Introduce support for using MemoryMapIDs (MMIDs) as an alternative to
Address Space IDs (ASIDs). The major difference between the two is that
MMIDs are global - ie. an MMID uniquely identifies an address space
across all coherent CPUs. In contrast ASIDs are non-global per-CPU IDs,
wherein each address space is allocated a separate ASID for each CPU
upon which it is used. This global namespace allows a new GINVT
instruction be used to globally invalidate TLB entries associated with a
particular MMID across all coherent CPUs in the system, removing the
need for IPIs to invalidate entries with separate ASIDs on each CPU.
The allocation scheme used here is largely borrowed from arm64 (see
arch/arm64/mm/context.c). In essence we maintain a bitmap to track
available MMIDs, and MMIDs in active use at the time of a rollover to a
new MMID version are preserved in the new version. The allocation scheme
requires efficient 64 bit atomics in order to perform reasonably, so
this support depends upon CONFIG_GENERIC_ATOMIC64=n (ie. currently it
will only be included in MIPS64 kernels).
The first, and currently only, available CPU with support for MMIDs is
the MIPS I6500. This CPU supports 16 bit MMIDs, and so for now we cap
our MMIDs to 16 bits wide in order to prevent the bitmap growing to
absurd sizes if any future CPU does implement 32 bit MMIDs as the
architecture manuals suggest is recommended.
When MMIDs are in use we also make use of GINVT instruction which is
available due to the global nature of MMIDs. By executing a sequence of
GINVT & SYNC 0x14 instructions we can avoid the overhead of an IPI to
each remote CPU in many cases. One complication is that GINVT will
invalidate wired entries (in all cases apart from type 0, which targets
the entire TLB). In order to avoid GINVT invalidating any wired TLB
entries we set up, we make sure to create those entries using a reserved
MMID (0) that we never associate with any address space.
Also of note is that KVM will require further work in order to support
MMIDs & GINVT, since KVM is involved in allocating IDs for guests & in
configuring the MMU. That work is not part of this patch, so for now
when MMIDs are in use KVM is disabled.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Cc: linux-mips@vger.kernel.org
If an mm does not have an ASID on the local CPU then drop_mmu_context()
is always redundant, since there's no context to "drop". Various callers
of drop_mmu_context() check whether the mm has been allocated an ASID
before making the call. Move that check into drop_mmu_context() and
remove it from callers to simplify them.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Cc: linux-mips@vger.kernel.org
The drop_mmu_context() function accepts a cpu argument, but it
implicitly expects that this is always equal to smp_processor_id() by
allocating & configuring an ASID on the local CPU when the mm is active
on the CPU indicated by the cpu argument.
All callers do provide the value of smp_processor_id() to the cpu
argument.
Remove the redundant argument and have drop_mmu_context() call
smp_processor_id() itself, making it clearer that the cpu variable
always represents the local CPU.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Cc: linux-mips@vger.kernel.org
There is a plan to build the kernel with -Wimplicit-fallthrough and
these places in the code produced warnings. Fix them up.
This patch produces no change in behaviour, but should be reviewed in
case these are actually bugs not intentional fallthoughs.
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Paul Burton <paul.burton@mips.com>
Cc: Kees Cook <keescook@google.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: James Hogan <jhogan@kernel.org>
Cc: linux-mips@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
For multi-node Loongson-3 (NUMA configuration), r4k_blast_scache() can
only flush Node-0's scache. So we add r4k_blast_scache_node() by using
(CAC_BASE | (node_id << NODE_ADDRSPACE_SHIFT)) instead of CKSEG0 as the
start address.
Signed-off-by: Huacai Chen <chenhc@lemote.com>
[paul.burton@mips.com: Include asm/mmzone.h from asm/r4kcache.h for
nid_to_addrbase(). Add asm/mach-generic/mmzone.h
to allow inclusion for all platforms.]
Signed-off-by: Paul Burton <paul.burton@mips.com>
Patchwork: https://patchwork.linux-mips.org/patch/21129/
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: James Hogan <james.hogan@mips.com>
Cc: Steven J . Hill <Steven.Hill@cavium.com>
Cc: linux-mips@linux-mips.org
Cc: Fuxin Zhang <zhangfx@lemote.com>
Cc: Zhangjin Wu <wuzhangjin@gmail.com>
Cc: <stable@vger.kernel.org> # 3.15+
While both option select a form of conditional dma coherence they don't
actually share any code in the implementation, so untangle them.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Paul Burton <paul.burton@mips.com>
If a driver causes DMA cache maintenance with a zero length then we
currently BUG and kill the kernel. As this is a scenario that we may
well be able to recover from, WARN & return in the condition instead.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Patchwork: https://patchwork.linux-mips.org/patch/14623/
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
In systems that include a MIPS Coherency Manager, the icache always
fills from a cache which is coherent across all CPUs. In I6400 & I6500
systems the icache fills from the dcache which is coherent across all
CPUs. In all other CM-based systems the icache fills from the L2 cache
which is shared between all cores.
