| Commit message (Collapse) | Author | Files | Lines |
|---|
|
Clients may still be active in the early phase of system PM, thus we
need to move the suspend operations to the late system PM phase.
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
A channel can accommodate more than one transaction, each consisting of
multiple descriptors, the last of which has the DCMD_ENDIRQEN bit set.
In order to report the channel's residue, we hence have to walk the
list of running descriptors, look for those which match the cookie,
and then try to find the descriptor which defines upper and lower
boundaries that embrace the current transport pointer. Once it is found,
walk forward until we find the descriptor that tells us about the end of
a transaction via a set DCMD_ENDIRQEN bit.
Signed-off-by: Daniel Mack <zonque@gmail.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
Make sure to handle register context save/restore when needed from
system PM callbacks.
Previously we solely trusted the device to reside in in-active state
while the system suspend callback were invoked, which is just too
optimistic.
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
Converting to the PM macros makes us simplify and remove some code.
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
While probing, don't rely on CONFIG_PM_RUNTIME to be configured.
Instead, let's power up the device and make it fully operational.
Update the runtime PM status to reflect the active state.
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
The runtime PM resume callback needs to be executed while holding the
spinlock, make sure to maintain this for the pause operation as well.
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
The commit 4501fe61 "dma: dw: Add suspend and resume handling for PCI mode
DW_DMAC." introduces system power management callbacks. Regarding to commit
f78c4cff "PM / Sleep: Add macro to define common late/early system PM
callbacks" we have nice macro to setup dev_pm_ops structure. This patch
converts a driver to use the macro.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
There is no need to use *_noirq version of suspend and resume PM callbacks. The
suspend_late / resume_early suit better (it was discussed in [1]) and in future
could be used for runtime PM support.
[1] http://www.spinics.net/lists/kernel/msg1650974.html
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
Fix mpc_dma_probe() error path and mpc_dma_remove(): manually free IRQs and
dispose IRQ mappings before devm_* takes care of other resources.
Moreover replace devm_request_irq() with request_irq() since there is no need
to use it because the original code always frees IRQ manually with
devm_free_irq(). Replace devm_free_irq() with free_irq() accordingly.
Signed-off-by: Alexander Popov <a13xp0p0v88@gmail.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
MPC512x and MPC8308 have similar DMA controllers, but are independent SoCs.
DMA controller driver should have separate 'compatible' values for these SoCs.
Signed-off-by: Alexander Popov <a13xp0p0v88@gmail.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
Concentrate the specific code for MPC8308 in the 'if' branch
and handle MPC512x in the 'else' branch.
This modification only reorders instructions but doesn't change behaviour.
Signed-off-by: Alexander Popov <a13xp0p0v88@gmail.com>
Acked-by: Anatolij Gustschin <agust@denx.de>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
These functions will be modified in the next patch in the series. By moving the
function in a patch separate from the changes, it will make review easier.
Signed-off-by: Hongbo Zhang <hongbo.zhang@freescale.com>
Signed-off-by: Qiang Liu <qiang.liu@freescale.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
There are several places where descriptors are freed using identical code.
This patch puts this code into a function to reduce code duplication.
Signed-off-by: Hongbo Zhang <hongbo.zhang@freescale.com>
Signed-off-by: Qiang Liu <qiang.liu@freescale.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
Delete attribute DMA_INTERRUPT because fsldma doesn't support this function,
exception will be thrown if talitos is used to offload xor at the same time.
Signed-off-by: Hongbo Zhang <hongbo.zhang@freescale.com>
Signed-off-by: Qiang Liu <qiang.liu@freescale.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
Methods of accessing DMA controller registers are inconsistent, some registers
are accessed by DMA_IN/OUT directly, while others are accessed by functions
get/set_* which are wrappers of DMA_IN/OUT, and even for the BCR register, it
is read by get_bcr but written by DMA_OUT.
This patch unifies the inconsistent methods, all registers are accessed by
get/set_* now.
Signed-off-by: Hongbo Zhang <hongbo.zhang@freescale.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
Some codes are calling chan_dbg with FSL_DMA_LD_DEBUG surrounded, it is really
unnecessary to use such a macro because chan_dbg is a wrapper of dev_dbg, we do
have corresponding DEBUG macro to switch on/off dev_dbg, and most of the other
codes are also calling chan_dbg directly without using FSL_DMA_LD_DEBUG.
