spi: prefix modalias with "spi:"

This makes it consistent with other buses (platform, i2c, vio, ...).  I'm
not sure why we use the prefixes, but there must be a reason.

This was easy enough to do it, and I did it.

Signed-off-by: Anton Vorontsov <[email protected]>
Cc: David Brownell <[email protected]>
Cc: David Woodhouse <[email protected]>
Cc: Grant Likely <[email protected]>
Cc: Jean Delvare <[email protected]>
Cc: Ben Dooks <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Dmitry Torokhov <[email protected]>
Cc: Samuel Ortiz <[email protected]>
Cc: "John W. Linville" <[email protected]>
Acked-by: Mike Frysinger <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

hwmon: lm70: convert to device table matching

Make the code a little bit nicer, and shorter.

Signed-off-by: Anton Vorontsov <[email protected]>
Cc: Kaiwan N Billimoria <[email protected]>
Cc: David Brownell <[email protected]>
Cc: David Woodhouse <[email protected]>
Cc: Grant Likely <[email protected]>
Cc: Jean Delvare <[email protected]>
Cc: Ben Dooks <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

hwmon: adxx: convert to device table matching

Make the code a little bit nicer, and shorter.

Signed-off-by: Anton Vorontsov <[email protected]>
Cc: Marc Pignat <[email protected]>
Cc: David Brownell <[email protected]>
Cc: David Woodhouse <[email protected]>
Cc: Grant Likely <[email protected]>
Cc: Jean Delvare <[email protected]>
Cc: Ben Dooks <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Acked-by: Marc Pignat <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

of: remove "stm,m25p40" alias

The alias isn't needed any longer since the m25p80 driver converted to the
module device table matching.

Signed-off-by: Anton Vorontsov <[email protected]>
Cc: David Brownell <[email protected]>
Cc: David Woodhouse <[email protected]>
Cc: Grant Likely <[email protected]>
Cc: Jean Delvare <[email protected]>
Cc: Ben Dooks <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

spi: add support for device table matching

With this patch spi drivers can use standard spi_driver.id_table and
MODULE_DEVICE_TABLE() mechanisms to bind against the devices.  Just like
we do with I2C drivers.

This is useful when a single driver supports several variants of devices
but it is not possible to detect them in run-time (like non-JEDEC chips
probing in drivers/mtd/devices/m25p80.c), and when platform_data usage is
overkill.

This patch also makes life a lot easier on OpenFirmware platforms, since
with OF we extensively use proper device IDs in modaliases.

Signed-off-by: Anton Vorontsov <[email protected]>
Cc: David Brownell <[email protected]>
Cc: David Woodhouse <[email protected]>
Cc: Grant Likely <[email protected]>
Cc: Jean Delvare <[email protected]>
Cc: Ben Dooks <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

spi: add SPI driver for most known i.MX SoCs

This driver has been tested on i.MX1/i.MX27/i.MX35 with an AT25 type
EEPROM and on i.MX27/i.MX31 with a Freescale MC13783 PMIC.

Signed-off-by: Sascha Hauer <[email protected]>
Tested-by: Guennadi Liakhovetski <[email protected]>
Acked-by: David Brownell <[email protected]>
Cc: Andrea Paterniani <[email protected]>
Cc: Russell King <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

spi: add default selection of PL022 for ARM reference platforms

This makes the PL022 driver a default choice for any RealView and
Versatile boards plus the integrator IMPD1 which all contain the PL022
PrimeCell.  This will make it a default choice if and only if a user
selects SPI support for their board.

Signed-off-by: Linus Walleij <[email protected]>
Signed-off-by: David Brownell <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

spi.h: add missing kernel-doc for struct spi_master

Add missing kernel-doc notation in spi.h for struct spi_master:

Warning(include/linux/spi/spi.h:289): No description found for parameter 'mode_bits'
Warning(include/linux/spi/spi.h:289): No description found for parameter 'flags'

Signed-off-by: Randy Dunlap <[email protected]>
Acked-by: David Brownell <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

spi: add spi_ppc4xx driver

This adds a SPI driver for the SPI controller found in the IBM/AMCC
4xx PowerPC's.

Signed-off-by: Stefan Roese <[email protected]>
Signed-off-by: Wolfgang Ocker <[email protected]>
Acked-by: Josh Boyer <[email protected]>
Signed-off-by: Steven A. Falco <[email protected]>
Signed-off-by: David Brownell <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

spi: omap2_mcspi use BIT(n)

Convert bit shifted values into BIT format

Signed-off-by: Jouni Hogander <[email protected]>
Signed-off-by: Tony Lindgren <[email protected]>
Signed-off-by: David Brownell <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

spi: remove i.MX SPI driver

This driver is in a non working state at the moment and will be replaced
by a bitbang driver which can also handle the newer i.MX variants

Signed-off-by: Sascha Hauer <[email protected]>
Cc: Guennadi Liakhovetski <[email protected]>
Acked-by: David Brownell <[email protected]>
Acked-by: Andrea Paterniani <[email protected]>
Cc: Russell King <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

ncpfs: fix wrong check in __ncp_ioctl()

We want to check for s_inode's existence, not inode's one (inode is always
valid in this function).

This takes care of the following entry from Dan's list:

fs/ncpfs/ioctl.c +445 __ncp_ioctl(180) warning: variable derefenced before check 'inode'

Reported-by: Dan Carpenter <[email protected]>
Cc: Julia Lawall <[email protected]>
Signed-off-by: Bartlomiej Zolnierkiewicz <[email protected]>
Cc: Petr Vandrovec <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

ncpfs: remove dead URL from documentation

Noticed-by: Joe Perches <[email protected]>
Cc: Petr Vandrovec <[email protected]>
Signed-off-by: Bartlomiej Zolnierkiewicz <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

ncpfs: read buffer overflow

This function uses signed integers for the unix_date and local variables -
if a negative number is supplied and the leap-year condition is not met,
month will be 0, leading to a later read of day_n[-1]

Signed-off-by: Roel Kluin <[email protected]>
Cc: Petr Vandrovec <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

ramfs: move RAMFS_MAGIC to include/linux/magic.h

initramfs userspace likes to use this magic number.

Cc: Hugh Dickins <[email protected]>
Signed-off-by: maximilian attems <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

/proc/kcore: update stat.st_size after memory hotplug

After memory hotplug (or other events in future), kcore size can be
modified.

To update inode->i_size, we have to know inode/dentry but we can't get it
from inside /proc directly.  But considerinyg memory hotplug, kcore image
is updated only when it's opened.  Then, updating inode->i_size at open()
is enough.

Signed-off-by: KAMEZAWA Hiroyuki <[email protected]>
Acked-by: WANG Cong <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

/proc/kcore: fix stat.st_size

Presently the size of /proc/kcore which can be read by 'ls -l' is 0.  But
it's not the correct value.

