| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
Allocate subpage protect related variables only if we use the feature.
This helps in reducing the hash related mm context struct by around 4K
Before the patch
sizeof(struct hash_mm_context) = 8288
After the patch
sizeof(struct hash_mm_context) = 4160
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Currently, our mm_context_t on book3s64 include all hash specific
context details like slice mask and subpage protection details. We
can skip allocating these with radix translation. This will help us to save
8K per mm_context with radix translation.
With the patch applied we have
sizeof(mm_context_t) = 136
sizeof(struct hash_mm_context) = 8288
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
In preparation of next patch which generalises the use of
pte_fragment_alloc() for all, this patch moves the related functions
in a place that is common to all subarches.
The 8xx will need that for supporting 16k pages, as in that mode
page tables still have a size of 4k.
Since pte_fragment with only once fragment is not different
from what is done in the general case, we can easily migrate all
subarchs to pte fragments.
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
When switching processes, currently all user SLBEs are cleared, and a
few (exec_base, pc, and stack) are preloaded. In trivial testing with
small apps, this tends to miss the heap and low 256MB segments, and it
will also miss commonly accessed segments on large memory workloads.
Add a simple round-robin preload cache that just inserts the last SLB
miss into the head of the cache and preloads those at context switch
time. Every 256 context switches, the oldest entry is removed from the
cache to shrink the cache and require fewer slbmte if they are unused.
Much more could go into this, including into the SLB entry reclaim
side to track some LRU information etc, which would require a study of
large memory workloads. But this is a simple thing we can do now that
is an obvious win for common workloads.
With the full series, process switching speed on the context_switch
benchmark on POWER9/hash (with kernel speculation security masures
disabled) increases from 140K/s to 178K/s (27%).
POWER8 does not change much (within 1%), it's unclear why it does not
see a big gain like POWER9.
Booting to busybox init with 256MB segments has SLB misses go down
from 945 to 69, and with 1T segments 900 to 21. These could almost all
be eliminated by preloading a bit more carefully with ELF binary
loading.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
This will be used by the SLB code in the next patch, but for now this
sets the slb_addr_limit to the correct size for 32-bit tasks.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
This reverts commits:
5e46e29e6a97 ("powerpc/64s/hash: convert SLB miss handlers to C")
8fed04d0f6ae ("powerpc/64s/hash: remove user SLB data from the paca")
655deecf67b2 ("powerpc/64s/hash: SLB allocation status bitmaps")
2e1626744e8d ("powerpc/64s/hash: provide arch_setup_exec hooks for hash slice setup")
89ca4e126a3f ("powerpc/64s/hash: Add a SLB preload cache")
This series had a few bugs, and the fixes are not all trivial. So
revert most of it for now.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
When switching processes, currently all user SLBEs are cleared, and a
few (exec_base, pc, and stack) are preloaded. In trivial testing with
small apps, this tends to miss the heap and low 256MB segments, and it
will also miss commonly accessed segments on large memory workloads.
Add a simple round-robin preload cache that just inserts the last SLB
miss into the head of the cache and preloads those at context switch
time. Every 256 context switches, the oldest entry is removed from the
cache to shrink the cache and require fewer slbmte if they are unused.
Much more could go into this, including into the SLB entry reclaim
side to track some LRU information etc, which would require a study of
large memory workloads. But this is a simple thing we can do now that
is an obvious win for common workloads.
With the full series, process switching speed on the context_switch
benchmark on POWER9/hash (with kernel speculation security masures
disabled) increases from 140K/s to 178K/s (27%).
POWER8 does not change much (within 1%), it's unclear why it does not
see a big gain like POWER9.
Booting to busybox init with 256MB segments has SLB misses go down
from 945 to 69, and with 1T segments 900 to 21. These could almost all
be eliminated by preloading a bit more carefully with ELF binary
loading.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
This will be used by the SLB code in the next patch, but for now this
sets the slb_addr_limit to the correct size for 32-bit tasks.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Pull IDA updates from Matthew Wilcox:
"A better IDA API:
id = ida_alloc(ida, GFP_xxx);
ida_free(ida, id);
rather than the cumbersome ida_simple_get(), ida_simple_remove().
