diff --git a/os/StarryOS/kernel/src/ebpf/mod.rs b/os/StarryOS/kernel/src/ebpf/mod.rs
index e861046945..88bc8a4f25 100644
--- a/os/StarryOS/kernel/src/ebpf/mod.rs
+++ b/os/StarryOS/kernel/src/ebpf/mod.rs
@@ -20,6 +20,7 @@ use alloc::{collections::btree_map::BTreeMap, sync::Arc, vec};
 use ax_errno::{AxError, AxResult};
 use ax_io::Read;
 use ax_lazyinit::LazyInit;
+use ax_task::current;
 use kbpf_basic::{
     helper::RawBPFHelperFn,
     linux_bpf::{bpf_attr, bpf_cmd},
@@ -44,6 +45,7 @@ use crate::{
     kprobe::KernelRawMutex,
     mm::VmBytes,
     perf::raw_tracepoint::bpf_raw_tracepoint_open,
+    task::AsThread,
 };
 
 /// The global BPF helper-function table (id → `RawBPFHelperFn`). Populated by
@@ -51,6 +53,63 @@ use crate::{
 /// without re-running the kbpf-basic init on every call.
 pub static BPF_HELPER_FUN_SET: LazyInit<BTreeMap<u32, RawBPFHelperFn>> = LazyInit::new();
 
+/// BPF helper ID for `bpf_probe_read` (legacy, address-space-agnostic).
+const BPF_FUNC_PROBE_READ: u32 = 4;
+/// BPF helper ID for `bpf_get_current_pid_tgid`. kbpf-basic does not register
+/// this helper, so we implement it directly in StarryOS.
+const BPF_FUNC_GET_CURRENT_PID_TGID: u32 = 14;
+/// BPF helper ID for `bpf_get_current_comm`. kbpf-basic does not register
+/// this helper, so we implement it directly in StarryOS.
+const BPF_FUNC_GET_CURRENT_COMM: u32 = 16;
+/// BPF helper ID for `bpf_probe_read_kernel`. kbpf-basic only registers
+/// `bpf_probe_read` so we alias 113 onto the same raw helper.
+const BPF_FUNC_PROBE_READ_KERNEL: u32 = 113;
+
+/// `bpf_get_current_pid_tgid()` — returns `(tgid << 32) | tid` of the
+/// currently running task, matching the Linux kernel helper ABI.
+fn bpf_get_current_pid_tgid(_a: u64, _b: u64, _c: u64, _d: u64, _e: u64) -> u64 {
+    let task = current();
+    let tgid = task.as_thread().proc_data.proc.pid() as u64;
+    let pid = task.as_thread().tid() as u64;
+    (tgid << 32) | pid
+}
+
+/// `bpf_get_current_comm(char *buf, u32 size_of_buf)` — copies the current
+/// task's comm (name) into `buf` using `strscpy_pad` semantics: at most
+/// `size_of_buf - 1` bytes are copied, a NUL terminator is always written,
+/// and remaining bytes are zero-padded. Returns 0 on success or `-EINVAL`
+/// when `size_of_buf` is 0, matching the Linux kernel helper ABI.
+fn bpf_get_current_comm(buf: u64, size_of_buf: u64, _c: u64, _d: u64, _e: u64) -> u64 {
+    let size = size_of_buf as usize;
+    if buf == 0 {
+        return 0;
+    }
+
+    let task = current();
+    let comm = task.name();
+    let comm_bytes = comm.as_bytes();
+
+    if size == 0 {
+        return (-22i64) as u64; // -EINVAL
+    }
+
+    // Copy at most size-1 bytes to leave room for the NUL terminator.
+    let copy_len = comm_bytes.len().min(size.saturating_sub(1));
+
+    // SAFETY: `buf` is a kernel-space pointer validated by the eBPF verifier
+    // before the helper is invoked.
+    unsafe {
+        core::ptr::copy_nonoverlapping(comm_bytes.as_ptr(), buf as *mut u8, copy_len);
+        // Always NUL-terminate at `buf[copy_len]`.
+        (buf as *mut u8).add(copy_len).write(0);
+        // Zero-pad the remainder.
+        if copy_len + 1 < size {
+            core::ptr::write_bytes((buf as *mut u8).add(copy_len + 1), 0, size - copy_len - 1);
+        }
+    }
+    0
+}
+
 /// Initialize the BPF subsystem: build the helper-function table from
 /// `kbpf-basic`. Must be called before `sys_bpf(BPF_PROG_LOAD)` so loaded
 /// programs can resolve the helper ids referenced in their instructions.
@@ -61,9 +120,14 @@ pub fn init_ebpf() {
     // tracepoint program uses to pull fields out of its sample buffer.
     // kbpf-basic only registers the legacy `bpf_probe_read` (id 4), whose raw
     // reader is address-space-agnostic in this VM, so alias 113 onto it.
-    if let Some(&probe_read) = set.get(&4) {
-        set.entry(113).or_insert(probe_read);
+    if let Some(&probe_read) = set.get(&BPF_FUNC_PROBE_READ) {
+        set.entry(BPF_FUNC_PROBE_READ_KERNEL).or_insert(probe_read);
     }
+    // Register helpers that kbpf-basic does not yet provide (#14, #16).
+    set.entry(BPF_FUNC_GET_CURRENT_PID_TGID)
+        .or_insert(bpf_get_current_pid_tgid);
+    set.entry(BPF_FUNC_GET_CURRENT_COMM)
+        .or_insert(bpf_get_current_comm);
     BPF_HELPER_FUN_SET.init_once(set);
 }
 