On x86-64, ls -l shows
 ... root root 140737486266368 2009-09-17 10:29 /proc/kcore
Then, 7FFFFFFE02000. This comes from vmalloc area's size.
(*) This shows "core" size, not  memory size.

This patch shows the size by updating "size" field in struct
proc_dir_entry.  Later, lookup routine will create inode and fill
inode->i_size based on this value.  Then, this has a problem.

 - Once inode is cached, inode->i_size will never be updated.

Then, this patch is not memory-hotplug-aware.

To update inode->i_size, we have to know dentry or inode.
But there is no way to lookup them by inside kernel. Hmmm....
Next patch will try it.

Signed-off-by: KAMEZAWA Hiroyuki <[email protected]>
Acked-by: WANG Cong <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

kcore: more fixes for init

proc_kcore_init() doesn't check NULL case.  fix it and remove unnecessary
comments.

Signed-off-by: KAMEZAWA Hiroyuki <[email protected]>
Acked-by: WANG Cong <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

kcore: register module area in generic way

Some archs define MODULED_VADDR/MODULES_END which is not in VMALLOC area.
This is handled only in x86-64.  This patch make it more generic.  And we
can use vread/vwrite to access the area.  Fix it.

Signed-off-by: KAMEZAWA Hiroyuki <[email protected]>
Cc: Jiri Slaby <[email protected]>
Cc: Ralf Baechle <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: WANG Cong <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

kcore: register vmemmap range

Benjamin Herrenschmidt <[email protected]> pointed out that vmemmap
range is not included in KCORE_RAM, KCORE_VMALLOC ....

This adds KCORE_VMEMMAP if SPARSEMEM_VMEMMAP is used.  By this, vmemmap
can be readable via /proc/kcore

Because it's not vmalloc area, vread/vwrite cannot be used.  But the range
is static against the memory layout, this patch handles vmemmap area by
the same scheme with physical memory.

This patch assumes SPARSEMEM_VMEMMAP range is not in VMALLOC range.  It's
correct now.

[[email protected]: fix typo]
Signed-off-by: KAMEZAWA Hiroyuki <[email protected]>
Cc: Jiri Slaby <[email protected]>
Cc: Ralf Baechle <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: WANG Cong <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Randy Dunlap <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

kcore: use registerd physmem information

For /proc/kcore, each arch registers its memory range by kclist_add().
In usual,

- range of physical memory
- range of vmalloc area
- text, etc...

are registered but "range of physical memory" has some troubles.  It
doesn't updated at memory hotplug and it tend to include unnecessary
memory holes.  Now, /proc/iomem (kernel/resource.c) includes required
physical memory range information and it's properly updated at memory
hotplug.  Then, it's good to avoid using its own code(duplicating
information) and to rebuild kclist for physical memory based on
/proc/iomem.

Signed-off-by: KAMEZAWA Hiroyuki <[email protected]>
Signed-off-by: Jiri Slaby <[email protected]>
Cc: Ralf Baechle <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: WANG Cong <[email protected]>
Cc: KOSAKI Motohiro <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

walk system ram range

Originally, walk_memory_resource() was introduced to traverse all memory
of "System RAM" for detecting memory hotplug/unplug range.  For doing so,
flags of IORESOUCE_MEM|IORESOURCE_BUSY was used and this was enough for
memory hotplug.

But for using other purpose, /proc/kcore, this may includes some firmware
area marked as IORESOURCE_BUSY | IORESOUCE_MEM.  This patch makes the
check strict to find out busy "System RAM".

Note: PPC64 keeps their own walk_memory_resouce(), which walk through
ppc64's lmb informaton.  Because old kclist_add() is called per lmb, this
patch makes no difference in behavior, finally.

And this patch removes CONFIG_MEMORY_HOTPLUG check from this function.
Because pfn_valid() just show "there is memmap or not* and cannot be used
for "there is physical memory or not", this function is useful in generic
to scan physical memory range.

Signed-off-by: KAMEZAWA Hiroyuki <[email protected]>
Cc: Ralf Baechle <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: WANG Cong <[email protected]>
Cc: Américo Wang <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Roland Dreier <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

kcore: register text area in generic way

Some 64bit arch has special segment for mapping kernel text.  It should be
entried to /proc/kcore in addtion to direct-linear-map, vmalloc area.
This patch unifies KCORE_TEXT entry scattered under x86 and ia64.

I'm not familiar with other archs (mips has its own even after this patch)
but range of [_stext ..._end) is a valid area of text and it's not in
direct-map area, defining CONFIG_ARCH_PROC_KCORE_TEXT is only a necessary
thing to do.

Note: I left mips as it is now.

Signed-off-by: KAMEZAWA Hiroyuki <[email protected]>
Cc: Ralf Baechle <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: WANG Cong <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

kcore: register vmalloc area in generic way

For /proc/kcore, vmalloc areas are registered per arch.  But, all of them
registers same range of [VMALLOC_START...VMALLOC_END) This patch unifies
them.  By this.  archs which have no kclist_add() hooks can see vmalloc
area correctly.

Signed-off-by: KAMEZAWA Hiroyuki <[email protected]>
Cc: Ralf Baechle <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: WANG Cong <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

kcore: add kclist types

Presently, kclist_add() only eats start address and size as its arguments.
Considering to make kclist dynamically reconfigulable, it's necessary to
know which kclists are for System RAM and which are not.

This patch add kclist types as
  KCORE_RAM
  KCORE_VMALLOC
  KCORE_TEXT
  KCORE_OTHER

This "type" is used in a patch following this for detecting KCORE_RAM.

Signed-off-by: KAMEZAWA Hiroyuki <[email protected]>
Cc: Ralf Baechle <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: WANG Cong <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

kcore: use usual list for kclist

This patchset is for /proc/kcore.  With this,

 - many per-arch hooks are removed.

 - /proc/kcore will know really valid physical memory area.

 - /proc/kcore will be aware of memory hotplug.

 - /proc/kcore will be architecture independent i.e.
   if an arch supports CONFIG_MMU, it can use /proc/kcore.
   (if the arch uses usual memory layout.)

This patch:

/proc/kcore uses its own list handling codes. It's better to use
generic list codes.

No changes in logic. just clean up.

Signed-off-by: KAMEZAWA Hiroyuki <[email protected]>
Cc: Ralf Baechle <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: WANG Cong <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

procfs: provide stack information for threads

A patch to give a better overview of the userland application stack usage,
especially for embedded linux.

Currently you are only able to dump the main process/thread stack usage
which is showed in /proc/pid/status by the "VmStk" Value.  But you get no
information about the consumed stack memory of the the threads.

