Commit
5944ce092b ("arch_topology: Build cacheinfo from primary CPU")
adds functionality that architectures can use to optionally allocate and
build cacheinfo early during boot. Commit
6539cffa94 ("cacheinfo: Add arch specific early level initializer")
lets secondary CPUs correct (and reallocate memory) cacheinfo data if
needed.
If the early build functionality is not used and cacheinfo does not need
correction, memory for cacheinfo is never allocated. x86 does not use
the early build functionality. Consequently, during the cacheinfo CPU
hotplug callback, last_level_cache_is_valid() attempts to dereference
a NULL pointer:
BUG: kernel NULL pointer dereference, address: 0000000000000100
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not present page
PGD 0 P4D 0
Oops: 0000 [#1] PREEPMT SMP NOPTI
CPU: 0 PID 19 Comm: cpuhp/0 Not tainted 6.4.0-rc2 #1
RIP: 0010: last_level_cache_is_valid+0x95/0xe0a
Allocate memory for cacheinfo during the cacheinfo CPU hotplug callback
if not done earlier.
Moreover, before determining the validity of the last-level cache info,
ensure that it has been allocated. Simply checking for non-zero
cache_leaves() is not sufficient, as some architectures (e.g., Intel
processors) have non-zero cache_leaves() before allocation.
Dereferencing NULL cacheinfo can occur in update_per_cpu_data_slice_size().
This function iterates over all online CPUs. However, a CPU may have come
online recently, but its cacheinfo may not have been allocated yet.
While here, remove an unnecessary indentation in allocate_cache_info().
[ bp: Massage. ]
Fixes: 6539cffa94 ("cacheinfo: Add arch specific early level initializer")
Signed-off-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: Radu Rendec <rrendec@redhat.com>
Reviewed-by: Nikolay Borisov <nik.borisov@suse.com>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Reviewed-by: Sudeep Holla <sudeep.holla@arm.com>
Cc: stable@vger.kernel.org # 6.3+
Link: https://lore.kernel.org/r/20241128002247.26726-2-ricardo.neri-calderon@linux.intel.com
Introduce the __free attribute for scope-based resource management.
Resources allocated with __free are automatically released at the end of
the scope. This enhancement aims to mitigate memory management issues
associated with forgetting to release resources by utilizing __free
instead of of_node_put().
To introduce this feature, some modifications to the code structure were
necessary. The original pattern:
```
prev = np;
while(...) {
[...]
np = of_find_next_cache_node(np);
of_node_put(prev);
prev = np;
[...]
}
```
has been updated to:
```
while(...) {
[...]
struct device_node __free(device_node) *prev = np;
np = of_find_next_cache_node(np)
[...]
}
```
With this change, the previous node is automatically cleaned up at the end
of each iteration, allowing the elimination of all of_node_put() calls and
some goto statements.
Suggested-by: Julia Lawall <julia.lawall@inria.fr>
Signed-off-by: Vincenzo Mezzela <vincenzo.mezzela@gmail.com>
Link: https://lore.kernel.org/r/20240719151335.869145-1-vincenzo.mezzela@gmail.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
For each CPU hotplug event, we will update per-CPU data slice size and
corresponding PCP configuration for every online CPU to make the
implementation simple. But, Kyle reported that this takes tens seconds
during boot on a machine with 34 zones and 3840 CPUs.
So, in this patch, for each CPU hotplug event, we only update per-CPU data
slice size and corresponding PCP configuration for the CPUs that share
caches with the hotplugged CPU. With the patch, the system boot time
reduces 67 seconds on the machine.
Link: https://lkml.kernel.org/r/20240126081944.414520-1-ying.huang@intel.com
Fixes: 362d37a106 ("mm, pcp: reduce lock contention for draining high-order pages")
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Originally-by: Kyle Meyer <kyle.meyer@hpe.com>
Reported-and-tested-by: Kyle Meyer <kyle.meyer@hpe.com>
Cc: Sudeep Holla <sudeep.holla@arm.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
In commit f26b3fa046 ("mm/page_alloc: limit number of high-order pages
on PCP during bulk free"), the PCP (Per-CPU Pageset) will be drained when
PCP is mostly used for high-order pages freeing to improve the cache-hot
pages reusing between page allocating and freeing CPUs.
