vfio_container_ioctl() was a bad interface that bypassed abstraction
boundaries, had semantics that sat uneasily with its name, and was unsafe
in many realistic circumstances. Now that spapr-pci-vfio-host-bridge has
been folded into spapr-pci-host-bridge, there are no more users, so remove
it.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Acked-by: Alex Williamson <alex.williamson@redhat.com>
Now that spapr-pci-vfio-host-bridge is reduced to just a stub, there is
only one implementation of the finish_realize hook in sPAPRPHBClass. So,
we can fold that implementation into its (single) caller, and remove the
hook. That's the last thing left in sPAPRPHBClass, so that can go away as
well.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Now that the regular spapr-pci-host-bridge can handle EEH, there are only
two things that spapr-pci-vfio-host-bridge does differently:
1. automatically sizes its DMA window to match the host IOMMU
2. checks if the attached VFIO container is backed by the
VFIO_SPAPR_TCE_IOMMU type on the host
(1) is not particularly useful, since the default window used by the
regular host bridge will work with the host IOMMU configuration on all
current systems anyway.
Plus, automatically changing guest visible configuration (such as the DMA
window) based on host settings is generally a bad idea. It's not
definitively broken, since spapr-pci-vfio-host-bridge is only supposed to
support VFIO devices which can't be migrated anyway, but still.
(2) is not really useful, because if a guest tries to configure EEH on a
different host IOMMU, the first call will fail and that will be that.
It's possible there are scripts or tools out there which expect
spapr-pci-vfio-host-bridge, so we don't remove it entirely. This patch
reduces it to just a stub for backwards compatibility.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Now that the EEH code is independent of the special
spapr-vfio-pci-host-bridge device, we can allow it on all spapr PCI
host bridges instead. We do this by changing spapr_phb_eeh_available()
to be based on the vfio_eeh_as_ok() call instead of the host bridge class.
Because the value of vfio_eeh_as_ok() can change with devices being
hotplugged or unplugged, this can potentially lead to some strange edge
cases where the guest starts using EEH, then it starts failing because
of a change in status.
However, it's not really any worse than the current situation. Cases that
would have worked previously will still work (i.e. VFIO devices from at
most one VFIO IOMMU group per vPHB), it's just that it's no longer
necessary to use spapr-vfio-pci-host-bridge with the groupid pre-specified.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
The EEH operations in the spapr-vfio-pci-host-bridge no longer rely on the
special groupid field in sPAPRPHBVFIOState. So we can simplify, removing
the class specific callbacks with direct calls based on a simple
spapr_phb_eeh_enabled() helper. For now we implement that in terms of
a boolean in the class, but we'll continue to clean that up later.
On its own this is a rather strange way of doing things, but it's a useful
intermediate step to further cleanups.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
This switches all EEH on VFIO operations in spapr_pci_vfio.c from the
broken vfio_container_ioctl() interface to the new vfio_as_eeh_op()
interface.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
At present the code handling IBM's Enhanced Error Handling (EEH) interface
on VFIO devices operates by bypassing the usual VFIO logic with
vfio_container_ioctl(). That's a poorly designed interface with unclear
semantics about exactly what can be operated on.
In particular it operates on a single vfio container internally (hence the
name), but takes an address space and group id, from which it deduces the
container in a rather roundabout way. groupids are something that code
outside vfio shouldn't even be aware of.
This patch creates new interfaces for EEH operations. Internally we
have vfio_eeh_container_op() which takes a VFIOContainer object
directly. For external use we have vfio_eeh_as_ok() which determines
if an AddressSpace is usable for EEH (at present this means it has a
single container with exactly one group attached), and vfio_eeh_as_op()
which will perform an operation on an AddressSpace in the unambiguous case,
and otherwise returns an error.
This interface still isn't great, but it's enough of an improvement to
allow a number of cleanups in other places.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Acked-by: Alex Williamson <alex.williamson@redhat.com>
fa48b43 "target-ppc: Remove hack for ppc_hash64_load_hpte*() with HV KVM"
purports to remove a hack in the handling of hash page tables (HPTs)
managed by KVM instead of qemu. However, it actually went in the wrong
direction.
That patch requires anything looking for an external HPT (that is one not
managed by the guest itself) to check both env->external_htab (for a qemu
managed HPT) and kvmppc_kern_htab (for a KVM managed HPT). That's a
problem because kvmppc_kern_htab is local to mmu-hash64.c, but some places
which need to check for an external HPT are outside that, such as
kvm_arch_get_registers(). The latter was subtly broken by the earlier
patch such that gdbstub can no longer access memory.
Basically a KVM managed HPT is much more like a qemu managed HPT than it is
like a guest managed HPT, so the original "hack" was actually on the right
track.
This partially reverts fa48b43, so we again mark a KVM managed external HPT
by putting a special but non-NULL value in env->external_htab. It then
goes further, using that marker to eliminate the kvmppc_kern_htab global
entirely. The ppc_hash64_set_external_hpt() helper function is extended
to set that marker if passed a NULL value (if you're setting an external
HPT, but don't have an actual HPT to set, the assumption is that it must
be a KVM managed HPT).
This also has some flow-on changes to the HPT access helpers, required by
the above changes.
Reported-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Tested-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
When a Power cpu with 64-bit hash MMU has it's hash page table (HPT)
pointer updated by a write to the SDR1 register we need to update some
derived variables. Likewise, when the cpu is configured for an external
HPT (one not in the guest memory space) some derived variables need to be
updated.
Currently the logic for this is (partially) duplicated in ppc_store_sdr1()
and in spapr_cpu_reset(). In future we're going to need it in some other
places, so make some common helpers for this update.
