riscv: detect assembler support for .option arch

Some extensions use .option arch directive to selectively enable certain
extensions in parts of its assembly code. For example, Zbb uses it to
inform assmebler to emit bit manipulation instructions. However,
supporting of this directive only exist on GNU assembler and has not
landed on clang at the moment, making TOOLCHAIN_HAS_ZBB depend on
AS_IS_GNU.

While it is still under review at https://reviews.llvm.org/D123515, the
upcoming Vector patch also requires this feature in assembler. Thus,
provide Kconfig AS_HAS_OPTION_ARCH to detect such feature. Then
TOOLCHAIN_HAS_XXX will be turned on automatically when the feature land.

Suggested-by: Nathan Chancellor <[email protected]>
Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Conor Dooley <[email protected]>
Reviewed-by: Nathan Chancellor <[email protected]>
Reviewed-by: Heiko Stuebner <[email protected]>
Tested-by: Heiko Stuebner <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: Add sysctl to set the default vector rule for new processes

To support Vector extension, the series exports variable-length vector
registers on the signal frame. However, this potentially breaks abi if
processing vector registers is required in the signal handler for old
binaries. For example, there is such need if user-level context switch
is triggerred via signals[1].

For this reason, it is best to leave a decision to distro maintainers,
where the enablement of userspace Vector for new launching programs can
be controlled. Developers may also need the switch to experiment with.
The parameter is configurable through sysctl interface so a distro may
turn off Vector early at init script if the break really happens in the
wild.

The switch will only take effects on new execve() calls once set. This
will not effect existing processes that do not call execve(), nor
processes which has been set with a non-default vstate_ctrl by making
explicit PR_RISCV_V_SET_CONTROL prctl() calls.

Link: https://lore.kernel.org/all/[email protected]/
Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Greentime Hu <[email protected]>
Reviewed-by: Vincent Chen <[email protected]>
Reviewed-by: Björn Töpel <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: Add prctl controls for userspace vector management

This patch add two riscv-specific prctls, to allow usespace control the
use of vector unit:

 * PR_RISCV_V_SET_CONTROL: control the permission to use Vector at next,
   or all following execve for a thread. Turning off a thread's Vector
   live is not possible since libraries may have registered ifunc that
   may execute Vector instructions.
 * PR_RISCV_V_GET_CONTROL: get the same permission setting for the
   current thread, and the setting for following execve(s).

Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Greentime Hu <[email protected]>
Reviewed-by: Vincent Chen <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: hwcap: change ELF_HWCAP to a function

Using a function is flexible to represent ELF_HWCAP. So the kernel may
encode hwcap reflecting supported hardware features just at the moment of
the start of each program.

This will be helpful when we introduce prctl/sysctl interface to control
per-process availability of Vector extension in following patches.
Programs started with V disabled should see V masked off in theirs
ELF_HWCAP.

Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Conor Dooley <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: KVM: Add vector lazy save/restore support

This patch adds vector context save/restore for guest VCPUs. To reduce the
impact on KVM performance, the implementation imitates the FP context
switch mechanism to lazily store and restore the vector context only when
the kernel enters/exits the in-kernel run loop and not during the KVM
world switch.

Signed-off-by: Vincent Chen <[email protected]>
Signed-off-by: Greentime Hu <[email protected]>
Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
Acked-by: Anup Patel <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: kvm: Add V extension to KVM ISA

Add V extension to KVM isa extension list to enable supporting of V
extension on VCPUs.

Signed-off-by: Vincent Chen <[email protected]>
Signed-off-by: Greentime Hu <[email protected]>
Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Conor Dooley <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
Acked-by: Anup Patel <[email protected]>
Reviewed-by: Heiko Stuebner <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: prevent stack corruption by reserving task_pt_regs(p) early

Early function calls, such as setup_vm(), relocate_enable_mmu(),
soc_early_init() etc, are free to operate on stack. However,
PT_SIZE_ON_STACK bytes at the head of the kernel stack are purposedly
reserved for the placement of per-task register context pointed by
task_pt_regs(p). Those functions may corrupt task_pt_regs if we overlap
the $sp with it. In fact, we had accidentally corrupted sstatus.VS in some
tests, treating the kernel to save V context before V was actually
allocated, resulting in a kernel panic.

Thus, we should skip PT_SIZE_ON_STACK for $sp before making C function
calls from the top-level assembly.

Co-developed-by: ShihPo Hung <[email protected]>
Signed-off-by: ShihPo Hung <[email protected]>
Co-developed-by: Vincent Chen <[email protected]>
Signed-off-by: Vincent Chen <[email protected]>
Signed-off-by: Greentime Hu <[email protected]>
Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Conor Dooley <[email protected]>
Reviewed-by: Heiko Stuebner <[email protected]>
Tested-by: Heiko Stuebner <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: signal: validate altstack to reflect Vector

Some extensions, such as Vector, dynamically change footprint on a
signal frame, so MINSIGSTKSZ is no longer accurate. For example, an
RV64V implementation with vlen = 512 may occupy 2K + 40 + 12 Bytes of a
signal frame with the upcoming support. And processes that do not
execute any vector instructions do not need to reserve the extra
sigframe. So we need a way to guard the allocation size of the sigframe
at process runtime according to current status of V.

Thus, provide the function sigaltstack_size_valid() to validate its size
based on current allocation status of supported extensions.

Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Conor Dooley <[email protected]>
Reviewed-by: Heiko Stuebner <[email protected]>
Tested-by: Heiko Stuebner <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: signal: Report signal frame size to userspace via auxv

The vector register belongs to the signal context. They need to be stored
and restored as entering and leaving the signal handler. According to the
V-extension specification, the maximum length of the vector registers can
be 2^16. Hence, if userspace refers to the MINSIGSTKSZ to create a
sigframe, it may not be enough. To resolve this problem, this patch refers
to the commit 94b07c1f8c39c
("arm64: signal: Report signal frame size to userspace via auxv") to enable
userspace to know the minimum required sigframe size through the auxiliary
vector and use it to allocate enough memory for signal context.

Note that auxv always reports size of the sigframe as if V exists for
all starting processes, whenever the kernel has CONFIG_RISCV_ISA_V. The
reason is that users usually reference this value to allocate an
alternative signal stack, and the user may use V anytime. So the user
must reserve a space for V-context in sigframe in case that the signal
handler invokes after the kernel allocating V.

Signed-off-by: Greentime Hu <[email protected]>
Signed-off-by: Vincent Chen <[email protected]>
Signed-off-by: Andy Chiu <[email protected]>
Acked-by: Conor Dooley <[email protected]>
Reviewed-by: Björn Töpel <[email protected]>
Reviewed-by: Guo Ren <[email protected]>
Reviewed-by: Heiko Stuebner <[email protected]>
Tested-by: Heiko Stuebner <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: signal: Add sigcontext save/restore for vector

This patch facilitates the existing fp-reserved words for placement of
the first extension's context header on the user's sigframe. A context
header consists of a distinct magic word and the size, including the
header itself, of an extension on the stack. Then, the frame is followed
by the context of that extension, and then a header + context body for
another extension if exists. If there is no more extension to come, then
the frame must be ended with a null context header. A special case is
rv64gc, where the kernel support no extensions requiring to expose
additional regfile to the user. In such case the kernel would place the
null context header right after the first reserved word of
__riscv_q_ext_state when saving sigframe. And the kernel would check if
all reserved words are zeros when a signal handler returns.

__riscv_q_ext_state---->| |<-__riscv_extra_ext_header
~ ~
.reserved[0]--->|0 |<- .reserved
<-------|magic |<- .hdr
| |size |_______ end of sc_fpregs
| |ext-bdy|
| ~ ~
+)size ------->|magic |<- another context header
|size |
|ext-bdy|
~ ~
|magic:0|<- null context header
|size:0 |

The vector registers will be saved in datap pointer. The datap pointer
will be allocated dynamically when the task needs in kernel space. On
the other hand, datap pointer on the sigframe will be set right after
the __riscv_v_ext_state data structure.

Co-developed-by: Vincent Chen <[email protected]>
Signed-off-by: Vincent Chen <[email protected]>
Signed-off-by: Greentime Hu <[email protected]>
Suggested-by: Vineet Gupta <[email protected]>
Suggested-by: Richard Henderson <[email protected]>
Co-developed-by: Andy Chiu <[email protected]>
Signed-off-by: Andy Chiu <[email protected]>
Acked-by: Conor Dooley <[email protected]>
Acked-by: Heiko Stuebner <[email protected]>
Tested-by: Heiko Stuebner <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: signal: check fp-reserved words unconditionally

In order to let kernel/user locate and identify an extension context on
the existing sigframe, we are going to utilize reserved space of fp and
encode the information there. And since the sigcontext has already
preserved a space for fp context w or w/o CONFIG_FPU, we move those
reserved words checking/setting routine back into generic code.

This commit also undone an additional logical change carried by the
refactor commit 007f5c3589578
("Refactor FPU code in signal setup/return procedures"). Originally we
did not restore fp context if restoring of gpr have failed. And it was
fine on the other side. In such way the kernel could keep the regfiles
intact, and potentially react at the failing point of restore.

Signed-off-by: Andy Chiu <[email protected]>
Acked-by: Conor Dooley <[email protected]>
Acked-by: Heiko Stuebner <[email protected]>
Tested-by: Heiko Stuebner <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: Add ptrace vector support

This patch adds ptrace support for riscv vector. The vector registers will
be saved in datap pointer of __riscv_v_ext_state. This pointer will be set
right after the __riscv_v_ext_state data structure then it will be put in
ubuf for ptrace system call to get or set. It will check if the datap got
from ubuf is set to the correct address or not when the ptrace system call
is trying to set the vector registers.

Co-developed-by: Vincent Chen <[email protected]>
Signed-off-by: Vincent Chen <[email protected]>
Signed-off-by: Greentime Hu <[email protected]>
Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Conor Dooley <[email protected]>
Reviewed-by: Palmer Dabbelt <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: Allocate user's vector context in the first-use trap

Vector unit is disabled by default for all user processes. Thus, a
process will take a trap (illegal instruction) into kernel at the first
time when it uses Vector. Only after then, the kernel allocates V
context and starts take care of the context for that user process.

Suggested-by: Richard Henderson <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Conor Dooley <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: Add task switch support for vector

This patch adds task switch support for vector. It also supports all
lengths of vlen.

Suggested-by: Andrew Waterman <[email protected]>
Co-developed-by: Nick Knight <[email protected]>
Signed-off-by: Nick Knight <[email protected]>
Co-developed-by: Guo Ren <[email protected]>
Signed-off-by: Guo Ren <[email protected]>
Co-developed-by: Vincent Chen <[email protected]>
Signed-off-by: Vincent Chen <[email protected]>
Co-developed-by: Ruinland Tsai <[email protected]>
Signed-off-by: Ruinland Tsai <[email protected]>
Signed-off-by: Greentime Hu <[email protected]>
Signed-off-by: Vineet Gupta <[email protected]>
Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Conor Dooley <[email protected]>
Reviewed-by: Björn Töpel <[email protected]>
Reviewed-by: Heiko Stuebner <[email protected]>
Tested-by: Heiko Stuebner <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: Introduce struct/helpers to save/restore per-task Vector state

Add vector state context struct to be added later in thread_struct. And
prepare low-level helper functions to save/restore vector contexts.

This include Vector Regfile and CSRs holding dynamic configuration state
(vstart, vl, vtype, vcsr). The Vec Register width could be implementation
defined, but same for all processes, so that is saved separately.

This is not yet wired into final thread_struct - will be done when
__switch_to actually starts doing this in later patches.

Given the variable (and potentially large) size of regfile, they are
saved in dynamically allocated memory, pointed to by datap pointer in
__riscv_v_ext_state.

Co-developed-by: Vincent Chen <[email protected]>
Signed-off-by: Vincent Chen <[email protected]>
Signed-off-by: Greentime Hu <[email protected]>
Signed-off-by: Vineet Gupta <[email protected]>
Signed-off-by: Andy Chiu <[email protected]>
Acked-by: Conor Dooley <[email protected]>
Reviewed-by: Guo Ren <[email protected]>
Reviewed-by: Björn Töpel <[email protected]>
Reviewed-by: Heiko Stuebner <[email protected]>
Tested-by: Heiko Stuebner <[email protected]>
Reviewed-by: Palmer Dabbelt <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: Introduce riscv_v_vsize to record size of Vector context

This patch is used to detect the size of CPU vector registers and use
riscv_v_vsize to save the size of all the vector registers. It assumes all
harts has the same capabilities in a SMP system. If a core detects VLENB
that is different from the boot core, then it warns and turns off V
support for user space.

Co-developed-by: Guo Ren <[email protected]>
Signed-off-by: Guo Ren <[email protected]>
Co-developed-by: Vincent Chen <[email protected]>
Signed-off-by: Vincent Chen <[email protected]>
Signed-off-by: Greentime Hu <[email protected]>
Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Conor Dooley <[email protected]>
Reviewed-by: Heiko Stuebner <[email protected]>
Tested-by: Heiko Stuebner <[email protected]>
Reviewed-by: Palmer Dabbelt <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: Introduce Vector enable/disable helpers

These are small and likely to be frequently called so implement as
inline routines (vs. function call).

