Kajol Jain [Fri, 13 Aug 2021 08:21:58 +0000 (13:51 +0530)]
powerpc/perf/hv-gpci: Fix counter value parsing
H_GetPerformanceCounterInfo (0xF080) hcall returns the counter data in
the result buffer. Result buffer has specific format defined in the PAPR
specification. One of the fields is counter offset and width of the
counter data returned.
Counter data are returned in a unsigned char array in big endian byte
order. To get the final counter data, the values must be left shifted
byte at a time. But commit a037f249dbe62 ("powerpc/perf: Add support for
the hv gpci (get performance counter info) interface") made the shifting
bitwise and also assumed little endian order. Because of that, hcall
counters values are reported incorrectly.
In particular this can lead to counters go backwards which messes up the
counter prev vs now calculation and leads to huge counter value
reporting:
Lukas Bulwahn [Thu, 19 Aug 2021 11:39:53 +0000 (13:39 +0200)]
powerpc/kvm: Remove obsolete and unneeded select
Commit 57b1c73b1b3b ("powerpc: Fix compile issue with force DAWR")
selects the non-existing config PPC_DAWR_FORCE_ENABLE for config
KVM_BOOK3S_64_HANDLER. As this commit also introduces a config PPC_DAWR
and this config PPC_DAWR is selected with PPC if PPC64, there is no
need for any further select in the KVM_BOOK3S_64_HANDLER.
Remove an obsolete and unneeded select in config KVM_BOOK3S_64_HANDLER.
The issue was identified with ./scripts/checkkconfigsymbols.py.
Fixes: 57b1c73b1b3b ("powerpc: Fix compile issue with force DAWR") Signed-off-by: Lukas Bulwahn <lukas.bulwahn@gmail.com> Reviewed-by: Daniel Axtens <dja@axtens.net> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20210819113954.17515-2-lukas.bulwahn@gmail.com
Wan Jiabing [Tue, 23 Mar 2021 06:29:05 +0000 (14:29 +0800)]
powerpc: Remove duplicate includes
interrupt.c: asm/interrupt.h has been included at line 12, so remove the
duplicate one at line 10.
time.c: linux/sched/clock.h has been included at line 33,so remove the
duplicate one at line 56 and move sched/cputime.h under sched including
segament.
Joel Stanley [Tue, 17 Aug 2021 04:54:06 +0000 (14:24 +0930)]
powerpc/config: Renable MTD_PHYSMAP_OF
CONFIG_MTD_PHYSMAP_OF is not longer enabled as it depends on
MTD_PHYSMAP which is not enabled.
This is a regression from commit 9dbf99ff9466 ("mtd: maps: Merge
physmap_of.c into physmap-core.c"), which added the extra dependency.
Add CONFIG_MTD_PHYSMAP=y so this stays in the config, as Christophe said
it is useful for build coverage.
Fixes: 9dbf99ff9466 ("mtd: maps: Merge physmap_of.c into physmap-core.c") Signed-off-by: Joel Stanley <joel@jms.id.au> Acked-by: Christophe Leroy <christophe.leroy@csgroup.eu> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20210817045407.2445664-3-joel@jms.id.au
Fangrui Song [Fri, 13 Aug 2021 20:05:11 +0000 (13:05 -0700)]
powerpc: Add "-z notext" flag to disable diagnostic
Object files used to link .tmp_vmlinux.kallsyms1 have many
R_PPC64_ADDR64 relocations in non-SHF_WRITE sections. There are many
text relocations (e.g. in .rela___ksymtab_gpl+* and .rela__mcount_loc
sections) in a -pie link and are disallowed by LLD:
ld.lld: error: can't create dynamic relocation R_PPC64_ADDR64 against local symbol in readonly segment; recompile object files with -fPIC or pass '-Wl,-z,notext' to allow text relocations in the output
>>> defined in arch/powerpc/kernel/head_64.o
>>> referenced by arch/powerpc/kernel/head_64.o:(__restart_table+0x10)
Newer GNU ld configured with "--enable-textrel-check=error" will report
an error as well:
Add "-z notext" to suppress the errors. Non-CONFIG_RELOCATABLE builds
use the default -no-pie mode and thus R_PPC64_ADDR64 relocations can be
resolved at link-time.
Reported-by: Itaru Kitayama <itaru.kitayama@riken.jp> Co-developed-by: Bill Wendling <morbo@google.com> Signed-off-by: Fangrui Song <maskray@google.com> Signed-off-by: Bill Wendling <morbo@google.com> Reviewed-by: Nick Desaulniers <ndesaulniers@google.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20210813200511.1905703-1-morbo@google.com
powerpc/bug: Remove specific powerpc BUG_ON() and WARN_ON() on PPC32
powerpc BUG_ON() and WARN_ON() are based on using twnei instruction.
For catching simple conditions like a variable having value 0, this
is efficient because it does the test and the trap at the same time.
But most conditions used with BUG_ON or WARN_ON are more complex and
forces GCC to format the condition into a 0 or 1 value in a register.
This will usually require 2 to 3 instructions.
The most efficient solution would be to use __builtin_trap() because
GCC is able to optimise the use of the different trap instructions
based on the requested condition, but this is complex if not
impossible for the following reasons:
- __builtin_trap() is a non-recoverable instruction, so it can't be
used for WARN_ON
- Knowing which line of code generated the trap would require the
analysis of DWARF information. This is not a feature we have today.
As mentioned in commit c952e9f1dbf2 ("Fix WARN_ON() on bitfield ops")
the way WARN_ON() is implemented is suboptimal. That commit also
mentions an issue with 'long long' condition. It fixed it for
WARN_ON() but the same problem still exists today with BUG_ON() on
PPC32. It will be fixed by using the generic implementation.
By using the generic implementation, gcc will naturally generate a
branch to the unconditional trap generated by BUG().
