Currently, the refill path for RX buffers will always allocate the
buffers as GFP_ATOMIC, even if we are in process context. This will
fail to apply memory pressure as the worker thread will not contribute
to the freeing of memory.
Fix this by changing add_recvbuf_small to use the gfp variant allocator,
__netdev_alloc_skb_ip_align().
Signed-off-by: Mike Waychison <[email protected]>
Acked-by: Rusty Russell <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Commit 0945d4b228 tried to fix the get_msr path for the
HV_X64_MSR_APIC_ASSIST_PAGE msr, but was poorly tested. We should be
returning 0 if the read succeeded, and passing the value back to the
caller via the pdata out argument, not returning the value directly.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>
"get" support for the HV_X64_MSR_APIC_ASSIST_PAGE msr was missing, even
though it is explicitly enumerated as something the vmm should save in
msrs_to_save and reported to userland via the KVM_GET_MSR_INDEX_LIST
ioctl.
Add "get" support for HV_X64_MSR_APIC_ASSIST_PAGE. We simply return the
guest visible value of this register, which seems to be correct as a set
on the register is validated for us already.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Marcelo Tosatti <[email protected]>
Even though the Linux kernel does not use the tm_isdst field, it is
exposed as part of the ABI. This field can accidentally be left
initialized, which is why we currently memset buffers returned to
userland in rtc_read_time.
There is a case however where the field can return garbage from the
stack though when using the RTC_ALM_READ ioctl on the rtc device. This
ioctl invokes rtc_read_alarm, which is careful to memset the rtc_wkalrm
buffer that is copied to userland, but it then uses a struct copy to
assign to alarm->time given the return value from rtc_ktime_to_tm().
rtc_ktime_to_tm() is implemented by calling rtc_time_to_tm using a
derivative seconds counds from ktime, but rtc_time_to_tm does not assign
a value to ->tm_isdst. This results in garbage from rtc_ktime_to_tm()'s
frame ending up being copied out to userland as part of the returned
rtc_wkalrm.
Fix this by initializing rtc_time->tm_isdst to 0 in rtc_time_to_tm.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: John Stultz <[email protected]>
Using a gcc 4.4.3, warnings are emitted for a possibly uninitialized use
of ecn_ok.
This can happen if cookie_check_timestamp() returns due to not having
seen a timestamp. Defaulting to ecn off seems like a reasonable thing
to do in this case, so initialized ecn_ok to false.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Consolidate the attributes listed for pstore operations in one place,
PSTORE_EFI_ATTRIBUTES.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Tony Luck <[email protected]>
Instead of open-coding the string operations for comparing the prefix of
the variable names, use the provided utf16_* string functions.
This patch also changes the calls to efi.set_variable to
efivars->ops->set_variable so that the right function gets called in the
case of gsmi (which doesn't have a valid efi structure).
As well, make sure that we only consider variables with the right vendor
string.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Tony Luck <[email protected]>
Introduce utf16_strncmp which is used in the next patch. Semantics
should be the same as the strncmp C function.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Tony Luck <[email protected]>
Fix the string functions in the efivars driver to be called utf16_*
instead of utf8_* as the encoding is utf16, not utf8.
As well, rename utf16_strlen to utf16_strnlen as it takes a maxlength
argument and the name should be consistent with the standard C function
names. utf16_strlen is still provided for convenience in a subsequent
patch.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Tony Luck <[email protected]>
It doesn't make sense to ever see a half-initialized kvm structure on
mmu notifier callbacks. Previously, 85722cda changed the ordering to
ensure that the mmu_lock was initialized before mmu notifier
registration, but there is still a race where the mmu notifier could
come in and try accessing other portions of struct kvm before they are
intialized.
Solve this by moving the mmu notifier registration to occur after the
structure is completely initialized.