In addition the new ppc_hash64_set_external_hpt() helper also updates
SDR1 in KVM - it's not updated by the normal runtime KVM <-> qemu CPU
synchronization. In a sense this belongs logically in the
ppc_hash64_set_sdr1() helper, but that is called from
kvm_arch_get_registers() so can't itself call cpu_synchronize_state()
without infinite recursion. In practice this doesn't matter because
the only other caller is TCG specific.
Currently there aren't situations where updating SDR1 at runtime in KVM
matters, but there are going to be in future.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Currently the getting and setting of Power MMU registers (sregs) take up
large inline chunks of the kvm_arch_get_registers() and
kvm_arch_put_registers() functions. Especially since there are two
variants (for Book-E and Book-S CPUs), only one of which will be used in
practice, this is pretty hard to read.
This patch splits these out into helper functions for clarity. No
functional change is expected.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Commit 4b23699 "pseries: Add pseries-2.6 machine type" added a new
SPAPR_COMPAT_2_5 macro in the usual way. However, it didn't add this
macro to the existing SPAPR_COMPAT_2_4 macro so that pseries-2.4
inherits newer compatibility properties which are needed for 2.5 and
earlier.
This corrects the oversight.
Reported-by: Laszlo Ersek <lersek@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
htab_save_first_pass could return without finishing its work due to
timeout. The patch checks if another invocation of it is necessary and
will call it in htab_save_complete if necessary.
Signed-off-by: Jianjun Duan <duanj@linux.vnet.ibm.com>
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
[removed overlong line]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
With HV KVM, the guest's hash page table (HPT) is managed by the kernel and
not directly accessible to QEMU. This means that spapr->htab is NULL
and normally env->external_htab would also be NULL for each cpu.
However, that would cause ppc_hash64_load_hpte*() to do the wrong thing in
the few cases where QEMU does need to load entries from the in-kernel HPT.
Specifically, seeing external_htab is NULL, they would look for an HPT
within the guest's address space instead.
To stop that we have an ugly hack in the pseries machine type code to
set external htab to (void *)1 instead.
This patch removes that hack by having ppc_hash64_load_hpte*() explicitly
check kvmppc_kern_htab instead, which makes more sense.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
At the moment the size of the hash page table (HPT) is fixed based on the
maximum memory allowed to the guest. As such, we allocate the table during
machine construction, and just clear it at reset.
However, we're planning to implement a PAPR extension allowing the hash
page table to be resized at runtime. This will mean that on reset we want
to revert it to the default size. It also means that when migrating, we
need to make sure the destination allocates an HPT of size matching the
host, since the guest could have changed it before the migration.
This patch replaces the spapr_alloc_htab() and spapr_reset_htab() functions
with a new spapr_reallocate_hpt() function. This is called at reset and
inbound migration only, not during machine init any more.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
At present we calculate the recommended hash page table (HPT) size for a
pseries guest just once in ppc_spapr_init() before allocating the HPT.
In future patches we're going to want this calculation in other places, so
this splits it out into a helper function. While we're at it, change the
calculation to use ctz() instead of an explicit loop.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
When migrating the 'pseries' machine type with KVM, we use a special fd
to access the hash page table stored within KVM. Usually, this fd is
opened at the beginning of migration, and kept open until the migration
is complete.
However, if there is a guest reset during the migration, the fd can become
stale and we need to re-open it. At the moment we use an 'htab_fd_stale'
flag in sPAPRMachineState to signal this, which is checked in the migration
iterators.
But that's rather ugly. It's simpler to just close and invalidate the
fd on reset, and lazily re-open it in migration if necessary. This patch
implements that change.
This requires a small addition to the machine state's instance_init,
so that htab_fd is initialized to -1 (telling the migration code it
needs to open it) instead of 0, which could be a valid fd.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
The HMP command "info registers" produces somewhat different information on
different ppc cpu variants. For those with a hash MMU it's supposed to
include the SDR1, DAR and DSISR registers related to the MMU. However,
the switch is missing a couple of MMU model variants, meaning we will
miss out this information on certain CPUs which should have it.
This patch corrects the oversight. (Really these MMU model IDs need a big
cleanup, but we might as well fix the bug in the interim).
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
The 'base' field of MemoryHotplugState is ram_addr_t, which indicates that
it exists in the abstract address space of RAM regions.
However, the actual usage of this field indicates that it is a concrete
physical address (it's passed as an offset to memory_region_add_subgregion
for example).
So, correct its type to 'hwaddr'.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Acked-by: Eduardo Habkost <ehabkost@redhat.com>
Now that the TCG and spapr code has been extended to allow (semi-)
arbitrary page encodings in the CPU's 'sps' table, we can add the many
page sizes supported by real POWER7 and POWER8 hardware that we previously
didn't support in TCG.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Alexander Graf <agraf@suse.de>
h_enter() in the spapr code needs to know the page size of the HPTE it's
about to insert. Unlike other paths that do this, it doesn't have access
to the SLB, so at the moment it determines this with some open-coded
tests which assume POWER7 or POWER8 page size encodings.
To make this more flexible add ppc_hash64_hpte_page_shift_noslb() to
determine both the "base" page size per segment, and the individual
effective page size from an HPTE alone.
This means that the spapr code should now be able to handle any page size
listed in the env->sps table.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Alexander Graf <agraf@suse.de>
When HPTEs are removed or modified by hypercalls on spapr, we need to
invalidate the relevant pages in the qemu TLB.
Currently we do that by doing some complicated calculations to work out the
right encoding for the tlbie instruction, then passing that to
ppc_tlb_invalidate_one()... which totally ignores the argument and flushes
the whole tlb.
Avoid that by adding a new flush-by-hpte helper in mmu-hash64.c.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Alexander Graf <agraf@suse.de>
Currently both the tlbiva instruction (used on 44x chips) and the tlbie
instruction (used on hash MMU chips) are both handled via
ppc_tlb_invalidate_one(). This is silly, because they're invoked from
different places, and do different things.