As modern powerpc implement zero-cycle branch,
that's even more efficient.
And for the functions using WARN_ON() and its return, the test
on return from WARN_ON() is now also used for the WARN_ON() itself.
On PPC64 we don't want it because we want to be able to use CFAR
register to track how we entered the code that trapped. The CFAR
register would be clobbered by the branch.
A simple test function:
unsigned long test9w(unsigned long a, unsigned long b)
{
if (WARN_ON(!b))
return 0;
return a / b;
}
powerpc/pseries: Add support for FORM2 associativity
PAPR interface currently supports two different ways of communicating resource
grouping details to the OS. These are referred to as Form 0 and Form 1
associativity grouping. Form 0 is the older format and is now considered
deprecated. This patch adds another resource grouping named FORM2.
powerpc/pseries: Add a helper for form1 cpu distance
This helper is only used with the dispatch trace log collection.
A later patch will add Form2 affinity support and this change helps
in keeping that simpler. Also add a comment explaining we don't expect
the code to be called with FORM0
powerpc/pseries: Consolidate different NUMA distance update code paths
The associativity details of the newly added resourced are collected from
the hypervisor via "ibm,configure-connector" rtas call. Update the numa
distance details of the newly added numa node after the above call.
Instead of updating NUMA distance every time we lookup a node id
from the associativity property, add helpers that can be used
during boot which does this only once. Also remove the distance
update from node id lookup helpers.
Currently, we duplicate parsing code for ibm,associativity and
ibm,associativity-lookup-arrays in the kernel. The associativity array provided
by these device tree properties are very similar and hence can use
a helper to parse the node id and numa distance details.
powerpc: rename powerpc_debugfs_root to arch_debugfs_dir
No functional change in this patch. arch_debugfs_dir is the generic kernel
name declared in linux/debugfs.h for arch-specific debugfs directory.
Architectures like x86/s390 already use the name. Rename powerpc
specific powerpc_debugfs_root to arch_debugfs_dir.
cpufreq: powernv: Fix init_chip_info initialization in numa=off
In the numa=off kernel command-line configuration init_chip_info() loops
around the number of chips and attempts to copy the cpumask of that node
which is NULL for all iterations after the first chip.
Hence, store the cpu mask for each chip instead of derving cpumask from
node while populating the "chips" struct array and copy that to the
chips[i].mask
Fixes: 92a1fc1dae4d ("cpufreq: powernv: Handle throttling due to Pmax capping at chip level") Cc: stable@vger.kernel.org # v4.3+ Reported-by: Shirisha Ganta <shirisha.ganta1@ibm.com> Signed-off-by: Pratik R. Sampat <psampat@linux.ibm.com> Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
[mpe: Rename goto label to out_free_chip_cpu_mask] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20210728120500.87549-2-psampat@linux.ibm.com
powerpc: wii.dts: Reduce the size of the control area
This is wrong, but needed in order to avoid overlapping ranges with the
OTP area added in the next commit. A refactor of this part of the
device tree is needed: according to Wiibrew[1], this area starts at
0x0d800000 and spans 0x400 bytes (that is, 0x100 32-bit registers),
encompassing PIC and GPIO registers, amongst the ones already exposed in
this device tree, which should become children of the control@d800000
node.
Marc Zyngier [Mon, 2 Aug 2021 16:26:28 +0000 (17:26 +0100)]
powerpc: Bulk conversion to generic_handle_domain_irq()
Wherever possible, replace constructs that match either
generic_handle_irq(irq_find_mapping()) or
generic_handle_irq(irq_linear_revmap()) to a single call to
generic_handle_domain_irq().
KVM: PPC: Book3S HV: XIVE: Add support for automatic save-restore
On P10, the feature doing an automatic "save & restore" of a VCPU
interrupt context is set by default in OPAL. When a VP context is
pulled out, the state of the interrupt registers are saved by the XIVE
interrupt controller under the internal NVP structure representing the
VP. This saves a costly store/load in guest entries and exits.
If OPAL advertises the "save & restore" feature in the device tree,
it should also have set the 'H' bit in the CAM line. Check that when
vCPUs are connected to their ICP in KVM before going any further.
There is no need to use the lockup detector ("noirqdebug") for IPIs.
The ipistorm benchmark measures a ~10% improvement on high systems
when this flag is set.
powerpc/xive: Use XIVE domain under xmon and debugfs
The default domain of the PCI/MSIs is not the XIVE domain anymore. To
list the IRQ mappings under XMON and debugfs, query the IRQ data from
the low level XIVE domain.
KVM: PPC: Book3S HV: XICS: Fix mapping of passthrough interrupts
PCI MSIs now live in an MSI domain but the underlying calls, which
will EOI the interrupt in real mode, need an HW IRQ number mapped in
the XICS IRQ domain. Grab it there.
pnv_opal_pci_msi_eoi() is called from KVM to EOI passthrough interrupts
when in real mode. Adding MSI domain broke the hack using the
'ioda.irq_chip' field to deduce the owning PHB. Fix that by using the
IRQ chip data in the MSI domain.
The 'ioda.irq_chip' field is now unused and could be removed from the
pnv_phb struct.
powerpc/powernv/pci: Set the IRQ chip data for P8/CXL devices
Before MSI domains, the default IRQ chip of PHB3 MSIs was patched by
pnv_set_msi_irq_chip() with the custom EOI handler pnv_ioda2_msi_eoi()
and the owning PHB was deduced from the 'ioda.irq_chip' field. This
path has been deprecated by the MSI domains but it is still in use by
the P8 CAPI 'cxl' driver.
Rewriting this driver to support MSI would be a waste of time.
