Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB
CVE-2026-93259 Overview
CVE-2026-93259 affects the Linux kernel's PowerPC interrupt handling code when built with CONFIG_PPC_KERNEL_PCREL enabled. The flaw resides in call_do_irq() and call_do_softirq(), which use inline assembly to perform stack switching but fail to list register r2 in their clobber lists. In PCREL mode, r2 is no longer reserved for the Table of Contents (TOC) pointer and becomes a caller-saved register. The compiler assumes r2 is preserved across the inline assembly calls, leading to register corruption when __do_irq and __do_softirq clobber it.
Critical Impact
Register corruption during interrupt handling causes kernel crashes and prevents the system from booting on PCREL-enabled PowerPC builds compiled with GCC 14.
Affected Products
- Linux kernel builds for the PowerPC architecture with CONFIG_PPC_KERNEL_PCREL enabled
- Kernels compiled using GCC 14, which allocates r2 for values spanning function calls
- Stable kernel branches referenced by commits 00be690, 602ee44, 7b0093b, 90d9530, and dcc442a
Discovery Timeline
- 2026-09-24 - CVE-2026-93259 published to NVD
- 2026-09-24 - Last updated in NVD database
Technical Details for CVE-2026-93259
Vulnerability Analysis
The defect is a kernel-level register handling bug [Kernel Vulnerability] in PowerPC interrupt dispatch. Both call_do_irq() and call_do_softirq() perform stack switching through inline assembly before invoking __do_irq and __do_softirq. The inline assembly's clobber list omits r2, so the compiler is free to keep live values in r2 across the call boundary. In PCREL builds the called functions legitimately use r2 as a scratch register, overwriting any caller value stored there.
The practical result is memory corruption when the caller later dereferences the clobbered register. The upstream report captured the failure as a kernel data access fault at 0xc000000404697638 originating from __do_IRQ+0x6c/0xc0, oopsing with signal 11 and halting boot.
Root Cause
In CONFIG_PPC_KERNEL_PCREL mode, register r2 is reclassified as a caller-saved general purpose register rather than the dedicated TOC pointer. The inline assembly in call_do_irq() and call_do_softirq() was written under the earlier ABI assumption and never updated to declare r2 as clobbered. Older GCC releases hid the bug by conservatively assigning callee-saved registers such as r31 for cross-call values. GCC 14 selects r2 for the same role, exposing the missing constraint.
Attack Vector
The vulnerability triggers automatically during normal interrupt processing on affected builds; no attacker interaction is required. The impact is loss of availability through a boot-time or runtime kernel crash. The reporter did not identify an exploitation path for privilege escalation or information disclosure, and no public proof-of-concept exploit is available. The condition is a stability and reliability defect rather than a remotely reachable security primitive.
Detection Methods for CVE-2026-93259
Indicators of Compromise
- Kernel oops with the message BUG: Unable to handle kernel data access on write originating from __do_IRQ or related softirq paths on PowerPC hosts
- Boot failures or spontaneous panics on PowerPC systems after upgrading the toolchain to GCC 14 and rebuilding the kernel with CONFIG_PPC_KERNEL_PCREL=y
- Faulting instruction addresses within the __do_IRQ+0x6c/0xc0 range in crash logs
Detection Strategies
- Inspect kernel build configuration for CONFIG_PPC_KERNEL_PCREL=y combined with a GCC 14 toolchain and a kernel source predating the fix commits
- Collect kernel crash dumps and dmesg output from PowerPC fleet members and search for faults in __do_irq or __do_softirq
- Compare running kernel package versions against the stable series containing commits 00be690, 602ee44, 7b0093b, 90d9530, or dcc442a
Monitoring Recommendations
- Forward kernel log messages from PowerPC hosts to a centralized logging system and alert on repeated oops events in interrupt handlers
- Track reboot counters and uptime metrics on PowerPC servers to identify crash loops consistent with early-boot interrupt failures
- Monitor configuration management records for kernel builds that enable PCREL mode without the corresponding patch
How to Mitigate CVE-2026-93259
Immediate Actions Required
- Apply the upstream Linux kernel patches referenced by commits 00be690, 602ee44, 7b0093b, 90d9530, and dcc442a to add r2 to the clobber list in call_do_irq() and call_do_softirq()
- Rebuild and redeploy PowerPC kernels after applying the fix, prioritizing systems using GCC 14 or newer
- Validate interrupt handling stability through boot and stress testing before promoting the patched kernel to production
Patch Information
The fix adds r2 to the clobber list of the inline assembly in both call_do_irq() and call_do_softirq() when CONFIG_PPC_KERNEL_PCREL is enabled. This informs the compiler that r2 is not preserved across the call, preventing it from storing live values there. The patch is distributed across the stable kernel branches via the five referenced commits on git.kernel.org.
Workarounds
- Disable CONFIG_PPC_KERNEL_PCREL and rebuild the kernel without position-independent code support until patches can be applied
- Downgrade the kernel build toolchain to a GCC version prior to 14, which conservatively allocates callee-saved registers and avoids the trigger condition
- Pin PowerPC hosts to a previously stable kernel package until the patched build is validated
# Verify whether the running PowerPC kernel was built with PCREL enabled
zcat /proc/config.gz | grep CONFIG_PPC_KERNEL_PCREL
# Confirm installed kernel version against the fixed stable releases
uname -r
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB

