Convert the arch_atomic_xchg define to a C function so that proper
type checking is provided.
Reviewed-by: Juergen Christ <jchrist@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Since gcc 14 flag output operands are supported also for s390.
Provide an arch_try_cmpxchg128() implementation so that all existing
try_cmpxchg128() variants provide slightly better code, if compiled
with gcc 14 or newer.
Reviewed-by: Juergen Christ <jchrist@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Just like x86 and arm64 provide a trivial arch_cmpxchg128_local()
implementation by mapping it to arch_cmpxchg128().
Reviewed-by: Juergen Christ <jchrist@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Get rid of the arch_xchg() inline assemblies by converting the inline
assemblies to C functions which make use of arch_try_cmpxchg().
With flag output operand support the generated code is at least as good as
the previous version. Without it is slightly worse, however getting rid of
all the inline assembly code is worth it.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Since gcc 14 flag output operands are supported also for s390.
Provide an arch_try_cmpxchg() implementation so that all existing
try_cmpxchg() variants provide slightly better code, if compiled
with gcc 14 or newer.
Reviewed-by: Juergen Christ <jchrist@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Rewrite __cmpxchg() in order to get rid of the large inline
assemblies. Convert the one and two byte inline assemblies to
C functions.
The generated code of the new implementation is nearly as good or bad as
the old variant, but easier to read.
Note that the new variants are quite close to the generic cmpxchg_emu_u8()
implementation, however a conversion to the generic variant will not follow
since with mm/vmstat.c there is heavy user of one byte cmpxchg(). A not
inlined variant would have a negative performance impact.
Also note that the calls within __arch_cmpxchg() come with rather pointless
"& 0xff..." operations. They exist only to avoid false positive sparse
warnings like "warning: cast truncates bits from constant value ...".
Reviewed-by: Juergen Christ <jchrist@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Dump tools from s390-tools such as zipl need to know the correct dump area
size of the machine they run on in order to be able to create valid
standalone dumper images. Therefore, allow it to be obtained through
the new sysfs read-only attribute /sys/firmware/dump/dump_area_size.
Suggested-by: Heiko Carstens <hca@linux.ibm.com>
Suggested-by: Peter Oberparleiter <oberpar@linux.ibm.com>
Signed-off-by: Alexander Egorenkov <egorenar@linux.ibm.com>
Reviewed-by: Peter Oberparleiter <oberpar@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Get rid of sparse warnings:
CHECK drivers/s390/char/con3270.c
drivers/s390/char/con3270.c:531:15: warning: Using plain integer as NULL pointer
drivers/s390/char/con3270.c:749:15: warning: Using plain integer as NULL pointer
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
While pci_iov.h has include guards it doesn't include the necessary
header for struct zpci_dev, pci_bus.h on the other hand lacks even basic
include guards. This isn't only fragile and breaks convention but also
confuses tooling such as clang-analyzer. Add both include guards and the
necessary includes.
Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com>
Reviewed-by: Gerd Bayer <gbayer@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Add a trival phys_to_target_node() implementation which always returns 0 if
CONFIG_NUMA is enabled, since the s390 NUMA implementation only supports
node 0.
This is similar to memory_add_physaddr_to_nid() in order to avoid runtime
warnings.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
David Hildenbrand says:
====================
virtio-mem: s390 support
Let's finally add s390 support for virtio-mem; my last RFC was sent
4 years ago, and a lot changed in the meantime.
The latest QEMU series is available at [1], which contains some more
details and a usage example on s390 (last patch).
There is not too much in here: The biggest part is querying a new diag(500)
STORAGE_LIMIT hypercall to obtain the proper "max_physmem_end".
The last three patches are not strictly required but certainly nice-to-have.
Note that -- in contrast to standby memory -- virtio-mem memory must be
configured to be automatically onlined as soon as hotplugged. The easiest
approach is using the "memhp_default_state=" kernel parameter or by using
proper udev rules. More details can be found at [2].
