cxl/pci: Add support for RCH RAS error handling
Subsystems (3)
v8 cover cxl/pci: Add support for RCH RAS error handling
This patchset enables CXL RCH error handling. This is necessary because RCH downstream port protocol error handling is implemented uniquely and not currently supported. These patches address the following:
* Discovery and mapping of RCH downstream port AER registers.
* AER portdrv changes to support CXL RCH protocol errors.
* Interrupt setup specific to RCH mode: enabling RCEC internal errors and disabling root port interrupts.
Changes in V8:
- Rebased onto: commit
0c0df63177e3 ("Merge branch 'for-6.5/cxl-rch-eh' into for-6.5/cxl") - cxl/port: Pre-initialize component register mappings
- Added patch to pre-initialize component register mappings.
- cxl/pci: Remove Component Register base address from
- Separated removal of Component Register base address in struct cxl_dev_state to not break functionality.
- cxl/hdm: Use stored Component Register mappings to map HDM decoder capability
- Implemented a less strict check in devm_cxl_setup_hdm(), be tolerant if HDM decoder registers are not implemented.
- cxl/pci: Map RCH downstream AER registers for logging protocol errors
- Fixed uninitialized access of map->dev in cxl_dport_map_regs().
- PCI/AER: Forward RCH downstream port-detected errors to the CXL.mem dev handler
- Fix typo in patch description.
- cxl/pci: Update CXL error logging to use RAS register address
- Fix typo in patch description.
Changes in V7:
- cxl: Updates for CXL Test to work with RCH
- Removed Robert's DSO.
- cxl/core/regs: Add @dev to cxl_register_map
- Corrected typo in patch description.
- PCI/AER: Unmask RCEC internal errors to enable RCH downstream port error handling.
- Changed int variable to bool.
- cxl/pci: Map RCH downstream AER registers for logging protocol errors
- Corrected patch description.
- cxl/pci: Add RCH downstream port AER register discovery
- Reverted empty line removal.
- cxl/port: Store the port's Component Register mappings in struct cxl_port
- Update to use structure initialization in cxl_setup_comp_regs().
- Remove first patch (already in the tree) and added patch 27/27.
- Was a one-off error caused when merging branches during internal review.
Changes in V6:
- Added patch for cxl test fixes: 'cxl: Update CXl Test to Work with RCH'. Patch from Dan.
- Simplified: 'cxl/rch: Prepare for caching the MMIO mapped PCIe AER capability'. Patch from Dan.
- Added patch: 'cxl: Rename 'uport' to 'uport_dev''
- Updated patch: 'cxl: Rename member @dport of struct cxl_dport to
@dport_dev'
- Updated *map assignment to use structure init in 'cxl/core/regs: Add
@dev to cxl_register_map'. Also fixed whitespace.
- Removed extra whitespace in 'cxl/core/regs: Add @dev to
cxl_register_map'
- Updated patch subject: 'cxl/acpi: Move add_host_bridge_uport() after
cxl_get_chbs()' - Changes to work with CXL test. 'cxl/acpi: Directly bind the CEDT detected CHBCR to the Host Bridge's port'
- 'cxl/pci: Early setup RCH dport component registers from RCRB'
- Removed parameter from cxl_rcrb_get_comp_regs().
- Changed return value to EPROBE_DEFER for retry during ACPI initialization.
- Changed map to us struct initialization.
- Remove ENODEV check in 'cxl/port: Store the downstream port's Component Register mappings in struct cxl_dport'
- 'cxl/port: Remove Component Register base address from struct cxl_dport'
- Moved earlier with same removal for cxl_port.
- cxl/pci: Add RCH downstream port AER register discovery
- Flattened {request,release}_mem_region() and ioremap() into cxl_rcrb_to_aer().
- Add check if OS is assigned AER handling before discovering AER.
- Added CXL namespace import to cxl_core (drivers/cxl/core/port.c). Needed for using pci_print_aer(). In 'PCI/AER: Refactor cper_print_aer() for use by CXL driver module'.
- cxl/pci: Map RCH downstream AER registers for logging protocol errors
- Changed dport device used in devm_cxl_iomap_block() call to be port->dev.
- Removed ENODEV check.
- cxl/pci: Disable root port interrupts in RCH mode
- Removed unnecessary 'rch' check.
- Moved cxl_disable_rch_root_ints() into core/pci.c.
- Added OSC AER assignment check before accessing AER registers.
- cxl/pci: Update CXL error logging to use RAS register address
- Renamed function handlers.
- cxl/pci: Add RCH downstream port error logging
- Moved RCD check to caller.
- Added put_dev() after call to cxl_pci_find_port().
Changes in V5:
- Split 'cxl/rch: Prepare for logging RCH downstream port protocol errors' patch into 2 patches.
- Added:
cxl/core/regs: Rename phys_addr in cxl_map_component_regs() cxl/mem: Prepare for early RCH dport component register setup
- Correct comments CXL3.0 to CXL 3.0.
- changed cxl_port_get_comp_map() to static.
Changes in V4:
- Made port RAS register discovery common and called from
__devm_cxl_add_dport().
- Changed RCH AER register discovery to be called from
__devm_cxl_add_dport().
- Changed RAS and RCH AER register mapping to be called from
__devm_cxl_add_dport(). - Changed component register mapping to support all CXL component mapping, cxl_map_component_regs().
- Added cxl_regs to 'struct cxl_dport' for providing RCH downstream port mapped registers USED in error handler.
- PCI/AER:
- Improved description of PCIEAER_CXL option in Kconfig.
- Renamed function to pci_aer_unmask_internal_errors(), added pcie_aer_is_native() check.
- Improved comments and added spec refs.
- Renamed functions to cxl_rch_handle_error*().
- Modified cxl_rch_handle_error_iter() to only call the handler callbacks, this also simplifies refcounting of the pdev.
- Refactored handle_error_source(), created pci_aer_handle_error().
- Changed printk messages to pci_*() variants.
- Added check for pcie_aer_is_native() to the RCEC.
- Introduced function cxl_rch_enable_rcec().
- Updated patch description ("PCI/AER: Forward RCH downstream port-detected errors to the CXL.mem dev handler").
Changes in V3:
- Correct base commit in cover sheet.
- Change hardcoded return 0 to NULL in regs.c.
- Remove calls to pci_disable_pcie_error_reporting(pdev) and
pci_enable_pcie_error_reporting(pdev) in mem.c;
- Move RCEC interrupt unmask to PCIe port AER driver's probe.
- Fixes missing PCIEAER and PCIEPORTBUS config option error.
- Rename cxl_rcrb_setup() to cxl_setup_rcrb() in mem.c.
- Update cper_mem_err_unpack() patch subject and description.
Changes in V2:
- Refactor RCH initialization into cxl_mem driver.
- Includes RCH RAS and AER register discovery and mapping.
- Add RCEC protocol error interrupt forwarding to CXL endpoint handler.
- Change AER and RAS logging to use existing trace routines.
- Enable RCEC AER internal errors.
Robert Richter (8):
cxl/port: Pre-initialize component register mappings cxl/pci: Store the endpoint's Component Register mappings in struct
cxl_dev_state
cxl/hdm: Use stored Component Register mappings to map HDM decoder
capability
cxl/pci: Remove Component Register base address from struct
cxl_dev_state
cxl/port: Remove Component Register base address from struct cxl_port PCI/AER: Forward RCH downstream port-detected errors to the CXL.mem
dev handler
PCI/AER: Unmask RCEC internal errors to enable RCH downstream port
error handling
cxl/core/regs: Rename phys_addr in cxl_map_component_regs()
Terry Bowman (6):
cxl/pci: Add RCH downstream port AER register discovery PCI/AER: Refactor cper_print_aer() for use by CXL driver module cxl/pci: Update CXL error logging to use RAS register address cxl/pci: Map RCH downstream AER registers for logging protocol errors cxl/pci: Add RCH downstream port error logging
cxl/pci: Disable root port interrupts in RCH mode
base-commit: 0c0df63177e37ae826d803280eb2c5b6b6a7a9a4
drivers/cxl/core/core.h | 7 ++ drivers/cxl/core/hdm.c | 64 +++++++------ drivers/cxl/core/pci.c | 174 ++++++++++++++++++++++++++++++++--- drivers/cxl/core/port.c | 58 ++++++++++-- drivers/cxl/core/regs.c | 42 ++++++++- drivers/cxl/cxl.h | 14 ++- drivers/cxl/cxlmem.h | 4 +- drivers/cxl/mem.c | 4 +- drivers/cxl/pci.c | 10 +- drivers/pci/pcie/Kconfig | 12 +++ drivers/pci/pcie/aer.c | 162 ++++++++++++++++++++++++++++++-- include/linux/aer.h | 2 +- tools/testing/cxl/test/mem.c | 1 - 13 files changed, 481 insertions(+), 73 deletions(-)
Signature · 1 line
2.34.1
PATCH v8 1/14 cxl/port: Pre-initialize component register mappings
From: Robert Richter <[email protected]>
The component registers of a component may not exist or are not needed. The setup may fail for that reason. In some cases the initialization should continue anyway. Thus, always initialize struct cxl_register_map with valid values. In case of errors, zero it, set a value for @dev and make @resource a the valid value using CXL_RESOURCE_NONE.