Clean this up by separating out the tlbiva instruction into its own
handling. In fact the implementation is only a stub anyway.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Alexander Graf <agraf@suse.de>
ppc_tlb_invalidate_one() has a big switch handling many different MMU
types. However, most of those branches can never be reached:
It is called from 3 places: from remove_hpte() and h_protect() in
spapr_hcall.c (which always has a 64-bit hash MMU type), and from
helper_tlbie() in mmu_helper.c.
Calls to helper_tlbie() are generated from gen_tlbiel, gen_tlbiel and
gen_tlbiva. The first two are only used with the PPC_MEM_TLBIE flag,
set only with 32-bit or 64-bit hash MMU models, and gen_tlbiva() is
used only on 440 and 460 models with the BookE mmu model.
These means the exhaustive list of MMU types which may call
ppc_tlb_invalidate_one() is: POWERPC_MMU_SOFT_6xx, POWERPC_MMU_601,
POWERPC_MMU_32B, POWERPC_MMU_SOFT_74xx, POWERPC_MMU_64B, POWERPC_MMU_2_03,
POWERPC_MMU_2_06, POWERPC_MMU_2_07 and POWERPC_MMU_BOOKE.
Clean up by removing logic for all other MMU types from
ppc_tlb_invalidate_one().
This means that ppc4xx_tlb_invalidate_virt() now has no callers, or rather,
makes it obvious that it has no callers. So, we remove that function as
well.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
At present the 64-bit hash MMU code uses information from the SLB to
determine the page size of a translation. We do need that information to
correctly look up the hash table. However the MMU also allows a
possibly larger page size to be encoded into the HPTE itself, which is used
to populate the TLB. At present qemu doesn't check that, and so doesn't
support the MPSS "Multiple Page Size per Segment" feature.
This makes a start on allowing this, by adding an hpte_page_shift()
function which looks up the page size of an HPTE. We use this to validate
page sizes encodings on faults, and populate the qemu TLB with larger
page sizes when appropriate.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Alexander Graf <agraf@suse.de>
Currently, the ppc_hash64_page_shift() function looks up a page size based
on information in an SLB entry. It open codes the bit translation for
existing CPUs, however different CPU models can have different SLB
encodings. We already store those in the 'sps' table in CPUPPCState, but
we don't currently enforce that that actually matches the logic in
ppc_hash64_page_shift.
This patch reworks lookup of page size from SLB in several ways:
* ppc_store_slb() will now fail (triggering an illegal instruction
exception) if given a bad SLB page size encoding
* On success ppc_store_slb() stores a pointer to the relevant entry in
the page size table in the SLB entry. This is looked up directly from
the published table of page size encodings, so can't get out ot sync.
* ppc_hash64_htab_lookup() and others now use this precached page size
information rather than decoding the SLB values
* Now that callers have easy access to the page_shift,
ppc_hash64_pte_raddr() amounts to just a deposit64(), so remove it and
have the callers use deposit64() directly.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Alexander Graf <agraf@suse.de>
ppc_store_slb updates the SLB for PPC cpus with 64-bit hash MMUs.
Currently it takes two parameters, which contain values encoded as the
register arguments to the slbmte instruction, one register contains the
ESID portion of the SLBE and also the slot number, the other contains the
VSID portion of the SLBE.
We're shortly going to want to do some SLB updates from other code where
it is more convenient to supply the slot number and ESID separately, so
rework this function and its callers to work this way.
As a bonus, this slightly simplifies the emulation of segment registers for
when running a 32-bit OS on a 64-bit CPU.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Alexander Graf <agraf@suse.de>
Like a lot of places these files include a mixture of functions taking
both the older CPUPPCState *env and newer PowerPCCPU *cpu. Move a step
closer to cleaning this up by standardizing on PowerPCCPU, except for the
helper_* functions which are called with the CPUPPCState * from tcg.
Callers and some related functions are updated as well, the boundaries of
what's changed here are a bit arbitrary.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Alexander Graf <agraf@suse.de>
This stub function is in the !KVM ifdef in target-ppc/kvm_ppc.h. However
no such function exists on the KVM side, or is ever used.
I think this originally referenced a function which read host page size
information from /proc, for we we now use the KVM GET_SMMU_INFO extension
instead.
In any case, it has no function now, so remove it.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Alexander Graf <agraf@suse.de>
The implementation of the H_ENTER hypercall for PAPR guests needs to
enforce correct access attributes on the inserted HPTE. This means
determining if the HPTE's real address is a regular RAM address (which
requires attributes for coherent access) or an IO address (which requires
attributes for cache-inhibited access).
At the moment this check is implemented with (raddr < machine->ram_size),
but that only handles addresses in the base RAM area, not any hotplugged
RAM.
This patch corrects the problem with a new helper.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
The functions for migrating the hash page table on pseries machine type
(htab_save_setup() and htab_load()) can report some errors with an
explicit fprintf() before returning an appropriate error code. Change some
of these to use error_report() instead. htab_save_setup() is omitted for
now to avoid conflicts with some other in-progress work.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
This function includes a number of explicit fprintf()s for errors.
Change these to use error_report() instead.
Also replace the single exit(EXIT_FAILURE) with an explicit exit(1), since
the latter is the more usual idiom in qemu by a large margin.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Use the error handling infrastructure to pass an error out from
try_create_xics() instead of assuming &error_abort - the caller is in a
better position to decide on error handling policy.
Also change the error handling from an &error_abort to &error_fatal, since
this occurs during the initial machine construction and could be triggered
by bad configuration rather than a program error.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
The errors detected in this function necessarily indicate bugs in the rest
of the qemu code, rather than an external or configuration problem.
So, a simple assert() is more appropriate than any more complex error
reporting.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Use error_setg() to return an error rather than an explicit exit().