There is an enhancement in the /proc/<pid>/{task/*,}/*maps and which marks
the vm mapping where the thread stack pointer reside with "[thread stack
xxxxxxxx]".  xxxxxxxx is the maximum size of stack.  This is a value
information, because libpthread doesn't set the start of the stack to the
top of the mapped area, depending of the pthread usage.

A sample output of /proc/<pid>/task/<tid>/maps looks like:

08048000-08049000 r-xp 00000000 03:00 8312       /opt/z
08049000-0804a000 rw-p 00001000 03:00 8312       /opt/z
0804a000-0806b000 rw-p 00000000 00:00 0          [heap]
a7d12000-a7d13000 ---p 00000000 00:00 0
a7d13000-a7f13000 rw-p 00000000 00:00 0          [thread stack: 001ff4b4]
a7f13000-a7f14000 ---p 00000000 00:00 0
a7f14000-a7f36000 rw-p 00000000 00:00 0
a7f36000-a8069000 r-xp 00000000 03:00 4222       /lib/libc.so.6
a8069000-a806b000 r--p 00133000 03:00 4222       /lib/libc.so.6
a806b000-a806c000 rw-p 00135000 03:00 4222       /lib/libc.so.6
a806c000-a806f000 rw-p 00000000 00:00 0
a806f000-a8083000 r-xp 00000000 03:00 14462      /lib/libpthread.so.0
a8083000-a8084000 r--p 00013000 03:00 14462      /lib/libpthread.so.0
a8084000-a8085000 rw-p 00014000 03:00 14462      /lib/libpthread.so.0
a8085000-a8088000 rw-p 00000000 00:00 0
a8088000-a80a4000 r-xp 00000000 03:00 8317       /lib/ld-linux.so.2
a80a4000-a80a5000 r--p 0001b000 03:00 8317       /lib/ld-linux.so.2
a80a5000-a80a6000 rw-p 0001c000 03:00 8317       /lib/ld-linux.so.2
afaf5000-afb0a000 rw-p 00000000 00:00 0          [stack]
ffffe000-fffff000 r-xp 00000000 00:00 0          [vdso]

Also there is a new entry "stack usage" in /proc/<pid>/{task/*,}/status
which will you give the current stack usage in kb.

A sample output of /proc/self/status looks like:

Name: cat
State: R (running)
Tgid: 507
Pid: 507
.
.
.
CapBnd: fffffffffffffeff
voluntary_ctxt_switches: 0
nonvoluntary_ctxt_switches: 0
Stack usage: 12 kB

I also fixed stack base address in /proc/<pid>/{task/*,}/stat to the base
address of the associated thread stack and not the one of the main
process.  This makes more sense.

[[email protected]: fs/proc/array.c now needs walk_page_range()]
Signed-off-by: Stefani Seibold <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Alexey Dobriyan <[email protected]>
Cc: "Eric W. Biederman" <[email protected]>
Cc: Randy Dunlap <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

fs/proc/base.c: fix proc_fault_inject_write() input sanity check

Remove obfuscated zero-length input check and return -EINVAL instead of
-EIO error to make the error message clear to user.  Add whitespace
stripping.  No functionality changes.

The old code:

echo  1  > /proc/pid/make-it-fail (ok)
echo 1foo > /proc/pid/make-it-fail (-bash: echo: write error: Input/output error)

The new code:

echo  1  > /proc/pid/make-it-fail (ok)
echo 1foo > /proc/pid/make-it-fail (-bash: echo: write error: Invalid argument)

This patch is conservative in changes to not breaking existing
scripts/applications.

Signed-off-by: Vincent Li <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

fs/proc/task_mmu.c v1: fix clear_refs_write() input sanity check

Andrew Morton pointed out similar string hacking and obfuscated check for
zero-length input at the end of the function, David Rientjes suggested to
use strict_strtol to replace simple_strtol, this patch cover above
suggestions, add removing of leading and trailing whitespace from user
input.  It does not change function behavious.

Signed-off-by: Vincent Li <[email protected]>
Acked-by: David Rientjes <[email protected]>
Cc: Matt Mackall <[email protected]>
Cc: Amerigo Wang <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

kcore: fix /proc/kcore's stat.st_size

In 9063c61fd5cbd ("x86, 64-bit: Clean up user address masking") Linus
fixed the wrong size of /proc/kcore problem.

But its size still looks insane, since it never equals the size of
physical memory.

Signed-off-by: WANG Cong <[email protected]>
Cc: "Eric W. Biederman" <[email protected]>
Cc: Tao Ma <[email protected]>
Cc: <[email protected]>
Acked-by: KAMEZAWA Hiroyuki <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

proc_flush_task: flush /proc/tid/task/pid when a sub-thread exits

The exiting sub-thread flushes /proc/pid only, but this doesn't buy too
much: ps and friends mostly use /proc/tid/task/pid.

Remove "if (thread_group_leader())" checks from proc_flush_task() path,
this means we always remove /proc/tid/task/pid dentry on exit, and this
actually matches the comment above proc_flush_task().

The test-case:

static void* tfunc(void *arg)
{
char name[256];

sprintf(name, "/proc/%d/task/%ld/status", getpid(), gettid());
close(open(name, O_RDONLY));

return NULL;
}

int main(void)
{
pthread_t t;

for (;;) {
if (!pthread_create(&t, NULL, &tfunc, NULL))
pthread_join(t, NULL);
}
}

slabtop shows that pid/proc_inode_cache/etc grow quickly and
"indefinitely" until the task is killed or shrink_slab() is called, not
good.  And the main thread needs a lot of time to exit.

The same can happen if something like "ps -efL" runs continuously, while
some application spawns short-living threads.

Reported-by: "James M. Leddy" <[email protected]>
Signed-off-by: Oleg Nesterov <[email protected]>
Cc: Alexey Dobriyan <[email protected]>
Cc: "Eric W. Biederman" <[email protected]>
Cc: Dominic Duval <[email protected]>
Cc: Frank Hirtz <[email protected]>
Cc: "Fuller, Johnray" <[email protected]>
Cc: Larry Woodman <[email protected]>
Cc: Paul Batkowski <[email protected]>
Cc: Roland McGrath <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

proc: fix reported unit for RLIMIT_CPU

/proc/$pid/limits should show RLIMIT_CPU as seconds, which is the unit
used in kernel/posix-cpu-timers.c:

        unsigned long psecs = cputime_to_secs(ptime);
        ...
        if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
                ...
                __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);

Signed-off-by: Kees Cook <[email protected]>
Acked-by: WANG Cong <[email protected]>
Acked-by: Neil Horman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

sdhci: increase timeout for internal clock stabilization.

On an OLPC XO-1.5 development board with Via VX855 chipset, the sdhci
controller can take up to 12ms to stabilize its clock, but the current
timeout at which we give up on the controller is 10ms.