--- drivers/cxl/core/port.c | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-)
drivers/cxl/core/port.c +6 -5
@@ -693,16 +693,17 @@ static struct cxl_port *cxl_port_alloc(struct device *uport_dev,static int cxl_setup_comp_regs (struct device *dev,struct cxl_register_map *map,resource_size_t component_reg_phys) { -if (component_reg_phys ==CXL_RESOURCE_NONE ) -return 0 ; - *map = (struct cxl_register_map ) { .dev = dev, - .reg_type =CXL_REGLOC_RBI_COMPONENT , .resource = component_reg_phys, - .max_size =CXL_COMPONENT_REG_BLOCK_SIZE , }; +if (component_reg_phys ==CXL_RESOURCE_NONE ) +return 0 ; + + map->reg_type =CXL_REGLOC_RBI_COMPONENT ; + map->max_size =CXL_COMPONENT_REG_BLOCK_SIZE ; +return cxl_setup_regs (map); }
Signature · 1 line
2.34.1
Re: [PATCH v8 01/14] cxl/port: Pre-initialize component register mappings
Quoted 11 lines · On Fri, 30 Jun 2023 18:16:22 -0500 Terry Bowman <[email protected]> wrote:
From: Robert Richter <[email protected]>
The component registers of a component may not exist or are not needed. The setup may fail for that reason. In some cases the initialization should continue anyway. Thus, always initialize struct
cxl_register_mapwith valid values. In case of errors, zero it, set a value for @dev and make @resource a the valid value usingCXL_RESOURCE_NONE.Signed-off-by Robert Richter <[email protected]>Signed-off-by Terry Bowman <[email protected]>
Seems reasonable to me.
Quoted 31 lines
--- drivers/cxl/core/port.c | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-)drivers/cxl/core/port.c +6
-5@@ -693,16 +693,17 @@ static struct cxl_port *cxl_port_alloc(struct device *uport_dev,static int cxl_setup_comp_regs (struct device *dev,struct cxl_register_map *map,resource_size_t component_reg_phys) { -if (component_reg_phys ==CXL_RESOURCE_NONE ) -return 0 ; - *map = (struct cxl_register_map ) { .dev = dev, - .reg_type =CXL_REGLOC_RBI_COMPONENT , .resource = component_reg_phys, - .max_size =CXL_COMPONENT_REG_BLOCK_SIZE , }; +if (component_reg_phys ==CXL_RESOURCE_NONE ) +return 0 ; + + map->reg_type =CXL_REGLOC_RBI_COMPONENT ; + map->max_size =CXL_COMPONENT_REG_BLOCK_SIZE ; +return cxl_setup_regs (map); }
PATCH v8 2/14 cxl/pci: Store the endpoint's Component Register mappings in struct cxl_dev_state
In reply to Terry BowmanFrom: Robert Richter <[email protected]>
Same as for ports and dports, also store the endpoint's Component Register mappings, use struct cxl_dev_state for that.
Keep the Component Register base address @component_reg_phys a bit to not break functionality. It will be removed after the transition in a later patch.
--- drivers/cxl/cxlmem.h | 2 ++ drivers/cxl/pci.c | 9 +++++---- 2 files changed, 7 insertions(+), 4 deletions(-)
drivers/cxl/cxlmem.h +2 -0
@@ -382,6 +382,7 @@ enum cxl_devtype {* * @dev: The device associated with this CXL state * @cxlmd: The device representing the CXL.mem capabilities of @dev +* @comp_map: component register capability mappings * @regs: Parsed register blocks * @cxl_dvsec: Offset to the PCIe device DVSEC * @rcd: operating in RCD mode (CXL 3.0 9.11.8 CXL Devices Attached to an RCH)
@@ -396,6 +397,7 @@ enum cxl_devtype {struct cxl_dev_state {struct device *dev;struct cxl_memdev *cxlmd; +struct cxl_register_map comp_map;struct cxl_regs regs;int cxl_dvsec;bool rcd;
drivers/cxl/pci.c +5 -4
@@ -836,15 +836,16 @@ static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)* still be useful for management functions so don't return an error. */ cxlds->component_reg_phys =CXL_RESOURCE_NONE ; - rc =cxl_pci_setup_regs (pdev,CXL_REGLOC_RBI_COMPONENT , &map); + rc =cxl_pci_setup_regs (pdev,CXL_REGLOC_RBI_COMPONENT , + &cxlds->comp_map);if (rc)dev_warn (&pdev->dev,"No component registers (%d)\n" , rc); -else if (!map.component_map.ras.valid) +else if (!cxlds->comp_map.component_map.ras.valid)dev_dbg (&pdev->dev,"RAS registers not found\n" ); - cxlds->component_reg_phys = map.resource; + cxlds->component_reg_phys = cxlds->comp_map.resource; - rc =cxl_map_component_regs (&map, &cxlds->regs.component, + rc =cxl_map_component_regs (&cxlds->comp_map, &cxlds->regs.component,BIT (CXL_CM_CAP_CAP_ID_RAS ));if (rc)dev_dbg (&pdev->dev,"Failed to map RAS capability.\n" );
Signature · 1 line
2.34.1
PATCH v8 3/14 cxl/hdm: Use stored Component Register mappings to map HDM decoder capability
In reply to Terry BowmanFrom: Robert Richter <[email protected]>
Now, that the Component Register mappings are stored, use them to enable and map the HDM decoder capabilities. The Component Registers do not need to be probed again for this, remove probing code.
The HDM capability applies to Endpoints, USPs and VH Host Bridges. The Endpoint's component register mappings are located in the cxlds and else in the port's structure. Provide a helper function cxl_port_get_comp_map() to locate the mappings depending on the component's type.
--- drivers/cxl/core/hdm.c | 64 +++++++++++++++++++++++------------------- 1 file changed, 35 insertions(+), 29 deletions(-)
drivers/cxl/core/hdm.c +35 -29
@@ -81,26 +81,6 @@ static void parse_hdm_decoder_caps(struct cxl_hdm *cxlhdm) cxlhdm->interleave_mask |=GENMASK (14 ,12 ); } -static int map_hdm_decoder_regs (struct cxl_port *port,void __iomem *crb, -struct cxl_component_regs *regs) -{ -struct cxl_register_map map = { - .dev = &port->dev, - .resource = port->component_reg_phys, - .base = crb, - .max_size =CXL_COMPONENT_REG_BLOCK_SIZE , - }; - -cxl_probe_component_regs (&port->dev, crb, &map.component_map); -if (!map.component_map.hdm_decoder.valid) { -dev_dbg (&port->dev,"HDM decoder registers not implemented\n" ); -/* unique error code to indicate no HDM decoder capability */ -return -ENODEV ; - } - -return cxl_map_component_regs (&map, regs,BIT (CXL_CM_CAP_CAP_ID_HDM )); -} -static bool should_emulate_decoders (struct cxl_endpoint_dvsec_info *info) {struct cxl_hdm *cxlhdm;
@@ -145,6 +125,22 @@ static bool should_emulate_decoders(struct cxl_endpoint_dvsec_info *info)return true; } +static struct cxl_register_map *cxl_port_get_comp_map (struct cxl_port *port) +{ +/* +* HDM capability applies to Endpoints, USPs and VH Host +* Bridges. The Endpoint's component register mappings are +* located in the cxlds. +*/ +if (is_cxl_endpoint (port)) { +struct cxl_memdev *memdev =to_cxl_memdev (port->uport_dev); + +return &memdev->cxlds->comp_map; + } + +return &port->comp_map; +} +/** * devm_cxl_setup_hdm - map HDM decoder component registers * @port: cxl_port to map
@@ -155,7 +151,7 @@ struct cxl_hdm *devm_cxl_setup_hdm(struct cxl_port *port, {struct device *dev = &port->dev;struct cxl_hdm *cxlhdm; -void __iomem *crb; +struct cxl_register_map *comp_map;int rc; cxlhdm =devm_kzalloc (dev,sizeof (*cxlhdm),GFP_KERNEL );
@@ -164,19 +160,29 @@ struct cxl_hdm *devm_cxl_setup_hdm(struct cxl_port *port, cxlhdm->port = port;dev_set_drvdata (dev, cxlhdm); - crb =ioremap (port->component_reg_phys,CXL_COMPONENT_REG_BLOCK_SIZE ); -if (!crb && info && info->mem_enabled) { - cxlhdm->decoder_count = info->ranges; -return cxlhdm; - }else if (!crb) { + comp_map =cxl_port_get_comp_map (port); + +if (comp_map->resource ==CXL_RESOURCE_NONE ) { +if (info && info->mem_enabled) { + cxlhdm->decoder_count = info->ranges; +return cxlhdm; + }dev_err (dev,"No component registers mapped\n" );return ERR_PTR (-ENXIO ); } - rc =map_hdm_decoder_regs (port, crb, &cxlhdm->regs); -iounmap (crb); -if (rc) +if (!comp_map->component_map.hdm_decoder.valid) { +dev_dbg (&port->dev,"HDM decoder registers not implemented\n" ); +/* unique error code to indicate no HDM decoder capability */ +return ERR_PTR (-ENODEV ); + } + + rc =cxl_map_component_regs (comp_map, &cxlhdm->regs, +BIT (CXL_CM_CAP_CAP_ID_HDM )); +if (rc) { +dev_dbg (dev,"Failed to map HDM capability.\n" );return ERR_PTR (rc); + }parse_hdm_decoder_caps (cxlhdm);if (cxlhdm->decoder_count ==0 ) {
Signature · 1 line
2.34.1
RE: [PATCH v8 03/14] cxl/hdm: Use stored Component Register mappings to map HDM decoder capability
Quoted 11 lines · Terry Bowman
From: Robert Richter <[email protected]>
Now, that the Component Register mappings are stored, use them to enable and map the HDM decoder capabilities. The Component Registers do not need to be probed again for this, remove probing code.