Previously it was an exit(0) instead of a non-zero exit code, which was
simply a bug. Also improve the error message.
While we're at it change the type of spapr_vga_init() to bool since that's
how we're using it anyway.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Use error_setg() and return an error, rather than using an explicit exit().
Also improve messages, and be more explicit about which constraint failed.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Currently spapr_cpu_init() is hardcoded to handle any errors as fatal.
That works for now, since it's only called from initial setup where an
error here means we really can't proceed.
However, we'll want to handle this more flexibly for cpu hotplug in future
so generalize this using the error reporting infrastructure. While we're
at it make a small cleanup in a related part of ppc_spapr_init() to use
error_report() instead of an old-style explicit fprintf().
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Current ppc_set_compat() returns -1 for errors, and also (unconditionally)
reports an error message. The caller in h_client_architecture_support()
may then report it again using an outdated fprintf().
Clean this up by using the modern error reporting mechanisms. Also add
strerror(errno) to the error message.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
h_client_architecture_support() uses rtas_ld() for general purpose memory
access, despite the fact that it's not an RTAS routine at all and rtas_ld
makes things more awkward.
Clean this up by replacing rtas_ld() calls with appropriate ldXX_phys()
calls.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
rtas_st_buffer_direct() is a not particularly useful wrapper around
cpu_physical_memory_write(). All the callers are in
rtas_ibm_configure_connector, where it's better handled by local helper.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
rtas_st_buffer() appears in spapr.h as though it were a widely used helper,
but in fact it is only used for saving data in a format used by
rtas_ibm_get_system_parameter(). This changes it to a local helper more
specifically for that function.
While we're there fix a couple of small defects in
rtas_ibm_get_system_parameter:
- For the string value SPLPAR_CHARACTERISTICS, it wasn't including the
terminating \0 in the length which it should according to LoPAPR
7.3.16.1
- It now checks that the supplied buffer has at least enough space for
the length of the returned data, and returns an error if it does not.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
This tweaks the way the default machine version is controlled, so that
there will be a bit less churn when each new version is introduced.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Currently each of the *_class_options() functions for the pseries-2.1 ..
pseries-2.5 machine types are standalone. This will become harder to
maintain as new versions are added.
This patch restructures them similarly to x86 where each function calls
the one from the next version, then overrides anything necessary for
compatibility with the specific version and older.
The default behaviour - that for the most recent machine are set up in
the base class initializer spapr_machine_class_init(). Previously it had
some things set up to default to older behaviour with the more recent
machines overriding it.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
At the moment all the class_init functions and TypeInfo structures for the
various versioned pseries machine types are open-coded. As more versions
are created this is getting increasingly clumsy.
This patch borrows the approach used in PC, using a DEFINE_SPAPR_MACHINE()
macro to construct most of the boilerplate from simpler 'class_options' and
'instance_options' functions.
This patch makes a small semantic change - the versioned machine types are
now registered through machine_init() instead of type_init(). Since the
new way is how PC already did it, I'm assuming that's correct.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Sam Bobroff <sam.bobroff@au1.ibm.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
pc.h defines a SET_MACHINE_COMPAT macro to make setting up compat_props
for the various PC machine versions less verbose. There's nothing
inherently PC specific about it, though, so move it to boards.h where other
versioned machine types (like pseries-*) can use it.
While we're doing that, change it's indentation to be a bit more regular.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Currently, the versioned spapr machine types put the machine type version
into the description string. PC does not do this, using just the name
itself to distinguish. Doing the same lets us move setting the description
into the common base class, simplifying the code slightly.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
The instance_init() functions for several of the pseries-x.y versioned
machine types explicitly call spapr_machine_initfn(). But that's the
instance_init function for the common parent of all those machine types,
so will already have been called beforehand by the QOM infrastructure.
Remove the redundant calls.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
hw/ppc/spapr.c has a number of definitions related to the various versioned
machine types ("pseries-2.1" .. "pseries-2.5") it defines. These are
mostly arranged by type of function first, then machine version second, and
it's not consistent about whether it goes in increasing or decreasing
version order.
This rearranges the code to keep all the definitions for a particular
machine version together, and arrange then consistently in order most
recent to least recent.
This brings us closer to matching the way PC does things, and makes later
cleanups easier to follow.
Apart from adding some comments marking each section, this is a pure
mechanical rearrangement with no semantic changes.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
98cec76 "machine: Set MachineClass::name automatically" removed the setting
of mc->name for the pseries machine types, since it can be derived
automatically from the type names constructed with MACHINE_TYPE_NAME().
Unfortunately fb0fc8f "spapr: Create pseries-2.5 machine" went in later and
brought one of them back.
This removes it again.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
The core VFIO infrastructure more or less allows VFIO devices to work
on any normal guest PCI host bridge (PHB) without extra logic.
However, the "spapr-pci-host-bridge" device (as opposed to the special
"spapr-pci-vfio-host-bridge" device) breaks this by using a partially
KVM accelerated implementation of the guest kernel IOMMU which won't
work with VFIO devices, without additional kernel support.
This patch allows VFIO devices to work on the spapr-pci-host-bridge,
by having it switch off KVM TCE acceleration when a VFIO device is
added to the PHB (either on startup, or by hotplug).
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Because of the way non-VFIO guest IOMMU operations are KVM accelerated, not
all TCE tables (guest IOMMU contexts) can support VFIO devices. Currently,
this is decided at creation time.
To support hotplug of VFIO devices, we need to allow a TCE table which
previously didn't allow VFIO devices to be switched so that it can. This
patch adds an spapr_tce_set_need_vfio() function to do this, by
reallocating the table in userspace if necessary.
