Since 902bdc5745, get_pci_dev() calls pci_get_domain_bus_and_slot(). This
has the effect of incrementing the reference count of the PCI device, as
explained in drivers/pci/search.c:
* Given a PCI domain, bus, and slot/function number, the desired PCI
* device is located in the list of PCI devices. If the device is
* found, its reference count is increased and this function returns a
* pointer to its data structure. The caller must decrement the
* reference count by calling pci_dev_put(). If no device is found,
* %NULL is returned.
Nothing was done to call pci_dev_put() and the reference count of GPU and
NPU PCI devices rockets up.
A natural way to fix this would be to teach the callers about the change,
so that they call pci_dev_put() when done with the pointer. This turns
out to be quite intrusive, as it affects many paths in npu-dma.c,
pci-ioda.c and vfio_pci_nvlink2.c. Also, the issue appeared in 4.16 and
some affected code got moved around since then: it would be problematic
to backport the fix to stable releases.
All that code never cared for reference counting anyway. Call pci_dev_put()
from get_pci_dev() to revert to the previous behavior.
Fixes: 902bdc5745 ("powerpc/powernv/idoa: Remove unnecessary pcidev from pci_dn")
Cc: stable@vger.kernel.org # v4.16
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Use new helper pci_dev_id() to simplify the code.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
The change_pte() notifier was designed to use as a quick path to
update secondary MMU PTEs on write permission changes or PFN changes.
For KVM, it could reduce the vm-exits when vcpu faults on the pages
that was touched up by KSM. It's not used to do cache invalidations,
for example, if we see the notifier will be called before the real PTE
update after all (please see set_pte_at_notify that set_pte_at was
called later).
All the necessary cache invalidation should all be done in
invalidate_range() already.
Signed-off-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Alistair Popple <alistair@popple.id.au>
Reviewed-by: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Now that we've switched all the powerpc nommu and swiotlb methods to
use the generic dma_direct_* calls we can remove these ops vectors
entirely and rely on the common direct mapping bypass that avoids
indirect function calls entirely. This also allows to remove a whole
lot of boilerplate code related to setting up these operations.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Tested-by: Christian Zigotzky <chzigotzky@xenosoft.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
TCE_KILL_INVAL_ALL has moved long ago but the comment was forgotted so
finish the move and remove the comment.
Fixes: 0bbcdb437d "powerpc/powernv/npu: TCE Kill helpers cleanup"
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
There is a typo so we accidentally allocate enough memory for a pointer
when we wanted to allocate enough for a struct.
Fixes: 0bd971676e ("powerpc/powernv/npu: Add compound IOMMU groups")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
When a page fault happens in a GPU, the GPU signals the OS and the GPU
driver calls the fault handler which populated a page table; this allows
the GPU to complete an ATS request.
On the bare metal get_user_pages() is enough as it adds a pte to
the kernel page table but under KVM the partition scope tree does not get
updated so ATS will still fail.
This reads a byte from an effective address which causes HV storage
interrupt and KVM updates the partition scope tree.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
A broken device tree might contain more than 8 values and introduce hard
to debug memory corruption bug. This adds the boundary check.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
In order to make ATS work and translate addresses for arbitrary
LPID and PID, we need to program an NPU with LPID and allow PID wildcard
matching with a specific MSR mask.
This implements a helper to assign a GPU to LPAR and program the NPU
with a wildcard for PID and a helper to do clean-up. The helper takes
MSR (only DR/HV/PR/SF bits are allowed) to program them into NPU2 for
ATS checkout requests support.
This exports pnv_npu2_unmap_lpar_dev() as following patches will use it
from the VFIO driver.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
At the moment the powernv platform registers an IOMMU group for each
PE. There is an exception though: an NVLink bridge which is attached
to the corresponding GPU's IOMMU group making it a master.
Now we have POWER9 systems with GPUs connected to each other directly
bypassing PCI. At the moment we do not control state of these links so
we have to put such interconnected GPUs to one IOMMU group which means
that the old scheme with one GPU as a master won't work - there will
be up to 3 GPUs in such group.
This introduces a npu_comp struct which represents a compound IOMMU
group made of multiple PEs - PCI PEs (for GPUs) and NPU PEs (for
NVLink bridges). This converts the existing NVLink1 code to use the
new scheme. >From now on, each PE must have a valid
iommu_table_group_ops which will either be called directly (for a
single PE group) or indirectly from a compound group handlers.