The HDM capability applies to Endpoints, USPs and VH Host Bridges. The Endpoint's component register mappings are located in the cxlds and else in the port's structure. Provide a helper function
cxl_port_get_comp_map()to locate the mappings depending on the component's type.
This causes a regression that cxl_test tripped over. It's likely something you could reproduce without cxl_test just be reloading the cxl_port driver. Root cause below, but here is what I see on a test run:
# meson test -C build --suite cxl ninja: no work to do. ninja: Entering directory `/root/git/ndctl/build' [113/113] Linking target ndctl/ndctl 1/6 ndctl:cxl / cxl-topology.sh OK 12.78s 2/6 ndctl:cxl / cxl-region-sysfs.sh OK 7.72s 3/6 ndctl:cxl / cxl-labels.sh FAIL 1.53s (exit status 129 or signal 1 SIGHUP)
Quoted 1 line
Quoted 1 line
Quoted 1 line
LD_LIBRARY_PATH=/root/git/ndctl/build/cxl/lib:/root/git/ndctl/build/ndctl/lib:/root/git/ndctl/build/daxctl/lib NDCTL=/root/git/ndctl/build/ndctl/ndctl DAXCTL=/root/git/ndctl/build/daxctl/daxctl MALLOC_PERTURB_=67 TEST_PATH=/root/git/ndctl/build/test DATA_PATH=/root/git/ndctl/test
/bin/bash/root/git/ndctl/test/cxl-labels.sh
4/6 ndctl:cxl / cxl-create-region.sh OK 17.05s 5/6 ndctl:cxl / cxl-xor-region.sh OK 5.18s 6/6 ndctl:cxl / cxl-security.sh OK 4.68s
Ok: 5 Expected Fail: 0 Fail: 1 Unexpected Pass: 0 Skipped: 0 Timeout: 0
Full log written to /root/git/ndctl/build/meson-logs/testlog.txt
It's not that cxl-labels.sh causes the error, it is loading and unloading the port driver trips over the problem below:
Quoted 45 lines
Signed-off-by Robert Richter <[email protected]>Signed-off-by Terry Bowman <[email protected]>Reviewed-by Jonathan Cameron <[email protected]>Reviewed-by Dave Jiang <[email protected]>--- drivers/cxl/core/hdm.c | 64 +++++++++++++++++++++++------------------- 1 file changed, 35 insertions(+), 29 deletions(-)drivers/cxl/core/hdm.c +1
-20@@ -81,26 +81,6 @@ static void parse_hdm_decoder_caps(struct cxl_hdm *cxlhdm) cxlhdm->interleave_mask |=GENMASK (14 ,12 ); } -static int map_hdm_decoder_regs (struct cxl_port *port,void __iomem *crb, -struct cxl_component_regs *regs) -{ -struct cxl_register_map map = { - .dev = &port->dev, - .resource = port->component_reg_phys, - .base = crb, - .max_size =CXL_COMPONENT_REG_BLOCK_SIZE , - }; - -cxl_probe_component_regs (&port->dev, crb, &map.component_map); -if (!map.component_map.hdm_decoder.valid) { -dev_dbg (&port->dev,"HDM decoder registers not implemented\n" ); -/* unique error code to indicate no HDM decoder capability */ -return -ENODEV ; - } - -return cxl_map_component_regs (&map, regs,BIT (CXL_CM_CAP_CAP_ID_HDM )); -} -static bool should_emulate_decoders (struct cxl_endpoint_dvsec_info *info) {struct cxl_hdm *cxlhdm;@@ -145,6 +125,22 @@ static bool should_emulate_decoders(struct cxl_endpoint_dvsec_info *info)return true; } +static struct cxl_register_map *cxl_port_get_comp_map (struct cxl_port *port)
drivers/cxl/core/hdm.c static bool should_emulate_decoders(struct cxl_endpoint_dvsec_info *info) · line 138
Quoted 10 lines
drivers/cxl/core/hdm.c +10
-0+{ +/* +* HDM capability applies to Endpoints, USPs and VH Host +* Bridges. The Endpoint's component register mappings are +* located in the cxlds. +*/ +if (is_cxl_endpoint (port)) { +struct cxl_memdev *memdev =to_cxl_memdev (port->uport_dev); + +return &memdev->cxlds->comp_map;
...but why? The issue here is that the @dev argument in that map is the memdev parent PCI device. However, in this context the @dev for devm operations wants to be &port->dev.
drivers/cxl/core/hdm.c static bool should_emulate_decoders(struct cxl_endpoint_dvsec_info *info) · line 141
Quoted 3 lines
drivers/cxl/core/hdm.c +3
-0+ } + +return &port->comp_map;
...so this is fine, and folding in the following resolves the test failure.
drivers/cxl/core/hdm.c +1 -18
@@ -125,22 +125,6 @@ static bool should_emulate_decoders(struct cxl_endpoint_dvsec_info *info)return true; } -static struct cxl_register_map *cxl_port_get_comp_map (struct cxl_port *port) -{ -/* -* HDM capability applies to Endpoints, USPs and VH Host -* Bridges. The Endpoint's component register mappings are -* located in the cxlds. -*/ -if (is_cxl_endpoint (port)) { -struct cxl_memdev *memdev =to_cxl_memdev (port->uport_dev); - -return &memdev->cxlds->comp_map; - } - -return &port->comp_map; -} -/** * devm_cxl_setup_hdm - map HDM decoder component registers * @port: cxl_port to map
@@ -160,8 +144,7 @@ struct cxl_hdm *devm_cxl_setup_hdm(struct cxl_port *port, cxlhdm->port = port;dev_set_drvdata (dev, cxlhdm); - comp_map =cxl_port_get_comp_map (port); - + comp_map = &port->comp_map;if (comp_map->resource ==CXL_RESOURCE_NONE ) {if (info && info->mem_enabled) { cxlhdm->decoder_count = info->ranges;
Am I missing why the cxlds->comp_map needs to be reused?
Note that I am out and may not be able to respond further until next week.
Re: [PATCH v8 03/14] cxl/hdm: Use stored Component Register mappings to map HDM decoder capability
Dan,
Quoted 17 lines · Dan Williams
Quoted 11 lines · Terry Bowman
From: Robert Richter <[email protected]>
Now, that the Component Register mappings are stored, use them to enable and map the HDM decoder capabilities. The Component Registers do not need to be probed again for this, remove probing code.