Currently this doesn't allow the KVM acceleration to be re-enabled if all
the VFIO devices are removed. That's an optimization for another time.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
The vfio_accel parameter used when creating a new TCE table (guest IOMMU
context) has a confusing name. What it really means is whether we need the
TCE table created to be able to support VFIO devices.
VFIO is relevant, because when available we use in-kernel acceleration of
the TCE table, but that may not work with VFIO devices because updates to
the table are handled in kernel, bypass qemu and so don't hit qemu's
infrastructure for keeping the VFIO host IOMMU state in sync with the guest
IOMMU state.
Rename the parameter to "need_vfio" throughout. This is a cosmetic change,
with no impact on the logic.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
At present the PCI host bridge (PHB) for the pseries machine type has a
fixed DMA window from 0..1GB (in PCI address space) which is mapped to real
memory via the PAPR paravirtualized IOMMU.
For better support of VFIO devices, we're going to want to allow for
different configurations of the DMA window.
Eventually we'll want to allow the guest itself to reconfigure the window
via the PAPR dynamic DMA window interface, but as a preliminary this patch
allows the user to reconfigure the window with new properties on the PHB
device.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
ed173cb ".travis.yml: remove "make check" from main matrix" stopped running
make check for all the Travis build targets for various reasons. It
continued to run make check on one Travis build, which builds for a big
list of all (? nearly all) our supported softmmu targets.
Unfortunately, due to a spacing / quoting error it only actually builds for
the alpha, arm, aarch64 and cris targets. Specifically, the list of
targets is split over several lines. Even with YAML folding, this will
leave spaces in the list, meaning $TARGETS won't have the value we need.
I had a look at the YAML spec and I couldn't quickly see a way of splitting
the list so that it doesn't end up with spaces, so this patch fixes the
problem by putting the whole list on one huge line.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
At present the memory listener used by vfio to keep host IOMMU mappings
in sync with the guest memory image assumes that if a guest IOMMU
appears, then it has no existing mappings.
This may not be true if a VFIO device is hotplugged onto a guest bus
which didn't previously include a VFIO device, and which has existing
guest IOMMU mappings.
Therefore, use the memory_region_register_iommu_notifier_replay()
function in order to fix this case, replaying existing guest IOMMU
mappings, bringing the host IOMMU into sync with the guest IOMMU.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
When we have guest visible IOMMUs, we allow notifiers to be registered
which will be informed of all changes to IOMMU mappings. This is used by
vfio to keep the host IOMMU mappings in sync with guest IOMMU mappings.
However, unlike with a memory region listener, an iommu notifier won't be
told about any mappings which already exist in the (guest) IOMMU at the
time it is registered. This can cause problems if hotplugging a VFIO
device onto a guest bus which had existing guest IOMMU mappings, but didn't
previously have an VFIO devices (and hence no host IOMMU mappings).
This adds a memory_region_iommu_replay() function to handle this case. It
replays any existing mappings in an IOMMU memory region to a specified
notifier. Because the IOMMU memory region doesn't internally remember the
granularity of the guest IOMMU it has a small hack where the caller must
specify a granularity at which to replay mappings.
If there are finer mappings in the guest IOMMU these will be reported in
the iotlb structures passed to the notifier which it must handle (probably
causing it to flag an error). This isn't new - the VFIO iommu notifier
must already handle notifications about guest IOMMU mappings too short
for it to represent in the host IOMMU.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Depending on the host IOMMU type we determine and record the available page
sizes for IOMMU translation. We'll need this for other validation in
future patches.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
The current vfio core code assumes that the host IOMMU is capable of
mapping any IOVA the guest wants to use to where we need. However, real
IOMMUs generally only support translating a certain range of IOVAs (the
"DMA window") not a full 64-bit address space.
The common x86 IOMMUs support a wide enough range that guests are very
unlikely to go beyond it in practice, however the IOMMU used on IBM Power
machines - in the default configuration - supports only a much more limited
IOVA range, usually 0..2GiB.
If the guest attempts to set up an IOVA range that the host IOMMU can't
map, qemu won't report an error until it actually attempts to map a bad
IOVA. If guest RAM is being mapped directly into the IOMMU (i.e. no guest
visible IOMMU) then this will show up very quickly. If there is a guest
visible IOMMU, however, the problem might not show up until much later when
the guest actually attempt to DMA with an IOVA the host can't handle.
This patch adds a test so that we will detect earlier if the guest is
attempting to use IOVA ranges that the host IOMMU won't be able to deal
with.
For now, we assume that "Type1" (x86) IOMMUs can support any IOVA, this is
incorrect, but no worse than what we have already. We can't do better for
now because the Type1 kernel interface doesn't tell us what IOVA range the
IOMMU actually supports.
For the Power "sPAPR TCE" IOMMU, however, we can retrieve the supported
IOVA range and validate guest IOVA ranges against it, and this patch does
so.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
If a DMA mapping operation fails in vfio_listener_region_add() it
checks to see if we've already completed initial setup of the
container. If so it reports an error so the setup code can fail
gracefully, otherwise throws a hw_error().
There are other potential failure cases in vfio_listener_region_add()
which could benefit from the same logic, so move it to its own
fail: block. Later patches can use this to extend other failure cases
to fail as gracefully as possible under the circumstances.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Currently the VFIOContainer iommu_data field contains a union with
different information for different host iommu types. However:
* It only actually contains information for the x86-like "Type1" iommu
* Because we have a common listener the Type1 fields are actually used
on all IOMMU types, including the SPAPR TCE type as well
In fact we now have a general structure for the listener which is unlikely
to ever need per-iommu-type information, so this patch removes the union.
In a similar way we can unify the setup of the vfio memory listener in
vfio_connect_container() that is currently split across a switch on iommu
type, but is effectively the same in both cases.
The iommu_data.release pointer was only needed as a cleanup function
which would handle potentially different data in the union. With the
union gone, it too can be removed.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
At present, if guest numa nodes are requested, but the cpus in each node
are not specified, spapr just uses the default behaviour or assigning each
vcpu round-robin to nodes.