Signed-off-by: Hongbo Zhang <hongbo.zhang@freescale.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
This patch add cyclic transfer support and enables dmaengine_prep_dma_cyclic()
Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
[reflown changelog for readablity]
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
Current shdma is using "last" which indicates last desc which needs to have
callback function. But that desc's chunks is always 1, we can use it as finder
Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
[reflown changelog for readablity]
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
Use sizeof(*var) instead of sizeof(type) when calling devm_k*alloc().
This avoids using the wrong type as was done to allocate the physical
channels array.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
As the physical channel and virtual channel point to each other,
pchan->phy->vchan is always equal to pchan. Simplify the code
accordingly.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
The property is optional and defaults to 32. Document it as such.
Cc: devicetree@vger.kernel.org
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
edma param struct is now within an edma_pset struct introduced in Thomas
Gleixner's edma tx status series. Update memcpy function for the same.
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
The granular residue accounting code uses certain variables specifically
for residue accounting. Document these in the structure declaration.
Also move around some elements and group them together.
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
The first slot in the ParamRAM of EDMA holds the current active
subtransfer. Depending on the direction we read either the source or
the destination address from there. In the internal psets we have the
address of the buffer(s).
In the cyclic case we only use the internal pset[0] which holds the
start address of the circular buffer and calculate the remaining room
to the end of the buffer.
In the SG case we read the current address and compare it to the
internal psets address and length.
- If the current address is outside of this range, the pset has been
processed already and we mark it done, update the residue_stat value
and process the next set. That avoids that we need to walk all
processed psets for every invocation of tx_status.
- If its inside the range we know that we look at the current active
set and stop the walk.
- In case of intermediate transfers we update the stats in the
interrupt callback function before starting the next batch of
transfers. The tx_status callback and the interrupt callback are
serialized via vchan.lock.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
[joelf@ti.com: Hunk #2 in original patch manually applied]
Signed-off-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
As Joel pointed out, edma_read_position() uses memcpy_fromio() to read
the parameter ram. That's not synchronized with the internal update as
it does a byte by byte copy. We need to do a 32bit read to get a
consistent value.
Further reading destination and source is pointless. In DEV_TO_MEM
transfers we are only interested in the destination, in MEM_TO_DEV we
care about the source. In MEM_TO_MEM it really does not matter which
one you read.
Simple solution: Remove the pointers, select dest/source via a bool
and return the read value.
Remove the export of this function while at it. The only potential
user is the dmaengine and that's always builtin.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Sekhar Nori <nsekhar@ti.com>
Signed-off-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
For granular accounting we need to store the direction and the
information for the individual psets:
- source or destination address, depending on direction
- length
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
Preparatory patch to support finer grained accounting.
Move the edma_params array out of edma_desc so we can add further per
pset data to it.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
[joelf@ti.com: Fixed up hunk #3 in original patch to apply]
Signed-off-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
It's likely that the caller investigates the status of a currently
active descriptor. Make that simple check first and only rumage in the
vchan list if that fails.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
The residue reporting in edma_tx_status() is just broken. It blindly
walks the psets and recalculates the lenght of the transfer from the
hardware parameters. For cyclic transfers it adds the link pset, which
results in interestingly large residues. For non-cyclic it adds the
dummy pset, which is stupid as well.
Aside of that it's silly to walk through the pset params when the per
descriptor residue is known at the point of creating it.
Store the information in edma_desc and use it.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
It helps to identify issues if we have some information regarding to the
channel which the event is associated.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Acked-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
The vchan lock in edma_callback is acquired in hard interrupt context. As
interrupts are already disabled, there's no point in save/restoring interrupt
mask bit or cpsr flags.
Get rid of flags local variable and use spin_lock instead of spin_lock_irqsave.
Signed-off-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
We add DMA memcpy support to EDMA driver. Successful tests performed using
dmatest kernel module. Copy alignment is set to DMA_SLAVE_BUSWIDTH_4_BYTES and
users must ensure length is aligned so that copy is performed fully.
Signed-off-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
In case of not supported direction it is better to print the direction also.