This moves IOMMU group registration for NVLink-connected GPUs to
npu-dma.c. For POWER8, this stores a new compound group pointer in the
PE (so a GPU is still a master); for POWER9 the new group pointer is
stored in an NPU (which is allocated per a PCI host controller).
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
[mpe: Initialise npdev to NULL in pnv_try_setup_npu_table_group()]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
At the moment NPU IOMMU is manipulated directly from the IODA2 PCI
PE code; PCI PE acts as a master to NPU PE. Soon we will have compound
IOMMU groups with several PEs from several different PHB (such as
interconnected GPUs and NPUs) so there will be no single master but
a one big IOMMU group.
This makes a first step and converts an NPU PE with a set of extern
function to a table group.
This should cause no behavioral change. Note that
pnv_npu_release_ownership() has never been implemented.
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>
Normal PCI PEs have 2 TVEs, one per a DMA window; however NPU PE has only
one which points to one of two tables of the corresponding PCI PE.
So whenever a new DMA window is programmed to PEs, the NPU PE needs to
release old table in order to use the new one.
Commit d41ce7b1bc ("powerpc/powernv/npu: Do not try invalidating 32bit
table when 64bit table is enabled") did just that but in pci-ioda.c
while it actually belongs to npu-dma.c.
This moves the single TVE handling to npu-dma.c. This does not implement
restoring though as it is highly unlikely that we can set the table to
PCI PE and cannot to NPU PE and if that fails, we could only set 32bit
table to NPU PE and this configuration is not really supported or wanted.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
When introduced, the NPU context init/destroy helpers called OPAL which
enabled/disabled PID (a userspace memory context ID) filtering in an NPU
per a GPU; this was a requirement for P9 DD1.0. However newer chip
revision added a PID wildcard support so there is no more need to
call OPAL every time a new context is initialized. Also, since the PID
wildcard support was added, skiboot does not clear wildcard entries
in the NPU so these remain in the hardware till the system reboot.
This moves LPID and wildcard programming to the PE setup code which
executes once during the booting process so NPU2 context init/destroy
won't need to do additional configuration.
This replaces the check for FW_FEATURE_OPAL with a check for npu!=NULL as
this is the way to tell if the NPU support is present and configured.
This moves pnv_npu2_init() declaration as pseries should be able to use it.
This keeps pnv_npu2_map_lpar() in powernv as pseries is not allowed to
call that. This exports pnv_npu2_map_lpar_dev() as following patches
will use it from the VFIO driver.
While at it, replace redundant list_for_each_entry_safe() with
a simpler list_for_each_entry().
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The powernv PCI code stores NPU data in the pnv_phb struct. The latter
is referenced by pci_controller::private_data. We are going to have NPU2
support in the pseries platform as well but it does not store any
private_data in in the pci_controller struct; and even if it did,
it would be a different data structure.
This makes npu a pointer and stores it one level higher in
the pci_controller struct.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The macro and few headers are not used so remove them.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Acked-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The NPU IOMMU is setup to mirror the parent PCIe device IOMMU
setup. Therefore it does not make sense to call dma operations such as
dma_map_page(), etc. directly on these devices. The existing dma_ops
simply print a warning if they are ever called, however this is
unnecessary and the warnings are likely to go unnoticed.
It is instead simpler to remove these operations and let the generic
DMA code print warnings (eg. via a NULL pointer deref) in cases of
buggy drivers attempting dma operations on NVLink devices.
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
This threshold is no longer used now that all invalidates issue a single
ATSD to each active NPU.
Signed-off-by: Mark Hairgrove <mhairgrove@nvidia.com>
Reviewed-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Prior to this change only two types of ATSDs were issued to the NPU:
invalidates targeting a single page and invalidates targeting the whole
address space. The crossover point happened at the configurable
atsd_threshold which defaulted to 2M. Invalidates that size or smaller
would issue per-page invalidates for the whole range.
The NPU supports more invalidation sizes however: 64K, 2M, 1G, and all.
These invalidates target addresses aligned to their size. 2M is a common
invalidation size for GPU-enabled applications because that is a GPU
page size, so reducing the number of invalidates by 32x in that case is a
clear improvement.
ATSD latency is high in general so now we always issue a single invalidate
rather than multiple. This will over-invalidate in some cases, but for any
invalidation size over 2M it matches or improves the prior behavior.
There's also an improvement for single-page invalidates since the prior
version issued two invalidates for that case instead of one.
With this change all issued ATSDs now perform a flush, so the flush
parameter has been removed from all the helpers.
To show the benefit here are some performance numbers from a
microbenchmark which creates a 1G allocation then uses mprotect with
PROT_NONE to trigger invalidates in strides across the allocation.