The patch increases the timeout delay rather than using a device-specific
quirk -- since we exit the loop when the clock comes up, increasing the
timeout value will only make us mdelay() longer in the errant case of a
device with a clock that is not stabilizing, which it seems worth waiting
a little longer for in general.

Signed-off-by: Chris Ball <[email protected]>
Cc: Harald Welte <[email protected]>
Acked-by: Pierre Ossman <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

sdhci: support for ADMA only hosts

Add support for ADMA on SDHCI hosts, not supporting SDMA.

According to the SDHCI specifications a host can support ADMA but not SDMA

Signed-off-by: Richard Röjfors <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mmc: make SDIO device/driver struct accessors public

Especially with the PM framework, those are quite handy to have in driver
code too.

Signed-off-by: Nicolas Pitre <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mmc: propagate error codes back from bus drivers' suspend/resume methods

Especially for SDIO drivers which may have special conditions/errors to
report, it is a good thing to relay the returned error code back to upper
layers.

This also allows for the rationalization of the resume path where code to
"remove" a no-longer-existing or replaced card was duplicated into the
MMC, SD and SDIO bus drivers.

In the SDIO case, if a function suspend method returns an error, then all
previously suspended functions are resumed and the error returned.  An
exception is made for -ENOSYS which the core interprets as "we don't
support suspend so just kick the card out for suspend and return success".

When resuming SDIO cards, the core code only validates the manufacturer
and product IDs to make sure the same kind of card is still present before
invoking functions resume methods.  It's the function driver's
responsibility to perform further tests to confirm that the actual same
card is present (same MAC address, etc.) and return an error otherwise.

Signed-off-by: Nicolas Pitre <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mmc: core SDIO suspend/resume support

Currently, all SDIO cards are virtually removed upon a suspend, and
completely reprobed upon a resume.  This adds the suspend and resume
methods to the SDIO bus driver so to be able to dispatch those events to
the actual SDIO function drivers for real suspend/resume instead.

All active functions on a card must have a driver with both a suspend and
a resume method though.  Failing that, we fall back to the current
behavior of simply "removing" the card when suspending.

When resuming, we make sure the same card is still inserted by comparing
the vendor and product IDs.  If there is a mismatch, or if there is simply
no card anymore in the slot, then the previous card is "removed" and the
new card is detected.  This is further enhanced with the next patch.

[[email protected]: fix warnings]
Signed-off-by: Nicolas Pitre <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mmc_spi: fail gracefully if host or card do not support the switch command

Some time ago, I have send a patch to the mmc_spi subsystem changing the
error codes.  This was after a discussion with Pierre about using EINVAL
only for non-recoverable errors.  This patch was accepted as

http://git.kernel.org/linus/fdd858db7113ca64132de390188d7ca00701013d

Unfortunately, several weeks later, I realized that this patch has opened
a little can of worms because there are SD cards on the market which

a) claim that they support the switch command
AND
b) refuse to execute this command if operating in SPI mode.

So, such a card would get unusuable in an embedded linux system in SPI
mode, because the init sequence terminates with an error.

This patch adds the missing error codes to the caller of the switch
command and restores the old behaviour to fail gracefully if these
commands can not execute.

Signed-off-by: Wolfgang Muees <[email protected]>
Cc: <[email protected]>
Cc: <[email protected]> [2.6.31.x]
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap4: mmc driver support on OMAP4

Add basic support for all 5 MMC controllers on OMAP4.

This patch doesn't include mmc-regulator support

Signed-off-by: Kishore Kadiyala <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Acked-by: Madhusudhan Chikkature <[email protected]>
Cc: Russell King <[email protected]>
Acked-by: Tony Lindgren <[email protected]>
Cc: Hiroshi DOYU <[email protected]>
Cc: Sakari Ailus <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

sdhci: add no-card-no-reset quirk for Ricoh R5C822/Sony Z11

Card insertion detection is broken without this quirk on a Sony Vaio
Z11, as discussed on linux-mmc here: http://marc.info/?t=125017355000008

Signed-off-by: Chris Ball <[email protected]>
Tested-by: Norbert Preining <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

AT91: atmel-mci: Platform configuration to the the atmel-mci driver

Created a modified version of the at91sam9g20 evaluation kit platform
(board-sam9g20ek-2slot-mmc.c) and device support to make use of the
updated atmel-mci driver.

As the use of two slots modify GPIO pin allocation, we create another
board file.

This requires getting the most updated arch/arm/tools/mach-types from
http://www.arm.linux.org.uk/developer/machines/download.php to have the machine
type for the at91sam9g20ek-2slot-mmc board.

[[email protected]: printk, slot_count modification in at91sam9260_devices.c file]
[[email protected]: coding-style fixes]
Signed-off-by: Rob Emanuele <[email protected]>
Signed-off-by: Nicolas Ferre <[email protected]>
Cc: Haavard Skinnemoen <[email protected]>
Cc: Andrew Victor <[email protected]>
Cc: Russell King <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

atmel-mci: unified Atmel MCI drivers (AVR32 & AT91)

Unification of the atmel-mci driver to support the AT91 processors MCI
interface.  The atmel-mci driver currently supports the AVR32 and this
patch adds AT91 support.

Add read/write proof selection switch dependent on chip availability of
this feature.

To use this new driver on a at91 the platform driver for your board needs
to be updated.

[[email protected] indent, Kconfig comment and one printk modification]
Signed-off-by: Rob Emanuele <[email protected]>
Signed-off-by: Nicolas Ferre <[email protected]>
Cc: Haavard Skinnemoen <[email protected]>
Cc: Andrew Victor <[email protected]>
Cc: Russell King <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

sdio: add MMC_QUIRK_LENIENT_FN0

Normally writes to SDIO function 0 outside the vendor specific CCCR
registers are prohibited.

To support embedded devices that require writes to SDIO function 0 outside
this range (e.g.  TI WL127x embedded sdio wifi device),
MMC_QUIRK_LENIENT_FN0 is introduced.

A card quirks field is added to `struct mmc_card' to support non-standard
devices (e.g.  embedded sdio devices).

[[email protected]: code in C, not cpp!]
Signed-off-by: Ohad Ben-Cohen <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

sdhci-of: cleanup eSDHC's set_clock() a little bit

- Get rid of incomprehensible "if { for { if } }" construction for the
  exponential divisor calculation. The first if statement isn't correct
  at all, since it should check for "host->max_clk / pre_div / 16 >
  clock". The error doesn't cause any bugs because the check in the for
  loop does the right thing, and so the outer check becomes useless;

- For the linear divisor do the same: a single while statement is more
  readable than for + if construction;

- Add dev_dbg() that prints desired and actual clock frequency.

Signed-off-by: Anton Vorontsov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Kumar Gala <[email protected]>
Cc: David Vrabel <[email protected]>
Cc: Ben Dooks <[email protected]>
Cc: Sascha Hauer <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

sdhci-of: don't hard-code inverted write-protect quirk

MPC85xx SOCs have normal write-protect state reporting, so we shouldn't
hard-code the quirk.

