Recent changes allowed the pSeries machine to rollback the hotunplug
process for the DIMM when the guest kernel signals, via a
reconfiguration of the DR connector, that it's not going to release the
LMBs.
Let's also warn QAPI listerners about it. One place to do it would be
right after the unplug state is cleaned up,
spapr_clear_pending_dimm_unplug_state(). This would mean that the
function is now doing more than cleaning up the pending dimm state
though.
This patch does the following changes in spapr.c:
- send a QAPI event to inform that we experienced a failure in the
hotunplug of the DIMM;
- rename spapr_clear_pending_dimm_unplug_state() to
spapr_memory_unplug_rollback(). This is a better fit for what the
function is now doing, and it makes callers care more about what the
function goal is and less about spapr.c internals such as clearing
the pending dimm unplug state.
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210302141019.153729-3-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Handling errors in memory hotunplug in the pSeries machine is more
complex than any other device type, because there are all the
complications that other devices has, and more.
For instance, determining a timeout for a DIMM hotunplug must consider
if it's a Hash-MMU or a Radix-MMU guest, because Hash guests takes
longer to hotunplug DIMMs. The size of the DIMM is also a factor, given
that longer DIMMs naturally takes longer to be hotunplugged from the
kernel. And there's also the guest memory usage to be considered: if
there's a process that is consuming memory that would be lost by the
DIMM unplug, the kernel will postpone the unplug process until the
process finishes, and then initiate the regular hotunplug process. The
first two considerations are manageable, but the last one is a deal
breaker.
There is no sane way for the pSeries machine to determine the memory
load in the guest when attempting a DIMM hotunplug - and even if there
was a way, the guest can start using all the RAM in the middle of the
unplug process and invalidate our previous assumptions - and in result
we can't even begin to calculate a timeout for the operation. This means
that we can't implement a viable timeout mechanism for memory unplug in
pSeries.
Going back to why we would consider an unplug timeout, the reason is
that we can't know if the kernel is giving up the unplug. Turns out
that, sometimes, we can. Consider a failed memory hotunplug attempt
where the kernel will error out with the following message:
'pseries-hotplug-mem: Memory indexed-count-remove failed, adding any
removed LMBs'
This happens when there is a LMB that the kernel gave up in removing,
and the LMBs previously marked for removal are now being added back.
This happens in the pseries kernel in [1], dlpar_memory_remove_by_ic()
into dlpar_add_lmb(), and after that update_lmb_associativity_index().
In this function, the kernel is configuring the LMB DRC connector again.
Note that this is a valid usage in LOPAR, as stated in section
"ibm,configure-connector RTAS Call":
'A subsequent sequence of calls to ibm,configure-connector with the same
entry from the “ibm,drc-indexes” or “ibm,drc-info” property will restart
the configuration of devices which were not completely configured.'
We can use this kernel behavior in our favor. If a DRC connector
reconfiguration for a LMB that we marked as unplug pending happens, this
indicates that the kernel changed its mind about the unplug and is
reasserting that it will keep using all the LMBs of the DIMM. In this
case, it's safe to assume that the whole DIMM device unplug was
cancelled.
This patch hops into rtas_ibm_configure_connector() and, in the scenario
described above, clear the unplug state for the DIMM device. This will
not solve all the problems we still have with memory unplug, but it will
cover this case where the kernel reconfigures LMBs after a failed
unplug. We are a bit more resilient, without using an unreliable
timeout, and we didn't make the remaining error cases any worse.
[1] arch/powerpc/platforms/pseries/hotplug-memory.c
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210222194531.62717-6-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This function is used only in spapr_numa.c.
Tested-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210128174213.1349181-2-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The function is called only inside spapr_hcall.c.
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210114180628.1675603-3-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This whitespace was messing with lots of diffs if you happen
to use an editor that eliminates trailing whitespaces on file
save.
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210114180628.1675603-2-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Since commit 1e8b5b1aa1 ("spapr: Allow memory unplug to always succeed")
trying to unplug memory from a guest that doesn't support it (eg. rhel6)
no longer generates an error like it used to. Instead, it leaves the
memory around : only a subsequent reboot or manual use of drmgr within
the guest can complete the hot-unplug sequence. A flag was added to
SpaprMachineClass so that this new behavior only applies to the default
machine type.
We can do better. CAS processes all pending hot-unplug requests. This
means that we don't really care about what the guest supports if
the hot-unplug request happens before CAS.
