From 695cca8763078c743417886e80af8ccb78bec9f2 Mon Sep 17 00:00:00 2001 From: Mike Waychison Date: Fri, 18 Mar 2011 16:50:03 -0700 Subject: firmware: Fix grammar in sysfs-firmware-dmi doc Fix the grammar in describing the position attribute of DMI entries in the dmi-sysfs module. While here, make a couple other small clarifying fixups to the docs. Reported-by: Signed-off-by: Mike Waychison Signed-off-by: Greg Kroah-Hartman --- Documentation/ABI/testing/sysfs-firmware-dmi | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-) (limited to 'Documentation/ABI') diff --git a/Documentation/ABI/testing/sysfs-firmware-dmi b/Documentation/ABI/testing/sysfs-firmware-dmi index ba9da9503c23..c78f9ab01e56 100644 --- a/Documentation/ABI/testing/sysfs-firmware-dmi +++ b/Documentation/ABI/testing/sysfs-firmware-dmi @@ -14,14 +14,15 @@ Description: DMI is structured as a large table of entries, where each entry has a common header indicating the type and - length of the entry, as well as 'handle' that is - supposed to be unique amongst all entries. + length of the entry, as well as a firmware-provided + 'handle' that is supposed to be unique amongst all + entries. Some entries are required by the specification, but many others are optional. In general though, users should never expect to find a specific entry type on their system unless they know for certain what their firmware - is doing. Machine to machine will vary. + is doing. Machine to machine experiences will vary. Multiple entries of the same type are allowed. In order to handle these duplicate entry types, each entry is @@ -67,25 +68,24 @@ Description: and the two terminating nul characters. type : The type of the entry. This value is the same as found in the directory name. It indicates - how the rest of the entry should be - interpreted. + how the rest of the entry should be interpreted. instance: The instance ordinal of the entry for the given type. This value is the same as found in the parent directory name. - position: The position of the entry within the entirety - of the entirety. + position: The ordinal position (zero-based) of the entry + within the entirety of the DMI entry table. === Entry Specialization === Some entry types may have other information available in - sysfs. + sysfs. Not all types are specialized. --- Type 15 - System Event Log --- This entry allows the firmware to export a log of events the system has taken. This information is typically backed by nvram, but the implementation - details are abstracted by this table. This entries data + details are abstracted by this table. This entry's data is exported in the directory: /sys/firmware/dmi/entries/15-0/system_event_log -- cgit v1.2.3 From 088ab0b4d855d68a0f0c16b72fb8e492a533aaa1 Mon Sep 17 00:00:00 2001 From: Ludwig Nussel Date: Mon, 28 Feb 2011 15:57:17 +0100 Subject: kernel/ksysfs.c: expose file_caps_enabled in sysfs A kernel booted with no_file_caps allows to install fscaps on a binary but doesn't actually honor the fscaps when running the binary. Userspace currently has no sane way to determine whether installing fscaps actually has any effect. Since parsing /proc/cmdline is fragile this patch exposes the current setting (1 or 0) via /sys/kernel/fscaps Signed-off-by: Ludwig Nussel Signed-off-by: Greg Kroah-Hartman --- Documentation/ABI/testing/sysfs-kernel-fscaps | 8 ++++++++ kernel/ksysfs.c | 10 ++++++++++ 2 files changed, 18 insertions(+) create mode 100644 Documentation/ABI/testing/sysfs-kernel-fscaps (limited to 'Documentation/ABI') diff --git a/Documentation/ABI/testing/sysfs-kernel-fscaps b/Documentation/ABI/testing/sysfs-kernel-fscaps new file mode 100644 index 000000000000..50a3033b5e15 --- /dev/null +++ b/Documentation/ABI/testing/sysfs-kernel-fscaps @@ -0,0 +1,8 @@ +What: /sys/kernel/fscaps +Date: February 2011 +KernelVersion: 2.6.38 +Contact: Ludwig Nussel +Description + Shows whether file system capabilities are honored + when executing a binary + diff --git a/kernel/ksysfs.c b/kernel/ksysfs.c index 0b624e791805..3b053c04dd86 100644 --- a/kernel/ksysfs.c +++ b/kernel/ksysfs.c @@ -16,6 +16,7 @@ #include #include #include +#include #define KERNEL_ATTR_RO(_name) \ static struct kobj_attribute _name##_attr = __ATTR_RO(_name) @@ -131,6 +132,14 @@ KERNEL_ATTR_RO(vmcoreinfo); #endif /* CONFIG_KEXEC */ +/* whether file capabilities are enabled */ +static ssize_t fscaps_show(struct kobject *kobj, + struct kobj_attribute *attr, char *buf) +{ + return sprintf(buf, "%d\n", file_caps_enabled); +} +KERNEL_ATTR_RO(fscaps); + /* * Make /sys/kernel/notes give the raw contents of our kernel .notes section. */ @@ -158,6 +167,7 @@ struct kobject *kernel_kobj; EXPORT_SYMBOL_GPL(kernel_kobj); static struct attribute * kernel_attrs[] = { + &fscaps_attr.attr, #if defined(CONFIG_HOTPLUG) &uevent_seqnum_attr.attr, &uevent_helper_attr.attr, -- cgit v1.2.3 From 74c5b31c6618f01079212332b2e5f6c42f2d6307 Mon Sep 17 00:00:00 2001 From: Mike Waychison Date: Fri, 29 Apr 2011 17:39:19 -0700 Subject: driver: Google EFI SMI The "gsmi" driver bridges userland with firmware specific routines for accessing hardware. Currently, this driver only supports NVRAM and eventlog information. Deprecated functions have been removed from the driver, though their op-codes are left in place so that they are not re-used. This driver works by trampolining into the firmware via the smi_command outlined in the FADT table. Three protocols are used due to various limitations over time, but all are included herein. This driver should only ever load on Google boards, identified by either a "Google, Inc." board vendor string in DMI, or "GOOGLE" in the OEM strings of the FADT ACPI table. This logic happens in gsmi_system_valid(). Signed-off-by: Duncan Laurie Signed-off-by: Aaron Durbin Signed-off-by: Mike Waychison Signed-off-by: Greg Kroah-Hartman --- Documentation/ABI/testing/sysfs-firmware-gsmi | 58 ++ drivers/firmware/Kconfig | 2 + drivers/firmware/Makefile | 2 + drivers/firmware/google/Kconfig | 9 + drivers/firmware/google/Makefile | 2 + drivers/firmware/google/gsmi.c | 940 ++++++++++++++++++++++++++ 6 files changed, 1013 insertions(+) create mode 100644 Documentation/ABI/testing/sysfs-firmware-gsmi create mode 100644 drivers/firmware/google/Kconfig create mode 100644 drivers/firmware/google/Makefile create mode 100644 drivers/firmware/google/gsmi.c (limited to 'Documentation/ABI') diff --git a/Documentation/ABI/testing/sysfs-firmware-gsmi b/Documentation/ABI/testing/sysfs-firmware-gsmi