This means that an icache will always see stores from remote CPUs
without needing to write them back any further than that L2, which is
what the cpu_icache_snoops_remote_store feature is used to test. In
order for it to return 1 without needing a per-platform override (which
is what Malta has relied upon so far) set the MIPS_IC_SNOOPS_REMOTE flag
when a CM is present.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/16200/
CONFIG_DMA_MAYBE_COHERENT already selects CONFIG_DMA_NONCOHERENT, so we
can remove the extra conditions.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Paul Burton <paul.burton@mips.com>
Patchwork: https://patchwork.linux-mips.org/patch/19529/
Signed-off-by: Paul Burton <paul.burton@mips.com>
Cc: Florian Fainelli <f.fainelli@gmail.com>
Cc: David Daney <david.daney@cavium.com>
Cc: Kevin Cernekee <cernekee@gmail.com>
Cc: Jiaxun Yang <jiaxun.yang@flygoat.com>
Cc: Tom Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Huacai Chen <chenhc@lemote.com>
Cc: iommu@lists.linux-foundation.org
Cc: linux-mips@linux-mips.org
When DMA will be performed to a MIPS32 1004K CPS, the L1-cache for the
range needs to be flushed and invalidated first.
The code currently takes one of two approaches.
1/ If the range is less than the size of the dcache, then HIT type
requests flush/invalidate cache lines for the particular addresses.
HIT-type requests a globalised by the CPS so this is safe on SMP.
2/ If the range is larger than the size of dcache, then INDEX type
requests flush/invalidate the whole cache. INDEX type requests affect
the local cache only. CPS does not propagate them in any way. So this
invalidation is not safe on SMP CPS systems.
Data corruption due to '2' can quite easily be demonstrated by
repeatedly "echo 3 > /proc/sys/vm/drop_caches" and then sha1sum a file
that is several times the size of available memory. Dropping caches
means that large contiguous extents (large than dcache) are more likely.
This was not a problem before Linux-4.8 because option 2 was never used
if CONFIG_MIPS_CPS was defined. The commit which removed that apparently
didn't appreciate the full consequence of the change.
We could, in theory, globalize the INDEX based flush by sending an IPI
to other cores. These cache invalidation routines can be called with
interrupts disabled and synchronous IPI require interrupts to be
enabled. Asynchronous IPI may not trigger writeback soon enough. So we
cannot use IPI in practice.
We can already test if IPI would be needed for an INDEX operation with
r4k_op_needs_ipi(R4K_INDEX). If this is true then we mustn't try the
INDEX approach as we cannot use IPI. If this is false (e.g. when there
is only one core and hence one L1 cache) then it is safe to use the
INDEX approach without IPI.
This patch avoids options 2 if r4k_op_needs_ipi(R4K_INDEX), and so
eliminates the corruption.
Fixes: c00ab4896e ("MIPS: Remove cpu_has_safe_index_cacheops")
Signed-off-by: NeilBrown <neil@brown.name>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Paul Burton <paul.burton@mips.com>
Cc: linux-mips@linux-mips.org
Cc: <stable@vger.kernel.org> # 4.8+
Patchwork: https://patchwork.linux-mips.org/patch/19259/
Signed-off-by: James Hogan <jhogan@kernel.org>
With Coherence Manager (CM) 3.5 information about the topology of the
system, which has previously only been available through & accessed from
the CM, is now also provided by the Cluster Power Controller (CPC). This
includes a new CPC_CONFIG register mirroring GCR_CONFIG, and similarly a
new CPC_Cx_CONFIG register mirroring GCR_Cx_CONFIG.