On system with small per-CPU data cache slice, pages shouldn't be cached
before draining to guarantee cache-hot. But on a system with large
per-CPU data cache slice, some pages can be cached before draining to
reduce zone lock contention.
So, in this patch, instead of draining without any caching, "pcp->batch"
pages will be cached in PCP before draining if the size of the per-CPU
data cache slice is more than "3 * batch".
In theory, if the size of per-CPU data cache slice is more than "2 *
batch", we can reuse cache-hot pages between CPUs. But considering the
other usage of cache (code, other data accessing, etc.), "3 * batch" is
used.
Note: "3 * batch" is chosen to make sure the optimization works on recent
x86_64 server CPUs. If you want to increase it, please check whether it
breaks the optimization.
On a 2-socket Intel server with 128 logical CPU, with the patch, the
network bandwidth of the UNIX (AF_UNIX) test case of lmbench test suite
with 16-pair processes increase 70.5%. The cycles% of the spinlock
contention (mostly for zone lock) decreases from 46.1% to 21.3%. The
number of PCP draining for high order pages freeing (free_high) decreases
89.9%. The cache miss rate keeps 0.2%.
Link: https://lkml.kernel.org/r/20231016053002.756205-4-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Sudeep Holla <sudeep.holla@arm.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: David Hildenbrand <david@redhat.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Until commit 5c2712387d ("cacheinfo: Fix LLC is not exported through
sysfs"), cacheinfo called populate_cache_leaves() for CPU coming online
which let the arch specific functions handle (at least on x86)
populating the shared_cpu_map. However, with the changes in the
aforementioned commit, populate_cache_leaves() is not called when a CPU
comes online as a result of hotplug since last_level_cache_is_valid()
returns true as the cacheinfo data is not discarded. The CPU coming
online is not present in shared_cpu_map, however, it will not be added
since the cpu_cacheinfo->cpu_map_populated flag is set (it is set in
populate_cache_leaves() when cacheinfo is first populated for x86)
This can lead to inconsistencies in the shared_cpu_map when an offlined
CPU comes online again. Example below depicts the inconsistency in the
shared_cpu_list in cacheinfo when CPU8 is offlined and onlined again on
a 3rd Generation EPYC processor:
# for i in /sys/devices/system/cpu/cpu8/cache/index*/shared_cpu_list; do echo -n "$i: "; cat $i; done
/sys/devices/system/cpu/cpu8/cache/index0/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu8/cache/index1/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu8/cache/index2/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu8/cache/index3/shared_cpu_list: 8-15,136-143
# echo 0 > /sys/devices/system/cpu/cpu8/online
# echo 1 > /sys/devices/system/cpu/cpu8/online
# for i in /sys/devices/system/cpu/cpu8/cache/index*/shared_cpu_list; do echo -n "$i: "; cat $i; done
/sys/devices/system/cpu/cpu8/cache/index0/shared_cpu_list: 8
/sys/devices/system/cpu/cpu8/cache/index1/shared_cpu_list: 8
/sys/devices/system/cpu/cpu8/cache/index2/shared_cpu_list: 8
/sys/devices/system/cpu/cpu8/cache/index3/shared_cpu_list: 8
# cat /sys/devices/system/cpu/cpu136/cache/index0/shared_cpu_list
136
# cat /sys/devices/system/cpu/cpu136/cache/index3/shared_cpu_list
9-15,136-143
Clear the flag when the CPU is removed from shared_cpu_map when
cache_shared_cpu_map_remove() is called during CPU hotplug. This will
allow cache_shared_cpu_map_setup() to add the CPU coming back online in
the shared_cpu_map. Set the flag again when the shared_cpu_map is setup.
