Commit af3e4fd37a "ARM: 6859/1: Add writethrough dcache support for
ARM926EJS processor" broke Thumb2 compilation by omitting to maintain
the wide encoding for the added branch instructions which made the
ARM926EJ-S record smaller than expected, breaking the record walk code.
Signed-off-by: Nicolas Pitre <[email protected]>
Cc: Mark A. Greer <[email protected]>
Signed-off-by: Russell King <[email protected]>
If the DT physical address is zero, this is equivalent to no DT.
Especially when the actual RAM physical address is not located at zero,
the result of phys_to_virt() would point to la-la-land and crash the
kernel, which crash is completely silent this early during boot.
Signed-off-by: Nicolas Pitre <[email protected]>
Acked-by: Grant Likely <[email protected]>
Signed-off-by: Russell King <[email protected]>
To be able to relocate the .bss section at run time independently from
the rest of the code, we must make sure that no GOTOFF relocations are
used with .bss symbols. This usually means that no global variables can
be marked static unless they're also const.
To enforce this, suffice to fail the build whenever a private symbol
is allocated to .bss and list those symbols for convenience.
The user_stack and user_stack_end labels in head.S were converted into
non exported symbols to remove false positives.
Signed-off-by: Nicolas Pitre <[email protected]>
Tested-by: Tony Lindgren <[email protected]>
To be able to relocate the .bss section at run time independently from
the rest of the code, we must make sure that no GOTOFF relocations are
used with .bss symbols. This usually means that no global variables can
be marked static unless they're also const.
Let's remove the static qualifier from current offenders, or turn them
into const variables when possible. Next commit will ensure the build
fails if one of those is reintroduced due to otherwise enforced coding
standards for the kernel.
Signed-off-by: Nicolas Pitre <[email protected]>
Tested-by: Tony Lindgren <[email protected]>
If decompress() returns an error without calling error(), we must
not attempt to boot the resulting kernel.
Signed-off-by: Nicolas Pitre <[email protected]>
Tested-by: Shawn Guo <[email protected]>
Tested-by: Tony Lindgren <[email protected]>
The return value for decompress_kernel() is no longer used. Furthermore,
this was obtained and stored in a variable called output_ptr which is
a complete misnomer for what is actually the size of the decompressed
kernel image. Let's get rid of it.
Signed-off-by: Nicolas Pitre <[email protected]>
Tested-by: Shawn Guo <[email protected]>
Tested-by: Tony Lindgren <[email protected]>
In commit d239b1dc09 the hardcoded 4x estimate for the decompressed
kernel size was replaced by the exact Image file size and passed to
the linker as a symbol value. Turns out that this is unneeded as the
size is already included at the end of the compressed piggy data.
For those compressed formats that don't include this data, the build
system already takes care of appending it using size_append in
scripts/Makefile.lib. So let's use that instead.
Signed-off-by: Nicolas Pitre <[email protected]>
Tested-by: Shawn Guo <[email protected]>
Tested-by: Tony Lindgren <[email protected]>
For correctness, the initial page table located right before the
decompressed kernel should be considered when determining if relocation
is required.
Signed-off-by: Nicolas Pitre <[email protected]>
Tested-by: Shawn Guo <[email protected]>
Acked-by: Tony Lindgren <[email protected]>
If the zImage load address is slightly below the relocation address,
there is a risk for the copied data to overwrite the copy loop or
cache flush code that the relocation process requires. Always
bump the relocation address by the size of that code to avoid this
issue.
Noticed by Tony Lindgren <[email protected]>.
While at it, let's start the copy from the restart symbol which makes
the above code size computation possible by the assembler directly
(same sections), given that we don't need to preserve the code before
that point anyway. And therefore we don't need to carry the _start
pointer in r5 anymore.
Signed-off-by: Nicolas Pitre <[email protected]>
Tested-by: Tony Lindgren <[email protected]>
With ARMv5+ and EABI, the compiler expects a 64-bit aligned stack so
instructions like STRD and LDRD can be used. Without this, mysterious
boot failures were seen semi randomly with the LZMA decompressor.
While at it, let's align .bss as well.