diff --git a/os/StarryOS/kernel/src/perf/bpf.rs b/os/StarryOS/kernel/src/perf/bpf.rs
index d52369ebc4..c6a6013244 100644
--- a/os/StarryOS/kernel/src/perf/bpf.rs
+++ b/os/StarryOS/kernel/src/perf/bpf.rs
@@ -37,6 +37,11 @@ use crate::{
     file::FileLike,
 };
 
+/// Number of 4K pages reserved for x86_64 BPF JIT executable memory.
+/// Each JIT-compiled eBPF program fits within this space; the allocation
+/// is sized generously (16 KiB) since programs are typically < 1 page.
+const BPF_JIT_MEM_PAGES: usize = 4;
+
 /// Wraps `kbpf_basic::perf::bpf::BpfPerfEvent` with kernel state: a poll
 /// set so readers can wait for new records, and a weak handle to the
 /// backing pages produced by `device_mmap` (Some after the first
@@ -285,14 +290,17 @@ impl OwnedEbpfVm {
         // TODO: not all of the address space is accessible to a BPF program;
         // allowing the full `0..u64::MAX` range disables rbpf's bounds check
         // and lets a buggy/hostile program read arbitrary kernel memory via
-        // direct loads. Narrow this to the legitimately-reachable context /
-        // map / stack ranges once kbpf-basic exposes the per-program bounds.
+        // direct loads.
+        //
+        // FIXME: narrow this to the legitimately-reachable context /
+        // map / stack ranges once kbpf-basic exposes per-program
+        // bounds.
         vm.register_allowed_memory(0..u64::MAX);
 
         #[cfg(target_arch = "x86_64")]
         {
             // TODO: calculate a more precise size.
-            let mut jit_exec_memory = BPFJitMemory::new(4)?;
+            let mut jit_exec_memory = BPFJitMemory::new(BPF_JIT_MEM_PAGES)?;
             // SAFETY: `jit_exec_memory` is moved into the returned
             // `OwnedEbpfVm` after `vm`; field drop order guarantees `vm` is
             // destroyed before the executable mapping is unmapped.
diff --git a/os/StarryOS/kernel/src/perf/kprobe.rs b/os/StarryOS/kernel/src/perf/kprobe.rs
index c66d3cad18..d29a3e1047 100644
--- a/os/StarryOS/kernel/src/perf/kprobe.rs
+++ b/os/StarryOS/kernel/src/perf/kprobe.rs
@@ -16,6 +16,14 @@ use axpoll::Pollable;
 use kbpf_basic::perf::{PerfProbeArgs, PerfProbeConfig};
 use kprobe::{CallBackFunc, KretprobeBuilder, ProbeBuilder, PtRegs};
 
+/// Config value for entry probes (kprobe/uprobe), per Linux PERF_TYPE_PROBE ABI.
+pub const PROBE_CONFIG_ENTRY: u64 = 0;
+/// Config value for return probes (kretprobe/uretprobe), per Linux PERF_TYPE_PROBE ABI.
+pub const PROBE_CONFIG_RETURN: u64 = 1;
+/// Maximum number of concurrently active kretprobe instances, matching
+/// Linux's `max(10, 2*NR_CPUS)` default for single-CPU configurations.
+const KRETPROBE_MAX_ACTIVE: u32 = 10;
+
 use crate::{
     file::FileLike,
     kprobe::{
@@ -182,25 +190,24 @@ fn perf_probe_arg_to_kretprobe_builder(
 ) -> AxResult<KretprobeBuilder<KernelRawMutex>> {
     let symbol = &args.name;
     let addr = lookup_symbol_addr(symbol)?;
-    Ok(KretprobeBuilder::<KernelRawMutex>::new(10)
-        .with_symbol(symbol.clone())
-        .with_symbol_addr(addr))
+    Ok(
+        KretprobeBuilder::<KernelRawMutex>::new(KRETPROBE_MAX_ACTIVE)
+            .with_symbol(symbol.clone())
+            .with_symbol_addr(addr),
+    )
 }
 