I have reviving+upstreaming a systemd service to handle configuring
that on my todo list, but for some reason I keep getting distracted ...
I tested various things, including:
* Various memory hotplug/hotunplug combinations
* Device hotplug/hotunplug
* /proc/iomem output
* reboot
* kexec
* kdump: make sure we properly enter the "kdump mode" in the virtio-mem
driver
kdump support for virtio-mem memory on s390 will be sent out separately.
v2 -> v3
* "s390/kdump: make is_kdump_kernel() consistently return "true" in kdump
environments only"
-> Sent out separately [3]
* "s390/physmem_info: query diag500(STORAGE LIMIT) to support QEMU/KVM memory
devices"
-> No query function for diag500 for now.
-> Update comment above setup_ident_map_size().
-> Optimize/rewrite diag500_storage_limit() [Heiko]
-> Change handling in detect_physmem_online_ranges [Alexander]
-> Improve documentation.
* "s390/sparsemem: provide memory_add_physaddr_to_nid() with CONFIG_NUMA"
-> Added after testing on systems with CONFIG_NUMA=y
v1 -> v2:
* Document the new diag500 subfunction
* Use "s390" instead of "s390x" consistently
[1] https://lkml.kernel.org/r/20241008105455.2302628-1-david@redhat.com
[2] https://virtio-mem.gitlab.io/user-guide/user-guide-linux.html
[3] https://lkml.kernel.org/r/20241023090651.1115507-1-david@redhat.com
====================
Link: https://lore.kernel.org/r/20241025141453.1210600-1-david@redhat.com/
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
virtio-mem uses memory_add_physaddr_to_nid() to determine the NID to use
for memory it adds.
We currently fallback to the dummy implementation in mm/numa.c with
CONFIG_NUMA, which will end up triggering an undesired pr_info_once():
Unknown online node for memory at 0x100000000, assuming node 0
On s390, we map all cpus and memory to node 0, so let's add a simple
memory_add_physaddr_to_nid() implementation that does exactly that,
but without complaining.
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Tested-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@linux.ibm.com>
Link: https://lore.kernel.org/r/20241025141453.1210600-8-david@redhat.com
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Ever since commit 421c175c4d ("[S390] Add support for memory hot-add.")
we've been using a section size of 256 MiB on s390 and 32 MiB on s390.
Before that, we were using a section size of 32 MiB on both
architectures.
Now that we have a new mechanism to expose additional memory to a VM --
virtio-mem -- reduce the section size to 128 MiB to allow for more
flexibility and reduce the metadata overhead when dealing with hot(un)plug
granularity smaller than 256 MiB.
128 MiB has been used by x86-64 since the very beginning. arm64 with 4k
base pages switched to 128 MiB as well: it's just big enough on these
architectures to allows for using a huge page (2 MiB) in the vmemmap in
sane setups with sizeof(struct page) == 64 bytes and a huge page mapping
in the direct mapping, while still allowing for small hot(un)plug
granularity.
For s390, we could even switch to a 64 MiB section size, as our huge page
size is 1 MiB: but the smaller the section size, the more sections we'll
have to manage especially on bigger machines. Making it consistent with
x86-64 and arm64 feels like the right thing for now.
Note that the smallest memory hot(un)plug granularity is also limited by
the memory block size, determined by extracting the memory increment
size from SCLP. Under QEMU/KVM, implementing virtio-mem, we expose 0;
therefore, we'll end up with a memory block size of 128 MiB with a
128 MiB section size.
Tested-by: Mario Casquero <mcasquer@redhat.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Tested-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@linux.ibm.com>
Link: https://lore.kernel.org/r/20241025141453.1210600-7-david@redhat.com
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Now that s390 code is prepared for memory devices that reside above the
maximum storage increment exposed through SCLP, everything is in place
to unlock virtio-mem support.
