This is the last implementation of iommu_table_ops::exchange() which
we are about to remove.
This implements xchg_no_kill() for pseries. Since it is paravirtual
platform, the hypervisor does TCE invalidations and we do not have
to deal with it here, hence no tce_kill() hook.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190829085252.72370-5-aik@ozlabs.ru
Invalidating a TCE cache entry for each updated TCE is quite expensive.
This makes use of the new iommu_table_ops::xchg_no_kill()/tce_kill()
callbacks to bring down the time spent in mapping a huge guest DMA window;
roughly 20s to 10s for each guest's 100GB of DMA space.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Acked-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190829085252.72370-3-aik@ozlabs.ru
At the moment updates in a TCE table are made by iommu_table_ops::exchange
which update one TCE and invalidates an entry in the PHB/NPU TCE cache
via set of registers called "TCE Kill" (hence the naming).
Writing a TCE is a simple xchg() but invalidating the TCE cache is
a relatively expensive OPAL call. Mapping a 100GB guest with PCI+NPU
passed through devices takes about 20s.
Thankfully we can do better. Since such big mappings happen at the boot
time and when memory is plugged/onlined (i.e. not often), these requests
come in 512 pages so we call call OPAL 512 times less which brings 20s
from the above to less than 10s. Also, since TCE caches can be flushed
entirely, calling OPAL for 512 TCEs helps skiboot [1] to decide whether
to flush the entire cache or not.
This implements 2 new iommu_table_ops callbacks:
- xchg_no_kill() to update a single TCE with no TCE invalidation;
- tce_kill() to invalidate multiple TCEs.
This uses the same xchg_no_kill() callback for IODA1/2.
This implements 2 new wrappers on top of the new callbacks similar to
the existing iommu_tce_xchg().
This does not use the new callbacks yet, the next patches will;
so this should not cause any behavioral change.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190829085252.72370-2-aik@ozlabs.ru
H_PUT_TCE_INDIRECT handlers receive a page with up to 512 TCEs from
a guest. Although we verify correctness of TCEs before we do anything
with the existing tables, there is a small window when a check in
kvmppc_tce_validate might pass and right after that the guest alters
the page with TCEs which can cause early exit from the handler and
leave srcu_read_lock(&vcpu->kvm->srcu) (virtual mode) or lock_rmap(rmap)
(real mode) locked.
This fixes the bug by jumping to the common exit code with an appropriate
unlock.
Fixes: 121f80ba68 ("KVM: PPC: VFIO: Add in-kernel acceleration for VFIO")
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190826045520.92153-1-aik@ozlabs.ru
The existing code uses bunch of hardcoded values from the PCI Bus
Binding to IEEE Std 1275 spec; and it does so in quite non-obvious
way.
This defines fields from the cell#0 of the "reg" property of a PCI
device and uses them for parsing.
This should cause no behavioral change.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
[mpe: Unsplit some 80/81 char lines, space the code with some newlines]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190829084417.71873-1-aik@ozlabs.ru
This switches to using common code for the DMA allocations, including
potential use of the CMA allocator if configured.
Switching to the generic code enables DMA allocations from atomic
context, which is required by the DMA API documentation, and also
adds various other minor features drivers start relying upon. It
also makes sure we have on tested code base for all architectures
that require uncached pte bits for coherent DMA allocations.
Another advantage is that consistent memory allocations now share
the general vmalloc pool instead of needing an explicit careout
from it.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Tested-by: Christophe Leroy <christophe.leroy@c-s.fr> # tested on 8xx
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190814132230.31874-2-hch@lst.de
Powerpc 601 is rather old powerpc which as some important
limitations compared to other book3s/32 powerpcs:
- No Timebase.
- Common BATs for instruction and data.
- No execution protection in segment registers.
- No RI bit in MSR
- ...
It is starting to be difficult and cumbersome to maintain
kernels that are compatible both with 601 and other 6xx cores.
Create a compiletime option to exclusively select either powerpc 601
or other 6xx.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/d644eaf7dff8cc149260066802af230bdf34fded.1566834712.git.christophe.leroy@c-s.fr
Prior to commit 1bd98d7fbaf5 ("ppc64: Update BUG handling based on
ppc32"), BUG() family was using BUG_ILLEGAL_INSTRUCTION which
was an invalid instruction opcode to trap into program check
exception.
That commit converted them to using standard trap instructions,
but prom/prom_init and their PROM_BUG() macro were left over.
head_64.S and exception-64s.S were left aside as well.