All guests that we care for, even old ones, set enough bits in OV5
that lead to a non-empty bitmap in spapr->ov5_cas. Use that as a
heuristic to decide if CAS has already occured or not.
Always accept unplug requests that happen before CAS since CAS will
process them. Restore the previous behavior of rejecting them after
CAS when we know that the guest doesn't support memory hot-unplug.
This behavior is suitable for all machine types : this allows to
drop the pre_6_0_memory_unplug flag.
Fixes: 1e8b5b1aa1 ("spapr: Allow memory unplug to always succeed")
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <161012708715.801107.11418801796987916516.stgit@bahia.lan>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Running a guest with 128 NUMA nodes crashes QEMU:
../../util/error.c:59: error_setv: Assertion `*errp == NULL' failed.
The crash happens when setting the FWNMI migration blocker:
2861 if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2862 /* Create the error string for live migration blocker */
2863 error_setg(&spapr->fwnmi_migration_blocker,
2864 "A machine check is being handled during migration. The handler"
2865 "may run and log hardware error on the destination");
2866 }
Inspection reveals that papr->fwnmi_migration_blocker isn't NULL:
(gdb) p spapr->fwnmi_migration_blocker
$1 = (Error *) 0x8000000004000000
Since this is the only place where papr->fwnmi_migration_blocker is
set, this means someone wrote there in our back. Further analysis
points to spapr_numa_associativity_init(), especially the part
that initializes the associative arrays for NVLink GPUs:
max_nodes_with_gpus = nb_numa_nodes + NVGPU_MAX_NUM;
ie. max_nodes_with_gpus = 128 + 6, but the array isn't sized to
accommodate the 6 extra nodes:
struct SpaprMachineState {
.
.
.
uint32_t numa_assoc_array[MAX_NODES][NUMA_ASSOC_SIZE];
Error *fwnmi_migration_blocker;
};
and the following loops happily overwrite spapr->fwnmi_migration_blocker,
and probably more:
for (i = nb_numa_nodes; i < max_nodes_with_gpus; i++) {
spapr->numa_assoc_array[i][0] = cpu_to_be32(MAX_DISTANCE_REF_POINTS);
for (j = 1; j < MAX_DISTANCE_REF_POINTS; j++) {
uint32_t gpu_assoc = smc->pre_5_1_assoc_refpoints ?
SPAPR_GPU_NUMA_ID : cpu_to_be32(i);
spapr->numa_assoc_array[i][j] = gpu_assoc;
}
spapr->numa_assoc_array[i][MAX_DISTANCE_REF_POINTS] = cpu_to_be32(i);
}
Fix the size of the array. This requires "hw/ppc/spapr.h" to see
NVGPU_MAX_NUM. Including "hw/pci-host/spapr.h" introduces a
circular dependency that breaks the build, so this moves the
definition of NVGPU_MAX_NUM to "hw/ppc/spapr.h" instead.
Reported-by: Min Deng <mdeng@redhat.com>
BugLink: https://bugzilla.redhat.com/show_bug.cgi?id=1908693
Fixes: dd7e1d7ae4 ("spapr_numa: move NVLink2 associativity handling to spapr_numa.c")
Cc: danielhb413@gmail.com
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <160829960428.734871.12634150161215429514.stgit@bahia.lan>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
It is currently impossible to hot-unplug a memory device between
machine reset and CAS.
(qemu) device_del dimm1
Error: Memory hot unplug not supported for this guest
This limitation was introduced in order to provide an explicit
error path for older guests that didn't support hot-plug event
sources (and thus memory hot-unplug).
The linux kernel has been supporting these since 4.11. All recent
enough guests are thus capable of handling the removal of a memory
device at all time, including during early boot.
Lift the limitation for the latest machine type. This means that
trying to unplug memory from a guest that doesn't support it will
likely just do nothing and the memory will only get removed at
next reboot. Such older guests can still get the existing behavior
by using an older machine type.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <160794035064.23292.17560963281911312439.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This allows to drop a user of qdev_get_machine().