Co-developed-by: Guo Ren <[email protected]>
Signed-off-by: Guo Ren <[email protected]>
Co-developed-by: Vincent Chen <[email protected]>
Signed-off-by: Vincent Chen <[email protected]>
Signed-off-by: Greentime Hu <[email protected]>
Signed-off-by: Vineet Gupta <[email protected]>
Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Conor Dooley <[email protected]>
Reviewed-by: Heiko Stuebner <[email protected]>
Tested-by: Heiko Stuebner <[email protected]>
Reviewed-by: Palmer Dabbelt <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: Disable Vector Instructions for kernel itself

Disable vector instructions execution for kernel mode at its entrances.
This helps find illegal uses of vector in the kernel space, which is
similar to the fpu.

Signed-off-by: Guo Ren <[email protected]>
Co-developed-by: Vincent Chen <[email protected]>
Signed-off-by: Vincent Chen <[email protected]>
Co-developed-by: Han-Kuan Chen <[email protected]>
Signed-off-by: Han-Kuan Chen <[email protected]>
Co-developed-by: Greentime Hu <[email protected]>
Signed-off-by: Greentime Hu <[email protected]>
Signed-off-by: Vineet Gupta <[email protected]>
Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Conor Dooley <[email protected]>
Reviewed-by: Heiko Stuebner <[email protected]>
Tested-by: Heiko Stuebner <[email protected]>
Reviewed-by: Palmer Dabbelt <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: Clear vector regfile on bootup

clear vector registers on boot if kernel supports V.

Signed-off-by: Greentime Hu <[email protected]>
Signed-off-by: Vineet Gupta <[email protected]>
Signed-off-by: Andy Chiu <[email protected]>
Acked-by: Conor Dooley <[email protected]>
Reviewed-by: Heiko Stuebner <[email protected]>
Tested-by: Heiko Stuebner <[email protected]>
Reviewed-by: Palmer Dabbelt <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: Add new csr defines related to vector extension

Follow the riscv vector spec to add new csr numbers.

Acked-by: Guo Ren <[email protected]>
Co-developed-by: Guo Ren <[email protected]>
Signed-off-by: Guo Ren <[email protected]>
Co-developed-by: Vincent Chen <[email protected]>
Signed-off-by: Vincent Chen <[email protected]>
Signed-off-by: Greentime Hu <[email protected]>
Reviewed-by: Palmer Dabbelt <[email protected]>
Suggested-by: Vineet Gupta <[email protected]>
Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Conor Dooley <[email protected]>
Reviewed-by: Heiko Stuebner <[email protected]>
Tested-by: Heiko Stuebner <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: hwprobe: Add support for probing V in RISCV_HWPROBE_KEY_IMA_EXT_0

Probing kernel support for Vector extension is available now. This only
add detection for V only. Extenions like Zvfh, Zk are not in this scope.

Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Conor Dooley <[email protected]>
Reviewed-by: Evan Green <[email protected]>
Reviewed-by: Palmer Dabbelt <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: Extending cpufeature.c to detect V-extension

Add V-extension into riscv_isa_ext_keys array and detect it with isa
string parsing.

Signed-off-by: Guo Ren <[email protected]>
Signed-off-by: Guo Ren <[email protected]>
Signed-off-by: Greentime Hu <[email protected]>
Suggested-by: Vineet Gupta <[email protected]>
Co-developed-by: Andy Chiu <[email protected]>
Signed-off-by: Andy Chiu <[email protected]>
Reviewed-by: Conor Dooley <[email protected]>
Reviewed-by: Heiko Stuebner <[email protected]>
Tested-by: Heiko Stuebner <[email protected]>
Reviewed-by: Palmer Dabbelt <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

riscv: Rename __switch_to_aux() -> fpu

The name of __switch_to_aux() is not clear and rename it with the
determine function: __switch_to_fpu(). Next we could add other regs'
switch.

Signed-off-by: Guo Ren <[email protected]>
Signed-off-by: Guo Ren <[email protected]>
Signed-off-by: Greentime Hu <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
Reviewed-by: Palmer Dabbelt <[email protected]>
Signed-off-by: Andy Chiu <[email protected]>
Tested-by: Heiko Stuebner <[email protected]>
Reviewed-by: Heiko Stuebner <[email protected]>
Reviewed-by: Conor Dooley <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Palmer Dabbelt <[email protected]>

Linux 6.4-rc1

Merge tag 'perf-tools-for-v6.4-3-2023-05-06' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux

Pull perf tool updates from Arnaldo Carvalho de Melo:
 "Third version of perf tool updates, with the build problems with with
  using a 'vmlinux.h' generated from the main build fixed, and the bpf
  skeleton build disabled by default.

  Build:

   - Require libtraceevent to build, one can disable it using
     NO_LIBTRACEEVENT=1.

     It is required for tools like 'perf sched', 'perf kvm', 'perf
     trace', etc.

     libtraceevent is available in most distros so installing
     'libtraceevent-devel' should be a one-time event to continue
     building perf as usual.

     Using NO_LIBTRACEEVENT=1 produces tooling that is functional and
     sufficient for lots of users not interested in those libtraceevent
     dependent features.

   - Allow Python support in 'perf script' when libtraceevent isn't
     linked, as not all features requires it, for instance Intel PT does
     not use tracepoints.

   - Error if the python interpreter needed for jevents to work isn't
     available and NO_JEVENTS=1 isn't set, preventing a build without
     support for JSON vendor events, which is a rare but possible
     condition. The two check error messages:

        $(error ERROR: No python interpreter needed for jevents generation. Install python or build with NO_JEVENTS=1.)
        $(error ERROR: Python interpreter needed for jevents generation too old (older than 3.6). Install a newer python or build with NO_JEVENTS=1.)

   - Make libbpf 1.0 the minimum required when building with out of
     tree, distro provided libbpf.

   - Use libsdtc++'s and LLVM's libcxx's __cxa_demangle, a portable C++
     demangler, add 'perf test' entry for it.

   - Make binutils libraries opt in, as distros disable building with it
     due to licensing, they were used for C++ demangling, for instance.

   - Switch libpfm4 to opt-out rather than opt-in, if libpfm-devel (or
     equivalent) isn't installed, we'll just have a build warning:

       Makefile.config:1144: libpfm4 not found, disables libpfm4 support. Please install libpfm4-dev

   - Add a feature test for scandirat(), that is not implemented so far
     in musl and uclibc, disabling features that need it, such as
     scanning for tracepoints in /sys/kernel/tracing/events.

  perf BPF filters:

   - New feature where BPF can be used to filter samples, for instance:

      $ sudo ./perf record -e cycles --filter 'period > 1000' true
      $ sudo ./perf script
           perf-exec 2273949 546850.708501:       5029 cycles:  ffffffff826f9e25 finish_wait+0x5 ([kernel.kallsyms])
           perf-exec 2273949 546850.708508:      32409 cycles:  ffffffff826f9e25 finish_wait+0x5 ([kernel.kallsyms])
           perf-exec 2273949 546850.708526:     143369 cycles:  ffffffff82b4cdbf xas_start+0x5f ([kernel.kallsyms])
           perf-exec 2273949 546850.708600:     372650 cycles:  ffffffff8286b8f7 __pagevec_lru_add+0x117 ([kernel.kallsyms])
           perf-exec 2273949 546850.708791:     482953 cycles:  ffffffff829190de __mod_memcg_lruvec_state+0x4e ([kernel.kallsyms])
                true 2273949 546850.709036:     501985 cycles:  ffffffff828add7c tlb_gather_mmu+0x4c ([kernel.kallsyms])
                true 2273949 546850.709292:     503065 cycles:      7f2446d97c03 _dl_map_object_deps+0x973 (/usr/lib/x86_64-linux-gnu/ld-linux-x86-64.so.2)

   - In addition to 'period' (PERF_SAMPLE_PERIOD), the other
     PERF_SAMPLE_ can be used for filtering, and also some other sample
     accessible values, from tools/perf/Documentation/perf-record.txt:

        Essentially the BPF filter expression is:

        <term> <operator> <value> (("," | "||") <term> <operator> <value>)*

     The <term> can be one of:
        ip, id, tid, pid, cpu, time, addr, period, txn, weight, phys_addr,
        code_pgsz, data_pgsz, weight1, weight2, weight3, ins_lat, retire_lat,
        p_stage_cyc, mem_op, mem_lvl, mem_snoop, mem_remote, mem_lock,
        mem_dtlb, mem_blk, mem_hops

     The <operator> can be one of:
        ==, !=, >, >=, <, <=, &

     The <value> can be one of:
        <number> (for any term)
        na, load, store, pfetch, exec (for mem_op)
        l1, l2, l3, l4, cxl, io, any_cache, lfb, ram, pmem (for mem_lvl)
        na, none, hit, miss, hitm, fwd, peer (for mem_snoop)
        remote (for mem_remote)
        na, locked (for mem_locked)
        na, l1_hit, l1_miss, l2_hit, l2_miss, any_hit, any_miss, walk, fault (for mem_dtlb)
        na, by_data, by_addr (for mem_blk)
        hops0, hops1, hops2, hops3 (for mem_hops)

  perf lock contention:

   - Show lock type with address.

   - Track and show mmap_lock, siglock and per-cpu rq_lock with address.
     This is done for mmap_lock by following the current->mm pointer:

      $ sudo ./perf lock con -abl -- sleep 10
       contended   total wait     max wait     avg wait            address   symbol
       ...
           16344    312.30 ms      2.22 ms     19.11 us   ffff8cc702595640
           17686    310.08 ms      1.49 ms     17.53 us   ffff8cc7025952c0
               3     84.14 ms     45.79 ms     28.05 ms   ffff8cc78114c478   mmap_lock
            3557     76.80 ms     68.75 us     21.59 us   ffff8cc77ca3af58
               1     68.27 ms     68.27 ms     68.27 ms   ffff8cda745dfd70
               9     54.53 ms      7.96 ms      6.06 ms   ffff8cc7642a48b8   mmap_lock
           14629     44.01 ms     60.00 us      3.01 us   ffff8cc7625f9ca0
            3481     42.63 ms    140.71 us     12.24 us   ffffffff937906ac   vmap_area_lock
           16194     38.73 ms     42.15 us      2.39 us   ffff8cd397cbc560
              11     38.44 ms     10.39 ms      3.49 ms   ffff8ccd6d12fbb8   mmap_lock
               1      5.43 ms      5.43 ms      5.43 ms   ffff8cd70018f0d8
            1674      5.38 ms    422.93 us      3.21 us   ffffffff92e06080   tasklist_lock
             581      4.51 ms    130.68 us      7.75 us   ffff8cc9b1259058
               5      3.52 ms      1.27 ms    703.23 us   ffff8cc754510070
             112      3.47 ms     56.47 us     31.02 us   ffff8ccee38b3120
             381      3.31 ms     73.44 us      8.69 us   ffffffff93790690   purge_vmap_area_lock
             255      3.19 ms     36.35 us     12.49 us   ffff8d053ce30c80

   - Update default map size to 16384.

   - Allocate single letter option -M for --map-nr-entries, as it is
     proving being frequently used.

   - Fix struct rq lock access for older kernels with BPF's CO-RE
     (Compile once, run everywhere).

   - Fix problems found with MSAn.

  perf report/top:

   - Add inline information when using --call-graph=fp or lbr, as was
     already done to the --call-graph=dwarf callchain mode.

   - Improve the 'srcfile' sort key performance by really using an
     optimization introduced in 6.2 for the 'srcline' sort key that
     avoids calling addr2line for comparision with each sample.

  perf sched:

   - Make 'perf sched latency/map/replay' to use "sched:sched_waking"
     instead of "sched:sched_waking", consistent with 'perf record'
     since d566a9c2d482 ("perf sched: Prefer sched_waking event when it
     exists").

  perf ftrace:

   - Make system wide the default target for latency subcommand, run the
     following command then generate some network traffic and press
     control+C:

       # perf ftrace latency -T __kfree_skb
     ^C
         DURATION     |      COUNT | GRAPH                                          |
          0 - 1    us |         27 | #############                                  |
          1 - 2    us |         22 | ###########                                    |
          2 - 4    us |          8 | ####                                           |
          4 - 8    us |          5 | ##                                             |
          8 - 16   us |         24 | ############                                   |
         16 - 32   us |          2 | #                                              |
         32 - 64   us |          1 |                                                |
         64 - 128  us |          0 |                                                |
        128 - 256  us |          0 |                                                |
        256 - 512  us |          0 |                                                |
        512 - 1024 us |          0 |                                                |
          1 - 2    ms |          0 |                                                |
          2 - 4    ms |          0 |                                                |
          4 - 8    ms |          0 |                                                |
          8 - 16   ms |          0 |                                                |
         16 - 32   ms |          0 |                                                |
         32 - 64   ms |          0 |                                                |
         64 - 128  ms |          0 |                                                |
        128 - 256  ms |          0 |                                                |
        256 - 512  ms |          0 |                                                |
        512 - 1024 ms |          0 |                                                |
          1 - ...   s |          0 |                                                |
       #

  perf top:

   - Add --branch-history (LBR: Last Branch Record) option, just like
     already available for 'perf record'.

   - Fix segfault in thread__comm_len() where thread->comm was being
     used outside thread->comm_lock.

  perf annotate:

   - Allow configuring objdump and addr2line in ~/.perfconfig., so that
     you can use alternative binaries, such as llvm's.

  perf kvm:

   - Add TUI mode for 'perf kvm stat report'.