One NPU (1 GPU):
mprotect rate (GB/s)
Stride Before After Speedup
64K 5.3 5.6 5%
1M 39.3 57.4 46%
2M 49.7 82.6 66%
4M 286.6 285.7 0%
Two NPUs (6 GPUs):
mprotect rate (GB/s)
Stride Before After Speedup
64K 6.5 7.4 13%
1M 33.4 67.9 103%
2M 38.7 93.1 141%
4M 356.7 354.6 -1%
Anything over 2M is roughly the same as before since both cases issue a
single ATSD.
Signed-off-by: Mark Hairgrove <mhairgrove@nvidia.com>
Reviewed-By: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
There are two types of ATSDs issued to the NPU: invalidates targeting a
specific virtual address and invalidates targeting the whole address
space. In both cases prior to this change, the sequence was:
for each NPU
- Write the target address to the XTS_ATSD_AVA register
- EIEIO
- Write the launch value to issue the ATSD
First, a target address is not required when invalidating the whole
address space, so that write and the EIEIO have been removed. The AP
(size) field in the launch is not needed either.
Second, for per-address invalidates the above sequence is inefficient in
the common case of multiple NPUs because an EIEIO is issued per NPU. This
unnecessarily forces the launches of later ATSDs to be ordered with the
launches of earlier ones. The new sequence only issues a single EIEIO:
for each NPU
- Write the target address to the XTS_ATSD_AVA register
EIEIO
for each NPU
- Write the launch value to issue the ATSD
Performance results were gathered using a microbenchmark which creates a
1G allocation then uses mprotect with PROT_NONE to trigger invalidates in
strides across the allocation.
With only a single NPU active (one GPU) the difference is in the noise for
both types of invalidates (+/-1%).
With two NPUs active (on a 6-GPU system) the effect is more noticeable:
mprotect rate (GB/s)
Stride Before After Speedup
64K 5.9 6.5 10%
1M 31.2 33.4 7%
2M 36.3 38.7 7%
4M 322.6 356.7 11%
Signed-off-by: Mark Hairgrove <mhairgrove@nvidia.com>
Reviewed-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
We've encountered a performance issue when multiple processors stress
{get,put}_mmio_atsd_reg(). These functions contend for
mmio_atsd_usage, an unsigned long used as a bitmask.
The accesses to mmio_atsd_usage are done using test_and_set_bit_lock()
and clear_bit_unlock(). As implemented, both of these will require
a (successful) stwcx to that same cache line.
What we end up with is thread A, attempting to unlock, being slowed by
other threads repeatedly attempting to lock. A's stwcx instructions
fail and retry because the memory reservation is lost every time a
different thread beats it to the punch.
There may be a long-term way to fix this at a larger scale, but for
now resolve the immediate problem by gating our call to
test_and_set_bit_lock() with one to test_bit(), which is obviously
implemented without using a store.
Fixes: 1ab66d1fba ("powerpc/powernv: Introduce address translation services for Nvlink2")
Signed-off-by: Reza Arbab <arbab@linux.ibm.com>
Acked-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The threshold at which it becomes more efficient to coalesce a range
of ATSDs into a single per-PID ATSD is currently not well understood
due to a lack of real-world work loads. This patch adds a debugfs
parameter allowing the threshold to be altered at runtime in order to
aid future development and refinement of the value.
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Acked-by: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
This allows us to squash some sparse warnings and also avoids having
to do explicity endian conversions in the code.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
The NPU has a limited number of address translation shootdown (ATSD)
registers and the GPU has limited bandwidth to process ATSDs. This can
result in contention of ATSD registers leading to soft lockups on some
threads, particularly when invalidating a large address range in
pnv_npu2_mn_invalidate_range().
At some threshold it becomes more efficient to flush the entire GPU
TLB for the given MM context (PID) than individually flushing each
address in the range. This patch will result in ranges greater than
2MB being converted from 32+ ATSDs into a single ATSD which will flush
the TLB for the given PID on each GPU.
Fixes: 1ab66d1fba ("powerpc/powernv: Introduce address translation services for Nvlink2")
Cc: stable@vger.kernel.org # v4.12+
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Acked-by: Balbir Singh <bsingharora@gmail.com>
Tested-by: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
There is a single npu context per set of callback parameters. Callers
should be prevented from overwriting existing callback values so
instead return an error if different parameters are passed.
