// $Id: pgp_pmc.c,v 1.2 2014/02/03 01:30:35 daviddu Exp $ // $Source: /afs/awd/projects/eclipz/KnowledgeBase/.cvsroot/eclipz/chips/p8/working/procedures/ssx/pgp/pgp_pmc.c,v $ //----------------------------------------------------------------------------- // *! (C) Copyright International Business Machines Corp. 2013 // *! All Rights Reserved -- Property of IBM // *! *** IBM Confidential *** //----------------------------------------------------------------------------- /// \file pgp_pmc.c /// \brief PgP procedures and support for PMC operations /// /// SCOM Operations /// /// The PMC provides an indirect bridge from the OCI to the PIB/PCB. OCC /// firmware therefore has the ability to do immediate putscom()/getscom() /// operations in addition to the capabilities provided by the PORE-GPE /// engines. In PgP, SCOM latency from OCC is expected to be in the range of /// 150 - 1000 ns. The maximum latency of a PIB operation has a hard upper /// bound derived from the hardware implementation. The putscom()/getscom() /// drivers here take advantage of this upper bound and implement tight /// timeouts, enforced by polling the timebase while waiting for the SCOM /// operations to complete. /// /// The latencies are small enough and so well understood that the /// getscom()/putscom() procedures operate in SSX_CRITICAL critical /// sections. There should be no problem for thread-based procedures to be /// written using getscom()/putscom() directly - in fact for short procedures /// it may be less overhead to use getscom()/putscom() than queuing a PORE-GPE /// program. All mainline procedures used by hard real-time code should /// remain as PORE-GPE programs however. /// /// SCOM operations return non-zero error codes that may or may not indicate /// an actual error, depending on which SCOM is being accessed. This error /// code (or 0 for success) is returned as the value of getscom()/putscom(). /// The error severity increases with the severity of the error: /// \code /// /// #define PCB_ERROR_NONE 0 /// #define PCB_ERROR_RESOURCE_OCCUPIED 1 /// #define PCB_ERROR_CHIPLET_OFFLINE 2 /// #define PCB_ERROR_PARTIAL_GOOD 3 /// #define PCB_ERROR_ADDRESS_ERROR 4 /// #define PCB_ERROR_CLOCK_ERROR 5 /// #define PCB_ERROR_PACKET_ERROR 6 /// #define PCB_ERROR_TIMEOUT 7 /// \endcode /// /// The default configuration variable SCOM_ERROR_LIMIT defines the maximum /// error code that will be returned - error codes above the limit (plus hard /// timeouts and other protocol errors) cause an immediate kernel panic. In /// the event of a non-0 error code, getscom() always sets the returned data /// to 0. /// /// In addition to getscom()/putscom() that implement the above defined error /// protocols, the raw APIs _getscom()/_putscom() are also available and /// allow the application full control over timeouts and error handling on a /// SCOM-by-SCOM basis. /// /// \bug Modify getscom/putscom to return the SSX error codes rather than /// 1-7. /// /// \bug Implement and use a generic poll_with_timeout(f, arg, t) #include "ssx.h" #include "pgp_pmc.h" //////////////////////////////////////////////////////////////////////////// // SCOM //////////////////////////////////////////////////////////////////////////// // Common SCOM polling loop with software timeout. The PMC is always polled // at least twice to guarantee that we always poll once after a timeout. static int poll_scom(SsxInterval timeout, pmc_o2p_ctrl_status_reg_t *cs) { SsxTimebase start; int timed_out; start = ssx_timebase_get(); timed_out = 0; do { cs->value = in32(PMC_O2P_CTRL_STATUS_REG); if (!