The new IDA API is similar to ida_simple_get() but better named. The
internal restructuring of the IDA code removes the bitmap
preallocation nonsense.
I hope the net -200 lines of code is convincing"
* 'ida-4.19' of git://git.infradead.org/users/willy/linux-dax: (29 commits)
ida: Change ida_get_new_above to return the id
ida: Remove old API
test_ida: check_ida_destroy and check_ida_alloc
test_ida: Convert check_ida_conv to new API
test_ida: Move ida_check_max
test_ida: Move ida_check_leaf
idr-test: Convert ida_check_nomem to new API
ida: Start new test_ida module
target/iscsi: Allocate session IDs from an IDA
iscsi target: fix session creation failure handling
drm/vmwgfx: Convert to new IDA API
dmaengine: Convert to new IDA API
ppc: Convert vas ID allocation to new IDA API
media: Convert entity ID allocation to new IDA API
ppc: Convert mmu context allocation to new IDA API
Convert net_namespace to new IDA API
cb710: Convert to new IDA API
rsxx: Convert to new IDA API
osd: Convert to new IDA API
sd: Convert to new IDA API
...
|
|
ida_alloc_range is the perfect fit for this use case. Eliminates
a custom spinlock, a call to ida_pre_get and a local check for the
allocated ID exceeding a maximum.
Signed-off-by: Matthew Wilcox <willy@infradead.org>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
|
|
The page table fragment allocator uses the main page refcount racily
with respect to speculative references. A customer observed a BUG due
to page table page refcount underflow in the fragment allocator. This
can be caused by the fragment allocator set_page_count stomping on a
speculative reference, and then the speculative failure handler
decrements the new reference, and the underflow eventually pops when
the page tables are freed.
Fix this by using a dedicated field in the struct page for the page
table fragment allocator.
Fixes: 5c1f6ee9a31c ("powerpc: Reduce PTE table memory wastage")
Cc: stable@vger.kernel.org # v3.10+
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
The kernel page table caches are tied to init_mm, so there is no
more need for them after userspace is finished.
destroy_context() gets called when we drop the last reference for an
mm, which can be much later than the task exit due to other lazy mm
references to it. We can free the page table cache pages on task exit
because they only cache the userspace page tables and kernel threads
should not access user space addresses.
The mapping for kernel threads itself is maintained in init_mm and
page table cache for that is attached to init_mm.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
[mpe: Merge change log additions from Aneesh]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
POWER9 DD1 was never a product. It is no longer supported by upstream
firmware, and it is not effectively supported in Linux due to lack of
testing.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Michael Ellerman <mpe@ellerman.id.au>
[mpe: Remove arch_make_huge_pte() entirely]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
4K config use one full page at level 4 of the pagetable. Add support for single
fragment allocation in pagetable fragment code and and use that for 4K config.
This makes both 4k and 64k use the same code path. Later we will switch pmd to
use the page table fragment code. This is done only for 64bit platforms which
is using page table fragment support.
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
For addresses above 512TB we allocate additional mmu contexts. To make
it all easy, addresses above 512TB are handled with IR/DR=1 and with
stack frame setup.
The mmu_context_t is also updated to track the new extended_ids. To
support upto 4PB we need a total 8 contexts.
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
[mpe: Minor formatting tweaks and comment wording, switch BUG to WARN
in get_ea_context().]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Merge our fixes branch from the 4.16 cycle.
There were a number of important fixes merged, in particular some Power9
workarounds that we want in next for testing purposes. There's also been
some conflicting changes in the CPU features code which are best merged
and tested before going upstream.
|
|
Currently, when using coprocessors (which use the Nest MMU), we
simply increment the active_cpu count to force all TLB invalidations
to be come broadcast.
Unfortunately, due to an errata in POWER9, we will need to know
more specifically that coprocessors are in use.
This maintains a separate copros counter in the MMU context for
that purpose.
NB. The commit mentioned in the fixes tag below is not at fault for
the bug we're fixing in this commit and the next, but this fix applies
on top the infrastructure it introduced.
Fixes: 03b8abedf4f4 ("cxl: Enable global TLBIs for cxl contexts")
Cc: stable@vger.kernel.org # v4.15+
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Tested-by: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
The slice state of an mm gets zeroed then initialised upon exec.