Instead, look for "sdhci,wp-inverted" property, plus check for
mpc837x_{rdb,mds} machines since older device trees don't specify the new
property.

Signed-off-by: Anton Vorontsov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Kumar Gala <[email protected]>
Cc: David Vrabel <[email protected]>
Cc: Ben Dooks <[email protected]>
Cc: Sascha Hauer <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

powerpc: introduce and document sdhci,wp-inverted property for eSDHC

eSDHC block in MPC837x SOCs reports inverted write-protect state, soon
sdhci-of driver will look for sdhci,wp-inverted properties to decide
whether apply a specific quirk.

So, document the property and add it to device tree source files.

Signed-off-by: Anton Vorontsov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Kumar Gala <[email protected]>
Cc: David Vrabel <[email protected]>
Cc: Ben Dooks <[email protected]>
Cc: Sascha Hauer <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

sdhci-of: fix high-speed cards recognition

eSDHC fails to recognize some SDHS cards, throwing timeout errors:

  mmc0: error -110 whilst initialising SD card

That's because we calculate timeout value in a wrong way: on eSDHC hosts
the timeout clock is derivied from the SD clock, which is set dynamically.

As David Vrabel suggested, deriving timeout clock from SD clock is a
common scheme, so let's implement DATA_TIMEOUT_USES_SDCLK quirk and use it
for eSDHC hosts.

Also, from now on we don't need esdhc_get_timeout_clock() callback, so
remove it.

Signed-off-by: Anton Vorontsov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Kumar Gala <[email protected]>
Cc: David Vrabel <[email protected]>
Cc: Ben Dooks <[email protected]>
Cc: Sascha Hauer <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

sdhci-of: avoid writing reserved bits into host control register

SDHCI core tries to write HISPD bit into the host control register, but
the eSDHC controllers don't have that bit, and that causes all sorts of
misbehaviour when using 4-bit mode capable SD cards.

Signed-off-by: Anton Vorontsov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Kumar Gala <[email protected]>
Cc: David Vrabel <[email protected]>
Cc: Ben Dooks <[email protected]>
Cc: Sascha Hauer <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

sdhci-of: fix SD clock calculation

Linear divisor's values in a register start at 0 (zero means "divide by
1").  Before this patch the code didn't account that fact, so SD cards
were running underclocked.

Signed-off-by: Anton Vorontsov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Kumar Gala <[email protected]>
Cc: David Vrabel <[email protected]>
Cc: Ben Dooks <[email protected]>
Cc: Sascha Hauer <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

sdio: fix read buffer overflow

Avoid buffer underrun when parsing an invalid CISTPL_VERS_1.

Signed-off-by: Roel Kluin <[email protected]>
Cc: David Vrabel <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

sdhci: be more strict with get_min_clock() usage

get_min_clock() makes sense only with NONSTANDARD_CLOCK quirk and when
set_clock() callback is specified.

The patch should cause no functional changes, it just makes the code
self-documented and avoids any possible misuse of get_min_clock().

Suggested-by: Pierre Ossman <[email protected]>
Signed-off-by: Anton Vorontsov <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: Matt Fleming <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: Pierre Ossman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

sdio: add CD disable support

Add support to disconnect the pull-up resistor on CD/DAT[3] (pin 1)
of the card. This may be desired on certain setups of boards,
controllers and embedded sdio devices which do not need the card's
pull-up. As a result, card detection is disabled and power is saved.

[[email protected]: simplify sdio_disable_cd() a bit]
Signed-off-by: Ohad Ben-Cohen <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: David Vrabel <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

MAINTAINERS: update for TI OMAP hsmmc driver

Update maintainers entry for TI OMAP HS MMC support.

Signed-off-by: Madhusudhan Chikkature <[email protected]>
Acked-by: Kevin Hilman <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

ARM: OMAP: RX51: set MMC capabilities and power-saving flag

Specify MMC capabilities and set the power-saving flag for RX51.

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: set a large data timeout for commands with busy signal

Commands like SWITCH (CMD6) send a response and then signal busy while the
operation is completed.  These commands are expected to always succeed
(otherwise the response would have indicated an error).

Set an arbitrarily large data timeout value (100ms) for these commands to
ensure that premature timeouts do not occur.

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: ensure all clock enables and disables are paired

[[email protected]: fix for the db clock failure message]
Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Madhusudhan Chikkature <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: protect the card when the cover is open

Depending on the manufacturer, there is a small possibility that removing
a card while it is being written to, can render the card permanently
unusable.  To prevent that, the card is made inaccessible when the cover
is open.

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: code refactoring

Functions', structures', variables' names are changed to start with
omap_hsmmc_ prefix.

Signed-off-by: Denis Karpov <[email protected]>
Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: prevent races with irq handler

If an unexpected interrupt occurs while preparing the next request, an
oops can occur.

For example, a new request is setting up DMA for data transfer so
host->data is not NULL.  An unexpected transfer complete (TC) interrupt
comes along and the interrupt handler sets host->data to NULL.  Oops!

Prevent that by adding a spinlock.

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: cater for weird CMD6 behaviour

Sometimes the controller unexpectedly produces a TC (transfer complete)
interrupt before the CC (command complete) interrupt for command 6
(SWITCH).  This is a problem because the CC interrupt can get mixed up
with the next request.  Add a hack for CMD6.

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: clear interrupt status after init sequence

Clear the interrupt status after sending the initialization sequence, as
specified in the TRM.

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: cleanup macro usage

Use macro mmc_slot() in omap_hsmmc.

Signed-off-by: Denis Karpov <[email protected]>
Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: fix NULL pointer dereference

Do not call 'mmc_omap_xfer_done()' if the request is already done.

Signed-off-by: Jarkko Lavinen <[email protected]>
Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: add mmc card sleep and awake support

After 1 second of inactivity, put card and/or regulator to sleep.  After 8
seconds of inactivity, turn off the power.

Signed-off-by: Jarkko Lavinen <[email protected]>
Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: put MMC regulator to sleep

When a card is not in use, the voltage regulator can be put to sleep.
This is an alternative to powering the card off, when powering off is not
safe because the card might be replaced without the driver being aware of
it.

That situation happens if:
- the card is removable i.e. not eMMC
- and there is no card detect
- and there is a cover switch but the cover is open

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

ARM: OMAP: mmc-twl4030: add regulator sleep / wake function

Add the ability for the driver to put the card power regulators to sleep
and wake them up again.

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: support for deeper power saving states

Support for multi-level dynamic power saving states in omap_hsmmc
(ENABLED->DISABLED->OFF).  In the "deepest" state (OFF) we switch off the
voltage regulators.