new file mode 100644 index 000000000000..0faa0aaf4b6a --- /dev/null +++ b/Documentation/ABI/testing/sysfs-firmware-gsmi @@ -0,0 +1,58 @@ +What: /sys/firmware/gsmi +Date: March 2011 +Contact: Mike Waychison +Description: + Some servers used internally at Google have firmware + that provides callback functionality via explicit SMI + triggers. Some of the callbacks are similar to those + provided by the EFI runtime services page, but due to + historical reasons this different entry-point has been + used. + + The gsmi driver implements the kernel's abstraction for + these firmware callbacks. Currently, this functionality + is limited to handling the system event log and getting + access to EFI-style variables stored in nvram. + + Layout: + + /sys/firmware/gsmi/vars: + + This directory has the same layout (and + underlying implementation as /sys/firmware/efi/vars. + See Documentation/ABI/*/sysfs-firmware-efi-vars + for more information on how to interact with + this structure. + + /sys/firmware/gsmi/append_to_eventlog - write-only: + + This file takes a binary blob and passes it onto + the firmware to be timestamped and appended to + the system eventlog. The binary format is + interpreted by the firmware and may change from + platform to platform. The only kernel-enforced + requirement is that the blob be prefixed with a + 32bit host-endian type used as part of the + firmware call. + + /sys/firmware/gsmi/clear_config - write-only: + + Writing any value to this file will cause the + entire firmware configuration to be reset to + "factory defaults". Callers should assume that + a reboot is required for the configuration to be + cleared. + + /sys/firmware/gsmi/clear_eventlog - write-only: + + This file is used to clear out a portion/the + whole of the system event log. Values written + should be values between 1 and 100 inclusive (in + ASCII) representing the fraction of the log to + clear. Not all platforms support fractional + clearing though, and this writes to this file + will error out if the firmware doesn't like your + submitted fraction. + + Callers should assume that a reboot is needed + for this operation to complete. diff --git a/drivers/firmware/Kconfig b/drivers/firmware/Kconfig index b3a25a55ba23..efba163595db 100644 --- a/drivers/firmware/Kconfig +++ b/drivers/firmware/Kconfig @@ -157,4 +157,6 @@ config SIGMA If unsure, say N here. Drivers that need these helpers will select this option automatically. +source "drivers/firmware/google/Kconfig" + endmenu diff --git a/drivers/firmware/Makefile b/drivers/firmware/Makefile index 00bb0b80a79f..d7d600992d4a 100644 --- a/drivers/firmware/Makefile +++ b/drivers/firmware/Makefile @@ -13,3 +13,5 @@ obj-$(CONFIG_ISCSI_IBFT_FIND) += iscsi_ibft_find.o obj-$(CONFIG_ISCSI_IBFT) += iscsi_ibft.o obj-$(CONFIG_FIRMWARE_MEMMAP) += memmap.o obj-$(CONFIG_SIGMA) += sigma.o + +obj-y += google/ diff --git a/drivers/firmware/google/Kconfig b/drivers/firmware/google/Kconfig new file mode 100644 index 000000000000..4a0383537f5d --- /dev/null +++ b/drivers/firmware/google/Kconfig @@ -0,0 +1,9 @@ +config GOOGLE_SMI + tristate "SMI interface for Google platforms" + depends on ACPI && DMI + select EFI_VARS + help + Say Y here if you want to enable SMI callbacks for Google + platforms. This provides an interface for writing to and + clearing the EFI event log and reading and writing NVRAM + variables. diff --git a/drivers/firmware/google/Makefile b/drivers/firmware/google/Makefile new file mode 100644 index 000000000000..fb127d7b3c7a --- /dev/null +++ b/drivers/firmware/google/Makefile @@ -0,0 +1,2 @@ + +obj-$(CONFIG_GOOGLE_SMI) += gsmi.o diff --git a/drivers/firmware/google/gsmi.c b/drivers/firmware/google/gsmi.c new file mode 100644 index 000000000000..fa7f0b3e81dd --- /dev/null +++ b/drivers/firmware/google/gsmi.c @@ -0,0 +1,940 @@ +/* + * Copyright 2010 Google Inc. All Rights Reserved. + * Author: dlaurie@google.com (Duncan Laurie) + * + * Re-worked to expose sysfs APIs by mikew@google.com (Mike Waychison) + * + * EFI SMI interface for Google platforms + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#define GSMI_SHUTDOWN_CLEAN 0 /* Clean Shutdown */ +/* TODO(mikew@google.com): Tie in HARDLOCKUP_DETECTOR with NMIWDT */ +#define GSMI_SHUTDOWN_NMIWDT 1 /* NMI Watchdog */ +#define GSMI_SHUTDOWN_PANIC 2 /* Panic */ +#define GSMI_SHUTDOWN_OOPS 3 /* Oops */ +#define GSMI_SHUTDOWN_DIE 4 /* Die -- No longer meaningful */ +#define GSMI_SHUTDOWN_MCE 5 /* Machine Check */ +#define GSMI_SHUTDOWN_SOFTWDT 6 /* Software Watchdog */ +#define GSMI_SHUTDOWN_MBE 7 /* Uncorrected ECC */ +#define GSMI_SHUTDOWN_TRIPLE 8 /* Triple Fault */ + +#define DRIVER_VERSION "1.0" +#define GSMI_GUID_SIZE 16 +#define GSMI_BUF_SIZE 1024 +#define GSMI_BUF_ALIGN sizeof(u64) +#define GSMI_CALLBACK 0xef + +/* SMI return codes */ +#define GSMI_SUCCESS 0x00 +#define GSMI_UNSUPPORTED2 0x03 +#define GSMI_LOG_FULL 0x0b +#define GSMI_VAR_NOT_FOUND 0x0e +#define GSMI_HANDSHAKE_SPIN 0x7d +#define GSMI_HANDSHAKE_CF 0x7e +#define GSMI_HANDSHAKE_NONE 0x7f +#define GSMI_INVALID_PARAMETER 0x82 +#define GSMI_UNSUPPORTED 0x83 +#define GSMI_BUFFER_TOO_SMALL 0x85 +#define GSMI_NOT_READY 0x86 +#define GSMI_DEVICE_ERROR 0x87 +#define GSMI_NOT_FOUND 0x8e + +#define QUIRKY_BOARD_HASH 0x78a30a50 + +/* Internally used commands passed to the firmware */ +#define GSMI_CMD_GET_NVRAM_VAR 0x01 +#define GSMI_CMD_GET_NEXT_VAR 0x02 +#define GSMI_CMD_SET_NVRAM_VAR 0x03 +#define GSMI_CMD_SET_EVENT_LOG 0x08 +#define GSMI_CMD_CLEAR_EVENT_LOG 0x09 +#define GSMI_CMD_CLEAR_CONFIG 0x20 +#define GSMI_CMD_HANDSHAKE_TYPE 0xC1 + +/* Magic entry type for kernel events */ +#define GSMI_LOG_ENTRY_TYPE_KERNEL 0xDEAD + +/* SMI buffers must be in 32bit physical address space */ +struct gsmi_buf { + u8 *start; /* start of buffer */ + size_t length; /* length of buffer */ + dma_addr_t handle; /* dma allocation handle */ + u32 address; /* physical address of buffer */ +}; + +struct gsmi_device { + struct platform_device *pdev; /* platform device */ + struct gsmi_buf *name_buf; /* variable name buffer */ + struct gsmi_buf *data_buf; /* generic data buffer */ + struct gsmi_buf *param_buf; /* parameter buffer */ + spinlock_t lock; /* serialize access to SMIs */ + u16 smi_cmd; /* SMI command port */ + int handshake_type; /* firmware handler interlock type */ + struct dma_pool *dma_pool; /* DMA buffer pool */ +} gsmi_dev; + +/* Packed structures for communicating with the firmware */ +struct gsmi_nvram_var_param { + efi_guid_t guid; + u32 name_ptr; + u32 attributes; + u32 data_len; + u32 data_ptr; +} __packed; + +struct gsmi_get_next_var_param { + u8 guid[GSMI_GUID_SIZE]; + u32 name_ptr; + u32 name_len; +} __packed; + +struct gsmi_set_eventlog_param { + u32 data_ptr; + u32 data_len; + u32 type; +} __packed; + +/* Event log formats */ +struct gsmi_log_entry_type_1 { + u16 type; + u32 instance; +} __packed; + + +/* + * Some platforms don't have explicit SMI handshake + * and need to wait for SMI to complete. + */ +#define GSMI_DEFAULT_SPINCOUNT 0x10000 +static unsigned int spincount = GSMI_DEFAULT_SPINCOUNT; +module_param(spincount, uint, 0600); +MODULE_PARM_DESC(spincount, + "The number of loop iterations to use when using the spin handshake."); + +static struct gsmi_buf *gsmi_buf_alloc(void) +{ + struct gsmi_buf *smibuf; + + smibuf = kzalloc(sizeof(*smibuf), GFP_KERNEL); + if (!smibuf) { + printk(KERN_ERR "gsmi: out of memory\n"); + return NULL; + } + + /* allocate buffer in 32bit address space */ + smibuf->start = dma_pool_alloc(gsmi_dev.dma_pool, GFP_KERNEL, + &smibuf->handle); + if (!smibuf->start) { + printk(KERN_ERR "gsmi: failed to allocate name buffer\n"); + kfree(smibuf); + return NULL; + } + + /* fill in the buffer handle */ + smibuf->length = GSMI_BUF_SIZE; + smibuf->address = (u32)virt_to_phys(smibuf->start); + + return smibuf; +} + +static void gsmi_buf_free(struct gsmi_buf *smibuf) +{ + if (smibuf) { + if (smibuf->start) + dma_pool_free(gsmi_dev.dma_pool, smibuf->start, + smibuf->handle); + kfree(smibuf); + } +} + +/* + * Make a call to gsmi func(sub). GSMI error codes are translated to + * in-kernel errnos (0 on success, -ERRNO on error). + */ +static int gsmi_exec(u8 func, u8 sub) +{ + u16 cmd = (sub << 8) | func; + u16 result = 0; + int rc = 0; + + /* + * AH : Subfunction number + * AL : Function number + * EBX : Parameter block address + * DX : SMI command port + * + * Three protocols here. See also the comment in gsmi_init(). + */ + if (gsmi_dev.handshake_type == GSMI_HANDSHAKE_CF) { + /* + * If handshake_type == HANDSHAKE_CF then set CF on the + * way in and wait for the handler to clear it; this avoids + * corrupting register state on those chipsets which have + * a delay between writing the SMI trigger register and + * entering SMM. + */ + asm volatile ( + "stc\n" + "outb %%al, %%dx\n" + "1: jc 1b\n" + : "=a" (result) + : "0" (cmd), + "d" (gsmi_dev.smi_cmd), + "b" (gsmi_dev.param_buf->address) + : "memory", "cc" + ); + } else if (gsmi_dev.handshake_type == GSMI_HANDSHAKE_SPIN) { + /* + * If handshake_type == HANDSHAKE_SPIN we spin a + * hundred-ish usecs to ensure the SMI has triggered. + */ + asm volatile ( + "outb %%al, %%dx\n" + "1: loop 1b\n" + : "=a" (result) + : "0" (cmd), + "d" (gsmi_dev.smi_cmd), + "b" (gsmi_dev.param_buf->address), + "c" (spincount) + : "memory", "cc" + ); + } else { + /* + * If handshake_type == HANDSHAKE_NONE we do nothing; + * either we don't need to or it's legacy firmware that + * doesn't understand the CF protocol. + */ + asm volatile ( + "outb %%al, %%dx\n\t" + : "=a" (result) + : "0" (cmd), + "d" (gsmi_dev.smi_cmd), + "b" (gsmi_dev.param_buf->address) + : "memory", "cc" + ); + } + + /* check return code from SMI handler */ + switch (result) { + case GSMI_SUCCESS: + break; + case GSMI_VAR_NOT_FOUND: + /* not really an error, but let the caller know */ + rc = 1; + break; + case GSMI_INVALID_PARAMETER: + printk(KERN_ERR "gsmi: exec 0x%04x: Invalid parameter\n", cmd); + rc = -EINVAL; + break; + case GSMI_BUFFER_TOO_SMALL: + printk(KERN_ERR "gsmi: exec 0x%04x: Buffer too small\n", cmd); + rc = -ENOMEM; + break; + case GSMI_UNSUPPORTED: + case GSMI_UNSUPPORTED2: + if (sub != GSMI_CMD_HANDSHAKE_TYPE) + printk(KERN_ERR "gsmi: exec 0x%04x: Not supported\n", + cmd); + rc = -ENOSYS; + break; + case GSMI_NOT_READY: + printk(KERN_ERR "gsmi: exec 0x%04x: Not ready\n", cmd); + rc = -EBUSY; + break; + case GSMI_DEVICE_ERROR: + printk(KERN_ERR "gsmi: exec 0x%04x: Device error\n", cmd); + rc = -EFAULT; + break; + case GSMI_NOT_FOUND: + printk(KERN_ERR "gsmi: exec 0x%04x: Data not found\n", cmd); + rc = -ENOENT; + break; + case GSMI_LOG_FULL: + printk(KERN_ERR "gsmi: exec 0x%04x: Log full\n", cmd); + rc = -ENOSPC; + break; + case GSMI_HANDSHAKE_CF: + case GSMI_HANDSHAKE_SPIN: + case GSMI_HANDSHAKE_NONE: + rc = result; + break; + default: + printk(KERN_ERR "gsmi: exec 0x%04x: Unknown error 0x%04x\n", + cmd, result); + rc = -ENXIO; + } + + return rc; +} + +/* Return the number of unicode characters in data */ +static size_t +utf16_strlen(efi_char16_t *data, unsigned long maxlength) +{ + unsigned long length = 0; + + while (*data++ != 0 && length < maxlength) + length++; + return length; +} + +static efi_status_t gsmi_get_variable(efi_char16_t *name, + efi_guid_t *vendor, u32 *attr, + unsigned long *data_size, + void *data) +{ + struct gsmi_nvram_var_param param = { + .name_ptr = gsmi_dev.name_buf->address, + .data_ptr = gsmi_dev.data_buf->address, + .data_len = (u32)*data_size, + }; + efi_status_t ret = EFI_SUCCESS; + unsigned long flags; + size_t name_len = utf16_strlen(name, GSMI_BUF_SIZE / 2); + int rc; + + if (name_len >= GSMI_BUF_SIZE / 2) + return EFI_BAD_BUFFER_SIZE; + + spin_lock_irqsave(&gsmi_dev.lock, flags); + + /* Vendor guid */ + memcpy(¶m.guid, vendor, sizeof(param.guid)); + + /* variable name, already in UTF-16 */ + memset(gsmi_dev.name_buf->start, 0, gsmi_dev.name_buf->length); + memcpy(gsmi_dev.name_buf->start, name, name_len * 2); + + /* data pointer */ + memset(gsmi_dev.data_buf->start, 0, gsmi_dev.data_buf->length); + + /* parameter buffer */ + memset(gsmi_dev.param_buf->start, 0, gsmi_dev.param_buf->length); + memcpy(gsmi_dev.param_buf->start, ¶m, sizeof(param)); + + rc = gsmi_exec(GSMI_CALLBACK, GSMI_CMD_GET_NVRAM_VAR); + if (rc < 0) { + printk(KERN_ERR "gsmi: Get Variable failed\n"); + ret = EFI_LOAD_ERROR; + } else if (rc == 1) { + /* variable was not found */ + ret = EFI_NOT_FOUND; + } else { + /* Get the arguments back */ + memcpy(¶m, gsmi_dev.param_buf->start, sizeof(param)); + + /* The size reported is the min of all of our buffers */ + *data_size = min(*data_size, gsmi_dev.data_buf->length); + *data_size = min_t(unsigned long, *data_size, param.data_len); + + /* Copy data back to return buffer. */ + memcpy(data, gsmi_dev.data_buf->start, *data_size); + + /* All variables are have the following attributes */ + *attr = EFI_VARIABLE_NON_VOLATILE | + EFI_VARIABLE_BOOTSERVICE_ACCESS | + EFI_VARIABLE_RUNTIME_ACCESS; + } + + spin_unlock_irqrestore(&gsmi_dev.lock, flags); + + return ret; +} + +static efi_status_t gsmi_get_next_variable(unsigned long *name_size, + efi_char16_t *name, + efi_guid_t *vendor) +{ + struct gsmi_get_next_var_param param = { + .name_ptr = gsmi_dev.name_buf->address, + .name_len = gsmi_dev.name_buf->length, + }; + efi_status_t ret = EFI_SUCCESS; + int rc; + unsigned long flags; + + /* For the moment, only support buffers that exactly match in size */ + if (*name_size != GSMI_BUF_SIZE) + return EFI_BAD_BUFFER_SIZE; + + /* Let's make sure the thing is at least null-terminated */ + if (utf16_strlen(name, GSMI_BUF_SIZE / 2) == GSMI_BUF_SIZE / 2) + return EFI_INVALID_PARAMETER; + + spin_lock_irqsave(&gsmi_dev.lock, flags); + + /* guid */ + memcpy(¶m.guid, vendor, sizeof(param.guid)); + + /* variable name, already in UTF-16 */ + memcpy(gsmi_dev.name_buf->start, name, *name_size); + + /* parameter buffer */ + memset(gsmi_dev.param_buf->start, 0, gsmi_dev.param_buf->length); + memcpy(gsmi_dev.param_buf->start, ¶m, sizeof(param)); + + rc = gsmi_exec(GSMI_CALLBACK, GSMI_CMD_GET_NEXT_VAR); + if (rc < 0) { + printk(KERN_ERR "gsmi: Get Next Variable Name failed\n"); + ret = EFI_LOAD_ERROR; + } else if (rc == 1) { + /* variable not found -- end of list */ + ret = EFI_NOT_FOUND; + } else { + /* copy variable data back to return buffer */ + memcpy(¶m, gsmi_dev.param_buf->start, sizeof(param)); + + /* Copy the name back */ + memcpy(name, gsmi_dev.name_buf->start, GSMI_BUF_SIZE); + *name_size = utf16_strlen(name, GSMI_BUF_SIZE / 2) * 2; + + /* copy guid to return buffer */ + memcpy(vendor, ¶m.guid, sizeof(param.guid)); + ret = EFI_SUCCESS; + } + + spin_unlock_irqrestore(&gsmi_dev.lock, flags); + + return ret; +} + +static efi_status_t gsmi_set_variable(efi_char16_t *name, + efi_guid_t *vendor, + unsigned long attr, + unsigned long data_size, + void *data) +{ + struct gsmi_nvram_var_param param = { + .name_ptr = gsmi_dev.name_buf->address, + .data_ptr = gsmi_dev.data_buf->address, + .data_len = (u32)data_size, + .attributes = EFI_VARIABLE_NON_VOLATILE | + EFI_VARIABLE_BOOTSERVICE_ACCESS | + EFI_VARIABLE_RUNTIME_ACCESS, + }; + size_t name_len = utf16_strlen(name, GSMI_BUF_SIZE / 2); + efi_status_t ret = EFI_SUCCESS; + int rc; + unsigned long flags; + + if (name_len >= GSMI_BUF_SIZE / 2) + return EFI_BAD_BUFFER_SIZE; + + spin_lock_irqsave(&gsmi_dev.lock, flags); + + /* guid */ + memcpy(¶m.guid, vendor, sizeof(param.guid)); + + /* variable name, already in UTF-16 */ + memset(gsmi_dev.name_buf->start, 0, gsmi_dev.name_buf->length); + memcpy(gsmi_dev.name_buf->start, name, name_len * 2); + + /* data pointer */ + memset(gsmi_dev.data_buf->start, 0, gsmi_dev.data_buf->length); + memcpy(gsmi_dev.data_buf->start, data, data_size); + + /* parameter buffer */ + memset(gsmi_dev.param_buf->start, 0, gsmi_dev.param_buf->length); + memcpy(gsmi_dev.param_buf->start, ¶m, sizeof(param)); + + rc = gsmi_exec(GSMI_CALLBACK, GSMI_CMD_SET_NVRAM_VAR); + if (rc < 0) { + printk(KERN_ERR "gsmi: Set Variable failed\n"); + ret = EFI_INVALID_PARAMETER; + } + + spin_unlock_irqrestore(&gsmi_dev.lock, flags); + + return ret; +} + +static const struct efivar_operations efivar_ops = { + .get_variable = gsmi_get_variable, + .set_variable = gsmi_set_variable, + .get_next_variable = gsmi_get_next_variable, +}; + +static ssize_t eventlog_write(struct file *filp, struct kobject *kobj, + struct bin_attribute *bin_attr, + char *buf, loff_t pos, size_t count) +{ + struct gsmi_set_eventlog_param param = { + .data_ptr = gsmi_dev.data_buf->address, + }; + int rc = 0; + unsigned long flags; + + /* Pull the type out */ + if (count < sizeof(u32)) + return -EINVAL; + param.type = *(u32 *)buf; + count -= sizeof(u32); + buf += sizeof(u32); + + /* The remaining buffer is the data payload */ + if (count > gsmi_dev.data_buf->length) + return -EINVAL; + param.data_len = count - sizeof(u32); + + spin_lock_irqsave(&gsmi_dev.lock, flags); + + /* data pointer */ + memset(gsmi_dev.data_buf->start, 0, gsmi_dev.data_buf->length); + memcpy(gsmi_dev.data_buf->start, buf, param.data_len); + + /* parameter buffer */ + memset(gsmi_dev.param_buf->start, 0, gsmi_dev.param_buf->length); + memcpy(gsmi_dev.param_buf->start, ¶m, sizeof(param)); + + rc = gsmi_exec(GSMI_CALLBACK, GSMI_CMD_SET_EVENT_LOG); + if (rc < 0) + printk(KERN_ERR "gsmi: Set Event Log failed\n"); + + spin_unlock_irqrestore(&gsmi_dev.lock, flags); + + return rc; + +} + +static struct bin_attribute eventlog_bin_attr = { + .attr = {.name = "append_to_eventlog", .mode = 0200}, + .write = eventlog_write, +}; + +static ssize_t gsmi_clear_eventlog_store(struct kobject *kobj, + struct kobj_attribute *attr, + const char *buf, size_t count) +{ + int rc; + unsigned long flags; + unsigned long val; + struct { + u32 percentage; + u32 data_type; + } param; + + rc = strict_strtoul(buf, 0, &val); + if (rc) + return rc; + + /* + * Value entered is a percentage, 0 through 100, anything else + * is invalid. + */ + if (val > 100) + return -EINVAL; + + /* data_type here selects the smbios event log. */ + param.percentage = val; + param.data_type = 0; + + spin_lock_irqsave(&gsmi_dev.lock, flags); + + /* parameter buffer */ + memset(gsmi_dev.param_buf->start, 0, gsmi_dev.param_buf->length); + memcpy(gsmi_dev.param_buf->start, ¶m, sizeof(param)); + + rc = gsmi_exec(GSMI_CALLBACK, GSMI_CMD_CLEAR_EVENT_LOG); + + spin_unlock_irqrestore(&gsmi_dev.lock, flags); + + if (rc) + return rc; + return count; +} + +static struct kobj_attribute gsmi_clear_eventlog_attr = { + .attr = {.name = "clear_eventlog", .mode = 0200}, + .store = gsmi_clear_eventlog_store, +}; + +static ssize_t gsmi_clear_config_store(struct kobject *kobj, + struct kobj_attribute *attr, + const char *buf, size_t count) +{ + int rc; + unsigned long flags; + + spin_lock_irqsave(&gsmi_dev.lock, flags); + + /* clear parameter buffer */ + memset(gsmi_dev.param_buf->start, 0, gsmi_dev.param_buf->length); + + rc = gsmi_exec(GSMI_CALLBACK, GSMI_CMD_CLEAR_CONFIG); + + spin_unlock_irqrestore(&gsmi_dev.lock, flags); + + if (rc) + return rc; + return count; +} + +static struct kobj_attribute gsmi_clear_config_attr = { + .attr = {.name = "clear_config", .mode = 0200}, + .store = gsmi_clear_config_store, +}; + +static const struct attribute *gsmi_attrs[] = { + &gsmi_clear_config_attr.attr, + &gsmi_clear_eventlog_attr.attr, + NULL, +}; + +static int gsmi_shutdown_reason(int reason) +{ + struct