  Reference counting:

   - Add reference count checking infrastructure to check for use after
     free, done to the 'cpumap', 'namespaces', 'maps' and 'map' structs,
     more to come.

     To build with it use -DREFCNT_CHECKING=1 in the make command line
     to build tools/perf. Documented at:

       https://perf.wiki.kernel.org/index.php/Reference_Count_Checking

   - The above caught, for instance, fix, present in this series:

        - Fix maps use after put in 'perf test "Share thread maps"':

          'maps' is copied from leader, but the leader is put on line 79
          and then 'maps' is used to read the reference count below - so
          a use after put, with the put of maps happening within
          thread__put.

     Fixed by reversing the order of puts so that the leader is put
     last.

   - Also several fixes were made to places where reference counts were
     not being held.

   - Make this one of the tests in 'make -C tools/perf build-test' to
     regularly build test it and to make sure no direct access to the
     reference counted structs are made, doing that via accessors to
     check the validity of the struct pointer.

  ARM64:

   - Fix 'perf report' segfault when filtering coresight traces by
     sparse lists of CPUs.

   - Add support for 'simd' as a sort field for 'perf report', to show
     ARM's NEON SIMD's predicate flags: "partial" and "empty".

  arm64 vendor events:

   - Add N1 metrics.

  Intel vendor events:

   - Add graniterapids, grandridge and sierraforrest events.

   - Refresh events for: alderlake, aldernaken, broadwell, broadwellde,
     broadwellx, cascadelakx, haswell, haswellx, icelake, icelakex,
     jaketown, meteorlake, knightslanding, sandybridge, sapphirerapids,
     silvermont, skylake, tigerlake and westmereep-dp

   - Refresh metrics for alderlake-n, broadwell, broadwellde,
     broadwellx, haswell, haswellx, icelakex, ivybridge, ivytown and
     skylakex.

  perf stat:

   - Implement --topdown using JSON metrics.

   - Add TopdownL1 JSON metric as a default if present, but disable it
     for now for some Intel hybrid architectures, a series of patches
     addressing this is being reviewed and will be submitted for v6.5.

   - Use metrics for --smi-cost.

   - Update topdown documentation.

  Vendor events (JSON) infrastructure:

   - Add support for computing and printing metric threshold values. For
     instance, here is one found in thesapphirerapids json file:

       {
           "BriefDescription": "Percentage of cycles spent in System Management Interrupts.",
           "MetricExpr": "((msr@aperf@ - cycles) / msr@aperf@ if msr@smi@ > 0 else 0)",
           "MetricGroup": "smi",
           "MetricName": "smi_cycles",
           "MetricThreshold": "smi_cycles > 0.1",
           "ScaleUnit": "100%"
       },

   - Test parsing metric thresholds with the fake PMU in 'perf test
     pmu-events'.

   - Support for printing metric thresholds in 'perf list'.

   - Add --metric-no-threshold option to 'perf stat'.

   - Add rand (reverse and) and has_pmem (optane memory) support to
     metrics.

   - Sort list of input files to avoid depending on the order from
     readdir() helping in obtaining reproducible builds.

  S/390:

   - Add common metrics: - CPI (cycles per instruction), prbstate (ratio
     of instructions executed in problem state compared to total number
     of instructions), l1mp (Level one instruction and data cache misses
     per 100 instructions).

   - Add cache metrics for z13, z14, z15 and z16.

   - Add metric for TLB and cache.

  ARM:

   - Add raw decoding for SPE (Statistical Profiling Extension) v1.3 MTE
     (Memory Tagging Extension) and MOPS (Memory Operations) load/store.

  Intel PT hardware tracing:

   - Add event type names UINTR (User interrupt delivered) and UIRET
     (Exiting from user interrupt routine), documented in table 32-50
     "CFE Packet Type and Vector Fields Details" in the Intel Processor
     Trace chapter of The Intel SDM Volume 3 version 078.

   - Add support for new branch instructions ERETS and ERETU.

   - Fix CYC timestamps after standalone CBR

  ARM CoreSight hardware tracing:

   - Allow user to override timestamp and contextid settings.

   - Fix segfault in dso lookup.

   - Fix timeless decode mode detection.

   - Add separate decode paths for timeless and per-thread modes.

  auxtrace:

   - Fix address filter entire kernel size.

  Miscellaneous:

   - Fix use-after-free and unaligned bugs in the PLT handling routines.

   - Use zfree() to reduce chances of use after free.

   - Add missing 0x prefix for addresses printed in hexadecimal in 'perf
     probe'.

   - Suppress massive unsupported target platform errors in the unwind
     code.

   - Fix return incorrect build_id size in elf_read_build_id().

   - Fix 'perf scripts intel-pt-events.py' IPC output for Python 2 .

   - Add missing new parameter in kfree_skb tracepoint to the python
     scripts using it.

   - Add 'perf bench syscall fork' benchmark.

   - Add support for printing PERF_MEM_LVLNUM_UNC (Uncached access) in
     'perf mem'.

   - Fix wrong size expectation for perf test 'Setup struct
     perf_event_attr' caused by the patch adding
     perf_event_attr::config3.

   - Fix some spelling mistakes"

* tag 'perf-tools-for-v6.4-3-2023-05-06' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux: (365 commits)
  Revert "perf build: Make BUILD_BPF_SKEL default, rename to NO_BPF_SKEL"
  Revert "perf build: Warn for BPF skeletons if endian mismatches"
  perf metrics: Fix SEGV with --for-each-cgroup
  perf bpf skels: Stop using vmlinux.h generated from BTF, use subset of used structs + CO-RE
  perf stat: Separate bperf from bpf_profiler
  perf test record+probe_libc_inet_pton: Fix call chain match on x86_64
  perf test record+probe_libc_inet_pton: Fix call chain match on s390
  perf tracepoint: Fix memory leak in is_valid_tracepoint()
  perf cs-etm: Add fix for coresight trace for any range of CPUs
  perf build: Fix unescaped # in perf build-test
  perf unwind: Suppress massive unsupported target platform errors
  perf script: Add new parameter in kfree_skb tracepoint to the python scripts using it
  perf script: Print raw ip instead of binary offset for callchain
  perf symbols: Fix return incorrect build_id size in elf_read_build_id()
  perf list: Modify the warning message about scandirat(3)
  perf list: Fix memory leaks in print_tracepoint_events()
  perf lock contention: Rework offset calculation with BPF CO-RE
  perf lock contention: Fix struct rq lock access
  perf stat: Disable TopdownL1 on hybrid
  perf stat: Avoid SEGV on counter->name
  ...

Merge tag 'core-debugobjects-2023-05-06' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull debugobjects fix from Thomas Gleixner:
 "A single fix for debugobjects:

  The recent fix to ensure atomicity of lookup and allocation
  inadvertently broke the pool refill mechanism, so that debugobject
  OOMs now in certain situations. The reason is that the functions which
  got updated no longer invoke debug_objecs_init(), which is now the
  only place to care about refilling the tracking object pool.

  Restore the original behaviour by adding explicit refill opportunities
  to those places"

* tag 'core-debugobjects-2023-05-06' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  debugobject: Ensure pool refill (again)

Merge tag 'v6.4-p2' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6

Pull crypto fixes from Herbert Xu:

 - A long-standing bug in crypto_engine

 - A buggy but harmless check in the sun8i-ss driver

 - A regression in the CRYPTO_USER interface

* tag 'v6.4-p2' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
  crypto: api - Fix CRYPTO_USER checks for report function
  crypto: engine - fix crypto_queue backlog handling
  crypto: sun8i-ss - Fix a test in sun8i_ss_setup_ivs()

Merge tag '6.4-rc-smb3-client-fixes-part2' of git://git.samba.org/sfrench/cifs-2.6

Pull cifs fixes from Steve French:
 "smb3 client fixes, mostly DFS or reconnect related:

   - Two DFS connection sharing fixes

   - DFS refresh fix

   - Reconnect fix

   - Two potential use after free fixes

   - Also print prefix patch in mount debug msg

   - Two small cleanup fixes"

* tag '6.4-rc-smb3-client-fixes-part2' of git://git.samba.org/sfrench/cifs-2.6:
  cifs: Remove unneeded semicolon
  cifs: fix sharing of DFS connections
  cifs: avoid potential races when handling multiple dfs tcons
  cifs: protect access of TCP_Server_Info::{origin,leaf}_fullpath
  cifs: fix potential race when tree connecting ipc
  cifs: fix potential use-after-free bugs in TCP_Server_Info::hostname
  cifs: print smb3_fs_context::source when mounting
  cifs: protect session status check in smb2_reconnect()
  SMB3.1.1: correct definition for app_instance_id create contexts

Merge tag 'clk-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/clk/linux

Pull clk fixes from Stephen Boyd:
 "A couple more patches that would be good to get into -rc1:

   - Revert an i.MX patch that's causing video failures because division
     math goes sideways

   - Fix a clang + W=1 build isue where FIELD_PREP() is taking a 32-bit
     variable instead of the usual u64 type

   - Fix a Kconfig bug in the StarFive JH7110 clk config that selects a
     reset controller when it can't be selected"

* tag 'clk-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/clk/linux:
  clk: starfive: Fix RESET_STARFIVE_JH7110 can't be selected in a specified case
  clk: sp7021: Adjust width of _m in HWM_FIELD_PREP()
  Revert "clk: imx: composite-8m: Add support to determine_rate"

Merge tag 'mailbox-v6.4' of git://git.linaro.org/landing-teams/working/fujitsu/integration

Pull mailbox updates from Jassi Brar:

 - mailbox api: allow direct registration to a channel and convert omap
   and pcc to use mbox_bind_client

 - omap and hi6220 : use of_property_read_bool

 - test: fix double-free and use spinlock header

 - rockchip and bcm-pdc: drop of_match_ptr

 - mpfs: change config symbol

 - mediatek gce: support MT6795

 - qcom apcs: consolidate of_device_id and support IPQ9574

* tag 'mailbox-v6.4' of git://git.linaro.org/landing-teams/working/fujitsu/integration:
  dt-bindings: mailbox: qcom: add compatible for IPQ9574 SoC
  mailbox: qcom-apcs-ipc: do not grow the of_device_id
  dt-bindings: mailbox: qcom,apcs-kpss-global: use fallbacks for few variants
  dt-bindings: mailbox: mediatek,gce-mailbox: Add support for MT6795
  mailbox: mpfs: convert SOC_MICROCHIP_POLARFIRE to ARCH_MICROCHIP_POLARFIRE
  mailbox: bcm-pdc: drop of_match_ptr for ID table
  mailbox: rockchip: drop of_match_ptr for ID table
  mailbox: mailbox-test: Fix potential double-free in mbox_test_message_write()
  mailbox: mailbox-test: Explicitly include header for spinlock support
  mailbox: Use of_property_read_bool() for boolean properties
  mailbox: pcc: Use mbox_bind_client
  mailbox: omap: Use mbox_bind_client
  mailbox: Allow direct registration to a channel

Merge tag 'for-6.4/io_uring-2023-05-07' of git://git.kernel.dk/linux

Pull more io_uring updates from Jens Axboe:
 "Nothing major in here, just two different parts:

   - A small series from Breno that enables passing the full SQE down
     for ->uring_cmd().

     This is a prerequisite for enabling full network socket operations.
     Queued up a bit late because of some stylistic concerns that got
     resolved, would be nice to have this in 6.4-rc1 so the dependent
     work will be easier to handle for 6.5.

   - Fix for the huge page coalescing, which was a regression introduced
     in the 6.3 kernel release (Tobias)"

* tag 'for-6.4/io_uring-2023-05-07' of git://git.kernel.dk/linux:
  io_uring: Remove unnecessary BUILD_BUG_ON
  io_uring: Pass whole sqe to commands
  io_uring: Create a helper to return the SQE size
  io_uring/rsrc: check for nonconsecutive pages

Revert "perf build: Make BUILD_BPF_SKEL default, rename to NO_BPF_SKEL"

This reverts commit a980755beb5aca9002e1c95ba519b83a44242b5b.

We need to better polish building with BPF skels, so revert back to
making it an experimental feature that has to be explicitely enabled
using BUILD_BPF_SKEL=1.

Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>

Revert "perf build: Warn for BPF skeletons if endian mismatches"

This reverts commit 51924ae69eea5bc90b5da525fbcf4bbd5f8551b3.

We need to better polish building with BPF skels, so revert back to
making it an experimental feature that has to be explicitely enabled
using BUILD_BPF_SKEL=1.

Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>

Merge tag 'mm-stable-2023-05-06-10-49' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm

Pull dmapool updates - again - from Andrew Morton:
 "Reinstate the dmapool changes which were accidentally removed by a
  mishap on the last commit in the previous attempt at the series"

Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup").
[ The whole old series: def8574308ed..2d55c16c0c54 results in an empty
  diff because that last commit ended up being just a revert of all that
  came everything before it.     - Linus ]

* tag 'mm-stable-2023-05-06-10-49' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm:
  dmapool: link blocks across pages
  dmapool: don't memset on free twice
  dmapool: simplify freeing
  dmapool: consolidate page initialization
  dmapool: rearrange page alloc failure handling
  dmapool: move debug code to own functions
  dmapool: speedup DMAPOOL_DEBUG with init_on_alloc
  dmapool: cleanup integer types
  dmapool: use sysfs_emit() instead of scnprintf()
  dmapool: remove checks for dev == NULL

Merge tag 'mm-hotfixes-stable-2023-05-06-10-45' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm

Pull hotfixes from Andrew Morton:
 "Five hotfixes.

