stopped by zeroing out the source bits, meaning, there is no source for these events, which effectively stops counting. RK_PERF_LVLK, RK_PERF_SIBLK and RK_PERF_YANK counters are never stopped. Therefore, a different algorithm is used to stop these counters. Explanation of this algorithm is beyond the scope of this document. 5.2: SAMPLERS: In every function the "Sampler type" follows the same convention as counter type (Refer section 5.1). 5.2.1: Trap#: FAST_TRAP Function: RK_PERF_SAMPLE_INIT (0x135) arg0: Sampler Type arg1: Ring-buffer Real Address ret0: EOK: Success EINVAL: Invalid sampler type ENOADDR: Invalid address ENOACCESS: Sampler is in use by a strand in a different guest EWOULDBLOCK: Sampler is in use by a strand in the same guest ENOCPU: Source or destination CPU-ID is not a part of current guest This interface will provide exclusive access of sampler to the strand on success and initializes different configuration registers. Validates the real address provided by the guest. 5.2.2: Trap#: FAST_TRAP Function: RK_PERF_SAMPLE_RELEASE (0x136) arg0: Sampler Type ret0: EOK: Success EINVAL: Invalid sampler type Sampler un-initialized ENOACCESS: Current strand is not the owner of the sampler This interfaces stops any counting activity and removes exclusive access to the sampler. It also clears off associated configuration values. 5.2.3: Trap#: FAST_TRAP Function: RK_PERF_SAMPLE_CONFIG (0x137) arg0: Sampler Type arg1: VA of the configuration ASI arg2: Raw value for the configuration ASI ret0: EOK: Success EINVAL: Invalid Sampler type Un-initialized sampler Invalid ASI-VA specified ENOACCESS: Current strand is not the owner of the sampler Hypervisor maintains a list of configuration ASIs that can be configured by the guest. This fucntion validates the ASI specified by the guest against its own list before setting the value. It also validates the value specified on need basis, for example, too low candidate counter value can lead to denial of service. 5.2.4: Trap#: FAST_TRAP Function: RK_PERF_SAMPLE_START (0x138) arg0: Sampler Type arg1: Frequency (Candidate counter value) arg2: Size (Num of bytes for the VA list in arg3. A multiple of x-words) arg3: Start address of the VA list ret0: EOK: Success EINVAL: Invalid Sampler type Un-initialized sampler Frequency out of range ENOADDR: Invalid address in arg3 ENOACCESS: Current strand is not the owner of the sampler This function validates the address range specified by the guest. It then validates all the VAs in the list specified. The guest should populate a list of VAs that it is interested in in its memory and specify the size and start address of the list via this API. Since each VA should be specified in a x-word, size is a multiple of x-words in bytes. All the VAs in the list should be valid for the call to be successful. Hypervisor saves this information in its belly, which is used by the performance event trap handler while recording samples. 5.2.5: Trap#: FAST_TRAP Function: RK_PERF_SAMPLE_PENDING (0x139) arg0: Sampler Type ret0: EOK: Success EINVAL: Invalid Sampler type Un-initialized sampler ENOACCESS: Current strand is not the owner of the sampler ret1: Sample count This function returns the number of unclaimed samples in the ring-buffer since the last call to this function. As a side effect clears the count. 5.2.6: Trap#: FAST_TRAP Function: RK_PERF_SAMPLE_STOP (0x13A) arg0: Sampler Type ret0: EOK: Success EINVAL: Invalid Sampler type Un-initialized sampler ENOACCESS: Current strand is not the owner of the sampler Stops taking any further samples. Unclaimed samples that are in the ring-buffer are untouched. 