gsmi_log_entry_type_1 entry = { + .type = GSMI_LOG_ENTRY_TYPE_KERNEL, + .instance = reason, + }; + struct gsmi_set_eventlog_param param = { + .data_len = sizeof(entry), + .type = 1, + }; + static int saved_reason; + int rc = 0; + unsigned long flags; + + /* avoid duplicate entries in the log */ + if (saved_reason & (1 << reason)) + return 0; + + spin_lock_irqsave(&gsmi_dev.lock, flags); + + saved_reason |= (1 << reason); + + /* data pointer */ + memset(gsmi_dev.data_buf->start, 0, gsmi_dev.data_buf->length); + memcpy(gsmi_dev.data_buf->start, &entry, sizeof(entry)); + + /* parameter buffer */ + param.data_ptr = gsmi_dev.data_buf->address; + memset(gsmi_dev.param_buf->start, 0, gsmi_dev.param_buf->length); + memcpy(gsmi_dev.param_buf->start, ¶m, sizeof(param)); + + rc = gsmi_exec(GSMI_CALLBACK, GSMI_CMD_SET_EVENT_LOG); + + spin_unlock_irqrestore(&gsmi_dev.lock, flags); + + if (rc < 0) + printk(KERN_ERR "gsmi: Log Shutdown Reason failed\n"); + else + printk(KERN_EMERG "gsmi: Log Shutdown Reason 0x%02x\n", + reason); + + return rc; +} + +static int gsmi_reboot_callback(struct notifier_block *nb, + unsigned long reason, void *arg) +{ + gsmi_shutdown_reason(GSMI_SHUTDOWN_CLEAN); + return NOTIFY_DONE; +} + +static struct notifier_block gsmi_reboot_notifier = { + .notifier_call = gsmi_reboot_callback +}; + +static int gsmi_die_callback(struct notifier_block *nb, + unsigned long reason, void *arg) +{ + if (reason == DIE_OOPS) + gsmi_shutdown_reason(GSMI_SHUTDOWN_OOPS); + return NOTIFY_DONE; +} + +static struct notifier_block gsmi_die_notifier = { + .notifier_call = gsmi_die_callback +}; + +static int gsmi_panic_callback(struct notifier_block *nb, + unsigned long reason, void *arg) +{ + gsmi_shutdown_reason(GSMI_SHUTDOWN_PANIC); + return NOTIFY_DONE; +} + +static struct notifier_block gsmi_panic_notifier = { + .notifier_call = gsmi_panic_callback, +}; + +/* + * This hash function was blatantly copied from include/linux/hash.h. + * It is used by this driver to obfuscate a board name that requires a + * quirk within this driver. + * + * Please do not remove this copy of the function as any changes to the + * global utility hash_64() function would break this driver's ability + * to identify a board and provide the appropriate quirk -- mikew@google.com + */ +static u64 __init local_hash_64(u64 val, unsigned bits) +{ + u64 hash = val; + + /* Sigh, gcc can't optimise this alone like it does for 32 bits. */ + u64 n = hash; + n <<= 18; + hash -= n; + n <<= 33; + hash -= n; + n <<= 3; + hash += n; + n <<= 3; + hash -= n; + n <<= 4; + hash += n; + n <<= 2; + hash += n; + + /* High bits are more random, so use them. */ + return hash >> (64 - bits); +} + +static u32 __init hash_oem_table_id(char s[8]) +{ + u64 input; + memcpy(&input, s, 8); + return local_hash_64(input, 32); +} + +static struct dmi_system_id gsmi_dmi_table[] __initdata = { + { + .ident = "Google Board", + .matches = { + DMI_MATCH(DMI_BOARD_VENDOR, "Google, Inc."), + }, + }, + {} +}; +MODULE_DEVICE_TABLE(dmi, gsmi_dmi_table); + +static __init int gsmi_system_valid(void) +{ + u32 hash; + + if (!dmi_check_system(gsmi_dmi_table)) + return -ENODEV; + + /* + * Only newer firmware supports the gsmi interface. All older + * firmware that didn't support this interface used to plug the + * table name in the first four bytes of the oem_table_id field. + * Newer firmware doesn't do that though, so use that as the + * discriminant factor. We have to do this in order to + * whitewash our board names out of the public driver. + */ + if (!strncmp(acpi_gbl_FADT.header.oem_table_id, "FACP", 4)) { + printk(KERN_INFO "gsmi: Board is too old\n"); + return -ENODEV; + } + + /* Disable on board with 1.0 BIOS due to Google bug 2602657 */ + hash = hash_oem_table_id(acpi_gbl_FADT.header.oem_table_id); + if (hash == QUIRKY_BOARD_HASH) { + const char *bios_ver = dmi_get_system_info(DMI_BIOS_VERSION); + if (strncmp(bios_ver, "1.0", 3) == 0) { + pr_info("gsmi: disabled on this board's BIOS %s\n", + bios_ver); + return -ENODEV; + } + } + + /* check for valid SMI command port in ACPI FADT */ + if (acpi_gbl_FADT.smi_command == 0) { + pr_info("gsmi: missing smi_command\n"); + return -ENODEV; + } + + /* Found */ + return 0; +} + +static struct kobject *gsmi_kobj; +static struct efivars efivars; + +static __init int gsmi_init(void) +{ + unsigned long flags; + int ret; + + ret = gsmi_system_valid(); + if (ret) + return ret; + + gsmi_dev.smi_cmd = acpi_gbl_FADT.smi_command; + + /* register device */ + gsmi_dev.pdev = platform_device_register_simple("gsmi", -1, NULL, 0); + if (IS_ERR(gsmi_dev.pdev)) { + printk(KERN_ERR "gsmi: unable to register platform device\n"); + return PTR_ERR(gsmi_dev.pdev); + } + + /* SMI access needs to be serialized */ + spin_lock_init(&gsmi_dev.lock); + + /* SMI callbacks require 32bit addresses */ + gsmi_dev.pdev->dev.coherent_dma_mask = DMA_BIT_MASK(32); + gsmi_dev.pdev->dev.dma_mask = + &gsmi_dev.pdev->dev.coherent_dma_mask; + ret = -ENOMEM; + gsmi_dev.dma_pool = dma_pool_create("gsmi", &gsmi_dev.pdev->dev, + GSMI_BUF_SIZE, GSMI_BUF_ALIGN, 0); + if (!gsmi_dev.dma_pool) + goto out_err; + + /* + * pre-allocate buffers because sometimes we are called when + * this is not feasible: oops, panic, die, mce, etc + */ + gsmi_dev.name_buf = gsmi_buf_alloc(); + if (!gsmi_dev.name_buf) { + printk(KERN_ERR "gsmi: failed to allocate name buffer\n"); + goto out_err; + } + + gsmi_dev.data_buf = gsmi_buf_alloc(); + if (!gsmi_dev.data_buf) { + printk(KERN_ERR "gsmi: failed to allocate data buffer\n"); + goto out_err; + } + + gsmi_dev.param_buf = gsmi_buf_alloc(); + if (!gsmi_dev.param_buf) { + printk(KERN_ERR "gsmi: failed to allocate param buffer\n"); + goto out_err; + } + + /* + * Determine type of handshake used to serialize the SMI + * entry. See also gsmi_exec(). + * + * There's a "behavior" present on some chipsets where writing the + * SMI trigger register in the southbridge doesn't result in an + * immediate SMI. Rather, the processor can execute "a few" more + * instructions before the SMI takes effect. To ensure synchronous + * behavior, implement a handshake between the kernel driver and the + * firmware handler to spin until released. This ioctl determines + * the type of handshake. + * + * NONE: The firmware handler does not implement any + * handshake. Either it doesn't need to, or it's legacy firmware + * that doesn't know it needs to and never will. + * + * CF: The firmware handler will clear the CF in the