Convert them to using the standard BUG infrastructure.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/cdaf4bbbb64c288a077845846f04b12683f8875a.1566817807.git.christophe.leroy@c-s.fr
Booting w/ppc64le_defconfig + CONFIG_PREEMPT on bare metal results in
the oops below due to calling into __spin_yield() when not running in
an SPLPAR, which means lppaca pointers are NULL.
We fixed a similar case previously in commit a6201da34f ("powerpc:
Fix oops due to bad access of lppaca on bare metal"), by adding SPLPAR
checks in lppaca_shared_proc(). However when PREEMPT is enabled we can
call __spin_yield() directly from arch_spin_yield().
To fix it add spin_yield() and rw_yield() which check that
shared-processor LPAR is enabled before calling the SPLPAR-only
implementation of each.
BUG: Kernel NULL pointer dereference at 0x00000100
Faulting instruction address: 0xc000000000097f88
Oops: Kernel access of bad area, sig: 7 [#1]
LE PAGE_SIZE=64K MMU=Radix MMU=Hash PREEMPT SMP NR_CPUS=2048 NUMA PowerNV
Modules linked in:
CPU: 0 PID: 2 Comm: kthreadd Not tainted 5.2.0-rc6-00491-g249155c20f9b #28
NIP: c000000000097f88 LR: c000000000c07a88 CTR: c00000000015ca10
REGS: c0000000727079f0 TRAP: 0300 Not tainted (5.2.0-rc6-00491-g249155c20f9b)
MSR: 9000000002009033 <SF,HV,VEC,EE,ME,IR,DR,RI,LE> CR: 84000424 XER: 20040000
CFAR: c000000000c07a84 DAR: 0000000000000100 DSISR: 00080000 IRQMASK: 1
GPR00: c000000000c07a88 c000000072707c80 c000000001546300 c00000007be38a80
GPR04: c0000000726f0c00 0000000000000002 c00000007279c980 0000000000000100
GPR08: c000000001581b78 0000000080000001 0000000000000008 c00000007279c9b0
GPR12: 0000000000000000 c000000001730000 c000000000142558 0000000000000000
GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
GPR20: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
GPR24: c00000007be38a80 c000000000c002f4 0000000000000000 0000000000000000
GPR28: c000000072221a00 c0000000726c2600 c00000007be38a80 c00000007be38a80
NIP [c000000000097f88] __spin_yield+0x48/0xa0
LR [c000000000c07a88] __raw_spin_lock+0xb8/0xc0
Call Trace:
[c000000072707c80] [c000000072221a00] 0xc000000072221a00 (unreliable)
[c000000072707cb0] [c000000000bffb0c] __schedule+0xbc/0x850
[c000000072707d70] [c000000000c002f4] schedule+0x54/0x130
[c000000072707da0] [c0000000001427dc] kthreadd+0x28c/0x2b0
[c000000072707e20] [c00000000000c1cc] ret_from_kernel_thread+0x5c/0x70
Instruction dump:
4d9e0020 552a043e 210a07ff 79080fe0 0b080000 3d020004 3908b878 794a1f24
e8e80000 7ce7502a e8e70000 38e70100 <7ca03c2c> 70a70001 78a50020 4d820020
---[ end trace 474d6b2b8fc5cb7e ]---
Fixes: 499dcd4137 ("powerpc/64s: Allocate LPPACAs individually")
Signed-off-by: Christopher M. Riedl <cmr@informatik.wtf>
[mpe: Reword change log a bit]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190813031314.1828-4-cmr@informatik.wtf
On POWER9, when userspace reads the value of the DPDES register on a
vCPU, it is possible for 0 to be returned although there is a doorbell
interrupt pending for the vCPU. This can lead to a doorbell interrupt
being lost across migration. If the guest kernel uses doorbell
interrupts for IPIs, then it could malfunction because of the lost
interrupt.
This happens because a newly-generated doorbell interrupt is signalled
by setting vcpu->arch.doorbell_request to 1; the DPDES value in
vcpu->arch.vcore->dpdes is not updated, because it can only be updated
when holding the vcpu mutex, in order to avoid races.
To fix this, we OR in vcpu->arch.doorbell_request when reading the
DPDES value.
Cc: stable@vger.kernel.org # v4.13+
Fixes: 579006944e ("KVM: PPC: Book3S HV: Virtualize doorbell facility on POWER9")
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Tested-by: Alexey Kardashevskiy <aik@ozlabs.ru>
When we are running multiple vcores on the same physical core, they
could be from different VMs and so it is possible that one of the
VMs could have its arch.mmu_ready flag cleared (for example by a
concurrent HPT resize) when we go to run it on a physical core.
