{"id":"CVE-2026-98020","published":"2026-09-25T11:17:30.480","lastModified":"2026-09-25T11:17:30.480","description":"In the Linux kernel, the following vulnerability has been resolved:\n\npds_core: fix cmd_regs access racing BAR unmap on reset\n\npdsc_reset_prepare() and pdsc_reset_done()'s pdsc_map_bars() error path\nclear/iounmap cmd_regs without devcmd_lock, and\npdsc_legacy_firmware_update()'s download loop derefs cmd_regs after\ndropping and retaking the lock without re-checking. An FLR concurrent\nwith a devlink flash can unmap cmd_regs under an in-flight devcmd,\ncausing a NULL deref or a write to unmapped MMIO.\n\nTake devcmd_lock across the BAR unmap/remap, and re-check cmd_regs in\nthe download loop. Only the PF maps cmd_regs and runs devcmd, so skip\nthe unmap on a VF, as pdsc_remove() and pdsc_reset_done() already do.\n\nA reset that completes entirely within the unlocked window is not a\ncorrectness problem for the image: the device clears its update session,\nso a resumed download is rejected, and it verifies the staged image\nbefore writing a flash slot, reporting PDS_RC_BAD_FW rather than\nactivating it.\n\npdsc_unmap_bars() also clears info_regs, intr_status and intr_ctrl. The\ninterrupt and start/stop readers of those are quiesced before the unmap\nby pdsc_fw_down(), which frees the interrupts and tears down the queues.\nThe debugfs readers are not, since those files outlive a reset; that is\npre-existing and out of scope here.","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/09f831bfe39de5b8026fefb3d106b5cc93272170","tags":[]},{"url":"https://git.kernel.org/stable/c/2cc697565fd19b5ba2d100cdd4a20dd6d263abc2","tags":[]},{"url":"https://git.kernel.org/stable/c/7980325b2f71e3f65c1323c39792e2455da6fab6","tags":[]},{"url":"https://git.kernel.org/stable/c/fa31bd14c5042c6315bb2182c963f03ca6e79ca4","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}