saved + * state before returning. The driver may set the CF and test for + * it to clear before proceeding. + * + * SPIN: The firmware handler does not implement any handshake + * but the driver should spin for a hundred or so microseconds + * to ensure the SMI has triggered. + * + * Finally, the handler will return -ENOSYS if + * GSMI_CMD_HANDSHAKE_TYPE is unimplemented, which implies + * HANDSHAKE_NONE. + */ + spin_lock_irqsave(&gsmi_dev.lock, flags); + gsmi_dev.handshake_type = GSMI_HANDSHAKE_SPIN; + gsmi_dev.handshake_type = + gsmi_exec(GSMI_CALLBACK, GSMI_CMD_HANDSHAKE_TYPE); + if (gsmi_dev.handshake_type == -ENOSYS) + gsmi_dev.handshake_type = GSMI_HANDSHAKE_NONE; + spin_unlock_irqrestore(&gsmi_dev.lock, flags); + + /* Remove and clean up gsmi if the handshake could not complete. */ + if (gsmi_dev.handshake_type == -ENXIO) { + printk(KERN_INFO "gsmi version " DRIVER_VERSION + " failed to load\n"); + ret = -ENODEV; + goto out_err; + } + + printk(KERN_INFO "gsmi version " DRIVER_VERSION " loaded\n"); + + /* Register in the firmware directory */ + ret = -ENOMEM; + gsmi_kobj = kobject_create_and_add("gsmi", firmware_kobj); + if (!gsmi_kobj) { + printk(KERN_INFO "gsmi: Failed to create firmware kobj\n"); + goto out_err; + } + + /* Setup eventlog access */ + ret = sysfs_create_bin_file(gsmi_kobj, &eventlog_bin_attr); + if (ret) { + printk(KERN_INFO "gsmi: Failed to setup eventlog"); + goto out_err; + } + + /* Other attributes */ + ret = sysfs_create_files(gsmi_kobj, gsmi_attrs); + if (ret) { + printk(KERN_INFO "gsmi: Failed to add attrs"); + goto out_err; + } + + if (register_efivars(&efivars, &efivar_ops, gsmi_kobj)) { + printk(KERN_INFO "gsmi: Failed to register efivars\n"); + goto out_err; + } + + register_reboot_notifier(&gsmi_reboot_notifier); + register_die_notifier(&gsmi_die_notifier); + atomic_notifier_chain_register(&panic_notifier_list, + &gsmi_panic_notifier); + + return 0; + + out_err: + kobject_put(gsmi_kobj); + gsmi_buf_free(gsmi_dev.param_buf); + gsmi_buf_free(gsmi_dev.data_buf); + gsmi_buf_free(gsmi_dev.name_buf); + if (gsmi_dev.dma_pool) + dma_pool_destroy(gsmi_dev.dma_pool); + platform_device_unregister(gsmi_dev.pdev); + pr_info("gsmi: failed to load: %d\n", ret); + return ret; +} + +static void __exit gsmi_exit(void) +{ + unregister_reboot_notifier(&gsmi_reboot_notifier); + unregister_die_notifier(&gsmi_die_notifier); + atomic_notifier_chain_unregister(&panic_notifier_list, + &gsmi_panic_notifier); + unregister_efivars(&efivars); + + kobject_put(gsmi_kobj); + gsmi_buf_free(gsmi_dev.param_buf); + gsmi_buf_free(gsmi_dev.data_buf); + gsmi_buf_free(gsmi_dev.name_buf); + dma_pool_destroy(gsmi_dev.dma_pool); + platform_device_unregister(gsmi_dev.pdev); +} + +module_init(gsmi_init); +module_exit(gsmi_exit); + +MODULE_AUTHOR("Google, Inc."); +MODULE_LICENSE("GPL"); -- cgit v1.2.3 From e561bc45920aade3f8a5aad9058a00e750af1345 Mon Sep 17 00:00:00 2001 From: Mike Waychison Date: Fri, 29 Apr 2011 17:39:25 -0700 Subject: driver: Google Memory Console This patch introduces the 'memconsole' driver. Our firmware gives us access to an in-memory log of the firmware's output. This gives us visibility in a data-center of headless machines as to what the firmware is doing. The memory console is found by the driver by finding a header block in the EBDA. The buffer is then copied out, and is exported to userland in the file /sys/firmware/log. Signed-off-by: San Mehat Signed-off-by: Mike Waychison Signed-off-by: Greg Kroah-Hartman --- Documentation/ABI/testing/sysfs-firmware-log | 7 ++ drivers/firmware/google/Kconfig | 8 ++ drivers/firmware/google/Makefile | 1 + drivers/firmware/google/memconsole.c | 166 +++++++++++++++++++++++++++ 4 files changed, 182 insertions(+) create mode 100644 Documentation/ABI/testing/sysfs-firmware-log create mode 100644 drivers/firmware/google/memconsole.c (limited to 'Documentation/ABI') diff --git a/Documentation/ABI/testing/sysfs-firmware-log b/Documentation/ABI/testing/sysfs-firmware-log new file mode 100644 index 000000000000..9b58e7c5365f --- /dev/null +++ b/Documentation/ABI/testing/sysfs-firmware-log @@ -0,0 +1,7 @@ +What: /sys/firmware/log +Date: February 2011 +Contact: Mike Waychison +Description: + The /sys/firmware/log is a binary file that represents a + read-only copy of the firmware's log if one is + available. diff --git a/drivers/firmware/google/Kconfig b/drivers/firmware/google/Kconfig index 4a0383537f5d..640dc6bedd69 100644 --- a/drivers/firmware/google/Kconfig +++ b/drivers/firmware/google/Kconfig @@ -7,3 +7,11 @@ config GOOGLE_SMI platforms. This provides an interface for writing to and clearing the EFI event log and reading and writing NVRAM variables. + +config GOOGLE_MEMCONSOLE + tristate "Firmware Memory Console" + depends on DMI + help + This option enables the kernel to search for a firmware log in + the EBDA on Google servers. If found, this log is exported to + userland in the file /sys/firmware/log. diff --git a/drivers/firmware/google/Makefile b/drivers/firmware/google/Makefile index fb127d7b3c7a..54a294e3cb61 100644 --- a/drivers/firmware/google/Makefile +++ b/drivers/firmware/google/Makefile @@ -1,2 +1,3 @@ obj-$(CONFIG_GOOGLE_SMI) += gsmi.o +obj-$(CONFIG_GOOGLE_MEMCONSOLE) += memconsole.o diff --git a/drivers/firmware/google/memconsole.c b/drivers/firmware/google/memconsole.c new file mode 100644 index 000000000000..2a90ba613613 --- /dev/null +++ b/drivers/firmware/google/memconsole.c @@ -0,0 +1,166 @@ +/* + * memconsole.c + * + * Infrastructure for importing the BIOS memory based console + * into the kernel log ringbuffer. + * + * Copyright 2010 Google Inc. All rights reserved. + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#define BIOS_MEMCONSOLE_V1_MAGIC 0xDEADBABE +#define BIOS_MEMCONSOLE_V2_MAGIC (('M')|('C'<<8)|('O'<<16)|('N'<<24)) + +struct biosmemcon_ebda { + u32 signature; + union { + struct { + u8 enabled; + u32 buffer_addr; + u16 start; + u16 end; + u16 num_chars; + u8 wrapped; + } __packed v1; + struct { + u32 buffer_addr; + /* Misdocumented as number of pages! */ + u16 num_bytes; + u16 start; + u16 end; + } __packed v2; + }; +} __packed; + +static char *memconsole_baseaddr; +static size_t memconsole_length; + +static ssize_t memconsole_read(struct file *filp, struct kobject *kobp, + struct bin_attribute *bin_attr, char *buf, + loff_t pos, size_t count) +{ + return memory_read_from_buffer(buf, count, &pos, memconsole_baseaddr, + memconsole_length); +} + +static struct