]> Git Repo - linux.git/log
linux.git
9 months agoperf/x86/intel: Support Perfmon MSRs aliasing
Kan Liang [Wed, 26 Jun 2024 14:35:40 +0000 (07:35 -0700)]
perf/x86/intel: Support Perfmon MSRs aliasing

The architectural performance monitoring V6 supports a new range of
counters' MSRs in the 19xxH address range. They include all the GP
counter MSRs, the GP control MSRs, and the fixed counter MSRs.

The step between each sibling counter is 4. Add intel_pmu_addr_offset()
to calculate the correct offset.

Add fixedctr in struct x86_pmu to store the address of the fixed counter
0. It can be used to calculate the rest of the fixed counters.

The MSR address of the fixed counter control is not changed.

Signed-off-by: Kan Liang <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Reviewed-by: Andi Kleen <[email protected]>
Reviewed-by: Ian Rogers <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]
9 months agoperf/x86/intel: Support PERFEVTSEL extension
Kan Liang [Wed, 26 Jun 2024 14:35:39 +0000 (07:35 -0700)]
perf/x86/intel: Support PERFEVTSEL extension

Two new fields (the unit mask2, and the equal flag) are added in the
IA32_PERFEVTSELx MSRs. They can be enumerated by the CPUID.23H.0.EBX.

Update the config_mask in x86_pmu and x86_hybrid_pmu for the true layout
of the PERFEVTSEL.
Expose the new formats into sysfs if they are available. The umask
extension reuses the same format attr name "umask" as the previous
umask. Add umask2_show to determine/display the correct format
for the current machine.

Co-developed-by: Dapeng Mi <[email protected]>
Signed-off-by: Dapeng Mi <[email protected]>
Signed-off-by: Kan Liang <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Reviewed-by: Andi Kleen <[email protected]>
Reviewed-by: Ian Rogers <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]
9 months agoperf/x86: Add config_mask to represent EVENTSEL bitmask
Kan Liang [Wed, 26 Jun 2024 14:35:38 +0000 (07:35 -0700)]
perf/x86: Add config_mask to represent EVENTSEL bitmask

Different vendors may support different fields in EVENTSEL MSR, such as
Intel would introduce new fields umask2 and eq bits in EVENTSEL MSR
since Perfmon version 6. However, a fixed mask X86_RAW_EVENT_MASK is
used to filter the attr.config.

Introduce a new config_mask to record the real supported EVENTSEL
bitmask.
Only apply it to the existing code now. No functional change.

Co-developed-by: Dapeng Mi <[email protected]>
Signed-off-by: Dapeng Mi <[email protected]>
Signed-off-by: Kan Liang <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Reviewed-by: Andi Kleen <[email protected]>
Reviewed-by: Ian Rogers <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]
9 months agoperf/x86/intel: Support new data source for Lunar Lake
Kan Liang [Wed, 26 Jun 2024 14:35:37 +0000 (07:35 -0700)]
perf/x86/intel: Support new data source for Lunar Lake

A new PEBS data source format is introduced for the p-core of Lunar
Lake. The data source field is extended to 8 bits with new encodings.

A new layout is introduced into the union intel_x86_pebs_dse.
Introduce the lnl_latency_data() to parse the new format.
Enlarge the pebs_data_source[] accordingly to include new encodings.

Only the mem load and the mem store events can generate the data source.
Introduce INTEL_HYBRID_LDLAT_CONSTRAINT and
INTEL_HYBRID_STLAT_CONSTRAINT to mark them.

Add two new bits for the new cache-related data src, L2_MHB and MSC.
The L2_MHB is short for L2 Miss Handling Buffer, which is similar to
LFB (Line Fill Buffer), but to track the L2 Cache misses.
The MSC stands for the memory-side cache.

Signed-off-by: Kan Liang <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Reviewed-by: Andi Kleen <[email protected]>
Reviewed-by: Ian Rogers <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]
9 months agoperf/x86/intel: Rename model-specific pebs_latency_data functions
Kan Liang [Wed, 26 Jun 2024 14:35:36 +0000 (07:35 -0700)]
perf/x86/intel: Rename model-specific pebs_latency_data functions

The model-specific pebs_latency_data functions of ADL and MTL use the
"small" as a postfix to indicate the e-core. The postfix is too generic
for a model-specific function. It cannot provide useful information that
can directly map it to a specific uarch, which can facilitate the
development and maintenance.
Use the abbr of the uarch to rename the model-specific functions.

Suggested-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Kan Liang <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Reviewed-by: Ian Rogers <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]
9 months agoperf/x86: Add Lunar Lake and Arrow Lake support
Kan Liang [Wed, 26 Jun 2024 14:35:35 +0000 (07:35 -0700)]
perf/x86: Add Lunar Lake and Arrow Lake support

From PMU's perspective, Lunar Lake and Arrow Lake are similar to the
previous generation Meteor Lake. Both are hybrid platforms, with e-core
and p-core.

The key differences include:
- The e-core supports 3 new fixed counters
- The p-core supports an updated PEBS Data Source format
- More GP counters (Updated event constraint table)
- New Architectural performance monitoring V6
  (New Perfmon MSRs aliasing, umask2, eq).
- New PEBS format V6 (Counters Snapshotting group)
- New RDPMC metrics clear mode

The legacy features, the 3 new fixed counters and updated event
constraint table are enabled in this patch.

The new PEBS data source format, the architectural performance
monitoring V6, the PEBS format V6, and the new RDPMC metrics clear mode
are supported in the following patches.

Signed-off-by: Kan Liang <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Reviewed-by: Andi Kleen <[email protected]>
Reviewed-by: Ian Rogers <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]
9 months agoperf/x86: Support counter mask
Kan Liang [Wed, 26 Jun 2024 14:35:34 +0000 (07:35 -0700)]
perf/x86: Support counter mask

The current perf assumes that both GP and fixed counters are contiguous.
But it's not guaranteed on newer Intel platforms or in a virtualization
environment.

Use the counter mask to replace the number of counters for both GP and
the fixed counters. For the other ARCHs or old platforms which don't
support a counter mask, using GENMASK_ULL(num_counter - 1, 0) to
replace. There is no functional change for them.

The interface to KVM is not changed. The number of counters still be
passed to KVM. It can be updated later separately.

Signed-off-by: Kan Liang <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Reviewed-by: Andi Kleen <[email protected]>
Reviewed-by: Ian Rogers <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]
9 months agoperf/x86/intel: Support the PEBS event mask
Kan Liang [Wed, 26 Jun 2024 14:35:33 +0000 (07:35 -0700)]
perf/x86/intel: Support the PEBS event mask

The current perf assumes that the counters that support PEBS are
contiguous. But it's not guaranteed with the new leaf 0x23 introduced.
The counters are enumerated with a counter mask. There may be holes in
the counter mask for future platforms or in a virtualization
environment.

Store the PEBS event mask rather than the maximum number of PEBS
counters in the x86 PMU structures.

Signed-off-by: Kan Liang <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Reviewed-by: Andi Kleen <[email protected]>
Reviewed-by: Ian Rogers <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]
9 months agoperf/x86/intel/cstate: Add Lunarlake support
Zhang Rui [Fri, 28 Jun 2024 03:17:58 +0000 (11:17 +0800)]
perf/x86/intel/cstate: Add Lunarlake support

Compared with previous client platforms, PC8 is removed from Lunarlake.
It supports CC1/CC6/CC7 and PC2/PC3/PC6/PC10 residency counters.

Signed-off-by: Zhang Rui <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Reviewed-by: Kan Liang <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agoperf/x86/intel/cstate: Add Arrowlake support
Zhang Rui [Fri, 28 Jun 2024 03:17:57 +0000 (11:17 +0800)]
perf/x86/intel/cstate: Add Arrowlake support

Like Alderlake, Arrowlake supports CC1/CC6/CC7 and PC2/PC3/PC6/PC8/PC10.

Signed-off-by: Zhang Rui <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Reviewed-by: Kan Liang <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agoperf/x86/intel/cstate: Fix Alderlake/Raptorlake/Meteorlake
Zhang Rui [Fri, 28 Jun 2024 03:17:56 +0000 (11:17 +0800)]
perf/x86/intel/cstate: Fix Alderlake/Raptorlake/Meteorlake

For Alderlake, the spec changes after the patch submitted and PC7/PC9
are removed.

Raptorlake and Meteorlake, which copy the Alderlake cstate PMU, also
don't have PC7/PC9.

Remove PC7/PC9 support for Alderlake/Raptorlake/Meteorlake.

Fixes: d0ca946bcf84 ("perf/x86/cstate: Add Alder Lake CPU support")
Signed-off-by: Zhang Rui <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Reviewed-by: Kan Liang <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agoMerge branch 'tip/x86/cpu'
Peter Zijlstra [Thu, 4 Jul 2024 14:00:24 +0000 (16:00 +0200)]
Merge branch 'tip/x86/cpu'

The Lunarlake patches rely on the new VFM stuff.

Signed-off-by: Peter Zijlstra <[email protected]>
9 months agoperf: Make rb_alloc_aux() return an error immediately if nr_pages <= 0
Adrian Hunter [Mon, 24 Jun 2024 20:11:01 +0000 (23:11 +0300)]
perf: Make rb_alloc_aux() return an error immediately if nr_pages <= 0

rb_alloc_aux() should not be called with nr_pages <= 0. Make it more robust
and readable by returning an error immediately in that case.

Signed-off-by: Adrian Hunter <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agoperf: Fix default aux_watermark calculation
Adrian Hunter [Mon, 24 Jun 2024 20:11:00 +0000 (23:11 +0300)]
perf: Fix default aux_watermark calculation

The default aux_watermark is half the AUX area buffer size. In general,
on a 64-bit architecture, the AUX area buffer size could be a bigger than
fits in a 32-bit type, but the calculation does not allow for that
possibility.

However the aux_watermark value is recorded in a u32, so should not be
more than U32_MAX either.

Fix by doing the calculation in a correctly sized type, and limiting the
result to U32_MAX.

Fixes: d68e6799a5c8 ("perf: Cap allocation order at aux_watermark")
Signed-off-by: Adrian Hunter <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agoperf: Prevent passing zero nr_pages to rb_alloc_aux()
Adrian Hunter [Mon, 24 Jun 2024 20:10:59 +0000 (23:10 +0300)]
perf: Prevent passing zero nr_pages to rb_alloc_aux()

nr_pages is unsigned long but gets passed to rb_alloc_aux() as an int,
and is stored as an int.

Only power-of-2 values are accepted, so if nr_pages is a 64_bit value, it
will be passed to rb_alloc_aux() as zero.

That is not ideal because:
 1. the value is incorrect
 2. rb_alloc_aux() is at risk of misbehaving, although it manages to
 return -ENOMEM in that case, it is a result of passing zero to get_order()
 even though the get_order() result is documented to be undefined in that
 case.

Fix by simply validating the maximum supported value in the first place.
Use -ENOMEM error code for consistency with the current error code that
is returned in that case.

Fixes: 45bfb2e50471 ("perf: Add AUX area to ring buffer for raw data streams")
Signed-off-by: Adrian Hunter <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agoperf: Fix perf_aux_size() for greater-than 32-bit size
Adrian Hunter [Mon, 24 Jun 2024 20:10:58 +0000 (23:10 +0300)]
perf: Fix perf_aux_size() for greater-than 32-bit size

perf_buffer->aux_nr_pages uses a 32-bit type, so a cast is needed to
calculate a 64-bit size.

Fixes: 45bfb2e50471 ("perf: Add AUX area to ring buffer for raw data streams")
Signed-off-by: Adrian Hunter <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agoperf/x86/intel/pt: Fix pt_topa_entry_for_page() address calculation
Adrian Hunter [Mon, 24 Jun 2024 20:10:57 +0000 (23:10 +0300)]
perf/x86/intel/pt: Fix pt_topa_entry_for_page() address calculation

Currently, perf allocates an array of page pointers which is limited in
size by MAX_PAGE_ORDER. That in turn limits the maximum Intel PT buffer
size to 2GiB. Should that limitation be lifted, the Intel PT driver can
support larger sizes, except for one calculation in
pt_topa_entry_for_page(), which is limited to 32-bits.

Fix pt_topa_entry_for_page() address calculation by adding a cast.

Fixes: 39152ee51b77 ("perf/x86/intel/pt: Get rid of reverse lookup table for ToPA")
Signed-off-by: Adrian Hunter <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agoperf/x86/intel/pt: Fix a topa_entry base address calculation
Adrian Hunter [Mon, 24 Jun 2024 20:10:56 +0000 (23:10 +0300)]
perf/x86/intel/pt: Fix a topa_entry base address calculation

topa_entry->base is a bit-field. Bit-fields are not promoted to a 64-bit
type, even if the underlying type is 64-bit, and so, if necessary, must
be cast to a larger type when calculations are done.

Fix a topa_entry->base address calculation by adding a cast.

Without the cast, the address was limited to 36-bits i.e. 64GiB.

The address calculation is used on systems that do not support Multiple
Entry ToPA (only Broadwell), and affects physical addresses on or above
64GiB. Instead of writing to the correct address, the address comprising
the first 36 bits would be written to.

Intel PT snapshot and sampling modes are not affected.

Fixes: 52ca9ced3f70 ("perf/x86/intel/pt: Add Intel PT PMU driver")
Reported-by: Dave Hansen <[email protected]>
Signed-off-by: Adrian Hunter <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Cc: [email protected]
Link: https://lore.kernel.org/r/[email protected]
9 months agoperf/x86/intel/pt: Fix topa_entry base length
Marco Cavenati [Mon, 24 Jun 2024 20:10:55 +0000 (23:10 +0300)]
perf/x86/intel/pt: Fix topa_entry base length

topa_entry->base needs to store a pfn.  It obviously needs to be
large enough to store the largest possible x86 pfn which is
MAXPHYADDR-PAGE_SIZE (52-12).  So it is 4 bits too small.

Increase the size of topa_entry->base from 36 bits to 40 bits.

Note, systems where physical addresses can be 256TiB or more are affected.

[ Adrian: Amend commit message as suggested by Dave Hansen ]

Fixes: 52ca9ced3f70 ("perf/x86/intel/pt: Add Intel PT PMU driver")
Signed-off-by: Marco Cavenati <[email protected]>
Signed-off-by: Adrian Hunter <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Reviewed-by: Adrian Hunter <[email protected]>
Cc: [email protected]
Link: https://lore.kernel.org/r/[email protected]
9 months agox86/cpu/intel: Drop stray FAM6 check with new Intel CPU model defines
Andrew Cooper [Wed, 29 May 2024 18:36:05 +0000 (19:36 +0100)]
x86/cpu/intel: Drop stray FAM6 check with new Intel CPU model defines

The outer if () should have been dropped when switching to c->x86_vfm.

Fixes: 6568fc18c2f6 ("x86/cpu/intel: Switch to new Intel CPU model defines")
Signed-off-by: Andrew Cooper <[email protected]>
Signed-off-by: Borislav Petkov (AMD) <[email protected]>
Acked-by: Tony Luck <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agox86/cpufeatures: Flip the /proc/cpuinfo appearance logic
Borislav Petkov (AMD) [Thu, 20 Jun 2024 18:47:46 +0000 (20:47 +0200)]
x86/cpufeatures: Flip the /proc/cpuinfo appearance logic

I'm getting tired of telling people to put a magic "" in the

  #define X86_FEATURE /* "" ... */

comment to hide the new feature flag from the user-visible
/proc/cpuinfo.

Flip the logic to make it explicit: an explicit "<name>" in the comment
adds the flag to /proc/cpuinfo and otherwise not, by default.

Add the "<name>" of all the existing flags to keep backwards
compatibility with userspace.

There should be no functional changes resulting from this.

Acked-by: Dave Hansen <[email protected]>
Signed-off-by: Borislav Petkov (AMD) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agoperf/x86/intel/uncore: Support HBM and CXL PMON counters
Kan Liang [Fri, 14 Jun 2024 13:46:31 +0000 (06:46 -0700)]
perf/x86/intel/uncore: Support HBM and CXL PMON counters

Unknown uncore PMON types can be found in both SPR and EMR with HBM or
CXL.

 $ls /sys/devices/ | grep type
 uncore_type_12_16
 uncore_type_12_18
 uncore_type_12_2
 uncore_type_12_4
 uncore_type_12_6
 uncore_type_12_8
 uncore_type_13_17
 uncore_type_13_19
 uncore_type_13_3
 uncore_type_13_5
 uncore_type_13_7
 uncore_type_13_9

The unknown PMON types are HBM and CXL PMON. Except for the name, the
other information regarding the HBM and CXL PMON counters can be
retrieved via the discovery table. Add them into the uncores tables for
SPR and EMR.