Signed-off-by: Denis Karpov <[email protected]>
Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: make use of new MMC_CAP_NONREMOVABLE host capability

Let the board specify that a card is nonremovable e.g. eMMC

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: fix scatter-gather list sanity checking

Do not use host->dma_len when it is uninitialzed.  Finish the request with
an error if the mmc_omap_prepare_data() fails.

Signed-off-by: Jarkko Lavinen <[email protected]>
Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: ensure workqueues are empty before suspend

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: set open drain bit correctly

The code could set the bit to 1 but not reset it to 0.

Signed-off-by: Denis Karpov <[email protected]>
Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: context save/restore support

Keep the context over PM dynamic OFF states.

Signed-off-by: Denis Karpov <[email protected]>
Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: keep track of power mode

This patch is preparation for adding context save and restore support.

Keep track of the current power mode so that the context restore function
can avoid restoring the context for a card if the power has been switched
off.  If the power is off, the card must be reinitialized anyway which
will re-establish the context.

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

ARM: OMAP: mmc-twl4030: add context loss counter support

PM dynamic OFF state results in context loss.  That is, the host
controller has been powered off at some point, which means the registers
have been reset.  The driver must detect when this happens, and restore
the context.  This patch adds the means to detect context loss.

Note, the PM side is not yet implemented.

Signed-off-by: Denis Karpov <[email protected]>
Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: make use of new enable/disable interface

For the moment enable / disable just turns the fclk on and off.

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

omap_hsmmc: add debugfs entry (host registers)

Adds <debugfs_root>/kernel/debug/mmc<N>/regs entry, contents show
registers' state and some driver internal state variables.

Signed-off-by: Denis Karpov <[email protected]>
Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mmc: check status after MMC SWITCH command

According to the standard, the SWITCH command should be followed by a
SEND_STATUS command to check for errors.

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mmc: power off once at removal

Fix MMC host stop sequence: power off once.

Signed-off-by: Denis Karpov <[email protected]>
Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mmc: add mmc card sleep and awake support

Add support for the new MMC command SLEEP_AWAKE.

Signed-off-by: Jarkko Lavinen <[email protected]>
Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mmc: add ability to save power by powering off cards

Power can be saved by powering off cards that are not in use.  This is
similar to suspend / resume except it is under the control of the driver,
and does not require any power management support.  It can only be used
when the driver can monitor whether the card is removed, otherwise it is
unsafe.  This is possible because, unlike suspend, the driver still
receives card detect and / or cover switch interrupts.

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mmc: add MMC_CAP_NONREMOVABLE host capability

eMMC's are not removable, so unsafe resume is OK always.

To permit this a new host capability MMC_CAP_NONREMOVABLE has been added
and suspend / resume updated accordingly.

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mmc: allow host claim / release nesting

This change allows the MMC host to be claimed in situations where the host
may or may not have already been claimed.  Also 'mmc_try_claim_host()' is
now exported.

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mmc: add 'enable' and 'disable' methods to mmc host

MMC hosts that support power saving can use the 'enable' and 'disable'
methods to exit and enter power saving states.  An explanation of their
use is provided in the comments added to include/linux/mmc/host.h.

Signed-off-by: Adrian Hunter <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Jarkko Lavinen <[email protected]>
Cc: Denis Karpov <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Philip Langdale <[email protected]>
Cc: "Madhusudhan" <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

sdio: do not ignore MMC_VDD_165_195

This is needed for 1.8V embedded SDIO devices and supporting host controllers
(e.g. TI 127x and ZOOM2 boards)

Signed-off-by: Ohad Ben-Cohen <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Philip Langdale <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mmc: register mmci-omap-hs using platform_driver_probe

omap_mmc_probe lives in .init.text, so using platform_driver_register to
register it is wrong because binding a device after the init memory is
discarded (e.g.  via sysfs) results in an oops.

As requested by David Brownell platform_driver_probe is used instead of
moving the probe function to .devinit.text as proposed initially.  This
saves some memory, but devices registered after the driver is probed are
not bound (probably there are none) and binding via sysfs isn't possible.

Signed-off-by: Uwe Kleine-König <[email protected]>
Cc: Jean Pihet <[email protected]>
Cc: Tony Lindgren <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Andy Lowe <[email protected]>
Cc: Adrian Hunter <[email protected]>
Cc: Andrew Morton <[email protected]>
Acked-by: David Brownell <[email protected]>
Cc: Madhusudhan Chikkature<[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Philip Langdale <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

msm_sdcc.c: move overly indented code to separate function

Signed-off-by: Joe Perches <[email protected]>
Cc: Pavel Machek <[email protected]>
Cc: Brian Swetland <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: San Mehat <[email protected]>
Cc: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Philip Langdale <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

msm_sdcc.c: stylistic cleaning

Make it a bit more like typical kernel style.

Signed-off-by: Joe Perches <[email protected]>
Cc: Pavel Machek <[email protected]>
Cc: Brian Swetland <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: San Mehat <[email protected]>
Cc: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Philip Langdale <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

msm_sdcc.c: convert printk(KERN_<level> to pr_<level>(

Signed-off-by: Joe Perches <[email protected]>
Cc: Pavel Machek <[email protected]>
Cc: Brian Swetland <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: San Mehat <[email protected]>
Cc: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Philip Langdale <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mmc: msm_sdccc: driver for HTC Dream

MMC Driver for HTC Dream.  I picked the code up from Google git trees,
removed stuff not strictly necessary, and did a few cleanups.  It still
works :-).

Signed-off-by: Pavel Machek <[email protected]>
Cc: Brian Swetland <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Joe Perches <[email protected]>
Cc: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Philip Langdale <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

OMAP: HSMMC: do not enable buffer ready interrupt if using DMA

This considerably reduces the number of interrupts during a transfer
and ought to result in some power saving.

Signed-off-by: Anand Gadiyar <[email protected]>
Signed-off-by: Santosh Shilimkar <[email protected]>
Acked-by: Kishore Kadiyala <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Matt Fleming <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Philip Langdale <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mmc: in mmc_power_up(), use previously selected ocr if available

When mmc_power_up is called during unsafe resume, host->ocr should be used
instead of host->ocr_avail.

Signed-off-by: Balaji Rao <[email protected]>
Cc: Andy Green <[email protected]>
Cc: Pierre Ossman <[email protected]>
Cc: Ian Molton <[email protected]>
Cc: "Roberto A. Foglietta" <[email protected]>
Cc: Philip Langdale <[email protected]>
Acked-by: Matt Fleming <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Blackfin: override text/data checking functions

Signed-off-by: Mike Frysinger <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Robin Getz <[email protected]>
Cc: Sam Ravnborg <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Arnd Bergmann <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

lockdep: use new arch_is_kernel_data()

This allows lockdep to locate symbols that are in arch-specific data
sections (such as data in Blackfin on-chip SRAM regions).