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <20201209170052.1431440-4-groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Read documentation in "qapi/error.h" and changelog of commit
e3fe3988d7 ("error: Document Error API usage rules") for
rationale.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <20201120234208.683521-7-groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
spapr_reallocate_hpt() has three users, two of which pass &error_fatal
and the third one, htab_load(), passes &local_err, uses it to detect
failures and simply propagates -EINVAL up to vmstate_load(), which will
cause QEMU to exit. It is thus confusing that spapr_reallocate_hpt()
doesn't return right away when an error is detected in some cases. Also,
the comment suggesting that the caller is welcome to try to carry on
seems like a remnant in this respect.
This can be improved:
- change spapr_reallocate_hpt() to always report a negative errno on
failure, either as reported by KVM or -ENOSPC if the HPT is smaller
than what was asked,
- use that to detect failures in htab_load() which is preferred over
checking &local_err,
- propagate this negative errno to vmstate_load() because it is more
accurate than propagating -EINVAL for all possible errors.
[dwg: Fix compile error due to omitted prelim patch]
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <160371605460.305923.5890143959901241157.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The changes to come to NUMA support are all guest visible. In
theory we could just create a new 5_1 class option flag to
avoid the changes to cascade to 5.1 and under. The reality is that
these changes are only relevant if the machine has more than one
NUMA node. There is no need to change guest behavior that has
been around for years needlesly.
This new helper will be used by the next patches to determine
whether we should retain the (soon to be) legacy NUMA behavior
in the pSeries machine. The new behavior will only be exposed
if:
- machine is pseries-5.2 and newer;
- more than one NUMA node is declared in NUMA state.
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20201007172849.302240-2-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
As recommended in "qapi/error.h", return true on success and false on
failure. This allows to reduce error propagation overhead in the callers.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <20200914123505.612812-14-groug@kaod.org>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
As recommended in "qapi/error.h", return true on success and false on
failure. This allows to reduce error propagation overhead in the callers.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <20200914123505.612812-11-groug@kaod.org>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Some typedefs and macros are defined after the type check macros.
This makes it difficult to automatically replace their
definitions with OBJECT_DECLARE_TYPE.
Patch generated using:
$ ./scripts/codeconverter/converter.py -i \
--pattern=QOMStructTypedefSplit $(git grep -l '' -- '*.[ch]')
which will split "typdef struct { ... } TypedefName"
declarations.
Followed by:
$ ./scripts/codeconverter/converter.py -i --pattern=MoveSymbols \
$(git grep -l '' -- '*.[ch]')
which will:
- move the typedefs and #defines above the type check macros
- add missing #include "qom/object.h" lines if necessary
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Message-Id: <20200831210740.126168-9-ehabkost@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Message-Id: <20200831210740.126168-10-ehabkost@redhat.com>
Message-Id: <20200831210740.126168-11-ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
The work to be done in h_home_node_associativity() intersects
with what is already done in spapr_numa_fixup_cpu_dt(). This
patch creates a new helper, spapr_numa_get_vcpu_assoc(), to
be used for both spapr_numa_fixup_cpu_dt() and
h_home_node_associativity().
While we're at it, use memcpy() instead of loop assignment
to created the returned array.
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20200904172422.617460-3-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The next step to centralize all NUMA/associativity handling in
the spapr machine is to create a 'one stop place' for all
things ibm,associativity.
This patch introduces numa_assoc_array, a 2 dimensional array
that will store all ibm,associativity arrays of all NUMA nodes.
This array is initialized in a new spapr_numa_associativity_init()
function, called in spapr_machine_init(). It is being initialized
with the same values used in other ibm,associativity properties
around spapr files (i.e. all zeros, last value is node_id).
The idea is to remove all hardcoded definitions and FDT writes
of ibm,associativity arrays, doing instead a call to the new
helper spapr_numa_write_associativity_dt() helper, that will
be able to write the DT with the correct values.
We'll start small, handling the trivial cases first. The
remaining instances of ibm,associativity will be handled
next.
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20200903220639.563090-2-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Move the typedef from spapr_irq.h to spapr.h, and use "struct
SpaprMachineState" in the spapr_*.h headers (to avoid circular
header dependencies).
This will make future conversion to OBJECT_DECLARE* easier.
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Tested-By: Roman Bolshakov <r.bolshakov@yadro.com>
Message-Id: <20200825192110.3528606-28-ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
NUMA nodes corresponding to GPU memory currently have the same
affinity/distance as normal memory nodes. Add a third NUMA associativity
reference point enabling us to give GPU nodes more distance.
This is guest visible information, which shouldn't change under a
running guest across migration between different qemu versions, so make
the change effective only in new (pseries > 5.0) machine types.
