{"dataType":"CVE_RECORD","dataVersion":"5.2","cveMetadata":{"cveId":"CVE-2026-99485","assignerOrgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","state":"PUBLISHED","assignerShortName":"Linux","dateReserved":"2026-09-25T10:25:14.418Z","datePublished":"2026-11-19T14:43:23.095Z","dateUpdated":"2026-11-22T11:08:38.616Z"},"containers":{"cna":{"providerMetadata":{"orgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","shortName":"Linux","dateUpdated":"2026-11-22T11:08:38.616Z"},"descriptions":[{"lang":"en","value":"In the Linux kernel, the following vulnerability has been resolved:\n\naccel/amdxdna: clear the mailbox channel pointer when starting it fails\n\naie2_create_context() allocates a mailbox channel, starts it, and on a\nfailed start frees it again:\n\n\thwctx->priv->mbox_chann = xdna_mailbox_alloc_channel(ndev->mbox);\n\t...\n\tret = xdna_mailbox_start_channel(hwctx->priv->mbox_chann, ...);\n\tif (ret)\n\t\tgoto free_channel;\n\t...\nfree_channel:\n\txdna_mailbox_free_channel(hwctx->priv->mbox_chann);\n\nhwctx->priv->mbox_chann keeps pointing at the freed channel. That\npointer is the driver's own test for whether a hardware context has a\nusable channel: aie2_destroy_context() returns early on NULL,\naie2_sched_job_run() refuses to run a job on NULL, and the message\nhelpers return -ENODEV on NULL. A stale pointer passes all of them.\n\nIt matters on the restart path. aie2_hwctx_restart() calls\naie2_create_context() again on a hardware context that is already live,\nafter aie2_hwctx_stop() has torn its channel down, and the context\nsurvives a failure there: aie2_sched_job_timedout() discards the return\nvalue entirely, and aie2_hwctx_resume() only propagates it. What is\nleft is a live hardware context holding a freed channel, and every\nlater user of it takes the non-NULL branch:\n\n  - the next job submitted reaches xdna_mailbox_send_msg() on the freed\n    channel, via the !mbox_chann guard in aie2_sched_job_run(),\n  - a second command timeout reaches aie2_hwctx_stop() ->\n    aie2_destroy_context(),\n  - closing the device reaches aie2_hwctx_fini() ->\n    aie2_release_resource() -> xrs_release_resource() ->\n    aie2_xrs_unload() -> aie2_destroy_context(),\n\nand aie2_destroy_context() then calls xdna_mailbox_stop_channel()\nfollowed by xdna_mailbox_free_channel() on memory that was already\nfreed: a use-after-free, and a second free of the same channel.\n\nThe AIE4 management channel already does this correctly -\naie4_mailbox_start() clears ndev->aie.mgmt_chann right after freeing\nit. Do the same here.\n\nThe create path itself is not affected: when aie2_xrs_load() fails,\nxrs_allocate_resource() removes the solver node without calling\n->unload, and aie2_hwctx_init() frees hwctx->priv, so the stale pointer\nnever outlives the structure holding it.\n\nStarting a channel can fail today without this patch: the ring buffer\nsizes firmware reports are rejected unless both are powers of two, and\nrequest_irq() can fail."}],"metrics":[{"cvssV3_1":{"version":"3.1","vectorString":"CVSS:3.1/AV:L/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H","baseScore":7,"baseSeverity":"HIGH"},"scenarios":[{"lang":"en","value":"AV:L - aie2_create_context() frees hwctx->priv->mbox_chann on the free_channel path when xdna_mailbox_start_channel() fails and, before the fix, leaves that pointer set. The surviving caller is aie2_hwctx_restart(), from aie2_sched_job_timedout() or aie2_hw_resume(), after AMDXDNA_CREATE_HWCTX and AMDXDNA_EXEC_CMD on the accel node opened by accel_open().