The event config registers for all CXL related units are 8-byte apart.
Add SPR_UNCORE_MMIO_OFFS8_COMMON_FORMAT to specially handle it.

Signed-off-by: Kan Liang <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Tested-by: Yunying Sun <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agoperf/x86/uncore: Cleanup unused unit structure
Kan Liang [Fri, 14 Jun 2024 13:46:30 +0000 (06:46 -0700)]
perf/x86/uncore: Cleanup unused unit structure

The unit control and ID information are retrieved from the unit control
RB tree. No one uses the old structure anymore. Remove them.

Signed-off-by: Kan Liang <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Tested-by: Yunying Sun <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agoperf/x86/uncore: Apply the unit control RB tree to PCI uncore units
Kan Liang [Fri, 14 Jun 2024 13:46:29 +0000 (06:46 -0700)]
perf/x86/uncore: Apply the unit control RB tree to PCI uncore units

The unit control RB tree has the unit control and unit ID information
for all the PCI units. Use them to replace the box_ctls/pci_offsets to
get an accurate unit control address for PCI uncore units.

The UPI/M3UPI units in the discovery table are ignored. Please see the
commit 65248a9a9ee1 ("perf/x86/uncore: Add a quirk for UPI on SPR").
Manually allocate a unit control RB tree for UPI/M3UPI.
Add cleanup_extra_boxes to release such manual allocation.

Signed-off-by: Kan Liang <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Tested-by: Yunying Sun <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agoperf/x86/uncore: Apply the unit control RB tree to MSR uncore units
Kan Liang [Fri, 14 Jun 2024 13:46:28 +0000 (06:46 -0700)]
perf/x86/uncore: Apply the unit control RB tree to MSR uncore units

The unit control RB tree has the unit control and unit ID information
for all the MSR units. Use them to replace the box_ctl and
uncore_msr_box_ctl() to get an accurate unit control address for MSR
uncore units.

Add intel_generic_uncore_assign_hw_event(), which utilizes the accurate
unit control address from the unit control RB tree to calculate the
config_base and event_base.

The unit id related information should be retrieved from the unit
control RB tree as well.

Signed-off-by: Kan Liang <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Tested-by: Yunying Sun <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agoperf/x86/uncore: Apply the unit control RB tree to MMIO uncore units
Kan Liang [Fri, 14 Jun 2024 13:46:27 +0000 (06:46 -0700)]
perf/x86/uncore: Apply the unit control RB tree to MMIO uncore units

The unit control RB tree has the unit control and unit ID information
for all the units. Use it to replace the box_ctls/mmio_offsets to get
an accurate unit control address for MMIO uncore units.

Signed-off-by: Kan Liang <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Tested-by: Yunying Sun <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agoperf/x86/uncore: Retrieve the unit ID from the unit control RB tree
Kan Liang [Fri, 14 Jun 2024 13:46:26 +0000 (06:46 -0700)]
perf/x86/uncore: Retrieve the unit ID from the unit control RB tree

The box_ids only save the unit ID for the first die. If a unit, e.g., a
CXL unit, doesn't exist in the first die. The unit ID cannot be
retrieved.

The unit control RB tree also stores the unit ID information.
Retrieve the unit ID from the unit control RB tree

Signed-off-by: Kan Liang <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Tested-by: Yunying Sun <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agoperf/x86/uncore: Support per PMU cpumask
Kan Liang [Fri, 14 Jun 2024 13:46:25 +0000 (06:46 -0700)]
perf/x86/uncore: Support per PMU cpumask

The cpumask of some uncore units, e.g., CXL uncore units, may be wrong
under some configurations. Perf may access an uncore counter of a
non-existent uncore unit.

The uncore driver assumes that all uncore units are symmetric among
dies. A global cpumask is shared among all uncore PMUs. However, some
CXL uncore units may only be available on some dies.

A per PMU cpumask is introduced to track the CPU mask of this PMU.
The driver searches the unit control RB tree to check whether the PMU is
available on a given die, and updates the per PMU cpumask accordingly.

Signed-off-by: Kan Liang <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Tested-by: Yunying Sun <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agoperf/x86/uncore: Save the unit control address of all units
Kan Liang [Fri, 14 Jun 2024 13:46:24 +0000 (06:46 -0700)]
perf/x86/uncore: Save the unit control address of all units

The unit control address of some CXL units may be wrongly calculated
under some configuration on a EMR machine.

The current implementation only saves the unit control address of the
units from the first die, and the first unit of the rest of dies. Perf
assumed that the units from the other dies have the same offset as the
first die. So the unit control address of the rest of the units can be
calculated. However, the assumption is wrong, especially for the CXL
units.

Introduce an RB tree for each uncore type to save the unit control
address and three kinds of ID information (unit ID, PMU ID, and die ID)
for all units.
The unit ID is a physical ID of a unit.
The PMU ID is a logical ID assigned to a unit. The logical IDs start
from 0 and must be contiguous. The physical ID and the logical ID are
1:1 mapping. The units with the same physical ID in different dies share
the same PMU.
The die ID indicates which die a unit belongs to.

The RB tree can be searched by two different keys (unit ID or PMU ID +
die ID). During the RB tree setup, the unit ID is used as a key to look
up the RB tree. The perf can create/assign a proper PMU ID to the unit.
Later, after the RB tree is setup, PMU ID + die ID is used as a key to
look up the RB tree to fill the cpumask of a PMU. It's used more
frequently, so PMU ID + die ID is compared in the unit_less().
The uncore_find_unit() has to be O(N). But the RB tree setup only occurs
once during the driver load time. It should be acceptable.

Compared with the current implementation, more space is required to save
the information of all units. The extra size should be acceptable.
For example, on EMR, there are 221 units at most. For a 2-socket machine,
the extra space is ~6KB at most.

Signed-off-by: Kan Liang <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
9 months agox86/CPU/AMD: Always inline amd_clear_divider()
Mateusz Guzik [Thu, 13 Jun 2024 08:26:37 +0000 (10:26 +0200)]
x86/CPU/AMD: Always inline amd_clear_divider()

The routine is used on syscall exit and on non-AMD CPUs is guaranteed to
be empty.

It probably does not need to be a function call even on CPUs which do need the
mitigation.

  [ bp: Make sure it is always inlined so that noinstr marking works. ]

Signed-off-by: Mateusz Guzik <[email protected]>
Signed-off-by: Borislav Petkov (AMD) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
10 months agox86/mce/inject: Add missing MODULE_DESCRIPTION() line
Jeff Johnson [Fri, 31 May 2024 00:20:20 +0000 (17:20 -0700)]
x86/mce/inject: Add missing MODULE_DESCRIPTION() line

make W=1 C=1 warns:

  WARNING: modpost: missing MODULE_DESCRIPTION() in arch/x86/kernel/cpu/mce/mce-inject.o

Add the missing MODULE_DESCRIPTION().

Signed-off-by: Jeff Johnson <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
10 months agoperf/x86/rapl: Switch to new Intel CPU model defines
Tony Luck [Mon, 20 May 2024 22:46:14 +0000 (15:46 -0700)]
perf/x86/rapl: Switch to new Intel CPU model defines

New CPU #defines encode vendor and family as well as model.

Signed-off-by: Tony Luck <[email protected]>
Signed-off-by: Dave Hansen <[email protected]>
Link: https://lore.kernel.org/all/20240520224620.9480-44-tony.luck%40intel.com
10 months agox86/boot: Switch to new Intel CPU model defines
Tony Luck [Mon, 20 May 2024 22:46:05 +0000 (15:46 -0700)]
x86/boot: Switch to new Intel CPU model defines

New CPU #defines encode vendor and family as well as model but
boot code doesn't have all the infrastructure to use them. Hard
code the one CPU model number used here.

Signed-off-by: Tony Luck <[email protected]>
Signed-off-by: Dave Hansen <[email protected]>
Link: https://lore.kernel.org/all/20240520224620.9480-35-tony.luck%40intel.com
10 months agox86/cpu: Switch to new Intel CPU model defines
Tony Luck [Mon, 20 May 2024 22:46:04 +0000 (15:46 -0700)]
x86/cpu: Switch to new Intel CPU model defines

New CPU #defines encode vendor and family as well as model.

Update INTEL_CPU_DESC() to work with vendor/family/model.

Signed-off-by: Tony Luck <[email protected]>
Signed-off-by: Dave Hansen <[email protected]>
Link: https://lore.kernel.org/all/20240520224620.9480-34-tony.luck%40intel.com
10 months agoperf/x86/intel: Switch to new Intel CPU model defines
Tony Luck [Mon, 20 May 2024 22:46:02 +0000 (15:46 -0700)]
perf/x86/intel: Switch to new Intel CPU model defines

New CPU #defines encode vendor and family as well as model.

Signed-off-by: Tony Luck <[email protected]>
Signed-off-by: Dave Hansen <[email protected]>
Link: https://lore.kernel.org/all/20240520224620.9480-32-tony.luck%40intel.com
10 months agox86/virt/tdx: Switch to new Intel CPU model defines
Tony Luck [Mon, 20 May 2024 22:46:01 +0000 (15:46 -0700)]
x86/virt/tdx: Switch to new Intel CPU model defines

New CPU #defines encode vendor and family as well as model.

Signed-off-by: Tony Luck <[email protected]>
Signed-off-by: Dave Hansen <[email protected]>
Link: https://lore.kernel.org/all/20240520224620.9480-31-tony.luck%40intel.com
10 months agox86/PCI: Switch to new Intel CPU model defines
Tony Luck [Mon, 20 May 2024 22:46:00 +0000 (15:46 -0700)]
x86/PCI: Switch to new Intel CPU model defines

New CPU #defines encode vendor and family as well as model.

Signed-off-by: Tony Luck <[email protected]>
Signed-off-by: Dave Hansen <[email protected]>
Link: https://lore.kernel.org/all/20240520224620.9480-30-tony.luck%40intel.com
10 months agox86/cpu/intel: Switch to new Intel CPU model defines
Tony Luck [Mon, 20 May 2024 22:45:59 +0000 (15:45 -0700)]
x86/cpu/intel: Switch to new Intel CPU model defines

New CPU #defines encode vendor and family as well as model.

Signed-off-by: Tony Luck <[email protected]>
Signed-off-by: Dave Hansen <[email protected]>
Link: https://lore.kernel.org/all/20240520224620.9480-29-tony.luck%40intel.com
10 months agox86/platform/intel-mid: Switch to new Intel CPU model defines
Tony Luck [Tue, 21 May 2024 16:10:01 +0000 (09:10 -0700)]
x86/platform/intel-mid: Switch to new Intel CPU model defines

New CPU #defines encode vendor and family as well as model.

N.B. Drop Haswell. CPU model 0x3C was included by mistake
in upstream code.

Signed-off-by: Tony Luck <[email protected]>
Signed-off-by: Dave Hansen <[email protected]>
Acked-by: Andy Shevchenko <[email protected]>
Link: https://lore.kernel.org/all/20240521161002.12866-1-tony.luck%40intel.com
10 months agox86/pconfig: Remove unused MKTME pconfig code
Alison Schofield [Tue, 7 May 2024 04:24:22 +0000 (21:24 -0700)]
x86/pconfig: Remove unused MKTME pconfig code

Code supporting Intel PCONFIG targets was an early piece of enabling
for MKTME (Multi-Key Total Memory Encryption).

Since MKTME feature enablement did not follow into the kernel, remove
the unused PCONFIG code.

Signed-off-by: Alison Schofield <[email protected]>
Signed-off-by: Dave Hansen <[email protected]>
Acked-by: Kirill A. Shutemov <[email protected]>
Acked-by: Kai Huang <[email protected]>
Link: https://lore.kernel.org/all/4ddff30d466785b4adb1400f0518783012835141.1715054189.git.alison.schofield%40intel.com
10 months agox86/cpu: Remove useless work in detect_tme_early()
Alison Schofield [Tue, 7 May 2024 04:24:21 +0000 (21:24 -0700)]
x86/cpu: Remove useless work in detect_tme_early()

TME (Total Memory Encryption) and MKTME (Multi-Key Total Memory
Encryption) BIOS detection were introduced together here [1] and
are loosely coupled in the Intel CPU init code.

TME is a hardware only feature and its BIOS status is all that needs
to be shared with the kernel user: enabled or disabled. The TME
algorithm the BIOS is using and whether or not the kernel recognizes
that algorithm is useless to the kernel user.

MKTME is a hardware feature that requires kernel support. MKTME
detection code was added in advance of broader kernel support for
MKTME that never followed. So, rather than continuing to spew
needless and confusing messages about BIOS MKTME status, remove
most of the MKTME pieces from detect_tme_early().

Keep one useful message: alert the user when BIOS enabled MKTME
reduces the available physical address bits. Recovery of the MKTME
consumed bits requires a reboot with MKTME disabled in BIOS.

There is no functional change for the user, only a change in boot
messages. Below is one example when both TME and MKTME are enabled
in BIOS with AES_XTS_256 which is unknown to the detect tme code.

Before:
[] x86/tme: enabled by BIOS
[] x86/tme: Unknown policy is active: 0x2
[] x86/mktme: No known encryption algorithm is supported: 0x4
[] x86/mktme: enabled by BIOS
[] x86/mktme: 127 KeyIDs available

After:
[] x86/tme: enabled by BIOS
[] x86/mktme: BIOS enable: x86_phys_bits reduced by 8

[1]
commit cb06d8e3d020 ("x86/tme: Detect if TME and MKTME is activated by BIOS")

Signed-off-by: Alison Schofield <[email protected]>
Signed-off-by: Dave Hansen <[email protected]>
Acked-by: Kirill A. Shutemov <[email protected]>
Link: https://lore.kernel.org/all/86dfdf6ced8c9b790f9376bf6c7e22b5608f47c2.1715054189.git.alison.schofield%40intel.com
10 months agoLinux 6.10-rc1 v6.10-rc1
Linus Torvalds [Sun, 26 May 2024 22:20:12 +0000 (15:20 -0700)]
Linux 6.10-rc1

10 months agomm: percpu: Include smp.h in alloc_tag.h
Kent Overstreet [Fri, 24 May 2024 15:42:09 +0000 (11:42 -0400)]
mm: percpu: Include smp.h in alloc_tag.h

percpu.h depends on smp.h, but doesn't include it directly because of
circular header dependency issues; percpu.h is needed in a bunch of low
level headers.

This fixes a randconfig build error on mips:

  include/linux/alloc_tag.h: In function '__alloc_tag_ref_set':
  include/asm-generic/percpu.h:31:40: error: implicit declaration of function 'raw_smp_processor_id' [-Werror=implicit-function-declaration]

Reported-by: kernel test robot <[email protected]>
Fixes: 24e44cc22aa3 ("mm: percpu: enable per-cpu allocation tagging")
Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/
Signed-off-by: Kent Overstreet <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
10 months agoMerge tag 'perf-tools-fixes-for-v6.10-1-2024-05-26' of git://git.kernel.org/pub/scm...
Linus Torvalds [Sun, 26 May 2024 16:54:26 +0000 (09:54 -0700)]
Merge tag 'perf-tools-fixes-for-v6.10-1-2024-05-26' of git://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools

Pull perf tool fix from Arnaldo Carvalho de Melo:
 "Revert a patch causing a regression.

  This made a simple 'perf record -e cycles:pp make -j199' stop working
  on the Ampere ARM64 system Linus uses to test ARM64 kernels".

* tag 'perf-tools-fixes-for-v6.10-1-2024-05-26' of git://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools:
  Revert "perf parse-events: Prefer sysfs/JSON hardware events over legacy"

10 months agoRevert "perf parse-events: Prefer sysfs/JSON hardware events over legacy"
Arnaldo Carvalho de Melo [Sun, 26 May 2024 11:13:21 +0000 (08:13 -0300)]
Revert "perf parse-events: Prefer sysfs/JSON hardware events over legacy"

This reverts commit 617824a7f0f73e4de325cf8add58e55b28c12493.

This made a simple 'perf record -e cycles:pp make -j199' stop working on
the Ampere ARM64 system Linus uses to test ARM64 kernels, as discussed
at length in the threads in the Link tags below.