Before, `numactl -H` output in a guest with 4 GPUs (nodes 2-5):
node distances:
node 0 1 2 3 4 5
0: 10 40 40 40 40 40
1: 40 10 40 40 40 40
2: 40 40 10 40 40 40
3: 40 40 40 10 40 40
4: 40 40 40 40 10 40
5: 40 40 40 40 40 10
After:
node distances:
node 0 1 2 3 4 5
0: 10 40 80 80 80 80
1: 40 10 80 80 80 80
2: 80 80 10 80 80 80
3: 80 80 80 10 80 80
4: 80 80 80 80 10 80
5: 80 80 80 80 80 10
These are the same distances as on the host, mirroring the change made
to host firmware in skiboot commit f845a648b8cb ("numa/associativity:
Add a new level of NUMA for GPU's").
Signed-off-by: Reza Arbab <arbab@linux.ibm.com>
Message-Id: <20200716225655.24289-1-arbab@linux.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
On reboot, all memory that was previously added using object_add and
device_add is placed in this DIMM area.
The new SPAPR_LMB_FLAGS_HOTREMOVABLE flag helps Linux to put this memory in
the correct memory zone, so no unmovable allocations are made there,
allowing the object to be easily hot-removed by device_del and
object_del.
This new flag was accepted in Power Architecture documentation.
Signed-off-by: Leonardo Bras <leobras.c@gmail.com>
Reviewed-by: Bharata B Rao <bharata@linux.ibm.com>
Message-Id: <20200511200201.58537-1-leobras.c@gmail.com>
[dwg: Fixed syntax error spotted by Cédric Le Goater]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Several functions can't fail anymore: ich9_pm_add_properties(),
device_add_bootindex_property(), ppc_compat_add_property(),
spapr_caps_add_properties(), PropertyInfo.create(). Drop their @errp
parameter.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-16-armbru@redhat.com>
The CAS reboot flag is false by default and all the locations that
could set it to true have been dropped. This means that all code
blocks depending on the flag being set is dead code and the other
code blocks should be executed always.
Just do that and drop the now uneeded CAS reboot flag. Fix a
comment on the way to make checkpatch happy.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <158514994893.478799.11772512888322840990.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
At the moment "ibm,client-architecture-support" ("CAS") is implemented
in SLOF and QEMU assists via the custom H_CAS hypercall which copies
an updated flatten device tree (FDT) blob to the SLOF memory which
it then uses to update its internal tree.
When we enable the OpenFirmware client interface in QEMU, we won't need
to copy the FDT to the guest as the client is expected to fetch
the device tree using the client interface.
This moves FDT rebuild out to a separate helper which is going to be
called from the "ibm,client-architecture-support" handler and leaves
writing FDT to the guest in the H_CAS handler.
This should not cause any behavioral change.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Message-Id: <20200310050733.29805-3-aik@ozlabs.ru>
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <158514994229.478799.2178881312094922324.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The FWNMI option must deliver system reset interrupts to their
registered address, and there are a few constraints on the handler
addresses specified in PAPR. Add the system reset address state and
checks.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Message-Id: <20200316142613.121089-4-npiggin@gmail.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviwed-by: Mahesh Salgaonkar <mahesh@linux.ibm.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The option is called "FWNMI", and it involves more than just machine
checks, also machine checks can be delivered without the FWNMI option,
so re-name various things to reflect that.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Message-Id: <20200316142613.121089-3-npiggin@gmail.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
At the moment SLOF reserves space for RTAS and instantiates the RTAS blob
which is 20 bytes binary blob calling an hypercall. The rest of the RTAS
area is a log which SLOF has no idea about but QEMU does.
This moves RTAS sizing to QEMU and this overrides the size from SLOF.
The only remaining problem is that SLOF copies the number of bytes it
reserved (2KB for now) so QEMU needs to reserve at least this much;
SLOF will be fixed separately to check that rtas-size from QEMU is
enough for those 20 bytes for the H_RTAS hcall.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Message-Id: <20200316011841.99970-1-aik@ozlabs.ru>
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
In spapr_machine_init() we clamp the size of the RMA to 16GiB and the
comment saying why doesn't make a whole lot of sense. In fact, this was
done because the real mode handling code elsewhere limited the RMA in TCG
mode to the maximum value configurable in LPCR[RMLS], 16GiB.