Signed-off-by: Nicolas Pitre <[email protected]>
Tested-by: Shawn Guo <[email protected]>
Acked-by: Tony Lindgren <[email protected]>
CC: [email protected]
The Marvell PJ4 is ARMv7 capable, so we don't support it in
ARMv6 mode anymore.
Signed-off-by: Nicolas Pitre <[email protected]>
Acked-by: Saeed Bishara <[email protected]>
Acked-by: Haojian Zhuang <[email protected]>
There are optional bits that may complement a personality ID. It is
therefore wrong to simply test against the absolute current->personality
value to determine the effective personality. The PER_LINUX_32BIT is
itself just PER_LINUX with one of those optional bits set.
Signed-off-by: Nicolas Pitre <[email protected]>
Signed-off-by: Russell King <[email protected]>
Our SET_PERSONALITY() implementation was overwriting all existing
personality flags, including ADDR_NO_RANDOMIZE, making them unavailable
to processes being exec'd after a call to personality() in user space.
This prevents the gdb test suite from running successfully.
Signed-off-by: Nicolas Pitre <[email protected]>
Signed-off-by: Russell King <[email protected]>
Adding Thumb2 support to the runtime patching of the virt_to_phys and
phys_to_virt opcodes.
Tested both the 8-bit and the 16-bit fixups, using different placements
in memory to exercize all code paths.
Signed-off-by: Nicolas Pitre <[email protected]>
Reviewed-by: Dave Martin <[email protected]>
Signed-off-by: Russell King <[email protected]>
The simply expanded variable may be evaluated before the target file for
the stat command is up to date or even exists. Switching to a recursively
expanded variable move the execution of the stat command to the location
where LDFLAGS_vmlinux is actually used, fixing the dependency issue
introduced by patch #6746/1.
Signed-off-by: Nicolas Pitre <[email protected]>
Signed-off-by: Russell King <[email protected]>
We currently presume a 4x expansion to guess the decompressed kernel size
in order to determine if the decompressed kernel is in conflict with
the location where zImage is loaded. This guess may cause many issues
by overestimating the final kernel image size:
- This may force a needless relocation if the location of zImage was
fine, wasting some precious microseconds of boot time.
- The relocation may be located way too far, possibly overwriting the
initrd image in RAM.
- If the kernel image includes a large already-compressed initramfs image
then the problem is even more exacerbated.
And if by some strange means the 4x guess is too low then we may overwrite
ourselves with the decompressed image.
So let's use the exact decompressed kernel image size instead. For that
we need to rely on the stat command, but this is hardly a new build
dependency as the kernel already depends on many external commands
to be built provided by the coreutils package where stat is found.
Signed-off-by: Nicolas Pitre <[email protected]>
Signed-off-by: Russell King <[email protected]>
In the case of a conflict between the memory used by the compressed
kernel with its decompressor code and the memory used for the
decompressed kernel, we currently store the later after the former and
relocate it afterwards.
This would be more efficient to do this the other way around i.e.
relocate the compressed data up front instead, resulting in a smaller
copy. That also has the advantage of making the code smaller and more
straight forward.
Signed-off-by: Nicolas Pitre <[email protected]>
Signed-off-by: Russell King <[email protected]>
Some installers would binary patch the kernel zImage to replace the
first few nops with custom instructions. This breaks the Thumb2 kernel
as the mode switch is right at the beginning. Let's move it towards the
end of the nop sequence instead.
Signed-off-by: Nicolas Pitre <[email protected]>
Signed-off-by: Russell King <[email protected]>
In commit e616c59140, highmem support was
deactivated for SMP platforms without hardware TLB ops broadcast because
usage of kmap_high_get() requires that IRQs be disabled when kmap_lock
is locked which is incompatible with the IPI mechanism used by the
software TLB ops broadcast invoked through flush_all_zero_pkmaps().
The reason for kmap_high_get() is to ensure that the currently kmap'd
page usage count does not decrease to zero while we're using its
existing virtual mapping in an atomic context. With a VIVT cache this
is essential to do due to cache coherency issues, but with a VIPT cache
this is only an optimization so not to pay the price of establishing a
second mapping if an existing one can be used. However, on VIPT
platforms without hardware TLB maintenance we can give up on that
optimization in order to be able to use highmem.