The fix provided by Ian wasn't acceptable and work to fix this will take
time we don't have at this point, so lets revert this and work on it on
the next devel cycle.

Reported-by: Linus Torvalds <[email protected]>
Cc: Adrian Hunter <[email protected]>
Cc: Bhaskar Chowdhury <[email protected]>
Cc: Ethan Adams <[email protected]>
Cc: Ian Rogers <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: James Clark <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: Kan Liang <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Thomas Richter <[email protected]>
Cc: Tycho Andersen <[email protected]>
Cc: Yang Jihong <[email protected]>
Link: https://lore.kernel.org/lkml/CAHk-=wi5Ri=yR2jBVk-4HzTzpoAWOgstr1LEvg_-OXtJvXXJOA@mail.gmail.com
Link: https://lore.kernel.org/lkml/CAHk-=wiWvtFyedDNpoV7a8Fq_FpbB+F5KmWK2xPY3QoYseOf_A@mail.gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>
10 months agoMerge tag '6.10-rc-smb3-fixes-part2' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Sun, 26 May 2024 05:33:10 +0000 (22:33 -0700)]
Merge tag '6.10-rc-smb3-fixes-part2' of git://git.samba.org/sfrench/cifs-2.6

Pull smb client fixes from Steve French:

 - two important netfs integration fixes - including for a data
   corruption and also fixes for multiple xfstests

 - reenable swap support over SMB3

* tag '6.10-rc-smb3-fixes-part2' of git://git.samba.org/sfrench/cifs-2.6:
  cifs: Fix missing set of remote_i_size
  cifs: Fix smb3_insert_range() to move the zero_point
  cifs: update internal version number
  smb3: reenable swapfiles over SMB3 mounts

10 months agoMerge tag 'mm-hotfixes-stable-2024-05-25-09-13' of git://git.kernel.org/pub/scm/linux...
Linus Torvalds [Sat, 25 May 2024 22:10:33 +0000 (15:10 -0700)]
Merge tag 'mm-hotfixes-stable-2024-05-25-09-13' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm

Pull misc fixes from Andrew Morton:
 "16 hotfixes, 11 of which are cc:stable.

  A few nilfs2 fixes, the remainder are for MM: a couple of selftests
  fixes, various singletons fixing various issues in various parts"

* tag 'mm-hotfixes-stable-2024-05-25-09-13' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm:
  mm/ksm: fix possible UAF of stable_node
  mm/memory-failure: fix handling of dissolved but not taken off from buddy pages
  mm: /proc/pid/smaps_rollup: avoid skipping vma after getting mmap_lock again
  nilfs2: fix potential hang in nilfs_detach_log_writer()
  nilfs2: fix unexpected freezing of nilfs_segctor_sync()
  nilfs2: fix use-after-free of timer for log writer thread
  selftests/mm: fix build warnings on ppc64
  arm64: patching: fix handling of execmem addresses
  selftests/mm: compaction_test: fix bogus test success and reduce probability of OOM-killer invocation
  selftests/mm: compaction_test: fix incorrect write of zero to nr_hugepages
  selftests/mm: compaction_test: fix bogus test success on Aarch64
  mailmap: update email address for Satya Priya
  mm/huge_memory: don't unpoison huge_zero_folio
  kasan, fortify: properly rename memintrinsics
  lib: add version into /proc/allocinfo output
  mm/vmalloc: fix vmalloc which may return null if called with __GFP_NOFAIL

10 months agoMerge tag 'irq-urgent-2024-05-25' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 25 May 2024 21:48:40 +0000 (14:48 -0700)]
Merge tag 'irq-urgent-2024-05-25' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull irq fixes from Ingo Molnar:

 - Fix x86 IRQ vector leak caused by a CPU offlining race

 - Fix build failure in the riscv-imsic irqchip driver
   caused by an API-change semantic conflict

 - Fix use-after-free in irq_find_at_or_after()

* tag 'irq-urgent-2024-05-25' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  genirq/irqdesc: Prevent use-after-free in irq_find_at_or_after()
  genirq/cpuhotplug, x86/vector: Prevent vector leak during CPU offline
  irqchip/riscv-imsic: Fixup riscv_ipi_set_virq_range() conflict

10 months agoMerge tag 'x86-urgent-2024-05-25' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 25 May 2024 21:40:09 +0000 (14:40 -0700)]
Merge tag 'x86-urgent-2024-05-25' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull x86 fixes from Ingo Molnar:

 - Fix regressions of the new x86 CPU VFM (vendor/family/model)
   enumeration/matching code

 - Fix crash kernel detection on buggy firmware with
   non-compliant ACPI MADT tables

 - Address Kconfig warning

* tag 'x86-urgent-2024-05-25' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/cpu: Fix x86_match_cpu() to match just X86_VENDOR_INTEL
  crypto: x86/aes-xts - switch to new Intel CPU model defines
  x86/topology: Handle bogus ACPI tables correctly
  x86/kconfig: Select ARCH_WANT_FRAME_POINTERS again when UNWINDER_FRAME_POINTER=y

10 months agoMerge tag 'for-linus-6.10-1' of https://github.com/cminyard/linux-ipmi
Linus Torvalds [Sat, 25 May 2024 21:32:29 +0000 (14:32 -0700)]
Merge tag 'for-linus-6.10-1' of https://github.com/cminyard/linux-ipmi

Pull ipmi updates from Corey Minyard:
 "Mostly updates for deprecated interfaces, platform.remove and
  converting from a tasklet to a BH workqueue.

  Also use HAS_IOPORT for disabling inb()/outb()"

* tag 'for-linus-6.10-1' of https://github.com/cminyard/linux-ipmi:
  ipmi: kcs_bmc_npcm7xx: Convert to platform remove callback returning void
  ipmi: kcs_bmc_aspeed: Convert to platform remove callback returning void
  ipmi: ipmi_ssif: Convert to platform remove callback returning void
  ipmi: ipmi_si_platform: Convert to platform remove callback returning void
  ipmi: ipmi_powernv: Convert to platform remove callback returning void
  ipmi: bt-bmc: Convert to platform remove callback returning void
  char: ipmi: handle HAS_IOPORT dependencies
  ipmi: Convert from tasklet to BH workqueue

10 months agoMerge tag 'ceph-for-6.10-rc1' of https://github.com/ceph/ceph-client
Linus Torvalds [Sat, 25 May 2024 21:23:58 +0000 (14:23 -0700)]
Merge tag 'ceph-for-6.10-rc1' of https://github.com/ceph/ceph-client

Pull ceph updates from Ilya Dryomov:
 "A series from Xiubo that adds support for additional access checks
  based on MDS auth caps which were recently made available to clients.

  This is needed to prevent scenarios where the MDS quietly discards
  updates that a UID-restricted client previously (wrongfully) acked to
  the user.

  Other than that, just a documentation fixup"

* tag 'ceph-for-6.10-rc1' of https://github.com/ceph/ceph-client:
  doc: ceph: update userspace command to get CephFS metadata
  ceph: add CEPHFS_FEATURE_MDS_AUTH_CAPS_CHECK feature bit
  ceph: check the cephx mds auth access for async dirop
  ceph: check the cephx mds auth access for open
  ceph: check the cephx mds auth access for setattr
  ceph: add ceph_mds_check_access() helper
  ceph: save cap_auths in MDS client when session is opened

10 months agoMerge tag 'ntfs3_for_6.10' of https://github.com/Paragon-Software-Group/linux-ntfs3
Linus Torvalds [Sat, 25 May 2024 21:19:01 +0000 (14:19 -0700)]
Merge tag 'ntfs3_for_6.10' of https://github.com/Paragon-Software-Group/linux-ntfs3

Pull ntfs3 updates from Konstantin Komarov:
 "Fixes:
   - reusing of the file index (could cause the file to be trimmed)
   - infinite dir enumeration
   - taking DOS names into account during link counting
   - le32_to_cpu conversion, 32 bit overflow, NULL check
   - some code was refactored

  Changes:
   - removed max link count info display during driver init

  Remove:
   - atomic_open has been removed for lack of use"

* tag 'ntfs3_for_6.10' of https://github.com/Paragon-Software-Group/linux-ntfs3:
  fs/ntfs3: Break dir enumeration if directory contents error
  fs/ntfs3: Fix case when index is reused during tree transformation
  fs/ntfs3: Mark volume as dirty if xattr is broken
  fs/ntfs3: Always make file nonresident on fallocate call
  fs/ntfs3: Redesign ntfs_create_inode to return error code instead of inode
  fs/ntfs3: Use variable length array instead of fixed size
  fs/ntfs3: Use 64 bit variable to avoid 32 bit overflow
  fs/ntfs3: Check 'folio' pointer for NULL
  fs/ntfs3: Missed le32_to_cpu conversion
  fs/ntfs3: Remove max link count info display during driver init
  fs/ntfs3: Taking DOS names into account during link counting
  fs/ntfs3: remove atomic_open
  fs/ntfs3: use kcalloc() instead of kzalloc()

10 months agoMerge tag '6.10-rc-ksmbd-server-fixes' of git://git.samba.org/ksmbd
Linus Torvalds [Sat, 25 May 2024 21:15:39 +0000 (14:15 -0700)]
Merge tag '6.10-rc-ksmbd-server-fixes' of git://git.samba.org/ksmbd

Pull smb server fixes from Steve French:
 "Two ksmbd server fixes, both for stable"

* tag '6.10-rc-ksmbd-server-fixes' of git://git.samba.org/ksmbd:
  ksmbd: ignore trailing slashes in share paths
  ksmbd: avoid to send duplicate oplock break notifications

10 months agoMerge tag 'rtc-6.10' of git://git.kernel.org/pub/scm/linux/kernel/git/abelloni/linux
Linus Torvalds [Sat, 25 May 2024 20:33:53 +0000 (13:33 -0700)]
Merge tag 'rtc-6.10' of git://git.kernel.org/pub/scm/linux/kernel/git/abelloni/linux

Pull RTC updates from Alexandre Belloni:
 "There is one new driver and then most of the changes are the device
  tree bindings conversions to yaml.

  New driver:
   - Epson RX8111

  Drivers:
   - Many Device Tree bindings conversions to dtschema
   - pcf8563: wakeup-source support"

* tag 'rtc-6.10' of git://git.kernel.org/pub/scm/linux/kernel/git/abelloni/linux:
  pcf8563: add wakeup-source support
  rtc: rx8111: handle VLOW flag
  rtc: rx8111: demote warnings to debug level
  rtc: rx6110: Constify struct regmap_config
  dt-bindings: rtc: convert trivial devices into dtschema
  dt-bindings: rtc: stmp3xxx-rtc: convert to dtschema
  dt-bindings: rtc: pxa-rtc: convert to dtschema
  rtc: Add driver for Epson RX8111
  dt-bindings: rtc: Add Epson RX8111
  rtc: mcp795: drop unneeded MODULE_ALIAS
  rtc: nuvoton: Modify part number value
  rtc: test: Split rtc unit test into slow and normal speed test
  dt-bindings: rtc: nxp,lpc1788-rtc: convert to dtschema
  dt-bindings: rtc: digicolor-rtc: move to trivial-rtc
  dt-bindings: rtc: alphascale,asm9260-rtc: convert to dtschema
  dt-bindings: rtc: armada-380-rtc: convert to dtschema
  rtc: cros-ec: provide ID table for avoiding fallback match

10 months agoMerge tag 'i3c/for-6.10' of git://git.kernel.org/pub/scm/linux/kernel/git/i3c/linux
Linus Torvalds [Sat, 25 May 2024 20:28:29 +0000 (13:28 -0700)]
Merge tag 'i3c/for-6.10' of git://git.kernel.org/pub/scm/linux/kernel/git/i3c/linux

Pull i3c updates from Alexandre Belloni:
 "Runtime PM (power management) is improved and hot-join support has
  been added to the dw controller driver.

  Core:
   - Allow device driver to trigger controller runtime PM

  Drivers:
   - dw: hot-join support
   - svc: better IBI handling"

* tag 'i3c/for-6.10' of git://git.kernel.org/pub/scm/linux/kernel/git/i3c/linux:
  i3c: dw: Add hot-join support.
  i3c: master: Enable runtime PM for master controller
  i3c: master: svc: fix invalidate IBI type and miss call client IBI handler
  i3c: master: svc: change ENXIO to EAGAIN when IBI occurs during start frame
  i3c: Add comment for -EAGAIN in i3c_device_do_priv_xfers()

10 months agoMerge tag 'jffs2-for-linus-6.10-rc1' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 25 May 2024 20:23:42 +0000 (13:23 -0700)]
Merge tag 'jffs2-for-linus-6.10-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rw/ubifs

Pull jffs2 updates from Richard Weinberger:

 - Fix illegal memory access in jffs2_free_inode()

 - Kernel-doc fixes

 - print symbolic error names

* tag 'jffs2-for-linus-6.10-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rw/ubifs:
  jffs2: Fix potential illegal address access in jffs2_free_inode
  jffs2: Simplify the allocation of slab caches
  jffs2: nodemgmt: fix kernel-doc comments
  jffs2: print symbolic error name instead of error code

10 months agoMerge tag 'uml-for-linus-6.10-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 25 May 2024 20:17:48 +0000 (13:17 -0700)]
Merge tag 'uml-for-linus-6.10-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/uml/linux

Pull UML updates from Richard Weinberger:

 - Fixes for -Wmissing-prototypes warnings and further cleanup

 - Remove callback returning void from rtc and virtio drivers

 - Fix bash location

* tag 'uml-for-linus-6.10-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/uml/linux: (26 commits)
  um: virtio_uml: Convert to platform remove callback returning void
  um: rtc: Convert to platform remove callback returning void
  um: Remove unused do_get_thread_area function
  um: Fix -Wmissing-prototypes warnings for __vdso_*
  um: Add an internal header shared among the user code
  um: Fix the declaration of kasan_map_memory
  um: Fix the -Wmissing-prototypes warning for get_thread_reg
  um: Fix the -Wmissing-prototypes warning for __switch_mm
  um: Fix -Wmissing-prototypes warnings for (rt_)sigreturn
  um: Stop tracking host PID in cpu_tasks
  um: process: remove unused 'n' variable
  um: vector: remove unused len variable/calculation
  um: vector: fix bpfflash parameter evaluation
  um: slirp: remove set but unused variable 'pid'
  um: signal: move pid variable where needed
  um: Makefile: use bash from the environment
  um: Add winch to winch_handlers before registering winch IRQ
  um: Fix -Wmissing-prototypes warnings for __warp_* and foo
  um: Fix -Wmissing-prototypes warnings for text_poke*
  um: Move declarations to proper headers
  ...

10 months agoMerge tag 'drm-next-2024-05-25' of https://gitlab.freedesktop.org/drm/kernel
Linus Torvalds [Sat, 25 May 2024 00:28:02 +0000 (17:28 -0700)]
Merge tag 'drm-next-2024-05-25' of https://gitlab.freedesktop.org/drm/kernel

Pull drm fixes from Dave Airlie:
 "Some fixes for the end of the merge window, mostly amdgpu and panthor,
  with one nouveau uAPI change that fixes a bad decision we made a few
  months back.

  nouveau:
   - fix bo metadata uAPI for vm bind

  panthor:
   - Fixes for panthor's heap logical block.
   - Reset on unrecoverable fault
   - Fix VM references.
   - Reset fix.

  xlnx:
   - xlnx compile and doc fixes.

  amdgpu:
   - Handle vbios table integrated info v2.3

  amdkfd:
   - Handle duplicate BOs in reserve_bo_and_cond_vms
   - Handle memory limitations on small APUs

  dp/mst:
   - MST null deref fix.

  bridge:
   - Don't let next bridge create connector in adv7511 to make probe
     work"

* tag 'drm-next-2024-05-25' of https://gitlab.freedesktop.org/drm/kernel:
  drm/amdgpu/atomfirmware: add intergrated info v2.3 table
  drm/mst: Fix NULL pointer dereference at drm_dp_add_payload_part2
  drm/amdkfd: Let VRAM allocations go to GTT domain on small APUs
  drm/amdkfd: handle duplicate BOs in reserve_bo_and_cond_vms
  drm/bridge: adv7511: Attach next bridge without creating connector
  drm/buddy: Fix the warn on's during force merge
  drm/nouveau: use tile_mode and pte_kind for VM_BIND bo allocations
  drm/panthor: Call panthor_sched_post_reset() even if the reset failed
  drm/panthor: Reset the FW VM to NULL on unplug
  drm/panthor: Keep a ref to the VM at the panthor_kernel_bo level
  drm/panthor: Force an immediate reset on unrecoverable faults
  drm/panthor: Document drm_panthor_tiler_heap_destroy::handle validity constraints
  drm/panthor: Fix an off-by-one in the heap context retrieval logic
  drm/panthor: Relax the constraints on the tiler chunk size
  drm/panthor: Make sure the tiler initial/max chunks are consistent
  drm/panthor: Fix tiler OOM handling to allow incremental rendering
  drm: xlnx: zynqmp_dpsub: Fix compilation error
  drm: xlnx: zynqmp_dpsub: Fix few function comments

10 months agocifs: Fix missing set of remote_i_size
David Howells [Fri, 24 May 2024 14:23:36 +0000 (15:23 +0100)]
cifs: Fix missing set of remote_i_size

Occasionally, the generic/001 xfstest will fail indicating corruption in
one of the copy chains when run on cifs against a server that supports
FSCTL_DUPLICATE_EXTENTS_TO_FILE (eg. Samba with a share on btrfs).  The
problem is that the remote_i_size value isn't updated by cifs_setsize()
when called by smb2_duplicate_extents(), but i_size *is*.