Nevertheless, we can still remove the IRQ chip patch and set the IRQ
chip data instead. This is cleaner.
desc->irq_data points to the top level IRQ data descriptor which is
not necessarily in the XICS IRQ domain. MSIs are in another domain for
instance. Fix that by looking for a mapping on the low level XICS IRQ
domain.
powerpc/powernv/pci: Customize the MSI EOI handler to support PHB3
PHB3s need an extra OPAL call to EOI the interrupt. The call takes an
OPAL HW IRQ number but it is translated into a vector number in OPAL.
Here, we directly use the vector number of the in-the-middle "PNV-MSI"
domain instead of grabbing the OPAL HW IRQ number in the XICS parent
domain.
powerpc/xics: Rename the map handler in a check handler
This moves the IRQ initialization done under the different ICS backends
in the common part of XICS. The 'map' handler becomes a simple 'check'
on the HW IRQ at the FW level.
As we don't need an ICS anymore in xics_migrate_irqs_away(), the XICS
domain does not set a chip data for the IRQ.
We always had only one ICS per machine. Simplify the XICS driver by
removing the ICS list.
The ICS stored in the chip data of the XICS domain becomes useless and
we don't need it anymore to migrate away IRQs from a CPU. This will be
removed in a subsequent patch.
KVM: PPC: Book3S HV: XIVE: Fix mapping of passthrough interrupts
PCI MSI interrupt numbers are now mapped in a PCI-MSI domain but the
underlying calls handling the passthrough of the interrupt in the
guest need a number in the XIVE IRQ domain.
Use the IRQ data mapped in the XIVE IRQ domain and not the one in the
PCI-MSI domain.
The routine kvmppc_set_passthru_irq() calls kvmppc_xive_set_mapped()
and kvmppc_xive_clr_mapped() with an IRQ descriptor. Use directly the
host IRQ number to remove a useless conversion.
KVM: PPC: Book3S HV: Use the new IRQ chip to detect passthrough interrupts
Passthrough PCI MSI interrupts are detected in KVM with a check on a
specific EOI handler (P8) or on XIVE (P9). We can now check the
PCI-MSI IRQ chip which is cleaner.
This is very similar to the MSI domains of the pSeries platform. The
MSI allocator is directly handled under the Linux PHB in the
in-the-middle "PNV-MSI" domain.
Only the XIVE (P9/P10) parent domain is supported for now. Support for
XICS will come later.
powerpc/pseries/pci: Add a msi_free() handler to clear XIVE data
The MSI domain clears the IRQ with msi_domain_free(), which calls
irq_domain_free_irqs_top(), which clears the handler data. This is a
problem for the XIVE controller since we need to unmap MMIO pages and
free a specific XIVE structure.
The 'msi_free()' handler is called before irq_domain_free_irqs_top()
when the handler data is still available. Use that to clear the XIVE
controller data.
powerpc/xive: Remove irqd_is_started() check when setting the affinity
In the early days of XIVE support, commit b8c61a0f1980 ("powerpc/xive:
Ensure active irqd when setting affinity") tried to fix an issue
related to interrupt migration. If the root cause was related to CPU
unplug, it should have been fixed and there is no reason to keep the
irqd_is_started() check. This test is also breaking affinity setting
of MSIs which can set before starting the associated IRQ.
Two IRQ domains are added on top of default machine IRQ domain.
First, the top level "pSeries-PCI-MSI" domain deals with the MSI
specificities. In this domain, the HW IRQ numbers are generated by the
PCI MSI layer, they compose a unique ID for an MSI source with the PCI
device identifier and the MSI vector number.
These numbers can be quite large on a pSeries machine running under
the IBM Hypervisor and /sys/kernel/irq/ and /proc/interrupts will
require small fixes to show them correctly.
Second domain is the in-the-middle "pSeries-MSI" domain which acts as
a proxy between the PCI MSI subsystem and the machine IRQ subsystem.
It usually allocate the MSI vector numbers but, on pSeries machines,
this is done by the RTAS FW and RTAS returns IRQ numbers in the IRQ
number space of the machine. This is why the in-the-middle "pSeries-MSI"
domain has the same HW IRQ numbers as its parent domain.
Only the XIVE (P9/P10) parent domain is supported for now. We still
need to add support for IRQ domain hierarchy under XICS.
powerpc/xive: Add support for IRQ domain hierarchy
This adds handlers to allocate/free IRQs in a domain hierarchy. We
could try to use xive_irq_domain_map() in xive_irq_domain_alloc() but
we rely on xive_irq_alloc_data() to set the IRQ handler data and
duplicating the code is simpler.
xive_irq_free_data() needs to be called when IRQ are freed to clear
the MMIO mappings and free the XIVE handler data, xive_irq_data
structure. This is going to be a problem with MSI domains which we
will address later.
This splits the routine setting the MSIs in two parts: allocation of
MSIs for the PCI device at the FW level (RTAS) and the actual mapping
and activation of the IRQs.
rtas_prepare_msi_irqs() will serve as a handler for the PCI MSI domain.
KVM: PPC: Use arch_get_random_seed_long instead of powernv variant
The powernv_get_random_long() does not work in nested KVM (which is
pseries) and produces a crash when accessing in_be64(rng->regs) in
powernv_get_random_long().
This replaces powernv_get_random_long with the ppc_md machine hook
wrapper.
Jordan Niethe [Wed, 4 Aug 2021 01:37:24 +0000 (11:37 +1000)]
powerpc: Always inline radix_enabled() to fix build failure
This is the same as commit 1830fc1ac177 ("powerpc: Force inlining of
mmu_has_feature to fix build failure") but for radix_enabled(). The
config in the linked bugzilla causes the following build failure:
LD .tmp_vmlinux.kallsyms1
powerpc64-linux-ld: arch/powerpc/mm/pgtable.o: in function `.__ptep_set_access_flags':
pgtable.c:(.text+0x17c): undefined reference to `.radix__ptep_set_access_flags'
powerpc64-linux-ld: arch/powerpc/mm/pageattr.o: in function `.change_page_attr':
pageattr.c:(.text+0xc0): undefined reference to `.radix__flush_tlb_kernel_range'
etc.