From ARMv7 onwards the TLB ops are broadcasted in hardware, so let's
disable ARCH_NEEDS_KMAP_HIGH_GET only when CONFIG_SMP and
CONFIG_CPU_TLB_V6 are defined.
Signed-off-by: Nicolas Pitre <[email protected]>
Tested-by: Saeed Bishara <[email protected]>
Signed-off-by: Russell King <[email protected]>
The initial MMU table created in head.S contains a 1 MB mapping at the
start of memory to let the early kernel boot code access the boot params
specified by mdesc->boot_params.
When using CONFIG_ARM_PATCH_PHYS_VIRT it is possible for the kernel to
have a different idea of where the start of memory is at run time, making
the compile-time determined mdesc->boot_params pointing to a memory area
which is not mapped. Any access to the boot params in that case will
fault and silently hang the kernel at that point. It is therefore a
better idea to simply ignore mdesc->boot_params in that case and give
the kernel a chance to print some diagnostic on the console later.
If the bootloader provides a valid pointer in r2 to the kernel then this
is used instead of mdesc->boot_params, and an explicit mapping is already
created in the initial MMU table for it. It is therefore a good idea to
use that facility when using a relocated kernel.
Signed-off-by: Nicolas Pitre <[email protected]>
Signed-off-by: Russell King <[email protected]>
Since commit 6fc31d54 there is no callers for lookup_machine_type()
other than setup_machine(). And if the former fails it won't return,
therefore the error path in the later is dead code. Let's clean
things up by merging them together.
Signed-off-by: Nicolas Pitre <[email protected]>
Signed-off-by: Russell King <[email protected]>
Marcin Slusarz says:
> In arch/arm/kernel/kprobes-decode.c there's a function
> arm_kprobe_decode_insn which does:
>
> } else if ((insn & 0x0e000000) == 0x0c400000) {
> ...
>
> This is always false, so code below is dead.
> I found this bug by coccinelle (http://coccinelle.lip6.fr/).
Reported-by: Marcin Slusarz <[email protected]>
Signed-off-by: Nicolas Pitre <[email protected]>
Signed-off-by: Russell King <[email protected]>
The minsec argument to clocks_calc_mult_shift() is misnamed. It is used
to clamp the magnitude of the mult factor so that a multiplication with
any value in the given range won't overflow a 64 bit result. Let's
rename it to match the actual usage.
Signed-off-by: Nicolas Pitre <[email protected]>
Acked-by: John Stultz <[email protected]>
LKML-Reference: <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
The cnt32_to_63 algorithm relies on proper counter data evaluation
ordering to work properly. This was missing from the provided
documentation.
Let's augment the documentation with the missing usage constraint and
fix the only instance that got it wrong.
Signed-off-by: Nicolas Pitre <[email protected]>
Acked-by: David Howells <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Since commit 3e4d3af501 "mm: stack based kmap_atomic()", it is actively
wrong to rely on fixed kmap type indices (namely KM_L2_CACHE) as
kmap_atomic() totally ignores them and a concurrent instance of it may
happily reuse any slot for any purpose. Because kmap_atomic() is now
able to deal with reentrancy, we can get rid of the ad hoc mapping here.
While the code is made much simpler, there is a needless cache flush
introduced by the usage of __kunmap_atomic(). It is not clear if the
performance difference to remove that is worth the cost in code
maintenance (I don't think there are that many highmem users on that
platform anyway) but that should be reconsidered when/if someone cares
enough to do some measurements.
Signed-off-by: Nicolas Pitre <[email protected]>
Since commit 3e4d3af501 "mm: stack based kmap_atomic()", it is actively
wrong to rely on fixed kmap type indices (namely KM_L2_CACHE) as
kmap_atomic() totally ignores them and a concurrent instance of it may
happily reuse any slot for any purpose. Because kmap_atomic() is now
able to deal with reentrancy, we can get rid of the ad hoc mapping here,
and we even don't have to disable IRQs anymore (highmem case).
While the code is made much simpler, there is a needless cache flush
introduced by the usage of __kunmap_atomic(). It is not clear if the
performance difference to remove that is worth the cost in code
maintenance (I don't think there are that many highmem users on that
platform if at all anyway).
