X-Git-Url: http://wien.tomnetworks.com/gitweb/?p=coreboot.git;a=blobdiff_plain;f=src%2Fsouthbridge%2Fintel%2Fbd82x6x%2Fme_8.x.c;fp=src%2Fsouthbridge%2Fintel%2Fbd82x6x%2Fme_8.x.c;h=1bb2ce38dc151dc39ae5f2d59c73a918ffdf236b;hp=0000000000000000000000000000000000000000;hb=8e073829ec69ee89b3e91f4c040c96988084a526;hpb=cb91e1525eb0b81f9bc2e24e3404d6a9efc1cce3 diff --git a/src/southbridge/intel/bd82x6x/me_8.x.c b/src/southbridge/intel/bd82x6x/me_8.x.c new file mode 100644 index 000000000..1bb2ce38d --- /dev/null +++ b/src/southbridge/intel/bd82x6x/me_8.x.c @@ -0,0 +1,919 @@ +/* + * This file is part of the coreboot project. + * + * Copyright (C) 2011 The Chromium OS Authors. All rights reserved. + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public License as + * published by the Free Software Foundation; version 2 of + * the License. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, + * MA 02110-1301 USA + */ + +/* + * This is a ramstage driver for the Intel Management Engine found in the + * 6-series chipset. It handles the required boot-time messages over the + * MMIO-based Management Engine Interface to tell the ME that the BIOS is + * finished with POST. Additional messages are defined for debug but are + * not used unless the console loglevel is high enough. + */ + +#include +#include +#include +#include +#include +#include +#include +#include + +#ifdef __SMM__ +# include +# include +#else +# include +# include +#endif + +#include "me.h" +#include "pch.h" + +#if CONFIG_CHROMEOS +#include +#endif + +#ifndef __SMM__ +/* Path that the BIOS should take based on ME state */ +static const char *me_bios_path_values[] = { + [ME_NORMAL_BIOS_PATH] = "Normal", + [ME_S3WAKE_BIOS_PATH] = "S3 Wake", + [ME_ERROR_BIOS_PATH] = "Error", + [ME_RECOVERY_BIOS_PATH] = "Recovery", + [ME_DISABLE_BIOS_PATH] = "Disable", + [ME_FIRMWARE_UPDATE_BIOS_PATH] = "Firmware Update", +}; +static int intel_me_read_mbp(me_bios_payload *mbp_data); +#endif + +/* MMIO base address for MEI interface */ +static u32 mei_base_address; + +#if CONFIG_DEBUG_INTEL_ME +static void mei_dump(void *ptr, int dword, int offset, const char *type) +{ + struct mei_csr *csr; + + printk(BIOS_SPEW, "%-9s[%02x] : ", type, offset); + + switch (offset) { + case MEI_H_CSR: + case MEI_ME_CSR_HA: + csr = ptr; + if (!csr) { + printk(BIOS_SPEW, "ERROR: 0x%08x\n", dword); + break; + } + printk(BIOS_SPEW, "cbd=%u cbrp=%02u cbwp=%02u ready=%u " + "reset=%u ig=%u is=%u ie=%u\n", csr->buffer_depth, + csr->buffer_read_ptr, csr->buffer_write_ptr, + csr->ready, csr->reset, csr->interrupt_generate, + csr->interrupt_status, csr->interrupt_enable); + break; + case MEI_ME_CB_RW: + case MEI_H_CB_WW: + printk(BIOS_SPEW, "CB: 0x%08x\n", dword); + break; + default: + printk(BIOS_SPEW, "0x%08x\n", offset); + break; + } +} +#else +# define mei_dump(ptr,dword,offset,type) do {} while (0) +#endif + +/* + * ME/MEI access helpers using memcpy to avoid aliasing. + */ + +static inline void mei_read_dword_ptr(void *ptr, int offset) +{ + u32 dword = read32(mei_base_address + offset); + memcpy(ptr, &dword, sizeof(dword)); + mei_dump(ptr, dword, offset, "READ"); +} + +static inline void mei_write_dword_ptr(void *ptr, int offset) +{ + u32 dword = 0; + memcpy(&dword, ptr, sizeof(dword)); + write32(mei_base_address + offset, dword); + mei_dump(ptr, dword, offset, "WRITE"); +} + +#ifndef __SMM__ +static inline void pci_read_dword_ptr(device_t dev, void *ptr, int offset) +{ + u32 dword = pci_read_config32(dev, offset); + memcpy(ptr, &dword, sizeof(dword)); + mei_dump(ptr, dword, offset, "PCI READ"); +} +#endif + +static inline void read_host_csr(struct mei_csr *csr) +{ + mei_read_dword_ptr(csr, MEI_H_CSR); +} + +static inline void write_host_csr(struct mei_csr *csr) +{ + mei_write_dword_ptr(csr, MEI_H_CSR); +} + +static inline void read_me_csr(struct mei_csr *csr) +{ + mei_read_dword_ptr(csr, MEI_ME_CSR_HA); +} + +static inline void write_cb(u32 dword) +{ + write32(mei_base_address + MEI_H_CB_WW, dword); + mei_dump(NULL, dword, MEI_H_CB_WW, "WRITE"); +} + +static inline u32 read_cb(void) +{ + u32 dword = read32(mei_base_address + MEI_ME_CB_RW); + mei_dump(NULL, dword, MEI_ME_CB_RW, "READ"); + return dword; +} + +/* Wait for ME ready bit to be asserted */ +static int mei_wait_for_me_ready(void) +{ + struct mei_csr me; + unsigned try = ME_RETRY; + + while (try--) { + read_me_csr(&me); + if (me.ready) + return 0; + udelay(ME_DELAY); + } + + printk(BIOS_ERR, "ME: failed to become ready\n"); + return -1; +} + +static void mei_reset(void) +{ + struct mei_csr host; + + if (mei_wait_for_me_ready() < 0) + return; + + /* Reset host and ME circular buffers for next message */ + read_host_csr(&host); + host.reset = 1; + host.interrupt_generate = 1; + write_host_csr(&host); + + if (mei_wait_for_me_ready() < 0) + return; + + /* Re-init and indicate host is ready */ + read_host_csr(&host); + host.interrupt_generate = 1; + host.ready = 1; + host.reset = 0; + write_host_csr(&host); +} + +static int mei_send_msg(struct mei_header *mei, struct mkhi_header *mkhi, + void *req_data) +{ + struct mei_csr host; + unsigned ndata, n; + u32 *data; + + /* Number of dwords to write, ignoring MKHI */ + ndata = mei->length >> 2; + + /* Pad non-dword aligned request message length */ + if (mei->length & 3) + ndata++; + if (!ndata) { + printk(BIOS_DEBUG, "ME: request does not include MKHI\n"); + return -1; + } + ndata++; /* Add MEI header */ + + /* + * Make sure there is still room left in the circular buffer. + * Reset the buffer pointers if the requested message will not fit. + */ + read_host_csr(&host); + if ((host.buffer_depth - host.buffer_write_ptr) < ndata) { + printk(BIOS_ERR, "ME: circular buffer full, resetting...\n"); + mei_reset(); + read_host_csr(&host); + } + + /* + * This implementation does not handle splitting large messages + * across multiple transactions. Ensure the requested length + * will fit in the available circular buffer depth. + */ + if ((host.buffer_depth - host.buffer_write_ptr) < ndata) { + printk(BIOS_ERR, "ME: message (%u) too large for buffer (%u)\n", + ndata + 2, host.buffer_depth); + return -1; + } + + /* Write MEI header */ + mei_write_dword_ptr(mei, MEI_H_CB_WW); + ndata--; + + /* Write MKHI header */ + mei_write_dword_ptr(mkhi, MEI_H_CB_WW); + ndata--; + + /* Write message data */ + data = req_data; + for (n = 0; n < ndata; ++n) + write_cb(*data++); + + /* Generate interrupt to the ME */ + read_host_csr(&host); + host.interrupt_generate = 1; + write_host_csr(&host); + + /* Make sure ME is ready after sending request data */ + return mei_wait_for_me_ready(); +} + +static