Following are results of performing the same test as described above with
the changes:
# for i in /sys/devices/system/cpu/cpu8/cache/index*/shared_cpu_list; do echo -n "$i: "; cat $i; done
/sys/devices/system/cpu/cpu8/cache/index0/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu8/cache/index1/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu8/cache/index2/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu8/cache/index3/shared_cpu_list: 8-15,136-143
# echo 0 > /sys/devices/system/cpu/cpu8/online
# echo 1 > /sys/devices/system/cpu/cpu8/online
# for i in /sys/devices/system/cpu/cpu8/cache/index*/shared_cpu_list; do echo -n "$i: "; cat $i; done
/sys/devices/system/cpu/cpu8/cache/index0/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu8/cache/index1/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu8/cache/index2/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu8/cache/index3/shared_cpu_list: 8-15,136-143
# cat /sys/devices/system/cpu/cpu136/cache/index0/shared_cpu_list
8,136
# cat /sys/devices/system/cpu/cpu136/cache/index3/shared_cpu_list
8-15,136-143
Fixes: 5c2712387d ("cacheinfo: Fix LLC is not exported through sysfs")
Signed-off-by: K Prateek Nayak <kprateek.nayak@amd.com>
Reviewed-by: Yicong Yang <yangyicong@hisilicon.com>
Reviewed-by: Sudeep Holla <sudeep.holla@arm.com>
Link: https://lore.kernel.org/r/20230508084115.1157-3-kprateek.nayak@amd.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
While building the shared_cpu_map, check if the cache level and cache
type matches. On certain systems that build the cache topology based on
the instance ID, there are cases where the same ID may repeat across
multiple cache levels, leading inaccurate topology.
In event of CPU offlining, the cache_shared_cpu_map_remove() does not
consider if IDs at same level are being compared. As a result, when same
IDs repeat across different cache levels, the CPU going offline is not
removed from all the shared_cpu_map.
Below is the output of cache topology of CPU8 and it's SMT sibling after
CPU8 is offlined on a dual socket 3rd Generation AMD EPYC processor
(2 x 64C/128T) running kernel release v6.3:
# for i in /sys/devices/system/cpu/cpu8/cache/index*/shared_cpu_list; do echo -n "$i: "; cat $i; done
/sys/devices/system/cpu/cpu8/cache/index0/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu8/cache/index1/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu8/cache/index2/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu8/cache/index3/shared_cpu_list: 8-15,136-143
# echo 0 > /sys/devices/system/cpu/cpu8/online
# for i in /sys/devices/system/cpu/cpu136/cache/index*/shared_cpu_list; do echo -n "$i: "; cat $i; done
/sys/devices/system/cpu/cpu136/cache/index0/shared_cpu_list: 136
/sys/devices/system/cpu/cpu136/cache/index1/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu136/cache/index2/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu136/cache/index3/shared_cpu_list: 9-15,136-143
CPU8 is removed from index0 (L1i) but remains in the shared_cpu_list of
index1 (L1d) and index2 (L2). Since L1i, L1d, and L2 are shared by the
SMT siblings, and they have the same cache instance ID, CPU 2 is only
removed from the first index with matching ID which is index1 (L1i) in
this case. With this fix, the results are as expected when performing
the same experiment on the same system:
# for i in /sys/devices/system/cpu/cpu8/cache/index*/shared_cpu_list; do echo -n "$i: "; cat $i; done
/sys/devices/system/cpu/cpu8/cache/index0/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu8/cache/index1/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu8/cache/index2/shared_cpu_list: 8,136
/sys/devices/system/cpu/cpu8/cache/index3/shared_cpu_list: 8-15,136-143
# echo 0 > /sys/devices/system/cpu/cpu8/online
# for i in /sys/devices/system/cpu/cpu136/cache/index*/shared_cpu_list; do echo -n "$i: "; cat $i; done
/sys/devices/system/cpu/cpu136/cache/index0/shared_cpu_list: 136
/sys/devices/system/cpu/cpu136/cache/index1/shared_cpu_list: 136
/sys/devices/system/cpu/cpu136/cache/index2/shared_cpu_list: 136
/sys/devices/system/cpu/cpu136/cache/index3/shared_cpu_list: 9-15,136-143
When rebuilding topology, the same problem appears as
cache_shared_cpu_map_setup() implements a similar logic. Consider the
same 3rd Generation EPYC processor: CPUs in Core 1, that share the L1
and L2 caches, have L1 and L2 instance ID as 1. For all the CPUs on
the second chiplet, the L3 ID is also 1 leading to grouping on CPUs from
Core 1 (1, 17) and the entire second chiplet (8-15, 24-31) as CPUs
sharing one cache domain. This went undetected since x86 processors
depended on arch specific populate_cache_leaves() method to repopulate
the shared_cpus_map when CPU came back online until kernel release
v6.3-rc5.