This may cause cifs_remap_file_range() to then skip the bit after calling
->duplicate_extents() that sets sizes.

Fix this by calling netfs_resize_file() in smb2_duplicate_extents() before
calling cifs_setsize() to set i_size.

This means we don't then need to call netfs_resize_file() upon return from
->duplicate_extents(), but we also fix the test to compare against the pre-dup
inode size.

[Note that this goes back before the addition of remote_i_size with the
netfs_inode struct.  It should probably have been setting cifsi->server_eof
previously.]

Fixes: cfc63fc8126a ("smb3: fix cached file size problems in duplicate extents (reflink)")
Signed-off-by: David Howells <[email protected]>
cc: Steve French <[email protected]>
cc: Paulo Alcantara <[email protected]>
cc: Shyam Prasad N <[email protected]>
cc: Rohith Surabattula <[email protected]>
cc: Jeff Layton <[email protected]>
cc: [email protected]
cc: [email protected]
Signed-off-by: Steve French <[email protected]>
10 months agocifs: Fix smb3_insert_range() to move the zero_point
David Howells [Wed, 22 May 2024 08:38:48 +0000 (09:38 +0100)]
cifs: Fix smb3_insert_range() to move the zero_point

Fix smb3_insert_range() to move the zero_point over to the new EOF.
Without this, generic/147 fails as reads of data beyond the old EOF point
return zeroes.

Fixes: 3ee1a1fc3981 ("cifs: Cut over to using netfslib")
Signed-off-by: David Howells <[email protected]>
cc: Shyam Prasad N <[email protected]>
cc: Rohith Surabattula <[email protected]>
cc: Jeff Layton <[email protected]>
cc: [email protected]
cc: [email protected]
Signed-off-by: Steve French <[email protected]>
10 months agoMerge tag 'mm-stable-2024-05-24-11-49' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 24 May 2024 19:47:28 +0000 (12:47 -0700)]
Merge tag 'mm-stable-2024-05-24-11-49' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm

Pull more mm updates from Andrew Morton:
 "Jeff Xu's implementation of the mseal() syscall"

* tag 'mm-stable-2024-05-24-11-49' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm:
  selftest mm/mseal read-only elf memory segment
  mseal: add documentation
  selftest mm/mseal memory sealing
  mseal: add mseal syscall
  mseal: wire up mseal syscall

10 months agomm/ksm: fix possible UAF of stable_node
Chengming Zhou [Mon, 13 May 2024 03:07:56 +0000 (11:07 +0800)]
mm/ksm: fix possible UAF of stable_node

The commit 2c653d0ee2ae ("ksm: introduce ksm_max_page_sharing per page
deduplication limit") introduced a possible failure case in the
stable_tree_insert(), where we may free the new allocated stable_node_dup
if we fail to prepare the missing chain node.

Then that kfolio return and unlock with a freed stable_node set...  And
any MM activities can come in to access kfolio->mapping, so UAF.

Fix it by moving folio_set_stable_node() to the end after stable_node
is inserted successfully.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 2c653d0ee2ae ("ksm: introduce ksm_max_page_sharing per page deduplication limit")
Signed-off-by: Chengming Zhou <[email protected]>
Acked-by: David Hildenbrand <[email protected]>
Cc: Andrea Arcangeli <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: Stefan Roesch <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agomm/memory-failure: fix handling of dissolved but not taken off from buddy pages
Miaohe Lin [Thu, 23 May 2024 07:12:17 +0000 (15:12 +0800)]
mm/memory-failure: fix handling of dissolved but not taken off from buddy pages

When I did memory failure tests recently, below panic occurs:

page: refcount:0 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x8cee00
flags: 0x6fffe0000000000(node=1|zone=2|lastcpupid=0x7fff)
raw: 06fffe0000000000 dead000000000100 dead000000000122 0000000000000000
raw: 0000000000000000 0000000000000009 00000000ffffffff 0000000000000000
page dumped because: VM_BUG_ON_PAGE(!PageBuddy(page))
------------[ cut here ]------------
kernel BUG at include/linux/page-flags.h:1009!
invalid opcode: 0000 [#1] PREEMPT SMP NOPTI
RIP: 0010:__del_page_from_free_list+0x151/0x180
RSP: 0018:ffffa49c90437998 EFLAGS: 00000046
RAX: 0000000000000035 RBX: 0000000000000009 RCX: ffff8dd8dfd1c9c8
RDX: 0000000000000000 RSI: 0000000000000027 RDI: ffff8dd8dfd1c9c0
RBP: ffffd901233b8000 R08: ffffffffab5511f8 R09: 0000000000008c69
R10: 0000000000003c15 R11: ffffffffab5511f8 R12: ffff8dd8fffc0c80
R13: 0000000000000001 R14: ffff8dd8fffc0c80 R15: 0000000000000009
FS:  00007ff916304740(0000) GS:ffff8dd8dfd00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 000055eae50124c8 CR3: 00000008479e0000 CR4: 00000000000006f0
Call Trace:
 <TASK>
 __rmqueue_pcplist+0x23b/0x520
 get_page_from_freelist+0x26b/0xe40
 __alloc_pages_noprof+0x113/0x1120
 __folio_alloc_noprof+0x11/0xb0
 alloc_buddy_hugetlb_folio.isra.0+0x5a/0x130
 __alloc_fresh_hugetlb_folio+0xe7/0x140
 alloc_pool_huge_folio+0x68/0x100
 set_max_huge_pages+0x13d/0x340
 hugetlb_sysctl_handler_common+0xe8/0x110
 proc_sys_call_handler+0x194/0x280
 vfs_write+0x387/0x550
 ksys_write+0x64/0xe0
 do_syscall_64+0xc2/0x1d0
 entry_SYSCALL_64_after_hwframe+0x77/0x7f
RIP: 0033:0x7ff916114887
RSP: 002b:00007ffec8a2fd78 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
RAX: ffffffffffffffda RBX: 000055eae500e350 RCX: 00007ff916114887
RDX: 0000000000000004 RSI: 000055eae500e390 RDI: 0000000000000003
RBP: 000055eae50104c0 R08: 0000000000000000 R09: 000055eae50104c0
R10: 0000000000000077 R11: 0000000000000246 R12: 0000000000000004
R13: 0000000000000004 R14: 00007ff916216b80 R15: 00007ff916216a00
 </TASK>
Modules linked in: mce_inject hwpoison_inject
---[ end trace 0000000000000000 ]---

And before the panic, there had an warning about bad page state:

BUG: Bad page state in process page-types  pfn:8cee00
page: refcount:0 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x8cee00
flags: 0x6fffe0000000000(node=1|zone=2|lastcpupid=0x7fff)
page_type: 0xffffff7f(buddy)
raw: 06fffe0000000000 ffffd901241c0008 ffffd901240f8008 0000000000000000
raw: 0000000000000000 0000000000000009 00000000ffffff7f 0000000000000000
page dumped because: nonzero mapcount
Modules linked in: mce_inject hwpoison_inject
CPU: 8 PID: 154211 Comm: page-types Not tainted 6.9.0-rc4-00499-g5544ec3178e2-dirty #22
Call Trace:
 <TASK>
 dump_stack_lvl+0x83/0xa0
 bad_page+0x63/0xf0
 free_unref_page+0x36e/0x5c0
 unpoison_memory+0x50b/0x630
 simple_attr_write_xsigned.constprop.0.isra.0+0xb3/0x110
 debugfs_attr_write+0x42/0x60
 full_proxy_write+0x5b/0x80
 vfs_write+0xcd/0x550
 ksys_write+0x64/0xe0
 do_syscall_64+0xc2/0x1d0
 entry_SYSCALL_64_after_hwframe+0x77/0x7f
RIP: 0033:0x7f189a514887
RSP: 002b:00007ffdcd899718 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f189a514887
RDX: 0000000000000009 RSI: 00007ffdcd899730 RDI: 0000000000000003
RBP: 00007ffdcd8997a0 R08: 0000000000000000 R09: 00007ffdcd8994b2
R10: 0000000000000000 R11: 0000000000000246 R12: 00007ffdcda199a8
R13: 0000000000404af1 R14: 000000000040ad78 R15: 00007f189a7a5040
 </TASK>

The root cause should be the below race:

 memory_failure
  try_memory_failure_hugetlb
   me_huge_page
    __page_handle_poison
     dissolve_free_hugetlb_folio
     drain_all_pages -- Buddy page can be isolated e.g. for compaction.
     take_page_off_buddy -- Failed as page is not in the buddy list.
     -- Page can be putback into buddy after compaction.
    page_ref_inc -- Leads to buddy page with refcnt = 1.

Then unpoison_memory() can unpoison the page and send the buddy page back
into buddy list again leading to the above bad page state warning.  And
bad_page() will call page_mapcount_reset() to remove PageBuddy from buddy
page leading to later VM_BUG_ON_PAGE(!PageBuddy(page)) when trying to
allocate this page.

Fix this issue by only treating __page_handle_poison() as successful when
it returns 1.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: ceaf8fbea79a ("mm, hwpoison: skip raw hwpoison page in freeing 1GB hugepage")
Signed-off-by: Miaohe Lin <[email protected]>
Cc: Naoya Horiguchi <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agomm: /proc/pid/smaps_rollup: avoid skipping vma after getting mmap_lock again
Yuanyuan Zhong [Thu, 23 May 2024 18:35:31 +0000 (12:35 -0600)]
mm: /proc/pid/smaps_rollup: avoid skipping vma after getting mmap_lock again

After switching smaps_rollup to use VMA iterator, searching for next entry
is part of the condition expression of the do-while loop.  So the current
VMA needs to be addressed before the continue statement.

Otherwise, with some VMAs skipped, userspace observed memory
consumption from /proc/pid/smaps_rollup will be smaller than the sum of
the corresponding fields from /proc/pid/smaps.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: c4c84f06285e ("fs/proc/task_mmu: stop using linked list and highest_vm_end")
Signed-off-by: Yuanyuan Zhong <[email protected]>
Reviewed-by: Mohamed Khalfella <[email protected]>
Cc: David Hildenbrand <[email protected]>
Cc: Matthew Wilcox (Oracle) <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agonilfs2: fix potential hang in nilfs_detach_log_writer()
Ryusuke Konishi [Mon, 20 May 2024 13:26:21 +0000 (22:26 +0900)]
nilfs2: fix potential hang in nilfs_detach_log_writer()

Syzbot has reported a potential hang in nilfs_detach_log_writer() called
during nilfs2 unmount.

Analysis revealed that this is because nilfs_segctor_sync(), which
synchronizes with the log writer thread, can be called after
nilfs_segctor_destroy() terminates that thread, as shown in the call trace
below:

nilfs_detach_log_writer
  nilfs_segctor_destroy
    nilfs_segctor_kill_thread  --> Shut down log writer thread
    flush_work
      nilfs_iput_work_func
        nilfs_dispose_list
          iput
            nilfs_evict_inode
              nilfs_transaction_commit
                nilfs_construct_segment (if inode needs sync)
                  nilfs_segctor_sync  --> Attempt to synchronize with
                                          log writer thread
                           *** DEADLOCK ***

Fix this issue by changing nilfs_segctor_sync() so that the log writer
thread returns normally without synchronizing after it terminates, and by
forcing tasks that are already waiting to complete once after the thread
terminates.

The skipped inode metadata flushout will then be processed together in the
subsequent cleanup work in nilfs_segctor_destroy().

Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Ryusuke Konishi <[email protected]>
Reported-by: [email protected]
Closes: https://syzkaller.appspot.com/bug?extid=e3973c409251e136fdd0
Tested-by: Ryusuke Konishi <[email protected]>
Cc: <[email protected]>
Cc: "Bai, Shuangpeng" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agonilfs2: fix unexpected freezing of nilfs_segctor_sync()
Ryusuke Konishi [Mon, 20 May 2024 13:26:20 +0000 (22:26 +0900)]
nilfs2: fix unexpected freezing of nilfs_segctor_sync()

A potential and reproducible race issue has been identified where
nilfs_segctor_sync() would block even after the log writer thread writes a
checkpoint, unless there is an interrupt or other trigger to resume log
writing.

This turned out to be because, depending on the execution timing of the
log writer thread running in parallel, the log writer thread may skip
responding to nilfs_segctor_sync(), which causes a call to schedule()
waiting for completion within nilfs_segctor_sync() to lose the opportunity
to wake up.

The reason why waking up the task waiting in nilfs_segctor_sync() may be
skipped is that updating the request generation issued using a shared
sequence counter and adding an wait queue entry to the request wait queue
to the log writer, are not done atomically.  There is a possibility that
log writing and request completion notification by nilfs_segctor_wakeup()
may occur between the two operations, and in that case, the wait queue
entry is not yet visible to nilfs_segctor_wakeup() and the wake-up of
nilfs_segctor_sync() will be carried over until the next request occurs.

Fix this issue by performing these two operations simultaneously within
the lock section of sc_state_lock.  Also, following the memory barrier
guidelines for event waiting loops, move the call to set_current_state()
in the same location into the event waiting loop to ensure that a memory
barrier is inserted just before the event condition determination.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 9ff05123e3bf ("nilfs2: segment constructor")
Signed-off-by: Ryusuke Konishi <[email protected]>
Tested-by: Ryusuke Konishi <[email protected]>
Cc: <[email protected]>
Cc: "Bai, Shuangpeng" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agonilfs2: fix use-after-free of timer for log writer thread
Ryusuke Konishi [Mon, 20 May 2024 13:26:19 +0000 (22:26 +0900)]
nilfs2: fix use-after-free of timer for log writer thread

Patch series "nilfs2: fix log writer related issues".

This bug fix series covers three nilfs2 log writer-related issues,
including a timer use-after-free issue and potential deadlock issue on
unmount, and a potential freeze issue in event synchronization found
during their analysis.  Details are described in each commit log.

This patch (of 3):

A use-after-free issue has been reported regarding the timer sc_timer on
the nilfs_sc_info structure.

The problem is that even though it is used to wake up a sleeping log
writer thread, sc_timer is not shut down until the nilfs_sc_info structure
is about to be freed, and is used regardless of the thread's lifetime.

Fix this issue by limiting the use of sc_timer only while the log writer
thread is alive.

Link: https://lkml.kernel.org/r/[email protected]
Link: https://lkml.kernel.org/r/[email protected]
Fixes: fdce895ea5dd ("nilfs2: change sc_timer from a pointer to an embedded one in struct nilfs_sc_info")
Signed-off-by: Ryusuke Konishi <[email protected]>
Reported-by: "Bai, Shuangpeng" <[email protected]>
Closes: https://groups.google.com/g/syzkaller/c/MK_LYqtt8ko/m/8rgdWeseAwAJ
Tested-by: Ryusuke Konishi <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agoselftests/mm: fix build warnings on ppc64
Michael Ellerman [Tue, 21 May 2024 03:02:19 +0000 (13:02 +1000)]
selftests/mm: fix build warnings on ppc64

Fix warnings like:

  In file included from uffd-unit-tests.c:8:
  uffd-unit-tests.c: In function `uffd_poison_handle_fault':
  uffd-common.h:45:33: warning: format `%llu' expects argument of type
  `long long unsigned int', but argument 3 has type `__u64' {aka `long
  unsigned int'} [-Wformat=]

By switching to unsigned long long for u64 for ppc64 builds.

Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Michael Ellerman <[email protected]>
Cc: Shuah Khan <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agoarm64: patching: fix handling of execmem addresses
Will Deacon [Tue, 21 May 2024 21:38:13 +0000 (00:38 +0300)]
arm64: patching: fix handling of execmem addresses

Klara Modin reported warnings for a kernel configured with BPF_JIT but
without MODULES:

[   44.131296] Trying to vfree() bad address (000000004a17c299)
[   44.138024] WARNING: CPU: 1 PID: 193 at mm/vmalloc.c:3189 remove_vm_area (mm/vmalloc.c:3189 (discriminator 1))
[   44.146675] CPU: 1 PID: 193 Comm: kworker/1:2 Tainted: G      D W          6.9.0-01786-g2c9e5d4a0082 #25
[   44.158229] Hardware name: Raspberry Pi 3 Model B (DT)
[   44.164433] Workqueue: events bpf_prog_free_deferred
[   44.170492] pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[   44.178601] pc : remove_vm_area (mm/vmalloc.c:3189 (discriminator 1))
[   44.183705] lr : remove_vm_area (mm/vmalloc.c:3189 (discriminator 1))
[   44.188772] sp : ffff800082a13c70
[   44.193112] x29: ffff800082a13c70 x28: 0000000000000000 x27: 0000000000000000
[   44.201384] x26: 0000000000000000 x25: ffff00003a44efa0 x24: 00000000d4202000
[   44.209658] x23: ffff800081223dd0 x22: ffff00003a198a40 x21: ffff8000814dd880
[   44.217924] x20: 00000000d4202000 x19: ffff8000814dd880 x18: 0000000000000006
[   44.226206] x17: 0000000000000000 x16: 0000000000000020 x15: 0000000000000002
[   44.234460] x14: ffff8000811a6370 x13: 0000000020000000 x12: 0000000000000000
[   44.242710] x11: ffff8000811a6370 x10: 0000000000000144 x9 : ffff8000811fe370
[   44.250959] x8 : 0000000000017fe8 x7 : 00000000fffff000 x6 : ffff8000811fe370
[   44.259206] x5 : 0000000000000000 x4 : 0000000000000000 x3 : 0000000000000000
[   44.267457] x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffff000002203240
[   44.275703] Call trace:
[   44.279158] remove_vm_area (mm/vmalloc.c:3189 (discriminator 1))
[   44.283858] vfree (mm/vmalloc.c:3322)
[   44.287835] execmem_free (mm/execmem.c:70)
[   44.292347] bpf_jit_free_exec+0x10/0x1c
[   44.297283] bpf_prog_pack_free (kernel/bpf/core.c:1006)
[   44.302457] bpf_jit_binary_pack_free (kernel/bpf/core.c:1195)
[   44.307951] bpf_jit_free (include/linux/filter.h:1083 arch/arm64/net/bpf_jit_comp.c:2474)
[   44.312342] bpf_prog_free_deferred (kernel/bpf/core.c:2785)
[   44.317785] process_one_work (kernel/workqueue.c:3273)
[   44.322684] worker_thread (kernel/workqueue.c:3342 (discriminator 2) kernel/workqueue.c:3429 (discriminator 2))
[   44.327292] kthread (kernel/kthread.c:388)
[   44.331342] ret_from_fork (arch/arm64/kernel/entry.S:861)

The problem is because bpf_arch_text_copy() silently fails to write to the
read-only area as a result of patch_map() faulting and the resulting
-EFAULT being chucked away.

Update patch_map() to use CONFIG_EXECMEM instead of
CONFIG_STRICT_MODULE_RWX to check for vmalloc addresses.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 2c9e5d4a0082 ("bpf: remove CONFIG_BPF_JIT dependency on CONFIG_MODULES of")
Signed-off-by: Will Deacon <[email protected]>
Signed-off-by: Mike Rapoport (IBM) <[email protected]>
Reported-by: Klara Modin <[email protected]>
Closes: https://lore.kernel.org/all/[email protected]
Tested-by: Klara Modin <[email protected]>
Cc: Björn Töpel <[email protected]>
Cc: Luis Chamberlain <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agoselftests/mm: compaction_test: fix bogus test success and reduce probability of OOM...
Dev Jain [Tue, 21 May 2024 07:43:58 +0000 (13:13 +0530)]
selftests/mm: compaction_test: fix bogus test success and reduce probability of OOM-killer invocation

Reset nr_hugepages to zero before the start of the test.

If a non-zero number of hugepages is already set before the start of the
test, the following problems arise:

 - The probability of the test getting OOM-killed increases.  Proof:
   The test wants to run on 80% of available memory to prevent OOM-killing
   (see original code comments).  Let the value of mem_free at the start
   of the test, when nr_hugepages = 0, be x.  In the other case, when
   nr_hugepages > 0, let the memory consumed by hugepages be y.  In the
   former case, the test operates on 0.8 * x of memory.  In the latter,
   the test operates on 0.8 * (x - y) of memory, with y already filled,
   hence, memory consumed is y + 0.8 * (x - y) = 0.8 * x + 0.2 * y > 0.8 *
   x.  Q.E.D

 - The probability of a bogus test success increases.  Proof: Let the
   memory consumed by hugepages be greater than 25% of x, with x and y
   defined as above.  The definition of compaction_index is c_index = (x -
   y)/z where z is the memory consumed by hugepages after trying to
   increase them again.  In check_compaction(), we set the number of
   hugepages to zero, and then increase them back; the probability that
   they will be set back to consume at least y amount of memory again is
   very high (since there is not much delay between the two attempts of
   changing nr_hugepages).  Hence, z >= y > (x/4) (by the 25% assumption).
   Therefore, c_index = (x - y)/z <= (x - y)/y = x/y - 1 < 4 - 1 = 3
   hence, c_index can always be forced to be less than 3, thereby the test
   succeeding always.  Q.E.D

Link: https://lkml.kernel.org/r/[email protected]
Fixes: bd67d5c15cc1 ("Test compaction of mlocked memory")
Signed-off-by: Dev Jain <[email protected]>
Cc: <[email protected]>
Cc: Anshuman Khandual <[email protected]>
Cc: Shuah Khan <[email protected]>
Cc: Sri Jayaramappa <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agoselftests/mm: compaction_test: fix incorrect write of zero to nr_hugepages
Dev Jain [Tue, 21 May 2024 07:43:57 +0000 (13:13 +0530)]
selftests/mm: compaction_test: fix incorrect write of zero to nr_hugepages

Currently, the test tries to set nr_hugepages to zero, but that is not
actually done because the file offset is not reset after read().  Fix that
using lseek().

Link: https://lkml.kernel.org/r/[email protected]
Fixes: bd67d5c15cc1 ("Test compaction of mlocked memory")
Signed-off-by: Dev Jain <[email protected]>
Cc: <[email protected]>
Cc: Anshuman Khandual <[email protected]>
Cc: Shuah Khan <[email protected]>
Cc: Sri Jayaramappa <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agoselftests/mm: compaction_test: fix bogus test success on Aarch64
Dev Jain [Tue, 21 May 2024 07:43:56 +0000 (13:13 +0530)]
selftests/mm: compaction_test: fix bogus test success on Aarch64

Patch series "Fixes for compaction_test", v2.

The compaction_test memory selftest introduces fragmentation in memory
and then tries to allocate as many hugepages as possible. This series
addresses some problems.

On Aarch64, if nr_hugepages == 0, then the test trivially succeeds since
compaction_index becomes 0, which is less than 3, due to no division by
zero exception being raised. We fix that by checking for division by
zero.

Secondly, correctly set the number of hugepages to zero before trying
to set a large number of them.

Now, consider a situation in which, at the start of the test, a non-zero
number of hugepages have been already set (while running the entire
selftests/mm suite, or manually by the admin). The test operates on 80%
of memory to avoid OOM-killer invocation, and because some memory is
already blocked by hugepages, it would increase the chance of OOM-killing.
Also, since mem_free used in check_compaction() is the value before we
set nr_hugepages to zero, the chance that the compaction_index will
be small is very high if the preset nr_hugepages was high, leading to a
bogus test success.

This patch (of 3):

Currently, if at runtime we are not able to allocate a huge page, the test
will trivially pass on Aarch64 due to no exception being raised on
division by zero while computing compaction_index.  Fix that by checking
for nr_hugepages == 0.  Anyways, in general, avoid a division by zero by
exiting the program beforehand.  While at it, fix a typo, and handle the
case where the number of hugepages may overflow an integer.

Link: https://lkml.kernel.org/r/[email protected]
Link: https://lkml.kernel.org/r/[email protected]
Fixes: bd67d5c15cc1 ("Test compaction of mlocked memory")
Signed-off-by: Dev Jain <[email protected]>
Cc: Anshuman Khandual <[email protected]>
Cc: Shuah Khan <[email protected]>
Cc: Sri Jayaramappa <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agomailmap: update email address for Satya Priya
Satya Priya Kakitapalli [Wed, 15 May 2024 06:04:50 +0000 (11:34 +0530)]
mailmap: update email address for Satya Priya

Update mailmap with my latest email ID, [email protected]
is no longer active.

Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Satya Priya Kakitapalli <[email protected]>
Cc: Ajit Pandey <[email protected]>
Cc: Bjorn Andersson <[email protected]>
Cc: Imran Shaik <[email protected]>
Cc: Jagadeesh Kona <[email protected]>
Cc: Konrad Dybcio <[email protected]>
Cc: Taniya Das <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agomm/huge_memory: don't unpoison huge_zero_folio
Miaohe Lin [Thu, 16 May 2024 12:26:08 +0000 (20:26 +0800)]
mm/huge_memory: don't unpoison huge_zero_folio

When I did memory failure tests recently, below panic occurs:

 kernel BUG at include/linux/mm.h:1135!
 invalid opcode: 0000 [#1] PREEMPT SMP NOPTI
 CPU: 9 PID: 137 Comm: kswapd1 Not tainted 6.9.0-rc4-00491-gd5ce28f156fe-dirty #14
 RIP: 0010:shrink_huge_zero_page_scan+0x168/0x1a0
 RSP: 0018:ffff9933c6c57bd0 EFLAGS: 00000246
 RAX: 000000000000003e RBX: 0000000000000000 RCX: ffff88f61fc5c9c8
 RDX: 0000000000000000 RSI: 0000000000000027 RDI: ffff88f61fc5c9c0
 RBP: ffffcd7c446b0000 R08: ffffffff9a9405f0 R09: 0000000000005492
 R10: 00000000000030ea R11: ffffffff9a9405f0 R12: 0000000000000000
 R13: 0000000000000000 R14: 0000000000000000 R15: ffff88e703c4ac00
 FS:  0000000000000000(0000) GS:ffff88f61fc40000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 000055f4da6e9878 CR3: 0000000c71048000 CR4: 00000000000006f0
 Call Trace:
  <TASK>
  do_shrink_slab+0x14f/0x6a0
  shrink_slab+0xca/0x8c0
  shrink_node+0x2d0/0x7d0
  balance_pgdat+0x33a/0x720
  kswapd+0x1f3/0x410
  kthread+0xd5/0x100
  ret_from_fork+0x2f/0x50
  ret_from_fork_asm+0x1a/0x30
  </TASK>
 Modules linked in: mce_inject hwpoison_inject
 ---[ end trace 0000000000000000 ]---
 RIP: 0010:shrink_huge_zero_page_scan+0x168/0x1a0
 RSP: 0018:ffff9933c6c57bd0 EFLAGS: 00000246
 RAX: 000000000000003e RBX: 0000000000000000 RCX: ffff88f61fc5c9c8
 RDX: 0000000000000000 RSI: 0000000000000027 RDI: ffff88f61fc5c9c0
 RBP: ffffcd7c446b0000 R08: ffffffff9a9405f0 R09: 0000000000005492
 R10: 00000000000030ea R11: ffffffff9a9405f0 R12: 0000000000000000
 R13: 0000000000000000 R14: 0000000000000000 R15: ffff88e703c4ac00
 FS:  0000000000000000(0000) GS:ffff88f61fc40000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 000055f4da6e9878 CR3: 0000000c71048000 CR4: 00000000000006f0

The root cause is that HWPoison flag will be set for huge_zero_folio
without increasing the folio refcnt.  But then unpoison_memory() will
decrease the folio refcnt unexpectedly as it appears like a successfully
hwpoisoned folio leading to VM_BUG_ON_PAGE(page_ref_count(page) == 0) when
releasing huge_zero_folio.

Skip unpoisoning huge_zero_folio in unpoison_memory() to fix this issue.
We're not prepared to unpoison huge_zero_folio yet.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 478d134e9506 ("mm/huge_memory: do not overkill when splitting huge_zero_page")
Signed-off-by: Miaohe Lin <[email protected]>
Acked-by: David Hildenbrand <[email protected]>
Reviewed-by: Yang Shi <[email protected]>
Reviewed-by: Oscar Salvador <[email protected]>
Reviewed-by: Anshuman Khandual <[email protected]>
Cc: Naoya Horiguchi <[email protected]>
Cc: Xu Yu <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agokasan, fortify: properly rename memintrinsics
Andrey Konovalov [Fri, 17 May 2024 13:01:18 +0000 (15:01 +0200)]
kasan, fortify: properly rename memintrinsics

After commit 69d4c0d32186 ("entry, kasan, x86: Disallow overriding mem*()
functions") and the follow-up fixes, with CONFIG_FORTIFY_SOURCE enabled,
even though the compiler instruments meminstrinsics by generating calls to
__asan/__hwasan_ prefixed functions, FORTIFY_SOURCE still uses
uninstrumented memset/memmove/memcpy as the underlying functions.

As a result, KASAN cannot detect bad accesses in memset/memmove/memcpy.
This also makes KASAN tests corrupt kernel memory and cause crashes.

To fix this, use __asan_/__hwasan_memset/memmove/memcpy as the underlying
functions whenever appropriate.  Do this only for the instrumented code
(as indicated by __SANITIZE_ADDRESS__).

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 69d4c0d32186 ("entry, kasan, x86: Disallow overriding mem*() functions")
Fixes: 51287dcb00cc ("kasan: emit different calls for instrumentable memintrinsics")
Fixes: 36be5cba99f6 ("kasan: treat meminstrinsic as builtins in uninstrumented files")
Signed-off-by: Andrey Konovalov <[email protected]>
Reported-by: Erhard Furtner <[email protected]>
Reported-by: Nico Pache <[email protected]>
Closes: https://lore.kernel.org/all/[email protected]/
Reviewed-by: Marco Elver <[email protected]>
Tested-by: Nico Pache <[email protected]>
Acked-by: Nico Pache <[email protected]>
Cc: Alexander Potapenko <[email protected]>
Cc: Andrey Ryabinin <[email protected]>
Cc: Daniel Axtens <[email protected]>
Cc: Dmitry Vyukov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agolib: add version into /proc/allocinfo output
Suren Baghdasaryan [Tue, 14 May 2024 16:31:28 +0000 (09:31 -0700)]
lib: add version into /proc/allocinfo output

Add version string and a header at the beginning of /proc/allocinfo to
allow later format changes.  Example output:

> head /proc/allocinfo
allocinfo - version: 1.0
#     <size>  <calls> <tag info>
           0        0 init/main.c:1314 func:do_initcalls
           0        0 init/do_mounts.c:353 func:mount_nodev_root
           0        0 init/do_mounts.c:187 func:mount_root_generic
           0        0 init/do_mounts.c:158 func:do_mount_root
           0        0 init/initramfs.c:493 func:unpack_to_rootfs
           0        0 init/initramfs.c:492 func:unpack_to_rootfs
           0        0 init/initramfs.c:491 func:unpack_to_rootfs
         512        1 arch/x86/events/rapl.c:681 func:init_rapl_pmus
         128        1 arch/x86/events/rapl.c:571 func:rapl_cpu_online

[[email protected]: remove stray newline from struct allocinfo_private]
Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Suren Baghdasaryan <[email protected]>
Reviewed-by: Pasha Tatashin <[email protected]>
Reviewed-by: Kees Cook <[email protected]>
Cc: Kent Overstreet <[email protected]>
Cc: Vlastimil Babka <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agomm/vmalloc: fix vmalloc which may return null if called with __GFP_NOFAIL
Hailong.Liu [Fri, 10 May 2024 10:01:31 +0000 (18:01 +0800)]
mm/vmalloc: fix vmalloc which may return null if called with __GFP_NOFAIL

commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc")
includes support for __GFP_NOFAIL, but it presents a conflict with commit
dd544141b9eb ("vmalloc: back off when the current task is OOM-killed").  A
possible scenario is as follows:

process-a
__vmalloc_node_range(GFP_KERNEL | __GFP_NOFAIL)
    __vmalloc_area_node()
        vm_area_alloc_pages()
--> oom-killer send SIGKILL to process-a
        if (fatal_signal_pending(current)) break;
--> return NULL;

To fix this, do not check fatal_signal_pending() in vm_area_alloc_pages()
if __GFP_NOFAIL set.