As virtio-mem in Linux currently supports logically onlining/offlining
memory in pageblock granularity, we have an effective hot(un)plug
granularity of 1 MiB on s390.
As virito-mem adds/removes individual Linux memory blocks (256MB), we
will currently never use gigantic pages in the identity mapping.
It is worth noting that neither storage keys nor storage attributes (e.g.,
data / nodat) are touched when onlining memory blocks, which is good
because we are not supposed to touch these parts for unplugged device
blocks that are logically offline in Linux.
We will currently never initialize storage keys for virtio-mem
memory -- IOW, storage_key_init_range() is never called. It could be added
in the future when plugging device blocks. But as that function
essentially does nothing without modifying the code (changing
PAGE_DEFAULT_ACC), that's just fine for now.
kexec should work as intended and just like on other architectures that
support virtio-mem: we will never place kexec binaries on virtio-mem
memory, and never indicate virtio-mem memory to the 2nd kernel. The
device driver in the 2nd kernel can simply reset the device --
turning all memory unplugged, to then start plugging memory and adding
them to Linux, without causing trouble because the memory is already
used elsewhere.
The special s390 kdump mode, whereby the 2nd kernel creates the ELF
core header, won't currently dump virtio-mem memory. The virtio-mem
driver has a special kdump mode, from where we can detect memory ranges
to dump. Based on this, support for dumping virtio-mem memory can be
added in the future fairly easily.
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Tested-by: Mario Casquero <mcasquer@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Tested-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@linux.ibm.com>
Link: https://lore.kernel.org/r/20241025141453.1210600-5-david@redhat.com
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
To support memory devices under QEMU/KVM, such as virtio-mem,
we have to prepare our kernel virtual address space accordingly and
have to know the highest possible physical memory address we might see
later: the storage limit. The good old SCLP interface is not suitable for
this use case.
In particular, memory owned by memory devices has no relationship to
storage increments, it is always detected using the device driver, and
unaware OSes (no driver) must never try making use of that memory.
Consequently this memory is located outside of the "maximum storage
increment"-indicated memory range.
Let's use our new diag500 STORAGE_LIMIT subcode to query this storage
limit that can exceed the "maximum storage increment", and use the
existing interfaces (i.e., SCLP) to obtain information about the initial
memory that is not owned+managed by memory devices.
If a hypervisor does not support such memory devices, the address exposed
through diag500 STORAGE_LIMIT will correspond to the maximum storage
increment exposed through SCLP.
To teach kdump on s390 to include memory owned by memory devices, there
will be ways to query the relevant memory ranges from the device via a
driver running in special kdump mode (like virtio-mem already implements
to filter /proc/vmcore access so we don't end up reading from unplugged
device blocks).
Update setup_ident_map_size(), to clarify that there can be more than
just online and standby memory.
Tested-by: Mario Casquero <mcasquer@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com>
Tested-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@linux.ibm.com>
Link: https://lore.kernel.org/r/20241025141453.1210600-4-david@redhat.com
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Reservation of the PMU hardware is done at first event creation
and is protected by a pair of mutex_lock() and mutex_unlock().
After reservation of the PMU hardware the memory
required for the PMUs the event is to be installed on is
allocated by allocate_buffers() and alloc_sampling_buffer().
This done outside of the mutex protection.
Without mutex protection two or more concurrent invocations of
perf_event_init() may run in parallel.
This can lead to allocation of Sample Data Blocks (SDBs)
multiple times for the same PMU.
Prevent this and protect memory allocation of SDBs by
mutex.
Fixes: 8a6fe8f21e ("s390/cpum_sf: Use refcount_t instead of atomic_t")
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Combine the two VM_FAULT_ERROR checks in do_exception() and move them
to the exit path, similar to x86. Also remove a random blank line.
Suggested-by: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Reviewed-by: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
With GMAP_FAULT fault type gone, there are only KERNEL_FAULT and
USER_FAULT fault types left. Therefore there is no need for any fault
type switch statements left.