Fixes: 198102c910 ("cacheinfo: Fix shared_cpu_map to handle shared caches at different levels")
Signed-off-by: K Prateek Nayak <kprateek.nayak@amd.com>
Reviewed-by: Sudeep Holla <sudeep.holla@arm.com>
Link: https://lore.kernel.org/r/20230508084115.1157-2-kprateek.nayak@amd.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Pull driver core updates from Greg KH:
"Here is the large set of driver core changes for 6.4-rc1.
Once again, a busy development cycle, with lots of changes happening
in the driver core in the quest to be able to move "struct bus" and
"struct class" into read-only memory, a task now complete with these
changes.
This will make the future rust interactions with the driver core more
"provably correct" as well as providing more obvious lifetime rules
for all busses and classes in the kernel.
The changes required for this did touch many individual classes and
busses as many callbacks were changed to take const * parameters
instead. All of these changes have been submitted to the various
subsystem maintainers, giving them plenty of time to review, and most
of them actually did so.
Other than those changes, included in here are a small set of other
things:
- kobject logging improvements
- cacheinfo improvements and updates
- obligatory fw_devlink updates and fixes
- documentation updates
- device property cleanups and const * changes
- firwmare loader dependency fixes.
All of these have been in linux-next for a while with no reported
problems"
* tag 'driver-core-6.4-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core: (120 commits)
device property: make device_property functions take const device *
driver core: update comments in device_rename()
driver core: Don't require dynamic_debug for initcall_debug probe timing
firmware_loader: rework crypto dependencies
firmware_loader: Strip off \n from customized path
zram: fix up permission for the hot_add sysfs file
cacheinfo: Add use_arch[|_cache]_info field/function
arch_topology: Remove early cacheinfo error message if -ENOENT
cacheinfo: Check cache properties are present in DT
cacheinfo: Check sib_leaf in cache_leaves_are_shared()
cacheinfo: Allow early level detection when DT/ACPI info is missing/broken
cacheinfo: Add arm64 early level initializer implementation
cacheinfo: Add arch specific early level initializer
tty: make tty_class a static const structure
driver core: class: remove struct class_interface * from callbacks
driver core: class: mark the struct class in struct class_interface constant
driver core: class: make class_register() take a const *
driver core: class: mark class_release() as taking a const *
driver core: remove incorrect comment for device_create*
MIPS: vpe-cmp: remove module owner pointer from struct class usage.
...
Pull more devicetree updates from Rob Herring:
- First part of DT header detangling dropping cpu.h from of_device.h
and replacing some includes with forward declarations. A handful of
drivers needed some adjustment to their includes as a result.
- Refactor of_device.h to be used by bus drivers rather than various
device drivers. This moves non-bus related functions out of
of_device.h. The end goal is for of_platform.h and of_device.h to
stop including each other.