This is due to radix_enabled() not being inlined. See extract from
building with -Winline:
In file included from arch/powerpc/include/asm/lppaca.h:46,
from arch/powerpc/include/asm/paca.h:17,
from arch/powerpc/include/asm/current.h:13,
from include/linux/thread_info.h:23,
from include/asm-generic/preempt.h:5,
from ./arch/powerpc/include/generated/asm/preempt.h:1,
from include/linux/preempt.h:78,
from include/linux/spinlock.h:51,
from include/linux/mmzone.h:8,
from include/linux/gfp.h:6,
from arch/powerpc/mm/pgtable.c:21:
arch/powerpc/include/asm/book3s/64/pgtable.h: In function '__ptep_set_access_flags':
arch/powerpc/include/asm/mmu.h:327:20: error: inlining failed in call to 'radix_enabled': call is unlikely and code size would grow [-Werror=inline]
The code relies on constant folding of MMU_FTRS_POSSIBLE at buildtime
and elimination of non possible parts of code at compile time. For this
to work radix_enabled() must be inlined so make it __always_inline.
The functions get_online_cpus() and put_online_cpus() have been
deprecated during the CPU hotplug rework. They map directly to
cpus_read_lock() and cpus_read_unlock().
Replace deprecated CPU-hotplug functions with the official version.
The behavior remains unchanged.
powerpc/pseries: Prevent free CPU ids being reused on another node
When a CPU is hot added, the CPU ids are taken from the available mask
from the lower possible set. If that set of values was previously used
for a CPU attached to a different node, it appears to an application as
if these CPUs have migrated from one node to another node which is not
expected.
To prevent this, it is needed to record the CPU ids used for each node
and to not reuse them on another node. However, to prevent CPU hot plug
to fail, in the case the CPU ids is starved on a node, the capability to
reuse other nodes’ free CPU ids is kept. A warning is displayed in such
a case to warn the user.
A new CPU bit mask (node_recorded_ids_map) is introduced for each
possible node. It is populated with the CPU onlined at boot time, and
then when a CPU is hot plugged to a node. The bits in that mask remain
when the CPU is hot unplugged, to remind this CPU ids have been used for
this node.
If no id set was found, a retry is made without removing the ids used on
the other nodes to try reusing them. This is the way ids have been
allocated prior to this patch.
The effect of this patch can be seen by removing and adding CPUs using
the Qemu monitor. In the following case, the first CPU from the node 2
is removed, then the first one from the node 1 is removed too. Later,
the first CPU of the node 2 is added back. Without that patch, the
kernel will number these CPUs using the first CPU ids available which
are the ones freed when removing the second CPU of the node 0. This
leads to the CPU ids 16-23 to move from the node 1 to the node 2. With
the patch applied, the CPU ids 32-39 are used since they are the lowest
free ones which have not been used on another node.
Laurent Dufour [Mon, 17 May 2021 09:06:06 +0000 (11:06 +0200)]
pseries/drmem: update LMBs after LPM
After a LPM, the device tree node ibm,dynamic-reconfiguration-memory may be
updated by the hypervisor in the case the NUMA topology of the LPAR's
memory is updated.
This is handled by the kernel, but the memory's node is not updated because
there is no way to move a memory block between nodes from the Linux kernel
point of view.
If later a memory block is added or removed, drmem_update_dt() is called
and it is overwriting the DT node ibm,dynamic-reconfiguration-memory to
match the added or removed LMB. But the LMB's associativity node has not
been updated after the DT node update and thus the node is overwritten by
the Linux's topology instead of the hypervisor one.
Introduce a hook called when the ibm,dynamic-reconfiguration-memory node is
updated to force an update of the LMB's associativity. However, ignore the
call to that hook when the update has been triggered by drmem_update_dt().
Because, in that case, the LMB tree has been used to set the DT property
and thus it doesn't need to be updated back. Since drmem_update_dt() is
called under the protection of the device_hotplug_lock and the hook is
called in the same context, use a simple boolean variable to detect that
call.
Laurent Dufour [Tue, 11 May 2021 07:31:36 +0000 (09:31 +0200)]
powerpc/numa: Consider the max NUMA node for migratable LPAR
When a LPAR is migratable, we should consider the maximum possible NUMA
node instead of the number of NUMA nodes from the actual system.
The DT property 'ibm,current-associativity-domains' defines the maximum
number of nodes the LPAR can see when running on that box. But if the
LPAR is being migrated on another box, it may see up to the nodes
defined by 'ibm,max-associativity-domains'. So if a LPAR is migratable,
that value should be used.
Unfortunately, there is no easy way to know if an LPAR is migratable or
not. The hypervisor exports the property 'ibm,migratable-partition' in
the case it set to migrate partition, but that would not mean that the
current partition is migratable.
Without this patch, when a LPAR is started on a 2 node box and then
migrated to a 3 node box, the hypervisor may spread the LPAR's CPUs on
the 3rd node. In that case if a CPU from that 3rd node is added to the
LPAR, it will be wrongly assigned to the node because the kernel has
been set to use up to 2 nodes (the configuration of the departure node).
With this patch applies, the CPU is correctly added to the 3rd node.
Fixes: ba2c5eea4d6b ("powerpc/numa: Detect support for coregroup") Signed-off-by: Laurent Dufour <ldufour@linux.ibm.com> Reviewed-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20210511073136.17795-1-ldufour@linux.ibm.com
powerpc/non-smp: Unconditionaly call smp_mb() on switch_mm
Commit 321f5da583d4 ("powerpc, membarrier: Skip memory barrier in
switch_mm()") added some logic to skip the smp_mb() in
switch_mm_irqs_off() before the call to switch_mmu_context().