Signed-off-by: Nicolas Pitre <[email protected]>
Since commit 3e4d3af501 "mm: stack based kmap_atomic()", it is no longer
necessary to carry an ad hoc version of kmap_atomic() added in commit
7e5a69e83b "ARM: 6007/1: fix highmem with VIPT cache and DMA" to cope
with reentrancy.
In fact, it is now actively wrong to rely on fixed kmap type indices
(namely KM_L1_CACHE) as kmap_atomic() totally ignores them now and a
concurrent instance of it may reuse any slot for any purpose.
Signed-off-by: Nicolas Pitre <[email protected]>
A notifier chain is called whenever the vt code modifies a terminal
content, except for one case which is when the modification comes
through writes to /dev/vcs* devices. Let's add the missing notifier
invocation at the end of vcs_write() for that case too.
Signed-off-by: Nicolas Pitre <[email protected]>
Cc: Alan Cox <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
The /dev/vcs* devices are used, amongst other things, by accessibility
applications such as BRLTTY to display the screen content onto refreshable
braille displays. Currently this is performed by constantly reading from
/dev/vcsa0 whether or not the screen content has changed. Given the
default braille refresh rate of 25 times per second, this easily qualifies
as the biggest source of wake-up events preventing laptops from entering
deeper power saving states.
To avoid this periodic polling, let's add support for select()/poll() and
SIGIO with the /dev/vcs* devices. The implemented semantic is to report
data availability whenever the corresponding vt has seen some update after
the last read() operation. The application still has to lseek() back
as usual in order to read() the new data.
Not to create unwanted overhead, the needed data structure is allocated
and the vt notification callback is registered only when the poll or
fasync method is invoked for the first time per file instance.
Signed-off-by: Nicolas Pitre <[email protected]>
Cc: Andrew Morton <[email protected]>
Acked-by: Alan Cox <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
Commit 21f0ba90a4 "orion/kirkwood: reset PCIe unit on boot" made the
reset of the PCIe unit unconditional. While this may fix problems on some
targets, this also causes problems on other targets.
Saeed Bishara <[email protected]> said about the original problem: "We
couln't pinpoint the root cause of this issue, actually we failed to
reproduce that issue."
So let's restrict the reset of the PCIe unit only to the target where
the original problem was observed.
Signed-off-by: Nicolas Pitre <[email protected]>
Since we're now using addruart to establish the debug mapping, we can
remove the io_pg_offst and phys_io members of struct machine_desc.
The various declarations were removed using the following script:
grep -rl MACHINE_START arch/arm | xargs \
sed -i '/MACHINE_START/,/MACHINE_END/ { /\.\(phys_io\|io_pg_offst\)/d }'
[ Initial patch was from Jeremy Kerr, example script from Russell King ]
Signed-off-by: Nicolas Pitre <[email protected]>
Acked-by: Eric Miao <eric.miao at canonical.com>
As mentioned in the comment right at the top, the first four fields
are directly accessed by assembly code in head.S. Move nr_irqs so the
comment is true again.
Signed-off-by: Nicolas Pitre <[email protected]>
Since commits 990cb8acf2 and cc92c28b2d, it is possible to have full
address space layout randomization (ASLR) on ARM. Except that one small
change was missing for ASLR of PIE executables.
Signed-off-by: Nicolas Pitre <[email protected]>
Signed-off-by: Russell King <[email protected]>
The kernel makes the high vector page visible to user space. This page
contains (amongst others) small code segments that can be executed in
user space. Make this page visible through ptrace and /proc/<pid>/mem
in order to let gdb perform code parsing needed for proper unwinding.
For example, the ERESTART_RESTARTBLOCK handler actually has a stack
frame -- it returns to a PC value stored on the user's stack. To
unwind after a "sleep" system call was interrupted twice, GDB would
have to recognize this situation and understand that stack frame
layout -- which it currently cannot do.
We could fix this by hard-coding addresses in the vector page range into
GDB, but that isn't really portable as not all of those addresses are
guaranteed to remain stable across kernel releases. And having the gdb
process make an exception for this page and get content from its own
address space for it looks strange, and it is not future proof either.
Being located above PAGE_OFFSET, this vma cannot be deleted by
user space code.
