From 621ca384a7a5efb2cc7597504dc17b741cd2df10 Mon Sep 17 00:00:00 2001 From: efdesign98 Date: Mon, 20 Jun 2011 18:12:43 -0700 Subject: Move existing AMD Ffamily14 code to f14 folder This change moves the AMD Family14 cpu Agesa code to the vendorcode/amd/agesa/f14 folder to complete the transition to the family oriented folder structure. Change-Id: I211e80ee04574cc713f38b4cc1b767dbb2bfaa59 Signed-off-by: Frank Vibrans Signed-off-by: efdesign98 Reviewed-on: http://review.coreboot.org/52 Tested-by: build bot (Jenkins) Reviewed-by: Marc Jones --- .../f14/Proc/HT/Features/htFeatOptimization.c | 886 +++++++++++++++++++++ 1 file changed, 886 insertions(+) create mode 100644 src/vendorcode/amd/agesa/f14/Proc/HT/Features/htFeatOptimization.c (limited to 'src/vendorcode/amd/agesa/f14/Proc/HT/Features/htFeatOptimization.c') diff --git a/src/vendorcode/amd/agesa/f14/Proc/HT/Features/htFeatOptimization.c b/src/vendorcode/amd/agesa/f14/Proc/HT/Features/htFeatOptimization.c new file mode 100644 index 0000000000..bcd669f799 --- /dev/null +++ b/src/vendorcode/amd/agesa/f14/Proc/HT/Features/htFeatOptimization.c @@ -0,0 +1,886 @@ +/* $NoKeywords:$ */ +/** + * @file + * + * Link Optimization Routines. + * + * Contains routines for determining width, frequency, and other + * Link features + * + * @xrefitem bom "File Content Label" "Release Content" + * @e project: AGESA + * @e sub-project: HyperTransport + * @e \$Revision: 35136 $ @e \$Date: 2010-07-16 11:29:48 +0800 (Fri, 16 Jul 2010) $ + * + */ +/* + ***************************************************************************** + * + * Copyright (c) 2011, Advanced Micro Devices, Inc. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Advanced Micro Devices, Inc. nor the names of + * its contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL ADVANCED MICRO DEVICES, INC. BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * *************************************************************************** + * + */ + +/* + *---------------------------------------------------------------------------- + * MODULES USED + * + *---------------------------------------------------------------------------- + */ + + + +#include "AGESA.h" +#include "amdlib.h" +#include "Ids.h" +#include "Topology.h" +#include "htFeat.h" +#include "IdsHt.h" +#include "htInterface.h" +#include "htNb.h" +#include "htFeatOptimization.h" +#include "htNotify.h" +#include "Filecode.h" +CODE_GROUP (G1_PEICC) +RDATA_GROUP (G1_PEICC) + +#define FILECODE PROC_HT_FEATURES_HTFEATOPTIMIZATION_FILECODE + +extern CONST PF_HtIdsGetPortOverride ROMDATA pf_HtIdsGetPortOverride; + +/*---------------------------------------------------------------------------- + * DEFINITIONS AND MACROS + * + *---------------------------------------------------------------------------- + */ +#define PCI_CONFIG_COMMAND_REG04 4 +#define PCI_CONFIG_REVISION_REG08 8 + +/*---------------------------------------------------------------------------- + * TYPEDEFS AND STRUCTURES + * + *---------------------------------------------------------------------------- + */ + +/*---------------------------------------------------------------------------- + * PROTOTYPES OF LOCAL FUNCTIONS + * + *---------------------------------------------------------------------------- + */ + +/*---------------------------------------------------------------------------- + * EXPORTED FUNCTIONS + * + *---------------------------------------------------------------------------- + */ + +/*---------------------------------------------------------------------------- + * LOCAL FUNCTIONS + * + *---------------------------------------------------------------------------- + */ + +/*************************************************************************** + *** Link Optimization *** + ***************************************************************************/ + +/*----------------------------------------------------------------------------------------*/ +/** + * Given the bits set in the register field, return the width it represents. + * + * As invalid width values or encodings are rare except during debug, catch those using + * ASSERT(). This means theoretically we are returning an incorrect result if that + * happens. The default chosen for the result is arbitrarily 8 bits. This is likely + * not to be the actual correct width and may cause a crash, hang, or incorrect operation. + * Hardware often ignores writes of invalid width encodings. + * + * @note This routine is used for CPUs as well as IO devices, as all comply to the + * "HyperTransport I/O Link Specification ". + * + * @param[in] Value The bits for the register + * + * @return The width + */ +UINT8 +STATIC +ConvertBitsToWidth ( + IN UINT8 Value + ) +{ + UINT8 Result; + + Result = 0; + + switch (Value) { + + case 1: + Result = 16; + break; + + case 0: + Result = 8; + break; + + case 3: + Result = 32; + break; + + case 5: + Result = 4; + break; + + case 4: + Result = 2; + break; + + default: + ASSERT (FALSE); + } + return Result; +} + +/*----------------------------------------------------------------------------------------*/ +/** + * Translate a desired width setting to the bits to set in the register field. + * + * As invalid width values or encodings are rare except during debug, catch those using + * ASSERT(). This means theoretically we are returning an incorrect result if that + * happens. The default chosen for the result is arbitrarily 8 bits. This is likely + * not to be the actual correct width and may cause a crash, hang, or incorrect operation. + * Hardware often ignores writes of invalid width encodings. + * + * @note This routine is used for CPUs as well as IO devices, as all comply to the + * "HyperTransport I/O Link Specification ". + * + * @param[in] Value the width Value + * + * @return The bits for the register + */ +UINT8 +ConvertWidthToBits ( + IN UINT8 Value + ) +{ + UINT8 Result; + + Result = 8; + + switch (Value) { + + case 16: + Result = 1; + break; + + case 8: + Result = 0; + break; + + case 32: + Result = 3; + break; + + case 4: + Result = 5; + break; + + case 2: + Result = 4; + break; + + default: + ASSERT (FALSE); + } + return Result; +} + +/*----------------------------------------------------------------------------------------*/ +/** + * Access HT Link Control Register. + * + * @HtFeatMethod{::F_SET_HT_CONTROL_REGISTER_BITS} + * + * Provide a common routine for accessing the HT Link Control registers (84, a4, c4, + * e4), to enforce not clearing the HT CRC error bits. Replaces direct use of + * AmdPCIWriteBits(). + * + * @note: This routine is called for CPUs as well as IO Devices! All comply to the + * "HyperTransport I/O Link Specification ". + * + * @param[in] Reg the PCI config address the control register + * @param[in] HiBit the high bit number + * @param[in] LoBit the low bit number + * @param[in] Value the value to write to that bit range. Bit 0 => loBit. + * @param[in] State Our state, config handle for lib + */ +VOID +SetHtControlRegisterBits ( + IN PCI_ADDR Reg, + IN UINT8 HiBit, + IN UINT8 LoBit, + IN UINT32 *Value, + IN STATE_DATA *State + ) +{ + UINT32 Temp; + UINT32 mask; + + ASSERT ((HiBit < 32) && (LoBit < 32) && (HiBit >= LoBit) && ((Reg.AddressValue & 0x3) == 0)); + ASSERT ((HiBit < 8) || (LoBit > 9)); + + // A 1 << 32 == 1 << 0 due to x86 SHL instruction, so skip if that is the case + if ((HiBit - LoBit) != 31) { + mask = (((UINT32)1 << (HiBit - LoBit + 1)) - 1); + } else { + mask = (UINT32)0xFFFFFFFF; + } + + LibAmdPciRead (AccessWidth32, Reg, &Temp, State->ConfigHandle); + Temp &= ~(mask << LoBit); + Temp |= (*Value & mask) << LoBit; + Temp &= (UINT32)HT_CONTROL_CLEAR_CRC; + LibAmdPciWrite (AccessWidth32, Reg, &Temp, State->ConfigHandle); +} + +/*----------------------------------------------------------------------------------------*/ +/** + * Set HT Frequency register for IO Devices + * + * Provide a common routine for accessing the HT Link Frequency registers at offset 8 + * and 0x10, to enforce not clearing the HT Link error bits. Replaces direct use of + * AmdPCIWriteBits(). + * + * @note This routine is called for IO Devices only!! All comply to the + * "HyperTransport I/O Link Specification ". + * + * @param[in] Reg the PCI config address the control register + * @param[in] Hibit the high bit number + * @param[in] Lobit the low bit number + * @param[in] Value the value to write to that bit range. Bit 0 => loBit. + * @param[in] State Our state, config handle for lib + */ +VOID +STATIC +SetHtIoFrequencyRegisterBits ( + IN PCI_ADDR Reg, + IN UINT8 Hibit, + IN UINT8 Lobit, + IN UINT32 *Value, + IN STATE_DATA *State + ) +{ + UINT32 Mask; + UINT32 Temp; + + ASSERT ((Hibit < 32) && (Lobit < 32) && (Hibit >= Lobit) && ((Reg.AddressValue & 0x3) == 0)); + ASSERT ((Hibit < 12) || (Lobit > 14)); + + // A 1<<32 == 1<<0 due to x86 SHL instruction, so skip if that is the case + if ((Hibit - Lobit) != 31) { + Mask = (((UINT32)1 << ((Hibit - Lobit) + 1)) - 1); + } else { + Mask = (UINT32)0xFFFFFFFF; + } + + LibAmdPciRead (AccessWidth32, Reg, &Temp, State->ConfigHandle); + Temp &= ~(Mask << Lobit); + Temp |= (*Value & Mask) << Lobit; + Temp &= (UINT32)HT_FREQUENCY_CLEAR_LINK_ERRORS; + LibAmdPciWrite (AccessWidth32, Reg, &Temp, State->ConfigHandle); +} + +/*----------------------------------------------------------------------------------------*/ +/** + * Get Link features into system data structure. + * + * @HtFeatMethod{::F_GATHER_LINK_DATA} + * + * For all discovered Links, populate the port list with the frequency and width + * capabilities. Gather support data for: + * - Unit ID Clumping + * + * @param[in] State our global state, port list + */ +VOID +GatherLinkData ( + IN STATE_DATA *State + ) +{ + UINT8 i; + PCI_ADDR LinkBase; + PCI_ADDR Reg; + UINT32 Bits; + UINT8 Revision; + + // Get the capability base for whatever device type the link port is on + for (i = 0; i < (State->TotalLinks * 2); i++) { + if ((*State->PortList)[i].Type == PORTLIST_TYPE_CPU) { + LinkBase = State->Nb->MakeLinkBase ((*State->PortList)[i].NodeID, (*State->PortList)[i].Link, State->Nb); + (*State->PortList)[i].Pointer = LinkBase; + } else { + LinkBase = (*State->PortList)[i].Pointer; + if ((*State->PortList)[i].Link == 1) { + LinkBase.Address.Register += HTSLAVE_LINK01_OFFSET; + } + } + + // Getting the Width is standard across device types + Reg = LinkBase; + Reg.Address.Register += HTSLAVE_LINK_CONTROL_0_REG; + LibAmdPciReadBits (Reg, 22, 20, &Bits, State->ConfigHandle); + (*State->PortList)[i].PrvWidthOutCap = ConvertBitsToWidth ((UINT8)Bits); + + LibAmdPciReadBits (Reg, 18, 16, &Bits, State->ConfigHandle); + (*State->PortList)[i].PrvWidthInCap = ConvertBitsToWidth ((UINT8)Bits); + + // Get Frequency and other device type specific features + if ((*State->PortList)[i].Type == PORTLIST_TYPE_CPU) { + State->Nb->GatherLinkFeatures (&(*State->PortList)[i], State->HtInterface, State->PlatformConfiguration, State->Nb); + } else { + Reg = LinkBase; + Reg.Address.Register += HTSLAVE_FREQ_REV_0_REG; + LibAmdPciReadBits (Reg, 31, 16, &Bits, State->ConfigHandle); + (*State->PortList)[i].PrvFrequencyCap = Bits; + + // Unit ID Clumping Support + if (State->IsUsingUnitIdClumping) { + if (DoesDeviceHaveHtSubtypeCap (LinkBase, HT_UNITID_CAPABILITY, &Reg, State)) { + Reg.Address.Register += HTUNIT_SUPPORT_REG; + LibAmdPciReadBits (Reg, 31, 0, &Bits, State->ConfigHandle); + } else { + // Not there, that's ok, we don't know that it should have one. + // Check for Passive support. (Bit 0 won't be set if full support is implemented, + // so we can use it to indicate passive support in our portlist struct). + Reg = LinkBase; + Reg.Address.Register += HTSLAVE_FEATURECAP_REG; + Bits = 1; + LibAmdPciWriteBits (Reg, 5, 5, &Bits, State->ConfigHandle); + LibAmdPciReadBits (Reg, 5, 5, &Bits, State->ConfigHandle); + } + (*State->PortList)[i].ClumpingSupport = Bits; + } else { + (*State->PortList)[i].ClumpingSupport = HT_CLUMPING_DISABLE; + } + + Reg = LinkBase; + Reg.Address.Register = PCI_CONFIG_REVISION_REG08; + LibAmdPciReadBits ( LinkBase, 7, 0, &Bits, State->ConfigHandle); + Revision = (UINT8) Bits; + + LinkBase.Address.Register = 0; + LibAmdPciRead (AccessWidth32, LinkBase, &Bits, State->ConfigHandle); + + State->HtInterface->GetDeviceCapOverride ((*State->PortList)[i].NodeID, + (*State->PortList)[i].HostLink, + (*State->PortList)[i].HostDepth, + (*State->PortList)[i].Pointer, + Bits, + Revision, + (*State->PortList)[i].Link, + &((*State->PortList)[i].PrvWidthInCap), + &((*State->PortList)[i].PrvWidthOutCap), + &((*State->PortList)[i].PrvFrequencyCap), + &((*State->PortList)[i].ClumpingSupport), + State); + } + } +} + +/*----------------------------------------------------------------------------------------*/ +/** + * Optimize Links. + * + * @HtFeatMethod{::F_SELECT_OPTIMAL_WIDTH_AND_FREQUENCY} + * + * For all Links: + * Examine both sides of a Link and determine the optimal frequency and width, + * taking into account externally provided limits and enforcing any other limit + * or matching rules as applicable except subLink balancing. Update the port + * list data with the optimal settings. + * + * @note no hardware state changes in this routine. + * + * @param[in,out] State Process and update portlist + */ +VOID +SelectOptimalWidthAndFrequency ( + IN OUT STATE_DATA *State + ) +{ + UINT8 i; + UINT8 j; + UINT8 Freq; + UINT32 Temp; + UINT32 CbPcbFreqLimit; + UINT8 CbPcbABDownstreamWidth; + UINT8 CbPcbBAUpstreamWidth; + + for (i = 0; i < (State->TotalLinks * 2); i += 2) { + CbPcbFreqLimit = HT_FREQUENCY_NO_LIMIT; + CbPcbABDownstreamWidth = HT_WIDTH_16_BITS; + CbPcbBAUpstreamWidth = HT_WIDTH_16_BITS; + + if (((*State->PortList)[i].Type == PORTLIST_TYPE_CPU) && ((*State->PortList)[i + 1].Type == PORTLIST_TYPE_CPU)) { + State->HtInterface->GetCpu2CpuPcbLimits ((*State->PortList)[i].NodeID, + (*State->PortList)[i].Link, + (*State->PortList)[i + 1].NodeID, + (*State->PortList)[i + 1].Link, + &CbPcbABDownstreamWidth, + &CbPcbBAUpstreamWidth, + &CbPcbFreqLimit, + State + ); + } else { + State->HtInterface->GetIoPcbLimits ((*State->PortList)[i + 1].NodeID, + (*State->PortList)[i + 1].HostLink, + (*State->PortList)[i + 1].HostDepth, + &CbPcbABDownstreamWidth, + &CbPcbBAUpstreamWidth, + &CbPcbFreqLimit, + State + ); + } + + Temp = (*State->PortList)[i].PrvFrequencyCap; + Temp &= (*State->PortList)[i + 1].PrvFrequencyCap; + Temp &= CbPcbFreqLimit; + (*State->PortList)[i].CompositeFrequencyCap = (UINT32)Temp; + (*State->PortList)[i + 1].CompositeFrequencyCap = (UINT32)Temp; + + ASSERT (Temp != 0); + Freq = LibAmdBitScanReverse (Temp); + (*State->PortList)[i].SelFrequency = Freq; + (*State->PortList)[i + 1].SelFrequency = Freq; + + Temp = (*State->PortList)[i].PrvWidthOutCap; + if ((*State->PortList)[i + 1].PrvWidthInCap < Temp) { + Temp = (*State->PortList)[i + 1].PrvWidthInCap; + } + if (CbPcbABDownstreamWidth < Temp) { + Temp = CbPcbABDownstreamWidth; + } + (*State->PortList)[i].SelWidthOut = (UINT8)Temp; + (*State->PortList)[i + 1].SelWidthIn = (UINT8)Temp; + + Temp = (*State->PortList)[i].PrvWidthInCap; + if ((*State->PortList)[i + 1].PrvWidthOutCap < Temp) { + Temp = (*State->PortList)[i + 1].PrvWidthOutCap; + } + if (CbPcbBAUpstreamWidth < Temp) { + Temp = CbPcbBAUpstreamWidth; + } + (*State->PortList)[i].SelWidthIn = (UINT8)Temp; + (*State->PortList)[i + 1].SelWidthOut = (UINT8)Temp; + } + // Calculate unit id clumping + // + // Find the root of each IO Chain, process the chain for clumping support. + // The root is always the first link of the chain in the port list. + // Clumping is not device link specific, so we can just look at the upstream ports (j+1). Use ASSERTs to sanity + // check the downstream ports (j). If any device on the chain does not support clumping, the entire chain will be + // disabled for clumping. + // After analyzing the clumping support on the chain the CPU's portlist has the enable mask. Update all the + // IO Devices on the chain with the enable mask. If any device's only have passive support, that is already enabled. + // + if (State->IsUsingUnitIdClumping) { + for (i = 0; i < (State->TotalLinks * 2); i += 2) { + if (((*State->PortList)[i].Type == PORTLIST_TYPE_CPU) && ((*State->PortList)[i + 1].Type == PORTLIST_TYPE_IO)) { + (*State->PortList)[i].ClumpingSupport = HT_CLUMPING_DISABLE; + if ((*State->PortList)[i + 1].ClumpingSupport != HT_CLUMPING_DISABLE) { + (*State->PortList)[i].ClumpingSupport |= (*State->PortList)[i + 1].ClumpingSupport; + for (j = i + 2; j < (State->TotalLinks * 2); j += 2) { + if (((*State->PortList)[j].Type == PORTLIST_TYPE_IO) && ((*State->PortList)[j + 1].Type == PORTLIST_TYPE_IO)) { + if (((*State->PortList)[i].NodeID == (*State->PortList)[j + 1].NodeID) && + ((*State->PortList)[i].Link == (*State->PortList)[j + 1].HostLink)) { + ASSERT (((*State->PortList)[i].NodeID == (*State->PortList)[j + 1].NodeID) && + ((*State->PortList)[i].Link == (*State->PortList)[j].HostLink)); + if ((*State->PortList)[j + 1].ClumpingSupport != HT_CLUMPING_DISABLE) { + ASSERT ((((*State->PortList)[j + 1].ClumpingSupport & HT_CLUMPING_PASSIVE) == 0) || + (((*State->PortList)[j + 1].ClumpingSupport & ~(HT_CLUMPING_PASSIVE)) == 0)); + (*State->PortList)[i].ClumpingSupport |= (*State->PortList)[j + 1].ClumpingSupport; + } else { + (*State->PortList)[i].ClumpingSupport = HT_CLUMPING_DISABLE; + break; + } + } + } + } + if ((*State->PortList)[i + 1].ClumpingSupport != HT_CLUMPING_PASSIVE) { + (*State->PortList)[i + 1].ClumpingSupport = (*State->PortList)[i].ClumpingSupport; + } + for (j = i + 2; j < (State->TotalLinks * 2); j += 2) { + if (((*State->PortList)[j].Type == PORTLIST_TYPE_IO) && ((*State->PortList)[j + 1].Type == PORTLIST_TYPE_IO)) { + if (((*State->PortList)[i].NodeID == (*State->PortList)[j + 1].NodeID) && + ((*State->PortList)[i].Link == (*State->PortList)[j + 1].HostLink)) { + if ((*State->PortList)[j + 1].ClumpingSupport != HT_CLUMPING_PASSIVE) { + (*State->PortList)[j + 1].ClumpingSupport = (*State->PortList)[i].ClumpingSupport; + // The