However, on non SMP smp_mb() is just a compiler barrier and doing
it unconditionaly is simpler than the logic used to check whether the
barrier is needed or not.
Nicholas Piggin [Tue, 20 Jul 2021 14:15:04 +0000 (00:15 +1000)]
powerpc/64s/perf: Always use SIAR for kernel interrupts
If an interrupt is taken in kernel mode, always use SIAR for it rather than
looking at regs_sipr. This prevents samples piling up around interrupt
enable (hard enable or interrupt replay via soft enable) in PMUs / modes
where the PR sample indication is not in synch with SIAR.
This results in better sampling of interrupt entry and exit in particular.
Parth Shah [Wed, 28 Jul 2021 17:56:07 +0000 (23:26 +0530)]
powerpc/smp: Use existing L2 cache_map cpumask to find L3 cache siblings
On POWER10 systems, the "ibm,thread-groups" property "2" indicates the cpus
in thread-group share both L2 and L3 caches. Hence, use cache_property = 2
itself to find both the L2 and L3 cache siblings.
Hence, create a new thread_group_l3_cache_map to keep list of L3 siblings,
but fill the mask using same property "2" array.
powerpc/cacheinfo: Remove the redundant get_shared_cpu_map()
The helper function get_shared_cpu_map() was added in
'commit 288e2142824b ("powerpc/cacheinfo: Report the correct
shared_cpu_map on big-cores")'
and subsequently expanded upon in
'commit 127abbf0a52d ("powerpc/cacheinfo: Print correct cache-sibling
map/list for L2 cache")'
in order to help report the correct groups of threads sharing these caches
on big-core systems where groups of threads within a core can share
different sets of caches.
Now that powerpc/cacheinfo is aware of "ibm,thread-groups" property,
cache->shared_cpu_map contains the correct set of thread-siblings
sharing the cache. Hence we no longer need the functions
get_shared_cpu_map(). This patch removes this function. We also remove
the helper function index_dir_to_cpu() which was only called by
get_shared_cpu_map().
With these functions removed, we can still see the correct
cache-sibling map/list for L1 and L2 caches on systems with L1 and L2
caches distributed among groups of threads in a core.
With this patch, on a SMT8 POWER10 system where the L1 and L2 caches
are split between the two groups of threads in a core, for CPUs 8,9,
the L1-Data, L1-Instruction, L2, L3 cache CPU sibling list is as
follows:
powerpc/cacheinfo: Lookup cache by dt node and thread-group id
Currently the cacheinfo code on powerpc indexes the "cache" objects
(modelling the L1/L2/L3 caches) where the key is device-tree node
corresponding to that cache. On some of the POWER server platforms
thread-groups within the core share different sets of caches (Eg: On
SMT8 POWER9 systems, threads 0,2,4,6 of a core share L1 cache and
threads 1,3,5,7 of the same core share another L1 cache). On such
platforms, there is a single device-tree node corresponding to that
cache and the cache-configuration within the threads of the core is
indicated via "ibm,thread-groups" device-tree property.
Since the current code is not aware of the "ibm,thread-groups"
property, on the aforementoined systems, cacheinfo code still treats
all the threads in the core to be sharing the cache because of the
single device-tree node (In the earlier example, the cacheinfo code
would says CPUs 0-7 share L1 cache).
In this patch, we make the powerpc cacheinfo code aware of the
"ibm,thread-groups" property. We indexe the "cache" objects by the
key-pair (device-tree node, thread-group id). For any CPUX, for a
given level of cache, the thread-group id is defined to be the first
CPU in the "ibm,thread-groups" cache-group containing CPUX. For levels
of cache which are not represented in "ibm,thread-groups" property,
the thread-group id is -1.
[parth: Remove "static" keyword for the definition of "thread_group_l1_cache_map"
and "thread_group_l2_cache_map" to get rid of the compile error.]
powerpc: move the install rule to arch/powerpc/Makefile
Currently, the install target in arch/powerpc/Makefile descends into
arch/powerpc/boot/Makefile to invoke the shell script, but there is no
good reason to do so.
arch/powerpc/Makefile can run the shell script directly.
powerpc: remove unused zInstall target from arch/powerpc/boot/Makefile
Commit 2541f318dc5b ("powerpc/boot: Don't install zImage.* from make
install") added the zInstall target to arch/powerpc/boot/Makefile,
but you cannot use it since the corresponding hook is missing in
arch/powerpc/Makefile.
It has never worked since its addition. Nobody has complained about
it for 7 years, which means this code was unneeded.
With this removal, the install.sh will be passed in with 4 parameters.
Simplify the shell script.
cpuidle: pseries: Mark pseries_idle_proble() as __init
After commit 7cbd631d4dec ("cpuidle: pseries: Fixup CEDE0 latency only
for POWER10 onwards"), pseries_idle_probe() is no longer inlined when
compiling with clang, which causes a modpost warning:
WARNING: modpost: vmlinux.o(.text+0xc86a54): Section mismatch in
reference from the function pseries_idle_probe() to the function
.init.text:fixup_cede0_latency()
The function pseries_idle_probe() references
the function __init fixup_cede0_latency().
This is often because pseries_idle_probe lacks a __init
annotation or the annotation of fixup_cede0_latency is wrong.
pseries_idle_probe() is a non-init function, which calls
fixup_cede0_latency(), which is an init function, explaining the
mismatch. pseries_idle_probe() is only called from
pseries_processor_idle_init(), which is an init function, so mark
pseries_idle_probe() as __init so there is no more warning.
Michal Suchanek [Thu, 29 Jul 2021 18:01:03 +0000 (20:01 +0200)]
powerpc/stacktrace: Include linux/delay.h
commit 07e7fdc30d5f ("powerpc/stacktrace: Fix spurious "stale" traces in raise_backtrace_ipi()")
introduces udelay() call without including the linux/delay.h header.