bin_attribute memconsole_bin_attr = { + .attr = {.name = "log", .mode = 0444}, + .read = memconsole_read, +}; + + +static void found_v1_header(struct biosmemcon_ebda *hdr) +{ + printk(KERN_INFO "BIOS console v1 EBDA structure found at %p\n", hdr); + printk(KERN_INFO "BIOS console buffer at 0x%.8x, " + "start = %d, end = %d, num = %d\n", + hdr->v1.buffer_addr, hdr->v1.start, + hdr->v1.end, hdr->v1.num_chars); + + memconsole_length = hdr->v1.num_chars; + memconsole_baseaddr = phys_to_virt(hdr->v1.buffer_addr); +} + +static void found_v2_header(struct biosmemcon_ebda *hdr) +{ + printk(KERN_INFO "BIOS console v2 EBDA structure found at %p\n", hdr); + printk(KERN_INFO "BIOS console buffer at 0x%.8x, " + "start = %d, end = %d, num_bytes = %d\n", + hdr->v2.buffer_addr, hdr->v2.start, + hdr->v2.end, hdr->v2.num_bytes); + + memconsole_length = hdr->v2.end - hdr->v2.start; + memconsole_baseaddr = phys_to_virt(hdr->v2.buffer_addr + + hdr->v2.start); +} + +/* + * Search through the EBDA for the BIOS Memory Console, and + * set the global variables to point to it. Return true if found. + */ +static bool found_memconsole(void) +{ + unsigned int address; + size_t length, cur; + + address = get_bios_ebda(); + if (!address) { + printk(KERN_INFO "BIOS EBDA non-existent.\n"); + return false; + } + + /* EBDA length is byte 0 of EBDA (in KB) */ + length = *(u8 *)phys_to_virt(address); + length <<= 10; /* convert to bytes */ + + /* + * Search through EBDA for BIOS memory console structure + * note: signature is not necessarily dword-aligned + */ + for (cur = 0; cur < length; cur++) { + struct biosmemcon_ebda *hdr = phys_to_virt(address + cur); + + /* memconsole v1 */ + if (hdr->signature == BIOS_MEMCONSOLE_V1_MAGIC) { + found_v1_header(hdr); + return true; + } + + /* memconsole v2 */ + if (hdr->signature == BIOS_MEMCONSOLE_V2_MAGIC) { + found_v2_header(hdr); + return true; + } + } + + printk(KERN_INFO "BIOS console EBDA structure not found!\n"); + return false; +} + +static struct dmi_system_id memconsole_dmi_table[] __initdata = { + { + .ident = "Google Board", + .matches = { + DMI_MATCH(DMI_BOARD_VENDOR, "Google, Inc."), + }, + }, + {} +}; +MODULE_DEVICE_TABLE(dmi, memconsole_dmi_table); + +static int __init memconsole_init(void) +{ + int ret; + + if (!dmi_check_system(memconsole_dmi_table)) + return -ENODEV; + + if (!found_memconsole()) + return -ENODEV; + + memconsole_bin_attr.size = memconsole_length; + + ret = sysfs_create_bin_file(firmware_kobj, &memconsole_bin_attr); + + return ret; +} + +static void __exit memconsole_exit(void) +{ + sysfs_remove_bin_file(firmware_kobj, &memconsole_bin_attr); +} + +module_init(memconsole_init); +module_exit(memconsole_exit); + +MODULE_AUTHOR("Google, Inc."); +MODULE_LICENSE("GPL"); -- cgit v1.2.3 From eecaaba5b2e4ae762b4726fae2e3b22630e137ec Mon Sep 17 00:00:00 2001 From: Borislav Petkov Date: Mon, 16 May 2011 15:39:48 +0200 Subject: Documentation, ABI: Update L3 cache index disable text Change contact person to AMD kernel mailing list, update text and external references, drop "Users:" tag. Cc: Randy Dunlap Signed-off-by: Borislav Petkov Link: http://lkml.kernel.org/r/1305553188-21061-4-git-send-email-bp@amd64.org Acked-by: Greg Kroah-Hartman Signed-off-by: H. Peter Anvin --- Documentation/ABI/testing/sysfs-devices-system-cpu | 34 +++++++++++----------- 1 file changed, 17 insertions(+), 17 deletions(-) (limited to 'Documentation/ABI') diff --git a/Documentation/ABI/testing/sysfs-devices-system-cpu b/Documentation/ABI/testing/sysfs-devices-system-cpu index 7564e88bfa43..e7be75b96e4b 100644 --- a/Documentation/ABI/testing/sysfs-devices-system-cpu +++ b/Documentation/ABI/testing/sysfs-devices-system-cpu @@ -183,21 +183,21 @@ Description: Discover and change clock speed of CPUs to learn how to control the knobs. -What: /sys/devices/system/cpu/cpu*/cache/index*/cache_disable_X -Date: August 2008 +What: /sys/devices/system/cpu/cpu*/cache/index3/cache_disable_{0,1} +Date: August 2008 KernelVersion: 2.6.27 -Contact: mark.langsdorf@amd.com -Description: These files exist in every cpu's cache index directories. - There are currently 2 cache_disable_# files in each - directory. Reading from these files on a supported - processor will return that cache disable index value - for that processor and node. Writing to one of these - files will cause the specificed cache index to be disabled. - - Currently, only AMD Family 10h Processors support cache index - disable, and only for their L3 caches. See the BIOS and - Kernel Developer's Guide at - http://support.amd.com/us/Embedded_TechDocs/31116-Public-GH-BKDG_3-28_5-28-09.pdf - for formatting information and other details on the - cache index disable. -Users: joachim.deguara@amd.com +Contact: discuss@x86-64.org +Description: Disable L3 cache indices + + These files exist in every CPU's cache/index3 directory. Each + cache_disable_{0,1} file corresponds to one disable slot which + can be used to disable a cache index. Reading from these files + on a processor with this functionality will return the currently + disabled index for that node. There is one L3 structure per + node, or per internal node on MCM machines. Writing a valid + index to one of these files will cause the specificed cache + index to be disabled. + + All AMD processors with L3 caches provide this functionality. + For details, see BKDGs at + http://developer.amd.com/documentation/guides/Pages/default.aspx -- cgit v1.2.3 From ddeb648708108091a641adad0a438ec4fd8bf190 Mon Sep 17 00:00:00 2001 From: "Rafael J. Wysocki" Date: Sun, 15 May 2011 11:38:48 +0200 Subject: PM / Hibernate: Add sysfs knob to control size of memory for drivers Martin reports that on his system hibernation occasionally fails due to the lack of memory, because the radeon driver apparently allocates too much of it during the device freeze stage. It turns out that the amount of memory allocated by radeon during hibernation (and presumably during system suspend too) depends on the utilization of the GPU (e.g. hibernating while there are two KDE 4 sessions with compositing enabled causes radeon to allocate more memory than for one KDE 4 session). In principle it should be possible to use image_size to make the memory preallocation mechanism free enough memory for the radeon driver, but in practice it is not easy to guess the right value because of the way the preallocation code uses image_size. For this reason, it seems reasonable to allow users to control the