{"id":"CVE-2026-99202","published":"2026-10-28T11:17:41.377","lastModified":"2026-10-28T11:17:41.377","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nscsi: libsas: Fix SMP IO deadlock during HA resume\n\nsmp_execute_task_sg() calls pm_runtime_get_sync() on the host before\nissuing an SMP command. When that command is itself issued from the HA\nresume path, the get_sync() deadlocks: it waits for the ongoing resume\n(the device is RPM_RESUMING), while the resume is blocked in\nsas_drain_work() waiting for that same SMP IO to complete.\n\nThe deadlock needs an expander-attached SATA disk. During\nsas_resume_ha() -> sas_drain_work(), DISCE_RESUME -> sas_resume_sata()\n-> ata_sas_port_resume() requests ATA_EH_RESET, and the hard reset for\nsuch a disk is done via SMP PHY CONTROL (sas_ata_hard_reset() ->\nsas_phy_reset() -> sas_smp_phy_control() ->\nsmp_execute_task_sg()). Direct-attached SATA resets through\nlldd_control_phy() and SSP devices use TMFs, so neither hits this.\n\nReplace the get_sync()/put_sync() pair with\npm_runtime_get_noresume()/pm_runtime_put(). smp_execute_task_sg() only\nneeds to hold off autosuspend while the SMP is in flight, and it must\nnot try to resume the host: a sync resume issued from the HA resume path\nitself is what deadlocks, and by the time sas_resume_ha() runs, hw_init\nhas already reinitialized the hardware, so the device is accessible\nwithout one.\n\nThe usage reference is still required. Discovery work normally runs\ninside an event worker's PM reference, taken at sas_notify_port_event()\nnotify time and held until the handler has flushed the disco\nqueue. sas_rediscover_ex_phy() however requeues DISCE_REVALIDATE_DOMAIN\nfrom within the revalidation worker itself, and flush_workqueue() does\nnot wait for work items queued during execution, so that chained\nrevalidation runs with no outer PM reference - without the\nget_noresume(), its SMP could race autosuspend.\n\nFor the BSG path, sas_smp_handler() is the only caller which may find\nthe host autosuspended: expander SMP requests do not go through any SCSI\ndevice request queue, so nothing else in that path holds the host\nawake. Resume it there with pm_runtime_resume_and_get() and check the\nresult.","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/08b95ef36a64c3ae4490794bed3c35d0c3354e85","tags":[]},{"url":"https://git.kernel.org/stable/c/81e097a064b28b4a1f6ca1a8b433e659c1bd64c0","tags":[]},{"url":"https://git.kernel.org/stable/c/c9fb89246d9922ac0a45d504adb03e0bb971310c","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}