In preparation for adjusting functions such as mips_cm_numcores(), which
have previously only needed to access the CM, to also access the CPC
this patch modifies the way we use the various CPS headers. Rather than
having users include asm/mips-cm.h or asm/mips-cpc.h individually we
instead have users include asm/mips-cps.h which in turn includes
asm/mips-cm.h & asm/mips-cpc.h. This means that users will gain access
to both CM & CPC registers by including one header, and most importantly
it makes asm/mips-cps.h an ideal location for helper functions which
need to access the various components of the CPS.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/17015/
Patchwork: https://patchwork.linux-mips.org/patch/17217/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Introduce the I6500 PRID & probe it just the same way as I6400. The MIPS
I6500 is the latest in Imagination Technologies' I-Class range of CPUs,
with a focus on scalability & heterogeneity. It introduces the notion of
multiple clusters to the MIPS Coherent Processing System, allowing for a
far higher total number of cores & threads in a system when compared
with its predecessors. Clusters don't need to be identical, and may
contain differing numbers of cores & IOCUs, or cores with differing
properties.
This patch alone adds the basic support for booting Linux on an I6500
CPU without support for any of its new functionality, for which support
will be introduced in further patches.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/16190/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
If scache.waysize is 0, r4k___flush_cache_all() will do nothing and
then cause bugs. BTW, though vcache.waysize isn't being used by now,
we also fix its calculation.
Signed-off-by: Huacai Chen <chenhc@lemote.com>
Cc: John Crispin <john@phrozen.org>
Cc: Steven J . Hill <Steven.Hill@caviumnetworks.com>
Cc: Fuxin Zhang <zhangfx@lemote.com>
Cc: Zhangjin Wu <wuzhangjin@gmail.com>
Cc: linux-mips@linux-mips.org
Cc: stable@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/15756/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Physically indexed caches cannot suffer from virtual aliasing, so clear
the MIPS_CACHE_ALIASES bit in order to ensure we don't do extra work
avoiding aliasing that cannot happen.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/14017/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The L1 data cache in I6400 CPUs is indexed by physical address bits if
an entry for the address is present in the DTLB early enough in the
pipelined execution of a memory access instruction. If an entry is not
present then it's indexed by virtual address bits, but hardware will
check in a later pipeline stage when a DTLB entry has been created
whether the virtual address bits used match the physical address bits,
and if not will transparently restart the memory access instruction.
This means that although it isn't always physically indexed, it appears
so to software & we can treat the I6400 L1 data cache as being
physically indexed in order to avoid considering aliasing.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/14016/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
On some MIPS systems, a subset of devices may have DMA coherent with CPU
caches. For example in systems including a MIPS I/O Coherence Unit
(IOCU), some devices may be connected to that IOCU whilst others are
not.
Prior to this patch, we have a plat_device_is_coherent() function but no
implementation which does anything besides return a global true or
false, optionally chosen at runtime. For devices such as those described
above this is insufficient.
Fix this by tracking DMA coherence on a per-device basis with a
dma_coherent field in struct dev_archdata. Setting this from
arch_setup_dma_ops() takes care of devices which set the dma-coherent
property via device tree, and any PCI devices beneath a bridge described
in DT, automatically.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/14349/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The coherentio variable has previously been used as a boolean value,
indicating whether the user specified that coherent I/O should be
enabled or disabled. It failed to take into account the case where the
user does not specify any preference, in which case it makes sense that
we should default to coherent I/O if the hardware supports it
(hw_coherentio is non-zero).
Introduce an enum to clarify the 3 different values of coherentio & use
it throughout the code, modifying plat_device_is_coherent() &
r4k_cache_init() to take into account the default case.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Cc: Paul Burton <paul.burton@imgtec.com>
Patchwork: https://patchwork.linux-mips.org/patch/14347/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Historically a lot of these existed because we did not have
a distinction between what was modular code and what was providing
support to modules via EXPORT_SYMBOL and friends. That changed
when we forked out support for the latter into the export.h file.
This means we should be able to reduce the usage of module.h
in code that is obj-y Makefile or bool Kconfig. The advantage
in doing so is that module.h itself sources about 15 other headers;
adding significantly to what we feed cpp, and it can obscure what
headers we are effectively using.
Since module.h was the source for init.h (for __init) and for
export.h (for EXPORT_SYMBOL) we consider each obj-y/bool instance
for the presence of either and replace as needed.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/14033/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
flush_icache_range() flushes icache lines in a protected fashion for
kernel addresses, however this isn't correct with EVA where protected
cache ops only operate on user addresses, making flush_icache_range()
ineffective.
Split the implementations of __flush_icache_user_range() from
flush_icache_range(), changing the normal flush_icache_range() to use
unprotected normal cache ops.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Leonid Yegoshin <leonid.yegoshin@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/14156/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
flush_icache_range() is used for both user addresses (i.e.
cacheflush(2)), and kernel addresses (as the API documentation
describes).