  Three are cc:stable, two pertain to merge window changes"

* tag 'mm-hotfixes-stable-2023-05-06-10-45' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm:
  afs: fix the afs_dir_get_folio return value
  nilfs2: do not write dirty data after degenerating to read-only
  mm: do not reclaim private data from pinned page
  nilfs2: fix infinite loop in nilfs_mdt_get_block()
  mm/mmap/vma_merge: always check invariants

dmapool: link blocks across pages

The allocated dmapool pages are never freed for the lifetime of the pool.
There is no need for the two level list+stack lookup for finding a free
block since nothing is ever removed from the list.  Just use a simple
stack, reducing time complexity to constant.

The implementation inserts the stack linking elements and the dma handle
of the block within itself when freed.  This means the smallest possible
dmapool block is increased to at most 16 bytes to accommodate these
fields, but there are no exisiting users requesting a dma pool smaller
than that anyway.

Removing the list has a significant change in performance. Using the
kernel's micro-benchmarking self test:

Before:

  # modprobe dmapool_test
  dmapool test: size:16   blocks:8192   time:57282
  dmapool test: size:64   blocks:8192   time:172562
  dmapool test: size:256  blocks:8192   time:789247
  dmapool test: size:1024 blocks:2048   time:371823
  dmapool test: size:4096 blocks:1024   time:362237

After:

  # modprobe dmapool_test
  dmapool test: size:16   blocks:8192   time:24997
  dmapool test: size:64   blocks:8192   time:26584
  dmapool test: size:256  blocks:8192   time:33542
  dmapool test: size:1024 blocks:2048   time:9022
  dmapool test: size:4096 blocks:1024   time:6045

The module test allocates quite a few blocks that may not accurately
represent how these pools are used in real life.  For a more marco level
benchmark, running fio high-depth + high-batched on nvme, this patch shows
submission and completion latency reduced by ~100usec each, 1% IOPs
improvement, and perf record's time spent in dma_pool_alloc/free were
reduced by half.

[[email protected]: push new blocks in ascending order]
Link: https://lkml.kernel.org/r/[email protected]
Link: https://lkml.kernel.org/r/[email protected]
Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup")
Signed-off-by: Keith Busch <[email protected]>
Reviewed-by: Christoph Hellwig <[email protected]>
Tested-by: Bryan O'Donoghue <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Cc: Tony Battersby <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

dmapool: don't memset on free twice

If debug is enabled, dmapool will poison the range, so no need to clear it
to 0 immediately before writing over it.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup")
Signed-off-by: Keith Busch <[email protected]>
Reviewed-by: Christoph Hellwig <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Cc: Tony Battersby <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

dmapool: simplify freeing

The actions for busy and not busy are mostly the same, so combine these
and remove the unnecessary function.  Also, the pool is about to be freed
so there's no need to poison the page data since we only check for poison
on alloc, which can't be done on a freed pool.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup")
Signed-off-by: Keith Busch <[email protected]>
Reviewed-by: Christoph Hellwig <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Cc: Tony Battersby <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

dmapool: consolidate page initialization

Various fields of the dma pool are set in different places. Move it all
to one function.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup")
Signed-off-by: Keith Busch <[email protected]>
Reviewed-by: Christoph Hellwig <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Cc: Tony Battersby <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

dmapool: rearrange page alloc failure handling

Handle the error in a condition so the good path can be in the normal
flow.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup")
Signed-off-by: Keith Busch <[email protected]>
Reviewed-by: Christoph Hellwig <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Cc: Tony Battersby <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

dmapool: move debug code to own functions

Clean up the normal path by moving the debug code outside it.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup")
Signed-off-by: Keith Busch <[email protected]>
Reviewed-by: Christoph Hellwig <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Cc: Tony Battersby <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

dmapool: speedup DMAPOOL_DEBUG with init_on_alloc

Avoid double-memset of the same allocated memory in dma_pool_alloc() when
both DMAPOOL_DEBUG is enabled and init_on_alloc=1.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup")
Signed-off-by: Tony Battersby <[email protected]>
Signed-off-by: Keith Busch <[email protected]>
Reviewed-by: Christoph Hellwig <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

dmapool: cleanup integer types

To represent the size of a single allocation, dmapool currently uses
'unsigned int' in some places and 'size_t' in other places.  Standardize
on 'unsigned int' to reduce overhead, but use 'size_t' when counting all
the blocks in the entire pool.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup")
Signed-off-by: Tony Battersby <[email protected]>
Signed-off-by: Keith Busch <[email protected]>
Reviewed-by: Christoph Hellwig <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

dmapool: use sysfs_emit() instead of scnprintf()

Use sysfs_emit instead of scnprintf, snprintf or sprintf.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup")
Signed-off-by: Tony Battersby <[email protected]>
Signed-off-by: Keith Busch <[email protected]>
Reviewed-by: Christoph Hellwig <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

dmapool: remove checks for dev == NULL

dmapool originally tried to support pools without a device because
dma_alloc_coherent() supports allocations without a device.  But nobody
ended up using dma pools without a device, and trying to do so will result
in an oops.  So remove the checks for pool->dev == NULL since they are
unneeded bloat.

[[email protected]: add check for null dev on create]
Link: https://lkml.kernel.org/r/[email protected]
Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup")
Signed-off-by: Tony Battersby <[email protected]>
Signed-off-by: Keith Busch <[email protected]>
Reviewed-by: Christoph Hellwig <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

nfs: fix mis-merged __filemap_get_folio() error check

Fix another case of an incorrect check for the returned 'folio' value
from __filemap_get_folio().

The failure case used to return NULL, but was changed by commit
66dabbb65d67 ("mm: return an ERR_PTR from __filemap_get_folio").

But in the meantime, commit ec108d3cc766 ("NFS: Convert readdir page
array functions to use a folio") added a new user of that function.

And my merge of the two did not fix this up correctly.

The ext4 merge had the same issue, but that one had been caught in
linux-next and got properly fixed while merging.

Fixes: 0127f25b5dfc ("Merge tag 'nfs-for-6.4-1' of git://git.linux-nfs.org/projects/anna/linux-nfs")
Cc: Anna Schumaker <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

afs: fix the afs_dir_get_folio return value

Keep returning NULL on failure instead of letting an ERR_PTR escape to
callers that don't expect it.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 66dabbb65d67 ("mm: return an ERR_PTR from __filemap_get_folio")
Signed-off-by: Christoph Hellwig <[email protected]>
Reported-by: Jan Kara <[email protected]>
Reviewed-by: Jan Kara <[email protected]>
Reviewed-by: David Howells <[email protected]>
Tested-by: David Howells <[email protected]>
Cc: Marc Dionne <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

nilfs2: do not write dirty data after degenerating to read-only

According to syzbot's report, mark_buffer_dirty() called from
nilfs_segctor_do_construct() outputs a warning with some patterns after
nilfs2 detects metadata corruption and degrades to read-only mode.

After such read-only degeneration, page cache data may be cleared through
nilfs_clear_dirty_page() which may also clear the uptodate flag for their
buffer heads.  However, even after the degeneration, log writes are still
performed by unmount processing etc., which causes mark_buffer_dirty() to
be called for buffer heads without the "uptodate" flag and causes the
warning.

Since any writes should not be done to a read-only file system in the
first place, this fixes the warning in mark_buffer_dirty() by letting
nilfs_segctor_do_construct() abort early if in read-only mode.

This also changes the retry check of nilfs_segctor_write_out() to avoid
unnecessary log write retries if it detects -EROFS that
nilfs_segctor_do_construct() returned.

Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Ryusuke Konishi <[email protected]>
Tested-by: Ryusuke Konishi <[email protected]>
Reported-by: [email protected]
Link: https://syzkaller.appspot.com/bug?extid=2af3bc9585be7f23f290
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

mm: do not reclaim private data from pinned page

If the page is pinned, there's no point in trying to reclaim it.
Furthermore if the page is from the page cache we don't want to reclaim
fs-private data from the page because the pinning process may be writing
to the page at any time and reclaiming fs private info on a dirty page can
upset the filesystem (see link below).

Link: https://lore.kernel.org/linux-mm/[email protected]
Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Jan Kara <[email protected]>
Reviewed-by: Matthew Wilcox (Oracle) <[email protected]>
Reviewed-by: Lorenzo Stoakes <[email protected]>
Reviewed-by: Christoph Hellwig <[email protected]>
Reviewed-by: John Hubbard <[email protected]>
Acked-by: David Hildenbrand <[email protected]>
Acked-by: Peter Xu <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

nilfs2: fix infinite loop in nilfs_mdt_get_block()

If the disk image that nilfs2 mounts is corrupted and a virtual block
address obtained by block lookup for a metadata file is invalid,
nilfs_bmap_lookup_at_level() may return the same internal return code as
-ENOENT, meaning the block does not exist in the metadata file.

This duplication of return codes confuses nilfs_mdt_get_block(), causing
it to read and create a metadata block indefinitely.

In particular, if this happens to the inode metadata file, ifile,
semaphore i_rwsem can be left held, causing task hangs in lock_mount.

Fix this issue by making nilfs_bmap_lookup_at_level() treat virtual block
address translation failures with -ENOENT as metadata corruption instead
of returning the error code.

Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Ryusuke Konishi <[email protected]>
Tested-by: Ryusuke Konishi <[email protected]>
Reported-by: [email protected]
Link: https://syzkaller.appspot.com/bug?extid=221d75710bde87fa0e97
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

mm/mmap/vma_merge: always check invariants

We may still have inconsistent input parameters even if we choose not to
merge and the vma_merge() invariant checks are useful for checking this
with no production runtime cost (these are only relevant when
CONFIG_DEBUG_VM is specified).

Therefore, perform these checks regardless of whether we merge.

This is relevant, as a recent issue (addressed in commit "mm/mempolicy:
Correctly update prev when policy is equal on mbind") in the mbind logic
was only picked up in the 6.2.y stable branch where these assertions are
performed prior to determining mergeability.

Had this remained the same in mainline this issue may have been picked up
faster, so moving forward let's always check them.

Link: https://lkml.kernel.org/r/df548a6ae3fa135eec3b446eb3dae8eb4227da97.1682885809.git.lstoakes@gmail.com
Signed-off-by: Lorenzo Stoakes <[email protected]>
Acked-by: Vlastimil Babka <[email protected]>
Reviewed-by: David Hildenbrand <[email protected]>
Reviewed-by: Liam R. Howlett <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

filemap: Handle error return from __filemap_get_folio()

Smatch reports that filemap_fault() was missed in the conversion of
__filemap_get_folio() error returns from NULL to ERR_PTR.

Fixes: 66dabbb65d67 ("mm: return an ERR_PTR from __filemap_get_folio")
Reported-by: Dan Carpenter <[email protected]>
Reported-by: [email protected]
Reported-by: Christoph Hellwig <[email protected]>
Signed-off-by: Matthew Wilcox (Oracle) <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Merge tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi

Pull more SCSI updates from James Bottomley:
 "Six late arriving patches for the merge window. Five are minor
  assorted fixes and updates.

  The IPR driver change removes SATA support, which will now allow a
  major cleanup in the ATA subsystem because it was the only driver
  still using the old attachment mechanism. The driver is only used on
  power systems and SATA was used to support a DVD device, which has
  long been moved to a different hba. IBM chose this route instead of
  porting ipr to the newer SATA interfaces"

* tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: qedi: Fix use after free bug in qedi_remove()
  scsi: ufs: core: mcq: Fix &hwq->cq_lock deadlock issue
  scsi: ipr: Remove several unused variables
  scsi: pm80xx: Log device registration
  scsi: ipr: Remove SATA support
  scsi: scsi_debug: Abort commands from scsi_debug_device_reset()

Merge tag 'for-6.4/block-2023-05-06' of git://git.kernel.dk/linux

Pull more block updates from Jens Axboe:

 - MD pull request via Song:
      - Improve raid5 sequential IO performance on spinning disks, which
        fixes a regression since v6.0 (Jan Kara)
      - Fix bitmap offset types, which fixes an issue introduced in this
        merge window (Jonathan Derrick)

 - Cleanup of hweight type used for cgroup writeback (Maxim)

 - Fix a regression with the "has_submit_bio" changes across partitions
   (Ming)

 - Cleanup of QUEUE_FLAG_ADD_RANDOM clearing.