int mei_recv_msg(struct mkhi_header *mkhi, + void *rsp_data, int rsp_bytes) +{ + struct mei_header mei_rsp; + struct mkhi_header mkhi_rsp; + struct mei_csr me, host; + unsigned ndata, n/*, me_data_len*/; + unsigned expected; + u32 *data; + + /* Total number of dwords to read from circular buffer */ + expected = (rsp_bytes + sizeof(mei_rsp) + sizeof(mkhi_rsp)) >> 2; + if (rsp_bytes & 3) + expected++; + + /* + * The interrupt status bit does not appear to indicate that the + * message has actually been received. Instead we wait until the + * expected number of dwords are present in the circular buffer. + */ + for (n = ME_RETRY; n; --n) { + read_me_csr(&me); + if ((me.buffer_write_ptr - me.buffer_read_ptr) >= expected) + break; + udelay(ME_DELAY); + } + if (!n) { + printk(BIOS_ERR, "ME: timeout waiting for data: expected " + "%u, available %u\n", expected, + me.buffer_write_ptr - me.buffer_read_ptr); + return -1; + } + + /* Read and verify MEI response header from the ME */ + mei_read_dword_ptr(&mei_rsp, MEI_ME_CB_RW); + if (!mei_rsp.is_complete) { + printk(BIOS_ERR, "ME: response is not complete\n"); + return -1; + } + + /* Handle non-dword responses and expect at least MKHI header */ + ndata = mei_rsp.length >> 2; + if (mei_rsp.length & 3) + ndata++; + if (ndata != (expected - 1)) { + printk(BIOS_ERR, "ME: response is missing data %d != %d\n", + ndata, (expected - 1)); + return -1; + } + + /* Read and verify MKHI response header from the ME */ + mei_read_dword_ptr(&mkhi_rsp, MEI_ME_CB_RW); + if (!mkhi_rsp.is_response || + mkhi->group_id != mkhi_rsp.group_id || + mkhi->command != mkhi_rsp.command) { + printk(BIOS_ERR, "ME: invalid response, group %u ?= %u," + "command %u ?= %u, is_response %u\n", mkhi->group_id, + mkhi_rsp.group_id, mkhi->command, mkhi_rsp.command, + mkhi_rsp.is_response); + return -1; + } + ndata--; /* MKHI header has been read */ + + /* Make sure caller passed a buffer with enough space */ + if (ndata != (rsp_bytes >> 2)) { + printk(BIOS_ERR, "ME: not enough room in response buffer: " + "%u != %u\n", ndata, rsp_bytes >> 2); + return -1; + } + + /* Read response data from the circular buffer */ + data = rsp_data; + for (n = 0; n < ndata; ++n) + *data++ = read_cb(); + + /* Tell the ME that we have consumed the response */ + read_host_csr(&host); + host.interrupt_status = 1; + host.interrupt_generate = 1; + write_host_csr(&host); + + return mei_wait_for_me_ready(); +} + +static inline int mei_sendrecv(struct mei_header *mei, struct mkhi_header *mkhi, + void *req_data, void *rsp_data, int rsp_bytes) +{ + if (mei_send_msg(mei, mkhi, req_data) < 0) + return -1; + if (mei_recv_msg(mkhi, rsp_data, rsp_bytes) < 0) + return -1; + return 0; +} + +/* Send END OF POST message to the ME */ +int mkhi_end_of_post(void) +{ + struct mkhi_header mkhi = { + .group_id = MKHI_GROUP_ID_GEN, + .command = MKHI_END_OF_POST, + }; + struct mei_header mei = { + .is_complete = 1, + .host_address = MEI_HOST_ADDRESS, + .client_address = MEI_ADDRESS_MKHI, + .length = sizeof(mkhi), + }; + + u32 eop_ack; + + /* Send request and wait for response */ + printk(BIOS_NOTICE, "ME: %s\n", __FUNCTION__); + if (mei_sendrecv(&mei, &mkhi, NULL, &eop_ack, sizeof(eop_ack)) < 0) { + printk(BIOS_ERR, "ME: END OF POST message failed\n"); + return -1; + } + + printk(BIOS_INFO, "ME: END OF POST message successful (%d)\n", eop_ack); + return 0; +} + +#if (CONFIG_DEFAULT_CONSOLE_LOGLEVEL >= BIOS_DEBUG) && !defined(__SMM__) +static inline void print_cap(const char *name, int state) +{ + printk(BIOS_DEBUG, "ME Capability: %-41s : %sabled\n", + name, state ? " en" : "dis"); +} + +static void me_print_fw_version(mbp_fw_version_name *vers_name) +{ + if (!vers_name->major_version) { + printk(BIOS_ERR, "ME: mbp missing version report\n"); + return; + } + + printk(BIOS_DEBUG, "ME: found version %d.%d.%d.%d\n", + vers_name->major_version, vers_name->minor_version, + vers_name->hotfix_version, vers_name->build_version); +} + +/* Get ME Firmware Capabilities */ +static int mkhi_get_fwcaps(mefwcaps_sku *cap) +{ + u32 rule_id = 0; + struct me_fwcaps cap_msg; + struct mkhi_header mkhi = { + .group_id = MKHI_GROUP_ID_FWCAPS, + .command = MKHI_FWCAPS_GET_RULE, + }; + struct mei_header mei = { + .is_complete = 1, + .host_address = MEI_HOST_ADDRESS, + .client_address = MEI_ADDRESS_MKHI, + .length = sizeof(mkhi) + sizeof(rule_id), + }; + + /* Send request and wait for response */ + if (mei_sendrecv(&mei, &mkhi, &rule_id, &cap_msg, sizeof(cap_msg)) + < 0) { + printk(BIOS_ERR, "ME: GET FWCAPS message failed\n"); + return -1; + } + *cap = cap_msg.caps_sku; + return 0; +} + +/* Get ME Firmware Capabilities */ +static void me_print_fwcaps(mbp_fw_caps *caps_section) +{ + mefwcaps_sku *cap = &caps_section->fw_capabilities; + if (!caps_section->available) { + printk(BIOS_ERR, "ME: mbp missing fwcaps report\n"); + if (mkhi_get_fwcaps(cap)) + return; + } + + print_cap("Full Network manageability", cap->full_net); + print_cap("Regular Network manageability", cap->std_net); + print_cap("Manageability", cap->manageability); + print_cap("Small business technology", cap->small_business); + print_cap("Level III manageability", cap->l3manageability); + print_cap("IntelR Anti-Theft (AT)", cap->intel_at); + print_cap("IntelR Capability Licensing Service (CLS)", cap->intel_cls); + print_cap("IntelR Power Sharing Technology (MPC)", cap->intel_mpc); + print_cap("ICC Over Clocking", cap->icc_over_clocking); + print_cap("Protected Audio Video Path (PAVP)", cap->pavp); + print_cap("IPV6", cap->ipv6); + print_cap("KVM Remote Control (KVM)", cap->kvm); + print_cap("Outbreak Containment Heuristic (OCH)", cap->och); + print_cap("Virtual LAN (VLAN)", cap->vlan); + print_cap("TLS", cap->tls); + print_cap("Wireless LAN (WLAN)", cap->wlan); +} +#endif + +/* Tell ME to issue a global reset */ +int mkhi_global_reset(void) +{ + struct me_global_reset reset = { + .request_origin = GLOBAL_RESET_BIOS_POST, + .reset_type = CBM_RR_GLOBAL_RESET, + }; + struct mkhi_header mkhi = { + .group_id = MKHI_GROUP_ID_CBM, + .command = MKHI_GLOBAL_RESET, + }; + struct mei_header mei = { + .is_complete = 1, + .length = sizeof(mkhi) + sizeof(reset), + .host_address = MEI_HOST_ADDRESS, + .client_address = MEI_ADDRESS_MKHI, + }; + + /* Send request and wait for response */ + printk(BIOS_NOTICE, "ME: %s\n", __FUNCTION__); + if (mei_sendrecv(&mei, &mkhi, &reset, NULL, 0) < 0) { + /* No response means reset will happen shortly... */ + hlt(); + } + + /* If the ME responded it rejected the reset request */ + printk(BIOS_ERR, "ME: Global Reset failed\n"); + return -1; +} + +#ifdef __SMM__ + +void intel_me_finalize_smm(void) +{ + struct me_hfs hfs; + u32 reg32; + + mei_base_address = + pcie_read_config32(PCH_ME_DEV, PCI_BASE_ADDRESS_0) & ~0xf; + + /* S3 path will have hidden this device already */ + if (!mei_base_address || mei_base_address == 0xfffffff0) + return; + + /* Make sure ME is in a mode that expects EOP */ + reg32 = pcie_read_config32(PCH_ME_DEV, PCI_ME_HFS); + memcpy(&hfs, ®32, sizeof(u32)); + + /* Abort and leave device alone if not normal mode */ + if (hfs.fpt_bad || + hfs.working_state != ME_HFS_CWS_NORMAL || + hfs.operation_mode != ME_HFS_MODE_NORMAL) + return; + + /* Try to send EOP command so ME stops accepting other commands */ + mkhi_end_of_post(); + + /* Make sure IO is disabled */ + reg32 = pcie_read_config32(PCH_ME_DEV, PCI_COMMAND); + reg32 &= ~(PCI_COMMAND_MASTER | + PCI_COMMAND_MEMORY | PCI_COMMAND_IO); + pcie_write_config32(PCH_ME_DEV, PCI_COMMAND, reg32); + + /* Hide the PCI device */ + RCBA32_OR(FD2, PCH_DISABLE_MEI1); +} + +#else /* !__SMM__ */ + +/* Determine the path that we should take based on ME status */ +static me_bios_path intel_me_path(device_t dev) +{ + me_bios_path path = ME_DISABLE_BIOS_PATH; + struct me_hfs hfs; + struct me_gmes gmes; + +#if CONFIG_HAVE_ACPI_RESUME + /* S3 wake skips all MKHI messages */ + if (acpi_slp_type == 3) { + return ME_S3WAKE_BIOS_PATH; + } +#endif + + pci_read_dword_ptr(dev, &hfs, PCI_ME_HFS); + pci_read_dword_ptr(dev, &gmes, PCI_ME_GMES); + + /* Check and dump status */ + intel_me_status(&hfs, &gmes); + + /* Check for valid firmware */ + if (hfs.fpt_bad) + return ME_ERROR_BIOS_PATH; + + /* Check Current Working State */ + switch (hfs.working_state) { + case ME_HFS_CWS_NORMAL: + path = ME_NORMAL_BIOS_PATH; + /* check if the MBP is ready */ + if (!gmes.mbp_rdy) { + printk(BIOS_CRIT, "%s: mbp is not ready!\n", + __FUNCTION__); + return ME_ERROR_BIOS_PATH; + } + break; + case ME_HFS_CWS_REC: + path = ME_RECOVERY_BIOS_PATH; + break; + default: + path = ME_DISABLE_BIOS_PATH; + break; + } + + /* Check Current Operation Mode */ + switch (hfs.operation_mode) { + case ME_HFS_MODE_NORMAL: + break; + case ME_HFS_MODE_DEBUG: + case ME_HFS_MODE_DIS: + case ME_HFS_MODE_OVER_JMPR: + case ME_HFS_MODE_OVER_MEI: + default: + path = ME_DISABLE_BIOS_PATH; + break; + } + + /* Check for any error code */ + if (hfs.error_code) + path = ME_ERROR_BIOS_PATH; + + return path; +} + +/* Prepare ME for MEI messages */ +static int intel_mei_setup(device_t dev) +{ + struct resource *res; + struct mei_csr host; + u32 reg32; + + /* Find the MMIO base for the ME interface */ + res = find_resource(dev, PCI_BASE_ADDRESS_0); + if (!res || res->base == 0 || res->size == 0) { + printk(BIOS_DEBUG, "ME: MEI resource not present!\n"); + return -1; + } + mei_base_address = res->base; + + /* Ensure Memory and Bus Master bits are set */ + reg32 = pci_read_config32(dev, PCI_COMMAND); + reg32 |= PCI_COMMAND_MASTER | PCI_COMMAND_MEMORY; + pci_write_config32(dev, PCI_COMMAND, reg32); + + /* Clean up status for next message */ + read_host_csr(&host); + host.interrupt_generate = 1; + host.ready = 1; + host.reset = 0; + write_host_csr(&host); + + return 0; +} + +/* Read the Extend register hash of ME firmware */ +static int intel_me_extend_valid(device_t dev) +{ + struct me_heres status; + u32 extend; + int