Signed-off-by: Nicolas Pitre <[email protected]>
There are very few legitimate use cases, if any, for directly accessing
system RAM through /dev/mem. So let's mimic what they do on x86 and
forbid it when CONFIG_STRICT_DEVMEM is turned on.
Signed-off-by: Nicolas Pitre <[email protected]>
VMALLOC_END is supposed to be an absolute value, while PAGE_OFFSET may
vary depending on the selected user:kernel memory split mode through
CONFIG_VMSPLIT_*. In fact, the goal of moving PAGE_OFFSET down is to
accommodate more directly addressed RAM by the kernel below the vmalloc
area, and having VMALLOC_END move along PAGE_OFFSET is rather against
the very reason why PAGE_OFFSET can be moved in the first place.
Signed-off-by: Nicolas Pitre <[email protected]>
This allows for board specific issues to override decisions made in generic
code that might not be suitable due to some errata or the like, by making
the initcall hooks from those board specific files run after the core ones,
therefore avoiding ugly #ifdef's in core code.
Signed-off-by: Nicolas Pitre <[email protected]>
Tested-by: Dave Martin <[email protected]>
Tested-by: Jason Hui <[email protected]>
When the policy for user space is to ignore misaligned accesses from user
space, the processor then performs a documented rotation on the accessed
data. This is the result of the access being trapped, and the kernel
disabling the alignment trap before returning to user space again.
In kernel space we always want misaligned accesses to be fixed up. This
is enforced by always re-enabling the alignment trap on every entry into
kernel space from user space. No such re-enabling is performed when an
exception occurs while already in kernel space as the alignment trap is
always supposed to be enabled in that case.
There is however a small race window when a misaligned access in user
space is trapped and the alignment trap disabled, but the CPU didn't
return to user space just yet. Any exception would be entered from kernel
space at that point and the kernel would then execute with the alignment
trap disabled.
Thanks to Maxime Bizon <[email protected]> for providing a test module
that made this issue reproducible.
Signed-off-by: Nicolas Pitre <[email protected]>
Signed-off-by: Russell King <[email protected]>
A new random value for the canary is stored in the task struct whenever
a new task is forked. This is meant to allow for different canary values
per task. On ARM, GCC expects the canary value to be found in a global
variable called __stack_chk_guard. So this variable has to be updated
with the value stored in the task struct whenever a task switch occurs.
Because the variable GCC expects is global, this cannot work on SMP
unfortunately. So, on SMP, the same initial canary value is kept
throughout, making this feature a bit less effective although it is still
useful.
One way to overcome this GCC limitation would be to locate the
__stack_chk_guard variable into a memory page of its own for each CPU,
and then use TLB locking to have each CPU see its own page at the same
virtual address for each of them.
Signed-off-by: Nicolas Pitre <[email protected]>
This is the very basic stuff without the changing canary upon
task switch yet. Just the Kconfig option and a constant canary
value initialized at boot time.
Signed-off-by: Nicolas Pitre <[email protected]>
For this feature to take effect, CONFIG_COMPAT_BRK must be turned
off. This can safely be turned off for any EABI user space versions.
Signed-off-by: Nicolas Pitre <[email protected]>
When CONFIG_DEBUG_HIGHMEM is used, the fixmap entry used for a highmem page
by kmap_atomic() is always cleared by kunmap_atomic(). This helps find
bad usages such as dereferences after the unmap, or overflow into the
adjacent fixmap areas.
But this debugging aid is completely bypassed when a kmap for the same
page already exists as the kmap is reused instead. ON VIVT systems we
have no choice but to reuse that kmap due to cache coherency issues,
but on non VIVT systems we should always force the fixmap usage when
debugging is active.
Signed-off-by: Nicolas Pitre <[email protected]>
Signed-off-by: Russell King <[email protected]>
The VIVT cache of a highmem page is always flushed before the page
is unmapped. This cache flush is explicit through flush_cache_kmaps()
in flush_all_zero_pkmaps(), or through __cpuc_flush_dcache_area() in
kunmap_atomic(). There is also an implicit flush of those highmem pages
that were part of a process that just terminated making those pages free
as the whole VIVT cache has to be flushed on every task switch. Hence
unmapped highmem pages need no cache maintenance in that case.