This issue occurred during OPLUS KASAN TEST. Below is part of the log
-> oom-killer sends signal to process
[65731.222840] [ T1308] oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),cpuset=/,mems_allowed=0,global_oom,task_memcg=/apps/uid_10198,task=gs.intelligence,pid=32454,uid=10198

[65731.259685] [T32454] Call trace:
[65731.259698] [T32454]  dump_backtrace+0xf4/0x118
[65731.259734] [T32454]  show_stack+0x18/0x24
[65731.259756] [T32454]  dump_stack_lvl+0x60/0x7c
[65731.259781] [T32454]  dump_stack+0x18/0x38
[65731.259800] [T32454]  mrdump_common_die+0x250/0x39c [mrdump]
[65731.259936] [T32454]  ipanic_die+0x20/0x34 [mrdump]
[65731.260019] [T32454]  atomic_notifier_call_chain+0xb4/0xfc
[65731.260047] [T32454]  notify_die+0x114/0x198
[65731.260073] [T32454]  die+0xf4/0x5b4
[65731.260098] [T32454]  die_kernel_fault+0x80/0x98
[65731.260124] [T32454]  __do_kernel_fault+0x160/0x2a8
[65731.260146] [T32454]  do_bad_area+0x68/0x148
[65731.260174] [T32454]  do_mem_abort+0x151c/0x1b34
[65731.260204] [T32454]  el1_abort+0x3c/0x5c
[65731.260227] [T32454]  el1h_64_sync_handler+0x54/0x90
[65731.260248] [T32454]  el1h_64_sync+0x68/0x6c

[65731.260269] [T32454]  z_erofs_decompress_queue+0x7f0/0x2258
--> be->decompressed_pages = kvcalloc(be->nr_pages, sizeof(struct page *), GFP_KERNEL | __GFP_NOFAIL);
kernel panic by NULL pointer dereference.
erofs assume kvmalloc with __GFP_NOFAIL never return NULL.
[65731.260293] [T32454]  z_erofs_runqueue+0xf30/0x104c
[65731.260314] [T32454]  z_erofs_readahead+0x4f0/0x968
[65731.260339] [T32454]  read_pages+0x170/0xadc
[65731.260364] [T32454]  page_cache_ra_unbounded+0x874/0xf30
[65731.260388] [T32454]  page_cache_ra_order+0x24c/0x714
[65731.260411] [T32454]  filemap_fault+0xbf0/0x1a74
[65731.260437] [T32454]  __do_fault+0xd0/0x33c
[65731.260462] [T32454]  handle_mm_fault+0xf74/0x3fe0
[65731.260486] [T32454]  do_mem_abort+0x54c/0x1b34
[65731.260509] [T32454]  el0_da+0x44/0x94
[65731.260531] [T32454]  el0t_64_sync_handler+0x98/0xb4
[65731.260553] [T32454]  el0t_64_sync+0x198/0x19c

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 9376130c390a ("mm/vmalloc: add support for __GFP_NOFAIL")
Signed-off-by: Hailong.Liu <[email protected]>
Acked-by: Michal Hocko <[email protected]>
Suggested-by: Barry Song <[email protected]>
Reported-by: Oven <[email protected]>
Reviewed-by: Barry Song <[email protected]>
Reviewed-by: Uladzislau Rezki (Sony) <[email protected]>
Cc: Chao Yu <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Cc: Gao Xiang <[email protected]>
Cc: Lorenzo Stoakes <[email protected]>
Cc: Michal Hocko <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agoMerge tag 'riscv-for-linus-6.10-mw2' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 24 May 2024 17:46:35 +0000 (10:46 -0700)]
Merge tag 'riscv-for-linus-6.10-mw2' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linux

Pull more RISC-V updates from Palmer Dabbelt:

 - The compression format used for boot images is now configurable at
   build time, and these formats are shown in `make help`

 - access_ok() has been optimized

 - A pair of performance bugs have been fixed in the uaccess handlers

 - Various fixes and cleanups, including one for the IMSIC build failure
   and one for the early-boot ftrace illegal NOPs bug

* tag 'riscv-for-linus-6.10-mw2' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linux:
  riscv: Fix early ftrace nop patching
  irqchip: riscv-imsic: Fixup riscv_ipi_set_virq_range() conflict
  riscv: selftests: Add signal handling vector tests
  riscv: mm: accelerate pagefault when badaccess
  riscv: uaccess: Relax the threshold for fast path
  riscv: uaccess: Allow the last potential unrolled copy
  riscv: typo in comment for get_f64_reg
  Use bool value in set_cpu_online()
  riscv: selftests: Add hwprobe binaries to .gitignore
  riscv: stacktrace: fixed walk_stackframe()
  ftrace: riscv: move from REGS to ARGS
  riscv: do not select MODULE_SECTIONS by default
  riscv: show help string for riscv-specific targets
  riscv: make image compression configurable
  riscv: cpufeature: Fix extension subset checking
  riscv: cpufeature: Fix thead vector hwcap removal
  riscv: rewrite __kernel_map_pages() to fix sleeping in invalid context
  riscv: force PAGE_SIZE linear mapping if debug_pagealloc is enabled
  riscv: Define TASK_SIZE_MAX for __access_ok()
  riscv: Remove PGDIR_SIZE_L3 and TASK_SIZE_MIN

10 months agoMerge tag 'for-linus-6.10a-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 24 May 2024 17:24:49 +0000 (10:24 -0700)]
Merge tag 'for-linus-6.10a-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip

Pull xen updates from Juergen Gross:

 - a small cleanup in the drivers/xen/xenbus Makefile

 - a fix of the Xen xenstore driver to improve connecting to a late
   started Xenstore

 - an enhancement for better support of ballooning in PVH guests

 - a cleanup using try_cmpxchg() instead of open coding it

* tag 'for-linus-6.10a-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
  drivers/xen: Improve the late XenStore init protocol
  xen/xenbus: Use *-y instead of *-objs in Makefile
  xen/x86: add extra pages to unpopulated-alloc if available
  locking/x86/xen: Use try_cmpxchg() in xen_alloc_p2m_entry()

10 months agoMerge tag 'for-6.10-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux
Linus Torvalds [Fri, 24 May 2024 16:40:31 +0000 (09:40 -0700)]
Merge tag 'for-6.10-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux

Pull more btrfs updates from David Sterba:
 "A few more updates, mostly stability fixes or user visible changes:

   - fix race in zoned mode during device replace that can lead to
     use-after-free

   - update return codes and lower message levels for quota rescan where
     it's causing false alerts

   - fix unexpected qgroup id reuse under some conditions

   - fix condition when looking up extent refs

   - add option norecovery (removed in 6.8), the intended replacements
     haven't been used and some aplications still rely on the old one

   - build warning fixes"

* tag 'for-6.10-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: re-introduce 'norecovery' mount option
  btrfs: fix end of tree detection when searching for data extent ref
  btrfs: scrub: initialize ret in scrub_simple_mirror() to fix compilation warning
  btrfs: zoned: fix use-after-free due to race with dev replace
  btrfs: qgroup: fix qgroup id collision across mounts
  btrfs: qgroup: update rescan message levels and error codes

10 months agoMerge tag 'erofs-for-6.10-rc1-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 24 May 2024 16:31:50 +0000 (09:31 -0700)]
Merge tag 'erofs-for-6.10-rc1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/xiang/erofs

Pull more erofs updates from Gao Xiang:
 "The main ones are metadata API conversion to byte offsets by Al Viro.

  Another patch gets rid of unnecessary memory allocation out of DEFLATE
  decompressor. The remaining one is a trivial cleanup.

   - Convert metadata APIs to byte offsets

   - Avoid allocating DEFLATE streams unnecessarily

   - Some erofs_show_options() cleanup"

* tag 'erofs-for-6.10-rc1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/xiang/erofs:
  erofs: avoid allocating DEFLATE streams before mounting
  z_erofs_pcluster_begin(): don't bother with rounding position down
  erofs: don't round offset down for erofs_read_metabuf()
  erofs: don't align offset for erofs_read_metabuf() (simple cases)
  erofs: mechanically convert erofs_read_metabuf() to offsets
  erofs: clean up erofs_show_options()

10 months agoMerge tag 'bcachefs-2024-05-24' of https://evilpiepirate.org/git/bcachefs
Linus Torvalds [Fri, 24 May 2024 16:07:22 +0000 (09:07 -0700)]
Merge tag 'bcachefs-2024-05-24' of https://evilpiepirate.org/git/bcachefs

Pull bcachefs fixes from Kent Overstreet:
 "Nothing exciting, just syzbot fixes (except for the one
  FMODE_CAN_ODIRECT patch).

  Looks like syzbot reports have slowed down; this is all catch up from
  two weeks of conferences.

  Next hardening project is using Thomas's error injection tooling to
  torture test repair"

* tag 'bcachefs-2024-05-24' of https://evilpiepirate.org/git/bcachefs:
  bcachefs: Fix race path in bch2_inode_insert()
  bcachefs: Ensure we're RW before journalling
  bcachefs: Fix shutdown ordering
  bcachefs: Fix unsafety in bch2_dirent_name_bytes()
  bcachefs: Fix stack oob in __bch2_encrypt_bio()
  bcachefs: Fix btree_trans leak in bch2_readahead()
  bcachefs: Fix bogus verify_replicas_entry() assert
  bcachefs: Check for subvolues with bogus snapshot/inode fields
  bcachefs: bch2_checksum() returns 0 for unknown checksum type
  bcachefs: Fix bch2_alloc_ciphers()
  bcachefs: Add missing guard in bch2_snapshot_has_children()
  bcachefs: Fix missing parens in drop_locks_do()
  bcachefs: Improve bch2_assert_pos_locked()
  bcachefs: Fix shift overflows in replicas.c
  bcachefs: Fix shift overflow in btree_lost_data()
  bcachefs: Fix ref in trans_mark_dev_sbs() error path
  bcachefs: set FMODE_CAN_ODIRECT instead of a dummy direct_IO method
  bcachefs: Fix rcu splat in check_fix_ptrs()

10 months agoMerge tag 'input-for-v6.10-rc0' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 24 May 2024 16:01:21 +0000 (09:01 -0700)]
Merge tag 'input-for-v6.10-rc0' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input

Pull input updates from Dmitry Torokhov:

 - a change to input core to trim amount of keys data in modalias string
   in case when a device declares too many keys and they do not fit in
   uevent buffer instead of reporting an error which results in uevent
   not being generated at all

 - support for Machenike G5 Pro Controller added to xpad driver

 - support for FocalTech FT5452 and FT8719 added to edt-ft5x06

 - support for new SPMI vibrator added to pm8xxx-vibrator driver

 - missing locking added to cyapa touchpad driver

 - removal of unused fields in various driver structures

 - explicit initialization of i2c_device_id::driver_data to 0 dropped
   from input drivers

 - other assorted fixes and cleanups.

* tag 'input-for-v6.10-rc0' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input: (24 commits)
  Input: edt-ft5x06 - add support for FocalTech FT5452 and FT8719
  dt-bindings: input: touchscreen: edt-ft5x06: Document FT5452 and FT8719 support
  Input: xpad - add support for Machenike G5 Pro Controller
  Input: try trimming too long modalias strings
  Input: drop explicit initialization of struct i2c_device_id::driver_data to 0
  Input: zet6223 - remove an unused field in struct zet6223_ts
  Input: chipone_icn8505 - remove an unused field in struct icn8505_data
  Input: cros_ec_keyb - remove an unused field in struct cros_ec_keyb
  Input: lpc32xx-keys - remove an unused field in struct lpc32xx_kscan_drv
  Input: matrix_keypad - remove an unused field in struct matrix_keypad
  Input: tca6416-keypad - remove unused struct tca6416_drv_data
  Input: tca6416-keypad - remove an unused field in struct tca6416_keypad_chip
  Input: da7280 - remove an unused field in struct da7280_haptic
  Input: ff-core - prefer struct_size over open coded arithmetic
  Input: cyapa - add missing input core locking to suspend/resume functions
  input: pm8xxx-vibrator: add new SPMI vibrator support
  dt-bindings: input: qcom,pm8xxx-vib: add new SPMI vibrator module
  input: pm8xxx-vibrator: refactor to support new SPMI vibrator
  Input: pm8xxx-vibrator - correct VIB_MAX_LEVELS calculation
  Input: sur40 - convert le16 to cpu before use
  ...

10 months agoMerge tag 'sound-fix-6.10-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai...
Linus Torvalds [Fri, 24 May 2024 15:48:51 +0000 (08:48 -0700)]
Merge tag 'sound-fix-6.10-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound

Pull sound fixes from Takashi Iwai:
 "A collection of small fixes for 6.10-rc1. Most of changes are various
  device-specific fixes and quirks, while there are a few small changes
  in ALSA core timer and module / built-in fixes"

* tag 'sound-fix-6.10-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound:
  ALSA: hda/realtek: fix mute/micmute LEDs don't work for ProBook 440/460 G11.
  ALSA: core: Enable proc module when CONFIG_MODULES=y
  ALSA: core: Fix NULL module pointer assignment at card init
  ALSA: hda/realtek: Enable headset mic of JP-IK LEAP W502 with ALC897
  ASoC: dt-bindings: stm32: Ensure compatible pattern matches whole string
  ASoC: tas2781: Fix wrong loading calibrated data sequence
  ASoC: tas2552: Add TX path for capturing AUDIO-OUT data
  ALSA: usb-audio: Fix for sampling rates support for Mbox3
  Documentation: sound: Fix trailing whitespaces
  ALSA: timer: Set lower bound of start tick time
  ASoC: codecs: ES8326: solve hp and button detect issue
  ASoC: rt5645: mic-in detection threshold modification
  ASoC: Intel: sof_sdw_rt_sdca_jack_common: Use name_prefix for `-sdca` detection

10 months agoMerge tag 'char-misc-6.10-rc1-fix' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 24 May 2024 15:43:25 +0000 (08:43 -0700)]
Merge tag 'char-misc-6.10-rc1-fix' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc

Pull char/misc fix from Greg KH:
 "Here is one remaining bugfix for 6.10-rc1 that missed the 6.9-final
  merge window, and has been sitting in my tree and linux-next for quite
  a while now, but wasn't sent to you (my fault, travels...)

  It is a bugfix to resolve an error in the speakup code that could
  overflow a buffer.

  It has been in linux-next for a while with no reported problems"

* tag 'char-misc-6.10-rc1-fix' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc:
  speakup: Fix sizeof() vs ARRAY_SIZE() bug

10 months agoMerge tag 'tty-6.10-rc1-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/gregk...
Linus Torvalds [Fri, 24 May 2024 15:38:28 +0000 (08:38 -0700)]
Merge tag 'tty-6.10-rc1-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty

Pull tty/serial fixes from Greg KH:
 "Here are some small TTY and Serial driver fixes that missed the
  6.9-final merge window, but have been in my tree for weeks (my fault,
  travel caused me to miss this)

  These fixes include:

   - more n_gsm fixes for reported problems

   - 8520_mtk driver fix

   - 8250_bcm7271 driver fix

   - sc16is7xx driver fix

  All of these have been in linux-next for weeks without any reported
  problems"

* tag 'tty-6.10-rc1-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty:
  serial: sc16is7xx: fix bug in sc16is7xx_set_baud() when using prescaler
  serial: 8250_bcm7271: use default_mux_rate if possible
  serial: 8520_mtk: Set RTS on shutdown for Rx in-band wakeup
  tty: n_gsm: fix missing receive state reset after mode switch
  tty: n_gsm: fix possible out-of-bounds in gsm0_receive()

10 months agoMerge tag 'hardening-v6.10-rc1-fixes' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 24 May 2024 15:33:44 +0000 (08:33 -0700)]
Merge tag 'hardening-v6.10-rc1-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/kees/linux

Pull hardening fixes from Kees Cook:

 - loadpin: Prevent SECURITY_LOADPIN_ENFORCE=y without module
   decompression (Stephen Boyd)

 - ubsan: Restore dependency on ARCH_HAS_UBSAN

 - kunit/fortify: Fix memcmp() test to be amplitude agnostic

* tag 'hardening-v6.10-rc1-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/kees/linux:
  kunit/fortify: Fix memcmp() test to be amplitude agnostic
  ubsan: Restore dependency on ARCH_HAS_UBSAN
  loadpin: Prevent SECURITY_LOADPIN_ENFORCE=y without module decompression

10 months agoMerge tag 'trace-tracefs-v6.10' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 24 May 2024 15:27:34 +0000 (08:27 -0700)]
Merge tag 'trace-tracefs-v6.10' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace

Pull tracefs/eventfs updates from Steven Rostedt:
 "Bug fixes:

   - The eventfs directories need to have unique inode numbers. Make
     sure that they do not get the default file inode number.