i, count = 0; + + pci_read_dword_ptr(dev, &status, PCI_ME_HERES); + if (!status.extend_feature_present) { + printk(BIOS_ERR, "ME: Extend Feature not present\n"); + return -1; + } + + if (!status.extend_reg_valid) { + printk(BIOS_ERR, "ME: Extend Register not valid\n"); + return -1; + } + + switch (status.extend_reg_algorithm) { + case PCI_ME_EXT_SHA1: + count = 5; + printk(BIOS_DEBUG, "ME: Extend SHA-1: "); + break; + case PCI_ME_EXT_SHA256: + count = 8; + printk(BIOS_DEBUG, "ME: Extend SHA-256: "); + break; + default: + printk(BIOS_ERR, "ME: Extend Algorithm %d unknown\n", + status.extend_reg_algorithm); + return -1; + } + + /* + * TODO(dlaurie) Verify the hash against a saved good value. + */ + + for (i = 0; i < count; ++i) { + extend = pci_read_config32(dev, PCI_ME_HER(i)); + printk(BIOS_DEBUG, "%08x", extend); + } + printk(BIOS_DEBUG, "\n"); + + return 0; +} + +/* Hide the ME virtual PCI devices */ +static void intel_me_hide(device_t dev) +{ + dev->enabled = 0; + pch_enable(dev); +} + +/* Check whether ME is present and do basic init */ +static void intel_me_init(device_t dev) +{ + me_bios_path path = intel_me_path(dev); + me_bios_payload mbp_data; + + /* Do initial setup and determine the BIOS path */ + printk(BIOS_NOTICE, "ME: BIOS path: %s\n", me_bios_path_values[path]); + + switch (path) { + case ME_S3WAKE_BIOS_PATH: + intel_me_hide(dev); + break; + + case ME_NORMAL_BIOS_PATH: + /* Validate the extend register */ + if (intel_me_extend_valid(dev) < 0) + break; /* TODO: force recovery mode */ + + /* Prepare MEI MMIO interface */ + if (intel_mei_setup(dev) < 0) + break; + + if(intel_me_read_mbp(&mbp_data)) + break; + +#if CONFIG_CHROMEOS && 0 /* DISABLED */ + /* + * Unlock ME in recovery mode. + */ + if (recovery_mode_enabled()) { + /* Unlock ME flash region */ + mkhi_hmrfpo_enable(); + + /* Issue global reset */ + mkhi_global_reset(); + return; + } +#endif + +#if (CONFIG_DEFAULT_CONSOLE_LOGLEVEL >= BIOS_DEBUG) + me_print_fw_version(&mbp_data.fw_version_name); + me_print_fwcaps(&mbp_data.fw_caps_sku); +#endif + /* Tell ME that BIOS is done */ + mkhi_end_of_post(); + /* Hide the virtual PCI device */ + intel_me_hide(dev); + break; + + case ME_ERROR_BIOS_PATH: + case ME_RECOVERY_BIOS_PATH: + case ME_DISABLE_BIOS_PATH: + case ME_FIRMWARE_UPDATE_BIOS_PATH: + /* + * TODO(dlaurie) Force recovery mode if ME is unhappy? + */ + break; + } +} + +static void set_subsystem(device_t dev, unsigned vendor, unsigned device) +{ + if (!vendor || !device) { + pci_write_config32(dev, PCI_SUBSYSTEM_VENDOR_ID, + pci_read_config32(dev, PCI_VENDOR_ID)); + } else { + pci_write_config32(dev, PCI_SUBSYSTEM_VENDOR_ID, + ((device & 0xffff) << 16) | (vendor & 0xffff)); + } +} + +static struct pci_operations pci_ops = { + .set_subsystem = set_subsystem, +}; + +static struct device_operations device_ops = { + .read_resources = pci_dev_read_resources, + .set_resources = pci_dev_set_resources, + .enable_resources = pci_dev_enable_resources, + .init = intel_me_init, + .scan_bus = scan_static_bus, + .ops_pci = &pci_ops, +}; + +static const struct pci_driver intel_me __pci_driver = { + .ops = &device_ops, + .vendor = PCI_VENDOR_ID_INTEL, + .device = 0x1e3a, +}; + +/****************************************************************************** + * */ +static u32 