Fixes: 1ab66d1fba ("powerpc/powernv: Introduce address translation services for Nvlink2")
Cc: stable@vger.kernel.org # v4.12+
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Reviewed-by: Mark Hairgrove <mhairgrove@nvidia.com>
Tested-by: Mark Hairgrove <mhairgrove@nvidia.com>
Reviewed-by: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The pnv_npu2_init_context() and pnv_npu2_destroy_context() functions
are used to allocate/free contexts to allow address translation and
shootdown by the NPU on a particular GPU. Context initialisation is
implicitly safe as it is protected by the requirement mmap_sem be held
in write mode, however pnv_npu2_destroy_context() does not require
mmap_sem to be held and it is not safe to call with a concurrent
initialisation for a different GPU.
It was assumed the driver would ensure destruction was not called
concurrently with initialisation. However the driver may be simplified
by allowing concurrent initialisation and destruction for different
GPUs. As npu context creation/destruction is not a performance
critical path and the critical section is not large a single spinlock
is used for simplicity.
Fixes: 1ab66d1fba ("powerpc/powernv: Introduce address translation services for Nvlink2")
Cc: stable@vger.kernel.org # v4.12+
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Reviewed-by: Mark Hairgrove <mhairgrove@nvidia.com>
Tested-by: Mark Hairgrove <mhairgrove@nvidia.com>
Reviewed-by: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
pnv_npu2_init_context wasn't checking the return code from
__mmu_notifier_register. If __mmu_notifier_register failed, the
npu_context was still assigned to the mm and the caller wasn't given any
indication that things went wrong. Later on pnv_npu2_destroy_context would
be called, which in turn called mmu_notifier_unregister and dropped
mm->mm_count without having incremented it in the first place. This led to
various forms of corruption like mm use-after-free and mm double-free.
__mmu_notifier_register can fail with EINTR if a signal is pending, so
this case can be frequent.
This patch calls opal_npu_destroy_context on the failure paths, and makes
sure not to assign mm->context.npu_context until past the failure points.
Signed-off-by: Mark Hairgrove <mhairgrove@nvidia.com>
Acked-By: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
When sending TLB invalidates to the NPU we need to send extra flushes due
to a hardware issue. The original implementation would lock the all the
ATSD MMIO registers sequentially before unlocking and relocking each of
them sequentially to do the extra flush.
This introduced a deadlock as it is possible for one thread to hold one
ATSD register whilst waiting for another register to be freed while the
other thread is holding that register waiting for the one in the first
thread to be freed.
For example if there are two threads and two ATSD registers:
Thread A Thread B
----------------------
Acquire 1
Acquire 2
Release 1 Acquire 1
Wait 1 Wait 2
Both threads will be stuck waiting to acquire a register resulting in an
RCU stall warning or soft lockup.
This patch solves the deadlock by refactoring the code to ensure registers
are not released between flushes and to ensure all registers are either
acquired or released together and in order.
Fixes: bbd5ff50af ("powerpc/powernv/npu-dma: Add explicit flush when sending an ATSD")
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The pcidev value stored in pci_dn is only used for NPU/NPU2
initialization. We can easily drop the cached pointer and
use an ancient helper - pci_get_domain_bus_and_slot() instead in order
to reduce complexity.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Acked-by: Russell Currey <ruscur@russell.cc>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The NPU was already abstracted by opal as a virtual PHB for nvlink,
but it helps to be able to differentiate between a nvlink or opencapi
PHB, as it's not completely transparent to linux. In particular, PE
assignment differs and we'll also need the information in later
patches.
So rename existing PNV_PHB_NPU type to PNV_PHB_NPU_NVLINK and add a
new type PNV_PHB_NPU_OCAPI.
Signed-off-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com>
Signed-off-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The nest mmu required an explicit flush as a tlbi would not flush it in the
same way as the core. However an alternate firmware fix exists which should
eliminate the need for this flush, so instead add a device-tree property
(ibm,nmmu-flush) on the NVLink2 PHB to enable it only if required.
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Reviewed-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
With the optimisations introduced by commit a46cc7a908 ("powerpc/mm/radix:
Improve TLB/PWC flushes"), flush_tlb_mm() no longer flushes the page walk
cache with radix. Switch to using flush_all_mm() to ensure the pwc and tlb
are properly flushed on the nmmu.
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Nothing really major this release, despite quite a lot of activity. Just lots of
things all over the place.
Some things of note include:
- Access via perf to a new type of PMU (IMC) on Power9, which can count both
core events as well as nest unit events (Memory controller etc).
- Optimisations to the radix MMU TLB flushing, mostly to avoid unnecessary Page
Walk Cache (PWC) flushes when the structure of the tree is not changing.