But,
* Actually LPCR[RMLS] has been able to encode a 256GiB size for a very
long time, we just didn't implement it properly in the softmmu
* LPCR[RMLS] shouldn't really be relevant anyway, it only was because we
used to abuse the RMOR based translation mode in order to handle the
fact that we're not modelling the hypervisor parts of the cpu
We've now removed those limitations in the modelling so the 16GiB clamp no
longer serves a function. However, we can't just remove the limit
universally: that would break migration to earlier qemu versions, where
the 16GiB RMLS limit still applies, no matter how bad the reasons for it
are.
So, we replace the 16GiB clamp, with a clamp to a limit defined in the
machine type class. We set it to 16 GiB for machine types 4.2 and earlier,
but set it to 0 meaning unlimited for the new 5.0 machine type.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
The Real Mode Area (RMA) is the part of memory which a guest can access
when in real (MMU off) mode. Of course, for a guest under KVM, the MMU
isn't really turned off, it's just in a special translation mode - Virtual
Real Mode Area (VRMA) - which looks like real mode in guest mode.
The mechanics of how this works when using the hash MMU (HPT) put a
constraint on the size of the RMA, which depends on the size of the
HPT. So, the latter part of spapr_setup_hpt_and_vrma() clamps the RMA
we advertise to the guest based on this VRMA limit.
There are several things wrong with this:
1) spapr_setup_hpt_and_vrma() doesn't actually clamp, it takes the minimum
of Node 0 memory size and the VRMA limit. That will *often* work the
same as clamping, but there can be other constraints on RMA size which
supersede Node 0 memory size. We have real bugs caused by this
(currently worked around in the guest kernel)
2) Some callers of spapr_setup_hpt_and_vrma() are in a situation where
we're past the point that we can actually advertise an RMA limit to the
guest
3) But most fundamentally, the VRMA limit depends on host configuration
(page size) which shouldn't be visible to the guest, but this partially
exposes it. This can cause problems with migration in certain edge
cases, although we will mostly get away with it.
In practice, this clamping is almost never applied anyway. With 64kiB
pages and the normal rules for sizing of the HPT, the theoretical VRMA
limit will be 4x(guest memory size) and so never hit. It will hit with
4kiB pages, where it will be (guest memory size)/4. However all mainstream
distro kernels for POWER have used a 64kiB page size for at least 10 years.
So, simply replace this logic with a check that the RMA we've calculated
based only on guest visible configuration will fit within the host implied
VRMA limit. This can break if running HPT guests on a host kernel with
4kiB page size. As noted that's very rare. There also exist several
possible workarounds:
* Change the host kernel to use 64kiB pages
* Use radix MMU (RPT) guests instead of HPT
* Use 64kiB hugepages on the host to back guest memory
* Increase the guest memory size so that the RMA hits one of the fixed
limits before the RMA limit. This is relatively easy on POWER8 which
has a 16GiB limit, harder on POWER9 which has a 1TiB limit.
* Use a guest NUMA configuration which artificially constrains the RMA
within the VRMA limit (the RMA must always fit within Node 0).
Previously, on KVM, we also temporarily reduced the rma_size to 256M so
that the we'd load the kernel and initrd safely, regardless of the VRMA
limit. This was a) confusing, b) could significantly limit the size of
images we could load and c) introduced a behavioural difference between
KVM and TCG. So we remove that as well.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Greg Kurz <groug@kaod.org>
If a hot plug or unplug request is pending at CAS, we currently trigger
a CAS reboot, which severely increases the guest boot time. This is
because SLOF doesn't handle hot plug events and we had no way to fix
the FDT that gets presented to the guest.
We can do better thanks to recent changes in QEMU and SLOF:
- we now return a full FDT to SLOF during CAS
- SLOF was fixed to correctly detect any device that was either added or
removed since boot time and to update its internal DT accordingly.
The right solution is to process all pending hot plug/unplug requests
during CAS: convert hot plugged devices to cold plugged devices and
remove the hot unplugged ones, which is exactly what spapr_drc_reset()
does. Also clear all hot plug events that are currently queued since
they're no longer relevant.
Note that SLOF cannot currently populate hot plugged PCI bridges or PHBs
at CAS. Until this limitation is lifted, SLOF will reset the machine when
this scenario occurs : this will allow the FDT to be fully processed when
SLOF is started again (ie. the same effect as the CAS reboot that would
occur anyway without this patch).