   We used to set this flag on queues non blk-mq queues, and hence some
   drivers clear it unconditionally. Since all of these have since been
   converted to true blk-mq drivers, drop the useless clear as the bit
   is not set (Chaitanya)

 - Fix the flags being set in a bio for a flush for drbd (Christoph)

 - Cleanup and deduplication of the code handling setting block device
   capacity (Damien)

 - Fix for ublk handling IO timeouts (Ming)

 - Fix for a regression in blk-cgroup teardown (Tao)

 - NBD documentation and code fixes (Eric)

 - Convert blk-integrity to using device_attributes rather than a second
   kobject to manage lifetimes (Thomas)

* tag 'for-6.4/block-2023-05-06' of git://git.kernel.dk/linux:
  ublk: add timeout handler
  drbd: correctly submit flush bio on barrier
  mailmap: add mailmap entries for Jens Axboe
  block: Skip destroyed blkg when restart in blkg_destroy_all()
  writeback: fix call of incorrect macro
  md: Fix bitmap offset type in sb writer
  md/raid5: Improve performance for sequential IO
  docs nbd: userspace NBD now favors github over sourceforge
  block nbd: use req.cookie instead of req.handle
  uapi nbd: add cookie alias to handle
  uapi nbd: improve doc links to userspace spec
  blk-integrity: register sysfs attributes on struct device
  blk-integrity: convert to struct device_attribute
  blk-integrity: use sysfs_emit
  block/drivers: remove dead clear of random flag
  block: sync part's ->bd_has_submit_bio with disk's
  block: Cleanup set_capacity()/bdev_set_nr_sectors()

Merge tag 'pipe-nonblock-2023-05-06' of git://git.kernel.dk/linux

Pull nonblocking pipe io_uring support from Jens Axboe:
 "Here's the revised edition of the FMODE_NOWAIT support for pipes, in
  which we just flag it as such supporting FMODE_NOWAIT unconditionally,
  but clear it if we ever end up using splice/vmsplice on the pipe.

  The pipe read/write side is perfectly fine for nonblocking IO, however
  splice and vmsplice can potentially wait for IO with the pipe lock
  held"

* tag 'pipe-nonblock-2023-05-06' of git://git.kernel.dk/linux:
  pipe: set FMODE_NOWAIT on pipes
  splice: clear FMODE_NOWAIT on file if splice/vmsplice is used

Merge tag 'sound-fix-6.4-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 rc1.

  The only (LOC-wise) dominant change was ASoC Qualcomm fix, but most of
  it was merely a code shuffling.

  Another significant change here is for ALSA PCM core; it received a
  revert and a series of fixes for PCM auto-silencing where it caused a
  regression in the previous PR for rc1.

  Others are all small: ASoC Intel fixes, various quirks for ASoC AMD,
  HD-audio and USB-audio, the continued legacy emu10k1 code cleanup, and
  some documentation updates"

* tag 'sound-fix-6.4-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound: (23 commits)
  ALSA: pcm: use exit controlled loop in snd_pcm_playback_silence()
  ALSA: pcm: simplify top-up mode init in snd_pcm_playback_silence()
  ALSA: pcm: playback silence - move silence variable updates to separate function
  ALSA: pcm: playback silence - remove extra code
  ALSA: pcm: fix playback silence - correct incremental silencing
  ALSA: pcm: fix playback silence - use the actual new_hw_ptr for the threshold mode
  ALSA: pcm: Revert "ALSA: pcm: rewrite snd_pcm_playback_silence()"
  ALSA: hda/realtek: Fix mute and micmute LEDs for an HP laptop
  ALSA: caiaq: input: Add error handling for unsupported input methods in `snd_usb_caiaq_input_init`
  ALSA: usb-audio: Add quirk for Pioneer DDJ-800
  ALSA: hda/realtek: support HP Pavilion Aero 13-be0xxx Mute LED
  ASoC: Intel: soc-acpi-cht: Add quirk for Nextbook Ares 8A tablet
  ASoC: amd: yc: Add Asus VivoBook Pro 14 OLED M6400RC to the quirks list for acp6x
  ASoC: codecs: wcd938x: fix accessing regmap on unattached devices
  ALSA: docs: Fix code block indentation in ALSA driver example
  ALSA: docs: Extend module parameters description
  ALSA: hda/realtek: Add quirk for ASUS UM3402YAR using CS35L41
  ALSA: emu10k1: use more existing defines instead of open-coded numbers
  ASoC: amd: yc: Add ASUS M3402RA into DMI table
  ALSA: hda/realtek: Add quirk for ThinkPad P1 Gen 6
  ...

Merge tag 'regulator-fix-v6.4' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regulator

Pull regulator fix from Mark Brown:
 "A trivial typo fix that came in during the merge window"

* tag 'regulator-fix-v6.4' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regulator:
  regulator: consumer.rst: fix 'regulator_enable' typo.

Merge tag '6.4-rc-ksmbd-server-fixes-part2' of git://git.samba.org/ksmbd

Pull ksmbd server fixes from Steve French:
 "Ten ksmbd server fixes, including some important security fixes:

   - Two use after free fixes

   - Fix RCU callback race

   - Deadlock fix

   - Three patches to prevent session setup attacks

   - Prevent guest users from establishing multichannel sessions

   - Fix null pointer dereference in query FS info

   - Memleak fix"

* tag '6.4-rc-ksmbd-server-fixes-part2' of git://git.samba.org/ksmbd:
  ksmbd: call rcu_barrier() in ksmbd_server_exit()
  ksmbd: fix racy issue under cocurrent smb2 tree disconnect
  ksmbd: fix racy issue from smb2 close and logoff with multichannel
  ksmbd: not allow guest user on multichannel
  ksmbd: fix deadlock in ksmbd_find_crypto_ctx()
  ksmbd: block asynchronous requests when making a delay on session setup
  ksmbd: destroy expired sessions
  ksmbd: fix racy issue from session setup and logoff
  ksmbd: fix NULL pointer dereference in smb2_get_info_filesystem()
  ksmbd: fix memleak in session setup

Merge tag 'net-6.4-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net

Pull networking fixes from Jakub Kicinski:
 "Including fixes from netfilter.

  Current release - regressions:

   - sched: act_pedit: free pedit keys on bail from offset check

  Current release - new code bugs:

   - pds_core:
      - Kconfig fixes (DEBUGFS and AUXILIARY_BUS)
      - fix mutex double unlock in error path

  Previous releases - regressions:

   - sched: cls_api: remove block_cb from driver_list before freeing

   - nf_tables: fix ct untracked match breakage

   - eth: mtk_eth_soc: drop generic vlan rx offload

   - sched: flower: fix error handler on replace

  Previous releases - always broken:

   - tcp: fix skb_copy_ubufs() vs BIG TCP

   - ipv6: fix skb hash for some RST packets

   - af_packet: don't send zero-byte data in packet_sendmsg_spkt()

   - rxrpc: timeout handling fixes after moving client call connection
     to the I/O thread

   - ixgbe: fix panic during XDP_TX with > 64 CPUs

   - igc: RMW the SRRCTL register to prevent losing timestamp config

   - dsa: mt7530: fix corrupt frames using TRGMII on 40 MHz XTAL MT7621

   - r8152:
      - fix flow control issue of RTL8156A
      - fix the poor throughput for 2.5G devices
      - move setting r8153b_rx_agg_chg_indicate() to fix coalescing
      - enable autosuspend

   - ncsi: clear Tx enable mode when handling a Config required AEN

   - octeontx2-pf: macsec: fixes for CN10KB ASIC rev

  Misc:

   - 9p: remove INET dependency"

* tag 'net-6.4-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (69 commits)
  net: bcmgenet: Remove phy_stop() from bcmgenet_netif_stop()
  pds_core: fix mutex double unlock in error path
  net/sched: flower: fix error handler on replace
  Revert "net/sched: flower: Fix wrong handle assignment during filter change"
  net/sched: flower: fix filter idr initialization
  net: fec: correct the counting of XDP sent frames
  bonding: add xdp_features support
  net: enetc: check the index of the SFI rather than the handle
  sfc: Add back mailing list
  virtio_net: suppress cpu stall when free_unused_bufs
  ice: block LAN in case of VF to VF offload
  net: dsa: mt7530: fix network connectivity with multiple CPU ports
  net: dsa: mt7530: fix corrupt frames using trgmii on 40 MHz XTAL MT7621
  9p: Remove INET dependency
  netfilter: nf_tables: fix ct untracked match breakage
  af_packet: Don't send zero-byte data in packet_sendmsg_spkt().
  igc: read before write to SRRCTL register
  pds_core: add AUXILIARY_BUS and NET_DEVLINK to Kconfig
  pds_core: remove CONFIG_DEBUG_FS from makefile
  ionic: catch failure from devlink_alloc
  ...

Merge tag 'i2c-for-6.4-rc1-part2' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux

Pull more i2c updates from Wolfram Sang:
 "Some more driver bugfixes and a DT binding conversion"

* tag 'i2c-for-6.4-rc1-part2' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  dt-bindings: i2c: brcm,kona-i2c: convert to YAML
  i2c: gxp: fix build failure without CONFIG_I2C_SLAVE
  i2c: imx-lpi2c: avoid taking clk_prepare mutex in PM callbacks
  i2c: omap: Fix standard mode false ACK readings
  i2c: tegra: Fix PEC support for SMBUS block read

s390: remove the unneeded select GCC12_NO_ARRAY_BOUNDS

Commit 0da6e5fd6c37 ("gcc: disable '-Warray-bounds' for gcc-13 too") makes
config GCC11_NO_ARRAY_BOUNDS to be for disabling -Warray-bounds in any gcc
version 11 and upwards, and with that, removes the GCC12_NO_ARRAY_BOUNDS
config as it is now covered by the semantics of GCC11_NO_ARRAY_BOUNDS.

As GCC11_NO_ARRAY_BOUNDS is yes by default, there is no need for the s390
architecture to explicitly select GCC11_NO_ARRAY_BOUNDS. Hence, the select
GCC12_NO_ARRAY_BOUNDS in arch/s390/Kconfig can simply be dropped.

Remove the unneeded "select GCC12_NO_ARRAY_BOUNDS".

Signed-off-by: Lukas Bulwahn <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

perf metrics: Fix SEGV with --for-each-cgroup

Ensure the metric threshold is copied correctly or else a use of
uninitialized memory happens.

Fixes: d0a3052f6faefffc ("perf metric: Compute and print threshold values")
Reported-by: Namhyung Kim <[email protected]>
Signed-off-by: Ian Rogers <[email protected]>
Tested-by: Namhyung Kim <[email protected]>
Cc: Adrian Hunter <[email protected]>
Cc: Alexander Shishkin <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: Kajol Jain <[email protected]>
Cc: Mark Rutland <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>

perf bpf skels: Stop using vmlinux.h generated from BTF, use subset of used structs + CO-RE

Linus reported a build break due to using a vmlinux without a BTF elf
section to generate the vmlinux.h header with bpftool for use in the BPF
tools in tools/perf/util/bpf_skel/*.bpf.c.

Instead add a vmlinux.h file with the structs needed with the fields the
tools need, marking the structs with __attribute__((preserve_access_index)),
so that libbpf's CO-RE code can fixup the struct field offsets.

In some cases the vmlinux.h file that was being generated by bpftool
from the kernel BTF information was not needed at all, just including
linux/bpf.h, sometimes linux/perf_event.h was enough as non-UAPI
types were not being used.

To keep te patch small, include those UAPI headers from the trimmed down
vmlinux.h file, that then provides the tools with just the structs and
the subset of its fields needed for them.

Testing it:

  # perf lock contention -b find / > /dev/null
  ^C contended   total wait     max wait     avg wait         type   caller

           7     53.59 us     10.86 us      7.66 us     rwlock:R   start_this_handle+0xa0
           2     30.35 us     21.99 us     15.17 us  rwsem:R   iterate_dir+0x52
           1   9.04 us      9.04 us      9.04 us     rwlock:W   start_this_handle+0x291
           1   8.73 us      8.73 us      8.73 us     spinlock   raw_spin_rq_lock_nested+0x1e
  #
  # perf lock contention -abl find / > /dev/null
  ^C contended   total wait     max wait     avg wait            address   symbol