- Reworks/cleanups of do_page_fault() to modernise it and bring it closer to
other architectures where possible.
- Rework of our page table walking so that THP updates only need to send IPIs
to CPUs where the affected mm has run, rather than all CPUs.
- The size of our vmalloc area is increased to 56T on 64-bit hash MMU systems.
This avoids problems with the percpu allocator on systems with very sparse
NUMA layouts.
- STRICT_KERNEL_RWX support on PPC32.
- A new sched domain topology for Power9, to capture the fact that pairs of
cores may share an L2 cache.
- Power9 support for VAS, which is a new mechanism for accessing coprocessors,
and initial support for using it with the NX compression accelerator.
- Major work on the instruction emulation support, adding support for many new
instructions, and reworking it so it can be used to implement the emulation
needed to fixup alignment faults.
- Support for guests under PowerVM to use the Power9 XIVE interrupt controller.
And probably that many things again that are almost as interesting, but I had to
keep the list short. Plus the usual fixes and cleanups as always.
Thanks to:
Alexey Kardashevskiy, Alistair Popple, Andreas Schwab, Aneesh Kumar K.V, Anju
T Sudhakar, Arvind Yadav, Balbir Singh, Benjamin Herrenschmidt, Bhumika Goyal,
Breno Leitao, Bryant G. Ly, Christophe Leroy, Cédric Le Goater, Dan Carpenter,
Dou Liyang, Frederic Barrat, Gautham R. Shenoy, Geliang Tang, Geoff Levand,
Hannes Reinecke, Haren Myneni, Ivan Mikhaylov, John Allen, Julia Lawall, LABBE
Corentin, Laurentiu Tudor, Madhavan Srinivasan, Markus Elfring, Masahiro
Yamada, Matt Brown, Michael Neuling, Murilo Opsfelder Araujo, Nathan Fontenot,
Naveen N. Rao, Nicholas Piggin, Oliver O'Halloran, Paul Mackerras, Rashmica
Gupta, Rob Herring, Rui Teng, Sam Bobroff, Santosh Sivaraj, Scott Wood,
Shilpasri G Bhat, Sukadev Bhattiprolu, Suraj Jitindar Singh, Tobin C. Harding,
Victor Aoqui.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQIcBAABAgAGBQJZr83SAAoJEFHr6jzI4aWA6pUP/3CEaj2bSxNzWIwidqyYjuoS
O1moEsP0oYH7eBEWVHalYxvo0QPIIAhbFPaFyrOrgtfDH01Szwu9LcCALGb8orC5
Hg3IY8mpNG3Q1T8wEtTa56Ik4b5ZFty35S5+X9qLNSFoDUqSvGlSsLzhPNN7f2tl
XFm2hWqd8wXCwDsuVSFBCF61M3SAm+g6NMVNJ+VL2KIDCwBrOZLhKDPRoxLTAuMa
jjSdjVIozWyXjUrBFi8HVcoOWLxcT1HsNF0tRs51LwY/+Mlj2jAtFtsx+a06HZa6
f2p/Kcp/MEispSTk064Ap9cC1seXWI18zwZKpCUFqu0Ec2yTAiGdjOWDyYQldIp+
ttVPSHQ01YrVKwDFTtM9CiA0EET6fVPhWgAPkPfvH5TvtKwGkNdy0b+nQLuWrYip
BUmOXmjdIG3nujCzA9sv6/uNNhjhj2y+HWwuV7Qo002VFkhgZFL67u2SSUQLpYPj
PxdkY8pPVq+O+in94oDV3c36dYFF6+g6A6505Vn6eKUm/TLpszRFGkS3bKKA5vtn
74FR+guV/5RwYJcdZbfm04DgAocl7AfUDxpwRxibt6KtAK2VZKQuw4ugUTgYEd7W
mL2+AMmPKuajWXAMTHjCZPbUp9gFNyYyBQTFfGVX/XLiM8erKBnGfoa1/KzUJkhr
fVZLYIO/gzl34PiTIfgD
=UJtt
-----END PGP SIGNATURE-----
Merge tag 'powerpc-4.14-1' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux
Pull powerpc updates from Michael Ellerman:
"Nothing really major this release, despite quite a lot of activity.
Just lots of things all over the place.
Some things of note include:
- Access via perf to a new type of PMU (IMC) on Power9, which can
count both core events as well as nest unit events (Memory
controller etc).
- Optimisations to the radix MMU TLB flushing, mostly to avoid
unnecessary Page Walk Cache (PWC) flushes when the structure of the
tree is not changing.