- Refactor open coded parsing of "ranges" in some bus drivers to use DT
address parsing functions
- Add some new address parsing functions of_property_read_reg(),
of_range_count(), and of_range_to_resource() in preparation to
convert more open coded parsing of DT addresses to use them.
- Treewide clean-ups to use of_property_read_bool() and
of_property_present() as appropriate. The ones here are the ones that
didn't get picked up elsewhere.
* tag 'devicetree-for-6.4-2' of git://git.kernel.org/pub/scm/linux/kernel/git/robh/linux: (34 commits)
bus: tegra-gmi: Replace of_platform.h with explicit includes
hte: Use of_property_present() for testing DT property presence
w1: w1-gpio: Use of_property_read_bool() for boolean properties
virt: fsl: Use of_property_present() for testing DT property presence
soc: fsl: Use of_property_present() for testing DT property presence
sbus: display7seg: Use of_property_read_bool() for boolean properties
sparc: Use of_property_read_bool() for boolean properties
sparc: Use of_property_present() for testing DT property presence
bus: mvebu-mbus: Remove open coded "ranges" parsing
of/address: Add of_property_read_reg() helper
of/address: Add of_range_count() helper
of/address: Add support for 3 address cell bus
of/address: Add of_range_to_resource() helper
of: unittest: Add bus address range parsing tests
of: Drop cpu.h include from of_device.h
OPP: Adjust includes to remove of_device.h
irqchip: loongson-eiointc: Add explicit include for cpuhotplug.h
cpuidle: Adjust includes to remove of_device.h
cpufreq: sun50i: Add explicit include for cpu.h
cpufreq: Adjust includes to remove of_device.h
...
Sudeep writes:
cacheinfo and arch_topology updates for v6.4
The cache information can be extracted from either a Device Tree(DT),
the PPTT ACPI table, or arch registers (clidr_el1 for arm64).
When the DT is used but no cache properties are advertised, the current
code doesn't correctly fallback to using arch information. The changes
fixes the same and also assuse the that L1 data/instruction caches
are private and L2/higher caches are shared when the cache information
is missing in DT/ACPI and is derived form clidr_el1/arch registers.
Currently the cacheinfo is built from the primary CPU prior to secondary
CPUs boot, if the DT/ACPI description contains cache information.
However, if not present, it still reverts to the old behavior, which
allocates the cacheinfo memory on each secondary CPUs which causes
RT kernels to triggers a "BUG: sleeping function called from invalid
context".
The changes here attempts to enable automatic detection for RT kernels
when no DT/ACPI cache information is available, by pre-allocating
cacheinfo memory on the primary CPU.
* tag 'cacheinfo-updates-6.4' of git://git.kernel.org/pub/scm/linux/kernel/git/sudeep.holla/linux:
cacheinfo: Add use_arch[|_cache]_info field/function
arch_topology: Remove early cacheinfo error message if -ENOENT
cacheinfo: Check cache properties are present in DT
cacheinfo: Check sib_leaf in cache_leaves_are_shared()
cacheinfo: Allow early level detection when DT/ACPI info is missing/broken
cacheinfo: Add arm64 early level initializer implementation
cacheinfo: Add arch specific early level initializer
The cache information can be extracted from either a Device
Tree (DT), the PPTT ACPI table, or arch registers (clidr_el1
for arm64).
The clidr_el1 register is used only if DT/ACPI information is not
available. It does not states how caches are shared among CPUs.
Add a use_arch_cache_info field/function to identify when the
DT/ACPI doesn't provide cache information. Use this information
to assume L1 caches are privates and L2 and higher are shared among
all CPUs.
Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
Link: https://lore.kernel.org/r/20230414081453.244787-5-pierre.gondois@arm.com
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
If a Device Tree (DT) is used, the presence of cache properties is
assumed. Not finding any is not considered. For arm64 platforms,
cache information can be fetched from the clidr_el1 register.