downstream isn't really passive, just mark it so in order to write the device only once. + (*State->PortList)[j].ClumpingSupport = HT_CLUMPING_PASSIVE; + } + } + } + } + } + } + } + } +} + +/*----------------------------------------------------------------------------------------*/ +/** + * Change the hardware state for all Links according to the now optimized data in the + * port list data structure. + * + * @HtFeatMethod{::F_SET_LINK_DATA} + * + * @param[in] State our global state, port list + */ +VOID +SetLinkData ( + IN STATE_DATA *State + ) +{ + UINT8 i; + PCI_ADDR LinkBase; + PCI_ADDR Reg; + UINT32 Temp; + UINT32 Widthin; + UINT32 Widthout; + UINT32 Bits; + PCI_ADDR CurrentPtr; + HTIDS_PORT_OVERRIDE_LIST PortOverrides; + + PortOverrides = NULL; + + for (i = 0; i < (State->TotalLinks * 2); i++) { + + ASSERT ((*State->PortList)[i & 0xFE].SelWidthOut == (*State->PortList)[ (i & 0xFE) + 1].SelWidthIn); + ASSERT ((*State->PortList)[i & 0xFE].SelWidthIn == (*State->PortList)[ (i & 0xFE) + 1].SelWidthOut); + ASSERT ((*State->PortList)[i & 0xFE].SelFrequency == (*State->PortList)[ (i & 0xFE) + 1].SelFrequency); + + if ((*State->PortList)[i].SelRegang) { + ASSERT ((*State->PortList)[i].Type == PORTLIST_TYPE_CPU); + ASSERT ((*State->PortList)[i].Link < 4); + State->Nb->SetLinkRegang ( + (*State->PortList)[i].NodeID, + (*State->PortList)[i].Link, + State->Nb + ); + } + + // + // IDS port override for CPUs and IO Devices + // + pf_HtIdsGetPortOverride ((BOOLEAN) ((i & 1) == 0), &(*State->PortList)[i], &(*State->PortList)[i + 1], &PortOverrides, State); + + LinkBase = (*State->PortList)[i].Pointer; + if (((*State->PortList)[i].Type == PORTLIST_TYPE_IO) && ((*State->PortList)[i].Link == 1)) { + LinkBase.Address.Register += HTSLAVE_LINK01_OFFSET; + } + + // HT CRC Feature, set if configured. The default is not to set it, because with some chipsets it + // will lock up if done here. + if (State->IsSetHtCrcFlood) { + Temp = 1; + Reg = LinkBase; + Reg.Address.Register += HTHOST_LINK_CONTROL_REG; + State->HtFeatures->SetHtControlRegisterBits (Reg, 1, 1, &Temp, State); + if ((*State->PortList)[i].Type == PORTLIST_TYPE_IO) { + // IO Devices also need to have SERR enabled. + Reg = LinkBase; + Reg.Address.Register = PCI_CONFIG_COMMAND_REG04; + LibAmdPciWriteBits (Reg, 8, 8, &Temp, State->ConfigHandle); + } + } + + // Some IO devices don't work properly when setting widths, so write them in a single operation, + // rather than individually. + // + Widthout = ConvertWidthToBits ((*State->PortList)[i].SelWidthOut); + ASSERT (Widthout == 1 || Widthout == 0 || Widthout == 5 || Widthout == 4); + Widthin = ConvertWidthToBits ((*State->PortList)[i].SelWidthIn); + ASSERT (Widthin == 1 || Widthin == 0 || Widthin == 5 || Widthin == 4); + + Temp = (Widthin & 7) | ((Widthout & 7) << 4); + Reg = LinkBase; + Reg.Address.Register += HTHOST_LINK_CONTROL_REG; + State->HtFeatures->SetHtControlRegisterBits (Reg, 31, 24, &Temp, State); + + Temp = (*State->PortList)[i].SelFrequency; + if ((*State->PortList)[i].Type == PORTLIST_TYPE_CPU) { + State->Nb->SetLinkFrequency ( + (*State->PortList)[i].NodeID, + (*State->PortList)[i].Link, + (UINT8)Temp, + State->Nb + ); + } else { + ASSERT (Temp <= HT_FREQUENCY_2600M); + // Write the frequency setting + Reg = LinkBase; + Reg.Address.Register += HTSLAVE_FREQ_REV_0_REG; + SetHtIoFrequencyRegisterBits (Reg, 11, 8, &Temp, State); + + // Handle additional HT3 frequency requirements, if needed, + // or clear them if switching down to ht1 on a warm reset. + // Gen1 = 200Mhz -> 1000MHz, Gen3 = 1200MHz -> 2600MHz + // + // Even though we assert if debugging, we need to check that the capability was + // found always, since this is an unknown hardware device, also we are taking + // unqualified frequency from the external interface (could be trying to do ht3 + // on an ht1 IO device). + // + + if (Temp > HT_FREQUENCY_1000M) { + // Enabling features if gen 3 + Bits = 1; + } else { + // Disabling features if gen 1 + Bits = 0; + } + + // Retry Enable + if (DoesDeviceHaveHtSubtypeCap (LinkBase, HT_RETRY_CAPABILITY, &CurrentPtr, State)) { + ASSERT ((*State->PortList)[i].Link < 2); + CurrentPtr.Address.Register += HTRETRY_CONTROL_REG; + LibAmdPciWriteBits (CurrentPtr, + ((*State->PortList)[i].Link * 16), + ((*State->PortList)[i].Link * 16), + &Bits, + State->ConfigHandle); + } else { + // If we are turning it off, that may mean the device was only ht1 capable, + // so don't complain that we can't do it. + // + if (Bits != 0) { + NotifyWarningOptRequiredCapRetry ((*State->PortList)[i].NodeID, + (*State->PortList)[i].HostLink, + (*State->PortList)[i].HostDepth, + State); + } + } + + // Scrambling enable + if (DoesDeviceHaveHtSubtypeCap (LinkBase, HT_GEN3_CAPABILITY, &CurrentPtr, State)) { + ASSERT ((*State->PortList)[i].Link < 2); + CurrentPtr.Address.Register = CurrentPtr.Address.Register + + HTGEN3_LINK_TRAINING_0_REG + + ((*State->PortList)[i].Link * HTGEN3_LINK01_OFFSET); + LibAmdPciWriteBits (CurrentPtr, 3, 3, &Bits, State->ConfigHandle); + } else { + // If we are turning it off, that may mean the device was only ht1 capable, + // so don't complain that we can't do it. + // + if (Bits != 0) { + NotifyWarningOptRequiredCapGen3 ((*State->PortList)[i].NodeID, + (*State->PortList)[i].HostLink, + (*State->PortList)[i].HostDepth, + State); + } + } + } + // Enable Unit ID Clumping if supported. + if (State->IsUsingUnitIdClumping) { + if (((*State->PortList)[i].ClumpingSupport != HT_CLUMPING_PASSIVE) && + ((*State->PortList)[i].ClumpingSupport != HT_CLUMPING_DISABLE)) { + Bits = (*State->PortList)[i].ClumpingSupport; + if ((*State->PortList)[i].Type == PORTLIST_TYPE_CPU) { + State->Nb->SetLinkUnitIdClumping ( + (*State->PortList)[i].NodeID, + (*State->PortList)[i].Link, + (*State->PortList)[i].ClumpingSupport, + State->Nb + ); + } else { + if (DoesDeviceHaveHtSubtypeCap (LinkBase, HT_UNITID_CAPABILITY, &Reg, State)) { + Reg.Address.Register += HTUNIT_ENABLE_REG; + LibAmdPciWriteBits (Reg, 31, 0, &Bits, State->ConfigHandle); + } else { + // If we found one when gathering support, we have to find one now. + ASSERT (FALSE); + } + } + } + } + } +} + +/*------------------------------------------------------------------------------------------*/ +/** + * Find a specific HT capability type. + * + * Search all the PCI Config space capabilities on any type of device for an + * HT capability of the specific subtype. + * + * @param[in] DevicePointer A PCI Config address somewhere in the device config space + * @param[in] CapSubType The HT capability subtype to find + * @param[out] CapabilityBase The Config space base address of the capability, if found. + * @param[in] State Our State + * + * @retval TRUE the capability was found + * @retval FALSE the capability was not found + */ +BOOLEAN +DoesDeviceHaveHtSubtypeCap ( + IN PCI_ADDR DevicePointer, + IN UINT8 CapSubType, + OUT PCI_ADDR *CapabilityBase, + IN STATE_DATA *State + ) +{ + BOOLEAN IsFound; + BOOLEAN IsDone; + PCI_ADDR Reg; + UINT32 Temp; + UINT32 RegSubType; + UINT32 RegSubTypeMask; + + // Set the PCI Config Space base and the match value. + IsFound = FALSE; + IsDone = FALSE; + Reg = DevicePointer; + Reg.Address.Register = 0; + if (CapSubType < (HT_HOST_CAPABILITY + 1)) { + // HT Interface sub type + RegSubType = ((UINT32) (CapSubType << 29) | (UINT32)8); + RegSubTypeMask = HT_INTERFACE_CAP_SUBTYPE_MASK; + } else { + // Other HT capability subtype + RegSubType = ((UINT32) (CapSubType << 27) | (UINT32)8); + RegSubTypeMask = HT_CAP_SUBTYPE_MASK; + } + (*CapabilityBase).AddressValue = (UINT32)ILLEGAL_SBDFO; + + // Find it + do { + LibAmdPciFindNextCap (&Reg, State->ConfigHandle); + if (Reg.AddressValue != (UINT32)ILLEGAL_SBDFO) { + LibAmdPciRead (AccessWidth32, Reg, &Temp, State->ConfigHandle); + // HyperTransport and subtype capability ? + if ((Temp & RegSubTypeMask) == RegSubType) { + *CapabilityBase = Reg; + IsFound = TRUE; + } + // Some other capability, keep looking + } else { + // Not there + IsDone = TRUE; + } + } while (!IsFound && !IsDone); + + return IsFound; +} + +/*----------------------------------------------------------------------------------------*/ +/** + * Retry must be enabled on all coherent links if it is enabled on any coherent links. + * + * @HtFeatMethod{::F_SET_LINK_DATA} + * + * Effectively, this means HT3 on some links cannot be mixed with HT1 on others. + * Scan the CPU to CPU links for this condition and limit those frequencies to HT1 + * if it is detected. + * (Non-coherent links are independent.) + * + * @param[in,out] State global state, port frequency settings. + * + * @retval TRUE Fixup occurred, all coherent links HT1 + * @retval FALSE No changes + */ +BOOLEAN +IsCoherentRetryFixup ( + IN STATE_DATA *State + ) +{ + UINT8 Freq; + UINT8 i; + UINT8 DetectedFrequencyState; + BOOLEAN IsMixed; + UINT32 Temp; + + // + // detectedFrequencyState: + // 0 - initial state + // 1 - HT1 Frequencies detected + // 2 - HT3 Frequencies detected + // + IsMixed = FALSE; + DetectedFrequencyState = 0; + + // Scan coherent links for a mix of HT3 / HT1 + for (i = 0; i < (State->TotalLinks * 2); i += 2) { + if (((*State->PortList)[i].Type == PORTLIST_TYPE_CPU) && ((*State->PortList)[i + 1].Type == PORTLIST_TYPE_CPU)) { + // At this point, Frequency of port [i+1] must equal [i], so just check one of them. + switch (DetectedFrequencyState) { + case 0: + // Set current state to indicate what link frequency we found first + if ((*State->PortList)[i].SelFrequency > HT_FREQUENCY_1000M) { + // HT3 frequencies + DetectedFrequencyState = 2; + } else { + // HT1 frequencies + DetectedFrequencyState = 1; + } + break; + case 1: + // If HT1 frequency detected, fail any HT3 frequency + if ((*State->PortList)[i].SelFrequency > HT_FREQUENCY_1000M) { + IsMixed = TRUE; + } + break; + case 2: + // If HT3 frequency detected, fail any HT1 frequency + if ((*State->PortList)[i].SelFrequency <= HT_FREQUENCY_1000M) { + IsMixed = TRUE; + } + break; + default: + ASSERT (FALSE); + } + if (IsMixed) { + // Don't need to keep checking after we find a mix. + break; + } + } + } + + if (IsMixed) { + for (i = 0; i < (State->TotalLinks * 2); i += 2) { + if (((*State->PortList)[i].Type == PORTLIST_TYPE_CPU) && ((*State->PortList)[i + 1].Type == PORTLIST_TYPE_CPU)) { + // Limit coherent links to HT 1 frequencies. + Temp = (*State->PortList)[i].CompositeFrequencyCap & (*State->PortList)[i + 1].CompositeFrequencyCap; + Temp &= HT_FREQUENCY_LIMIT_HT1_ONLY; + ASSERT (Temp != 0); + (*State->PortList)[i].CompositeFrequencyCap = Temp; + (*State->PortList)[i + 1].CompositeFrequencyCap = Temp; + Freq = LibAmdBitScanReverse (Temp); + (*State->PortList)[i].SelFrequency = Freq; + (*State->PortList)[i + 1].SelFrequency = Freq; + } + } + } + return (IsMixed); +} -- cgit v1.2.3