           1    262.96 ms    262.96 ms    262.96 ms   ffff8e67502d0170    (mutex)
          12    244.24 us     39.91 us     20.35 us   ffff8e6af56f8070   mmap_lock (rwsem)
           7     30.28 us      6.85 us      4.33 us   ffff8e6c865f1d40   rq_lock (spinlock)
           3   7.42 us      4.03 us      2.47 us   ffff8e6c864b1d40   rq_lock (spinlock)
           2   3.72 us      2.19 us      1.86 us   ffff8e6c86571d40   rq_lock (spinlock)
           1   2.42 us      2.42 us      2.42 us   ffff8e6c86471d40   rq_lock (spinlock)
           4   2.11 us 559 ns       527 ns   ffffffff9a146c80   rcu_state (spinlock)
           3   1.45 us 818 ns       482 ns   ffff8e674ae8384c    (rwlock)
           1    870 ns 870 ns       870 ns   ffff8e68456ee060    (rwlock)
           1    663 ns 663 ns       663 ns   ffff8e6c864f1d40   rq_lock (spinlock)
           1    573 ns 573 ns       573 ns   ffff8e6c86531d40   rq_lock (spinlock)
           1    472 ns 472 ns       472 ns   ffff8e6c86431740    (spinlock)
           1    397 ns 397 ns       397 ns   ffff8e67413a4f04    (spinlock)
  #
  # perf test offcpu
  95: perf record offcpu profiling tests                              : Ok
  #
  # perf kwork latency --use-bpf
  Starting trace, Hit <Ctrl+C> to stop and report
  ^C
    Kwork Name                     | Cpu  | Avg delay     | Count     | Max delay     | Max delay start     | Max delay end   |
   --------------------------------------------------------------------------------------------------------------------------------
    (w)flush_memcg_stats_dwork     | 0000 |   1056.212 ms |         2 |   2112.345 ms |     550113.229573 s |     550115.341919 s |
    (w)toggle_allocation_gate    | 0000 |     10.144 ms |        62 |    416.389 ms |     550113.453518 s |     550113.869907 s |
    (w)0xffff8e6748e28080          | 0002 |  0.623 ms |         1 |      0.623 ms |     550110.989841 s |     550110.990464 s |
    (w)vmstat_shepherd             | 0000 |  0.586 ms |        10 |      2.828 ms |     550111.971536 s |     550111.974364 s |
    (w)vmstat_update               | 0007 |  0.363 ms |         5 |      1.634 ms |     550113.222520 s |     550113.224154 s |
    (w)vmstat_update               | 0000 |  0.324 ms |        10 |      2.827 ms |     550111.971526 s |     550111.974354 s |
    (w)0xffff8e674c5f4a58          | 0002 |  0.102 ms |         5 |      0.134 ms |     550110.989839 s |     550110.989972 s |
    (w)psi_avgs_work               | 0001 |  0.086 ms |         3 |      0.107 ms |     550114.957852 s |     550114.957959 s |
    (w)psi_avgs_work               | 0000 |  0.079 ms |         5 |      0.100 ms |     550118.605668 s |     550118.605768 s |
    (w)kfree_rcu_monitor           | 0006 |  0.079 ms |         1 |      0.079 ms |     550110.925821 s |     550110.925900 s |
    (w)psi_avgs_work               | 0004 |  0.079 ms |         1 |      0.079 ms |     550109.581835 s |     550109.581914 s |
    (w)psi_avgs_work               | 0001 |  0.078 ms |         1 |      0.078 ms |     550109.197809 s |     550109.197887 s |
    (w)psi_avgs_work               | 0002 |  0.077 ms |         5 |      0.086 ms |     550110.669819 s |     550110.669905 s |
  <SNIP>
  # strace -e bpf -o perf-stat-bpf-counters.output perf stat -e cycles --bpf-counters sleep 1

   Performance counter stats for 'sleep 1':

           6,197,983   cycles

         1.003922848 seconds time elapsed

         0.000000000 seconds user
         0.002032000 seconds sys

  # head -7 perf-stat-bpf-counters.output
  bpf(BPF_OBJ_GET, {pathname="/sys/fs/bpf/perf_attr_map", bpf_fd=0, file_flags=0}, 16) = 3
  bpf(BPF_OBJ_GET_INFO_BY_FD, {info={bpf_fd=3, info_len=88, info=0x7ffcead64990}}, 16) = 0
  bpf(BPF_MAP_LOOKUP_ELEM, {map_fd=3, key=0x24129e0, value=0x7ffcead65a48, flags=BPF_ANY}, 32) = 0
  bpf(BPF_LINK_GET_FD_BY_ID, {link_id=1252}, 12) = -1 ENOENT (No such file or directory)
  bpf(BPF_PROG_LOAD, {prog_type=BPF_PROG_TYPE_SOCKET_FILTER, insn_cnt=2, insns=0x7ffcead65780, license="GPL", log_level=0, log_size=0, log_buf=NULL, kern_version=KERNEL_VERSION(0, 0, 0), prog_flags=0, prog_name="", prog_ifindex=0, expected_attach_type=BPF_CGROUP_INET_INGRESS, prog_btf_fd=0, func_info_rec_size=0,
+func_info=NULL, func_info_cnt=0, line_info_rec_size=0, line_info=NULL, line_info_cnt=0, attach_btf_id=0, attach_prog_fd=0}, 116) = 4
  bpf(BPF_PROG_LOAD, {prog_type=BPF_PROG_TYPE_SOCKET_FILTER, insn_cnt=2, insns=0x7ffcead65920, license="GPL", log_level=0, log_size=0, log_buf=NULL, kern_version=KERNEL_VERSION(0, 0, 0), prog_flags=0, prog_name="", prog_ifindex=0, expected_attach_type=BPF_CGROUP_INET_INGRESS, prog_btf_fd=0, func_info_rec_size=0,
+func_info=NULL, func_info_cnt=0, line_info_rec_size=0, line_info=NULL, line_info_cnt=0, attach_btf_id=0, attach_prog_fd=0, fd_array=NULL}, 128) = 4
  bpf(BPF_BTF_LOAD, {btf="\237\353\1\0\30\0\0\0\0\0\0\0\20\0\0\0\20\0\0\0\5\0\0\0\1\0\0\0\0\0\0\1"..., btf_log_buf=NULL, btf_size=45, btf_log_size=0, btf_log_level=0}, 28) = 4
  #

Reported-by: Linus Torvalds <[email protected]>
Suggested-by: Andrii Nakryiko <[email protected]>
Tested-by: Namhyung Kim <[email protected]>
Tested-by: Song Liu <[email protected]>
Cc: Adrian Hunter <[email protected]>
Cc: Alexei Starovoitov <[email protected]>
Cc: Ian Rogers <[email protected]>
Cc: Jiri Olsa <[email protected]>
Co-developed-by: Jiri Olsa <[email protected]>
Link: https://lore.kernel.org/lkml/[email protected]
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>

Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next

There's a fix which landed in net-next, pull it in along
with the couple of minor cleanups.

Signed-off-by: Jakub Kicinski <[email protected]>

perf stat: Separate bperf from bpf_profiler

It seems that perf stat -b <prog id> doesn't produce any results:

    $ perf stat -e cycles -b 4 -I 10000 -vvv
    Control descriptor is not initialized
    cycles: 0 0 0
                time        counts unit      events
10.007641640    <not supported>      cycles

Looks like this happens because fentry/fexit progs are getting loaded, but the
corresponding perf event is not enabled and not added into the events bpf map.
I think there is some mixing up between two type of bpf support, one for bperf
and one for bpf_profiler. Both are identified via evsel__is_bpf, based on which
perf events are enabled, but for the latter (bpf_profiler) a perf event is
required. Using evsel__is_bperf to check only bperf produces expected results:

    $ perf stat -e cycles -b 4 -I 10000 -vvv
    Control descriptor is not initialized
    ------------------------------------------------------------
    perf_event_attr:
      size                             136
      sample_type                      IDENTIFIER
      read_format                      TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
      disabled                         1
      exclude_guest                    1
    ------------------------------------------------------------
    sys_perf_event_open: pid -1  cpu 0  group_fd -1  flags 0x8 = 3
    ------------------------------------------------------------
    [...perf_event_attr for other CPUs...]
    ------------------------------------------------------------
    cycles: 309426 169009 169009
time             counts unit events
10.010091271             309426      cycles

The final numbers correspond (at least in the level of magnitude) to the
same metric obtained via bpftool.

Fixes: 112cb56164bc2108 ("perf stat: Introduce config stat.bpf-counter-events")
Reviewed-by: Song Liu <[email protected]>
Signed-off-by: Dmitrii Dolgov <[email protected]>
Tested-by: Song Liu <[email protected]>
Cc: Ian Rogers <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Song Liu <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>

Merge tag 'devicetree-fixes-for-6.4-1' of git://git.kernel.org/pub/scm/linux/kernel/git/robh/linux

Pull devicetree fixes from Rob Herring:

 - Add Conor Dooley as a DT binding maintainer

 - Swap the order of parsing /memreserve/ and /reserved-memory nodes so
   that the /reserved-memory nodes which have more information are
   handled first

 - Fix some property dependencies in riscv,pmu binding

 - Update maintainers entries on a couple of bindings

* tag 'devicetree-fixes-for-6.4-1' of git://git.kernel.org/pub/scm/linux/kernel/git/robh/linux:
  MAINTAINERS: add Conor as a dt-bindings maintainer
  dt-bindings: perf: riscv,pmu: fix property dependencies
  dt-bindings: xilinx: Remove Naga from memory and mtd bindings
  of: fdt: Scan /memreserve/ last
  dt-bindings: clock: r9a06g032-sysctrl: Change maintainer to Fabrizio Castro
  dt-bindings: pinctrl: renesas,rzv2m: Change maintainer to Fabrizio Castro
  dt-bindings: pinctrl: renesas,rzn1: Change maintainer to Fabrizio Castro
  dt-bindings: i2c: renesas,rzv2m: Change maintainer to Fabrizio Castro

Merge tag 'docs-6.4-2' of git://git.lwn.net/linux

Pull more documentation updates from Jonathan Corbet:
 "A handful of late-arriving documentation fixes, plus one Spanish
  translation that has been ready for some time but got applied late"

* tag 'docs-6.4-2' of git://git.lwn.net/linux:
  docs/sp_SP: Add translation of process/adding-syscalls
  CREDITS: Update email address for Mat Martineau
  Documentation: update kernel stack for x86_64
  docs: Remove unnecessary unicode character
  docs: fix "Reviewd" typo
  Documentation: timers: hrtimers: Make hybrid union historical
  docs/admin-guide/mm/ksm.rst fix intraface -> interface typo
  doc:it_IT: fix some typos

Merge tag 'trace-v6.4-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace

Pull more tracing updates from Steven Rostedt:

 - Make buffer_percent read/write.

   The buffer_percent file is how users can state how long to block on
   the tracing buffer depending on how much is in the buffer. When it
   hits the "buffer_percent" it will wake the task waiting on the
   buffer. For some reason it was set to read-only.

   This was not noticed because testing was done as root without
   SELinux, but with SELinux it will prevent even root to write to it
   without having CAP_DAC_OVERRIDE.

 - The "touched_functions" was added this merge window, but one of the
   reasons for adding it was not implemented.

   That was to show what functions were not only touched, but had either
   a direct trampoline attached to it, or a kprobe or live kernel
   patching that can "hijack" the function to run a different function.
   The point is to know if there's functions in the kernel that may not
   be behaving as the kernel code shows. This can be used for debugging.

   TODO: Add this information to kernel oops too.

* tag 'trace-v6.4-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace:
  ftrace: Add MODIFIED flag to show if IPMODIFY or direct was attached
  tracing: Fix permissions for the buffer_percent file

Merge tag 'locking-core-2023-05-05' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull locking updates from Ingo Molnar:

 - Introduce local{,64}_try_cmpxchg() - a slightly more optimal
   primitive, which will be used in perf events ring-buffer code

 - Simplify/modify rwsems on PREEMPT_RT, to address writer starvation

 - Misc cleanups/fixes

* tag 'locking-core-2023-05-05' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  locking/atomic: Correct (cmp)xchg() instrumentation
  locking/x86: Define arch_try_cmpxchg_local()
  locking/arch: Wire up local_try_cmpxchg()
  locking/generic: Wire up local{,64}_try_cmpxchg()
  locking/atomic: Add generic try_cmpxchg{,64}_local() support
  locking/rwbase: Mitigate indefinite writer starvation
  locking/arch: Rename all internal __xchg() names to __arch_xchg()

Merge branch 'x86-uaccess-cleanup': x86 uaccess header cleanups

Merge my x86 uaccess updates branch.

The LAM ("Linear Address Masking") updates in this release made me
unhappy about how "access_ok()" was done, and it actually turned out to
have a couple of small bugs in it too.  This is my cleanup of the code:

 - use the sign bit of the __user pointer rather than masking the
   address and checking it against the TASK_SIZE range.

   We already did this part for the get/put_user() side, but
   'access_ok()' did the naïve "mask and range check" thing, which not
   only generates nasty code, but also ended up meaning that __access_ok
   itself didn't do a good job, and so copy_from_user_nmi() didn't get
   the check right.

 - move all the code that is 64-bit only into the 64-bit version of the
   header file, so that we don't unnecessarily pollute the shared x86
   code and make it look like LAM might work in 32-bit too.

 - fix a bug in the address masking (that doesn't end up mattering: in
   this case the fix was to just remove the buggy code entirely).

 - a couple of trivial cleanups and added commentary about the
   access_ok() rules.

* x86-uaccess-cleanup:
  x86-64: mm: clarify the 'positive addresses' user address rules
  x86: mm: remove 'sign' games from LAM untagged_addr*() macros
  x86: uaccess: move 32-bit and 64-bit parts into proper <asm/uaccess_N.h> header
  x86: mm: remove architecture-specific 'access_ok()' define
  x86-64: make access_ok() independent of LAM

Merge tag 'riscv-for-linus-6.4-mw2' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linux

Pull more RISC-V updates from Palmer Dabbelt:

 - Support for hibernation

 - The .rela.dyn section has been moved to the init area

 - A fix for the SBI probing to allow for implementation-defined
   behavior

 - Various other fixes and cleanups throughout the tree

* tag 'riscv-for-linus-6.4-mw2' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linux:
  RISC-V: include cpufeature.h in cpufeature.c
  riscv: Move .rela.dyn to the init sections
  dt-bindings: riscv: explicitly mention assumption of Zicsr & Zifencei support
  riscv: compat_syscall_table: Fixup compile warning
  RISC-V: fixup in-flight collision with ARCH_WANT_OPTIMIZE_VMEMMAP rename
  RISC-V: fix sifive and thead section mismatches in errata
  RISC-V: Align SBI probe implementation with spec
  riscv: mm: remove redundant parameter of create_fdt_early_page_table
  riscv: Adjust dependencies of HAVE_DYNAMIC_FTRACE selection
  RISC-V: Add arch functions to support hibernation/suspend-to-disk
  RISC-V: mm: Enable huge page support to kernel_page_present() function
  RISC-V: Factor out common code of __cpu_resume_enter()
  RISC-V: Change suspend_save_csrs and suspend_restore_csrs to public function

Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm

Pull more kvm updates from Paolo Bonzini:
 "This includes the 6.4 changes for RISC-V, and a few bugfix patches for
  other architectures. For x86, this closes a longstanding performance
  issue in the newer and (usually) more scalable page table management
  code.