This is the only caller of slice_set_user_psize now, so that can be
removed and instead implement a faster and simplified approach that
requires no locking or checking existing state.
This speeds up vfork+exec+exit performance on POWER8 by 3%.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Total 32 keys are available on power7 and above. However
pkey 0,1 are reserved. So effectively we have 30 pkeys.
On 4K kernels, we do not have 5 bits in the PTE to
represent all the keys; we only have 3bits. Two of those
keys are reserved; pkey 0 and pkey 1. So effectively we
have 6 pkeys.
This patch keeps track of reserved keys, allocated keys
and keys that are currently free.
Also it adds skeletal functions and macros, that the
architecture-independent code expects to be available.
Reviewed-by: Thiago Jung Bauermann <bauerman@linux.vnet.ibm.com>
Signed-off-by: Ram Pai <linuxram@us.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Pull powerpc updates from Michael Ellerman:
"A bit of a small release, I suspect in part due to me travelling for
KS. But my backlog of patches to review is smaller than usual, so I
think in part folks just didn't send as much this cycle.
Non-highlights:
- Five fixes for the >128T address space handling, both to fix bugs
in our implementation and to bring the semantics exactly into line
with x86.
Highlights:
- Support for a new OPAL call on bare metal machines which gives us a
true NMI (ie. is not masked by MSR[EE]=0) for debugging etc.
- Support for Power9 DD2 in the CXL driver.
- Improvements to machine check handling so that uncorrectable errors
can be reported into the generic memory_failure() machinery.
- Some fixes and improvements for VPHN, which is used under PowerVM
to notify the Linux partition of topology changes.
- Plumbing to enable TM (transactional memory) without suspend on
some Power9 processors (PPC_FEATURE2_HTM_NO_SUSPEND).
- Support for emulating vector loads form cache-inhibited memory, on
some Power9 revisions.
- Disable the fast-endian switch "syscall" by default (behind a
CONFIG), we believe it has never had any users.
- A major rework of the API drivers use when initiating and waiting
for long running operations performed by OPAL firmware, and changes
to the powernv_flash driver to use the new API.
- Several fixes for the handling of FP/VMX/VSX while processes are
using transactional memory.
- Optimisations of TLB range flushes when using the radix MMU on
Power9.
- Improvements to the VAS facility used to access coprocessors on
Power9, and related improvements to the way the NX crypto driver
handles requests.
- Implementation of PMEM_API and UACCESS_FLUSHCACHE for 64-bit.
Thanks to: Alexey Kardashevskiy, Alistair Popple, Allen Pais, Andrew
Donnellan, Aneesh Kumar K.V, Arnd Bergmann, Balbir Singh, Benjamin
Herrenschmidt, Breno Leitao, Christophe Leroy, Christophe Lombard,
Cyril Bur, Frederic Barrat, Gautham R. Shenoy, Geert Uytterhoeven,
Guilherme G. Piccoli, Gustavo Romero, Haren Myneni, Joel Stanley,
Kamalesh Babulal, Kautuk Consul, Markus Elfring, Masami Hiramatsu,
Michael Bringmann, Michael Neuling, Michal Suchanek, Naveen N. Rao,
Nicholas Piggin, Oliver O'Halloran, Paul Mackerras, Pedro Miraglia
Franco de Carvalho, Philippe Bergheaud, Sandipan Das, Seth Forshee,
Shriya, Stephen Rothwell, Stewart Smith, Sukadev Bhattiprolu, Tyrel
Datwyler, Vaibhav Jain, Vaidyanathan Srinivasan, and William A.