Signed-off-by: Mike Frysinger <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Robin Getz <[email protected]>
Cc: Sam Ravnborg <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Arnd Bergmann <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

kallsyms: use new arch_is_kernel_text()

This allows kallsyms to locate symbols that are in arch-specific text
sections (such as text in Blackfin on-chip SRAM regions).

Signed-off-by: Mike Frysinger <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Robin Getz <[email protected]>
Cc: Sam Ravnborg <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Arnd Bergmann <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

asm/sections: add text/data checking functions for arches to override

Some ports (like the Blackfin arch) have a discontiguous memory map which
means there may be text or data that falls outside of the standard range
of the start/end text/data symbols.  Creating some helper functions allows
these non-standard ports to declare these regions without adversely
affecting anyone else.

Signed-off-by: Mike Frysinger <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Robin Getz <[email protected]>
Cc: Sam Ravnborg <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Arnd Bergmann <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

kallsyms: fix segfault in prefix_underscores_count()

Commit b478b782e110fdb4135caa3062b6d687e989d994 "kallsyms, tracing: output
more proper symbol name" introduces a "bugfix" that introduces a segfault
in kallsyms in my configurations.

The cause is the introduction of prefix_underscores_count() which attempts
to count underscores, even in symbols that do not have them.  As a result,
it just uselessly runs past the end of the buffer until it crashes:

  CC      init/version.o
  LD      init/built-in.o
  LD      .tmp_vmlinux1
  KSYM    .tmp_kallsyms1.S
/bin/sh: line 1: 16934 Done                    sh-linux-gnu-nm -n .tmp_vmlinux1
     16935 Segmentation fault      | scripts/kallsyms > .tmp_kallsyms1.S
make: *** [.tmp_kallsyms1.S] Error 139

This simplifies the logic and just does a straightforward count.

Signed-off-by: Paul Mundt <[email protected]>
Reviewed-by: Li Zefan <[email protected]>
Cc: Lai Jiangshan <[email protected]>
Cc: Sam Ravnborg <[email protected]>
Cc: Paulo Marques <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: <[email protected]> [2.6.30.x, 2.6.31.x]
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

getrusage: fill ru_maxrss value

Make ->ru_maxrss value in struct rusage filled accordingly to rss hiwater
mark.  This struct is filled as a parameter to getrusage syscall.
->ru_maxrss value is set to KBs which is the way it is done in BSD
systems.  /usr/bin/time (gnu time) application converts ->ru_maxrss to KBs
which seems to be incorrect behavior.  Maintainer of this util was
notified by me with the patch which corrects it and cc'ed.

To make this happen we extend struct signal_struct by two fields.  The
first one is ->maxrss which we use to store rss hiwater of the task.  The
second one is ->cmaxrss which we use to store highest rss hiwater of all
task childs.  These values are used in k_getrusage() to actually fill
->ru_maxrss.  k_getrusage() uses current rss hiwater value directly if mm
struct exists.

Note:
exec() clear mm->hiwater_rss, but doesn't clear sig->maxrss.
it is intetionally behavior. *BSD getrusage have exec() inheriting.

test programs
========================================================

getrusage.c
===========
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include <sys/types.h>
 #include <sys/time.h>
 #include <sys/resource.h>
 #include <sys/types.h>
 #include <sys/wait.h>
 #include <unistd.h>
 #include <signal.h>
 #include <sys/mman.h>

 #include "common.h"

 #define err(str) perror(str), exit(1)

int main(int argc, char** argv)
{
int status;

printf("allocate 100MB\n");
consume(100);

printf("testcase1: fork inherit? \n");
printf("  expect: initial.self ~= child.self\n");
show_rusage("initial");
if (__fork()) {
wait(&status);
} else {
show_rusage("fork child");
_exit(0);
}
printf("\n");

printf("testcase2: fork inherit? (cont.) \n");
printf("  expect: initial.children ~= 100MB, but child.children = 0\n");
show_rusage("initial");
if (__fork()) {
wait(&status);
} else {
show_rusage("child");
_exit(0);
}
printf("\n");

printf("testcase3: fork + malloc \n");
printf("  expect: child.self ~= initial.self + 50MB\n");
show_rusage("initial");
if (__fork()) {
wait(&status);
} else {
printf("allocate +50MB\n");
consume(50);
show_rusage("fork child");
_exit(0);
}
printf("\n");

printf("testcase4: grandchild maxrss\n");
printf("  expect: post_wait.children ~= 300MB\n");
show_rusage("initial");
if (__fork()) {
wait(&status);
show_rusage("post_wait");
} else {
system("./child -n 0 -g 300");
_exit(0);
}
printf("\n");

printf("testcase5: zombie\n");
printf("  expect: pre_wait ~= initial, IOW the zombie process is not accounted.\n");
printf("          post_wait ~= 400MB, IOW wait() collect child's max_rss. \n");
show_rusage("initial");
if (__fork()) {
sleep(1); /* children become zombie */
show_rusage("pre_wait");
wait(&status);
show_rusage("post_wait");
} else {
system("./child -n 400");
_exit(0);
}
printf("\n");

printf("testcase6: SIG_IGN\n");
printf("  expect: initial ~= after_zombie (child's 500MB alloc should be ignored).\n");
show_rusage("initial");
signal(SIGCHLD, SIG_IGN);
if (__fork()) {
sleep(1); /* children become zombie */
show_rusage("after_zombie");
} else {
system("./child -n 500");
_exit(0);
}
printf("\n");
signal(SIGCHLD, SIG_DFL);

printf("testcase7: exec (without fork) \n");
printf("  expect: initial ~= exec \n");
show_rusage("initial");
execl("./child", "child", "-v", NULL);

return 0;
}

child.c
=======
 #include <sys/types.h>
 #include <unistd.h>
 #include <sys/types.h>
 #include <sys/wait.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include <sys/types.h>
 #include <sys/time.h>
 #include <sys/resource.h>

 #include "common.h"

int main(int argc, char** argv)
{
int status;
int c;
long consume_size = 0;
long grandchild_consume_size = 0;
int show = 0;

while ((c = getopt(argc, argv, "n:g:v")) != -1) {
switch (c) {
case 'n':
consume_size = atol(optarg);
break;
case 'v':
show = 1;
break;
case 'g':

grandchild_consume_size = atol(optarg);
break;
default:
break;
}
}

if (show)
show_rusage("exec");

if (consume_size) {
printf("child alloc %ldMB\n", consume_size);
consume(consume_size);
}

if (grandchild_consume_size) {
if (fork()) {
wait(&status);
} else {
printf("grandchild alloc %ldMB\n", grandchild_consume_size);
consume(grandchild_consume_size);

exit(0);
}
}

return 0;
}

common.c
========
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include <sys/types.h>
 #include <sys/time.h>
 #include <sys/resource.h>
 #include <sys/types.h>
 #include <sys/wait.h>
 #include <unistd.h>
 #include <signal.h>
 #include <sys/mman.h>