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <158257222352.4102917.8984214333937947307.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This allows moving the kernel in the guest memory. The option is useful
for step debugging (as Linux is linked at 0x0); it also allows loading
grub which is normally linked to run at 0x20000.
This uses the existing kernel address by default.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Message-Id: <20200203032943.121178-6-aik@ozlabs.ru>
Reviewed-by: Fabiano Rosas <farosas@linux.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This patch implements few of the necessary hcalls for the nvdimm support.
PAPR semantics is such that each NVDIMM device is comprising of multiple
SCM(Storage Class Memory) blocks. The guest requests the hypervisor to
bind each of the SCM blocks of the NVDIMM device using hcalls. There can
be SCM block unbind requests in case of driver errors or unplug(not
supported now) use cases. The NVDIMM label read/writes are done through
hcalls.
Since each virtual NVDIMM device is divided into multiple SCM blocks,
the bind, unbind, and queries using hcalls on those blocks can come
independently. This doesn't fit well into the qemu device semantics,
where the map/unmap are done at the (whole)device/object level granularity.
The patch doesnt actually bind/unbind on hcalls but let it happen at the
device_add/del phase itself instead.
The guest kernel makes bind/unbind requests for the virtual NVDIMM device
at the region level granularity. Without interleaving, each virtual NVDIMM
device is presented as a separate guest physical address range. So, there
is no way a partial bind/unbind request can come for the vNVDIMM in a
hcall for a subset of SCM blocks of a virtual NVDIMM. Hence it is safe to
do bind/unbind everything during the device_add/del.
Signed-off-by: Shivaprasad G Bhat <sbhat@linux.ibm.com>
Message-Id: <158131059899.2897.11515211602702956854.stgit@lep8c.aus.stglabs.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This patch includes migration support for machine check
handling. Especially this patch blocks VM migration
requests until the machine check error handling is
complete as these errors are specific to the source
hardware and is irrelevant on the target hardware.
Signed-off-by: Aravinda Prasad <arawinda.p@gmail.com>
[Do not set FWNMI cap in post_load, now its done in .apply hook]
Signed-off-by: Ganesh Goudar <ganeshgr@linux.ibm.com>
Message-Id: <20200130184423.20519-7-ganeshgr@linux.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This patch adds support in QEMU to handle "ibm,nmi-register"
and "ibm,nmi-interlock" RTAS calls.
The machine check notification address is saved when the
OS issues "ibm,nmi-register" RTAS call.
This patch also handles the case when multiple processors
experience machine check at or about the same time by
handling "ibm,nmi-interlock" call. In such cases, as per
PAPR, subsequent processors serialize waiting for the first
processor to issue the "ibm,nmi-interlock" call. The second
processor that also received a machine check error waits
till the first processor is done reading the error log.
The first processor issues "ibm,nmi-interlock" call
when the error log is consumed.
Signed-off-by: Aravinda Prasad <arawinda.p@gmail.com>
[Register fwnmi RTAS calls in core_rtas_register_types()
where other RTAS calls are registered]
Signed-off-by: Ganesh Goudar <ganeshgr@linux.ibm.com>
Message-Id: <20200130184423.20519-6-ganeshgr@linux.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Upon a machine check exception (MCE) in a guest address space,
KVM causes a guest exit to enable QEMU to build and pass the
error to the guest in the PAPR defined rtas error log format.
This patch builds the rtas error log, copies it to the rtas_addr
and then invokes the guest registered machine check handler. The
handler in the guest takes suitable action(s) depending on the type
and criticality of the error. For example, if an error is
unrecoverable memory corruption in an application inside the
guest, then the guest kernel sends a SIGBUS to the application.
For recoverable errors, the guest performs recovery actions and
logs the error.
Signed-off-by: Aravinda Prasad <arawinda.p@gmail.com>
[Assume SLOF has allocated enough room for rtas error log]
Signed-off-by: Ganesh Goudar <ganeshgr@linux.ibm.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Message-Id: <20200130184423.20519-5-ganeshgr@linux.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Memory error such as bit flips that cannot be corrected
by hardware are passed on to the kernel for handling.
If the memory address in error belongs to guest then
the guest kernel is responsible for taking suitable action.
Patch [1] enhances KVM to exit guest with exit reason
set to KVM_EXIT_NMI in such cases. This patch handles
KVM_EXIT_NMI exit.