It is unlikely, but in such an event it helps with the debugging.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Acked-by: Joel Fernandes <joelf@ti.com>
Reviewed-and-Tested-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
callbacks
prep_slave_sg and prep_dma_cyclic callbacks have mostly same failure cases
with the same texts printed in case we hit them. It helps when debugging if
we know exactly which callback generated the errors.
At the same time change the debug level for descriptor allocation failure
from dbg to err since all other error cases are dev_err and this failure is
similarly fatal as the other ones.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Acked-by: Joel Fernandes <joelf@ti.com>
Reviewed-and-Tested-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
With the callback implemented omap-dma can provide information to client
drivers regarding to supported address widths, directions, residue
granularity, etc.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Acked-by: Joel Fernandes <joelf@ti.com>
Reviewed-and-Tested-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
Do not print the paRAM information when verbose debugging is not asked and
also reduce the number of lines printed in edma_prep_dma_cyclic()
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Acked-by: Joel Fernandes <joelf@ti.com>
Reviewed-and-Tested-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
Indicate that the edma dmaengine driver has support for cyclic mode.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Acked-by: Joel Fernandes <joelf@ti.com>
Reviewed-and-Tested-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
Pause/Resume can be used by the audio stack when the stream is paused/resumed
The edma platform code has support for this and the legacy audio stack used
this.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Acked-by: Joel Fernandes <joelf@ti.com>
Reviewed-and-Tested-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
When clients asks for maxburst = 0 it is basically the same case as if they
were asking for maxburst = 1 since in both case ASYNC need to be used and
the eDMA is expected to write/read one word per DMA request.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Acked-by: Joel Fernandes <joelf@ti.com>
Reviewed-and-Tested-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
For later use save the number of queues available for the CC.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Acked-by: Joel Fernandes <joelf@ti.com>
Reviewed-and-Tested-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
The edmacc_param struct should follow the layout of the paRAM area in the
HW. Be explicit on the size of the fields (u32) and also mark the struct
as packed to avoid any padding on non 32bit architectures.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Acked-by: Joel Fernandes <joelf@ti.com>
Reviewed-and-Tested-by: Joel Fernandes <joelf@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
|
|
In the current version, when using perf record, if something goes
wrong in tools/perf/builtin-record.c:375
session = perf_session__new(file, false, NULL);
The error message:
"Not enough memory for reading per file header"
is issued. This error message seems to be outdated and is not very
helpful. This patch proposes to replace this error message by
"Perf session creation failed"
I believe this issue has been brought to lkml:
https://lkml.org/lkml/2014/2/24/458
although this patch only tackles a (small) part of the issue.
Additionnaly, this patch improves error reporting in
tools/perf/util/data.c open_file_write.
Currently, if the call to open fails, the user is unaware of it.
This patch logs the error, before returning the error code to
the caller.
Reported-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Adrien BAK <adrien.bak@metascale.org>
Link: http://lkml.kernel.org/r/1397786443.3093.4.camel@beast
[ Reorganize the changelog into paragraphs ]
[ Added empty line after fd declaration in open_file_write ]
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
|
|
pert-report doesn't resolve function names in VDSO:
$ perf report --stdio -g flat,0.0,15,callee --sort pid
...
8.76%
0x7fff6b1fe861
__gettimeofday
ACE_OS::gettimeofday()
...
In this case symbol values should be adjusted the same way as for executables,
relocatable objects and prelinked libraries.
After fix:
$ perf report --stdio -g flat,0.0,15,callee --sort pid
...
8.76%
__vdso_gettimeofday
__gettimeofday
ACE_OS::gettimeofday()
Signed-off-by: Vladimir Nikulichev <nvs@tbricks.com>
Tested-by: Namhyung Kim <namhyung@kernel.org>
Reviewed-by: Adrian Hunter <adrian.hunter@intel.com>
Link: http://lkml.kernel.org/r/969812.163009436-sendEmail@nvs
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
|
|
Every event in the perf-kvm has a 'stats' structure, which contains
max/min/average/etc times of handling this event.
The problem is that the 'perf-kvm stat report' command always shows
that 'min time' is 0us for every event. Example:
# perf kvm stat report
Analyze events for all VCPUs:
VM-EXIT Samples Samples% Time% Min Time Max Time Avg time
[..]