- Reworks/cleanups of do_page_fault() to modernise it and bring it
closer to other architectures where possible.
- Rework of our page table walking so that THP updates only need to
send IPIs to CPUs where the affected mm has run, rather than all
CPUs.
- The size of our vmalloc area is increased to 56T on 64-bit hash MMU
systems. This avoids problems with the percpu allocator on systems
with very sparse NUMA layouts.
- STRICT_KERNEL_RWX support on PPC32.
- A new sched domain topology for Power9, to capture the fact that
pairs of cores may share an L2 cache.
- Power9 support for VAS, which is a new mechanism for accessing
coprocessors, and initial support for using it with the NX
compression accelerator.
- Major work on the instruction emulation support, adding support for
many new instructions, and reworking it so it can be used to
implement the emulation needed to fixup alignment faults.
- Support for guests under PowerVM to use the Power9 XIVE interrupt
controller.
And probably that many things again that are almost as interesting,
but I had to keep the list short. Plus the usual fixes and cleanups as
always.
Thanks to: Alexey Kardashevskiy, Alistair Popple, Andreas Schwab,
Aneesh Kumar K.V, Anju T Sudhakar, Arvind Yadav, Balbir Singh,
Benjamin Herrenschmidt, Bhumika Goyal, Breno Leitao, Bryant G. Ly,
Christophe Leroy, Cédric Le Goater, Dan Carpenter, Dou Liyang,
Frederic Barrat, Gautham R. Shenoy, Geliang Tang, Geoff Levand, Hannes
Reinecke, Haren Myneni, Ivan Mikhaylov, John Allen, Julia Lawall,
LABBE Corentin, Laurentiu Tudor, Madhavan Srinivasan, Markus Elfring,
Masahiro Yamada, Matt Brown, Michael Neuling, Murilo Opsfelder Araujo,
Nathan Fontenot, Naveen N. Rao, Nicholas Piggin, Oliver O'Halloran,
Paul Mackerras, Rashmica Gupta, Rob Herring, Rui Teng, Sam Bobroff,
Santosh Sivaraj, Scott Wood, Shilpasri G Bhat, Sukadev Bhattiprolu,
Suraj Jitindar Singh, Tobin C. Harding, Victor Aoqui"
* tag 'powerpc-4.14-1' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux: (321 commits)
powerpc/xive: Fix section __init warning
powerpc: Fix kernel crash in emulation of vector loads and stores
powerpc/xive: improve debugging macros
powerpc/xive: add XIVE Exploitation Mode to CAS
powerpc/xive: introduce H_INT_ESB hcall
powerpc/xive: add the HW IRQ number under xive_irq_data
powerpc/xive: introduce xive_esb_write()
powerpc/xive: rename xive_poke_esb() in xive_esb_read()
powerpc/xive: guest exploitation of the XIVE interrupt controller
powerpc/xive: introduce a common routine xive_queue_page_alloc()
powerpc/sstep: Avoid used uninitialized error
axonram: Return directly after a failed kzalloc() in axon_ram_probe()
axonram: Improve a size determination in axon_ram_probe()
axonram: Delete an error message for a failed memory allocation in axon_ram_probe()
powerpc/powernv/npu: Move tlb flush before launching ATSD
powerpc/macintosh: constify wf_sensor_ops structures
powerpc/iommu: Use permission-specific DEVICE_ATTR variants
powerpc/eeh: Delete an error out of memory message at init time
powerpc/mm: Use seq_putc() in two functions
macintosh: Convert to using %pOF instead of full_name
...
The nest MMU tlb flush needs to happen before the GPU translation
shootdown is launched to avoid the GPU refilling its tlb with stale
nmmu translations prior to the nmmu flush completing.
Fixes: 1ab66d1fba ("powerpc/powernv: Introduce address translation services for Nvlink2")
Cc: stable@vger.kernel.org # v4.12+
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Calls to mmu_notifier_invalidate_page() were replaced by calls to
mmu_notifier_invalidate_range() and now are bracketed by calls to
mmu_notifier_invalidate_range_start()/end()
Remove now useless invalidate_page callback.
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Cc: linuxppc-dev@lists.ozlabs.org
Cc: Alistair Popple <alistair@popple.id.au>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
NPU2 requires an extra explicit flush to an active GPU PID when
sending address translation shoot downs (ATSDs) to reliably flush the
GPU TLB. This patch adds just such a flush at the end of each sequence
of ATSDs.