The HDM capability applies to Endpoints, USPs and VH Host Bridges. The Endpoint's component register mappings are located in the cxlds and else in the port's structure. Provide a helper function
cxl_port_get_comp_map()to locate the mappings depending on the component's type.This causes a regression that
cxl_testtripped over. It's likely something you could reproduce withoutcxl_testjust be reloading thecxl_portdriver. Root cause below, but here is what I see on a test run:
Quoted 8 lines
# meson test -C build --suite cxl ninja: no work to do. ninja: Entering directory `/root/git/ndctl/build' [113/113] Linking target ndctl/ndctl 1/6 ndctl:cxl / cxl-topology.sh OK 12.78s 2/6 ndctl:cxl / cxl-region-sysfs.sh OK 7.72s 3/6 ndctl:cxl / cxl-labels.sh FAIL 1.53s (exit status 129 or signal 1 SIGHUP)Quoted 1 line
Quoted 1 line
Quoted 1 line
LD_LIBRARY_PATH=/root/git/ndctl/build/cxl/lib:/root/git/ndctl/build/ndctl/lib:/root/git/ndctl/build/daxctl/lib NDCTL=/root/git/ndctl/build/ndctl/ndctl DAXCTL=/root/git/ndctl/build/daxctl/daxctl MALLOC_PERTURB_=67 TEST_PATH=/root/git/ndctl/build/test DATA_PATH=/root/git/ndctl/test
/bin/bash/root/git/ndctl/test/cxl-labels.sh
I have tried various combinations of revisions and setups (patch #3, #5 and #14, ndctl:pending/v77, cxl_test only with both, "ACPI0017" and cxl_pci_probe() disabled), but could not reproduce this. Tests always pass for me:
[51/51] Linking target ndctl/ndctl
1/6 ndctl:cxl / cxl-topology.sh OK 18.59s 2/6 ndctl:cxl / cxl-region-sysfs.sh OK 12.26s 3/6 ndctl:cxl / cxl-labels.sh OK 28.25s 4/6 ndctl:cxl / cxl-create-region.sh OK 19.56s 5/6 ndctl:cxl / cxl-xor-region.sh OK 10.57s 6/6 ndctl:cxl / cxl-security.sh OK 9.77s
Ok: 6 Expected Fail: 0 Fail: 0 Unexpected Pass: 0 Skipped: 0 Timeout: 0
Full log written to /root/ndctl/build/meson-logs/testlog.txt
Quoted 61 lines
4/6 ndctl:cxl / cxl-create-region.sh OK 17.05s 5/6 ndctl:cxl / cxl-xor-region.sh OK 5.18s 6/6 ndctl:cxl / cxl-security.sh OK 4.68sOk: 5 Expected Fail: 0 Fail: 1 Unexpected Pass: 0 Skipped: 0 Timeout: 0Full log written to
/root/git/ndctl/build/meson-logs/testlog.txtIt's not that
cxl-labels.shcauses the error, it is loading and unloading the port driver trips over the problem below:Quoted 44 lines
Signed-off-by Robert Richter <[email protected]>Signed-off-by Terry Bowman <[email protected]>Reviewed-by Jonathan Cameron <[email protected]>Reviewed-by Dave Jiang <[email protected]>--- drivers/cxl/core/hdm.c | 64 +++++++++++++++++++++++------------------- 1 file changed, 35 insertions(+), 29 deletions(-)drivers/cxl/core/hdm.c +0
-20@@ -81,26 +81,6 @@ static void parse_hdm_decoder_caps(struct cxl_hdm *cxlhdm) cxlhdm->interleave_mask |=GENMASK (14 ,12 ); } -static int map_hdm_decoder_regs (struct cxl_port *port,void __iomem *crb, -struct cxl_component_regs *regs) -{ -struct cxl_register_map map = { - .dev = &port->dev, - .resource = port->component_reg_phys, - .base = crb, - .max_size =CXL_COMPONENT_REG_BLOCK_SIZE , - }; - -cxl_probe_component_regs (&port->dev, crb, &map.component_map); -if (!map.component_map.hdm_decoder.valid) { -dev_dbg (&port->dev,"HDM decoder registers not implemented\n" ); -/* unique error code to indicate no HDM decoder capability */ -return -ENODEV ; - } - -return cxl_map_component_regs (&map, regs,BIT (CXL_CM_CAP_CAP_ID_HDM )); -} -static bool should_emulate_decoders (struct cxl_endpoint_dvsec_info *info) {struct cxl_hdm *cxlhdm;@@ -145,6 +125,22 @@ static bool should_emulate_decoders(struct cxl_endpoint_dvsec_info *info)return true; }
Quoted 15 lines
drivers/cxl/core/hdm.c static bool should_emulate_decoders(struct cxl_endpoint_dvsec_info *info) · line 138
Quoted 11 lines
drivers/cxl/core/hdm.c +11
-0+static struct cxl_register_map *cxl_port_get_comp_map (struct cxl_port *port) +{ +/* +* HDM capability applies to Endpoints, USPs and VH Host +* Bridges. The Endpoint's component register mappings are +* located in the cxlds. +*/ +if (is_cxl_endpoint (port)) { +struct cxl_memdev *memdev =to_cxl_memdev (port->uport_dev); + +return &memdev->cxlds->comp_map;...but why? The issue here is that the @dev argument in that map is the memdev parent PCI device. However, in this context the @dev for devm operations wants to be
&port->dev.
The cxl_pci driver stores the comp_map of the endpoint in the cxlds structure, struct cxl_port is not yet available at this point. See patch #2 of this series ("cxl/pci: Store the endpoint's Component Register mappings in struct cxl_dev_state").
Quoted 36 lines
drivers/cxl/core/hdm.c static bool should_emulate_decoders(struct cxl_endpoint_dvsec_info *info) · line 141
Quoted 4 lines
drivers/cxl/core/hdm.c +3
-0+ } + +return &port->comp_map;...so this is fine, and folding in the following resolves the test failure.
drivers/cxl/core/hdm.c +0
-16@@ -125,22 +125,6 @@ static bool should_emulate_decoders(struct cxl_endpoint_dvsec_info *info)return true; } -static struct cxl_register_map *cxl_port_get_comp_map (struct cxl_port *port) -{ -/* -* HDM capability applies to Endpoints, USPs and VH Host -* Bridges. The Endpoint's component register mappings are -* located in the cxlds. -*/ -if (is_cxl_endpoint (port)) { -struct cxl_memdev *memdev =to_cxl_memdev (port->uport_dev); - -return &memdev->cxlds->comp_map; - } - -return &port->comp_map; -} -/** * devm_cxl_setup_hdm - map HDM decoder component registers * @port: cxl_port to map
drivers/cxl/core/hdm.c struct cxl_hdm *devm_cxl_setup_hdm(struct cxl_port *port, · line 147
Quoted 6 lines
drivers/cxl/core/hdm.c +1
-2@@ -160,8 +144,7 @@ struct cxl_hdm *devm_cxl_setup_hdm(struct cxl_port *port, cxlhdm->port = port;dev_set_drvdata (dev, cxlhdm); - comp_map =cxl_port_get_comp_map (port); - + comp_map = &port->comp_map;
Can you check if the following works instead, I think the pre-initialization is missing in cxl_mock_mem_probe() for cxl_test:
drivers/cxl/core/mbox.c +2 -0
@@ -1333,6 +1333,8 @@ struct cxl_memdev_state *cxl_memdev_state_create(struct device *dev)mutex_init (&mds->mbox_mutex);mutex_init (&mds->event.log_lock); mds->cxlds.dev = dev; + mds->cxlds.comp_map.dev = dev; + mds->cxlds.comp_map.resource =CXL_RESOURCE_NONE ; mds->cxlds.type =CXL_DEVTYPE_CLASSMEM ;return mds;
Signature · 9 lines
2.30.2 The cxl_pci driver always has something valid or fails otherwise. If that works the change should be merged into patch #2. Thanks, -Robert
Quoted 9 lines
drivers/cxl/core/hdm.c +0
-0if (comp_map->resource ==CXL_RESOURCE_NONE ) {if (info && info->mem_enabled) { cxlhdm->decoder_count = info->ranges;Am I missing why the
cxlds->comp_mapneeds to be reused?Note that I am out and may not be able to respond further until next week.
Re: [PATCH v8 03/14] cxl/hdm: Use stored Component Register mappings to map HDM decoder capability
Quoted 27 lines · Robert Richter
Dan,
Quoted 24 lines · Dan Williams
Quoted 11 lines · Terry Bowman
From: Robert Richter <[email protected]>
Now, that the Component Register mappings are stored, use them to enable and map the HDM decoder capabilities. The Component Registers do not need to be probed again for this, remove probing code.
The HDM capability applies to Endpoints, USPs and VH Host Bridges. The Endpoint's component register mappings are located in the cxlds and else in the port's structure. Provide a helper function
cxl_port_get_comp_map()to locate the mappings depending on the component's type.This causes a regression that
cxl_testtripped over. It's likely something you could reproduce withoutcxl_testjust be reloading thecxl_portdriver. Root cause below, but here is what I see on a test run:# meson test -C build --suite cxl ninja: no work to do. ninja: Entering directory `/root/git/ndctl/build' [113/113] Linking target ndctl/ndctl 1/6 ndctl:cxl / cxl-topology.sh OK 12.78s 2/6 ndctl:cxl / cxl-region-sysfs.sh OK 7.72s 3/6 ndctl:cxl / cxl-labels.sh FAIL 1.53s (exit status 129 or signal 1 SIGHUP)
Quoted 6 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
LD_LIBRARY_PATH=/root/git/ndctl/build/cxl/lib:/root/git/ndctl/build/ndctl/lib:/root/git/ndctl/build/daxctl/lib NDCTL=/root/git/ndctl/build/ndctl/ndctl DAXCTL=/root/git/ndctl/build/daxctl/daxctl MALLOC_PERTURB_=67 TEST_PATH=/root/git/ndctl/build/test DATA_PATH=/root/git/ndctl/test
/bin/bash/root/git/ndctl/test/cxl-labels.shI have tried various combinations of revisions and setups (patch #3, #5 and #14, ndctl:pending/v77,
cxl_testonly with both, "ACPI0017" andcxl_pci_probe()disabled), but could not reproduce this. Tests always pass for me:
Apologies for the long delay in getting back to this. July had some major personal interrupts to attend.
Quoted 9 lines
[51/51] Linking target ndctl/ndctl
1/6 ndctl:cxl / cxl-topology.sh OK 18.59s 2/6 ndctl:cxl / cxl-region-sysfs.sh OK 12.26s 3/6 ndctl:cxl / cxl-labels.sh OK 28.25s 4/6 ndctl:cxl / cxl-create-region.sh OK 19.56s 5/6 ndctl:cxl / cxl-xor-region.sh OK 10.57s 6/6 ndctl:cxl / cxl-security.sh OK 9.77s
Quoted 8 lines
Ok: 6 Expected Fail: 0 Fail: 0 Unexpected Pass: 0 Skipped: 0 Timeout: 0Full log written to
/root/ndctl/build/meson-logs/testlog.txt
So it turns out it is not cxl_test that is failing instead it is the tests noticing a regression of the VH base case. I am running cxl-topology.sh in a QEMU environment with a local device defined, and that local device is hitting a probe regression.
When cxl_test goes to count the expected disabled devices it sees one more because the QEMU device is also disabled.
++ jq 'map(select(.state == "disabled")) | length'
+ count=5
+ (( count == 4 ))
+ err 170
++ basename /root/git/ndctl/test/cxl-topology.sh
+ echo test/cxl-topology.sh: failed at line 170
...i.e. only 4 devices are expected to be disabled, not the 5th one that was not on the cxl_test bus. I assume you are running without any other CXL devices?
I will look into making that a more formalized case so that failure is not QEMU configuration dependent, but please make sure that QEMU CXL configs do not regress.
