diff --git a/crates/memtrack/src/ebpf/memtrack/mod.rs b/crates/memtrack/src/ebpf/memtrack/mod.rs
index 8586872e..4c32a33a 100644
--- a/crates/memtrack/src/ebpf/memtrack/mod.rs
+++ b/crates/memtrack/src/ebpf/memtrack/mod.rs
@@ -136,6 +136,8 @@ impl MemtrackBpf {
     /// would detect. Either attaches given host privileges; the token only
     /// matters when `bpf()` is called from an unprivileged user namespace.
     pub fn with_variant(variant: BpfVariant, track_rmap: bool) -> Result<Self> {
+        crate::kernel::KernelBtf::ensure_available()?;
+
         let page_shift = page_shift()?;
         let rmap = if track_rmap {
             RmapSupport::detect()
diff --git a/crates/memtrack/src/kernel.rs b/crates/memtrack/src/kernel.rs
index bafffa33..c99cf045 100644
--- a/crates/memtrack/src/kernel.rs
+++ b/crates/memtrack/src/kernel.rs
@@ -1,6 +1,15 @@
 use crate::prelude::*;
 use std::fmt;
 
+/// The running kernel's full release, e.g. `6.12.8+`.
+fn kernel_release() -> Result<String> {
+    const OSRELEASE_PATH: &str = "/proc/sys/kernel/osrelease";
+
+    std::fs::read_to_string(OSRELEASE_PATH)
+        .map(|release| release.trim().to_owned())
+        .with_context(|| format!("Failed to read {OSRELEASE_PATH}"))
+}
+
 /// A kernel release, ordered by `(major, minor)`. The patch level is ignored:
 /// features are introduced in merge windows, never in a stable point release.
 #[derive(Debug, Copy, Clone, PartialEq, Eq, PartialOrd, Ord)]
@@ -16,12 +25,8 @@ impl KernelVersion {
 
     /// The running kernel's release.
     pub fn current() -> Result<Self> {
-        const PATH: &str = "/proc/sys/kernel/osrelease";
-
-        let release =
-            std::fs::read_to_string(PATH).with_context(|| format!("Failed to read {PATH}"))?;
-        Self::parse(&release)
-            .with_context(|| format!("Failed to parse kernel release {:?}", release.trim()))
+        let release = kernel_release()?;
+        Self::parse(&release).with_context(|| format!("Failed to parse kernel release {release:?}"))
     }
 
     /// Parse the leading `<major>.<minor>` of a release string, ignoring
@@ -46,6 +51,40 @@ impl fmt::Display for KernelVersion {
     }
 }
 
+/// Whether the running kernel exposes its own BTF.
+///
+/// libbpf needs it to resolve CO-RE relocations, and the kernel resolves the
+/// attach target of every `fentry`/`tp_btf` program against it, so a kernel
+/// without BTF cannot load the programs at all. Detecting it up front replaces
+/// libbpf's bare `-ESRCH` with something the reader can act on.
+///
+/// Minimal kernels built for fast boot — microVM images in particular — commonly
+/// drop `CONFIG_DEBUG_INFO_BTF`, so the message has to name the option.
+pub struct KernelBtf;
+
+impl KernelBtf {
+    /// Present only on a kernel built with `CONFIG_DEBUG_INFO_BTF`.
+    const PATH: &'static str = "/sys/kernel/btf/vmlinux";
+
+    pub fn is_available() -> bool {
+        std::path::Path::new(Self::PATH).exists()
+    }
+
+    pub fn ensure_available() -> Result<()> {
+        if Self::is_available() {
+            return Ok(());
+        }
+
+        let release = kernel_release().unwrap_or_else(|_| "unknown".to_owned());
+        bail!(
+            "Memory profiling is not supported on this runner: its kernel ({release}) \
+             was built without BTF, so {} does not exist. Use a runner whose kernel is \
+             built with CONFIG_DEBUG_INFO_BTF=y.",
+            Self::PATH
+        );
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
diff --git a/crates/memtrack/src/lib.rs b/crates/memtrack/src/lib.rs
index 1d3278d9..ccd93399 100644
--- a/crates/memtrack/src/lib.rs
+++ b/crates/memtrack/src/lib.rs
@@ -14,7 +14,7 @@ pub use ipc::{
     IpcCommand as MemtrackIpcCommand, IpcMessage as MemtrackIpcMessage,
     IpcResponse as MemtrackIpcResponse, MemtrackIpcClient, MemtrackIpcServer,
 };
-pub use kernel::KernelVersion;
+pub use kernel::{KernelBtf, KernelVersion};
 
 #[cfg(feature = "ebpf")]
 pub use ebpf::*;