Kennington III"
* tag 'powerpc-4.15-1' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux: (151 commits)
powerpc/64s: Fix Power9 DD2.0 workarounds by adding DD2.1 feature
powerpc/64s: Fix masking of SRR1 bits on instruction fault
powerpc/64s: mm_context.addr_limit is only used on hash
powerpc/64s/radix: Fix 128TB-512TB virtual address boundary case allocation
powerpc/64s/hash: Allow MAP_FIXED allocations to cross 128TB boundary
powerpc/64s/hash: Fix fork() with 512TB process address space
powerpc/64s/hash: Fix 128TB-512TB virtual address boundary case allocation
powerpc/64s/hash: Fix 512T hint detection to use >= 128T
powerpc: Fix DABR match on hash based systems
powerpc/signal: Properly handle return value from uprobe_deny_signal()
powerpc/fadump: use kstrtoint to handle sysfs store
powerpc/lib: Implement UACCESS_FLUSHCACHE API
powerpc/lib: Implement PMEM API
powerpc/powernv/npu: Don't explicitly flush nmmu tlb
powerpc/powernv/npu: Use flush_all_mm() instead of flush_tlb_mm()
powerpc/powernv/idle: Round up latency and residency values
powerpc/kprobes: refactor kprobe_lookup_name for safer string operations
powerpc/kprobes: Blacklist emulate_update_regs() from kprobes
powerpc/kprobes: Do not disable interrupts for optprobes and kprobes_on_ftrace
powerpc/kprobes: Disable preemption before invoking probe handler for optprobes
...
|
|
Most callers users of free_hot_cold_page claim the pages being released
are cache hot. The exception is the page reclaim paths where it is
likely that enough pages will be freed in the near future that the
per-cpu lists are going to be recycled and the cache hotness information
is lost. As no one really cares about the hotness of pages being
released to the allocator, just ditch the parameter.
The APIs are renamed to indicate that it's no longer about hot/cold
pages. It should also be less confusing as there are subtle differences
between them. __free_pages drops a reference and frees a page when the
refcount reaches zero. free_hot_cold_page handled pages whose refcount
was already zero which is non-obvious from the name. free_unref_page
should be more obvious.
No performance impact is expected as the overhead is marginal. The
parameter is removed simply because it is a bit stupid to have a useless
parameter copied everywhere.
[mgorman@techsingularity.net: add pages to head, not tail]
Link: http://lkml.kernel.org/r/20171019154321.qtpzaeftoyyw4iey@techsingularity.net
Link: http://lkml.kernel.org/r/20171018075952.10627-8-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
Radix keeps no meaningful state in addr_limit, so remove it from radix
code and rename to slb_addr_limit to make it clear it applies to hash
only.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Hash unconditionally resets the addr_limit to default (128TB) when the
mm context is initialised. If a process has > 128TB mappings when it
forks, the child will not get the 512TB addr_limit, so accesses to
valid > 128TB mappings will fail in the child.
Fix this by only resetting the addr_limit to default if it was 0. Non
zero indicates it was duplicated from the parent (0 means exec()).
Fixes: f4ea6dcb08ea ("powerpc/mm: Enable mappings above 128TB")
Cc: stable@vger.kernel.org # v4.12+
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
According to the architecture, the process table entry cache must be
flushed with tlbie RIC=2.
Currently the process table entry is set to invalid right before the
PID is returned to the allocator, with no invalidation. This works on
existing implementations that are known to not cache the process table
entry for any except the current PIDR.
It is architecturally correct and cleaner to invalidate with RIC=2
after clearing the process table entry and before the PID is returned
to the allocator. This can be done in arch_exit_mmap that runs before
the final flush, and to ensure the final flush (fullmm) is always a
RIC=2 variant.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Instead of comparing the whole CPU mask every time, let's
keep a counter of how many bits are set in the mask. Thus
testing for a local mm only requires testing if that counter
is 1 and the current CPU bit is set in the mask.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
There's a non-trivial dependency between some commits we want to put in
next and the KVM prefetch work around that went into fixes. So merge
fixes into next.
|
|
We have a whole pile of unused code to maintain the ACOP register,
allocate coprocessor PIDs and handle ACOP faults. This mechanism
was used for the HFI adapter on POWER7 which is dead and gone and
whose driver never went upstream. It was used on some A2 core based
stuff that also never saw the light of day.
Take out all that code.
There is still some POWER8 coprocessor code that uses icswx but it's
kernel only and thus doesn't use any of that infrastructure.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
There's a somewhat architectural issue with Radix MMU and KVM.
When coming out of a guest with AIL (Alternate Interrupt Location, ie,
MMU enabled), we start executing hypervisor code with the PID register
still containing whatever the guest has been using.