Quoted 77 lines
Quoted 61 lines
4/6 ndctl:cxl / cxl-create-region.sh OK 17.05s 5/6 ndctl:cxl / cxl-xor-region.sh OK 5.18s 6/6 ndctl:cxl / cxl-security.sh OK 4.68sOk: 5 Expected Fail: 0 Fail: 1 Unexpected Pass: 0 Skipped: 0 Timeout: 0Full log written to
/root/git/ndctl/build/meson-logs/testlog.txtIt's not that
cxl-labels.shcauses the error, it is loading and unloading the port driver trips over the problem below:Quoted 44 lines
Signed-off-by Robert Richter <[email protected]>Signed-off-by Terry Bowman <[email protected]>Reviewed-by Jonathan Cameron <[email protected]>Reviewed-by Dave Jiang <[email protected]>--- drivers/cxl/core/hdm.c | 64 +++++++++++++++++++++++------------------- 1 file changed, 35 insertions(+), 29 deletions(-)drivers/cxl/core/hdm.c +0
-20@@ -81,26 +81,6 @@ static void parse_hdm_decoder_caps(struct cxl_hdm *cxlhdm) cxlhdm->interleave_mask |=GENMASK (14 ,12 ); } -static int map_hdm_decoder_regs (struct cxl_port *port,void __iomem *crb, -struct cxl_component_regs *regs) -{ -struct cxl_register_map map = { - .dev = &port->dev, - .resource = port->component_reg_phys, - .base = crb, - .max_size =CXL_COMPONENT_REG_BLOCK_SIZE , - }; - -cxl_probe_component_regs (&port->dev, crb, &map.component_map); -if (!map.component_map.hdm_decoder.valid) { -dev_dbg (&port->dev,"HDM decoder registers not implemented\n" ); -/* unique error code to indicate no HDM decoder capability */ -return -ENODEV ; - } - -return cxl_map_component_regs (&map, regs,BIT (CXL_CM_CAP_CAP_ID_HDM )); -} -static bool should_emulate_decoders (struct cxl_endpoint_dvsec_info *info) {struct cxl_hdm *cxlhdm;@@ -145,6 +125,22 @@ static bool should_emulate_decoders(struct cxl_endpoint_dvsec_info *info)return true; }Quoted 15 lines
drivers/cxl/core/hdm.c static bool should_emulate_decoders(struct cxl_endpoint_dvsec_info *info) · line 138
Quoted 11 lines
drivers/cxl/core/hdm.c +11
-0+static struct cxl_register_map *cxl_port_get_comp_map (struct cxl_port *port) +{ +/* +* HDM capability applies to Endpoints, USPs and VH Host +* Bridges. The Endpoint's component register mappings are +* located in the cxlds. +*/ +if (is_cxl_endpoint (port)) { +struct cxl_memdev *memdev =to_cxl_memdev (port->uport_dev); + +return &memdev->cxlds->comp_map;...but why? The issue here is that the @dev argument in that map is the memdev parent PCI device. However, in this context the @dev for devm operations wants to be
&port->dev.
Quoted 2 lines
The
cxl_pcidriver stores thecomp_mapof the endpoint in the cxlds structure, structcxl_portis not yet available at this point.
When you say "this point" you mean "that point" in cxl_pci, right? I initially took that to mean literally "this" point in the quoted code above.
Quoted 56 lines
See patch #2 of this series ("cxl/pci: Store the endpoint's Component Register mappings in struct
cxl_dev_state").Quoted 36 lines
drivers/cxl/core/hdm.c static bool should_emulate_decoders(struct cxl_endpoint_dvsec_info *info) · line 141
Quoted 4 lines
drivers/cxl/core/hdm.c +3
-0+ } + +return &port->comp_map;...so this is fine, and folding in the following resolves the test failure.
drivers/cxl/core/hdm.c +0
-16@@ -125,22 +125,6 @@ static bool should_emulate_decoders(struct cxl_endpoint_dvsec_info *info)return true; } -static struct cxl_register_map *cxl_port_get_comp_map (struct cxl_port *port) -{ -/* -* HDM capability applies to Endpoints, USPs and VH Host -* Bridges. The Endpoint's component register mappings are -* located in the cxlds. -*/ -if (is_cxl_endpoint (port)) { -struct cxl_memdev *memdev =to_cxl_memdev (port->uport_dev); - -return &memdev->cxlds->comp_map; - } - -return &port->comp_map; -} -/** * devm_cxl_setup_hdm - map HDM decoder component registers * @port: cxl_port to map drivers/cxl/core/hdm.c struct cxl_hdm *devm_cxl_setup_hdm(struct cxl_port *port, · line 147
Quoted 6 lines
drivers/cxl/core/hdm.c +1
-2@@ -160,8 +144,7 @@ struct cxl_hdm *devm_cxl_setup_hdm(struct cxl_port *port, cxlhdm->port = port;dev_set_drvdata (dev, cxlhdm); - comp_map =cxl_port_get_comp_map (port); - + comp_map = &port->comp_map;Can you check if the following works instead, I think the pre-initialization is missing in
cxl_mock_mem_probe()forcxl_test:drivers/cxl/core/mbox.c +0
-0
drivers/cxl/core/mbox.c struct cxl_memdev_state *cxl_memdev_state_create(struct device *dev) · line 1337
Quoted 5 lines
drivers/cxl/core/mbox.c +2
-0@@ -1333,6 +1333,8 @@ struct cxl_memdev_state *cxl_memdev_state_create(struct device *dev)mutex_init (&mds->mbox_mutex);mutex_init (&mds->event.log_lock); mds->cxlds.dev = dev; + mds->cxlds.comp_map.dev = dev; + mds->cxlds.comp_map.resource =CXL_RESOURCE_NONE ;
This has the same problem. @dev is specifying the lifetime of the mapping. The lifetime of the mapping needs to be relative to the driver using the registers. So if the cxl_port driver is mapping the component registers the only valid device in the comp_map is &port->dev.
I notice that cxl_port_get_comp_map() endpoint port as an argument. That endpoint port was instantiated with @cxlmd, but it seems not with @cxlmd->cxlds->comp_map information which is available. Lets just fix that. I.e. move devm_cxl_add_endpoint() into the core and make it initialize @endpoint->comp_map with @cxlds->comp_map while switching @dev in the @endpoint->comp_map to be @endpoint->dev, and not @cxlds->dev.
...but it turns out that is the second bug in this patch. As I went to try to reproduce this failure for a single port VH configuration I notice another bug, a regression of this fix:
7bba261e0aa6 cxl/port: Scan single-target ports for decoders
...because there is no requirement for single port switches / host-bridges to have HDM decoders per the specification, and the original patch is turning HDM decoders not present as a hard failure.
Quoted 7 lines
drivers/cxl/core/mbox.c +0
-0mds->cxlds.type =CXL_DEVTYPE_CLASSMEM ;return mds;Signature · 3 lines
2.30.2 The cxl_pci driver always has something valid or fails otherwise.
Understood, just copy that map at __devm_cxl_add_port() time. I am thinking devm_cxl_add_endpoint() moves into core/port.c and it calls __devm_cxl_add_port() directly with a new parameter to take a passed in @comp_map or @component_reg_phys for the switch port case.
Quoted 17 lines
If that works the change should be merged into patch #2.
Thanks,
-Robert
Quoted 9 lines
drivers/cxl/core/hdm.c +0
-0if (comp_map->resource ==CXL_RESOURCE_NONE ) {if (info && info->mem_enabled) { cxlhdm->decoder_count = info->ranges;Am I missing why the
cxlds->comp_mapneeds to be reused?Note that I am out and may not be able to respond further until next week.
Re: [PATCH v8 03/14] cxl/hdm: Use stored Component Register mappings to map HDM decoder capability
Quoted 35 lines · Dan Williams
Quoted 10 lines · Robert Richter
On 03.07.23 17:39:38, Dan Williams wrote:
[51/51] Linking target ndctl/ndctl
1/6 ndctl:cxl / cxl-topology.sh OK 18.59s 2/6 ndctl:cxl / cxl-region-sysfs.sh OK 12.26s 3/6 ndctl:cxl / cxl-labels.sh OK 28.25s 4/6 ndctl:cxl / cxl-create-region.sh OK 19.56s 5/6 ndctl:cxl / cxl-xor-region.sh OK 10.57s 6/6 ndctl:cxl / cxl-security.sh OK 9.77sQuoted 8 lines
Ok: 6 Expected Fail: 0 Fail: 0 Unexpected Pass: 0 Skipped: 0 Timeout: 0Full log written to
/root/ndctl/build/meson-logs/testlog.txtSo it turns out it is not
cxl_testthat is failing instead it is the tests noticing a regression of the VH base case. I am runningcxl-topology.shin a QEMU environment with a local device defined, and that local device is hitting a probe regression.When
cxl_testgoes to count the expected disabled devices it sees one more because the QEMU device is also disabled.++ jq '
map(select(.state == "disabled"))| length'
+ count=5
+ (( count == 4 ))
+ err 170
++ basename/root/git/ndctl/test/cxl-topology.sh
+ echotest/cxl-topology.sh: failed at line 170
Quoted 7 lines
...i.e. only 4 devices are expected to be disabled, not the 5th one that was not on the
cxl_testbus. I assume you are running without any other CXL devices?I will look into making that a more formalized case so that failure is not QEMU configuration dependent, but please make sure that QEMU CXL configs do not regress.
I finally could reproduce and root cause the issue.