We can safely use PID 0 which is always reserved and active on the
GPU. PID 0 is only used for init_mm which will never be a user mm on
the GPU. To enforce this we add a check in pnv_npu2_init_context()
just in case someone tries to use PID 0 on the GPU.
Signed-off-by: Alistair Popple <alistair@popple.id.au>
[mpe: Use true/false for bool literals]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Commit 4c3b89effc ("powerpc/powernv: Add sanity checks to
pnv_pci_get_{gpu|npu}_dev") introduced explicit warnings in
pnv_pci_get_npu_dev() when a PCIe device has no associated device-tree
node. However not all PCIe devices have an of_node and
pnv_pci_get_npu_dev() gets indirectly called at least once for every
PCIe device in the system. This results in spurious WARN_ON()'s so
remove it.
The same situation should not exist for pnv_pci_get_gpu_dev() as any
NPU based PCIe device requires a device-tree node.
Fixes: 4c3b89effc ("powerpc/powernv: Add sanity checks to pnv_pci_get_{gpu|npu}_dev")
Reported-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
opal_npu_destroy_context() should be called with the NPU PHB, not the
PCIe PHB.
Fixes: 1ab66d1fba ("powerpc/powernv: Introduce address translation services for Nvlink2")
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Commit 616badd2fb ("powerpc/powernv: Use OPAL call for TCE kill on
NVLink2") forced all TCE kills to go via the OPAL call for
NVLink2. However the PHB3 implementation of TCE kill was still being
called directly from some functions which in some circumstances caused
a machine check.
This patch adds an equivalent IODA2 version of the function which uses
the correct invalidation method depending on PHB model and changes all
external callers to use it instead.
Fixes: 616badd2fb ("powerpc/powernv: Use OPAL call for TCE kill on NVLink2")
Cc: stable@vger.kernel.org # v4.11+
Signed-off-by: Alistair Popple <alistair@popple.id.au>
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>
The pnv_pci_get_{gpu|npu}_dev functions are used to find associations
between nvlink PCIe devices and standard PCIe devices. However they
lacked basic sanity checking which results in NULL pointer
dereferencing if they are incorrect called can be harder to spot than
an explicit WARN_ON.
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
PHB, PE (and by association MVE) numbers are printed as a mix of decimal
and hexadecimal throughout the kernel. This can be misleading, so make
them all hexadecimal.
Standardising on hex instead of dec because:
- PHB numbers are presented in hex in sysfs/debugfs (and lspci, etc)
- PE numbers are presented as hex in sysfs and parsed in hex in debugfs
The only place I think this could cause confusing are the messages during
boot, i.e.
pci 000a:01 : [PE# 000] Secondary bus 1 associated with PE#0
which can be a quick way to check PE numbers. pe_level_printk() will
only print two characters instead of three, so the above would be
pci 000a:01 : [PE# 00] Secondary bus 1 associated with PE#0
which gives a hint it's in hex.
Signed-off-by: Russell Currey <ruscur@russell.cc>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Squash a bunch of sparse warnings by making things static.
Reviewed-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The dma-mapping core and the implementations do not change the DMA
attributes passed by pointer. Thus the pointer can point to const data.
However the attributes do not have to be a bitfield. Instead unsigned
long will do fine:
1. This is just simpler. Both in terms of reading the code and setting
attributes. Instead of initializing local attributes on the stack
and passing pointer to it to dma_set_attr(), just set the bits.
2. It brings safeness and checking for const correctness because the
attributes are passed by value.
Semantic patches for this change (at least most of them):
virtual patch
virtual context
@r@
identifier f, attrs;
@@
f(...,
- struct dma_attrs *attrs
+ unsigned long attrs
, ...)
{
...
