{"id":"CVE-2026-99148","kind":"vulnerability","aliases":["CVE-2026-99148"],"related":[],"title":"Linux amdgpu: closing one BO handle drops eviction fence, user queues unsuspended","description":"In the Linux kernel, the following vulnerability has been resolved:\n\ndrm/amdgpu: keep the eviction fence until the last handle is closed\n\namdgpu_gem_object_close() detaches the user queue eviction fence from\nthe BO before it checks bo_va->ref_count, so closing any handle of a BO\nremoves the fence, even when the file still has another handle to the\nBO and the bo_va with its GPU mappings stays. The BO can then be evicted\nwithout suspending the user queues of the file, which keep accessing it\nthrough mappings that are only marked for update.\n\nSeen on navi31 with a SDMA user queue: a BO opened twice in the same\nfile (GEM_FLINK + GEM_OPEN) loses its eviction fence when the second\nhandle is closed (traced attach/detach of the fence):\n\n  attach bo=ffff8c78a6c84c00\n  GEM_CLOSE handle=7\n  detach bo=ffff8c78a6c84c00\n\nwhile handle 6 and the mapping are still in use.\n\nOnly detach the fence when the last reference to the bo_va goes away,\nmatching amdgpu_gem_object_open(), which attaches it only when the\nbo_va is created. With this the fence stays until the file is closed.\n\n(cherry picked from commit 8387982f03d4026b7dc11b0b5e3118298eee4d6a)","summary":"In the Linux kernel's AMD GPU driver, closing one of two handles to a graphics buffer removes the eviction fence even while another handle and its GPU mappings remain in use. The buffer can then be evicted without suspending the user queues that still access it, which can lead to memory corruption or unintended GPU memory access.","severity":"unknown","cvss":null,"epss":null,"exploited":false,"exploited_at":null,"products":[],"ecosystems":[],"tags":["amd-gpu","amdgpu","linux-kernel","other"],"not_affected_if":["no-amd-gpu"],"how_to_check":"- unknown","fix":"- Install a Linux kernel build that includes upstream commit 8387982f03d4026b7dc11b0b5e3118298eee4d6a (drm/amdgpu: keep the eviction fence until the last handle is closed).","status":"open","published_at":"2026-10-28T11:17:34+00:00","first_seen_at":"2026-10-28T11:32:35+00:00","triaged_at":"2026-10-28T11:35:59+00:00","updated_at":"2026-10-28T11:35:59+00:00","category":"other","confidence":0.5,"triage":{"at":"2026-10-28T11:35:59Z","model":"claude-haiku-5-5","output":{"fix":"- Install a Linux kernel build that includes upstream commit 8387982f03d4026b7dc11b0b5e3118298eee4d6a (drm/amdgpu: keep the eviction fence until the last handle is closed).","reach":"local","title":"Linux amdgpu: closing one BO handle drops eviction fence, user queues unsuspended","impact":"medium","summary":"In the Linux kernel's AMD GPU driver, closing one of two handles to a graphics buffer removes the eviction fence even while another handle and its GPU mappings remain in use. The buffer can then be evicted without suspending the user queues that still access it, which can lead to memory corruption or unintended GPU memory access.","category":"other","nicknames":[],"who_cares":["linux-kernel","amdgpu","amd-gpu"],"confidence":0.5,"needs_auth":"unknown","worst_case":"unknown, sources do not describe the outcome beyond unsuspended queue access","analyst_note":"Check the kernel.org stable commit 8387982f03d4026b7dc11b0b5e3118298eee4d6a and NVD for CVSS score and fixed stable versions.","how_to_check":"- unknown","needs_analyst":true,"not_affected_if":["no-amd-gpu"],"who_is_affected":"Systems running a Linux kernel with the amdgpu driver and AMD GPU user queues (for example navi31 with SDMA user queues) where a buffer object is opened through more than one handle in the same file. Systems without AMD GPUs using amdgpu user queues can likely skip it; the sources do not say more than that.","user_interaction":"no","needs_special_setup":"unknown","severity_if_unknown":"unknown"},"cost_usd":0.00066,"input_tokens":3656,"output_tokens":589,"severity_before":"unknown"},"analyst_status":null,"analyst_note":null,"search":"'-2026':14A,17A '-99148':15A,18A '6':228C '7':222C '8387982f03d4026b7dc11b0b5e3118298eee4d6a':279C 'access':71B,83B,171C 'amd':20B,31B 'amd-gpu':19B 'amdgpu':2A,22B,105C,251C 'anoth':49B,141C 'attach':216C,256C 'attach/detach':212C 'away':249C 'bo':5A,117C,121C,131C,145C,148C,156C,190C,217C,224C,246C,261C 'buffer':42B,59B 'check':120C 'cherri':275C 'close':3A,34B,104C,108C,126C,210C,220C,274C 'commit':278C 'corrupt':78B 'count':124C 'creat':264C 'cve':13A,16A 'detach':109C,223C,237C 'driver':33B 'drm/amdgpu':94C 'drop':7A 'even':47B,135C 'evict':8A,45B,63B,97C,113C,160C,203C 'fenc':9A,46B,98C,114C,134C,204C,215C,239C,268C 'ffff8c78a6c84c00':218C,225C 'file':138C,168C,196C,272C 'flink':198C 'follow':89C 'gem':106C,197C,199C,219C,252C 'goe':248C 'gpu':21B,32B,53B,81B,152C 'graphic':41B 'handl':6A,38B,50B,102C,128C,142C,208C,221C,227C 'keep':95C,170C 'kernel':25B,29B,87C 'last':101C,242C 'lead':75B 'linux':1A,24B,28B,86C 'linux-kernel':23B 'lose':201C 'map':54B,153C,174C,231C 'mark':178C 'match':250C 'memori':77B,82B 'navi31':183C 'object':107C,253C 'one':4A,35B 'open':191C,200C,254C 'pick':276C 'queue':11A,68B,112C,165C,188C 'ref':123C 'refer':243C 'remain':55B 'remov':43B,132C 'resolv':93C 'sdma':186C 'second':207C 'seen':181C 'stay':154C,269C 'still':70B,139C,233C 'suspend':65B,162C 'trace':211C 'twice':192C 'two':37B 'unintend':80B 'unsuspend':12A 'updat':180C 'use':57B,235C 'user':10A,67B,111C,164C,187C 'va':122C,149C,247C,262C 'vulner':90C 'without':64B,161C","nicknames":[],"mitigations":[],"who_affected":"Systems running a Linux kernel with the amdgpu driver and AMD GPU user queues (for example navi31 with SDMA user queues) where a buffer object is opened through more than one handle in the same file. Systems without AMD GPUs using amdgpu user queues can likely skip it; the sources do not say more than that.","tier":"C","verified":{},"analyst_revisit_at":null,"first_public_at":"2026-10-28T11:17:34+00:00","kev":null,"cvss_vector":null,"cvss_by":null,"poc_count":0,"sources":[{"source":"nvd","external_id":"CVE-2026-99148","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-99148","severity":null,"published_at":"2026-10-28T11:17:34+00:00","modified_at":"2026-10-28T11:17:34+00:00","first_seen_at":"2026-10-28T11:32:35+00:00","fetched_at":"2026-10-28T11:32:35+00:00","note":null}],"documents":[],"notice":"Free to read at a human pace. An API with keys and an MCP server are in preparation; early access: hello@infrastack.ai"}