Checking whether cache information is available in the DT
allows to switch to using clidr_el1.
init_of_cache_level()
\-of_count_cache_leaves()
will assume there a 2 cache leaves (L1 data/instruction caches), which
can be different from clidr_el1 information.
cache_setup_of_node() tries to read cache properties in the DT.
If there are none, this is considered a success. Knowing no
information was available would allow to switch to using clidr_el1.
Fixes: de0df442ee ("cacheinfo: Check 'cache-unified' property to count cache leaves")
Reported-by: Alexandre Ghiti <alexghiti@rivosinc.com>
Link: https://lore.kernel.org/all/20230404-hatred-swimmer-6fecdf33b57a@spud/
Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Link: https://lore.kernel.org/r/20230414081453.244787-3-pierre.gondois@arm.com
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
If there is no ACPI/DT information, it is assumed that L1 caches
are private and L2 (and higher) caches are shared. A cache is
'shared' between two CPUs if it is accessible from these two
CPUs.
Each CPU owns a representation (i.e. has a dedicated cacheinfo struct)
of the caches it has access to. cache_leaves_are_shared() tries to
identify whether two representations are designating the same actual
cache.
In cache_leaves_are_shared(), if 'this_leaf' is a L2 cache (or higher)
and 'sib_leaf' is a L1 cache, the caches are detected as shared as
only this_leaf's cache level is checked.
This is leads to setting sib_leaf as being shared with another CPU,
which is incorrect as this is a L1 cache.
Check 'sib_leaf->level'. Also update the comment as the function is
called when populating 'shared_cpu_map'.
Fixes: f16d1becf9 ("cacheinfo: Use cache identifiers to check if the caches are shared if available")
Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Link: https://lore.kernel.org/r/20230414081453.244787-2-pierre.gondois@arm.com
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
This patch gives architecture specific code the ability to initialize
the cache level and allocate cacheinfo memory early, when cache level
initialization runs on the primary CPU for all possible CPUs.
This is part of a patch series that attempts to further the work in
commit 5944ce092b ("arch_topology: Build cacheinfo from primary CPU").
Previously, in the absence of any DT/ACPI cache info, architecture
specific cache detection and info allocation for secondary CPUs would
happen in non-preemptible context during early CPU initialization and
trigger a "BUG: sleeping function called from invalid context" splat on
an RT kernel.
More specifically, this patch adds the early_cache_level() function,
which is called by fetch_cache_info() as a fallback when the number of
cache leaves cannot be extracted from DT/ACPI. In the default generic
(weak) implementation, this new function returns -ENOENT, which
preserves the original behavior for architectures that do not implement
the function.
Since early detection can get the number of cache leaves wrong in some
cases*, additional logic is added to still call init_cache_level() later
on the secondary CPU, therefore giving the architecture specific code an
opportunity to go back and fix the initial guess. Again, the original
behavior is preserved for architectures that do not implement the new
function.
* For example, on arm64, CLIDR_EL1 detection works only when it runs on
the current CPU. In other words, a CPU cannot detect the cache depth
for any other CPU than itself.
Signed-off-by: Radu Rendec <rrendec@redhat.com>
Reviewed-by: Pierre Gondois <pierre.gondois@arm.com>
Link: https://lore.kernel.org/r/20230412185759.755408-2-rrendec@redhat.com
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
After entering 6.3-rc1 the LLC cacheinfo is not exported on our ACPI
based arm64 server. This is because the LLC cacheinfo is partly reset
when secondary CPUs boot up. On arm64 the primary cpu will allocate
and setup cacheinfo:
init_cpu_topology()
for_each_possible_cpu()
fetch_cache_info() // Allocate cacheinfo and init levels
detect_cache_attributes()
cache_shared_cpu_map_setup()
if (!last_level_cache_is_valid()) // not valid, setup LLC
cache_setup_properties() // setup LLC
On secondary CPU boot up:
detect_cache_attributes()
populate_cache_leaves()
get_cache_type() // Get cache type from clidr_el1,
// for LLC type=CACHE_TYPE_NOCACHE
cache_shared_cpu_map_setup()
if (!last_level_cache_is_valid()) // Valid and won't go to this branch,
// leave LLC's type=CACHE_TYPE_NOCACHE
The last_level_cache_is_valid() use cacheinfo->{attributes, fw_token} to
test it's valid or not, but populate_cache_leaves() will only reset
LLC's type, so we won't try to re-setup LLC's type and leave it
CACHE_TYPE_NOCACHE and won't export it through sysfs.