If smp_threads != 1, that will assign adjacent threads in a core to
different NUMA nodes. As well as being just weird, that's a configuration
that can't be represented in the device tree we give to the guest, which
means the guest and qemu end up with different ideas of the NUMA topology.
This patch implements mc->cpu_index_to_socket_id in the spapr code to
make sure vcpus get assigned to nodes only at the socket granularity.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Enable memory hotplug for pseries 2.4 and add LMB DR connectors.
With memory hotplug, enforce RAM size, NUMA node memory size and maxmem
to be a multiple of SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the
granularity in which LMBs are represented and hot-added.
LMB DR connectors will be used by the memory hotplug code.
Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
[spapr_drc_reset implementation]
[since this missed the 2.4 cutoff, changing to only enable for 2.5]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The dynamic reconfiguration (hotplug) code for the pseries machine type
uses a "DR connector" QOM object for each resource it will be possible
to hotplug. Each of these is added to its owner using
object_property_add_child(owner, "dr-connector[*], ...);
That works ok, mostly, but it means that the property indices are
arbitrary, depending on the order in which the connectors are constructed.
That might line up to something useful, but it doesn't have to.
It will get worse once we add hotplug RAM support. That will add a DR
connector object for every 256MB of potential memory. So if maxmem=2T,
for example, there are 8192 objects under the same parent.
The QOM interfaces aren't really designed for this. In particular
object_property_add() with [*] has O(n^2) time complexity (in the number of
existing children): first it has a linear search through array indices to
find a free slot, each of which is attempted to a recursive call to
object_property_add() with a specific [N]. Those calls are O(n) because
there's a linear search through all properties to check for duplicates.
By using a meaningful index value, which we already know is unique we can
avoid the [*] special behaviour. That lets us reduce the total time for
creating the DR objects from O(n^3) to O(n^2).
O(n^2) is still kind of crappy, but it's enough to reduce the startup time
of qemu (with in-progress memory hotplug support) with maxmem=2T from ~20
minutes to ~4 seconds.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Cc: Bharata B Rao <bharata@linux.vnet.ibm.com>
Tested-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
The device tree presented to pseries machine type guests includes an
ibm,chip-id property which gives essentially the socket number of each
vcpu core (individual vcpu threads don't get a node in the device
tree).
To calculate this, it uses a vcpus_per_socket variable computed as
(smp_cpus / #sockets). This is correct for the usual case where
smp_cpus == smp_threads * smp_cores * #sockets.
However, you can start QEMU with the number of cores and threads
mismatching the total number of vcpus (whether that _should_ be
permitted is a topic for another day). It's a bit hard to say what
the "real" number of vcpus per socket here is, but for most purposes
(smp_threads * smp_cores) will more meaningfully match how QEMU
behaves with respect to socket boundaries.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
The DRC_INDEX_ID_MASK macro does a left shift on ~0, which is a signed
quantity, and therefore undefined behaviour according to the C spec. In
particular this causes warnings from the clang sanitizer.
This fixes it by calculating the same mask without using ~0 (I think the
new method is a more common idiom for generating masks anyway). For good
measure I also use 1ULL to force the expression's type to unsigned long
long, which should be good for assigning to anything we're going to want
to.
Reported-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
There's a call to object_dynamic_cast() in spapr_vty which uses the type
name "spapr-vty" directly, instead of the usual idiom of using the #defined
TYPE_VIO_SPAPR_VTY_DEVICE. Fix it.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
If a guest passes the reg property of a valid VIO object that is not a VTY
to either H_GET_TERM_CHAR or H_PUT_TERM_CHAR, QEMU hits a dynamic cast
assertion and aborts.
PAPR+ says "Hypervisor checks the termno parameter for validity against the
Vterm IOA unit addresses assigned to the partition, else return H_Parameter."
This patch adds a type check to ensure vty_lookup() either returns a pointer
to a valid VTY object or NULL. H_GET_TERM_CHAR and H_PUT_TERM_CHAR will
now return H_PARAMETER to the guest instead of crashing.
The patch has no effect on the reg == 0 hack used to implement the RTAS call
display-character.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
Currently although we have an sPAPRMachineState descended from MachineState
we don't have an sPAPRMAchineClass descended from MachineClass. So far it
hasn't been needed, but several upcoming features are going to want it,
so this patch creates a stub implementation.
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
The sPAPRMachineState structure includes an entry_point field containing
the initial PC value for starting the machine, even though this always has
the value 0x100.
I think this is a hangover from very early versions which bypassed the
firmware when using -kernel. In any case it has no function now, so remove
it.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
The ram_limit field was imported from sPAPREnvironment where it predates
the machine's ram size being available generically from machine->ram_size.
Worse, the existing code was inconsistent about where it got the ram size
from. Sometimes it used spapr->ram_limit, sometimes the global 'ram_size'
and sometimes a local 'ram_size' masking the global.
This cleans up the code to consistently use machine->ram_size, eliminating
spapr->ram_limit in the process.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
The code for -machine pseries maintains a global sPAPREnvironment structure
which keeps track of general state information about the guest platform.
This predates the existence of the MachineState structure, but performs
basically the same function.
Now that we have the generic MachineState, fold sPAPREnvironment into
sPAPRMachineState, the pseries specific subclass of MachineState.
This is mostly a matter of search and replace, although a few places which
relied on the global spapr variable are changed to find the structure via
qdev_get_machine().
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
At Alex Graf's request I'm now acting as sub-maintainer for the sPAPR
(-machine pseries) code. This updates MAINTAINERS accordingly.