Rename get_fault_type() into is_kernel_fault() and let it return a
boolean value. Change all switch statements to if statements. This
removes quite a bit of code.
Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com>
Link: https://lore.kernel.org/r/20241022120601.167009-11-imbrenda@linux.ibm.com
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
With the gmap code gone get_fault_type() can be simplified:
- every fault with user_mode(regs) == true must be a fault in user address
space
- every fault with user_mode(regs) == false is only a fault in user
address space if the used address space is the secondary address space
- every other fault is within the kernel address space
Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com>
Link: https://lore.kernel.org/r/20241022120601.167009-10-imbrenda@linux.ibm.com
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Any program interrupt that happens in the host during the execution of
a KVM guest will now short circuit the fault handler and return to KVM
immediately. Guest fault handling (including pfault) will happen
entirely inside KVM.
When sie64a() returns zero, current->thread.gmap_int_code will contain
the program interrupt number that caused the exit, or zero if the exit
was not caused by a host program interrupt.
KVM will now take care of handling all guest faults in vcpu_post_run().
Since gmap faults will not be visible by the rest of the kernel, remove
GMAP_FAULT, the linux fault handlers for secure execution faults, the
exception table entries for the sie instruction, the nop padding after
the sie instruction, and all other references to guest faults from the
s390 code.
Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Co-developed-by: Heiko Carstens <hca@linux.ibm.com>
Link: https://lore.kernel.org/r/20241022120601.167009-6-imbrenda@linux.ibm.com
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Now that the guest ASCE is passed as a parameter to __sie64a(),
_PIF_GUEST_FAULT can be used again to determine whether the fault was a
guest or host fault.
Since the guest ASCE will not be taken from the gmap pointer in lowcore
anymore, __GMAP_ASCE can be removed. For the same reason the guest
ASCE needs now to be saved into the cr1 save area unconditionally.
Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Acked-by: Janosch Frank <frankja@linux.ibm.com>
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Link: https://lore.kernel.org/r/20241022120601.167009-2-imbrenda@linux.ibm.com
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Setup_pmc_cpu() function body consists of one single switch
statement with two cases PMC_INIT and PMC_RELEASE.
In both of these cases sf_disable() is invoked to turn off the
CPU Measurement sampling facility.
Move sf_disable() out of the switch statement.
No functional change.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Reviewed-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Extend the existing paes cipher to exploit the full AES-XTS hardware
acceleration introduced with message-security assist extension 10.
The full AES-XTS mode requires a protected key of type
PKEY_KEYTYPE_AES_XTS_128 or PKEY_KEYTYPE_AES_XTS_256.
Reviewed-by: Harald Freudenberger <freude@linux.ibm.com>
Signed-off-by: Holger Dengler <dengler@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Store the input key material of paes-xts in a single key_blob
structure. The split of the input key material is postponed to the key
conversion. Split the key material only, if the returned protected
keytype requires a second protected key.
For clear key pairs, prepare a clearkey token for each key and convert
them separately to protected keys. Store the concatenated conversion
results as input key in the context. All other input keys are stored
as is.
Reviewed-by: Harald Freudenberger <freude@linux.ibm.com>
Signed-off-by: Holger Dengler <dengler@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Move the conversion of a clearkey blob to token into a separate
function.
The functionality of the paes module is not affected by this commit.
Reviewed-by: Harald Freudenberger <freude@linux.ibm.com>
Signed-off-by: Holger Dengler <dengler@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
As a preparation for multiple key tokens in a key_blob structure, use
separate pointer and length parameters for __paes_keyblob2pkey()
instead a pointer to the struct key_blob.
The functionality of the paes module is not affected by this commit.
Reviewed-by: Harald Freudenberger <freude@linux.ibm.com>
Signed-off-by: Holger Dengler <dengler@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
The paes module uses only AES related structures and constants of the
pkey module. As pkey also supports protected keys other than AES keys,
the structures and size constants of the pkey module may be
changed. Use module-local structures and size constants for paes to
prevent any unwanted side effect by such a change.