The problem is that the CPU can (and will) then start prefetching or
speculatively load from whatever host context has that same PID (if
any), thus bringing translations for that context into the TLB, which
Linux doesn't know about.
This can cause stale translations and subsequent crashes.
Fixing this in a way that is neither racy nor a huge performance
impact is difficult. We could just make the host invalidations always
use broadcast forms but that would hurt single threaded programs for
example.
We chose to fix it instead by partitioning the PID space between guest
and host. This is possible because today Linux only use 19 out of the
20 bits of PID space, so existing guests will work if we make the host
use the top half of the 20 bits space.
We additionally add support for a property to indicate to Linux the
size of the PID register which will be useful if we eventually have
processors with a larger PID space available.
There is still an issue with malicious guests purposefully setting the
PID register to a value in the hosts PID range. Hopefully future HW
can prevent that, but in the meantime, we handle it with a pair of
kludges:
- On the way out of a guest, before we clear the current VCPU in the
PACA, we check the PID and if it's outside of the permitted range
we flush the TLB for that PID.
- When context switching, if the mm is "new" on that CPU (the
corresponding bit was set for the first time in the mm cpumask), we
check if any sibling thread is in KVM (has a non-NULL VCPU pointer
in the PACA). If that is the case, we also flush the PID for that
CPU (core).
This second part is needed to handle the case where a process is
migrated (or starts a new pthread) on a sibling thread of the CPU
coming out of KVM, as there's a window where stale translations can
exist before we detect it and flush them out.
A future optimization could be added by keeping track of whether the
PID has ever been used and avoid doing that for completely fresh PIDs.
We could similarily mark PIDs that have been the subject of a global
invalidation as "fresh". But for now this will do.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
[mpe: Rework the asm to build with CONFIG_PPC_RADIX_MMU=n, drop
unneeded include of kvm_book3s_asm.h]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
When writing to the process table, we need to ensure the store is
visible to a subsequent access by the MMU. We assume we never have
the PID active while doing the update, so a ptesync/isync pair
should hopefully be a big enough hammer for our purpose.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
On radix, the process table entry we want to clear when destroying a
context is entry 0, not entry 1. This has no *immediate* consequence
on Power9, but it can cause other bugs to become worse.
Fixes: 7e381c0ff618 ("powerpc/mm/radix: Add mmu context handling callback for radix")
Cc: stable@vger.kernel.org # v4.7+
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Merge our fixes branch, a few of them are tripping people up while
working on top of next, and we also have a dependency between the CXL
fixes and new CXL code we want to merge into next.
|
|
From: Michael Neuling <mikey@neuling.org>
On P9 (Nimbus) DD2 and later, in radix mode, the move to the PID
register will implicitly invalidate the user space ERAT entries
and leave the kernel ones alone. Thus the only thing needed is
an isync() to synchronize this with subsequent uaccess's
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Supporting 512TB requires us to do a order 3 allocation for level 1 page
table (pgd). This results in page allocation failures with certain workloads.
For now limit 4k linux page size config to 64TB.
Fixes: f6eedbba7a26 ("powerpc/mm/hash: Increase VA range to 128TB")
Reported-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Nvlink2 supports address translation services (ATS) allowing devices
to request address translations from an mmu known as the nest MMU
which is setup to walk the CPU page tables.
To access this functionality certain firmware calls are required to
setup and manage hardware context tables in the nvlink processing unit
(NPU). The NPU also manages forwarding of TLB invalidates (known as
address translation shootdowns/ATSDs) to attached devices.
This patch exports several methods to allow device drivers to register
a process id (PASID/PID) in the hardware tables and to receive
notification of when a device should stop issuing address translation
requests (ATRs). It also adds a fault handler to allow device drivers
to demand fault pages in.
Signed-off-by: Alistair Popple <alistair@popple.id.au>
[mpe: Fix up comment formatting, use flush_tlb_mm()]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Not all user space application is ready to handle wide addresses. It's
known that at least some JIT compilers use higher bits in pointers to
encode their information. It collides with valid pointers with 512TB
addresses and leads to crashes.
To mitigate this, we are not going to allocate virtual address space
above 128TB by default.
But userspace can ask for allocation from full address space by
specifying hint address (with or without MAP_FIXED) above 128TB.