amount of memory reserved for driver allocations made after the hibernate preallocation, which currently is constant and amounts to 1 MB. Introduce a new sysfs file, /sys/power/reserved_size, whose value will be used as the amount of memory to reserve for the post-preallocation reservations made by device drivers, in bytes. For backwards compatibility, set its default (and initial) value to the currently used number (1 MB). References: https://bugzilla.kernel.org/show_bug.cgi?id=34102 Reported-and-tested-by: Martin Steigerwald Signed-off-by: Rafael J. Wysocki --- Documentation/ABI/testing/sysfs-power | 14 ++++++++++++++ kernel/power/hibernate.c | 23 +++++++++++++++++++++++ kernel/power/main.c | 1 + kernel/power/power.h | 4 ++++ kernel/power/snapshot.c | 25 ++++++++++++++++++++----- 5 files changed, 62 insertions(+), 5 deletions(-) (limited to 'Documentation/ABI') diff --git a/Documentation/ABI/testing/sysfs-power b/Documentation/ABI/testing/sysfs-power index 194ca446ac28..b464d12761ba 100644 --- a/Documentation/ABI/testing/sysfs-power +++ b/Documentation/ABI/testing/sysfs-power @@ -158,3 +158,17 @@ Description: successful, will make the kernel abort a subsequent transition to a sleep state if any wakeup events are reported after the write has returned. + +What: /sys/power/reserved_size +Date: May 2011 +Contact: Rafael J. Wysocki +Description: + The /sys/power/reserved_size file allows user space to control + the amount of memory reserved for allocations made by device + drivers during the "device freeze" stage of hibernation. It can + be written a string representing a non-negative integer that + will be used as the amount of memory to reserve for allocations + made by device drivers' "freeze" callbacks, in bytes. + + Reading from this file will display the current value, which is + set to 1 MB by default. diff --git a/kernel/power/hibernate.c b/kernel/power/hibernate.c index 50aae660174d..431721313b71 100644 --- a/kernel/power/hibernate.c +++ b/kernel/power/hibernate.c @@ -982,10 +982,33 @@ static ssize_t image_size_store(struct kobject *kobj, struct kobj_attribute *att power_attr(image_size); +static ssize_t reserved_size_show(struct kobject *kobj, + struct kobj_attribute *attr, char *buf) +{ + return sprintf(buf, "%lu\n", reserved_size); +} + +static ssize_t reserved_size_store(struct kobject *kobj, + struct kobj_attribute *attr, + const char *buf, size_t n) +{ + unsigned long size; + + if (sscanf(buf, "%lu", &size) == 1) { + reserved_size = size; + return n; + } + + return -EINVAL; +} + +power_attr(reserved_size); + static struct attribute * g[] = { &disk_attr.attr, &resume_attr.attr, &image_size_attr.attr, + &reserved_size_attr.attr, NULL, }; diff --git a/kernel/power/main.c b/kernel/power/main.c index de9aef8742f4..2981af4ce7cb 100644 --- a/kernel/power/main.c +++ b/kernel/power/main.c @@ -337,6 +337,7 @@ static int __init pm_init(void) if (error) return error; hibernate_image_size_init(); + hibernate_reserved_size_init(); power_kobj = kobject_create_and_add("power", NULL); if (!power_kobj) return -ENOMEM; diff --git a/kernel/power/power.h b/kernel/power/power.h index 03634be55f62..9a00a0a26280 100644 --- a/kernel/power/power.h +++ b/kernel/power/power.h @@ -15,6 +15,7 @@ struct swsusp_info { #ifdef CONFIG_HIBERNATION /* kernel/power/snapshot.c */ +extern void __init hibernate_reserved_size_init(void); extern void __init hibernate_image_size_init(void); #ifdef CONFIG_ARCH_HIBERNATION_HEADER @@ -55,6 +56,7 @@ extern int hibernation_platform_enter(void); #else /* !CONFIG_HIBERNATION */ +static inline void hibernate_reserved_size_init(void) {} static inline void hibernate_image_size_init(void) {} #endif /* !CONFIG_HIBERNATION */ @@ -72,6 +74,8 @@ static struct kobj_attribute _name##_attr = { \ /* Preferred image size in bytes (default 500 MB) */ extern unsigned long image_size; +/* Size of memory reserved for drivers (default SPARE_PAGES x PAGE_SIZE) */ +extern unsigned long reserved_size; extern int in_suspend; extern dev_t swsusp_resume_device; extern sector_t swsusp_resume_block; diff --git a/kernel/power/snapshot.c b/kernel/power/snapshot.c index ca0aacc24874..d69e3323a85d 100644 --- a/kernel/power/snapshot.c +++ b/kernel/power/snapshot.c @@ -40,6 +40,18 @@ static int swsusp_page_is_free(struct page *); static void swsusp_set_page_forbidden(struct page *); static void swsusp_unset_page_forbidden(struct page *); +/* + * Number of bytes to reserve for memory allocations made by device drivers + * from their ->freeze() and ->freeze_noirq() callbacks so that they don't + * cause image creation to fail (tunable via /sys/power/reserved_size). + */ +unsigned long reserved_size; + +void __init hibernate_reserved_size_init(void) +{ + reserved_size = SPARE_PAGES * PAGE_SIZE; +} + /* * Preferred image size in bytes (tunable via /sys/power/image_size). * When it is set to N, the image creating code will do its best to @@ -1263,11 +1275,13 @@ static unsigned long minimum_image_size(unsigned long saveable) * frame in use. We also need a number of page frames to be free during * hibernation for allocations made while saving the image and for device * drivers, in case they need to allocate memory from their hibernation - * callbacks (these two numbers are given by PAGES_FOR_IO and SPARE_PAGES, - * respectively, both of which are rough estimates). To make this happen, we - * compute the total number of available page frames and allocate at least + * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough + * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through + * /sys/power/reserved_size, respectively). To make this happen, we compute the + * total number of available page frames and allocate at least * - * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2 + 2 * SPARE_PAGES + * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2 + * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE) * * of them, which corresponds to the maximum size of a hibernation image. * @@ -1322,7 +1336,8 @@ int hibernate_preallocate_memory(void) count -= totalreserve_pages; /* Compute the maximum number of saveable pages to leave in memory. */ - max_size = (count - (size + PAGES_FOR_IO)) / 2 - 2 * SPARE_PAGES; + max_size = (count - (size + PAGES_FOR_IO)) / 2 + - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE); /* Compute the desired number of image pages specified by image_size. */ size = DIV_ROUND_UP(image_size, PAGE_SIZE); if (size > max_size) -- cgit v1.2.3