   - Update the inode uid and gid fields on remount.

     When a remount happens where a uid and/or gid is specified, all the
     tracefs files and directories should get the specified uid and/or
     gid. But this can be sporadic when some uids were assigned already.
     There's already a list of inodes that are allocated. Just update
     their uid and gid fields at the time of remount.

   - Update the eventfs_inodes on remount from the top level "events"
     descriptor.

     There was a bug where not all the eventfs files or directories
     where getting updated on remount. One fix was to clear the
     SAVED_UID/GID flags from the inode list during the iteration of the
     inodes during the remount. But because the eventfs inodes can be
     freed when the last referenced is released, not all the
     eventfs_inodes were being updated. This lead to the ownership
     selftest to fail if it was run a second time (the first time would
     leave eventfs_inodes with no corresponding tracefs_inode).

     Instead, for eventfs_inodes, only process the "events"
     eventfs_inode from the list iteration, as it is guaranteed to have
     a tracefs_inode (it's never freed while the "events" directory
     exists). As it has a list of its children, and the children have a
     list of their children, just iterate all the eventfs_inodes from
     the "events" descriptor and it is guaranteed to get all of them.

   - Clear the EVENT_INODE flag from the tracefs_drop_inode() callback.

     Currently the EVENTFS_INODE FLAG is cleared in the tracefs_d_iput()
     callback. But this is the wrong location. The iput() callback is
     called when the last reference to the dentry inode is hit. There
     could be a case where two dentry's have the same inode, and the
     flag will be cleared prematurely. The flag needs to be cleared when
     the last reference of the inode is dropped and that happens in the
     inode's drop_inode() callback handler.

  Cleanups:

   - Consolidate the creation of a tracefs_inode for an eventfs_inode

     A tracefs_inode is created for both files and directories of the
     eventfs system. It is open coded. Instead, consolidate it into a
     single eventfs_get_inode() function call.

   - Remove the eventfs getattr and permission callbacks.

     The permissions for the eventfs files and directories are updated
     when the inodes are created, on remount, and when the user sets
     them (via setattr). The inodes hold the current permissions so
     there is no need to have custom getattr or permissions callbacks as
     they will more likely cause them to be incorrect. The inode's
     permissions are updated when they should be updated. Remove the
     getattr and permissions inode callbacks.

   - Do not update eventfs_inode attributes on creation of inodes.

     The eventfs_inodes attribute field is used to store the permissions
     of the directories and files for when their corresponding inodes
     are freed and are created again. But when the creation of the
     inodes happen, the eventfs_inode attributes are recalculated. The
     recalculation should only happen when the permissions change for a
     given file or directory. Currently, the attribute changes are just
     being set to their current files so this is not a bug, but it's
     unnecessary and error prone. Stop doing that.

   - The events directory inode is created once when the events
     directory is created and deleted when it is deleted. It is now
     updated on remount and when the user changes the permissions.
     There's no need to use the eventfs_inode of the events directory to
     store the events directory permissions. But using it to store the
     default permissions for the files within the directory that have
     not been updated by the user can simplify the code"

* tag 'trace-tracefs-v6.10' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace:
  eventfs: Do not use attributes for events directory
  eventfs: Cleanup permissions in creation of inodes
  eventfs: Remove getattr and permission callbacks
  eventfs: Consolidate the eventfs_inode update in eventfs_get_inode()
  tracefs: Clear EVENT_INODE flag in tracefs_drop_inode()
  eventfs: Update all the eventfs_inodes from the events descriptor
  tracefs: Update inode permissions on remount
  eventfs: Keep the directories from having the same inode number as files

10 months agogenirq/irqdesc: Prevent use-after-free in irq_find_at_or_after()
dicken.ding [Fri, 24 May 2024 09:17:39 +0000 (17:17 +0800)]
genirq/irqdesc: Prevent use-after-free in irq_find_at_or_after()

irq_find_at_or_after() dereferences the interrupt descriptor which is
returned by mt_find() while neither holding sparse_irq_lock nor RCU read
lock, which means the descriptor can be freed between mt_find() and the
dereference:

    CPU0                            CPU1
    desc = mt_find()
                                    delayed_free_desc(desc)
    irq_desc_get_irq(desc)

The use-after-free is reported by KASAN:

    Call trace:
     irq_get_next_irq+0x58/0x84
     show_stat+0x638/0x824
     seq_read_iter+0x158/0x4ec
     proc_reg_read_iter+0x94/0x12c
     vfs_read+0x1e0/0x2c8

    Freed by task 4471:
     slab_free_freelist_hook+0x174/0x1e0
     __kmem_cache_free+0xa4/0x1dc
     kfree+0x64/0x128
     irq_kobj_release+0x28/0x3c
     kobject_put+0xcc/0x1e0
     delayed_free_desc+0x14/0x2c
     rcu_do_batch+0x214/0x720

Guard the access with a RCU read lock section.

Fixes: 721255b9826b ("genirq: Use a maple tree for interrupt descriptor management")
Signed-off-by: dicken.ding <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
Cc: [email protected]
Link: https://lore.kernel.org/r/[email protected]
10 months agofs/ntfs3: Break dir enumeration if directory contents error
Konstantin Komarov [Tue, 23 Apr 2024 14:21:58 +0000 (17:21 +0300)]
fs/ntfs3: Break dir enumeration if directory contents error

If we somehow attempt to read beyond the directory size, an error
is supposed to be returned.

However, in some cases, read requests do not stop and instead enter
into a loop.

To avoid this, we set the position in the directory to the end.

Signed-off-by: Konstantin Komarov <[email protected]>
Cc: [email protected]
10 months agofs/ntfs3: Fix case when index is reused during tree transformation
Konstantin Komarov [Tue, 23 Apr 2024 12:31:56 +0000 (15:31 +0300)]
fs/ntfs3: Fix case when index is reused during tree transformation

In most cases when adding a cluster to the directory index,
they are placed at the end, and in the bitmap, this cluster corresponds
to the last bit. The new directory size is calculated as follows:

data_size = (u64)(bit + 1) << indx->index_bits;

In the case of reusing a non-final cluster from the index,
data_size is calculated incorrectly, resulting in the directory size
differing from the actual size.

A check for cluster reuse has been added, and the size update is skipped.

Fixes: 82cae269cfa95 ("fs/ntfs3: Add initialization of super block")
Signed-off-by: Konstantin Komarov <[email protected]>
Cc: [email protected]
10 months agoselftest mm/mseal read-only elf memory segment
Jeff Xu [Mon, 15 Apr 2024 16:35:24 +0000 (16:35 +0000)]
selftest mm/mseal read-only elf memory segment

Sealing read-only of elf mapping so it can't be changed by mprotect.

[[email protected]: style change]
Link: https://lkml.kernel.org/r/[email protected]
[[email protected]: fix linker error for inline function]
Link: https://lkml.kernel.org/r/[email protected]
[[email protected]: fix compile warning]
Link: https://lkml.kernel.org/r/[email protected]
[[email protected]: fix arm build]
Link: https://lkml.kernel.org/r/[email protected]
Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Jeff Xu <[email protected]>
Signed-off-by: Amer Al Shanawany <[email protected]>
Reviewed-by: Kees Cook <[email protected]>
Reviewed-by: Liam R. Howlett <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Guenter Roeck <[email protected]>
Cc: Jann Horn <[email protected]>
Cc: Jeff Xu <[email protected]>
Cc: Jonathan Corbet <[email protected]>
Cc: Jorge Lucangeli Obes <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Matthew Wilcox (Oracle) <[email protected]>
Cc: Muhammad Usama Anjum <[email protected]>
Cc: Pedro Falcato <[email protected]>
Cc: Stephen Röttger <[email protected]>
Cc: Suren Baghdasaryan <[email protected]>
Cc: Amer Al Shanawany <[email protected]>
Cc: Javier Carrasco <[email protected]>
Cc: Shuah Khan <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agomseal: add documentation
Jeff Xu [Mon, 15 Apr 2024 16:35:23 +0000 (16:35 +0000)]
mseal: add documentation

Add documentation for mseal().

Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Jeff Xu <[email protected]>
Reviewed-by: Kees Cook <[email protected]>
Reviewed-by: Liam R. Howlett <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Guenter Roeck <[email protected]>
Cc: Jann Horn <[email protected]>
Cc: Jeff Xu <[email protected]>
Cc: Jonathan Corbet <[email protected]>
Cc: Jorge Lucangeli Obes <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Matthew Wilcox (Oracle) <[email protected]>
Cc: Muhammad Usama Anjum <[email protected]>
Cc: Pedro Falcato <[email protected]>
Cc: Stephen Röttger <[email protected]>
Cc: Suren Baghdasaryan <[email protected]>
Cc: Amer Al Shanawany <[email protected]>
Cc: Javier Carrasco <[email protected]>
Cc: Shuah Khan <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agoselftest mm/mseal memory sealing
Jeff Xu [Mon, 15 Apr 2024 16:35:22 +0000 (16:35 +0000)]
selftest mm/mseal memory sealing

selftest for memory sealing change in mmap() and mseal().

Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Jeff Xu <[email protected]>
Reviewed-by: Kees Cook <[email protected]>
Reviewed-by: Liam R. Howlett <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Guenter Roeck <[email protected]>
Cc: Jann Horn <[email protected]>
Cc: Jeff Xu <[email protected]>
Cc: Jonathan Corbet <[email protected]>
Cc: Jorge Lucangeli Obes <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Matthew Wilcox (Oracle) <[email protected]>
Cc: Muhammad Usama Anjum <[email protected]>
Cc: Pedro Falcato <[email protected]>
Cc: Stephen Röttger <[email protected]>
Cc: Suren Baghdasaryan <[email protected]>
Cc: Amer Al Shanawany <[email protected]>
Cc: Javier Carrasco <[email protected]>
Cc: Shuah Khan <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agomseal: add mseal syscall
Jeff Xu [Mon, 15 Apr 2024 16:35:21 +0000 (16:35 +0000)]
mseal: add mseal syscall

The new mseal() is an syscall on 64 bit CPU, and with following signature:

int mseal(void addr, size_t len, unsigned long flags)
addr/len: memory range.
flags: reserved.

mseal() blocks following operations for the given memory range.

1> Unmapping, moving to another location, and shrinking the size,
   via munmap() and mremap(), can leave an empty space, therefore can
   be replaced with a VMA with a new set of attributes.

2> Moving or expanding a different VMA into the current location,
   via mremap().

3> Modifying a VMA via mmap(MAP_FIXED).

4> Size expansion, via mremap(), does not appear to pose any specific
   risks to sealed VMAs. It is included anyway because the use case is
   unclear. In any case, users can rely on merging to expand a sealed VMA.

5> mprotect() and pkey_mprotect().

6> Some destructive madvice() behaviors (e.g. MADV_DONTNEED) for anonymous
   memory, when users don't have write permission to the memory. Those
   behaviors can alter region contents by discarding pages, effectively a
   memset(0) for anonymous memory.

Following input during RFC are incooperated into this patch:

Jann Horn: raising awareness and providing valuable insights on the
destructive madvise operations.
Linus Torvalds: assisting in defining system call signature and scope.
Liam R. Howlett: perf optimization.
Theo de Raadt: sharing the experiences and insight gained from
  implementing mimmutable() in OpenBSD.

Finally, the idea that inspired this patch comes from Stephen Röttger's
work in Chrome V8 CFI.

[[email protected]: add branch prediction hint, per Pedro]
Link: https://lkml.kernel.org/r/[email protected]
Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Jeff Xu <[email protected]>
Reviewed-by: Kees Cook <[email protected]>
Reviewed-by: Liam R. Howlett <[email protected]>
Cc: Pedro Falcato <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Guenter Roeck <[email protected]>
Cc: Jann Horn <[email protected]>
Cc: Jeff Xu <[email protected]>
Cc: Jonathan Corbet <[email protected]>
Cc: Jorge Lucangeli Obes <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Matthew Wilcox (Oracle) <[email protected]>
Cc: Muhammad Usama Anjum <[email protected]>
Cc: Pedro Falcato <[email protected]>
Cc: Stephen Röttger <[email protected]>
Cc: Suren Baghdasaryan <[email protected]>
Cc: Amer Al Shanawany <[email protected]>
Cc: Javier Carrasco <[email protected]>
Cc: Shuah Khan <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agomseal: wire up mseal syscall
Jeff Xu [Mon, 15 Apr 2024 16:35:20 +0000 (16:35 +0000)]
mseal: wire up mseal syscall

Patch series "Introduce mseal", v10.

This patchset proposes a new mseal() syscall for the Linux kernel.

In a nutshell, mseal() protects the VMAs of a given virtual memory range
against modifications, such as changes to their permission bits.

Modern CPUs support memory permissions, such as the read/write (RW) and
no-execute (NX) bits.  Linux has supported NX since the release of kernel
version 2.6.8 in August 2004 [1].  The memory permission feature improves
the security stance on memory corruption bugs, as an attacker cannot
simply write to arbitrary memory and point the code to it.  The memory
must be marked with the X bit, or else an exception will occur.
Internally, the kernel maintains the memory permissions in a data
structure called VMA (vm_area_struct).  mseal() additionally protects the
VMA itself against modifications of the selected seal type.

Memory sealing is useful to mitigate memory corruption issues where a
corrupted pointer is passed to a memory management system.  For example,
such an attacker primitive can break control-flow integrity guarantees
since read-only memory that is supposed to be trusted can become writable
or .text pages can get remapped.  Memory sealing can automatically be
applied by the runtime loader to seal .text and .rodata pages and
applications can additionally seal security critical data at runtime.  A
similar feature already exists in the XNU kernel with the
VM_FLAGS_PERMANENT [3] flag and on OpenBSD with the mimmutable syscall
[4].  Also, Chrome wants to adopt this feature for their CFI work [2] and
this patchset has been designed to be compatible with the Chrome use case.

Two system calls are involved in sealing the map:  mmap() and mseal().

The new mseal() is an syscall on 64 bit CPU, and with following signature:

int mseal(void addr, size_t len, unsigned long flags)
addr/len: memory range.
flags: reserved.

mseal() blocks following operations for the given memory range.

1> Unmapping, moving to another location, and shrinking the size,
   via munmap() and mremap(), can leave an empty space, therefore can
   be replaced with a VMA with a new set of attributes.

2> Moving or expanding a different VMA into the current location,
   via mremap().

3> Modifying a VMA via mmap(MAP_FIXED).

4> Size expansion, via mremap(), does not appear to pose any specific
   risks to sealed VMAs. It is included anyway because the use case is
   unclear. In any case, users can rely on merging to expand a sealed VMA.

5> mprotect() and pkey_mprotect().

6> Some destructive madvice() behaviors (e.g. MADV_DONTNEED) for anonymous
   memory, when users don't have write permission to the memory. Those
   behaviors can alter region contents by discarding pages, effectively a
   memset(0) for anonymous memory.

The idea that inspired this patch comes from Stephen Röttger’s work in
V8 CFI [5].  Chrome browser in ChromeOS will be the first user of this
API.

Indeed, the Chrome browser has very specific requirements for sealing,
which are distinct from those of most applications.  For example, in the
case of libc, sealing is only applied to read-only (RO) or read-execute
(RX) memory segments (such as .text and .RELRO) to prevent them from
becoming writable, the lifetime of those mappings are tied to the lifetime
of the process.

Chrome wants to seal two large address space reservations that are managed
by different allocators.  The memory is mapped RW- and RWX respectively
but write access to it is restricted using pkeys (or in the future ARM
permission overlay extensions).  The lifetime of those mappings are not
tied to the lifetime of the process, therefore, while the memory is
sealed, the allocators still need to free or discard the unused memory.
For example, with madvise(DONTNEED).