This patch tries to fix this by not re-populating the cache leaves if
the LLC is valid.
Fixes: 5944ce092b ("arch_topology: Build cacheinfo from primary CPU")
Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
Reviewed-by: Pierre Gondois <pierre.gondois@arm.com>
Reviewed-by: Sudeep Holla <sudeep.holla@arm.com>
Link: https://lore.kernel.org/r/20230328114915.33340-1-yangyicong@huawei.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The cacheinfo sets up the shared_cpu_map by checking whether the caches
with the same index are shared between CPUs. However, this will trigger
slab-out-of-bounds access if the CPUs do not have the same cache hierarchy.
Another problem is the mismatched shared_cpu_map when the shared cache does
not have the same index between CPUs.
CPU0 I D L3
index 0 1 2 x
^ ^ ^ ^
index 0 1 2 3
CPU1 I D L2 L3
This patch checks each cache is shared with all caches on other CPUs.
Reviewed-by: Pierre Gondois <pierre.gondois@arm.com>
Signed-off-by: Yong-Xuan Wang <yongxuan.wang@sifive.com>
Link: https://lore.kernel.org/r/20230117105133.4445-2-yongxuan.wang@sifive.com
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
commit 3fcbf1c77d ("arch_topology: Fix cache attributes detection
in the CPU hotplug path")
adds a call to detect_cache_attributes() to populate the cacheinfo
before updating the siblings mask. detect_cache_attributes() allocates
memory and can take the PPTT mutex (on ACPI platforms). On PREEMPT_RT
kernels, on secondary CPUs, this triggers a:
'BUG: sleeping function called from invalid context' [1]
as the code is executed with preemption and interrupts disabled.
The primary CPU was previously storing the cache information using
the now removed (struct cpu_topology).llc_id:
commit 5b8dc787ce ("arch_topology: Drop LLC identifier stash from
the CPU topology")
allocate_cache_info() tries to build the cacheinfo from the primary
CPU prior secondary CPUs boot, if the DT/ACPI description
contains cache information.
If allocate_cache_info() fails, then fallback to the current state
for the cacheinfo allocation. [1] will be triggered in such case.
When unplugging a CPU, the cacheinfo memory cannot be freed. If it
was, then the memory would be allocated early by the re-plugged
CPU and would trigger [1].
Note that populate_cache_leaves() might be called multiple times
due to populate_leaves being moved up. This is required since
detect_cache_attributes() might be called with per_cpu_cacheinfo(cpu)
being allocated but not populated.
[1]:
| BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:46
| in_atomic(): 1, irqs_disabled(): 128, non_block: 0, pid: 0, name: swapper/111
| preempt_count: 1, expected: 0
| RCU nest depth: 1, expected: 1
| 3 locks held by swapper/111/0:
| #0: (&pcp->lock){+.+.}-{3:3}, at: get_page_from_freelist+0x218/0x12c8
| #1: (rcu_read_lock){....}-{1:3}, at: rt_spin_trylock+0x48/0xf0
| #2: (&zone->lock){+.+.}-{3:3}, at: rmqueue_bulk+0x64/0xa80
| irq event stamp: 0
| hardirqs last enabled at (0): 0x0
| hardirqs last disabled at (0): copy_process+0x5dc/0x1ab8
| softirqs last enabled at (0): copy_process+0x5dc/0x1ab8
| softirqs last disabled at (0): 0x0
| Preemption disabled at:
| migrate_enable+0x30/0x130
| CPU: 111 PID: 0 Comm: swapper/111 Tainted: G W 6.0.0-rc4-rt6-[...]