This isn't really suitable however for Enhanced Virtual Addressing (EVA)
where cache operations on usermode addresses must use a different
instruction, and the protected cache ops assume user addresses, making
flush_icache_range() ineffective on kernel addresses.
Split out a new __flush_icache_user_range() and
__local_flush_icache_user_range() for users which actually want to flush
usermode addresses (note that flush_icache_user_range() already exists
on various architectures but with different arguments).
The implementation of flush_icache_range() will be changed in an
upcoming commit to use unprotected normal cache ops so as to always work
on the kernel mode address space.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Leonid Yegoshin <leonid.yegoshin@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/14152/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The EVA conditional bc_wback_inv() at the end of flush_icache_range() to
flush the modified code all the way back to RAM was apparently there for
debug purposes and to accommodate the Malta EVA configuration which
makes use of a physical alias, and didn't use the CP0_EBase.WG (Write
Gate) bit to put the exception vector in the same physical alias where
the exception vector code is written and is being flushed.
Now that CP0_EBase.WG is used, lets drop this flush.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Leonid Yegoshin <leonid.yegoshin@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/14151/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Commit f70ddc07b6 ("MIPS: c-r4k: Avoid small flush_icache_range SMP
calls") adds checks to force use of hit-type cache ops for small icache
flushes where they are globalised & index-type cache ops aren't, in
order to avoid the overhead of IPIs in those cases. However it
calculated the size of the region being flushed incorrectly, subtracting
the end address from the start address rather than the reverse. This
would have led to an overflow with size wrapping round to some large
value, and likely to the special case for avoiding IPIs not actually
being hit.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: James Hogan <james.hogan@imgtec.com>
Fixes: f70ddc07b6 ("MIPS: c-r4k: Avoid small flush_icache_range SMP calls")
Reviewed-by: James Hogan <james.hogan@imgtec.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Cc: Huacai Chen <chenhc@lemote.com>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/14211/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Pull MIPS updates from Ralf Baechle:
"This is the main pull request for MIPS for 4.8. Also includes is a
minor SSB cleanup as SSB code traditionally is merged through the MIPS
tree:
ATH25:
- MIPS: Add default configuration for ath25
Boot:
- For zboot, copy appended dtb to the end of the kernel
- store the appended dtb address in a variable
BPF:
- Fix off by one error in offset allocation
Cobalt code:
- Fix typos
Core code:
- debugfs_create_file returns NULL on error, so don't use IS_ERR for
testing for errors.
- Fix double locking issue in RM7000 S-cache code. This would only
affect RM7000 ARC systems on reboot.
- Fix page table corruption on THP permission changes.
- Use compat_sys_keyctl for 32 bit userspace on 64 bit kernels.
David says, there are no compatibility issues raised by this fix.
- Move some signal code around.
- Rewrite r4k count/compare clockevent device registration such that
min_delta_ticks/max_delta_ticks files are guaranteed to be
initialized.
- Only register r4k count/compare as clockevent device if we can
assume the clock to be constant.
- Fix MSA asm warnings in control reg accessors
- uasm and tlbex fixes and tweaking.
- Print segment physical address when EU=1.
- Define AT_VECTOR_SIZE_ARCH for ARCH_DLINFO.
- CP: Allow booting by VP other than VP 0
- Cache handling fixes and optimizations for r4k class caches
- Add hotplug support for R6 processors
- Cleanup hotplug bits in kconfig
- traps: return correct si code for accessing nonmapped addresses
- Remove cpu_has_safe_index_cacheops
Lantiq:
- Register IRQ handler for virtual IRQ number
- Fix EIU interrupt loading code
- Use the real EXIN count
- Fix build error.
Loongson 3:
- Increase HPET_MIN_PROG_DELTA and decrease HPET_MIN_CYCLES
Octeon:
- Delete built-in DTB pruning code for D-Link DSR-1000N.
- Clean up GPIO definitions in dlink_dsr-1000n.dts.
- Add more LEDs to the DSR-100n DTS
- Fix off by one in octeon_irq_gpio_map()
- Typo fixes
- Enable SATA by default in cavium_octeon_defconfig
- Support readq/writeq()
- Remove forced mappings of USB interrupts.
- Ensure DMA descriptors are always in the low 4GB
- Improve USB reset code for OCTEON II.
Pistachio:
- Add maintainers entry for pistachio SoC Support
- Remove plat_setup_iocoherency
Ralink:
- Fix pwm UART in spis group pinmux.
SSB:
- Change bare unsigned to unsigned int to suit coding style
Tools:
- Fix reloc tool compiler warnings.