5.3 Imported Interfaces: Interface Classification Comments ======================================================================== HV APIs Committed FWARC/2005/116 5.4 Exported Interface Table: Interface Classification Comments ======================================================================== RK_PERF_COUNT_INIT API Committed Function 0x108, group 0x206 as defined in section 5.1.1 RK_PERF_COUNT_RELEASE API Committed Function 0x109, group 0x206 as defined in section 5.1.2 RK_PERF_COUNT_SET API Committed Function 0x10A, group 0x206 as defined in section 5.1.3 RK_PERF_COUNT_GET API Committed Function 0x10B, group 0x206 as defined in section 5.1.4 RK_PERF_COUNT_START API Committed Function 0x10C, group 0x206 as defined in section 5.1.5 RK_PERF_COUNT_OVERFLOW API Committed Function 0x10D, group 0x206 as defined in section 5.1.6 RK_PERF_COUNT_STOP API Committed Function 0x10E, group 0x206 as defined in section 5.1.7 RK_PERF_SAMPLE_INIT API Committed Function 0x135, group 0x206 as defined in section 5.2.1 RK_PERF_SAMPLE_RELEASE API Committed Function 0x136, group 0x206 as defined in section 5.2.2 RK_PERF_SAMPLE_CONFIG API Committed Function 0x137, group 0x206 as defined in section 5.2.3 RK_PERF_SAMPLE_START API Committed Function 0x138, group 0x206 as defined in section 5.2.4 RK_PERF_SAMPLE_PENDING API Committed Function 0x139, group 0x206 as defined in section 5.2.5 RK_PERF_SAMPLE_STOP API Committed Function 0x13A, group 0x206 as defined in section 5.2.6 --Boundary_(ID_8gj8j8JFKaAipt3A455Lsw)-- From sacadmin Fri Sep 12 15:52:57 2008 Received: from sunmail4.singapore.sun.com (sunmail4.Singapore.Sun.COM [129.158.71.19]) by sac.sfbay.sun.com (8.13.8+Sun/8.13.8) with ESMTP id m8CMquwt011005 for ; Fri, 12 Sep 2008 15:52:56 -0700 (PDT) Received: from nwk-avmta-1.SFBay.Sun.COM (nwk-avmta-1.SFBay.Sun.COM [129.146.11.74]) by sunmail4.singapore.sun.com (8.13.4+Sun/8.13.3/ENSMAIL,v2.2) with ESMTP id m8CMqqOK016811 for <@sunmail2sca.sfbay.sun.com:fwarc@sun.com>; Sat, 13 Sep 2008 06:52:55 +0800 (SGT) Received: from pmxchannel-daemon.nwk-avmta-1.sfbay.Sun.COM by nwk-avmta-1.sfbay.Sun.COM (Sun Java System Messaging Server 6.2-3.04 (built Jul 15 2005)) id <0K7300905U862J00@nwk-avmta-1.sfbay.Sun.COM> for fwarc@sun.com (ORCPT fwarc@sun.com); Fri, 12 Sep 2008 15:52:54 -0700 (PDT) Received: from sca-es-mail-2.sun.com ([192.18.43.133]) by nwk-avmta-1.sfbay.Sun.COM (Sun Java System Messaging Server 6.2-3.04 (built Jul 15 2005)) with ESMTP id <0K7300332U85Q230@nwk-avmta-1.sfbay.Sun.COM> for fwarc@sun.com (ORCPT fwarc@sun.com); Fri, 12 Sep 2008 15:52:53 -0700 (PDT) Received: from fe-sfbay-10.sun.com ([192.18.43.129]) by sca-es-mail-2.sun.com (8.13.7+Sun/8.12.9) with ESMTP id m8CMqrIK013716 for ; Fri, 12 Sep 2008 15:52:53 -0700 (PDT) Received: from conversion-daemon.fe-sfbay-10.sun.com by fe-sfbay-10.sun.com (Sun Java System Messaging Server 6.2-8.04 (built Feb 28 2007)) id <0K7300B01U6FPH00@fe-sfbay-10.sun.com> (original mail from Hitendra.Zhangada@Sun.COM) for fwarc@sun.com (ORCPT fwarc@sun.com); Fri, 12 Sep 2008 15:52:53 -0700 (PDT) Received: from [129.150.32.24] by fe-sfbay-10.sun.com (Sun Java System Messaging Server 6.2-8.04 (built Feb 28 2007)) with ESMTPSA id <0K7300AACU844R80@fe-sfbay-10.sun.com> for fwarc@sun.com (ORCPT fwarc@sun.com); Fri, 12 Sep 2008 15:52:53 -0700 (PDT) Date: Fri, 12 Sep 2008 15:52:52 -0700 From: Hitendra Zhangada Subject: Re: Converting to Fast-track : 2008/399 - Rock Performance Counters/Samplers Hypervisor API In-reply-to: <48C1F864.8050803@sun.com> Sender: Hitendra.Zhangada@sun.com To: Firmware ARC Cc: Shesha Sreenivasamurthy , Prashanth Sreenivasa Message-id: <48CAF2C4.2090504@sun.com> MIME-version: 1.0 Content-type: text/plain; format=flowed; charset=ISO-8859-1 Content-transfer-encoding: 7BIT X-PMX-Version: 5.4.1.325704 References: <48C1F864.8050803@sun.com> User-Agent: Thunderbird 2.0.0.16 (Windows/20080708) Status: RO Content-Length: 751 Hitendra Zhangada wrote: > I am converting this case to fast-track case for Shesha Sreenivasamurthy. > This case describes ROCK performance counters and samplers along with > their associated Hypervisor APIs. See details in the attached > specification. > > This case will be approved for minor/micro/patch OS binding > and minor/micro binding for the firmware. > > The case timer is set to expire on September 12, 2008. This case has timed out and is now approved. I am update IAM file and HV registry later. Thanks. -- Hitendra Zhangada ============================================= SPS Common SW Features Engineering Software Group, Sun Microsystems, Inc. Work Ph# (858) 625 3757, Ext. x53757 SUN Internal homepage http://esp.west/~hitu