However unmapped pages may still be cached with a VIPT cache because the
cache is tagged with physical addresses. There is no need for a whole
cache flush during task switching for that reason, and despite the
explicit cache flushes in flush_all_zero_pkmaps() and kunmap_atomic(),
some highmem pages that were mapped in user space end up still cached
even when they become unmapped.
So, we do have to perform cache maintenance on those unmapped highmem
pages in the context of DMA when using a VIPT cache. Unfortunately,
it is not possible to perform that cache maintenance using physical
addresses as all the L1 cache maintenance coprocessor functions accept
virtual addresses only. Therefore we have no choice but to set up a
temporary virtual mapping for that purpose.
And of course the explicit cache flushing when unmapping a highmem page
on a system with a VIPT cache now can go, which should increase
performance.
While at it, because the code in __flush_dcache_page() has to be modified
anyway, let's also make sure the mapped highmem pages are pinned with
kmap_high_get() for the duration of the cache maintenance operation.
Because kunmap() does unmap highmem pages lazily, it was reported by
Gary King <[email protected]> that those pages ended up being unmapped
during cache maintenance on SMP causing segmentation faults.
Signed-off-by: Nicolas Pitre <[email protected]>
Signed-off-by: Russell King <[email protected]>
Some SDIO cards may suspend while keeping function interrupts active
especially in the powered suspend case. Upon resume we need to kick the
SDIO interrupt thread to check for pending interrupts and to restart card
IRQ detection at the host controller level.
Signed-off-by: Nicolas Pitre <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This patch series provides the core changes needed to allow SDIO cards to
remain powered and active while the host system is suspended, and let them
wake up the host system when needed. This is used to implement
wake-on-lan with SDIO wireless cards at the moment. Patches to add that
support to the libertas driver will be posted separately.
This patch:
Some SDIO cards have the ability to keep on running autonomously when the
host system is suspended, and wake it up when needed. This however
requires that the host controller preserve power to the card, and
configure itself appropriately for wake-up.
There is however 4 layers of abstractions involved: the host controller
driver, the MMC core code, the SDIO card management code, and the actual
SDIO function driver. To make things simple and manageable, host drivers
must advertise their PM capabilities with a feature bitmask, then function
drivers can query and set those features from their suspend method. Then
each layer in the suspend call chain is expected to act upon those bits
accordingly.
[[email protected]: fix typo in comment]
Signed-off-by: Nicolas Pitre <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This is still in use especially to develop SDIO device drivers on laptop
machines which are lacking SDIO slots. This adapter supports SDIO cards
only due to lack of 136-bit response capability.
Signed-off-by: Nicolas Pitre <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
There is not enough users to warrant its existence, and it is actually
an obstacle to progress with the new DMA API which cannot cover this
case properly.
To keep backward compatibility, let's perform the necessary custom
cache maintenance locally in the only driver affected.
Signed-off-by: Nicolas Pitre <[email protected]>
Signed-off-by: Russell King <[email protected]>
Now... testing reveals that the very first patch "sdio_uart: use
tty_port" causes a segmentation fault in sdio_uart_open():
Unable to handle kernel NULL pointer dereference at virtual address 00000084
pgd = dfb44000 [00000084] *pgd=1fb99031, *pte=00000000, *ppte=00000000
Internal error: Oops: 17 [#1] PREEMPT
last sysfs file:
/sys/devices/platform/mvsdio/mmc_host/mmc0/mmc0:f111/uevent
Modules linked in:
CPU: 0 Not tainted (2.6.32-rc5-next-20091102-00001-gb36eae9 #10)
PC is at sdio_uart_open+0x204/0x2cc
[...]
Signed-off-by: Alan Cox <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
That code was refactored a long time ago, but one particular label
didn't get adjusted properly which broke the listing of supported
machines.
Signed-off-by: Nicolas Pitre <[email protected]>
Signed-off-by: Russell King <[email protected]>
Commit 200b812d00 "Clear the exclusive monitor when returning from an
exception" broke the vast majority of ARM systems in the wild which are
still pre ARMv6. The kernel is crashing on the first occurrence of an
exception due to the removal of the actual return instruction for them.
Let's add it back.
Signed-off-by: Nicolas Pitre <[email protected]>
Acked-by: Uwe Kleine-König <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
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]>
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]>
Nicolas Pitre