Other:
- Delete use of ARCH_WANT_OPTIONAL_GPIOLIB"
* 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus: (61 commits)
MIPS: mm: Fix definition of R6 cache instruction
MIPS: tools: Fix relocs tool compiler warnings
MIPS: Cobalt: Fix typo
MIPS: Octeon: Fix typo
MIPS: Lantiq: Fix build failure
MIPS: Use CPHYSADDR to implement mips32 __pa
MIPS: Octeon: Dlink_dsr-1000n.dts: add more leds.
MIPS: Octeon: Clean up GPIO definitions in dlink_dsr-1000n.dts.
MIPS: Octeon: Delete built-in DTB pruning code for D-Link DSR-1000N.
MIPS: store the appended dtb address in a variable
MIPS: ZBOOT: copy appended dtb to the end of the kernel
MIPS: ralink: fix spis group pinmux
MIPS: Factor o32 specific code into signal_o32.c
MIPS: non-exec stack & heap when non-exec PT_GNU_STACK is present
MIPS: Use per-mm page to execute branch delay slot instructions
MIPS: Modify error handling
MIPS: c-r4k: Use SMP calls for CM indexed cache ops
MIPS: c-r4k: Avoid small flush_icache_range SMP calls
MIPS: c-r4k: Local flush_icache_range cache op override
MIPS: c-r4k: Split r4k_flush_kernel_vmap_range()
...
The MIPS Coherence Manager (CM) can propagate address-based ("hit")
cache operations to other cores in the coherent system, alleviating
software of the need to use SMP calls, however indexed cache operations
are not propagated by hardware since doing so makes no sense for
separate caches.
Update r4k_op_needs_ipi() to report that only hit cache operations are
globalized by the CM, requiring indexed cache operations to be
globalized by software via an SMP call.
r4k_on_each_cpu() previously had a special case for CONFIG_MIPS_MT_SMP,
intended to avoid the SMP calls when the only other CPUs in the system
were other VPEs in the same core, and hence sharing the same caches.
This was changed by commit cccf34e941 ("MIPS: c-r4k: Fix cache
flushing for MT cores") to apparently handle multi-core multi-VPE
systems, but it focussed mainly on hit cache ops, so the SMP calls were
still disabled entirely for CM systems.
This doesn't normally cause problems, but tests can be written to hit
these corner cases by using multiple threads, or changing task
affinities to force the process to migrate cores. For example the
failure of mprotect RW->RX to globally sync icaches (via
flush_cache_range) can be detected by modifying and mprotecting a code
page on one core, and migrating to a different core to execute from it.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paul Burton <paul.burton@imgtec.com>
Cc: Leonid Yegoshin <leonid.yegoshin@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13807/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Avoid SMP calls for flushing small icache ranges. On non-CM platforms,
and CM platforms too after we make r4k_on_each_cpu() take the cache op
type into account, it will be called on multiple CPUs due to the
possibility that local_r4k_flush_icache_range_ipi() could do
non-globalized indexed cache ops. This rougly copies the range size
check out into r4k_flush_icache_range(), which can disallow indexed
cache ops and allow r4k_on_each_cpu() to skip the SMP call.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paul Burton <paul.burton@imgtec.com>
Cc: Leonid Yegoshin <leonid.yegoshin@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13805/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Allow the permitted cache op types used by
local_r4k_flush_icache_range_ipi() to be overridden by the SMP caller.
This will allow SMP calls to be avoided under certain circumstances,
falling back to a single CPU performing globalized hit cache ops only.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paul Burton <paul.burton@imgtec.com>
Cc: Leonid Yegoshin <leonid.yegoshin@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13803/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Split the operation of r4k_flush_kernel_vmap_range() into separate
SMP callbacks for the indexed cache flush and hit cache flush cases,
since the logic to determine which to use can be determined by the
initiating CPU prior to doing any SMP calls.
This will help when we change r4k_on_each_cpu() to distinguish indexed
and hit cache ops in a later patch, preventing globalized hit cache ops
being performed redundantly on multiple CPUs.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paul Burton <paul.burton@imgtec.com>
Cc: Leonid Yegoshin <leonid.yegoshin@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13806/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
When performing SMP calls to foreign cores, exclude sibling CPUs from
the provided map, as we already handle the local core on the current
CPU. This prevents an SMP call from for example core 0, VPE 1 to VPE 0
on the same core.