  RISC-V:
   - ONE_REG interface to enable/disable SBI extensions
   - Zbb extension for Guest/VM
   - AIA CSR virtualization

  x86:
   - Fix a long-standing TDP MMU flaw, where unloading roots on a vCPU
     can result in the root being freed even though the root is
     completely valid and can be reused as-is (with a TLB flush).

  s390:
   - A couple of bugfixes"

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  KVM: s390: fix race in gmap_make_secure()
  KVM: s390: pv: fix asynchronous teardown for small VMs
  KVM: x86: Preserve TDP MMU roots until they are explicitly invalidated
  RISC-V: KVM: Virtualize per-HART AIA CSRs
  RISC-V: KVM: Use bitmap for irqs_pending and irqs_pending_mask
  RISC-V: KVM: Add ONE_REG interface for AIA CSRs
  RISC-V: KVM: Implement subtype for CSR ONE_REG interface
  RISC-V: KVM: Initial skeletal support for AIA
  RISC-V: KVM: Drop the _MASK suffix from hgatp.VMID mask defines
  RISC-V: Detect AIA CSRs from ISA string
  RISC-V: Add AIA related CSR defines
  RISC-V: KVM: Allow Zbb extension for Guest/VM
  RISC-V: KVM: Add ONE_REG interface to enable/disable SBI extensions
  RISC-V: KVM: Alphabetize selects
  KVM: RISC-V: Retry fault if vma_lookup() results become invalid

Merge tag 'acpi-6.4-rc1-3' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm

Pull ACPI fix from Rafael Wysocki:
 "Remove an ACPI backlight quirk for Lenovo ThinkPad W530, added during
  the 6.3 cycle, that turned out to do more harm than help (Hans de
  Goede)"

* tag 'acpi-6.4-rc1-3' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  ACPI: video: Remove acpi_backlight=video quirk for Lenovo ThinkPad W530

Merge tag 'thermal-6.4-rc1-3' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm

Pull thermal control fixes from Rafael Wysocki:
 "These fix a NULL pointer dereference in the Intel powerclamp driver
  introduced during the 6.3 cycle and update MAINTAINERS to match recent
  code changes.

  Specifics:

   - Fix NULL pointer access in the Intel powerclamp thermal driver that
     occurs on attempts to set the cooling device state to 0 in the
     default configuration (Srinivas Pandruvada)

   - Drop the stale MAINTAINERS entry for the Intel Menlow thermal
     driver that has been removed recently (Lukas Bulwahn)"

* tag 'thermal-6.4-rc1-3' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  MAINTAINERS: remove section INTEL MENLOW THERMAL DRIVER
  thermal: intel: powerclamp: Fix NULL pointer access issue

Merge tag 'phy-fixes-6.4-1' of git://git.kernel.org/pub/scm/linux/kernel/git/phy/linux-phy

Pull phy fixes from Vinod Koul:

 - Fix for mediatek driver warning for variable used uninitialized and
   for wrong pll math

* tag 'phy-fixes-6.4-1' of git://git.kernel.org/pub/scm/linux/kernel/git/phy/linux-phy:
  phy: mediatek: hdmi: mt8195: fix wrong pll calculus
  phy: mediatek: hdmi: mt8195: fix uninitialized variable usage in pll_calc

Merge tag 'drm-next-2023-05-05' of git://anongit.freedesktop.org/drm/drm

Pull more drm fixes from Dave Airlie:
 "This is the fixes for the last couple of weeks for i915 and last 3
  weeks for amdgpu, lots of them but pretty scattered around and all
  pretty small.

  amdgpu:
   - SR-IOV fixes
   - DCN 3.2 fixes
   - DC mclk handling fixes
   - eDP fixes
   - SubVP fixes
   - HDCP regression fix
   - DSC fixes
   - DC FP fixes
   - DCN 3.x fixes
   - Display flickering fix when switching between vram and gtt
   - Z8 power saving fix
   - Fix hang when skipping modeset
   - GPU reset fixes
   - Doorbell fix when resizing BARs
   - Fix spurious warnings in gmc
   - Locking fix for AMDGPU_SCHED IOCTL
   - SR-IOV fix
   - DCN 3.1.4 fix
   - DCN 3.2 fix
   - Fix job cleanup when CS is aborted

  i915:
   - skl pipe source size check
   - mtl transcoder mask fix
   - DSI power on sequence fix
   - GuC versioning corner case fix"

* tag 'drm-next-2023-05-05' of git://anongit.freedesktop.org/drm/drm: (48 commits)
  drm/amdgpu: drop redundant sched job cleanup when cs is aborted
  drm/amd/display: filter out invalid bits in pipe_fuses
  drm/amd/display: Change default Z8 watermark values
  drm/amdgpu: disable SDMA WPTR_POLL_ENABLE for SR-IOV
  drm/amdgpu: add a missing lock for AMDGPU_SCHED
  drm/amdgpu: fix an amdgpu_irq_put() issue in gmc_v9_0_hw_fini()
  drm/amdgpu: fix amdgpu_irq_put call trace in gmc_v10_0_hw_fini
  drm/amdgpu: fix amdgpu_irq_put call trace in gmc_v11_0_hw_fini
  drm/amdgpu: Enable doorbell selfring after resize FB BAR
  drm/amdgpu: Use the default reset when loading or reloading the driver
  drm/amdgpu: Fix mode2 reset for sienna cichlid
  drm/i915/dsi: Use unconditional msleep() instead of intel_dsi_msleep()
  drm/i915/mtl: Add the missing CPU transcoder mask in intel_device_info
  drm/i915/guc: Actually return an error if GuC version range check fails
  drm/amd/display: Lowering min Z8 residency time
  drm/amd/display: fix flickering caused by S/G mode
  drm/amd/display: Set min_width and min_height capability for DCN30
  drm/amd/display: Isolate remaining FPU code in DCN32
  drm/amd/display: Update bounding box values for DCN321
  drm/amd/display: Do not clear GPINT register when releasing DMUB from reset
  ...

ALSA: pcm: use exit controlled loop in snd_pcm_playback_silence()

We already know that `frames` is greater than zero, because we just
checked it. So we don't need to check the loop condition on the first
iteration.

Signed-off-by: Oswald Buddenhagen <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Takashi Iwai <[email protected]>

ALSA: pcm: simplify top-up mode init in snd_pcm_playback_silence()

Inline the remaining call of snd_pcm_playback_hw_avail(). This makes
the top-up branch more congruent with the thresholded one, and allows
simplifying the handling of the corner cases.

Signed-off-by: Oswald Buddenhagen <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Takashi Iwai <[email protected]>

ALSA: pcm: playback silence - move silence variable updates to separate function

The code tracking the added samples in thresholded mode and the code
tracking the just played samples in top-up mode are semantically
identical, so factor it out to a common function to enhance readability.

Co-developed-by: Oswald Buddenhagen <[email protected]>
Signed-off-by: Oswald Buddenhagen <[email protected]>
Signed-off-by: Jaroslav Kysela <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Takashi Iwai <[email protected]>

ALSA: pcm: playback silence - remove extra code

The removed condition handles de facto only one situation where
runtime->silence_filled variable is equal to runtime->buffer_size,
because this variable cannot go over the buffer size. This case is
implicitly caught by the required comparison of the noise distance
with the threshold.

Suggested-by: Oswald Buddenhagen <[email protected]>
Signed-off-by: Jaroslav Kysela <[email protected]>
Signed-off-by: Oswald Buddenhagen <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Takashi Iwai <[email protected]>

ALSA: pcm: fix playback silence - correct incremental silencing

Commit 9a826ddba6e ("[ALSA] pcm core: fix silence_start calculations")
came with exactly the right commit message, but the patch just made
things broken in a different way: We'd fill at a too low address if the
area was already partially zeroed, so we'd under-fill. This affected
both thresholded mode (where it was somewhat less likely) and top-up
mode (where it would be the case consistently).

Co-developed-by: Oswald Buddenhagen <[email protected]>
Signed-off-by: Oswald Buddenhagen <[email protected]>
Signed-off-by: Jaroslav Kysela <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Takashi Iwai <[email protected]>

ALSA: pcm: fix playback silence - use the actual new_hw_ptr for the threshold mode

The snd_pcm_playback_hw_avail() function uses runtime->status->hw_ptr.
Unfortunately, in case when we call this function from snd_pcm_update_hw_ptr0(),
this variable contains the previous hardware pointer. Use the new_hw_ptr
argument to calculate hw_avail (filled samples by the user space) to
correct the threshold comparison.

The new_hw_ptr argument may also be set to ULONG_MAX which means the
initialization phase. In this case, use runtime->status->hw_ptr.

Suggested-by: Oswald Buddenhagen <[email protected]>
Signed-off-by: Jaroslav Kysela <[email protected]>
Reviewed-by: Oswald Buddenhagen <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Takashi Iwai <[email protected]>

ALSA: pcm: Revert "ALSA: pcm: rewrite snd_pcm_playback_silence()"

This reverts commit 9f656705c5faa18afb26d922cfc64f9fd103c38d.

There was a regression (in the top-up mode). Unfortunately, the patch
provided from the author of this commit is not easy to review.

Keep the updated and new comments in headers.
Also add a new comment that documents the missed API constraint which
led to the regression.

Reported-by: Jeff Chua <[email protected]>
Link: https://lore.kernel.org/r/CAAJw_ZsbTVd3Es373x_wTNDF7RknGhCD0r+NKUSwAO7HpLAkYA@mail.gmail.com
Signed-off-by: Jaroslav Kysela <[email protected]>
Signed-off-by: Oswald Buddenhagen <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Takashi Iwai <[email protected]>

ALSA: hda/realtek: Fix mute and micmute LEDs for an HP laptop

There's another laptop that needs the fixup to enable mute and micmute
LEDs. So do it accordingly.

Signed-off-by: Kai-Heng Feng <[email protected]>
Cc: <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Takashi Iwai <[email protected]>

Merge tag 'asoc-fix-v6.4-rc1' of https://git.kernel.org/pub/scm/linux/kernel/git/broonie/sound into for-linus

ASoC: Fixes for v6.4

A small set of fixes and device quirks that have come in during the
merge window, the Qualcomm fix seems quite large but it's mainly code
motion so looks larger than it is.

ftrace: Add MODIFIED flag to show if IPMODIFY or direct was attached

If a function had ever had IPMODIFY or DIRECT attached to it, where this
is how live kernel patching and BPF overrides work, mark them and display
an "M" in the enabled_functions and touched_functions files. This can be
used for debugging. If a function had been modified and later there's a bug
in the code related to that function, this can be used to know if the cause
is possibly from a live kernel patch or a BPF program that changed the
behavior of the code.

Also update the documentation on the enabled_functions and
touched_functions output, as it was missing direct callers and CALL_OPS.
And include this new modify attribute.

Link: https://lore.kernel.org/linux-trace-kernel/[email protected]
Cc: Mark Rutland <[email protected]>
Acked-by: Masami Hiramatsu (Google) <[email protected]>
Signed-off-by: Steven Rostedt (Google) <[email protected]>

MAINTAINERS: remove section INTEL MENLOW THERMAL DRIVER

Commit 2b6a7409ac39 ("thermal: intel: menlow: Get rid of this driver")
removes the driver drivers/thermal/intel/intel_menlow.c, but misses to
remove its reference in MAINTAINERS.

Hence, ./scripts/get_maintainer.pl --self-test=patterns complains about a
broken reference.

Remove the INTEL MENLOW THERMAL DRIVER section in MAINTAINERS.

Fixes: 2b6a7409ac39 ("thermal: intel: menlow: Get rid of this driver")
Signed-off-by: Lukas Bulwahn <[email protected]>
Signed-off-by: Rafael J. Wysocki <[email protected]>

MAINTAINERS: add Conor as a dt-bindings maintainer

Rob asked if I would be interested in helping with the dt-bindings
maintenance, and since I am a glutton for punishment I accepted.

Signed-off-by: Conor Dooley <[email protected]>
Acked-by: Krzysztof Kozlowski <[email protected]>
Link: https://lore.kernel.org/r/20230504-renderer-alive-1c01d431b2a7@spud
Signed-off-by: Rob Herring <[email protected]>

Merge tag 'kvm-s390-next-6.4-2' of https://git.kernel.org/pub/scm/linux/kernel/git/kvms390/linux into HEAD

For 6.4

Merge tag 'kvm-x86-mmu-6.4-2' of https://github.com/kvm-x86/linux into HEAD

Fix a long-standing flaw in x86's TDP MMU where unloading roots on a vCPU can
result in the root being freed even though the root is completely valid and
can be reused as-is (with a TLB flush).