I have isolated this to the operation on the qemu mem0 device, without cxl_test involved:
root@qemu:~# cxl list
[
{
"memdev":"mem0",
"pmem_size":536870912,
"serial":0,
"host":"0000:35:00.0"
}
]
root@qemu:~# echo mem0 >/sys/bus/cxl/drivers/cxl_mem/unbind
[ 65.957423] nvdimm nmem0: trace
[ 66.001986] nd_bus ndbus0: nvdimm.remove(nmem0)
[ 66.041053] cxl_mem mem0: disconnect mem0 from port1
root@qemu:~# echo mem0 >/sys/bus/cxl/drivers/cxl_mem/bind
[ 75.076881] cxl_mem mem0: scan: iter: mem0 dport_dev: 0000:34:00.0 parent: pci0000:34
[ 75.077349] cxl_mem mem0: found already registered port port1:pci0000:34
[ 75.077653] cxl_mem mem0: host-bridge: pci0000:34
[ 75.078003] cxl_pci 0000:35:00.0: Failed to request region 0x00000000fe841110-0x00000000fe84119f
[ 75.078415] cxl_port endpoint2: Failed to map HDM capability.
[ 75.078683] cxl_port endpoint2: probe: -12
[ 75.078911] cxl_port: probe of endpoint2 failed with error -12
[ 75.079223] cxl_mem mem0: endpoint2 added to port1
[ 75.079469] cxl_mem mem0: endpoint2 failed probe
[ 75.079706] cxl_mem mem0: probe: -6
[ 75.079943] cxl_mem mem0: disconnect mem0 from port1
bash: echo: write error: No such device or address
root@qemu:~# cxl list -i
[
{
"memdev":"mem0",
"pmem_size":536870912,
"serial":0,
"host":"0000:35:00.0",
"state":"disabled"
}
]The issue causes ndctl cxl tests with the cxl_test module to fail too.
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
(quoted diff) +7
-0+static struct cxl_register_map *cxl_port_get_comp_map(struct cxl_port *port) +{ + /* + * HDM capability applies to Endpoints, USPs and VH Host + * Bridges. The Endpoint's component register mappings are + * located in the cxlds. + */
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Review
Quoted 4 lines
(quoted diff) +4
-0+ if (is_cxl_endpoint(port)) { + struct cxl_memdev *memdev = to_cxl_memdev(port->uport_dev); + + return &memdev->cxlds->comp_map;...but why? The issue here is that the @dev argument in that map is the memdev parent PCI device. However, in this context the @dev for devm operations wants to be
&port->dev.
Right, I fixed this by changing the function i/f to pass @dev for the devm operation. See below.
Quoted 7 lines
Quoted 3 lines
The
cxl_pcidriver stores thecomp_mapof the endpoint in the cxlds structure, structcxl_portis not yet available at this point.When you say "this point" you mean "that point" in
cxl_pci, right? I initially took that to mean literally "this" point in the quoted code above.
Yes, exactly. The mapping is in cxlds->comp_map, not port->comp_map.
Quoted 68 lines
Quoted 57 lines
See patch #2 of this series ("cxl/pci: Store the endpoint's Component Register mappings in struct
cxl_dev_state").Quoted 36 lines
Review
Quoted 4 lines
(quoted diff) +3
-0+ } + + return &port->comp_map;...so this is fine, and folding in the following resolves the test failure.
drivers/cxl/core/hdm.c +0
-16@@ -125,22 +125,6 @@ static bool should_emulate_decoders(struct cxl_endpoint_dvsec_info *info)return true; } -static struct cxl_register_map *cxl_port_get_comp_map (struct cxl_port *port) -{ -/* -* HDM capability applies to Endpoints, USPs and VH Host -* Bridges. The Endpoint's component register mappings are -* located in the cxlds. -*/ -if (is_cxl_endpoint (port)) { -struct cxl_memdev *memdev =to_cxl_memdev (port->uport_dev); - -return &memdev->cxlds->comp_map; - } - -return &port->comp_map; -} -/** * devm_cxl_setup_hdm - map HDM decoder component registers * @port: cxl_port to map drivers/cxl/core/hdm.c struct cxl_hdm *devm_cxl_setup_hdm(struct cxl_port *port, · line 147
Quoted 6 lines
drivers/cxl/core/hdm.c +1
-2@@ -160,8 +144,7 @@ struct cxl_hdm *devm_cxl_setup_hdm(struct cxl_port *port, cxlhdm->port = port;dev_set_drvdata (dev, cxlhdm); - comp_map =cxl_port_get_comp_map (port); - + comp_map = &port->comp_map;Can you check if the following works instead, I think the pre-initialization is missing in
cxl_mock_mem_probe()forcxl_test:drivers/cxl/core/mbox.c +0
-0drivers/cxl/core/mbox.c struct cxl_memdev_state *cxl_memdev_state_create(struct device *dev) · line 1337
Quoted 5 lines
drivers/cxl/core/mbox.c +2
-0@@ -1333,6 +1333,8 @@ struct cxl_memdev_state *cxl_memdev_state_create(struct device *dev)mutex_init (&mds->mbox_mutex);mutex_init (&mds->event.log_lock); mds->cxlds.dev = dev; + mds->cxlds.comp_map.dev = dev; + mds->cxlds.comp_map.resource =CXL_RESOURCE_NONE ;This has the same problem. @dev is specifying the lifetime of the mapping. The lifetime of the mapping needs to be relative to the driver using the registers. So if the
cxl_portdriver is mapping the component registers the only valid device in thecomp_mapis&port->dev.
Quoted 7 lines
I notice that
cxl_port_get_comp_map()endpoint port as an argument. That endpoint port was instantiated with @cxlmd, but it seems not with @cxlmd->cxlds->comp_map information which is available. Lets just fix that. I.e. movedevm_cxl_add_endpoint()into the core and make it initialize @endpoint->comp_map with @cxlds->comp_map while switching @dev in the @endpoint->comp_map to be @endpoint->dev, and not @cxlds->dev.
That does not work as we need the RAS cap early once cxl_pci is bound. There are multiple devices using the comp_map. For proper devm release we need to assign different devices depending on the users. That is, the RAS cap is released with cxl_pci driver unbinding and HDM with the memdev release. The fix I have chosen is to expand cxl_map_component_regs() to pass the devm device to release it together with the device using it.
Quoted 10 lines
...but it turns out that is the second bug in this patch. As I went to try to reproduce this failure for a single port VH configuration I notice another bug, a regression of this fix:
7bba261e0aa6cxl/port: Scan single-target ports for decoders...because there is no requirement for single port switches / host-bridges to have HDM decoders per the specification, and the original patch is turning HDM decoders not present as a hard failure.
But if the HDM is not present, a -ENODEV is still returned? cxl_switch_port_probe() does not fail then.
Quoted 7 lines
Quoted 7 lines
drivers/cxl/core/mbox.c +0
-0mds->cxlds.type =CXL_DEVTYPE_CLASSMEM ;return mds;Signature · 1 line
2.30.2
Quoted 6 lines
Quoted 1 line
The
cxl_pcidriver always has something valid or fails otherwise.Understood, just copy that map at
__devm_cxl_add_port()time. I am thinkingdevm_cxl_add_endpoint()moves intocore/port.cand it calls__devm_cxl_add_port()directly with a new parameter to take a passed in @comp_map or @component_reg_phys for the switch port case.
Please take a look at the cxl_map_component_regs() update in v9. We will send it out today. The changes compared to v8 are straightforward.
Thanks,
-Robert
Quoted 20 lines
Quoted 18 lines
If that works the change should be merged into patch #2.
Thanks,
-Robert
Quoted 9 lines
drivers/cxl/core/hdm.c +0
-0if (comp_map->resource ==CXL_RESOURCE_NONE ) {if (info && info->mem_enabled) { cxlhdm->decoder_count = info->ranges;Am I missing why the
cxlds->comp_mapneeds to be reused?Note that I am out and may not be able to respond further until next week.
PATCH v8 4/14 cxl/pci: Remove Component Register base address from struct cxl_dev_state
In reply to Terry BowmanFrom: Robert Richter <[email protected]>
The Component Register base address @component_reg_phys is no longer used after the rework of the Component Register setup which now uses struct member @comp_map instead. Remove the base address.