In the process the cpu_foreign_map cpumask is turned into an array of
cpumasks, so that each CPU has its own version of it which excludes
sibling CPUs. r4k_op_needs_ipi() is also updated to reflect that cache
management SMP calls are not needed when all CPUs are siblings (i.e.
there are no foreign CPUs according to the new cpu_foreign_map[]
semantics which exclude siblings).
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paul Burton <paul.burton@imgtec.com>
Cc: Leonid Yegoshin <leonid.yegoshin@imgtec.com>
Cc: Felix Fietkau <nbd@nbd.name>
Cc: Jayachandran C. <jchandra@broadcom.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13801/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Several cache operations are optimised to return early from the SMP call
handler if the memory map in question has no valid ASID on the current
CPU, or any online CPU in the case of MIPS_MT_SMP. The idea is that if a
memory map has never been used on a CPU it shouldn't have cache lines in
need of flushing.
However this doesn't cover all cases when ASIDs for other CPUs need to
be checked:
- Offline VPEs may have recently been online and brought lines into the
(shared) cache, so they should also be checked, rather than only
online CPUs.
- SMP systems with a Coherence Manager (CM), but with MT disabled still
have globalized hit cache ops, but don't use SMP calls, so all present
CPUs should be taken into account.
- R6 systems have a different multithreading implementation, so
MIPS_MT_SMP won't be set, but as above may still have a CM which
globalizes hit cache ops.
Additionally for non-globalized cache operations where an SMP call to a
single VPE in each foreign core is used, it is not necessary to check
every CPU in the system, only sibling CPUs sharing the same first level
cache.
Fix this by making has_valid_asid() take a cache op type argument like
r4k_on_each_cpu(), so it can determine whether r4k_on_each_cpu() will
have done SMP calls to other cores. It can then determine which set of
CPUs to check the ASIDs of based on that, excluding foreign CPUs if an
SMP call will have been performed.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paul Burton <paul.burton@imgtec.com>
Cc: Leonid Yegoshin <leonid.yegoshin@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13804/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The r4k_on_each_cpu() function calls the specified cache flush helper on
other CPUs if deemed necessary due to the cache ops not being
globalized by hardware. However this really depends on the cache op
addressing type, as the MIPS Coherence Manager (CM) if present will
globalize "hit" cache ops (addressed by virtual address), but not
"index" cache ops (addressed by cache index). This results in index
cache ops only being performed on a single CPU when CM is present.
Most (but not all) of the functions called by r4k_on_each_cpu() perform
cache operations exclusively with a single cache op type, so add a type
argument and modify the callers to pass in some combination of R4K_HIT
(global kernel virtual addressing or user virtual addressing
conditional upon matching active_mm) and R4K_INDEX (index into cache).
This will allow r4k_on_each_cpu() to later distinguish these cases and
decide whether to perform an SMP call based on it.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paul Burton <paul.burton@imgtec.com>
Cc: Leonid Yegoshin <leonid.yegoshin@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13798/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Avoid the dcache and scache flush in local_r4k_flush_cache_sigtramp() if
the icache fills straight from the dcache.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paul Burton <paul.burton@imgtec.com>
Cc: Leonid Yegoshin <Leonid.Yegoshin@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13802/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Fix r4k_flush_cache_sigtramp() and local_r4k_flush_cache_sigtramp() to
flush the delay slot emulation trampoline cacheline through a kmap
rather than directly when the active_mm doesn't match that of the task
initiating the flush, a bit like local_r4k_flush_cache_page() does.
This would fix a corner case on SMP systems without hardware globalized
hit cache ops, where a migration to another CPU after the flush, where
that CPU did not have the same mm active at the time of the flush, could
result in stale icache content being executed instead of the trampoline,
e.g. from a previous delay slot emulation with a similar stack pointer.
This case was artificially triggered by replacing the icache flush with
a full indexed flush (not globalized on CM systems) and forcing the SMP
call to take place, with a test program that alternated two FPU delay
slots with a parent process repeatedly changing scheduler affinity.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paul Burton <paul.burton@imgtec.com>
Cc: Leonid Yegoshin <Leonid.Yegoshin@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13797/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The SMP flush_tlb_*() functions may clear the memory map's ASIDs for
other CPUs if the mm has only a single user (the current CPU) in order
to avoid SMP calls. However this makes it appear to has_valid_asid(),
which is used by various cache flush functions, as if the CPUs have
never run in the mm, and therefore can't have cached any of its memory.