However, always allowing madvise(DONTNEED) on this range poses a security
risk.  For example if a jump instruction crosses a page boundary and the
second page gets discarded, it will overwrite the target bytes with zeros
and change the control flow.  Checking write-permission before the discard
operation allows us to control when the operation is valid.  In this case,
the madvise will only succeed if the executing thread has PKEY write
permissions and PKRU changes are protected in software by control-flow
integrity.

Although the initial version of this patch series is targeting the Chrome
browser as its first user, it became evident during upstream discussions
that we would also want to ensure that the patch set eventually is a
complete solution for memory sealing and compatible with other use cases.
The specific scenario currently in mind is glibc's use case of loading and
sealing ELF executables.  To this end, Stephen is working on a change to
glibc to add sealing support to the dynamic linker, which will seal all
non-writable segments at startup.  Once this work is completed, all
applications will be able to automatically benefit from these new
protections.

In closing, I would like to formally acknowledge the valuable
contributions received during the RFC process, which were instrumental in
shaping this patch:

Jann Horn: raising awareness and providing valuable insights on the
  destructive madvise operations.
Liam R. Howlett: perf optimization.
Linus Torvalds: assisting in defining system call signature and scope.
Theo de Raadt: sharing the experiences and insight gained from
  implementing mimmutable() in OpenBSD.

MM perf benchmarks
==================
This patch adds a loop in the mprotect/munmap/madvise(DONTNEED) to
check the VMAs’ sealing flag, so that no partial update can be made,
when any segment within the given memory range is sealed.

To measure the performance impact of this loop, two tests are developed.
[8]

The first is measuring the time taken for a particular system call,
by using clock_gettime(CLOCK_MONOTONIC). The second is using
PERF_COUNT_HW_REF_CPU_CYCLES (exclude user space). Both tests have
similar results.

The tests have roughly below sequence:
for (i = 0; i < 1000, i++)
    create 1000 mappings (1 page per VMA)
    start the sampling
    for (j = 0; j < 1000, j++)
        mprotect one mapping
    stop and save the sample
    delete 1000 mappings
calculates all samples.

Below tests are performed on Intel(R) Pentium(R) Gold 7505 @ 2.00GHz,
4G memory, Chromebook.

Based on the latest upstream code:
The first test (measuring time)
syscall__ vmas t t_mseal delta_ns per_vma %
munmap__   1 909 944 35 35 104%
munmap__   2 1398 1502 104 52 107%
munmap__   4 2444 2594 149 37 106%
munmap__   8 4029 4323 293 37 107%
munmap__   16 6647 6935 288 18 104%
munmap__   32 11811 12398 587 18 105%
mprotect 1 439 465 26 26 106%
mprotect 2 1659 1745 86 43 105%
mprotect 4 3747 3889 142 36 104%
mprotect 8 6755 6969 215 27 103%
mprotect 16 13748 14144 396 25 103%
mprotect 32 27827 28969 1142 36 104%
madvise_ 1 240 262 22 22 109%
madvise_ 2 366 442 76 38 121%
madvise_ 4 623 751 128 32 121%
madvise_ 8 1110 1324 215 27 119%
madvise_ 16 2127 2451 324 20 115%
madvise_ 32 4109 4642 534 17 113%

The second test (measuring cpu cycle)
syscall__ vmas cpu cmseal delta_cpu per_vma %
munmap__ 1 1790 1890 100 100 106%
munmap__ 2 2819 3033 214 107 108%
munmap__ 4 4959 5271 312 78 106%
munmap__ 8 8262 8745 483 60 106%
munmap__ 16 13099 14116 1017 64 108%
munmap__ 32 23221 24785 1565 49 107%
mprotect 1 906 967 62 62 107%
mprotect 2 3019 3203 184 92 106%
mprotect 4 6149 6569 420 105 107%
mprotect 8 9978 10524 545 68 105%
mprotect 16 20448 21427 979 61 105%
mprotect 32 40972 42935 1963 61 105%
madvise_ 1 434 497 63 63 115%
madvise_ 2 752 899 147 74 120%
madvise_ 4 1313 1513 200 50 115%
madvise_ 8 2271 2627 356 44 116%
madvise_ 16 4312 4883 571 36 113%
madvise_ 32 8376 9319 943 29 111%

Based on the result, for 6.8 kernel, sealing check adds
20-40 nano seconds, or around 50-100 CPU cycles, per VMA.

In addition, I applied the sealing to 5.10 kernel:
The first test (measuring time)
syscall__ vmas t tmseal delta_ns per_vma %
munmap__ 1 357 390 33 33 109%
munmap__ 2 442 463 21 11 105%
munmap__ 4 614 634 20 5 103%
munmap__ 8 1017 1137 120 15 112%
munmap__ 16 1889 2153 263 16 114%
munmap__ 32 4109 4088 -21 -1 99%
mprotect 1 235 227 -7 -7 97%
mprotect 2 495 464 -30 -15 94%
mprotect 4 741 764 24 6 103%
mprotect 8 1434 1437 2 0 100%
mprotect 16 2958 2991 33 2 101%
mprotect 32 6431 6608 177 6 103%
madvise_ 1 191 208 16 16 109%
madvise_ 2 300 324 24 12 108%
madvise_ 4 450 473 23 6 105%
madvise_ 8 753 806 53 7 107%
madvise_ 16 1467 1592 125 8 108%
madvise_ 32 2795 3405 610 19 122%

The second test (measuring cpu cycle)
syscall__ nbr_vma cpu cmseal delta_cpu per_vma %
munmap__ 1 684 715 31 31 105%
munmap__ 2 861 898 38 19 104%
munmap__ 4 1183 1235 51 13 104%
munmap__ 8 1999 2045 46 6 102%
munmap__ 16 3839 3816 -23 -1 99%
munmap__ 32 7672 7887 216 7 103%
mprotect 1 397 443 46 46 112%
mprotect 2 738 788 50 25 107%
mprotect 4 1221 1256 35 9 103%
mprotect 8 2356 2429 72 9 103%
mprotect 16 4961 4935 -26 -2 99%
mprotect 32 9882 10172 291 9 103%
madvise_ 1 351 380 29 29 108%
madvise_ 2 565 615 49 25 109%
madvise_ 4 872 933 61 15 107%
madvise_ 8 1508 1640 132 16 109%
madvise_ 16 3078 3323 245 15 108%
madvise_ 32 5893 6704 811 25 114%

For 5.10 kernel, sealing check adds 0-15 ns in time, or 10-30
CPU cycles, there is even decrease in some cases.

It might be interesting to compare 5.10 and 6.8 kernel
The first test (measuring time)
syscall__ vmas t_5_10 t_6_8 delta_ns per_vma %
munmap__ 1 357 909 552 552 254%
munmap__ 2 442 1398 956 478 316%
munmap__ 4 614 2444 1830 458 398%
munmap__ 8 1017 4029 3012 377 396%
munmap__ 16 1889 6647 4758 297 352%
munmap__ 32 4109 11811 7702 241 287%
mprotect 1 235 439 204 204 187%
mprotect 2 495 1659 1164 582 335%
mprotect 4 741 3747 3006 752 506%
mprotect 8 1434 6755 5320 665 471%
mprotect 16 2958 13748 10790 674 465%
mprotect 32 6431 27827 21397 669 433%
madvise_ 1 191 240 49 49 125%
madvise_ 2 300 366 67 33 122%
madvise_ 4 450 623 173 43 138%
madvise_ 8 753 1110 357 45 147%
madvise_ 16 1467 2127 660 41 145%
madvise_ 32 2795 4109 1314 41 147%

The second test (measuring cpu cycle)
syscall__ vmas cpu_5_10 c_6_8 delta_cpu per_vma %
munmap__ 1 684 1790 1106 1106 262%
munmap__ 2 861 2819 1958 979 327%
munmap__ 4 1183 4959 3776 944 419%
munmap__ 8 1999 8262 6263 783 413%
munmap__ 16 3839 13099 9260 579 341%
munmap__ 32 7672 23221 15549 486 303%
mprotect 1 397 906 509 509 228%
mprotect 2 738 3019 2281 1140 409%
mprotect 4 1221 6149 4929 1232 504%
mprotect 8 2356 9978 7622 953 423%
mprotect 16 4961 20448 15487 968 412%
mprotect 32 9882 40972 31091 972 415%
madvise_ 1 351 434 82 82 123%
madvise_ 2 565 752 186 93 133%
madvise_ 4 872 1313 442 110 151%
madvise_ 8 1508 2271 763 95 151%
madvise_ 16 3078 4312 1234 77 140%
madvise_ 32 5893 8376 2483 78 142%

From 5.10 to 6.8
munmap: added 250-550 ns in time, or 500-1100 in cpu cycle, per vma.
mprotect: added 200-750 ns in time, or 500-1200 in cpu cycle, per vma.
madvise: added 33-50 ns in time, or 70-110 in cpu cycle, per vma.

In comparison to mseal, which adds 20-40 ns or 50-100 CPU cycles, the
increase from 5.10 to 6.8 is significantly larger, approximately ten times
greater for munmap and mprotect.

When I discuss the mm performance with Brian Makin, an engineer who worked
on performance, it was brought to my attention that such performance
benchmarks, which measuring millions of mm syscall in a tight loop, may
not accurately reflect real-world scenarios, such as that of a database
service.  Also this is tested using a single HW and ChromeOS, the data
from another HW or distribution might be different.  It might be best to
take this data with a grain of salt.

This patch (of 5):

Wire up mseal syscall for all architectures.

Link: https://lkml.kernel.org/r/[email protected]
Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Jeff Xu <[email protected]>
Reviewed-by: Kees Cook <[email protected]>
Reviewed-by: Liam R. Howlett <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Guenter Roeck <[email protected]>
Cc: Jann Horn <[email protected]> [Bug #2]
Cc: Jeff Xu <[email protected]>
Cc: Jonathan Corbet <[email protected]>
Cc: Jorge Lucangeli Obes <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Matthew Wilcox (Oracle) <[email protected]>
Cc: Muhammad Usama Anjum <[email protected]>
Cc: Pedro Falcato <[email protected]>
Cc: Stephen Röttger <[email protected]>
Cc: Suren Baghdasaryan <[email protected]>
Cc: Amer Al Shanawany <[email protected]>
Cc: Javier Carrasco <[email protected]>
Cc: Shuah Khan <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
10 months agoMerge tag 'nfs-for-6.10-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
Linus Torvalds [Thu, 23 May 2024 20:51:09 +0000 (13:51 -0700)]
Merge tag 'nfs-for-6.10-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs

Pull NFS client updates from Trond Myklebust:
 "Stable fixes:
   - nfs: fix undefined behavior in nfs_block_bits()
   - NFSv4.2: Fix READ_PLUS when server doesn't support OP_READ_PLUS

  Bugfixes:
   - Fix mixing of the lock/nolock and local_lock mount options
   - NFSv4: Fixup smatch warning for ambiguous return
   - NFSv3: Fix remount when using the legacy binary mount api
   - SUNRPC: Fix the handling of expired RPCSEC_GSS contexts
   - SUNRPC: fix the NFSACL RPC retries when soft mounts are enabled
   - rpcrdma: fix handling for RDMA_CM_EVENT_DEVICE_REMOVAL

  Features and cleanups:
   - NFSv3: Use the atomic_open API to fix open(O_CREAT|O_TRUNC)
   - pNFS/filelayout: S layout segment range in LAYOUTGET
   - pNFS: rework pnfs_generic_pg_check_layout to check IO range
   - NFSv2: Turn off enabling of NFS v2 by default"

* tag 'nfs-for-6.10-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs:
  nfs: fix undefined behavior in nfs_block_bits()
  pNFS: rework pnfs_generic_pg_check_layout to check IO range
  pNFS/filelayout: check layout segment range
  pNFS/filelayout: fixup pNfs allocation modes
  rpcrdma: fix handling for RDMA_CM_EVENT_DEVICE_REMOVAL
  NFS: Don't enable NFS v2 by default
  NFS: Fix READ_PLUS when server doesn't support OP_READ_PLUS
  sunrpc: fix NFSACL RPC retry on soft mount
  SUNRPC: fix handling expired GSS context
  nfs: keep server info for remounts
  NFSv4: Fixup smatch warning for ambiguous return
  NFS: make sure lock/nolock overriding local_lock mount option
  NFS: add atomic_open for NFSv3 to handle O_TRUNC correctly.
  pNFS/filelayout: Specify the layout segment range in LAYOUTGET
  pNFS/filelayout: Remove the whole file layout requirement

10 months agoMerge tag 'block-6.10-20240523' of git://git.kernel.dk/linux
Linus Torvalds [Thu, 23 May 2024 20:44:47 +0000 (13:44 -0700)]
Merge tag 'block-6.10-20240523' of git://git.kernel.dk/linux

Pull more block updates from Jens Axboe:
 "Followup block updates, mostly due to NVMe being a bit late to the
  party. But nothing major in there, so not a big deal.

  In detail, this contains:

   - NVMe pull request via Keith:
       - Fabrics connection retries (Daniel, Hannes)
       - Fabrics logging enhancements (Tokunori)
       - RDMA delete optimization (Sagi)

   - ublk DMA alignment fix (me)

   - null_blk sparse warning fixes (Bart)

   - Discard support for brd (Keith)

   - blk-cgroup list corruption fixes (Ming)

   - blk-cgroup stat propagation fix (Waiman)

   - Regression fix for plugging stall with md (Yu)

   - Misc fixes or cleanups (David, Jeff, Justin)"

* tag 'block-6.10-20240523' of git://git.kernel.dk/linux: (24 commits)
  null_blk: fix null-ptr-dereference while configuring 'power' and 'submit_queues'
  blk-throttle: remove unused struct 'avg_latency_bucket'
  block: fix lost bio for plug enabled bio based device
  block: t10-pi: add MODULE_DESCRIPTION()
  blk-mq: add helper for checking if one CPU is mapped to specified hctx
  blk-cgroup: Properly propagate the iostat update up the hierarchy
  blk-cgroup: fix list corruption from reorder of WRITE ->lqueued
  blk-cgroup: fix list corruption from resetting io stat
  cdrom: rearrange last_media_change check to avoid unintentional overflow
  nbd: Fix signal handling
  nbd: Remove a local variable from nbd_send_cmd()
  nbd: Improve the documentation of the locking assumptions
  nbd: Remove superfluous casts
  nbd: Use NULL to represent a pointer
  brd: implement discard support
  null_blk: Fix two sparse warnings
  ublk_drv: set DMA alignment mask to 3
  nvme-rdma, nvme-tcp: include max reconnects for reconnect logging
  nvmet-rdma: Avoid o(n^2) loop in delete_ctrl
  nvme: do not retry authentication failures
  ...

10 months agoMerge tag 'io_uring-6.10-20240523' of git://git.kernel.dk/linux
Linus Torvalds [Thu, 23 May 2024 20:41:49 +0000 (13:41 -0700)]
Merge tag 'io_uring-6.10-20240523' of git://git.kernel.dk/linux

Pull io_uring fixes from Jens Axboe:
 "Single fix here for a regression in 6.9, and then a simple cleanup
  removing some dead code"

* tag 'io_uring-6.10-20240523' of git://git.kernel.dk/linux:
  io_uring: remove checks for NULL 'sq_offset'
  io_uring/sqpoll: ensure that normal task_work is also run timely

10 months agoMerge tag 'regulator-fix-v6.10-merge-window' of git://git.kernel.org/pub/scm/linux...
Linus Torvalds [Thu, 23 May 2024 20:39:42 +0000 (13:39 -0700)]
Merge tag 'regulator-fix-v6.10-merge-window' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regulator

Pull regulator fixes from Mark Brown:
 "A bunch of fixes that came in during the merge window.

  Matti found several issues with some of the more complexly configured
  Rohm regulators and the helpers they use and there were some errors in
  the specification of tps6594 when regulators are grouped together"

* tag 'regulator-fix-v6.10-merge-window' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regulator:
  regulator: tps6594-regulator: Correct multi-phase configuration
  regulator: tps6287x: Force writing VSEL bit
  regulator: pickable ranges: don't always cache vsel
  regulator: rohm-regulator: warn if unsupported voltage is set
  regulator: bd71828: Don't overwrite runtime voltages

This page took 0.164512 seconds and 4 git commands to generate.