\nAC:H - xdna_mailbox_start_channel() fails only if the firmware create-context cq_pair buf_size values are not powers of two or request_irq() rejects the MSI-X vector, and only aie2_hwctx_restart() keeps the hwctx alive afterward. The caller can idle into amdxdna_pm autosuspend or trip the 2s TDR, but cannot force those two start failures; a failed first aie2_xrs_load() frees hwctx->priv instead.\nPR:L - amdxdna_drm_create_hwctx_ioctl() and the AMDXDNA_EXEC_CMD handler amdxdna_drm_submit_execbuf() are registered with ioctl flags 0, and accel_open() files are DRM_MINOR_ACCEL so drm_ioctl_permit() does not demand DRM_RENDER_ALLOW or CAP_SYS_ADMIN. A local user who can open /dev/accel/accel* can create the context and submit the command; DRM_ROOT_ONLY applies only to AMDXDNA_SET_STATE, which this path does not use.\nUI:N - The attacking process opens the accel node, calls AMDXDNA_CREATE_HWCTX, and either drops the runtime-PM reference in aie2_hwctx_init() so the 5s autosuspend runs aie2_hw_suspend() then aie2_hw_resume(), or submits AMDXDNA_EXEC_CMD so aie2_sched_job_timedout() calls aie2_hwctx_restart(). Each step is on that process's own fd.\nS:U - The twice-freed object is the mailbox_channel from xdna_mailbox_alloc_channel() for this hwctx. aie2_init() bails out unless hypervisor_is_type() is X86_HYPER_NATIVE, and the heap corruption is a privilege escalation inside that same kernel.\nC:H - aie2_sched_job_run() treats the non-NULL mbox_chann as live and calls xdna_mailbox_send_msg(), whose mailbox_reg_read() and xa_alloc_cyclic_irq() follow mb, res[], and chan_xa in the freed channel. Reclaiming that kmalloc slot lets those reads target attacker-chosen kernel memory.\nI:H - mailbox_send_msg() writel()s and memcpy_toio()s through mb->res.mbox_base and the ring offsets in the freed channel, using the caller's exec payload, and aie2_destroy_context() then free_irq()s, drain_workqueue()s, and xdna_mailbox_free_channel()s that same pointer a second time.\nA:H - With the stale pointer set, close runs aie2_hwctx_fini() to aie2_xrs_unload() to aie2_destroy_context(), which drain_workqueue()s the work_q already destroyed in the first xdna_mailbox_free_channel() and kfree()s the channel again. aie2_sched_job_timedout() discards the restart error, so the live context remains to be torn down."}]}],"affected":[{"product":"Linux","vendor":"Linux","defaultStatus":"unaffected","repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["drivers/accel/amdxdna/aie2_message.c"],"versions":[{"version":"d5b8b0347fa8470b751a506fb801797e271d7548","lessThan":"35003a3a2221207ded92ca4c60a54c2f83136473","status":"affected","versionType":"git"},{"version":"d5b8b0347fa8470b751a506fb801797e271d7548","lessThan":"2117e3d96a4fd1ee13fa28b31494ff5f5db6ccf4","status":"affected","versionType":"git"}]},{"product":"Linux","vendor":"Linux","defaultStatus":"affected","repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["drivers/accel/amdxdna/aie2_message.c"],"versions":[{"version":"7.0","status":"affected"},{"version":"0","lessThan":"7.0","status":"unaffected","versionType":"semver"},{"version":"7.3.2","lessThanOrEqual":"7.3.*","status":"unaffected","versionType":"semver"},{"version":"7.4-rc1","lessThanOrEqual":"*","status":"unaffected","versionType":"original_commit_for_fix"}]}],"cpeApplicability":[{"nodes":[{"operator":"OR","negate":false,"cpeMatch":[{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"7.0","versionEndExcluding":"7.3.2"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"7.0","versionEndExcluding":"7.4-rc1"}]}]}],"references":[{"url":"https://git.kernel.org/stable/c/35003a3a2221207ded92ca4c60a54c2f83136473"},{"url":"https://git.kernel.org/stable/c/2117e3d96a4fd1ee13fa28b31494ff5f5db6ccf4"}],"title":"accel/amdxdna: clear the mailbox channel pointer when starting it fails","x_generator":{"engine":"bippy-1.2.0"}}}}