Google-Bug-Id: 452199
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Avi Kivity <[email protected]>
In order to keep Google's firmware drivers organized amongst themselves,
all Google firmware drivers are gated on CONFIG_GOOGLE_FIRMWARE=y, which
defaults to 'n' in the kernel build.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
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 <[email protected]>
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
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 <[email protected]>
Signed-off-by: Aaron Durbin <[email protected]>
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
Make the comments a bit clearer for get_bios_ebda so that it actually
tells us what it is returning.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
Add a wrapper routine that tells us the length of the EBDA if it is
present. This guy also ensures that the returned length doesn't let the
EBDA run past the 640KiB mark.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
When SND_HDA_NEEDS_RESUME is not defined, the compiler identifies that
the following symbols are static but not used:
restore_shutup_pins
hda_cleanup_all_streams
Fix warnings by adding SND_HDA_NEEDS_RESUME guards.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Takashi Iwai <[email protected]>
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: <[email protected]>
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
While changing our build system over to use the headers_install target
as part of our klibc build, the following message started showing up in
our logs:
make[2]: `scripts/unifdef' is up to date.
It turns out that the build blindly invokes a recursive make on this
target, which causes make to emit this message when the target is
already up to date. This isn't seen for most targets as the rest of the
build relies primarily on the default target and on PHONY targets when
invoking make recursively.
Silence the above message when building unifdef as part of
headers_install by hiding it behind a new PHONY target called
"build_unifdef" that has an empty recipe.
Signed-off-by: Mike Waychison <[email protected]>
Acked-by: WANG Cong <[email protected]>
Signed-off-by: Michal Marek <[email protected]>
This patch documents the interface exposed by the 'efivars' module.
Signed-off-by: Mike Waychison <[email protected]>
Cc: Matt Domsch <[email protected]>,
Signed-off-by: Greg Kroah-Hartman <[email protected]>
Instead of letting efivars access struct efi directly when dealing with
variables, use an operations structure. This allows a later change to
reuse the efivars logic without having to pretend to support everything
in struct efi.
Signed-off-by: Mike Waychison <[email protected]>
Cc: Matt Domsch <[email protected]>,
Signed-off-by: Greg Kroah-Hartman <[email protected]>
In anticipation of re-using the variable facilities in efivars from
elsewhere, split out the registration and unregistration of struct
efivars from the rest of the EFI specific sysfs code.
Signed-off-by: Mike Waychison <[email protected]>
Cc: Matt Domsch <[email protected]>,
Signed-off-by: Greg Kroah-Hartman <[email protected]>
Now that we all global variable state is encapsulated by struct efivars,
parameterize all functions to the efivars local to the control flow rather
than at file scope. We do this by removing the variable "efivars" at file
scope and move its storage down to the end of the file.
Variables get at efivars by storing the efivars pointer within each
efivar_entry. The "new_var" and "del_var" binary attribute files get at
the efivars through the private pointer.
Signed-off-by: Mike Waychison <[email protected]>
Cc: Matt Domsch <[email protected]>,
Signed-off-by: Greg Kroah-Hartman <[email protected]>
In preparation for encapsulating efivars, we need to have the
bin_attributes be dynamically allocated so that we can use their
->private fields to get back to the struct efivars structure.
Signed-off-by: Mike Waychison <[email protected]>
Cc: Matt Domsch <[email protected]>,
Signed-off-by: Greg Kroah-Hartman <[email protected]>
In preparation for abstracting out efivars to be usable by other similar
variable services, move the global lock, list and kset into a structure.
Later patches will change the scope of 'efivars' and have it be passed
by function argument.
Signed-off-by: Mike Waychison <[email protected]>
Cc: Matt Domsch <[email protected]>,
Signed-off-by: Greg Kroah-Hartman <[email protected]>
DMI entries are arranged in memory back to back with no alignment
guarantees. This means that the struct dmi_header passed to callbacks
from dmi_walk() itself isn't byte aligned. This causes problems on
architectures that expect aligned data, such as IA64.
The dmi-sysfs patchset introduced structure member accesses through this
passed in dmi_header. Fix this by memcpy()ing the structures to
temporary locations on stack when inspecting/copying them.
Signed-off-by: Mike Waychison <[email protected]>
Tested-by: Tony Luck <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
Document the new ABI added by the dmi-sysfs module.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
The System Event Log described by DMI entry type 15 may be backed by
either memory or may be indirectly accessed via an IO index/data
register pair.
In order to get read access to this log, expose it in the
"system_event_log" sub-directory of type 15 DMI entries, ie:
/sys/firmware/dmi/entries/15-0/system_event_log/raw_event_log.