This may happen to work on master but the header that declares the
functionshould be included nonetheless.
Fixes: 07e7fdc30d5f ("powerpc/stacktrace: Fix spurious "stale" traces in raise_backtrace_ipi()") Signed-off-by: Michal Suchanek <msuchanek@suse.de> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20210729180103.15578-1-msuchanek@suse.de
cpuidle: pseries: Do not cap the CEDE0 latency in fixup_cede0_latency()
Currently in fixup_cede0_latency() code, we perform the fixup the
CEDE(0) exit latency value only if minimum advertized extended CEDE
latency values are less than 10us. This was done so as to not break
the expected behaviour on POWER8 platforms where the advertised
latency was higher than the default 10us, which would delay the SMT
folding on the core.
However, after the earlier patch "cpuidle/pseries: Fixup CEDE0 latency
only for POWER10 onwards", we can be sure that the fixup of CEDE0
latency is going to happen only from POWER10 onwards. Hence
unconditionally use the minimum exit latency provided by the platform.
cpuidle: pseries: Fixup CEDE0 latency only for POWER10 onwards
Commit 59777e5341bb ("cpuidle: pseries: Fixup exit latency for
CEDE(0)") sets the exit latency of CEDE(0) based on the latency values
of the Extended CEDE states advertised by the platform
On POWER9 LPARs, the firmwares advertise a very low value of 2us for
CEDE1 exit latency on a Dedicated LPAR. The latency advertized by the
PHYP hypervisor corresponds to the latency required to wakeup from the
underlying hardware idle state. However the wakeup latency from the
LPAR perspective should include
1. The time taken to transition the CPU from the Hypervisor into the
LPAR post wakeup from platform idle state
2. Time taken to send the IPI from the source CPU (waker) to the idle
target CPU (wakee).
1. can be measured via timer idle test, where we queue a timer, say
for 1ms, and enter the CEDE state. When the timer fires, in the timer
handler we compute how much extra timer over the expected 1ms have we
consumed. On a a POWER9 LPAR the numbers are
CEDE latency measured using a timer (numbers in ns)
N Min Median Avg 90%ile 99%ile Max Stddev
400 2601 5677 5668.74 5917 6413 9299 455.01
1. and 2. combined can be determined by an IPI latency test where we
send an IPI to an idle CPU and in the handler compute the time
difference between when the IPI was sent and when the handler ran. We
see the following numbers on POWER9 LPAR.
CEDE latency measured using an IPI (numbers in ns)
N Min Median Avg 90%ile 99%ile Max Stddev
400 711 7564 7369.43 8559 9514 9698 1200.01
Suppose, we consider the 99th percentile latency value measured using
the IPI to be the wakeup latency, the value would be 9.5us This is in
the ballpark of the default value of 10us.
Hence, use the exit latency of CEDE(0) based on the latency values
advertized by platform only from POWER10 onwards. The values
advertized on POWER10 platforms is more realistic and informed by the
latency measurements. For earlier platforms stick to the default value
of 10us. The fix was suggested by Michael Ellerman.
Hari Bathini [Wed, 14 Jul 2021 12:47:58 +0000 (18:17 +0530)]
powerpc/kexec: blacklist functions called in real mode for kprobe
As kprobe does not handle events happening in real mode, blacklist the
functions that only get called in real mode or in kexec sequence with
MMU turned off.
Nicholas Piggin [Thu, 8 Jul 2021 11:26:22 +0000 (21:26 +1000)]
KVM: PPC: Book3S HV Nested: Sanitise H_ENTER_NESTED TM state
The H_ENTER_NESTED hypercall is handled by the L0, and it is a request
by the L1 to switch the context of the vCPU over to that of its L2
guest, and return with an interrupt indication. The L1 is responsible
for switching some registers to guest context, and the L0 switches
others (including all the hypervisor privileged state).
If the L2 MSR has TM active, then the L1 is responsible for
recheckpointing the L2 TM state. Then the L1 exits to L0 via the
H_ENTER_NESTED hcall, and the L0 saves the TM state as part of the exit,
and then it recheckpoints the TM state as part of the nested entry and
finally HRFIDs into the L2 with TM active MSR. Not efficient, but about
the simplest approach for something that's horrendously complicated.
Problems arise if the L1 exits to the L0 with a TM state which does not
match the L2 TM state being requested. For example if the L1 is
transactional but the L2 MSR is non-transactional, or vice versa. The
L0's HRFID can take a TM Bad Thing interrupt and crash.
Fix this by disallowing H_ENTER_NESTED in TM[T] state entirely, and then
ensuring that if the L1 is suspended then the L2 must have TM active,
and if the L1 is not suspended then the L2 must not have TM active.
Fixes: ab4fb3f93012 ("KVM: PPC: Book3S HV: Nested guest entry via hypercall") Cc: stable@vger.kernel.org # v4.20+ Reported-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Michael Neuling <mikey@neuling.org> Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Nicholas Piggin [Tue, 20 Jul 2021 10:43:09 +0000 (20:43 +1000)]
KVM: PPC: Book3S: Fix H_RTAS rets buffer overflow
The kvmppc_rtas_hcall() sets the host rtas_args.rets pointer based on
the rtas_args.nargs that was provided by the guest. That guest nargs
value is not range checked, so the guest can cause the host rets pointer
to be pointed outside the args array. The individual rtas function
handlers check the nargs and nrets values to ensure they are correct,
but if they are not, the handlers store a -3 (0xfffffffd) failure
indication in rets[0] which corrupts host memory.
Fix this by testing up front whether the guest supplied nargs and nret
would exceed the array size, and fail the hcall directly without storing
a failure indication to rets[0].
Also expand on a comment about why we kill the guest and try not to
return errors directly if we have a valid rets[0] pointer.