0xB2 MSCH 12 0.07% 0.00% 0us 8us 7.31us ( +- 2.11% )
0xB2 CHSC 12 0.07% 0.00% 0us 18us 9.39us ( +- 9.49% )
0xB2 STPX 8 0.05% 0.00% 0us 2us 1.88us ( +- 7.18% )
0xB2 STSI 7 0.04% 0.00% 0us 44us 16.49us ( +- 38.20% )
[..]
This happens because the 'stats' structure is not initialized and
stats->min equals to 0. Lets initialize the structure for every
event after its allocation using init_stats() function. This initializes
stats->min to -1 and makes 'Min time' statistics counting work:
# perf kvm stat report
Analyze events for all VCPUs:
VM-EXIT Samples Samples% Time% Min Time Max Time Avg time
[..]
0xB2 MSCH 12 0.07% 0.00% 6us 8us 7.31us ( +- 2.11% )
0xB2 CHSC 12 0.07% 0.00% 7us 18us 9.39us ( +- 9.49% )
0xB2 STPX 8 0.05% 0.00% 1us 2us 1.88us ( +- 7.18% )
0xB2 STSI 7 0.04% 0.00% 1us 44us 16.49us ( +- 38.20% )
[..]
Signed-off-by: Alexander Yarygin <yarygin@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Link: http://lkml.kernel.org/r/1397053319-2130-3-git-send-email-borntraeger@de.ibm.com
[ Fixing the perf examples changelog output ]
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
|
|
A va_list needs to be copied in case it needs to be used twice.
Thanks to Hugh for debugging this issue, leading to various panics.
Tested:
lpq84:~# echo "|/foobar12345 %h %h %h %h %h %h %h %h %h %h %h %h %h %h %h %h %h %h %h %h" >/proc/sys/kernel/core_pattern
'produce_core' is simply : main() { *(int *)0 = 1;}
lpq84:~# ./produce_core
Segmentation fault (core dumped)
lpq84:~# dmesg | tail -1
[ 614.352947] Core dump to |/foobar12345 lpq84 lpq84 lpq84 lpq84 lpq84 lpq84 lpq84 lpq84 lpq84 lpq84 lpq84 lpq84 lpq84 lpq84 lpq84 lpq84 lpq84 lpq84 lpq84 (null) pipe failed
Notice the last argument was replaced by a NULL (we were lucky enough to
not crash, but do not try this on your production machine !)
After fix :
lpq83:~# echo "|/foobar12345 %h %h %h %h %h %h %h %h %h %h %h %h %h %h %h %h %h %h %h %h" >/proc/sys/kernel/core_pattern
lpq83:~# ./produce_core
Segmentation fault
lpq83:~# dmesg | tail -1
[ 740.800441] Core dump to |/foobar12345 lpq83 lpq83 lpq83 lpq83 lpq83 lpq83 lpq83 lpq83 lpq83 lpq83 lpq83 lpq83 lpq83 lpq83 lpq83 lpq83 lpq83 lpq83 lpq83 lpq83 pipe failed
Fixes: 5fe9d8ca21cc ("coredump: cn_vprintf() has no reason to call vsnprintf() twice")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Diagnosed-by: Hugh Dickins <hughd@google.com>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Cc: Neil Horman <nhorman@tuxdriver.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: stable@vger.kernel.org # 3.11+
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
Sasha Levin has reported two THP BUGs[1][2]. I believe both of them
have the same root cause. Let's look to them one by one.
The first bug[1] is "kernel BUG at mm/huge_memory.c:1829!". It's
BUG_ON(mapcount != page_mapcount(page)) in __split_huge_page(). From my
testing I see that page_mapcount() is higher than mapcount here.
I think it happens due to race between zap_huge_pmd() and
page_check_address_pmd(). page_check_address_pmd() misses PMD which is
under zap:
CPU0 CPU1
zap_huge_pmd()
pmdp_get_and_clear()
__split_huge_page()
anon_vma_interval_tree_foreach()
__split_huge_page_splitting()
page_check_address_pmd()
mm_find_pmd()
/*
* We check if PMD present without taking ptl: no
* serialization against zap_huge_pmd(). We miss this PMD,
* it's not accounted to 'mapcount' in __split_huge_page().