This commit handles both IO accessed and memory access system event
logs. OEM specific access and GPNV support is explicitly not handled
and we error out in the logs when we do not recognize the access method.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
The optional type 15 entry of the DMI table describes a non-volatile
storage-backed system event log.
In preparation for the next commit which exposes the raw bits of the
event log to userland, create a new sub-directory within the dmi entry
called "system_event_log" and expose attribute files that describe the
event log itself.
Currently, only a single child object is permitted within a
dmi_sysfs_entry. We simply point at this child from the dmi_sysfs_entry
if it exists.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
Introduce a new module "dmi-sysfs" that exports the broken out entries
of the DMI table through sysfs.
Entries are enumerated via dmi_walk() on module load, and are populated
as kobjects rooted at /sys/firmware/dmi/entries.
Entries are named "<type>-<instance>", where:
<type> : is the type of the entry, and
<instance> : is the ordinal count within the DMI table of that
entry type. This instance is used in lieu the DMI
entry's handle as no assurances are made by the kernel
that handles are unique.
All entries export the following attributes:
length : The length of the formatted portion of the entry
handle : The handle given to this entry by the firmware
raw : The raw bytes of the entire entry, including the
formatted portion, the unformatted (strings) portion,
and the two terminating nul characters.
type : The DMI entry type
instance : The ordinal instance of this entry given its type.
position : The position ordinal of the entry within the table in
its entirety.
Entries in dmi-sysfs are kobject backed members called "struct
dmi_sysfs_entry" and belong to dmi_kset. They are threaded through
entry_list (protected by entry_list_lock) so that we can find them at
cleanup time.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
In preparation for the upcoming commits, introduce the DMI entry types to
the headers. These type names are based on those specified in the DMTF
SMBIOS specification version 2.7.1.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
This driver adds support for /dev/watchdog for boards using either the MCP51 or
MCP55 chipsets. These are also known as the nForce 430 and nForce 550. This
driver is likely to work on other chipsets as well, though those are the only
two that have been tested.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Wim Van Sebroeck <[email protected]>
The qla2xxx driver uses a port_id_t to mark the start of its enumerations. gcc
is complaining that wrap.b24 may be used uninitialized, but this doesn't look
to be possible. Silence it.
Signed-off-by: Giridhar Malavali <[email protected]>
Signed-off-by: James Bottomley <[email protected]>
Allow search for a contiguous block of iommu space to cross the next_bit
marker if we have already committed ourselves to flushing the gart.
There shouldn't be any reason why we'd restrict the search.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
We do this by removing a micro-optimization that tries to avoid grabbing
the iommu_bitmap_lock spinlock and using a bus-locked operation.
This still races with other simultaneous alloc_iommu or free_iommu(size >
1) which both use bus-unlocked operations.
The end result of this race is eventually ending up with an
iommu_gart_bitmap that has bits errornously set all over, making large
contiguous iommu space allocations fail with 'PCI-DMA: Out of IOMMU space'.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The attempt to fixup the lockless mce log buffer introduced an infinite loop
when trying to find a free entry.
And:
Using rcu_dereference() to load mcelog.next doesn't seem to be sufficient
enough to ensure that mcelog.next is loaded each time around the loop in
mce_log(). Instead, use an explicit rmb() to ensure that the compiler gets it
right.
AK: turned the smp_wmbs into true wmbs to make sure they are not
reordered by the compiler on UP.
Signed-off-by: Mike Waychison <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Change the /proc/sysvipc/shm|sem|msg files to use the generic seq_file
implementation for struct ipc_ids.
Signed-off-by: Mike Waychison <[email protected]>
Cc: Manfred Spraul <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The following two patches convert /proc/sysvipc/* to use seq_file.
This gives us the following:
- Self-consistent IPC records in proc.
- O(n) reading of the files themselves.
This patch:
Add a generic method for ipc types to be displayed using seq_file. This
patch abstracts out seq_file iterating over struct ipc_ids into ipc/util.c
Signed-off-by: Mike Waychison <[email protected]>
Cc: Manfred Spraul <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Mike Waychison