Fixes: d88dc3c0455d ("KVM: PPC: Book3S: Add infrastructure to implement kernel-side RTAS calls") Cc: stable@vger.kernel.org # v3.10+ Reported-by: Alexey Kardashevskiy <aik@ozlabs.ru> Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Merge tag 'perf-tools-fixes-for-v5.14-2021-07-18' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux
Pull perf tools fixes from Arnaldo Carvalho de Melo:
- Skip invalid hybrid PMU on hybrid systems when the atom (little) CPUs
are offlined.
- Fix 'perf test' problems related to the recently added hybrid
(BIG/little) code.
- Split ARM's coresight (hw tracing) decode by aux records to avoid
fatal decoding errors.
- Fix add event failure in 'perf probe' when running 32-bit perf in a
64-bit kernel.
- Fix 'perf sched record' failure when CONFIG_SCHEDSTATS is not set.
- Fix memory and refcount leaks detected by ASAn when running 'perf
test', should be clean of warnings now.
- Remove broken definition of __LITTLE_ENDIAN from tools'
linux/kconfig.h, which was breaking the build in some systems.
- Cast PTHREAD_STACK_MIN to int as it may turn into 'long
sysconf(__SC_THREAD_STACK_MIN_VALUE), breaking the build in some
systems.
- Fix libperf build error with LIBPFM4=1.
- Sync UAPI files changed by the memfd_secret new syscall.
* tag 'perf-tools-fixes-for-v5.14-2021-07-18' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux: (35 commits)
perf sched: Fix record failure when CONFIG_SCHEDSTATS is not set
perf probe: Fix add event failure when running 32-bit perf in a 64-bit kernel
perf data: Close all files in close_dir()
perf probe-file: Delete namelist in del_events() on the error path
perf test bpf: Free obj_buf
perf trace: Free strings in trace__parse_events_option()
perf trace: Free syscall tp fields in evsel->priv
perf trace: Free syscall->arg_fmt
perf trace: Free malloc'd trace fields on exit
perf lzma: Close lzma stream on exit
perf script: Fix memory 'threads' and 'cpus' leaks on exit
perf script: Release zstd data
perf session: Cleanup trace_event
perf inject: Close inject.output on exit
perf report: Free generated help strings for sort option
perf env: Fix memory leak of cpu_pmu_caps
perf test maps__merge_in: Fix memory leak of maps
perf dso: Fix memory leak in dso__new_map()
perf test event_update: Fix memory leak of unit
perf test event_update: Fix memory leak of evlist
...
Merge tag 'xfs-5.14-fixes-1' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux
Pull xfs fixes from Darrick Wong:
"A few fixes for issues in the new online shrink code, additional
corrections for my recent bug-hunt w.r.t. extent size hints on
realtime, and improved input checking of the GROWFSRT ioctl.
IOW, the usual 'I somehow got bored during the merge window and
resumed auditing the farther reaches of xfs':
- Fix shrink eligibility checking when sparse inode clusters enabled
- Reset '..' directory entries when unlinking directories to prevent
verifier errors if fs is shrinked later
- Don't report unusable extent size hints to FSGETXATTR
- Don't warn when extent size hints are unusable because the sysadmin
configured them that way
- Fix insufficient parameter validation in GROWFSRT ioctl
- Fix integer overflow when adding rt volumes to filesystem"
* tag 'xfs-5.14-fixes-1' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux:
xfs: detect misaligned rtinherit directory extent size hints
xfs: fix an integer overflow error in xfs_growfs_rt
xfs: improve FSGROWFSRT precondition checking
xfs: don't expose misaligned extszinherit hints to userspace
xfs: correct the narrative around misaligned rtinherit/extszinherit dirs
xfs: reset child dir '..' entry when unlinking child
xfs: check for sparse inode clusters that cross new EOAG when shrinking
Merge tag 'iomap-5.14-fixes-1' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux
Pull iomap fixes from Darrick Wong:
"A handful of bugfixes for the iomap code.
There's nothing especially exciting here, just fixes for UBSAN (not
KASAN as I erroneously wrote in the tag message) warnings about
undefined behavior in the SEEK_DATA/SEEK_HOLE code, and some
reshuffling of per-page block state info to fix some problems with
gfs2.
- Fix KASAN warnings due to integer overflow in SEEK_DATA/SEEK_HOLE
- Fix assertion errors when using inlinedata files on gfs2"
* tag 'iomap-5.14-fixes-1' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux:
iomap: Don't create iomap_page objects in iomap_page_mkwrite_actor
iomap: Don't create iomap_page objects for inline files
iomap: Permit pages without an iop to enter writeback
iomap: remove the length variable in iomap_seek_hole
iomap: remove the length variable in iomap_seek_data
Merge tag 'kbuild-fixes-v5.14' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild
Pull Kbuild fixes from Masahiro Yamada:
- Restore the original behavior of scripts/setlocalversion when
LOCALVERSION is set to empty.
- Show Kconfig prompts even for 'make -s'
- Fix the combination of COFNIG_LTO_CLANG=y and CONFIG_MODVERSIONS=y
for older GNU Make versions
* tag 'kbuild-fixes-v5.14' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild:
Documentation: Fix intiramfs script name
Kbuild: lto: fix module versionings mismatch in GNU make 3.X
kbuild: do not suppress Kconfig prompts for silent build
scripts/setlocalversion: fix a bug when LOCALVERSION is empty
Robert Richter [Thu, 15 Jul 2021 09:26:02 +0000 (11:26 +0200)]
Documentation: Fix intiramfs script name
Documentation was not changed when renaming the script in commit 63dd77bf5e60 ("initramfs: rename gen_initramfs_list.sh to
gen_initramfs.sh"). Fixing this.