*/
pmd_present(pmd) == 0
BUG_ON(mapcount != page_mapcount(page)) // CRASH!!!
page_remove_rmap(page)
atomic_add_negative(-1, &page->_mapcount)
The second bug[2] is "kernel BUG at mm/huge_memory.c:1371!".
It's VM_BUG_ON_PAGE(!PageHead(page), page) in zap_huge_pmd().
This happens in similar way:
CPU0 CPU1
zap_huge_pmd()
pmdp_get_and_clear()
page_remove_rmap(page)
atomic_add_negative(-1, &page->_mapcount)
__split_huge_page()
anon_vma_interval_tree_foreach()
__split_huge_page_splitting()
page_check_address_pmd()
mm_find_pmd()
pmd_present(pmd) == 0 /* The same comment as above */
/*
* No crash this time since we already decremented page->_mapcount in
* zap_huge_pmd().
*/
BUG_ON(mapcount != page_mapcount(page))
/*
* We split the compound page here into small pages without
* serialization against zap_huge_pmd()
*/
__split_huge_page_refcount()
VM_BUG_ON_PAGE(!PageHead(page), page); // CRASH!!!
So my understanding the problem is pmd_present() check in mm_find_pmd()
without taking page table lock.
The bug was introduced by me commit with commit 117b0791ac42. Sorry for
that. :(
Let's open code mm_find_pmd() in page_check_address_pmd() and do the
check under page table lock.
Note that __page_check_address() does the same for PTE entires
if sync != 0.
I've stress tested split and zap code paths for 36+ hours by now and
don't see crashes with the patch applied. Before it took <20 min to
trigger the first bug and few hours for second one (if we ignore
first).
[1] https://lkml.kernel.org/g/<53440991.9090001@oracle.com>
[2] https://lkml.kernel.org/g/<5310C56C.60709@oracle.com>
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reported-by: Sasha Levin <sasha.levin@oracle.com>
Tested-by: Sasha Levin <sasha.levin@oracle.com>
Cc: Bob Liu <lliubbo@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michel Lespinasse <walken@google.com>
Cc: Dave Jones <davej@redhat.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: <stable@vger.kernel.org> [3.13+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
Fix new kernel-doc warning in mm/filemap.c:
Warning(mm/filemap.c:2600): Excess function parameter 'ppos' description in '__generic_file_aio_write'
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
This appears to be a copy/paste error. Update the description to
reflect extra rbtree debug and checks for the config option instead of
duplicating CONFIG_DEBUG_VM.
Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
Cc: Aswin Chandramouleeswaran <aswin@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
David Vrabel identified a regression when using automatic NUMA balancing
under Xen whereby page table entries were getting corrupted due to the
use of native PTE operations. Quoting him
Xen PV guest page tables require that their entries use machine
addresses if the preset bit (_PAGE_PRESENT) is set, and (for
successful migration) non-present PTEs must use pseudo-physical
addresses. This is because on migration MFNs in present PTEs are
translated to PFNs (canonicalised) so they may be translated back
to the new MFN in the destination domain (uncanonicalised).
pte_mknonnuma(), pmd_mknonnuma(), pte_mknuma() and pmd_mknuma()
set and clear the _PAGE_PRESENT bit using pte_set_flags(),
pte_clear_flags(), etc.
In a Xen PV guest, these functions must translate MFNs to PFNs
when clearing _PAGE_PRESENT and translate PFNs to MFNs when setting
_PAGE_PRESENT.
His suggested fix converted p[te|md]_[set|clear]_flags to using
paravirt-friendly ops but this is overkill. He suggested an alternative
of using p[te|md]_modify in the NUMA page table operations but this is
does more work than necessary and would require looking up a VMA for
protections.
This patch modifies the NUMA page table operations to use paravirt
friendly operations to set/clear the flags of interest. Unfortunately
this will take a performance hit when updating the PTEs on
CONFIG_PARAVIRT but I do not see a way around it that does not break
Xen.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: David Vrabel <david.vrabel@citrix.com>
Tested-by: David Vrabel <david.vrabel@citrix.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Peter Anvin <hpa@zytor.com>
Cc: Fengguang Wu <fengguang.wu@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Steven Noonan <steven@uplinklabs.net>
Cc: Rik van Riel <riel@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Cyrill Gorcunov <gorcunov@gmail.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