For flush_tlb_mm() this doesn't sound unreasonable.
flush_tlb_range() corresponds to flush_cache_range() which does do full
indexed cache flushes, but only on the icache if the specified mapping
is executable, otherwise it doesn't guarantee that there are no cache
contents left for the mm.
flush_tlb_page() corresponds to flush_cache_page(), which will perform
address based cache ops on the specified page only, and also only
touches the icache if the page is executable. It does not guarantee that
there are no cache contents left for the mm.
For example, this affects flush_cache_range() which uses the
has_valid_asid() optimisation. It is required to flush the icache when
mappings are made executable (e.g. using mprotect) so they are
immediately usable. If some code is changed to non executable in order
to be modified then it will not be flushed from the icache during that
time, but the ASID on other CPUs may still be cleared for TLB flushing.
When the code is changed back to executable, flush_cache_range() will
assume the code hasn't run on those other CPUs due to the zero ASID, and
won't invalidate the icache on them.
This is fixed by clearing the other CPUs ASIDs to 1 instead of 0 for the
above two flush_tlb_*() functions when the corresponding cache flushes
are likely to be incomplete (non executable range flush, or any page
flush). This ASID appears valid to has_valid_asid(), but still triggers
ASID regeneration due to the upper ASID version bits being 0, which is
less than the minimum ASID version of 1 and so always treated as stale.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paul Burton <paul.burton@imgtec.com>
Cc: Leonid Yegoshin <leonid.yegoshin@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13795/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Very early versions of the 1004K had an hardware issue that made index
cache ops unsafe so they had to be avoided and hit ops be used instead.
This may significantly slow down cache maintenance operations. Only
very early FPGA versions of the 1004K were affected so let's get rid
of the workaround which was only implemented for the DMA cache
maintenance operations anyway.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The Config.VI bit specifies that the instruction cache is virtually
tagged, which is checked in c-r4k.c's probe_pcache(). Add a proper
definition for it in mipsregs.h and make use of it.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
New Loongson 3 CPU (since Loongson-3A R2, as opposed to Loongson-3A R1,
Loongson-3B R1 and Loongson-3B R2) has many enhancements, such as FTLB,
L1-VCache, EI/DI/Wait/Prefetch instruction, DSP/DSPv2 ASE, User Local
register, Read-Inhibit/Execute-Inhibit, SFB (Store Fill Buffer), Fast
TLB refill support, etc.
This patch introduce a config option, CONFIG_LOONGSON3_ENHANCEMENT, to
enable those enhancements which are not probed at run time. If you want
a generic kernel to run on all Loongson 3 machines, please say 'N'
here. If you want a high-performance kernel to run on new Loongson 3
machines only, please say 'Y' here.
Some additional explanations:
1) SFB locates between core and L1 cache, it causes memory access out
of order, so writel/outl (and other similar functions) need a I/O
reorder barrier.
2) Loongson 3 has a bug that di instruction can not save the irqflag,
so arch_local_irq_save() is modified. Since CPU_MIPSR2 is selected
by CONFIG_LOONGSON3_ENHANCEMENT, generic kernel doesn't use ei/di
at all.
3) CPU_HAS_PREFETCH is selected by CONFIG_LOONGSON3_ENHANCEMENT, so
MIPS_CPU_PREFETCH (used by uasm) probing is also put in this patch.
Signed-off-by: Huacai Chen <chenhc@lemote.com>
Cc: Aurelien Jarno <aurelien@aurel32.net>
Cc: Steven J . Hill <sjhill@realitydiluted.com>
Cc: Fuxin Zhang <zhangfx@lemote.com>
Cc: Zhangjin Wu <wuzhangjin@gmail.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/12755/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Loongson-3 maintains cache coherency by hardware, this means:
1) It's icache is coherent with dcache.
2) It's dcaches don't alias (maybe depend on PAGE_SIZE).
3) It maintains cache coherency across cores (and for DMA).
So we can skip most cache flush operations by setting relevant handlers
to `cache_noop' in `r4k_cache_init'.
Signed-off-by: Huacai Chen <chenhc@lemote.com>
Cc: Aurelien Jarno <aurelien@aurel32.net>
Cc: Steven J . Hill <sjhill@realitydiluted.com>
Cc: Fuxin Zhang <zhangfx@lemote.com>
Cc: Zhangjin Wu <wuzhangjin@gmail.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/12752/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
local_r4k___flush_cache_all() is missing a special check for BMIPS5000
processors, we need to blast the S-cache, just like other MTI processors
since we have an inclusive cache. We also need an additional __sync() to
make sure this is completed.
