[PATCH v4 00/18] PCI/P2PDMA: Fix ACS egress control handling
Leon Romanovsky <[email protected]>
| Newsgroups | org.kernel.vger.linux-pci,dev.linux.lists.iommu,org.kernel.vger.linux-doc,org.kernel.vger.linux-kernel |
|---|---|
| Message-ID | <[email protected]> |
PCI P2PDMA treats any enabled ACS P2P Egress Control bit as an upstream redirect. PCIe r7.0, sec 6.12.3, table 6-11 says the Egress Control Vector bit for the target port decides instead: a clear bit routes a peer request directly, regardless of P2P Request Redirect. Firmware can therefore enable Egress Control with a permissive vector while Linux incorrectly rejects a valid direct P2P path. Table 6-11, where E is ACS P2P Egress Control Enable, R is ACS P2P Request Redirect Enable and V the Egress Control Vector bit for the target port: E R V Required Handling for Peer-to-Peer Requests - - - ------------------------------------------ 0 0 x Route directly to peer-to-peer target 0 1 x Redirect Upstream 1 0 1 Handle as an ACS Violation 1 0 0 Route directly to peer-to-peer target 1 1 1 Redirect Upstream 1 1 0 Route directly to peer-to-peer target P2P Completion Redirect lies outside this table and forces host-bridge routing when set at the provider-side path divergence. The same interaction affects target-independent ACS isolation checks. Request Redirect does not guarantee that peer requests are forwarded upstream while Egress Control is enabled because a clear vector bit overrides it. Such checks cannot identify every potential target, so treat Request Redirect as ineffective while Egress Control is enabled, which merges the affected devices into one IOMMU group. ACS Direct Translated P2P routes a Request carrying a Translated address to the peer regardless of Request Redirect and Egress Control, so it voids the same guarantee unless Translation Blocking rejects the Request first. That last rule holds only for a caller that needs Request Redirect to isolate peers. pci_enable_pasid() asks for it so that a Request carrying a PASID reaches the translation agent (sec 2.2.10.4), and a Translated Request already carries an address the agent produced for that PASID (sec 10.1.3). pci_acs_enabled() and pci_acs_path_enabled() therefore take a scope, and Direct Translated P2P applies only to PCI_ACS_SCOPE_ALL. A pre-existing gap comes first. The routing analysis covers only Requests carrying an Untranslated address; ACS Direct Translated P2P overrides those controls, so that scope is now written down rather than implied. It is nearly impossible to test all possible combinations due to limited hardware availability, so I added KUnit coverage for ACS routing decisions, isolation checks, Egress Control Vector lookups, and provider-to-client path traversal over a fabricated PCIe fabric. Disclaimer: All patches were prepared with AI assistance, with a significant difference between the code changes and the KUnit tests. The code changes were thoroughly reviewed and rewritten. In contrast, the KUnit patches were produced entirely by AI with minimal human interaction, and multiple AI tools (Claude, Codex, and Gemini) with frontier models were used to verify that the tests comply with the PCI specification. Thanks Signed-off-by: Leon Romanovsky <[email protected]> --- Changes in v4: - Added debug prints (we can drop it) patch which is very useful for automatic root cause analysis. Just feed the output of these prints, together with topology and kernel boot command line to your favorite LLM and it will give you reliable RCA why ACS didn't work. - Reject ACS Violations and unreadable routing state instead of treating them as host-bridge redirects - Added Tested-by tags from Tushar Dave - Added support to asymmetric ACS routing - Limited redirect checks to the two ports at the path divergence - Added standalone ACS routing diagnostics for hardware retesting - Link to v3: https://patch.msgid.link/[email protected] Changes in v3: - Fixed pci_p2pdma_add_resource() error unwinding - Made pdev->p2pdma teardown wait unconditionally for RCU readers - Restricted pci_p2pmem_find_many() to pool-backed providers - Documented the pdev->p2pdma lifetime and RCU rules - Fixed calc_map_type_and_dist() handling of the verbose argument - Required the ACS port and target to share a bus before indexing the Egress Control Vector - Gave pci_acs_enabled() and pci_acs_path_enabled() a scope, so the ACS Direct Translated P2P rule no longer stops pci_enable_pasid() from enabling PASID - Dropped "Report ACS ports when the paths share no upstream bridge": the mapping type cannot change without a shared upstream bridge, so the pci=disable_acs_redir= hint was not actionable there and the ACS walk only cost config space reads - Folded the Request Redirect rule into pci_acs_rr_ineffective(), so pci_acs_flags_enabled() and the Intel SPT PCH quirk share one copy - Renamed pci_acs_egress_ctrl_set() to pci_acs_egress_ctrl_is_set(), it reads the bit rather than setting it - Reworded the blocked-path warning: ACS may also leave the direct route indeterminate rather than blocked - Added KUnit coverage for the shared-bus guard, a device with no ACS capability and an unreadable ACS Control register - Added the missing Fixes: tags, a second one on the pci_p2pdma_add_resource() unwinding fix (the dangling devres action dates to f58ef9d1d135) and one on the Egress Control isolation change - Link to v2: https://patch.msgid.link/[email protected] Changes in v2: - Added Logan's ROB tags - Added commas in Documentation patch - Link to v1: https://patch.msgid.link/[email protected] --- Leon Romanovsky (18): PCI/P2PDMA: Do not tear down the allocate attribute on registration failure PCI/P2PDMA: Wait for RCU readers before freeing state PCI/P2PDMA: Restrict the p2pmem search to pool backed providers PCI/P2PDMA: Safely terminate ACS redirect lists PCI/P2PDMA: Document the pdev->p2pdma lifetime and RCU rules PCI/P2PDMA: Gate the host bridge whitelist warning on verbose PCI/P2PDMA: Document the Address Type assumption PCI: Account for Direct Translated P2P in ACS isolation checks PCI: Add ACS egress control vector accessor PCI: Account for ACS egress control in isolation checks PCI/P2PDMA: Derive peer-to-peer routing from ACS control bits PCI/P2PDMA: Honor ACS egress control vectors PCI/P2PDMA: Document ACS egress control handling PCI/P2PDMA: Extract pure ACS routing decision helpers PCI/P2PDMA: Add KUnit tests for ACS routing decisions PCI/P2PDMA: Add KUnit coverage for the ACS P2P routing walk PCI: Add KUnit coverage for ACS isolation checks PCI/P2PDMA: Log detailed ACS routing diagnostics Documentation/admin-guide/kernel-parameters.txt | 9 +- Documentation/driver-api/pci/p2pdma.rst | 24 + drivers/iommu/iommu.c | 8 +- drivers/pci/Kconfig | 15 + drivers/pci/Makefile | 1 + drivers/pci/ats.c | 11 +- drivers/pci/p2pdma.c | 496 +++++++++++-- drivers/pci/pci.c | 115 ++- drivers/pci/pci.h | 69 +- drivers/pci/pci_acs_test.c | 920 ++++++++++++++++++++++++ drivers/pci/quirks.c | 62 +- include/linux/pci-p2pdma.h | 8 +- include/linux/pci.h | 30 +- 13 files changed, 1661 insertions(+), 107 deletions(-) --- base-commit: 43598807f71ac1c9164f26004acf2496d4038daf change-id: 20260821-fix-p2p-acs-v4-0-e72455e3a261 Best regards, -- Leon Romanovsky <[email protected]>