diff options
author | Linus Torvalds <torvalds@linux-foundation.org> | 2023-04-09 09:45:46 -0700 |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2023-04-09 09:45:46 -0700 |
commit | c08cfd6716a170c549c1140f1d4a0e749c888a79 (patch) | |
tree | d2e78d27dc00081775cf793071f4a3fedff7135e /drivers/pci | |
parent | cdc9718d5e590d6905361800b938b93f2b66818e (diff) | |
parent | ca712e47054678c5ce93a0e0f686353ad5561195 (diff) |
Merge tag 'cxl-fixes-6.3-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/cxl/cxl
Pull compute express link (cxl) fixes from Dan Williams:
"Several fixes for driver startup regressions that landed during the
merge window as well as some older bugs.
The regressions were due to a lack of testing with what the CXL
specification calls Restricted CXL Host (RCH) topologies compared to
the testing with Virtual Host (VH) CXL topologies. A VH topology is
typical PCIe while RCH topologies map CXL endpoints as Root Complex
Integrated endpoints. The impact is some driver crashes on startup.
This merge window also added compatibility for range registers (the
mechanism that CXL 1.1 defined for mapping memory) to treat them like
HDM decoders (the mechanism that CXL 2.0 defined for mapping
Host-managed Device Memory). That work collided with the new region
enumeration code that was tested with CXL 2.0 setups, and fails with
crashes at startup.
Lastly, the DOE (Data Object Exchange) implementation for retrieving
an ACPI-like data table from CXL devices is being reworked for v6.4.
Several fixes fell out of that work that are suitable for v6.3.
All of this has been in linux-next for a while, and all reported
issues [1] have been addressed.
Summary:
- Fix several issues with region enumeration in RCH topologies that
can trigger crashes on driver startup or shutdown.
- Fix CXL DVSEC range register compatibility versus region
enumeration that leads to startup crashes
- Fix CDAT endiannes handling
- Fix multiple buffer handling boundary conditions
- Fix Data Object Exchange (DOE) workqueue usage vs
CONFIG_DEBUG_OBJECTS warn splats"
Link: http://lore.kernel.org/r/20230405075704.33de8121@canb.auug.org.au [1]
* tag 'cxl-fixes-6.3-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/cxl/cxl:
cxl/hdm: Extend DVSEC range register emulation for region enumeration
cxl/hdm: Limit emulation to the number of range registers
cxl/region: Move coherence tracking into cxl_region_attach()
cxl/region: Fix region setup/teardown for RCDs
cxl/port: Fix find_cxl_root() for RCDs and simplify it
cxl/hdm: Skip emulation when driver manages mem_enable
cxl/hdm: Fix double allocation of @cxlhdm
PCI/DOE: Fix memory leak with CONFIG_DEBUG_OBJECTS=y
PCI/DOE: Silence WARN splat with CONFIG_DEBUG_OBJECTS=y
cxl/pci: Handle excessive CDAT length
cxl/pci: Handle truncated CDAT entries
cxl/pci: Handle truncated CDAT header
cxl/pci: Fix CDAT retrieval on big endian
Diffstat (limited to 'drivers/pci')
-rw-r--r-- | drivers/pci/doe.c | 30 |
1 files changed, 18 insertions, 12 deletions