The struct pkey_protkey is used to store the protected key blob
together with its length and type. The structure is only used locally,
it is not required for any pkey API call. So define the module-local
structure struct paes_protkey instead.
While at it, unify the names of struct paes_protkey variables on
stack.
The functionality of the paes module is not affected by this commit.
Reviewed-by: Harald Freudenberger <freude@linux.ibm.com>
Signed-off-by: Holger Dengler <dengler@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Reverse x-mas tree order for stack variables in paes module. While at
it, rename stack variables ret to rc.
The functionality of the paes module is not affected by this commit.
Reviewed-by: Harald Freudenberger <freude@linux.ibm.com>
Signed-off-by: Holger Dengler <dengler@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
The pkey handlers should only check, if the length of a key blob is big
enough for holding a key. Larger blobs should be tolerated.
Reviewed-by: Harald Freudenberger <freude@linux.ibm.com>
Signed-off-by: Holger Dengler <dengler@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
CPU hotplug remove handling triggers the following function
call sequence:
CPUHP_AP_PERF_S390_SF_ONLINE --> s390_pmu_sf_offline_cpu()
...
CPUHP_AP_PERF_ONLINE --> perf_event_exit_cpu()
The s390 CPUMF sampling CPU hotplug handler invokes:
s390_pmu_sf_offline_cpu()
+--> cpusf_pmu_setup()
+--> setup_pmc_cpu()
+--> deallocate_buffers()
This function de-allocates all sampling data buffers (SDBs) allocated
for that CPU at event initialization. It also clears the
PMU_F_RESERVED bit. The CPU is gone and can not be sampled.
With the event still being active on the removed CPU, the CPU event
hotplug support in kernel performance subsystem triggers the
following function calls on the removed CPU:
perf_event_exit_cpu()
+--> perf_event_exit_cpu_context()
+--> __perf_event_exit_context()
+--> __perf_remove_from_context()
+--> event_sched_out()
+--> cpumsf_pmu_del()
+--> cpumsf_pmu_stop()
+--> hw_perf_event_update()
to stop and remove the event. During removal of the event, the
sampling device driver tries to read out the remaining samples from
the sample data buffers (SDBs). But they have already been freed
(and may have been re-assigned). This may lead to a use after free
situation in which case the samples are most likely invalid. In the
best case the memory has not been reassigned and still contains
valid data.
Remedy this situation and check if the CPU is still in reserved
state (bit PMU_F_RESERVED set). In this case the SDBs have not been
released an contain valid data. This is always the case when
the event is removed (and no CPU hotplug off occured).
If the PMU_F_RESERVED bit is not set, the SDB buffers are gone.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Fix format string in pr_err() and use the built-in
hexadecimal prefix %#x to display a number with a leading
hexadecimal indicator 0x.
No functional change.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Reviewed-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Use sf_buffer_available() consistently throughtout the code
to test for the existence of sampling buffer.
No functional change.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Reviewed-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
When the sampling buffer allocation fails in
__hw_perf_event_init(), jump to the end of the function
and return the result. This is consistent with the other
error handling and return conditions in this function.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Reviewed-By: Sumanth Korikkar <sumanthk@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Event delete removes an event from the event list, but common
code invokes the PMU's enable function later on. This happens
in event_sched_out() and leads to the following call sequence:
event_sched_out()
+--> cpumsf_pmu_del()
+--> cpumsf_pmu_enable()
In cpumsf_pmu_enable() return immediately when the event is not
active. Also remove an unneeded if clause. That if() statement
is only reached when flag PMU_F_IN_USE has been set in
cpumsf_pmu_add(). And this function also sets cpuhw->event
to a valid value.
Remove WARN_ON_ONCE() statement which never triggered.
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Acked-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