[1] https://www.spinics.net/lists/kvm-ppc/msg12637.html
(e20bbd3d and related commits)
Signed-off-by: Aravinda Prasad <arawinda.p@gmail.com>
Signed-off-by: Ganesh Goudar <ganeshgr@linux.ibm.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Message-Id: <20200130184423.20519-4-ganeshgr@linux.ibm.com>
[dwg: #ifdefs to fix compile for 32-bit target]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Introduce fwnmi an spapr capability and add a helper function
which tries to enable it, which would be used by following patch
of the series. This patch by itself does not change the existing
behavior.
Signed-off-by: Aravinda Prasad <arawinda.p@gmail.com>
[eliminate cap_ppc_fwnmi, add fwnmi cap to migration state
and reprhase the commit message]
Signed-off-by: Ganesh Goudar <ganeshgr@linux.ibm.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Message-Id: <20200130184423.20519-3-ganeshgr@linux.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
spapr_h_cas_compose_response() handles the last piece of the PAPR feature
negotiation process invoked via the ibm,client-architecture-support OF
call. Its only caller is h_client_architecture_support() which handles
most of the rest of that process.
I believe it was placed in a separate file originally to handle some
fiddly dependencies between functions, but mostly it's just confusing
to have the CAS process split into two pieces like this. Now that
compose response is simplified (by just generating the whole device
tree anew), it's cleaner to just fold it into
h_client_architecture_support().
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Cedric Le Goater <clg@fr.ibm.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
For the benefit of peripheral device allocation, the number of available
irqs really wants to be the same on a given machine type version,
regardless of what irq backends we are using. That's the case now, but
only because we make sure the different SpaprIrq instances have the same
value except for the special legacy one.
Since this really only depends on machine type version, move the value to
SpaprMachineClass instead of SpaprIrq. This also puts the code to set it
to the lower value on old machine types right next to setting
legacy_irq_allocation, which needs to go hand in hand.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
spapr now has the mechanism of constructing both XICS and XIVE instances of
the SpaprInterruptController interface. However, only one of the interrupt
controllers will actually be active at any given time, depending on feature
negotiation with the guest. This is handled in the current code via
spapr_irq_current() which checks the OV5 vector from feature negotiation to
determine the current backend.
Determining the active controller at the point we need it like this
can be pretty confusing, because it makes it very non obvious at what
points the active controller can change. This can make it difficult
to reason about the code and where a change of active controller could
appear in sequence with other events.
Make this mechanism more explicit by adding an 'active_intc' pointer
and an explicit spapr_irq_update_active_intc() function to update it
from the CAS state. We also add hooks on the intc backend which will
get called when it is activated or deactivated.
For now we just introduce the switch and hooks, later patches will
actually start using them.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Support for setting VSMT is available in KVM since linux-4.13. Most distros
that support KVM on POWER already have it. It thus seem reasonable enough
to have the default machine to set VSMT to smp_threads.
This brings contiguous VCPU ids and thus brings their upper bound down to
the machine's max_cpus. This is especially useful for XIVE KVM devices,
which may thus allocate only one VP descriptor per VCPU.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <157010411885.246126.12610015369068227139.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
SLOF implements one itself so let's remove it from QEMU. It is one less
image and simpler setup as the RTAS blob never stays in its initial place
anyway as the guest OS always decides where to put it.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Certain old guest versions don't understand the radix MMU introduced with
POWER ISA 3.0, but incorrectly select it if presented with the option at
CAS time. We workaround this in qemu by explicitly excluding the radix
(and other ISA 3.0 linked) options if the guest doesn't explicitly note
support for ISA 3.0.
This is handled by the 'cas_legacy_guest_workaround' flag, which is pretty
vague. Rename it to 'cas_pre_isa3_guest' to be clearer about what it's for.
In addition, we unnecessarily call spapr_populate_pa_features() with
different options when initially constructing the device tree and when
adjusting it at CAS time. At the initial construct time cas_pre_isa3_guest
is already false, so we can still use the flag, rather than explicitly
overriding it to be false at the callsite.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
The pseries guests do not normally allocate PCI resources and rely on
the system firmware doing so. Furthermore at least at some point in
the past the pseries guests won't even allowed to change BARs, probably
it is still the case for phyp. So since the initial commit we have [1]
which prevents resource reallocation.