If hint address set above 128TB, but MAP_FIXED is not specified, we try
to look for unmapped area by specified address. If it's already
occupied, we look for unmapped area in *full* address space, rather than
from 128TB window.
This approach helps to easily make application's memory allocator aware
about large address space without manually tracking allocated virtual
address space.
This is going to be a per mmap decision. ie, we can have some mmaps with
larger addresses and other that do not.
A sample memory layout looks like:
10000000-10010000 r-xp 00000000 fc:00 9057045 /home/max_addr_512TB
10010000-10020000 r--p 00000000 fc:00 9057045 /home/max_addr_512TB
10020000-10030000 rw-p 00010000 fc:00 9057045 /home/max_addr_512TB
10029630000-10029660000 rw-p 00000000 00:00 0 [heap]
7fff834a0000-7fff834b0000 rw-p 00000000 00:00 0
7fff834b0000-7fff83670000 r-xp 00000000 fc:00 9177190 /lib/powerpc64le-linux-gnu/libc-2.23.so
7fff83670000-7fff83680000 r--p 001b0000 fc:00 9177190 /lib/powerpc64le-linux-gnu/libc-2.23.so
7fff83680000-7fff83690000 rw-p 001c0000 fc:00 9177190 /lib/powerpc64le-linux-gnu/libc-2.23.so
7fff83690000-7fff836a0000 rw-p 00000000 00:00 0
7fff836a0000-7fff836c0000 r-xp 00000000 00:00 0 [vdso]
7fff836c0000-7fff83700000 r-xp 00000000 fc:00 9177193 /lib/powerpc64le-linux-gnu/ld-2.23.so
7fff83700000-7fff83710000 r--p 00030000 fc:00 9177193 /lib/powerpc64le-linux-gnu/ld-2.23.so
7fff83710000-7fff83720000 rw-p 00040000 fc:00 9177193 /lib/powerpc64le-linux-gnu/ld-2.23.so
7fffdccf0000-7fffdcd20000 rw-p 00000000 00:00 0 [stack]
1000000000000-1000000010000 rw-p 00000000 00:00 0
1ffff83710000-1ffff83720000 rw-p 00000000 00:00 0
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Now that we use all the available virtual address range, we need to make
sure we don't generate VSID such that it overlaps with the reserved vsid
range. Reserved vsid range include the virtual address range used by the
adjunct partition and also the VRMA virtual segment. We find the context
value that can result in generating such a VSID and reserve it early in
boot.
We don't look at the adjunct range, because for now we disable the
adjunct usage in a Linux LPAR via CAS interface.
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
[mpe: Rewrite hash__reserve_context_id(), move the rest into pseries]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
In the followup patch, we will increase the slice array size to handle
512TB range, but will limit the max addr to 128TB. Avoid doing
unnecessary computation and avoid doing slice mask related operation
above address limit.
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Inorder to support large effective address range (512TB), we want to
increase the virtual address bits to 68. But we do have platforms like
p4 and p5 that can only do 65 bit VA. We support those platforms by
limiting context bits on them to 16.
The protovsid -> vsid conversion is verified to work with both 65 and 68
bit va values. I also documented the restrictions in a table format as
part of code comments.
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Currently we use the top 4 context ids (0x7fffc-0x7ffff) for the kernel.
Kernel VSIDs are built using these top context values and effective the
segement ID. In subsequent patches we want to increase the max effective
address to 512TB. We will achieve that by increasing the effective
segment IDs there by increasing virtual address range.
We will be switching to a 68bit virtual address in the following patch.
But platforms like Power4 and Power5 only support a 65 bit virtual
address. We will handle that by limiting the context bits to 16 instead
of 19 on those platforms. That means the max context id will have a
different value on different platforms.
So that we don't have to deal with the kernel context ids changing
between different platforms, move the kernel context ids down to use
context ids 1-4.
We can't use segment 0 of context-id 0, because that maps to VSID 0,
which we want to keep as invalid, so we avoid context-id 0 entirely.
Similarly we can't use the last segment of the maximum context, so we
avoid it too.
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
[mpe: Switch from 0-3 to 1-4 so VSID=0 remains invalid]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Complete the split of the radix vs hash mm context initialisation.