(cs->fields.o2p_ongoing)) { break; } if (timed_out) { return -SCOM_TIMEOUT_ERROR; } timed_out = ((timeout != SSX_WAIT_FOREVER) && ((ssx_timebase_get() - start) > timeout)); } while (1); return 0; } /// A raw getscom() through the PMC OCI/PIB bridge /// /// \param address A standard 32-bit SCOM address, including multicast /// addresses. /// /// \param data Points to a container for the returned data. /// /// \param timeout The software timeout as an SSX interval (timebase ticks), /// or the special value SSX_WAIT_FOREVER to indicate no software timeout. /// /// This routine executes in an SSX_CRITICAL critical section. /// /// Unlike most other APIs, this API returns both positive and negative error /// codes, as well as the 0 code for success. In the event of PCB errors, the /// returned \a data is obtained from the PMC O2P data registers. In the /// event of non-PCB errors, the caller \a data is not modified. /// /// If the transaction experiences a software timeout (controlled by the \a /// timeout parameter) or a protocol error, the PMC PIB master will be left in /// a state in which it is illegal to perform further SCOM access through the /// PMC until the ongoing transaction is finished. /// /// \retval 0 Success /// ///\ retval 1-7 A PCB error code. See \c pcb_common.h /// /// \retval -SCOM_TIMEOUT_ERROR The software timeout specified by the \a /// timeout parameter expired before the transaction completed. /// /// retval -SCOM_PROTOCOL_ERROR_GETSCOM_BUSY The PMC SCOM engine was busy when /// the call was made. int _getscom(uint32_t address, uint64_t *data, SsxInterval timeout) { pmc_o2p_addr_reg_t addr; pmc_o2p_ctrl_status_reg_t cs; SsxMachineContext ctx; Uint64 data64; int rc; ssx_critical_section_enter(SSX_CRITICAL, &ctx); // Check for a transaction already ongoing cs.value = in32(PMC_O2P_CTRL_STATUS_REG); if (cs.fields.o2p_ongoing) { ssx_critical_section_exit(&ctx); return -SCOM_PROTOCOL_ERROR_GETSCOM_BUSY; } // Start the read. The 'read' bit is forced into the address. Writing // the PMC_O2P_ADDR_REG starts the read. addr.value = address; addr.fields.o2p_read_not_write = 1; out32(PMC_O2P_ADDR_REG, addr.value); // Polling and return. rc = poll_scom(timeout, &cs); data64.word[0] = in32(PMC_O2P_RECV_DATA_HI_REG); data64.word[1] = in32(PMC_O2P_RECV_DATA_LO_REG); *data = data64.value; ssx_critical_section_exit(&ctx); if (rc) { return rc; } else { return cs.fields.o2p_scresp; } } /// getscom() through the PMC OCI/PIB bridge /// /// \param address A standard 32-bit SCOM address, including multicast /// addresses. /// /// \param data Points to a container for the returned data. /// /// This routine executes in an SSX_CRITICAL critical section. /// /// Unlike most other APIs, this API returns positive error /// codes, as well as the 0 code for success. In the event of PCB errors, the /// returned \a data is set to 0. /// /// If the transaction experiences a software timeout (controlled by the \a /// timeout parameter), a protocol error, or a PCB error greater than the /// configuration constant SCOM_ERROR_LIMIT this routine causes a kernel /// panic. This may leave the PMC PIB master in a state in which it is illegal /// to perform further SCOM access through the PMC (until the ongoing /// transaction is finished.) /// /// \retval 0 Success /// ///\ retval 1-7 A PCB error code. See \c pcb_common.h int getscom(uint32_t address, uint64_t *data) { int rc; rc = _getscom(address, data, SCOM_TIMEOUT); if (rc == 0) { return 0; } if ((rc > 0) && (rc <= SCOM_ERROR_LIMIT)) { *data = 0; } else { printk("getscom(0x%08x, %p) : Failed with error %d\n", address, data, rc); if (rc > 0) { switch (rc) { case 1: SSX_PANIC(SCOM_PCB_ERROR_1_GETSCOM); break; case 2: SSX_PANIC(SCOM_PCB_ERROR_2_GETSCOM); break; case 3: SSX_PANIC(SCOM_PCB_ERROR_3_GETSCOM); break; case 4: SSX_PANIC(SCOM_PCB_ERROR_4_GETSCOM); break; case 5: SSX_PANIC(SCOM_PCB_ERROR_5_GETSCOM); break; case 6: SSX_PANIC(SCOM_PCB_ERROR_6_GETSCOM); break; default: SSX_PANIC(SCOM_PCB_ERROR_7_GETSCOM); break; } } else if (rc == -SCOM_TIMEOUT_ERROR) { SSX_PANIC(SCOM_TIMEOUT_ERROR_GETSCOM); } else { SSX_PANIC(SCOM_PROTOCOL_ERROR_GETSCOM); } } return rc; } /// A raw putscom() through the PMC OCI/PIB bridge /// /// \param address A standard 32-bit SCOM address, including multicast /// addresses. /// /// \param data The SCOM write data /// /// \param timeout The software timeout as an SSX interval (timebase ticks), /// or the special value SSX_WAIT_FOREVER to indicate no timeout. /// /// This routine executes in an SSX_CRITICAL critical section. /// /// Unlike most other APIs, this API returns both positive and negative error /// codes, as well as the 0 code for success. /// /// If the transaction experiences a software timeout (controlled by the \a /// timeout parameter) or a protocol error, the PMC PIB master will be left in /// a state in which it is illegal to perform further SCOM access through the /// PMC until the ongoing transaction is finished. /// /// \retval 0 Success /// /// \retval 1-7 A PCB error code. See \c pcb_common.h /// /// \retval -SCOM_TIMEOUT The software timeout specified by the \a timeout /// parameter expired before the transaction completed. /// /// \retval -SCOM_PROTOCOL_ERROR_PUTSCOM_BUSY The PMC SCOM engine was busy when /// the call was made. int _putscom(uint32_t address, uint64_t data, SsxInterval timeout) { pmc_o2p_addr_reg_t addr; pmc_o2p_ctrl_status_reg_t cs; SsxMachineContext ctx; Uint64 data64; int rc; ssx_critical_section_enter(SSX_CRITICAL, &ctx); // Check for a transaction already ongoing cs.value = in32(PMC_O2P_CTRL_STATUS_REG); if (cs.fields.o2p_ongoing) { ssx_critical_section_exit(&ctx); return -SCOM_PROTOCOL_ERROR_PUTSCOM_BUSY; } // Start the write. The 'write' bit is cleared in the address. Here the // PIB write starts when the PMC_O2P_SEND_DATA_LO_REG is written. addr.value = address; addr.fields.o2p_read_not_write = 0; out32(PMC_O2P_ADDR_REG, addr.value); data64.value = data; out32(PMC_O2P_SEND_DATA_HI_REG, data64.word[0]); out32(PMC_O2P_SEND_DATA_LO_REG, data64.word[1]); // Poll and return. rc = poll_scom(timeout, &cs); ssx_critical_section_exit(&ctx); if (rc) { return rc; } else { return cs.fields.o2p_scresp; } } /// putscom() through the PMC OCI/PIB bridge /// /// \param address A standard 32-bit SCOM address, including multicast /// addresses. /// /// \param data The SCOM write data. /// /// This routine executes in an SSX_CRITICAL critical section. /// /// Unlike most other APIs, this API returns positive error /// codes, as well as the 0 code for success. /// /// If the transaction experiences a software timeout (controlled by the \a /// timeout parameter), a protocol error, or a PCB error greater than the /// configuration constant SCOM_ERROR_LIMIT this routine causes a kernel /// panic. This may leave the PMC PIB master in a state in which it is illegal /// to perform further SCOM access through the PMC (until the ongoing /// transaction is finished.) /// /// \retval 0 Success /// /// \retval 1-7 A PCB error code. See \c pcb_common.h int putscom(uint32_t address, uint64_t data) { int rc; rc = _putscom(address, data, SCOM_TIMEOUT); if ((rc == 0) || ((rc > 0) && (rc <= SCOM_ERROR_LIMIT))) { return rc; } printk("putscom(0x%08x, 0x%016llx) : Failed with error %d\n", address, data, rc); if (rc > 0) { switch (rc) { case 1: SSX_PANIC(SCOM_PCB_ERROR_1_PUTSCOM); break; case 2: SSX_PANIC(SCOM_PCB_ERROR_2_PUTSCOM); break; case 3: SSX_PANIC(SCOM_PCB_ERROR_3_PUTSCOM); break; case 4: SSX_PANIC(SCOM_PCB_ERROR_4_PUTSCOM); break; case 5: SSX_PANIC(SCOM_PCB_ERROR_5_PUTSCOM); break; case 6: SSX_PANIC(SCOM_PCB_ERROR_6_PUTSCOM); break; default: SSX_PANIC(SCOM_PCB_ERROR_7_PUTSCOM); break; } } else if (rc == -SCOM_TIMEOUT_ERROR) { SSX_PANIC(SCOM_TIMEOUT_ERROR_PUTSCOM); } else { SSX_PANIC(SCOM_PROTOCOL_ERROR_PUTSCOM); } return rc; }