Basically does:
$ sed -i -e s/gen_initramfs_list.sh/gen_initramfs.sh/g $(git grep -l gen_initramfs_list.sh)
Fixes: 63dd77bf5e60 ("initramfs: rename gen_initramfs_list.sh to gen_initramfs.sh") Signed-off-by: Robert Richter <rrichter@amd.com> Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Kbuild: lto: fix module versionings mismatch in GNU make 3.X
When building modules(CONFIG_...=m), I found some of module versions
are incorrect and set to 0.
This can be found in build log for first clean build which shows
WARNING: EXPORT symbol "XXXX" [drivers/XXX/XXX.ko] version generation failed,
symbol will not be versioned.
But in second build(incremental build), the WARNING disappeared and the
module version becomes valid CRC and make someone who want to change
modules without updating kernel image can't insert their modules.
The problematic code is
+ $(foreach n, $(filter-out FORCE,$^), \
+ $(if $(wildcard $(n).symversions), \
+ ; cat $(n).symversions >> $@.symversions))
`foreach n` shows nothing to `cat` into $(n).symversions because
`if $(wildcard $(n).symversions)` return nothing, but actually
they do exist during this line was executed.
-rw-r--r-- 1 root root 168580 Jun 13 19:10 fs/notify/fsnotify.o
-rw-r--r-- 1 root root 111 Jun 13 19:10 fs/notify/fsnotify.o.symversions
The reason is the $(n).symversions are generated at runtime, but
Makefile wildcard function expends and checks the file exist or not
during parsing the Makefile.
Thus fix this by use `test` shell command to check the file
existence in runtime.
Rebase from both:
1. [https://lore.kernel.org/lkml/20210616080252.32046-1-lecopzer.chen@mediatek.com/]
2. [https://lore.kernel.org/lkml/20210702032943.7865-1-lecopzer.chen@mediatek.com/]
kbuild: do not suppress Kconfig prompts for silent build
When a new CONFIG option is available, Kbuild shows a prompt to get
the user input.
$ make
[ snip ]
Core Scheduling for SMT (SCHED_CORE) [N/y/?] (NEW)
This is the only interactive place in the build process.
Commit a4714225623a ("kbuild: sink stdout from cmd for silent build")
suppressed Kconfig prompts as well because syncconfig is invoked by
the 'cmd' macro. You cannot notice the fact that Kconfig is waiting
for the user input.
Use 'kecho' to show the equivalent short log without suppressing stdout
from sub-make.
scripts/setlocalversion: fix a bug when LOCALVERSION is empty
The commit 7822fe915091 ("scripts/setlocalversion: simplify the short
version part") reduces indentation. Unfortunately, it also changes behavior
in a subtle way - if the user has empty "LOCALVERSION" variable, the plus
sign is appended to the kernel version. It wasn't appended before.
This patch reverts to the old behavior - we append the plus sign only if
the LOCALVERSION variable is not set.
Fixes: 7822fe915091 ("scripts/setlocalversion: simplify the short version part") Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Yang Jihong [Tue, 13 Jul 2021 11:23:58 +0000 (19:23 +0800)]
perf sched: Fix record failure when CONFIG_SCHEDSTATS is not set
The tracepoints trace_sched_stat_{wait, sleep, iowait} are not exposed to user
if CONFIG_SCHEDSTATS is not set, "perf sched record" records the three events.
As a result, the command fails.
Error: File /sys/kernel/tracing/events/sched/sched_stat_wait not found.
Hint: Perhaps this kernel misses some CONFIG_ setting to enable this feature?.
Run 'perf list' for a list of valid events
Usage: perf record [<options>] [<command>]
or: perf record [<options>] -- <command> [<options>]
-e, --event <event> event selector. use 'perf list' to list available events
Solution:
Check whether schedstat tracepoints are exposed. If no, these events are not recorded.
After:
# perf sched record sleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.163 MB perf.data (1091 samples) ]
# perf sched report
run measurement overhead: 4736 nsecs
sleep measurement overhead: 9059979 nsecs
the run test took 999854 nsecs
the sleep test took 8945271 nsecs
nr_run_events: 716
nr_sleep_events: 785
nr_wakeup_events: 0
...
------------------------------------------------------------
Fixes: 206bbb6df45ba ("sched/fair: do not expose some tracepoints to user if CONFIG_SCHEDSTATS is not set") Signed-off-by: Yang Jihong <yangjihong1@huawei.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt (VMware) <rostedt@goodmis.org> Cc: Yafang Shao <laoar.shao@gmail.com> Link: http://lore.kernel.org/lkml/20210713112358.194693-1-yangjihong1@huawei.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Yang Jihong [Thu, 15 Jul 2021 06:37:23 +0000 (14:37 +0800)]
perf probe: Fix add event failure when running 32-bit perf in a 64-bit kernel
The "address" member of "struct probe_trace_point" uses long data type.
If kernel is 64-bit and perf program is 32-bit, size of "address"
variable is 32 bits.
As a result, upper 32 bits of address read from kernel are truncated, an
error occurs during address comparison in kprobe_warn_out_range().
Before:
# perf probe -a schedule
schedule is out of .text, skip it.
Error: Failed to add events.
Solution:
Change data type of "address" variable to u64 and change corresponding
address printing and value assignment.
After:
# perf.new.new probe -a schedule
Added new event:
probe:schedule (on schedule)
Signed-off-by: Yang Jihong <yangjihong1@huawei.com> Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Frank Ch. Eigler <fche@redhat.com> Cc: Ian Rogers <irogers@google.com> Cc: Jianlin Lv <jianlin.lv@arm.com> Cc: Jin Yao <yao.jin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Li Huafei <lihuafei1@huawei.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ravi Bangoria <ravi.bangoria@linux.ibm.com> Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com> Link: http://lore.kernel.org/lkml/20210715063723.11926-1-yangjihong1@huawei.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
projects / kernel.git / log