 #include "common.h"
 #define err(str) perror(str), exit(1)

void show_rusage(char *prefix)
{
     int err, err2;
     struct rusage rusage_self;
     struct rusage rusage_children;

     printf("%s: ", prefix);
     err = getrusage(RUSAGE_SELF, &rusage_self);
     if (!err)
     printf("self %ld ", rusage_self.ru_maxrss);
     err2 = getrusage(RUSAGE_CHILDREN, &rusage_children);
     if (!err2)
     printf("children %ld ", rusage_children.ru_maxrss);

     printf("\n");
}

/* Some buggy OS need this worthless CPU waste. */
void make_pagefault(void)
{
void *addr;
int size = getpagesize();
int i;

for (i=0; i<1000; i++) {
addr = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON, -1, 0);
if (addr == MAP_FAILED)
err("make_pagefault");
memset(addr, 0, size);
munmap(addr, size);
}
}

void consume(int mega)
{
     size_t sz = mega * 1024 * 1024;
     void *ptr;

     ptr = malloc(sz);
     memset(ptr, 0, sz);
make_pagefault();
}

pid_t __fork(void)
{
pid_t pid;

pid = fork();
make_pagefault();

return pid;
}

common.h
========
void show_rusage(char *prefix);
void make_pagefault(void);
void consume(int mega);
pid_t __fork(void);

FreeBSD result (expected result)
========================================================
allocate 100MB
testcase1: fork inherit?
  expect: initial.self ~= child.self
initial: self 103492 children 0
fork child: self 103540 children 0

testcase2: fork inherit? (cont.)
  expect: initial.children ~= 100MB, but child.children = 0
initial: self 103540 children 103540
child: self 103564 children 0

testcase3: fork + malloc
  expect: child.self ~= initial.self + 50MB
initial: self 103564 children 103564
allocate +50MB
fork child: self 154860 children 0

testcase4: grandchild maxrss
  expect: post_wait.children ~= 300MB
initial: self 103564 children 154860
grandchild alloc 300MB
post_wait: self 103564 children 308720

testcase5: zombie
  expect: pre_wait ~= initial, IOW the zombie process is not accounted.
          post_wait ~= 400MB, IOW wait() collect child's max_rss.
initial: self 103564 children 308720
child alloc 400MB
pre_wait: self 103564 children 308720
post_wait: self 103564 children 411312

testcase6: SIG_IGN
  expect: initial ~= after_zombie (child's 500MB alloc should be ignored).
initial: self 103564 children 411312
child alloc 500MB
after_zombie: self 103624 children 411312

testcase7: exec (without fork)
  expect: initial ~= exec
initial: self 103624 children 411312
exec: self 103624 children 411312

Linux result (actual test result)
========================================================
allocate 100MB
testcase1: fork inherit?
  expect: initial.self ~= child.self
initial: self 102848 children 0
fork child: self 102572 children 0

testcase2: fork inherit? (cont.)
  expect: initial.children ~= 100MB, but child.children = 0
initial: self 102876 children 102644
child: self 102572 children 0

testcase3: fork + malloc
  expect: child.self ~= initial.self + 50MB
initial: self 102876 children 102644
allocate +50MB
fork child: self 153804 children 0

testcase4: grandchild maxrss
  expect: post_wait.children ~= 300MB
initial: self 102876 children 153864
grandchild alloc 300MB
post_wait: self 102876 children 307536

testcase5: zombie
  expect: pre_wait ~= initial, IOW the zombie process is not accounted.
          post_wait ~= 400MB, IOW wait() collect child's max_rss.
initial: self 102876 children 307536
child alloc 400MB
pre_wait: self 102876 children 307536
post_wait: self 102876 children 410076

testcase6: SIG_IGN
  expect: initial ~= after_zombie (child's 500MB alloc should be ignored).
initial: self 102876 children 410076
child alloc 500MB
after_zombie: self 102880 children 410076

testcase7: exec (without fork)
  expect: initial ~= exec
initial: self 102880 children 410076
exec: self 102880 children 410076

Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: KOSAKI Motohiro <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: Ingo Molnar <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

kmap_types.h: rename D macro

I tend to use a 'D' debugging macro a lot during debugging.  When I define
it before includes I often get conflicts with kmap_types.h's use of 'D'
too.  It's not very nice when a global include pollutes the name space
like this.

Rename the kmap_types.h D to KMAP_D.  It is only used temporarily in the
header so has no effect on anything else.

Signed-off-by: Andi Kleen <[email protected]>
Reviewed-by: WANG Cong <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Make sure the value in abs() does not get truncated if it is greater than 2^32

abs() will truncate the input if is it outside the 2^32 range.  Fix that
by assuming `long' input.

This might generate worse code in the common case.

Signed-off-by: Rolf Eike Beer <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

fix compat_sys_utimensat()

Compat utimensat() returns EINVAL when the tv_nsec is one of UTIME_OMIT or
UTIME_NOW and the tv_sec is set to non-zero.  As per man pages, the tv_sec
field should be ignored.

sys_utimensat() works fine in this case.

Test case:

#define _GNU_SOURCE
#define _ATFILE_SOURCE
#include <stdio.h>
#include <fcntl.h>
#include <unistd.h>
#include <sys/stat.h>
#include <stdlib.h>

main(int argc, char *argv[])
{
struct timespec ts[2];
struct timespec *tsp;

if (argc < 2) {
fprintf(stderr, "Usage : %s filename\n", argv[0]);
exit (-1);
}

ts[0].tv_nsec = ts[1].tv_nsec = UTIME_NOW;
ts[0].tv_sec = ts[1].tv_sec = 1;

tsp = ts;

if (utimensat(AT_FDCWD, argv[1],tsp,0) == -1)
perror("utimensat");
else
fprintf(stdout, "utimensat success\n");
return 0;
}
mjs22lp5:~ # cc -m64 utimensat-test.c -o utimensat_test64
mjs22lp5:~ # cc -m32 utimensat-test.c -o utimensat_test32
mjs22lp5:~ # ./utimensat_test32 /tmp/utimensat_test
utimensat: Invalid argument
mjs22lp5:~ # ./utimensat_test64 /tmp/utimensat_test
utimensat success
mjs22lp5:~ # uname -r
2.6.31-rc8

With the patch :

mjs22lp5:~ # ./utimensat_test64 /tmp/utimensat_test
utimensat success
mjs22lp5:~ # ./utimensat_test32 /tmp/utimensat_test
utimensat success
mjs22lp5:~ # uname -r
2.6.31-rc8utimensat

Signed-off-by: Suzuki K P <[email protected]>
Cc: Ulrich Drepper <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>