| Call trace:
| __kmalloc+0xbc/0x1e8
| detect_cache_attributes+0x2d4/0x5f0
| update_siblings_masks+0x30/0x368
| store_cpu_topology+0x78/0xb8
| secondary_start_kernel+0xd0/0x198
| __secondary_switched+0xb0/0xb4
Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
Reviewed-by: Sudeep Holla <sudeep.holla@arm.com>
Acked-by: Palmer Dabbelt <palmer@rivosinc.com>
Link: https://lore.kernel.org/r/20230104183033.755668-7-pierre.gondois@arm.com
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
Make init_of_cache_level() return an error code when the cache
information parsing fails to help detecting missing information.
init_of_cache_level() is only called for riscv. Returning an error
code instead of 0 will prevent detect_cache_attributes() to allocate
memory if an incomplete DT is parsed.
Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
Acked-by: Palmer Dabbelt <palmer@rivosinc.com>
Link: https://lore.kernel.org/r/20230104183033.755668-3-pierre.gondois@arm.com
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
fw_token is used for DT/ACPI systems to identify CPUs sharing caches.
For DT based systems, fw_token is set to a pointer to a DT node.
commit 3da72e1837 ("cacheinfo: Decrement refcount in
cache_setup_of_node()")
doesn't increment the refcount of fw_token anymore in
cache_setup_of_node(). fw_token is indeed used as a token and not
as a (struct device_node*), so no reference to fw_token should be
kept.
However, [1] is triggered when hotplugging a CPU multiple times
since cache_shared_cpu_map_remove() decrements the refcount to
fw_token at each CPU unplugging, eventually reaching 0.
Remove of_node_put() for fw_token in cache_shared_cpu_map_remove().
[1]
------------[ cut here ]------------
refcount_t: saturated; leaking memory.
WARNING: CPU: 4 PID: 32 at lib/refcount.c:22 refcount_warn_saturate (lib/refcount.c:22 (discriminator 3))
Modules linked in:
CPU: 4 PID: 32 Comm: cpuhp/4 Tainted: G W 6.1.0-rc1-14091-g9fdf2ca7b9c8 #76
Hardware name: ARM LTD ARM Juno Development Platform/ARM Juno Development Platform, BIOS EDK II Oct 31 2022
pstate: 600000c5 (nZCv daIF -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : refcount_warn_saturate (lib/refcount.c:22 (discriminator 3))
lr : refcount_warn_saturate (lib/refcount.c:22 (discriminator 3))
[...]
Call trace:
[...]
of_node_release (drivers/of/dynamic.c:335)
kobject_put (lib/kobject.c:677 lib/kobject.c:704 ./include/linux/kref.h:65 lib/kobject.c:721)
of_node_put (drivers/of/dynamic.c:49)
free_cache_attributes.part.0 (drivers/base/cacheinfo.c:712)
cacheinfo_cpu_pre_down (drivers/base/cacheinfo.c:718)
cpuhp_invoke_callback (kernel/cpu.c:247 (discriminator 4))
cpuhp_thread_fun (kernel/cpu.c:785)
smpboot_thread_fn (kernel/smpboot.c:164 (discriminator 3))
kthread (kernel/kthread.c:376)
ret_from_fork (arch/arm64/kernel/entry.S:861)
---[ end trace 0000000000000000 ]---
Fixes: 3da72e1837 ("cacheinfo: Decrement refcount in cache_setup_of_node()")
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Reported-by: Marek Szyprowski <m.szyprowski@samsung.com>
Tested-by: Geert Uytterhoeven <geert+renesas@glider.be>
Tested-by: Sudeep Holla <sudeep.holla@arm.com>
Reviewed-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
Link: https://lore.kernel.org/r/20221116094958.2141072-1-pierre.gondois@arm.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
