CVE-2026-89596

In the Linux kernel, the following vulnerability has been resolved:

forcedeth: fix off-by-one when saving/restoring non-PCI config space

nv suspend() and nv resume() walk the non-PCI configuration space with

for (i = 0; i <= np->register size/sizeof(u32); i++)

which runs one iteration too many. saved config space is declared as

u32 saved config space[NV PCI REGSZ MAX/4];

and NV PCI REGSZ VER3 is equal to NV PCI REGSZ MAX (0x604), so on a VER3 device register size/sizeof(u32) is exactly the array length and the last iteration addresses one element past the end.

The element it lands on is np->name rx[0..3]: saved config space[] is followed immediately by char name rx[IFNAMSIZ + 3], and char needs no padding. Nothing observable is corrupted by that, because nv request irq() rewrites name rx with sprintf() before it is ever passed to request irq(). The bug is the out-of-bounds access itself, which UBSAN reports and which CONFIG UBSAN TRAP=y turns into a trap that aborts the running kernel code, plus an MMIO read and, on resume, an MMIO writel() to base + 0x604, one dword past the range the driver mapped:

np->base = ioremap(addr, np->register size);

VER1 and VER2 devices stay inside the array, but they too get the stray read and the stray write one dword past their own window.

Caught by UBSAN on an Apple Macmini3,1 (MCP79) during a deep S3 cycle. The splat below is trimmed: the build path in the file name, the CPU and taint lines, the Workqueue line, the "?" hint frames, and the frames below device suspend are all cut. The kernel was tainted, with an out-of-tree nouveau and CPU OUT OF SPEC; forcedeth itself was the stock module.

UBSAN: array-index-out-of-bounds in drivers/net/ethernet/nvidia/forcedeth.c:6225:25 index 385 is out of range for type 'u32 [385]' Call Trace: dump stack lvl+0x5d/0x80 ubsan epilogue+0x5/0x2b ubsan handle out of bounds.cold+0x54/0x59 this module+0xe398c/0xe9010 [forcedeth] pci pm suspend+0x80/0x170 dpm run callback+0x51/0x160 device suspend+0x1a2/0x4a0 ...

Both loops are hit. UBSAN reports each source location only once per module load ( ubsan handle out of bounds() calls suppress report(), which does test and set bit(REPORTED BIT, ...) on the struct source location), so the two splats land in the first S3 cycle after the module is loaded and later cycles are silent even though the access still runs off the end every time. In that first cycle line 6225 is reported from pci pm suspend and line 6240 from pci pm resume.

The same off-by-one was fixed in nv get regs() by commit ba9aa134287f ("forcedeth: fix buffer overflow") in 2012; these two loops were missed. The suspend and resume side was reported on LKML in September 2013 by Marc Weber, with the same analysis and the same one-character fix, but the patch was attached rather than sent inline and the thread ended there.

Use < instead of <=, which saves and restores exactly register size bytes.