--- drivers/cxl/cxlmem.h | 2 -- drivers/cxl/mem.c | 4 ++-- drivers/cxl/pci.c | 3 --- tools/testing/cxl/test/mem.c | 1 - 4 files changed, 2 insertions(+), 8 deletions(-)
drivers/cxl/cxlmem.h +0 -2
@@ -390,7 +390,6 @@ enum cxl_devtype {* @dpa_res: Overall DPA resource tree for the device * @pmem_res: Active Persistent memory capacity configuration * @ram_res: Active Volatile memory capacity configuration -* @component_reg_phys: register base of component registers * @serial: PCIe Device Serial Number * @type: Generic Memory Class device or Vendor Specific Memory device */
@@ -405,7 +404,6 @@ struct cxl_dev_state {struct resource dpa_res;struct resource pmem_res;struct resource ram_res; -resource_size_t component_reg_phys;u64 serial;enum cxl_devtype type; };
drivers/cxl/mem.c +2 -2
@@ -49,7 +49,6 @@ static int devm_cxl_add_endpoint(struct device *host, struct cxl_memdev *cxlmd, struct cxl_dport *parent_dport) {struct cxl_port *parent_port = parent_dport->port; -struct cxl_dev_state *cxlds = cxlmd->cxlds;struct cxl_port *endpoint, *iter, *down;int rc;
@@ -65,8 +64,9 @@ static int devm_cxl_add_endpoint(struct device *host, struct cxl_memdev *cxlmd, ep->next = down; } +/* The Endpoint's component regs are located in cxlds. */ endpoint =devm_cxl_add_port (host, &cxlmd->dev, - cxlds->component_reg_phys, +CXL_RESOURCE_NONE , parent_dport);if (IS_ERR (endpoint))return PTR_ERR (endpoint);
drivers/cxl/pci.c +0 -3
@@ -835,7 +835,6 @@ static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)* If the component registers can't be found, the cxl_pci driver may * still be useful for management functions so don't return an error. */ - cxlds->component_reg_phys =CXL_RESOURCE_NONE ; rc =cxl_pci_setup_regs (pdev,CXL_REGLOC_RBI_COMPONENT , &cxlds->comp_map);if (rc)
@@ -843,8 +842,6 @@ static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)else if (!cxlds->comp_map.component_map.ras.valid)dev_dbg (&pdev->dev,"RAS registers not found\n" ); - cxlds->component_reg_phys = cxlds->comp_map.resource; - rc =cxl_map_component_regs (&cxlds->comp_map, &cxlds->regs.component,BIT (CXL_CM_CAP_CAP_ID_RAS ));if (rc)
tools/testing/cxl/test/mem.c +0 -1
@@ -1423,7 +1423,6 @@ static int cxl_mock_mem_probe(struct platform_device *pdev) cxlds->serial = pdev->id;if (is_rcd (pdev)) { cxlds->rcd = true; - cxlds->component_reg_phys =CXL_RESOURCE_NONE ; } rc =cxl_enumerate_cmds (mds);
Signature · 1 line
2.34.1
PATCH v8 5/14 cxl/port: Remove Component Register base address from struct cxl_port
In reply to Terry BowmanFrom: Robert Richter <[email protected]>
The Component Register base address @component_reg_phys is no longer used after the rework of the Component Register setup which now uses struct member @comp_map instead. Remove the base address.
--- drivers/cxl/core/port.c | 4 +--- drivers/cxl/cxl.h | 2 -- 2 files changed, 1 insertion(+), 5 deletions(-)
drivers/cxl/core/port.c +1 -3
@@ -618,7 +618,6 @@ static int devm_cxl_link_parent_dport(struct device *host,static struct lock_class_key cxl_port_key;static struct cxl_port *cxl_port_alloc (struct device *uport_dev, -resource_size_t component_reg_phys,struct cxl_dport *parent_dport) {struct cxl_port *port;
@@ -669,7 +668,6 @@ static struct cxl_port *cxl_port_alloc(struct device *uport_dev, }else dev->parent = uport_dev; - port->component_reg_phys = component_reg_phys;ida_init (&port->decoder_ida); port->hdm_end =-1 ; port->commit_end =-1 ;
@@ -730,7 +728,7 @@ static struct cxl_port *__devm_cxl_add_port(struct device *host,struct device *dev;int rc; - port =cxl_port_alloc (uport_dev, component_reg_phys, parent_dport); + port =cxl_port_alloc (uport_dev, parent_dport);if (IS_ERR (port))return port;
drivers/cxl/cxl.h +0 -2
@@ -576,7 +576,6 @@ struct cxl_dax_region {* @nr_dports: number of entries in @dports * @hdm_end: track last allocated HDM decoder instance for allocation ordering * @commit_end: cursor to track highest committed decoder for commit ordering -* @component_reg_phys: component register capability base address (optional) * @dead: last ep has been removed, force port re-creation * @depth: How deep this port is relative to the root. depth 0 is the root. * @cdat: Cached CDAT data
@@ -596,7 +595,6 @@ struct cxl_port {int nr_dports;int hdm_end;int commit_end; -resource_size_t component_reg_phys;bool dead;unsigned int depth;struct cxl_cdat {
Signature · 1 line
2.34.1
PATCH v8 6/14 cxl/pci: Add RCH downstream port AER register discovery
In reply to Terry BowmanRestricted CXL host (RCH) downstream port AER information is not currently logged while in the error state. One problem preventing the error logging is the AER and RAS registers are not accessible. The CXL driver requires changes to find RCH downstream port AER and RAS registers for purpose of error logging.
RCH downstream ports are not enumerated during a PCI bus scan and are instead discovered using system firmware, ACPI in this case.[1] The downstream port is implemented as a Root Complex Register Block (RCRB). The RCRB is a 4k memory block containing PCIe registers based on the PCIe root port.[2] The RCRB includes AER extended capability registers used for reporting errors. Note, the RCH's AER Capability is located in the RCRB memory space instead of PCI configuration space, thus its register access is different. Existing kernel PCIe AER functions can not be used to manage the downstream port AER capabilities and RAS registers because the port was not enumerated during PCI scan and the registers are not PCI config accessible.
Discover RCH downstream port AER extended capability registers. Use MMIO accesses to search for extended AER capability in RCRB register space.
[1] CXL 3.0 Spec, 9.11.2 - System Firmware View of CXL 1.1 Hierarchy
[2] CXL 3.0 Spec, 8.2.1.1 - RCH Downstream Port RCRB
--- drivers/cxl/core/core.h | 1 + drivers/cxl/core/port.c | 6 ++++++ drivers/cxl/core/regs.c | 35 +++++++++++++++++++++++++++++++++++ 3 files changed, 42 insertions(+)
drivers/cxl/core/core.h +1 -0
@@ -73,6 +73,7 @@ struct cxl_rcrb_info;resource_size_t __rcrb_to_component (struct device *dev,struct cxl_rcrb_info *ri,enum cxl_rcrb which); +u16 cxl_rcrb_to_aer (struct device *dev,resource_size_t rcrb);extern struct rw_semaphore cxl_dpa_rwsem;
drivers/cxl/core/port.c +6 -0
@@ -978,6 +978,8 @@ __devm_cxl_add_dport(struct cxl_port *port, struct device *dport_dev,return ERR_PTR (-ENOMEM );if (rcrb !=CXL_RESOURCE_NONE ) { +struct pci_host_bridge *host_bridge; + dport->rcrb.base = rcrb; component_reg_phys =__rcrb_to_component (dport_dev, &dport->rcrb,CXL_RCRB_DOWNSTREAM );
@@ -986,6 +988,10 @@ __devm_cxl_add_dport(struct cxl_port *port, struct device *dport_dev,return ERR_PTR (-ENXIO ); } + host_bridge =to_pci_host_bridge (dport_dev); +if (host_bridge->native_cxl_error) + dport->rcrb.aer_cap =cxl_rcrb_to_aer (dport_dev, dport->rcrb.base); + dport->rch = true; }
drivers/cxl/core/regs.c +35 -0
@@ -470,6 +470,41 @@ int cxl_setup_regs(struct cxl_register_map *map) }EXPORT_SYMBOL_NS_GPL (cxl_setup_regs,CXL ); +u16 cxl_rcrb_to_aer (struct device *dev,resource_size_t rcrb) +{ +void __iomem *addr; +u16 offset =0 ; +u32 cap_hdr; + +if (WARN_ON_ONCE (rcrb ==CXL_RESOURCE_NONE )) +return 0 ; + +if (!request_mem_region (rcrb,SZ_4K ,dev_name (dev))) +return 0 ; + + addr =ioremap (rcrb,SZ_4K ); +if (!addr) { +release_mem_region (rcrb,SZ_4K ); +return 0 ; + } + + cap_hdr =readl (addr + offset); +while (PCI_EXT_CAP_ID (cap_hdr) !=PCI_EXT_CAP_ID_ERR ) { + offset =PCI_EXT_CAP_NEXT (cap_hdr); +if (!offset) +break ; + cap_hdr =readl (addr + offset); + } + +if (offset) +dev_dbg (dev,"found AER extended capability (0x%x)\n" , offset); + +iounmap (addr); +release_mem_region (rcrb,SZ_4K ); + +return offset; +} +resource_size_t __rcrb_to_component (struct device *dev,struct cxl_rcrb_info *ri,enum cxl_rcrb which) {
Signature · 1 line
2.34.1
PATCH v8 7/14 PCI/AER: Refactor cper_print_aer() for use by CXL driver module
In reply to Terry BowmanThe CXL driver plans to use cper_print_aer() for logging restricted CXL host (RCH) AER errors. cper_print_aer() is not currently exported and therefore not usable by the CXL drivers built as loadable modules. Export the cper_print_aer() function. Use the EXPORT_SYMBOL_NS_GPL() variant to restrict the export to CXL drivers.
The CONFIG_ACPI_APEI_PCIEAER kernel config is currently used to enable cper_print_aer(). cper_print_aer() logs the AER registers and is useful in PCIE AER logging outside of APEI. Remove the CONFIG_ACPI_APEI_PCIEAER dependency to enable cper_print_aer().
The cper_print_aer() function name implies CPER specific use but is useful in non-CPER cases as well. Rename cper_print_aer() to pci_print_aer().
Also, update cxl_core to import CXL namespace imports.