This is not a problem until we want specific BAR alignments, for example,
PAGE_SIZE==64k to make sure we can still map MMIO BARs directly. For
the boot time devices we handle this in SLOF [2] but since QEMU's RTAS
does not allocate BARs, the guest does this instead and does not align
BARs even if Linux is given pci=resource_alignment=16@pci:0:0 as
PCI_PROBE_ONLY makes Linux ignore alignment requests.
ARM folks added a dial to control PCI_PROBE_ONLY via the device tree [3].
This makes use of the dial to advertise to the guest that we can handle
BAR reassignments. This limits the change to the latest pseries machine
to avoid old guests explosion.
We do not remove the flag from [1] as pseries guests are still supported
under phyp so having that removed may cause problems.
[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/powerpc/platforms/pseries/setup.c?h=v5.1#n773
[2] https://git.qemu.org/?p=SLOF.git;a=blob;f=board-qemu/slof/pci-phb.fs;h=06729bcf77a0d4e900c527adcd9befe2a269f65d;hb=HEAD#l338
[3] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=f81c11af
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Message-Id: <20190719043734.108462-1-aik@ozlabs.ru>
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This has been useful to modify and test the Linux pseries suspend
code but it requires modification to the guest to call it (due to
being gated by other unimplemented features). It is not otherwise
used by Linux yet, but work is slowly progressing there.
This allows a (lightly modified) guest kernel to suspend with
`echo mem > /sys/power/state` and be resumed with system_wakeup
monitor command.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Message-Id: <20190722061752.22114-2-npiggin@gmail.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This implements the H_TPM_COMM hypercall, which is used by an
Ultravisor to pass TPM commands directly to the host's TPM device, or
a TPM Resource Manager associated with the device.
This also introduces a new virtual device, spapr-tpm-proxy, which
is used to configure the host TPM path to be used to service
requests sent by H_TPM_COMM hcalls, for example:
-device spapr-tpm-proxy,id=tpmp0,host-path=/dev/tpmrm0
By default, no spapr-tpm-proxy will be created, and hcalls will return
H_FUNCTION.
The full specification for this hypercall can be found in
docs/specs/ppc-spapr-uv-hcalls.txt
Since SVM-related hcalls like H_TPM_COMM use a reserved range of
0xEF00-0xEF80, we introduce a separate hcall table here to handle
them.
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com
Message-Id: <20190717205842.17827-3-mdroth@linux.vnet.ibm.com>
[dwg: Corrected #include for upstream change]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Implement cpu_exec_enter/exit on ppc which calls into new methods of
the same name in PPCVirtualHypervisorClass. These are used by spapr
to implement the splpar VPA dispatch counter initially.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Message-Id: <20190718034214.14948-2-npiggin@gmail.com>
[dwg: Removed unnecessary CONFIG_USER_ONLY checks as suggested by gkurz]
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
QEMU may crash when running a spapr machine in 'dual' interrupt controller
mode on some older (but not that old, eg. ubuntu 18.04.2) KVMs with partial
XIVE support:
qemu-system-ppc64: hw/ppc/spapr_rtas.c:411: spapr_rtas_register:
Assertion `!name || !rtas_table[token].name' failed.
XICS is controlled by the guest thanks to a set of RTAS calls. Depending
on whether KVM XICS is used or not, the RTAS calls are handled by KVM or
QEMU. In both cases, QEMU needs to expose the RTAS calls to the guest
through the "rtas" node of the device tree.
The spapr_rtas_register() helper takes care of all of that: it adds the
RTAS call token to the "rtas" node and registers a QEMU callback to be
invoked when the guest issues the RTAS call. In the KVM XICS case, QEMU
registers a dummy callback that just prints an error since it isn't
supposed to be invoked, ever.
Historically, the XICS controller was setup during machine init and
released during final teardown. This changed when the 'dual' interrupt
controller mode was added to the spapr machine: in this case we need
to tear the XICS down and set it up again during machine reset. The
crash happens because we indeed have an incompatibility with older
KVMs that forces QEMU to fallback on emulated XICS, which tries to
re-registers the same RTAS calls.
This could be fixed by adding proper rollback that would unregister
RTAS calls on error. But since the emulated RTAS calls in QEMU can
now detect when they are mistakenly called while KVM XICS is in
use, it seems simpler to register them once and for all at machine
init. This fixes the crash and allows to remove some now useless
lines of code.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156044429963.125694.13710679451927268758.stgit@bahia.lab.toulouse-stg.fr.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