We currently check the arch.mmu_ready flag for the primary vcore
but not the flags for the other vcores that will be run alongside
it. This adds that check, and also a check when we select the
secondary vcores from the preempted vcores list.
Cc: stable@vger.kernel.org # v4.14+
Fixes: 38c53af853 ("KVM: PPC: Book3S HV: Fix exclusion between HPT resizing and other HPT updates")
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Today LOAD_REG_IMMEDIATE() is a basic #define which loads all
parts on a value into a register, including the parts that are NUL.
This means always 2 instructions on PPC32 and always 5 instructions
on PPC64. And those instructions cannot run in parallele as they are
updating the same register.
Ex: LOAD_REG_IMMEDIATE(r1,THREAD_SIZE) in head_64.S results in:
3c 20 00 00 lis r1,0
60 21 00 00 ori r1,r1,0
78 21 07 c6 rldicr r1,r1,32,31
64 21 00 00 oris r1,r1,0
60 21 40 00 ori r1,r1,16384
Rewrite LOAD_REG_IMMEDIATE() with GAS macro in order to skip
the parts that are NUL.
Rename existing LOAD_REG_IMMEDIATE() as LOAD_REG_IMMEDIATE_SYM()
and use that one for loading value of symbols which are not known
at compile time.
Now LOAD_REG_IMMEDIATE(r1,THREAD_SIZE) in head_64.S results in:
38 20 40 00 li r1,16384
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/d60ce8dd3a383c7adbfc322bf1d53d81724a6000.1566311636.git.christophe.leroy@c-s.fr
PPC32 and PPC64 are doing the same once SLAB is available.
Create a do_ioremap() function that calls get_vm_area and
do the mapping.
For PPC64, we add the 4K PFN hack sanity check to __ioremap_caller()
in order to avoid using __ioremap_at(). Other checks in __ioremap_at()
are irrelevant for __ioremap_caller().
On PPC64, VM area is allocated in the range [ioremap_bot ; IOREMAP_END]
On PPC32, VM area is allocated in the range [VMALLOC_START ; VMALLOC_END]
Lets define IOREMAP_START is ioremap_bot for PPC64, and alias
IOREMAP_START/END to VMALLOC_START/END on PPC32
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/42e7e36ad32e0fdf76692426cc642799c9f689b8.1566309263.git.christophe.leroy@c-s.fr
book3s64's ioremap_range() is almost same as fallback ioremap_range(),
except that it calls radix__ioremap_range() when radix is enabled.
radix__ioremap_range() is also very similar to the other ones, expect
that it calls ioremap_page_range when slab is available.
PPC32 __ioremap_caller() have a loop doing the same thing as
ioremap_range() so use it on PPC32 as well.
Lets keep only one version of ioremap_range() which calls
ioremap_page_range() on all platforms when slab is available.
At the same time, drop the nid parameter which is not used.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/4b1dca7096b01823b101be7338983578641547f1.1566309263.git.christophe.leroy@c-s.fr
ppc_md.ioremap() is only used for I/O workaround on CELL platform,
so indirect function call can be avoided.
This patch reworks the io-workaround and ioremap() functions to
use the global 'io_workaround_inited' flag for the activation
of io-workaround.
When CONFIG_PPC_IO_WORKAROUNDS or CONFIG_PPC_INDIRECT_MMIO are not
selected, the I/O workaround ioremap() voids and the global flag is
not used.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/5fa3ef069fbd0f152512afaae19e7a60161454cf.1566309262.git.christophe.leroy@c-s.fr
There are some POWER9 machines where the OPAL firmware does not support
the OPAL_XIVE_GET_QUEUE_STATE and OPAL_XIVE_SET_QUEUE_STATE calls.
The impact of this is that a guest using XIVE natively will not be able
to be migrated successfully. On the source side, the get_attr operation
on the KVM native device for the KVM_DEV_XIVE_GRP_EQ_CONFIG attribute
will fail; on the destination side, the set_attr operation for the same
attribute will fail.
This adds tests for the existence of the OPAL get/set queue state
functions, and if they are not supported, the XIVE-native KVM device
is not created and the KVM_CAP_PPC_IRQ_XIVE capability returns false.
Userspace can then either provide a software emulation of XIVE, or
else tell the guest that it does not have a XIVE controller available
to it.