This is mostly code movement, with the exception that we now limit the
context allocation to PRTB_ENTRIES - 1 on radix.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
The min and max context id values used in alloc_context_id() are
currently the right values for use on hash, and happen to also be safe
for use on radix.
But we need to change that in a subsequent patch, so make the min/max
ids parameters and pull the hash values into hsah__alloc_context_id().
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
KVM wants to be able to allocate an MMU context id, which it does
currently by calling __init_new_context().
We're about to rework that code, so provide a wrapper for KVM so it
can not worry about the details.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
At the moment the userspace tool is expected to request pinning of
the entire guest RAM when VFIO IOMMU SPAPR v2 driver is present.
When the userspace process finishes, all the pinned pages need to
be put; this is done as a part of the userspace memory context (MM)
destruction which happens on the very last mmdrop().
This approach has a problem that a MM of the userspace process
may live longer than the userspace process itself as kernel threads
use userspace process MMs which was runnning on a CPU where
the kernel thread was scheduled to. If this happened, the MM remains
referenced until this exact kernel thread wakes up again
and releases the very last reference to the MM, on an idle system this
can take even hours.
This moves preregistered regions tracking from MM to VFIO; insteads of
using mm_iommu_table_group_mem_t::used, tce_container::prereg_list is
added so each container releases regions which it has pre-registered.
This changes the userspace interface to return EBUSY if a memory
region is already registered in a container. However it should not
have any practical effect as the only userspace tool available now
does register memory region once per container anyway.
As tce_iommu_register_pages/tce_iommu_unregister_pages are called
under container->lock, this does not need additional locking.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Acked-by: Alex Williamson <alex.williamson@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
We are going to get rid of @current references in mmu_context_boos3s64.c
and cache mm_struct in the VFIO container. Since mm_context_t does not
have reference counting, we will be using mm_struct which does have
the reference counter.
This changes mm_iommu_init/mm_iommu_cleanup to receive mm_struct rather
than mm_context_t (which is embedded into mm).
This should not cause any behavioral change.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Update the PID switch as per ISA doc. slbia is needed in radix to
invalidate any implementation specific lookaside information.
We use the .long format due to build errors with the below compiler
version.
gcc (Ubuntu 5.3.1-14ubuntu2.1) 5.3.1 20160413
GNU assembler (GNU Binutils for Ubuntu) 2.26
CC arch/powerpc/mm//mmu_context_book3s64.o
{standard input}: Assembler messages:
{standard input}:506: Error: junk at end of line: `0x7'
scripts/Makefile.build:291: recipe for target 'arch/powerpc/mm//mmu_context_book3s64.o' failed
make[1]: *** [arch/powerpc/mm//mmu_context_book3s64.o] Error 1
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
ISA 3.0 updated it to be encoded as Radix tree size = 2^(RTS + 31). We
have it encoded as 2^(RTS + 28). Add a helper with the correct encoding
and use it instead of opencoding.
Fixes: 2bfd65e45e87 ("powerpc/mm/radix: Add radix callbacks for early init routines")
Reviewed-by: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
As part of the radix support we switched Book3s64 to use a value of ~0
for MMU_NO_CONTEXT. That is because id 0 is special on radix.
However that broke the logic in init_new_context(). The code there needs
to differentiate between a newly allocated context and one inherited via
fork. Previously it worked because a newly allocated context has an id
of zero (because it was just memset() to zero), which used to match
MMU_NO_CONTEXT, and therefore slice_mm_new_context() did the right
thing.
Instead check against a context.id value of zero instead of using
slice_mm_new_context().
Without this patch we never call slice_set_user_psize(), and end up with
a slice psize value of zero and we always end up using 4K HPTE.
Fixes: 1a472c9dba6b ("powerpc/mm/radix: Add tlbflush routines")
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
This file now contains both hash and radix specific code. Rename it to
indicate this better.
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
Aneesh Kumar K.V
Christophe Leroy
Nicholas Piggin
Michael Ellerman
Linus Torvalds
Matthew Wilcox
Aneesh Kumar K.V
Benjamin Herrenschmidt
Ram Pai
Mel Gorman
Alistair Popple
Alexey Kardashevskiy