While we're at it, change the label to mention pseries since that's the
actual name of the machine type, even if most of the C files use the sPAPR
name.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
qemu currently implements the hypercalls H_LOGICAL_CI_LOAD and
H_LOGICAL_CI_STORE as PAPR extensions. These are used by the SLOF firmware
for IO, because performing cache inhibited MMIO accesses with the MMU off
(real mode) is very awkward on POWER.
This approach breaks when SLOF needs to access IO devices implemented
within KVM instead of in qemu. The simplest example would be virtio-blk
using an iothread, because the iothread / dataplane mechanism relies on
an in-kernel implementation of the virtio queue notification MMIO.
To fix this, an in-kernel implementation of these hypercalls has been made,
(kernel commit 99342cf "kvmppc: Implement H_LOGICAL_CI_{LOAD,STORE} in KVM"
however, the hypercalls still need to be enabled from qemu. This performs
the necessary calls to do so.
It would be nice to provide some warning if we encounter a problematic
device with a kernel which doesn't support the new calls. Unfortunately,
I can't see a way to detect this case which won't either warn in far too
many cases that will probably work, or which is horribly invasive.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Now that 2.4 development has opened, create a new pseries machine type
variant. For now it is identical to the pseries-2.3 machine type, but
a number of new features are coming that will need to set backwards
compatibility options.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
spapr_pci.c contains a number of expressions of the form (uval == -1) or
(uval != -1), where 'uval' is an unsigned value.
This mostly works in practice, because as long as the width of uval is
greater or equal than that of (int), the -1 will be promoted to the
unsigned type, which is the expected outcome.
However, at least for the cases where uval is uint32_t, this would break
on platforms where sizeof(int) > 4 (and a few such do exist), because then
the uint32_t value would be promoted to the larger int type, and never be
equal to -1.
This patch fixes these errors. The fixes for the (uint32_t) cases are
necessary as described above. I've made similar fixes to (uint64_t) and
(hwaddr) cases. Those are strictly theoretical, since I don't know of any
platforms where sizeof(int) > 8, but hey, it's not that hard so we might
as well be strictly C standard compliant.
Reported-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
If the guest programs a sufficiently large timeout value an integer
overflow can occur in i6300esb_restart_timer(). e.g. if the maximum
possible timer preload value of 0xfffff is programmed then we end up with
the calculation:
timeout = get_ticks_per_sec() * (0xfffff << 15) / 33000000;
get_ticks_per_sec() returns 1000000000 (10^9) giving:
10^9 * (0xfffff * 2^15) == 0x1dcd632329b000000 (65 bits)
Obviously the division by 33MHz brings it back under 64-bits, but the
overflow has already occurred.
Since signed integer overflow has undefined behaviour in C, in theory this
could be arbitrarily bad. In practice, the overflowed value wraps around
to something negative, causing the watchdog to immediately expire, killing
the guest, which is still fairly bad.
The bug can be triggered by running a Linux guest, loading the i6300esb
driver with parameter "heartbeat=2046" and opening /dev/watchdog. The
watchdog will trigger as soon as the device is opened.
This patch corrects the problem by using muldiv64(), which effectively
allows a 128-bit intermediate value between the multiplication and
division.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Message-Id: <1427075508-12099-3-git-send-email-david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The IO operations for the i6300esb watchdog timer are marked as
DEVICE_NATIVE_ENDIAN. This is not correct, and - as a PCI device - should
be DEVICE_LITTLE_ENDIAN.
This allows i6300esb to work on ppc targets (yes, using an Intel ICH
derived device on ppc is a bit odd, but the driver exists on the guest
and there's no more obviously suitable watchdog device).
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Message-Id: <1427075508-12099-2-git-send-email-david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
When the guest switches the interrupt endian mode, which essentially
means a global machine endian switch, we want to change the VGA
framebuffer endian mode as well in order to be backward compatible
with existing guests who don't know about the new endian control
register.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
The VGA device model now supports having the framebuffer in either endian,
and can be switched between these by the guest via a register in the qext
region.
However, in some cases (e.g. LE OS on the pseries machine) we have
existing guest that don't know about the endian switch register, but other
parts of the qemu code have better information to set a default endianness
than the VGA code does of itself.
In order to allow them to set a correct default endianness in these cases,
without breaking abstraction walls, this patch exposes the VGA framebuffer
endianness via a writable QOM property.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Gerd Hoffmann <kraxel@redhat.com>
[agraf: use instance_init for property exposure]
Signed-off-by: Alexander Graf <agraf@suse.de>
On x86, the guest's RTC can be read with QMP, either from the RTC device's
"date" property or via the "rtc-time" property on the machine (which is an
alias to the former). This is set up in the mc146818rtc driver, and
doesn't work on other targets.
This patch adds a similar "date" property to the pseries machine's RTAS RTC
and adds a compatible alias to the machine.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
The initial creation of the PAPR RTC qdev class left a wart - the rtc's
offset was left in the sPAPREnvironment structure, accessed via a global.
This patch moves it into the RTC device's own state structure, were it
belongs. This requires a small change to the migration stream format. In
order to handle incoming streams from older versions, we also need to
retain the rtc_offset field in the sPAPREnvironment structure, so that it
can be loaded into via the vmsd, then pushed into the RTC device.
Since we're changing the migration format, this also takes the opportunity
to:
* Change the rtc offset from a value in seconds to a value in
nanoseconds, allowing nanosecond offsets between host and guest
rtc time, if desired.
* Remove both the already unused "next_irq" field and now unused
"rtc_offset" field from the new version of the spapr migration
stream
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
At present the PAPR RTC isn't a "device" as such - it's accessed only via
firmware/hypervisor calls, and is handled in the sPAPR core code. This
becomes inconvenient as we extend it in various ways.
This patch makes the PAPR RTC a separate device in the qemu device model.