 /// Build a `ProbePerfEvent` for a `PERF_TYPE_KPROBE` perf_event_open call.
-/// Config 0 = kprobe; config 1 = kretprobe (per kbpf-basic convention).
+/// Config `PROBE_CONFIG_ENTRY` (0) = kprobe; `PROBE_CONFIG_RETURN` (1) = kretprobe.
 pub fn perf_event_open_kprobe(args: PerfProbeArgs) -> AxResult<ProbePerfEvent> {
     let probe = match args.config {
-        PerfProbeConfig::Raw(val) => {
-            if val == 0 {
-                let builder = perf_probe_arg_to_kprobe_builder(&args)?;
-                ProbeTy::Kprobe(register_kprobe(builder))
-            } else if val == 1 {
-                let builder = perf_probe_arg_to_kretprobe_builder(&args)?;
-                ProbeTy::Kretprobe(register_kretprobe(builder))
-            } else {
-                return Err(AxError::InvalidInput);
-            }
+        PerfProbeConfig::Raw(PROBE_CONFIG_ENTRY) => {
+            let builder = perf_probe_arg_to_kprobe_builder(&args)?;
+            ProbeTy::Kprobe(register_kprobe(builder))
+        }
+        PerfProbeConfig::Raw(PROBE_CONFIG_RETURN) => {
+            let builder = perf_probe_arg_to_kretprobe_builder(&args)?;
+            ProbeTy::Kretprobe(register_kretprobe(builder))
         }
         _ => return Err(AxError::InvalidInput),
     };
diff --git a/os/StarryOS/kernel/src/perf/mod.rs b/os/StarryOS/kernel/src/perf/mod.rs
index 96e7c79c24..dd469626e7 100644
--- a/os/StarryOS/kernel/src/perf/mod.rs
+++ b/os/StarryOS/kernel/src/perf/mod.rs
@@ -32,7 +32,7 @@ use core::{
     any::Any,
     ffi::c_void,
     fmt::Debug,
-    sync::atomic::{AtomicU64, Ordering},
+    sync::atomic::{AtomicBool, AtomicU64, Ordering},
 };
 
 use ax_errno::{AxError, AxResult};
@@ -203,6 +203,10 @@ pub struct PerfEvent {
     /// Unique, stable perf-event id (see [`NEXT_PERF_EVENT_ID`]). Returned by
     /// `PERF_EVENT_IOC_ID` and used as the `read_format` `PERF_FORMAT_ID` value.
     id: u64,
+    /// O_NONBLOCK flag set via `fcntl(F_SETFL)`. When true, operations that
+    /// would block (e.g. reading from an empty ring buffer) should return
+    /// `EAGAIN` instead.
+    nonblocking: AtomicBool,
 }
 
 impl Debug for PerfEvent {
@@ -220,6 +224,7 @@ impl PerfEvent {
         PerfEvent {
             event: SpinNoPreempt::new(event),
             id,
+            nonblocking: AtomicBool::new(false),
         }
     }
 
@@ -396,6 +401,15 @@ impl FileLike for PerfEvent {
             Some(anchor),
         ))
     }
+
+    fn nonblocking(&self) -> bool {
+        self.nonblocking.load(Ordering::Acquire)
+    }
+
+    fn set_nonblocking(&self, on: bool) -> AxResult {
+        self.nonblocking.store(on, Ordering::Release);
+        Ok(())
+    }
 }
 
 /// `perf_event_open(2)` syscall entry. Copies the user `perf_event_attr` in
diff --git a/os/StarryOS/kernel/src/perf/uprobe.rs b/os/StarryOS/kernel/src/perf/uprobe.rs
index 84374b2775..428ce33ae0 100644
--- a/os/StarryOS/kernel/src/perf/uprobe.rs
+++ b/os/StarryOS/kernel/src/perf/uprobe.rs
@@ -15,7 +15,7 @@ use ax_errno::{AxError, AxResult};
 use kbpf_basic::perf::{PerfProbeArgs, PerfProbeConfig};
 use kprobe::ProbeBuilder;
 