Merge tag 'kvm-riscv-6.4-1' of https://github.com/kvm-riscv/linux into HEAD

KVM/riscv changes for 6.4

- ONE_REG interface to enable/disable SBI extensions
- Zbb extension for Guest/VM
- AIA CSR virtualization

cifs: Remove unneeded semicolon

./fs/cifs/smb2pdu.c:4140:2-3: Unneeded semicolon

Reported-by: Abaci Robot <[email protected]>
Link: https://bugzilla.openanolis.cn/show_bug.cgi?id=4863
Signed-off-by: Yang Li <[email protected]>
Signed-off-by: Steve French <[email protected]>

net: bcmgenet: Remove phy_stop() from bcmgenet_netif_stop()

The call to phy_stop() races with the later call to phy_disconnect(),
resulting in concurrent phy_suspend() calls being run from different
CPUs. The final call to phy_disconnect() ensures that the PHY is
stopped and suspended, too.

Fixes: c96e731c93ff ("net: bcmgenet: connect and disconnect from the PHY state machine")
Signed-off-by: Florian Fainelli <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

pds_core: fix mutex double unlock in error path

Fix a double unlock in an error handling path by unlocking as soon as
the error is seen and removing unlocks in the error cleanup path.

Link: https://lore.kernel.org/kernel-janitors/[email protected]/
Fixes: 523847df1b37 ("pds_core: add devcmd device interfaces")
Reported-by: Dan Carpenter <[email protected]>
Signed-off-by: Shannon Nelson <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge branch 'tc-action-fixes'

Vlad Buslov says:

====================
Fixes for miss to tc action series

Changes V1 -> V2:

- Added new patch reverting Ivan's fix for the same issue.
====================

Signed-off-by: David S. Miller <[email protected]>

net/sched: flower: fix error handler on replace

When replacing a filter (i.e. 'fold' pointer is not NULL) the insertion of
new filter to idr is postponed until later in code since handle is already
provided by the user. However, the error handling code in fl_change()
always assumes that the new filter had been inserted into idr. If error
handler is reached when replacing existing filter it may remove it from idr
therefore making it unreachable for delete or dump afterwards. Fix the
issue by verifying that 'fold' argument wasn't provided by caller before
calling idr_remove().

Fixes: 08a0063df3ae ("net/sched: flower: Move filter handle initialization earlier")
Signed-off-by: Vlad Buslov <[email protected]>
Reviewed-by: Pedro Tammela <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Revert "net/sched: flower: Fix wrong handle assignment during filter change"

This reverts commit 32eff6bacec2cb574677c15378169a9fa30043ef.

Superseded by the following commit in this series.

Signed-off-by: Vlad Buslov <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net/sched: flower: fix filter idr initialization

The cited commit moved idr initialization too early in fl_change() which
allows concurrent users to access the filter that is still being
initialized and is in inconsistent state, which, in turn, can cause NULL
pointer dereference [0]. Since there is no obvious way to fix the ordering
without reverting the whole cited commit, alternative approach taken to
first insert NULL pointer into idr in order to allocate the handle but
still cause fl_get() to return NULL and prevent concurrent users from
seeing the filter while providing miss-to-action infrastructure with valid
handle id early in fl_change().

[  152.434728] general protection fault, probably for non-canonical address 0xdffffc0000000000: 0000 [#1] SMP KASAN
[  152.436163] KASAN: null-ptr-deref in range [0x0000000000000000-0x0000000000000007]
[  152.437269] CPU: 4 PID: 3877 Comm: tc Not tainted 6.3.0-rc4+ #5
[  152.438110] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
[  152.439644] RIP: 0010:fl_dump_key+0x8b/0x1d10 [cls_flower]
[  152.440461] Code: 01 f2 02 f2 c7 40 08 04 f2 04 f2 c7 40 0c 04 f3 f3 f3 65 48 8b 04 25 28 00 00 00 48 89 84 24 00 01 00 00 48 89 c8 48 c1 e8 03 <0f> b6 04 10 84 c0 74 08 3c 03 0f 8e 98 19 00 00 8b 13 85 d2 74 57
[  152.442885] RSP: 0018:ffff88817a28f158 EFLAGS: 00010246
[  152.443851] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000000
[  152.444826] RDX: dffffc0000000000 RSI: ffffffff8500ae80 RDI: ffff88810a987900
[  152.445791] RBP: ffff888179d88240 R08: ffff888179d8845c R09: ffff888179d88240
[  152.446780] R10: ffffed102f451e48 R11: 00000000fffffff2 R12: ffff88810a987900
[  152.447741] R13: ffffffff8500ae80 R14: ffff88810a987900 R15: ffff888149b3c738
[  152.448756] FS:  00007f5eb2a34800(0000) GS:ffff88881ec00000(0000) knlGS:0000000000000000
[  152.449888] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  152.450685] CR2: 000000000046ad19 CR3: 000000010b0bd006 CR4: 0000000000370ea0
[  152.451641] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  152.452628] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  152.453588] Call Trace:
[  152.454032]  <TASK>
[  152.454447]  ? netlink_sendmsg+0x7a1/0xcb0
[  152.455109]  ? sock_sendmsg+0xc5/0x190
[  152.455689]  ? ____sys_sendmsg+0x535/0x6b0
[  152.456320]  ? ___sys_sendmsg+0xeb/0x170
[  152.456916]  ? do_syscall_64+0x3d/0x90
[  152.457529]  ? entry_SYSCALL_64_after_hwframe+0x46/0xb0
[  152.458321]  ? ___sys_sendmsg+0xeb/0x170
[  152.458958]  ? __sys_sendmsg+0xb5/0x140
[  152.459564]  ? do_syscall_64+0x3d/0x90
[  152.460122]  ? entry_SYSCALL_64_after_hwframe+0x46/0xb0
[  152.460852]  ? fl_dump_key_options.part.0+0xea0/0xea0 [cls_flower]
[  152.461710]  ? _raw_spin_lock+0x7a/0xd0
[  152.462299]  ? _raw_read_lock_irq+0x30/0x30
[  152.462924]  ? nla_put+0x15e/0x1c0
[  152.463480]  fl_dump+0x228/0x650 [cls_flower]
[  152.464112]  ? fl_tmplt_dump+0x210/0x210 [cls_flower]
[  152.464854]  ? __kmem_cache_alloc_node+0x1a7/0x330
[  152.465592]  ? nla_put+0x15e/0x1c0
[  152.466160]  tcf_fill_node+0x515/0x9a0
[  152.466766]  ? tc_setup_offload_action+0xf0/0xf0
[  152.467463]  ? __alloc_skb+0x13c/0x2a0
[  152.468067]  ? __build_skb_around+0x330/0x330
[  152.468814]  ? fl_get+0x107/0x1a0 [cls_flower]
[  152.469503]  tc_del_tfilter+0x718/0x1330
[  152.470115]  ? is_bpf_text_address+0xa/0x20
[  152.470765]  ? tc_ctl_chain+0xee0/0xee0
[  152.471335]  ? __kernel_text_address+0xe/0x30
[  152.471948]  ? unwind_get_return_address+0x56/0xa0
[  152.472639]  ? __thaw_task+0x150/0x150
[  152.473218]  ? arch_stack_walk+0x98/0xf0
[  152.473839]  ? __stack_depot_save+0x35/0x4c0
[  152.474501]  ? stack_trace_save+0x91/0xc0
[  152.475119]  ? security_capable+0x51/0x90
[  152.475741]  rtnetlink_rcv_msg+0x2c1/0x9d0
[  152.476387]  ? rtnl_calcit.isra.0+0x2b0/0x2b0
[  152.477042]  ? __sys_sendmsg+0xb5/0x140
[  152.477664]  ? do_syscall_64+0x3d/0x90
[  152.478255]  ? entry_SYSCALL_64_after_hwframe+0x46/0xb0
[  152.479010]  ? __stack_depot_save+0x35/0x4c0
[  152.479679]  ? __stack_depot_save+0x35/0x4c0
[  152.480346]  netlink_rcv_skb+0x12c/0x360
[  152.480929]  ? rtnl_calcit.isra.0+0x2b0/0x2b0
[  152.481517]  ? do_syscall_64+0x3d/0x90
[  152.482061]  ? netlink_ack+0x1550/0x1550
[  152.482612]  ? rhashtable_walk_peek+0x170/0x170
[  152.483262]  ? kmem_cache_alloc_node+0x1af/0x390
[  152.483875]  ? _copy_from_iter+0x3d6/0xc70
[  152.484528]  netlink_unicast+0x553/0x790
[  152.485168]  ? netlink_attachskb+0x6a0/0x6a0
[  152.485848]  ? unwind_next_frame+0x11cc/0x1a10
[  152.486538]  ? arch_stack_walk+0x61/0xf0
[  152.487169]  netlink_sendmsg+0x7a1/0xcb0
[  152.487799]  ? netlink_unicast+0x790/0x790
[  152.488355]  ? iovec_from_user.part.0+0x4d/0x220
[  152.488990]  ? _raw_spin_lock+0x7a/0xd0
[  152.489598]  ? netlink_unicast+0x790/0x790
[  152.490236]  sock_sendmsg+0xc5/0x190
[  152.490796]  ____sys_sendmsg+0x535/0x6b0
[  152.491394]  ? import_iovec+0x7/0x10
[  152.491964]  ? kernel_sendmsg+0x30/0x30
[  152.492561]  ? __copy_msghdr+0x3c0/0x3c0
[  152.493160]  ? do_syscall_64+0x3d/0x90
[  152.493706]  ___sys_sendmsg+0xeb/0x170
[  152.494283]  ? may_open_dev+0xd0/0xd0
[  152.494858]  ? copy_msghdr_from_user+0x110/0x110
[  152.495541]  ? __handle_mm_fault+0x2678/0x4ad0
[  152.496205]  ? copy_page_range+0x2360/0x2360
[  152.496862]  ? __fget_light+0x57/0x520
[  152.497449]  ? mas_find+0x1c0/0x1c0
[  152.498026]  ? sockfd_lookup_light+0x1a/0x140
[  152.498703]  __sys_sendmsg+0xb5/0x140
[  152.499306]  ? __sys_sendmsg_sock+0x20/0x20
[  152.499951]  ? do_user_addr_fault+0x369/0xd80
[  152.500595]  do_syscall_64+0x3d/0x90
[  152.501185]  entry_SYSCALL_64_after_hwframe+0x46/0xb0
[  152.501917] RIP: 0033:0x7f5eb294f887
[  152.502494] Code: 0a 00 f7 d8 64 89 02 48 c7 c0 ff ff ff ff eb b9 0f 1f 00 f3 0f 1e fa 64 8b 04 25 18 00 00 00 85 c0 75 10 b8 2e 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 51 c3 48 83 ec 28 89 54 24 1c 48 89 74 24 10
[  152.505008] RSP: 002b:00007ffd2c708f78 EFLAGS: 00000246 ORIG_RAX: 000000000000002e
[  152.506152] RAX: ffffffffffffffda RBX: 00000000642d9472 RCX: 00007f5eb294f887
[  152.507134] RDX: 0000000000000000 RSI: 00007ffd2c708fe0 RDI: 0000000000000003
[  152.508113] RBP: 0000000000000000 R08: 0000000000000001 R09: 0000000000000000
[  152.509119] R10: 00007f5eb2808708 R11: 0000000000000246 R12: 0000000000000001
[  152.510068] R13: 0000000000000000 R14: 00007ffd2c70d1b8 R15: 0000000000485400
[  152.511031]  </TASK>
[  152.511444] Modules linked in: cls_flower sch_ingress openvswitch nsh mlx5_vdpa vringh vhost_iotlb vdpa mlx5_ib mlx5_core rpcrdma rdma_ucm ib_iser libiscsi scsi_transport_iscsi ib_umad rdma_cm ib_ipoib iw_cm ib_cm ib_uverbs ib_core xt_conntrack xt_MASQUERADE nf_conntrack_netlink nfnetlink xt_addrtype iptable_nat nf_nat br_netfilter overlay zram zsmalloc fuse [last unloaded: mlx5_core]
[  152.515720] ---[ end trace 0000000000000000 ]---

Fixes: 08a0063df3ae ("net/sched: flower: Move filter handle initialization earlier")
Signed-off-by: Vlad Buslov <[email protected]>
Reviewed-by: Pedro Tammela <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: fec: correct the counting of XDP sent frames

In the current xdp_xmit implementation, if any single frame fails to
transmit due to insufficient buffer descriptors, the function nevertheless
reports success in sending all frames. This results in erroneously
indicating that frames were transmitted when in fact they were dropped.

This patch fixes the issue by ensureing the return value properly
indicates the actual number of frames successfully transmitted, rather than
potentially reporting success for all frames when some could not transmit.

Fixes: 6d6b39f180b8 ("net: fec: add initial XDP support")
Signed-off-by: Gagandeep Singh <[email protected]>
Signed-off-by: Shenwei Wang <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

bonding: add xdp_features support

Introduce xdp_features support for bonding driver according to the slave
devices attached to the master one. xdp_features is required whenever we
want to xdp_redirect traffic into a bond device and then into selected
slaves attached to it.

Reviewed-by: Simon Horman <[email protected]>
Acked-by: Jay Vosburgh <[email protected]>
Fixes: 66c0e13ad236 ("drivers: net: turn on XDP features")
Signed-off-by: Lorenzo Bianconi <[email protected]>
Reviewed-by: Jussi Maki <[email protected]>
Tested-by: Jussi Maki <[email protected]>
Signed-off-by: David S. Miller <[email protected]>