For now, the only piece of device state - the rtc_offset - is still kept in
the global sPAPREnvironment structure. That's clearly wrong, but leaving
it to be fixed in a following patch makes for a clearer separation between
the internal re-organization of the device, and the behavioural changes
(because the migration stream format needs to change slightly when the
offset is moved into the device's own state).
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
In the 'pseries' machine the real time clock is provided by a
paravirtualized firmware interface rather than a device per se; the RTAS
get-time-of-day and set-time-of-day calls.
Out current implementations of those work directly off host time (with
an offset), not respecting options such as clock=vm which can be
specified in the -rtc command line option.
This patch reworks the RTAS RTC code to respect those options, primarily
by basing them on the qemu_clock_get_ns(rtc_clock) function instead of
directly on qemu_get_timedate() (which essentially handles host time, not
virtual rtc time).
As a bonus, this means our get-time-of-day function now also returns
nanoseconds.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
The virtual RTC time is used in two places in the pseries machine. First
is in the RTAS get-time-of-day function which returns the RTC time to the
guest. Second is in the spapr events code which is used to timestamp
event messages from the hypervisor to the guest.
Currently both call qemu_get_timedate() directly, but we want to change
that so we can properly handle the various -rtc options. In preparation,
create a helper function to return the virtual RTC time.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
Currently, the RTAS time of day functions only partially validate the
number of parameters they receive and return. Because of how the
parameters are used, this is unlikely to lead to a crash, but it's messy.
This patch adds the missing checks.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
At the moment the RTAS (firmware/hypervisor) time of day functions are
implemented in spapr_rtas.c along with a bunch of other things. Since
we're going to be expanding these a bit, move the RTAS RTC related code
out into new file spapr_rtc.c. Also add its own initialization function,
spapr_rtc_init() called from the main machine init routine.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
Currently, vmstate.h includes helper macro variants for 8, 16 and 32-bit
unsigned integers which include a "test" function which can selectively
enable or disable the field's presence in the migration stream.
There aren't similar helpers for 64-bit unsigned integers, or any size of
signed integers. This patch remedies this.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
The mc146818rtc driver exposes the current RTC date and time via the "date"
property in QOM (which is also aliased to the machine's "rtc-time"
property). Currently it uses a custom visitor function rtc_get_date to
do this.
This patch introduces new helpers to the QOM core to expose struct tm
valued properties via a getter function, so that this functionality can be
more easily duplicated in other RTC implementations.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
pseries guests can have large numbers of PCI host bridges. To avoid the
user having to specify a number of different configuration values for every
one, the device supports an "index" property which is a shorthand setting
the various window and configuration addresses from a predefined sensible
set.
There are some problems with the details at present:
* The "index" propery is signed, but negative values will create PCI
windows below where we expect, potentially colliding with other devices
* No limit is imposed on the "index" property and large values can
translate to extremely large window addresses. With PCI passthrough in
particular this can mean we exceed various mapping and physical address
limits causing the guest host bridge to not work in strange ways.
This patch addresses this, by making "index" unsigned, and imposing a
limit. Currently the limit allows indices from 0..255 which is probably
enough host bridges for the time being. It's fairly easy to extend if
we discover we need more.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Currently the "platform-bus" device is included for all softmmu builds.
This bridge is intended for use on any platforms that require dynamic
creation of sysbus devices. However, at present it is used only for the
PPC E500 target, with plans for the ARM "virt" target in the immediate
future.
To avoid a not-very-useful entry appearing in "qemu -device ?" output on
other targets, this patch makes a specific config option for platform-bus
and enables it (for now) only on ppc configurations which include E500
and on ARM (which always includes the "virt" target).
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Message-Id: <1425017077-18487-3-git-send-email-david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The i82801b11, ioh3420 and xio3130 PCI Express devices are currently
included in the build unconditionally.
While they could theoretically appear on any target platform with PCI-E,
they're pretty unlikely to appear on platforms that aren't Intel derived.
Therefore, to avoid presenting unlikely-to-be-relevant devices to the user,
add config options to enable these components, and enable them by default
only on x86 and arm platforms.
(Note that this patch does include these for aarch64, via its inclusion of
arm-softmmu.mak).
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Message-Id: <1425017077-18487-2-git-send-email-david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
A number of ARM embedded boards include EHCI USB host controllers which
appear as directly mapped devices, rather than sitting on a PCI bus.
At present code to emulate such devices is included whenever EHCI support
is included. This patch adjusts teh config options to only include them
in builds targetting ARM by default.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
spapr_tce_table_finalize() can SEGV if the object was not previously
realized. In particular this can be triggered by running
qemu-system-ppc -device spapr-tce-table,?
The basic problem is that we have mismatched initialization versus
finalization: spapr_tce_table_finalize() is attempting to undo things that
are done in spapr_tce_table_realize(), not an instance_init function.
Therefore, replace spapr_tce_table_finalize() with
spapr_tce_table_unrealize().
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Cc: qemu-stable@nongnu.org
Signed-off-by: Alexander Graf <agraf@suse.de>
The 'config' field in the VirtIOSerial structure keeps a copy of the virtio
console's config space as visible to the guest, that is to say, in guest
endianness. This is fiddly to maintain, because on some targets, such as
powerpc, the "guest endianness" can change when a new guest OS boots.
In fact, there's no need to maintain such a guest view of config space -
instead we can reconstruct it from host-format data when it is accessed
with get_config.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
A number of places in the virtio_serial driver retrieve the number of ports
from vser->config.max_nr_ports, which is guest-endian. But for internal
users, we already have a host-endian copy of the number of ports in
vser->serial.max_virtserial_ports. Using that instead of the config field
removes the need for easy-to-forget byteswapping.
In particular this fixes a bug on incoming migration, where we don't adjust
the endianness vser->config correctly, because it hasn't yet been loaded
from the migration stream when virtio_serial_load_device() is called.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
David Gibson