}
@@
identifier r.f;
@@
f(...,
- NULL
+ 0
)
and
// Options: --all-includes
virtual patch
virtual context
@r@
identifier f, attrs;
type t;
@@
t f(..., struct dma_attrs *attrs);
@@
identifier r.f;
@@
f(...,
- NULL
+ 0
)
Link: http://lkml.kernel.org/r/1468399300-5399-2-git-send-email-k.kozlowski@samsung.com
Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>
Acked-by: Vineet Gupta <vgupta@synopsys.com>
Acked-by: Robin Murphy <robin.murphy@arm.com>
Acked-by: Hans-Christian Noren Egtvedt <egtvedt@samfundet.no>
Acked-by: Mark Salter <msalter@redhat.com> [c6x]
Acked-by: Jesper Nilsson <jesper.nilsson@axis.com> [cris]
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch> [drm]
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Acked-by: Joerg Roedel <jroedel@suse.de> [iommu]
Acked-by: Fabien Dessenne <fabien.dessenne@st.com> [bdisp]
Reviewed-by: Marek Szyprowski <m.szyprowski@samsung.com> [vb2-core]
Acked-by: David Vrabel <david.vrabel@citrix.com> [xen]
Acked-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> [xen swiotlb]
Acked-by: Joerg Roedel <jroedel@suse.de> [iommu]
Acked-by: Richard Kuo <rkuo@codeaurora.org> [hexagon]
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org> [m68k]
Acked-by: Gerald Schaefer <gerald.schaefer@de.ibm.com> [s390]
Acked-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Acked-by: Hans-Christian Noren Egtvedt <egtvedt@samfundet.no> [avr32]
Acked-by: Vineet Gupta <vgupta@synopsys.com> [arc]
Acked-by: Robin Murphy <robin.murphy@arm.com> [arm64 and dma-iommu]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The TCE invalidation functions are fairly implementation specific,
and while the IODA specs more/less describe the register, in practice
various implementation workarounds may be required. So name the
functions after the target PHB.
Note today and for the foreseeable future, there's a 1:1 relationship
between an IODA version and a PHB implementation. There exist another
variant of IODA1 (Torrent) but we never supported in with OPAL and
never will.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
IBM POWER8 NVlink systems come with Tesla K40-ish GPUs each of which
also has a couple of fast speed links (NVLink). The interface to links
is exposed as an emulated PCI bridge which is included into the same
IOMMU group as the corresponding GPU.
In the kernel, NPUs get a separate PHB of the PNV_PHB_NPU type and a PE
which behave pretty much as the standard IODA2 PHB except NPU PHB has
just a single TVE in the hardware which means it can have either
32bit window or 64bit window or DMA bypass but never two of these.
In order to make these links work when GPU is passed to the guest,
these bridges need to be passed as well; otherwise performance will
degrade.
This implements and exports API to manage NPU state in regard to VFIO;
it replicates iommu_table_group_ops.
This defines a new pnv_pci_ioda2_npu_ops which is assigned to
the IODA2 bridge if there are NPUs for a GPU on the bridge.
The new callbacks call the default IODA2 callbacks plus new NPU API.
This adds a gpe_table_group_to_npe() helper to find NPU PE for the IODA2
table_group, it is not expected to fail as the helper is only called
from the pnv_pci_ioda2_npu_ops.
This does not define NPU-specific .release_ownership() so after
VFIO is finished, DMA on NPU is disabled which is ok as the nvidia
driver sets DMA mask when probing which enable 32 or 64bit DMA on NPU.
This adds a pnv_pci_npu_setup_iommu() helper which adds NPUs to
the GPU group if any found. The helper uses helpers to look for
the "ibm,gpu" property in the device tree which is a phandle of
the corresponding GPU.
This adds an additional loop over PEs in pnv_ioda_setup_dma() as the main
loop skips NPU PEs as they do not have 32bit DMA segments.
As pnv_npu_set_window() and pnv_npu_unset_window() are started being used
by the new IODA2-NPU IOMMU group, this makes the helpers public and
adds the DMA window number parameter.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-By: Alistair Popple <alistair@popple.id.au>
[mpe: Add pnv_pci_ioda_setup_iommu_api() to fix build with IOMMU_API=n]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The pnv_ioda_pe struct keeps an array of peers. At the moment it is only
used to link GPU and NPU for 2 purposes:
1. Access NPU quickly when configuring DMA for GPU - this was addressed
in the previos patch by removing use of it as DMA setup is not what
the kernel would constantly do.
2. Invalidate TCE cache for NPU when it is invalidated for GPU.
GPU and NPU are in different PE. There is already a mechanism to
attach multiple iommu_table_group to the same iommu_table (used for VFIO),
we can reuse it here so does this patch.
This gets rid of peers[] array and PNV_IODA_PE_PEER flag as they are
not needed anymore.
While we are here, add TCE cache invalidation after enabling bypass.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-By: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The upcoming NVLink passthrough support will require NPU code to cope
with two DMA windows.
This adds a pnv_npu_set_window() helper which programs 32bit window to
the hardware. This also adds multilevel TCE support.
This adds a pnv_npu_unset_window() helper which removes the DMA window
from the hardware. This does not make difference now as the caller -
pnv_npu_dma_set_bypass() - enables bypass in the hardware but the next
patch will use it to manage TCE table lists for TCE Kill handling.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-By: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