Cc: stable@vger.kernel.org # v5.2+
Fixes: 3fab2d1058 ("KVM: PPC: Book3S HV: XIVE: Activate XIVE exploitation mode")
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
H_PUT_TCE_INDIRECT handlers receive a page with up to 512 TCEs from
a guest. Although we verify correctness of TCEs before we do anything
with the existing tables, there is a small window when a check in
kvmppc_tce_validate might pass and right after that the guest alters
the page of TCEs, causing an early exit from the handler and leaving
srcu_read_lock(&vcpu->kvm->srcu) (virtual mode) or lock_rmap(rmap)
(real mode) locked.
This fixes the bug by jumping to the common exit code with an appropriate
unlock.
Cc: stable@vger.kernel.org # v4.11+
Fixes: 121f80ba68 ("KVM: PPC: VFIO: Add in-kernel acceleration for VFIO")
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
The rmap array in the guest memslot is an array of size number of guest
pages, allocated at memslot creation time. Each rmap entry in this array
is used to store information about the guest page to which it
corresponds. For example for a hpt guest it is used to store a lock bit,
rc bits, a present bit and the index of a hpt entry in the guest hpt
which maps this page. For a radix guest which is running nested guests
it is used to store a pointer to a linked list of nested rmap entries
which store the nested guest physical address which maps this guest
address and for which there is a pte in the shadow page table.
As there are currently two uses for the rmap array, and the potential
for this to expand to more in the future, define a type field (being the
top 8 bits of the rmap entry) to be used to define the type of the rmap
entry which is currently present and define two values for this field
for the two current uses of the rmap array.
Since the nested case uses the rmap entry to store a pointer, define
this type as having the two high bits set as is expected for a pointer.
Define the hpt entry type as having bit 56 set (bit 7 IBM bit ordering).
Signed-off-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Fix the error below triggered by `-Wimplicit-fallthrough`, by tagging
it as an expected fall-through.
arch/powerpc/kvm/book3s_32_mmu.c: In function ‘kvmppc_mmu_book3s_32_xlate_pte’:
arch/powerpc/kvm/book3s_32_mmu.c:241:21: error: this statement may fall through [-Werror=implicit-fallthrough=]
pte->may_write = true;
~~~~~~~~~~~~~~~^~~~~~
arch/powerpc/kvm/book3s_32_mmu.c:242:5: note: here
case 3:
^~~~
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
This merges in fixes for the XIVE interrupt controller which touch both
generic powerpc and PPC KVM code. To avoid merge conflicts, these
commits will go upstream via the powerpc tree as well as the KVM tree.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Pull RCU and LKMM changes from Paul E. McKenney:
- A few more RCU flavor consolidation cleanups.
- Miscellaneous fixes.
- Updates to RCU's list-traversal macros improving lockdep usability.
- Torture-test updates.
- Forward-progress improvements for no-CBs CPUs: Avoid ignoring
incoming callbacks during grace-period waits.
- Forward-progress improvements for no-CBs CPUs: Use ->cblist
structure to take advantage of others' grace periods.
- Also added a small commit that avoids needlessly inflicting
scheduler-clock ticks on callback-offloaded CPUs.
- Forward-progress improvements for no-CBs CPUs: Reduce contention
on ->nocb_lock guarding ->cblist.
- Forward-progress improvements for no-CBs CPUs: Add ->nocb_bypass
list to further reduce contention on ->nocb_lock guarding ->cblist.
- LKMM updates.
Signed-off-by: Ingo Molnar <mingo@kernel.org>
We still treat devices without a DMA mask as defaulting to 32-bits for
both mask, but a few releases ago we've started warning about such
cases, as they require special cases to work around this sloppyness.
Add a dma_mask field to struct platform_device so that we can initialize
the dma_mask pointer in struct device and initialize both masks to
32-bits by default, replacing similar functionality in m68k and
powerpc. The arch_setup_pdev_archdata hooks is now unused and removed.
Note that the code looks a little odd with the various conditionals
because we have to support platform_device structures that are
statically allocated.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Link: https://lore.kernel.org/r/20190816062435.881-7-hch@lst.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
While reviewing lockdown patches, I discovered that we still enable
/dev/port (CONFIG_DEVPORT) in skiroot.
/dev/port is used for old x86 style IO accesses. It's set up in
drivers/char/mem.c, and is only created if arch_has_dev_port() returns
true. Per arch/powerpc/include/asm/io.h, on PPC64 with PCI, this is
only true if there's a legacy ISA bridge.
Even if a system has a legacy ISA bridge installed, we have no
business accessing it in skiroot.
Deselect CONFIG_DEVPORT for skiroot.
Signed-off-by: Daniel Axtens <dja@axtens.net>
[mpe: Incorporate emailed comments into the change log]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190627053008.29315-1-dja@axtens.net