-use super::kprobe::{ProbePerfEvent, ProbeTy};
+use super::kprobe::{PROBE_CONFIG_ENTRY, PROBE_CONFIG_RETURN, ProbePerfEvent, ProbeTy};
 use crate::{
     kprobe::KprobeAuxiliary,
     task::{AsThread, get_task},
@@ -74,13 +74,13 @@ fn perf_probe_arg_to_uprobe_builder(
 /// Build a uprobe perf event from `perf_event_open` args.
 pub fn perf_event_open_uprobe(args: PerfProbeArgs) -> AxResult<ProbePerfEvent> {
     let probe = match args.config {
-        PerfProbeConfig::Raw(0) => {
+        PerfProbeConfig::Raw(PROBE_CONFIG_ENTRY) => {
             let builder = perf_probe_arg_to_uprobe_builder(&args)?;
             ProbeTy::Uprobe(crate::uprobe::register_uprobe(builder))
         }
-        PerfProbeConfig::Raw(1) => {
+        PerfProbeConfig::Raw(PROBE_CONFIG_RETURN) => {
             // uretprobe — not implemented for user space yet.
-            warn!("uprobe: uretprobe (config=1) is not yet supported");
+            warn!("uprobe: uretprobe is not yet supported");
             return Err(AxError::Unsupported);
         }
         other => {
diff --git a/os/StarryOS/kernel/src/tracepoint/mod.rs b/os/StarryOS/kernel/src/tracepoint/mod.rs
index 488a420bf5..8ede91f25e 100644
--- a/os/StarryOS/kernel/src/tracepoint/mod.rs
+++ b/os/StarryOS/kernel/src/tracepoint/mod.rs
@@ -27,6 +27,11 @@ use crate::{
     task::AsThread,
 };
 
+/// Maximum number of trace records kept in the raw trace pipe ring buffer.
+const TRACE_RAW_PIPE_CAPACITY: usize = 4096;
+/// Maximum number of PID→cmdline entries in the command-line cache.
+const TRACE_CMDLINE_CACHE_SIZE: usize = 4096;
+
 // The registry entry is locked from the tracepoint fire path, which for
 // `sched:sched_switch` runs inside `axtask::switch_to` (IRQ off,
 // preemption disabled). A sleeping `ax_sync::Mutex` would trip the
@@ -71,7 +76,7 @@ impl TraceState {
     const fn new() -> Self {
         Self {
             point_map: LazyInit::new(),
-            raw_pipe: Mutex::new(TracePipeRaw::new(4096)),
+            raw_pipe: Mutex::new(TracePipeRaw::new(TRACE_RAW_PIPE_CAPACITY)),
             pipe_event: PollSet::new(),
             pipe_notify: IrqNotify::new(),
             cmdline_cache: LazyInit::new(),
@@ -277,7 +282,7 @@ pub fn tracepoint_init() -> AxResult<()> {
     TRACE_STATE
         .cmdline_cache
         .init_once(Mutex::new(TraceCmdLineCache::new(
-            NonZero::new(4096).unwrap(),
+            NonZero::new(TRACE_CMDLINE_CACHE_SIZE).unwrap(),
         )));
     start_trace_pipe_notify_worker();
     Ok(())
diff --git a/test-suit/starryos/qemu-smp1/system/KNOWN_FAILS.md b/test-suit/starryos/qemu-smp1/system/KNOWN_FAILS.md
index 8ce69302d3..9b3147cf2b 100644
--- a/test-suit/starryos/qemu-smp1/system/KNOWN_FAILS.md
+++ b/test-suit/starryos/qemu-smp1/system/KNOWN_FAILS.md
@@ -21,4 +21,6 @@ but the grouped CI runner only executes `/usr/bin/starry-test-suit/*`.
 - `test-tgsigqueueinfo`: queued thread signal delivery with `siginfo`.
 - `test-uid-gid-direct-setters`: uid/gid setter boundary matrix semantics.
 - `test-uid-gid-groups`: user namespace `setgroups=deny` behavior.
+- `test-ptrace-gdb`: 19/20 pass, 1 remaining assertion failure on aarch64.
 - `test-uid-gid-res-setters`: setresuid/setresgid boundary matrix semantics.
+- `test-copy-file-range`: 35/37 pass, 2 assertion failures on loongarch64 (cross-process fd: errno=9 EBADF).