me_to_host_words_pending(void) +{ + struct mei_csr me; + read_me_csr(&me); + if (!me.ready) + return 0; + return (me.buffer_write_ptr - me.buffer_read_ptr) & + (me.buffer_depth - 1); +} + +#if 0 +/* This function is not yet being used, keep it in for the future. */ +static u32 host_to_me_words_room(void) +{ + struct mei_csr csr; + + read_me_csr(&csr); + if (!csr.ready) + return 0; + + read_host_csr(&csr); + return (csr.buffer_read_ptr - csr.buffer_write_ptr - 1) & + (csr.buffer_depth - 1); +} +#endif +/* + * mbp seems to be following its own flow, let's retrieve it in a dedicated + * function. + */ +static int intel_me_read_mbp(me_bios_payload *mbp_data) +{ + mbp_header mbp_hdr; + mbp_item_header mbp_item_hdr; + u32 me2host_pending; + u32 mbp_item_id; + struct mei_csr host; + + me2host_pending = me_to_host_words_pending(); + if (!me2host_pending) { + printk(BIOS_ERR, "ME: no mbp data!\n"); + return -1; + } + + /* we know for sure that at least the header is there */ + mei_read_dword_ptr(&mbp_hdr, MEI_ME_CB_RW); + + if ((mbp_hdr.num_entries > (mbp_hdr.mbp_size / 2)) || + (me2host_pending < mbp_hdr.mbp_size)) { + printk(BIOS_ERR, "ME: mbp of %d entries, total size %d words" + " buffer contains %d words\n", + mbp_hdr.num_entries, mbp_hdr.mbp_size, + me2host_pending); + return -1; + } + + me2host_pending--; + memset(mbp_data, 0, sizeof(*mbp_data)); + + while (mbp_hdr.num_entries--) { + u32* copy_addr; + u32 copy_size, buffer_room; + void *p; + + if (!me2host_pending) { + printk(BIOS_ERR, "ME: no mbp data %d entries to go!\n", + mbp_hdr.num_entries + 1); + return -1; + } + + mei_read_dword_ptr(&mbp_item_hdr, MEI_ME_CB_RW); + + if (mbp_item_hdr.length > me2host_pending) { + printk(BIOS_ERR, "ME: insufficient mbp data %d " + "entries to go!\n", + mbp_hdr.num_entries + 1); + return -1; + } + + me2host_pending -= mbp_item_hdr.length; + + mbp_item_id = (((u32)mbp_item_hdr.item_id) << 8) + + mbp_item_hdr.app_id; + + copy_size = mbp_item_hdr.length - 1; + +#define SET_UP_COPY(field) { copy_addr = (u32 *)&mbp_data->field; \ + buffer_room = sizeof(mbp_data->field) / sizeof(u32); \ + break; \ + } + + p = &mbp_item_hdr; + printk(BIOS_INFO, "ME: MBP item header %8.8x\n", *((u32*)p)); + + switch(mbp_item_id) { + case 0x101: + SET_UP_COPY(fw_version_name); + + case 0x102: + SET_UP_COPY(icc_profile); + + case 0x103: + SET_UP_COPY(at_state); + + case 0x201: + mbp_data->fw_caps_sku.available = 1; + SET_UP_COPY(fw_caps_sku.fw_capabilities); + + case 0x301: + SET_UP_COPY(rom_bist_data); + + case 0x401: + SET_UP_COPY(platform_key); + + case 0x501: + mbp_data->fw_plat_type.available = 1; + SET_UP_COPY(fw_plat_type.rule_data); + + case 0x601: + SET_UP_COPY(mfsintegrity); + + default: + printk(BIOS_ERR, "ME: unknown mbp item id 0x%x!!!\n", + mbp_item_id); + return -1; + } + + if (buffer_room != copy_size) { + printk(BIOS_ERR, "ME: buffer room %d != %d copy size" + " for item 0x%x!!!\n", + buffer_room, copy_size, mbp_item_id); + return -1; + } + while(copy_size--) + *copy_addr++ = read_cb(); + } + + read_host_csr(&host); + host.interrupt_generate = 1; + write_host_csr(&host); + + { + int cntr = 0; + while(host.interrupt_generate) { + read_host_csr(&host); + cntr++; + } + printk(BIOS_SPEW, "ME: mbp read OK after %d cycles\n", cntr); + } + + return 0; +} + +#endif /* !__SMM__ */