Fixes: d74b0172e4 ("MIPS: BMIPS: Add special cache handling in c-r4k.c")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13012/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
BMIPS5000 and BMIPS5200 processor have no D cache aliases, and this is
properly handled by the per-CPU override added at the end of
r4k_cache_init(), the problem is that the output of probe_pcache()
disagrees with that, since this is too late:
Primary instruction cache 32kB, VIPT, 4-way, linesize 64 bytes.
Primary data cache 32kB, 4-way, VIPT, cache aliases, linesize 32 bytes
With the change moved earlier, we now have a consistent output with the
settings we are intending to have:
Primary instruction cache 32kB, VIPT, 4-way, linesize 64 bytes.
Primary data cache 32kB, 4-way, VIPT, no aliases, linesize 32 bytes
Fixes: d74b0172e4 ("MIPS: BMIPS: Add special cache handling in c-r4k.c")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13011/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
BMIPS5000 and BMIPS52000 processors have their I-cache filling from the
D-cache. Since BMIPS_GENERIC does not provide (yet) a
cpu-feature-overrides.h file, this was not set anywhere, so make sure
the R4K cache detection takes care of that.
Fixes: d74b0172e4 ("MIPS: BMIPS: Add special cache handling in c-r4k.c")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13010/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Add casses supporting the M6250 CPU to various switch statements in the
core MIPS kernel code that define behaviour dependent upon the CPU.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: Joshua Kinard <kumba@gentoo.org>
Cc: Leonid Yegoshin <Leonid.Yegoshin@imgtec.com>
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Maciej W. Rozycki <macro@codesourcery.com>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/12374/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Add cases supporting the P6600 CPU to various switch statements in
core MIPS kernel code that define behaviour dependent upon the CPU.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: Maciej W. Rozycki <macro@imgtec.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Joshua Kinard <kumba@gentoo.org>
Cc: Andrzej Hajda <a.hajda@samsung.com>
Cc: Leonid Yegoshin <Leonid.Yegoshin@imgtec.com>
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: James Hogan <james.hogan@imgtec.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Petri Gynther <pgynther@google.com>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/12343/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Coherence Manager 3 (CM3) as present in I6400 can fill icache lines
effectively from dirty dcaches, so there is no need to flush dirty lines
from dcaches through to L2 prior to icache invalidation.
Set the MIPS_CACHE_IC_F_DC flag such that cpu_has_ic_fills_f_dc
evaluates to true, which avoids those dcache flushes.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Leonid Yegoshin <leonid.yegoshin@imgtec.com>
Cc: Manuel Lauss <manuel.lauss@gmail.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/12180/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
It is still necessary to handle icache coherency in flush_cache_range()
and copy_to_user_page() when the icache fills from the dcache, even
though the dcache does not need to be written back. However when this
handling was added in commit 2eaa7ec286 ("[MIPS] Handle I-cache
coherency in flush_cache_range()"), it did not do any icache flushing
when it fills from dcache.
Therefore fix r4k_flush_cache_range() to run
local_r4k_flush_cache_range() without taking into account whether icache
fills from dcache, so that the icache coherency gets handled. Checks are
also added in local_r4k_flush_cache_range() so that the dcache blast
doesn't take place when icache fills from dcache.
A test to mmap a page PROT_READ|PROT_WRITE, modify code in it, and
mprotect it to VM_READ|VM_EXEC (similar to case described in above
commit) can hit this case quite easily to verify the fix.
A similar check was added in commit f8829caee3 ("[MIPS] Fix aliasing
bug in copy_to_user_page / copy_from_user_page"), so also fix
copy_to_user_page() similarly, to call flush_cache_page() without taking
into account whether icache fills from dcache, since flush_cache_page()
already takes that into account to avoid performing a dcache flush.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Leonid Yegoshin <leonid.yegoshin@imgtec.com>
Cc: Manuel Lauss <manuel.lauss@gmail.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/12179/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Let's define page_mapped() to be true for compound pages if any
sub-pages of the compound page is mapped (with PMD or PTE).
On other hand page_mapcount() return mapcount for this particular small
page.
This will make cases like page_get_anon_vma() behave correctly once we
allow huge pages to be mapped with PTE.
Most users outside core-mm should use page_mapcount() instead of
page_mapped().
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Tested-by: Sasha Levin <sasha.levin@oracle.com>
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: Jerome Marchand <jmarchan@redhat.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Steve Capper <steve.capper@linaro.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add a CPU_I6400 case to various switch statements, doing the same thing
as for CPU_P5600.
Signed-off-by: Markos Chandras <markos.chandras@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/10635/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