diff --git a/drivers/pci/doe.c b/drivers/pci/doe.c index 66d9ab288646..e5e9b287b976 100644 --- a/drivers/pci/doe.c +++ b/drivers/pci/doe.c @@ -128,7 +128,7 @@ static int pci_doe_send_req(struct pci_doe_mb *doe_mb, return -EIO; /* Length is 2 DW of header + length of payload in DW */ - length = 2 + task->request_pl_sz / sizeof(u32); + length = 2 + task->request_pl_sz / sizeof(__le32); if (length > PCI_DOE_MAX_LENGTH) return -EIO; if (length == PCI_DOE_MAX_LENGTH) @@ -141,9 +141,9 @@ static int pci_doe_send_req(struct pci_doe_mb *doe_mb, pci_write_config_dword(pdev, offset + PCI_DOE_WRITE, FIELD_PREP(PCI_DOE_DATA_OBJECT_HEADER_2_LENGTH, length)); - for (i = 0; i < task->request_pl_sz / sizeof(u32); i++) + for (i = 0; i < task->request_pl_sz / sizeof(__le32); i++) pci_write_config_dword(pdev, offset + PCI_DOE_WRITE, - task->request_pl[i]); + le32_to_cpu(task->request_pl[i])); pci_doe_write_ctrl(doe_mb, PCI_DOE_CTRL_GO); @@ -195,11 +195,11 @@ static int pci_doe_recv_resp(struct pci_doe_mb *doe_mb, struct pci_doe_task *tas /* First 2 dwords have already been read */ length -= 2; - payload_length = min(length, task->response_pl_sz / sizeof(u32)); + payload_length = min(length, task->response_pl_sz / sizeof(__le32)); /* Read the rest of the response payload */ for (i = 0; i < payload_length; i++) { - pci_read_config_dword(pdev, offset + PCI_DOE_READ, - &task->response_pl[i]); + pci_read_config_dword(pdev, offset + PCI_DOE_READ, &val); + task->response_pl[i] = cpu_to_le32(val); /* Prior to the last ack, ensure Data Object Ready */ if (i == (payload_length - 1) && !pci_doe_data_obj_ready(doe_mb)) return -EIO; @@ -217,13 +217,14 @@ static int pci_doe_recv_resp(struct pci_doe_mb *doe_mb, struct pci_doe_task *tas if (FIELD_GET(PCI_DOE_STATUS_ERROR, val)) return -EIO; - return min(length, task->response_pl_sz / sizeof(u32)) * sizeof(u32); + return min(length, task->response_pl_sz / sizeof(__le32)) * sizeof(__le32); } static void signal_task_complete(struct pci_doe_task *task, int rv) { task->rv = rv; task->complete(task); + destroy_work_on_stack(&task->work); } static void signal_task_abort(struct pci_doe_task *task, int rv) @@ -317,14 +318,16 @@ static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid, { u32 request_pl = FIELD_PREP(PCI_DOE_DATA_OBJECT_DISC_REQ_3_INDEX, *index); + __le32 request_pl_le = cpu_to_le32(request_pl); + __le32 response_pl_le; u32 response_pl; DECLARE_COMPLETION_ONSTACK(c); struct pci_doe_task task = { .prot.vid = PCI_VENDOR_ID_PCI_SIG, .prot.type = PCI_DOE_PROTOCOL_DISCOVERY, - .request_pl = &request_pl, + .request_pl = &request_pl_le, .request_pl_sz = sizeof(request_pl), - .response_pl = &response_pl, + .response_pl = &response_pl_le, .response_pl_sz = sizeof(response_pl), .complete = pci_doe_task_complete, .private = &c, @@ -340,6 +343,7 @@ static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid, if (task.rv != sizeof(response_pl)) return -EIO; + response_pl = le32_to_cpu(response_pl_le); *vid = FIELD_GET(PCI_DOE_DATA_OBJECT_DISC_RSP_3_VID, response_pl); *protocol = FIELD_GET(PCI_DOE_DATA_OBJECT_DISC_RSP_3_PROTOCOL, response_pl); @@ -520,6 +524,8 @@ EXPORT_SYMBOL_GPL(pci_doe_supports_prot); * task->complete will be called when the state machine is done processing this * task. * + * @task must be allocated on the stack. + * * Excess data will be discarded. * * RETURNS: 0 when task has been successfully queued, -ERRNO on error @@ -533,15 +539,15 @@ int pci_doe_submit_task(struct pci_doe_mb *doe_mb, struct pci_doe_task *task) * DOE requests must be a whole number of DW and the response needs to * be big enough for at least 1 DW */ - if (task->request_pl_sz % sizeof(u32) || - task->response_pl_sz < sizeof(u32)) + if (task->request_pl_sz % sizeof(__le32) || + task->response_pl_sz < sizeof(__le32)) return -EINVAL; if (test_bit(PCI_DOE_FLAG_DEAD, &doe_mb->flags)) return -EIO; task->doe_mb = doe_mb; - INIT_WORK(&task->work, doe_statemachine_work); + INIT_WORK_ONSTACK(&task->work, doe_statemachine_work); queue_work(doe_mb->work_queue, &task->work); return 0; } |