--- drivers/cxl/core/port.c | 1 + drivers/pci/pcie/aer.c | 9 +++++---- include/linux/aer.h | 2 +- 3 files changed, 7 insertions(+), 5 deletions(-)
drivers/cxl/core/port.c +1 -0
@@ -2062,3 +2062,4 @@ static void cxl_core_exit(void)subsys_initcall (cxl_core_init);module_exit (cxl_core_exit);MODULE_LICENSE ("GPL v2" ); +MODULE_IMPORT_NS (CXL );
drivers/pci/pcie/aer.c +5 -4
@@ -771,9 +771,10 @@ int cper_severity_to_aer(int cper_severity) } }EXPORT_SYMBOL_GPL (cper_severity_to_aer); +#endif -void cper_print_aer (struct pci_dev *dev,int aer_severity, -struct aer_capability_regs *aer) +void pci_print_aer (struct pci_dev *dev,int aer_severity, +struct aer_capability_regs *aer) {int layer, agent, tlp_header_valid =0 ;u32 status, mask;
@@ -812,7 +813,7 @@ void cper_print_aer(struct pci_dev *dev, int aer_severity,trace_aer_event (dev_name (&dev->dev), (status & ~mask), aer_severity, tlp_header_valid, &aer->header_log); } -#endif +EXPORT_SYMBOL_NS_GPL (pci_print_aer,CXL );/** * add_error_device - list device to be handled
@@ -1009,7 +1010,7 @@ static void aer_recover_work_func(struct work_struct *work) PCI_SLOT(entry.devfn), PCI_FUNC(entry.devfn));continue ; } -cper_print_aer (pdev, entry.severity, entry.regs); +pci_print_aer (pdev, entry.severity, entry.regs);if (entry.severity ==AER_NONFATAL )pcie_do_recovery (pdev, pci_channel_io_normal, aer_root_reset);
include/linux/aer.h +1 -1
@@ -64,7 +64,7 @@ static inline void pci_save_aer_state(struct pci_dev *dev) {}static inline void pci_restore_aer_state (struct pci_dev *dev) {}#endif -void cper_print_aer (struct pci_dev *dev,int aer_severity, +void pci_print_aer (struct pci_dev *dev,int aer_severity,struct aer_capability_regs *aer);int cper_severity_to_aer (int cper_severity); void aer_recover_queue(int domain, unsigned int bus, unsigned int devfn,
Signature · 1 line
2.34.1
PATCH v8 8/14 cxl/pci: Update CXL error logging to use RAS register address
In reply to Terry BowmanThe CXL error handler currently only logs endpoint RAS status. The CXL topology includes several components providing RAS details to be logged during error handling.[1] Update the current handler's RAS logging to use a RAS register address. Also, update the error handler function names to be consistent with correctable and uncorrectable RAS. This will allow for adding support to log other CXL component's RAS details in the future.
[1] CXL3.0 Table 8-22 CXL_Capability_ID Assignment
--- drivers/cxl/core/pci.c | 44 +++++++++++++++++++++++++++++------------- 1 file changed, 31 insertions(+), 13 deletions(-)
drivers/cxl/core/pci.c +31 -13
@@ -646,32 +646,36 @@ void read_cdat_data(struct cxl_port *port) }EXPORT_SYMBOL_NS_GPL (read_cdat_data,CXL ); -void cxl_cor_error_detected (struct pci_dev *pdev) +static void __cxl_handle_cor_ras (struct cxl_dev_state *cxlds, +void __iomem *ras_base) { -struct cxl_dev_state *cxlds =pci_get_drvdata (pdev);void __iomem *addr;u32 status; -if (!cxlds->regs.ras) +if (!ras_base)return ; - addr = cxlds->regs.ras +CXL_RAS_CORRECTABLE_STATUS_OFFSET ; + addr = ras_base +CXL_RAS_CORRECTABLE_STATUS_OFFSET ; status =readl (addr);if (status &CXL_RAS_CORRECTABLE_STATUS_MASK ) {writel (status &CXL_RAS_CORRECTABLE_STATUS_MASK , addr);trace_cxl_aer_correctable_error (cxlds->cxlmd, status); } } -EXPORT_SYMBOL_NS_GPL (cxl_cor_error_detected,CXL ); + +static void cxl_handle_endpoint_cor_ras (struct cxl_dev_state *cxlds) +{ +return __cxl_handle_cor_ras (cxlds, cxlds->regs.ras); +}/* CXL spec rev3.0 8.2.4.16.1 */ -static void header_log_copy (struct cxl_dev_state *cxlds,u32 *log) +static void header_log_copy (void __iomem *ras_base,u32 *log) {void __iomem *addr;u32 *log_addr;int i, log_u32_size =CXL_HEADERLOG_SIZE /sizeof (u32); - addr = cxlds->regs.ras +CXL_RAS_HEADER_LOG_OFFSET ; + addr = ras_base +CXL_RAS_HEADER_LOG_OFFSET ; log_addr = log;for (i =0 ; i < log_u32_size; i++) {
@@ -685,17 +689,18 @@ static void header_log_copy(struct cxl_dev_state *cxlds, u32 *log)* Log the state of the RAS status registers and prepare them to log the * next error status. Return 1 if reset needed. */ -static bool cxl_report_and_clear (struct cxl_dev_state *cxlds) +static bool __cxl_handle_ras (struct cxl_dev_state *cxlds, +void __iomem *ras_base) {u32 hl[CXL_HEADERLOG_SIZE_U32 ];void __iomem *addr;u32 status;u32 fe; -if (!cxlds->regs.ras) +if (!ras_base)return false; - addr = cxlds->regs.ras +CXL_RAS_UNCORRECTABLE_STATUS_OFFSET ; + addr = ras_base +CXL_RAS_UNCORRECTABLE_STATUS_OFFSET ; status =readl (addr);if (!(status &CXL_RAS_UNCORRECTABLE_STATUS_MASK ))return false;
@@ -703,7 +708,7 @@ static bool cxl_report_and_clear(struct cxl_dev_state *cxlds)/* If multiple errors, log header points to first error from ctrl reg */ if (hweight32 (status) >1 ) {void __iomem *rcc_addr = - cxlds->regs.ras +CXL_RAS_CAP_CONTROL_OFFSET ; + ras_base +CXL_RAS_CAP_CONTROL_OFFSET ; fe =BIT (FIELD_GET (CXL_RAS_CAP_CONTROL_FE_MASK ,readl (rcc_addr)));
@@ -711,13 +716,26 @@ static bool cxl_report_and_clear(struct cxl_dev_state *cxlds) fe = status; } -header_log_copy (cxlds, hl); +header_log_copy (ras_base, hl);trace_cxl_aer_uncorrectable_error (cxlds->cxlmd, status, fe, hl);writel (status &CXL_RAS_UNCORRECTABLE_STATUS_MASK , addr);return true; } +static bool cxl_handle_endpoint_ras (struct cxl_dev_state *cxlds) +{ +return __cxl_handle_ras (cxlds, cxlds->regs.ras); +} + +void cxl_cor_error_detected (struct pci_dev *pdev) +{ +struct cxl_dev_state *cxlds =pci_get_drvdata (pdev); + +cxl_handle_endpoint_cor_ras (cxlds); +} +EXPORT_SYMBOL_NS_GPL (cxl_cor_error_detected,CXL ); +pci_ers_result_t cxl_error_detected (struct pci_dev *pdev,pci_channel_state_t state) {
@@ -732,7 +750,7 @@ pci_ers_result_t cxl_error_detected(struct pci_dev *pdev,* chance the situation is recoverable dump the status of the RAS * capability registers and bounce the active state of the memdev. */ - ue =cxl_report_and_clear (cxlds); + ue =cxl_handle_endpoint_ras (cxlds);switch (state) {case pci_channel_io_normal:
Signature · 1 line
2.34.1
PATCH v8 9/14 cxl/pci: Map RCH downstream AER registers for logging protocol errors
In reply to Terry BowmanThe restricted CXL host (RCH) error handler will log protocol errors using AER and RAS status registers. The AER and RAS registers need to be virtually memory mapped before enabling interrupts. Update __devm_cxl_add_dport() to include RCH RAS and AER mapping.
Add 'struct cxl_regs' to 'struct cxl_dport' for saving a pointer to the RCH downstream port's AER and RAS registers.
--- drivers/cxl/core/port.c | 33 +++++++++++++++++++++++++++++++++ drivers/cxl/core/regs.c | 1 + drivers/cxl/cxl.h | 12 ++++++++++++ 3 files changed, 46 insertions(+)
drivers/cxl/core/port.c +33 -0
@@ -8,6 +8,7 @@#include <linux/pci.h> #include <linux/slab.h> #include <linux/idr.h> +#include <linux/aer.h> #include <cxlmem.h> #include <cxlpci.h> #include <cxl.h>
@@ -948,6 +949,36 @@ static void cxl_dport_unlink(void *data)sysfs_remove_link (&port->dev.kobj, link_name); } +static void cxl_dport_map_rch_aer (struct cxl_dport *dport) +{ +struct cxl_rcrb_info *ri = &dport->rcrb; +struct cxl_port *port = dport->port; +void __iomem *dport_aer =NULL ; +resource_size_t aer_phys; + +if (dport->rch && ri->aer_cap) { + aer_phys = ri->aer_cap + ri->base; + dport_aer =devm_cxl_iomap_block (&port->dev, aer_phys, +sizeof (struct aer_capability_regs )); + } + + dport->regs.dport_aer = dport_aer; +} + +static void cxl_dport_map_regs (struct cxl_dport *dport) +{ +struct cxl_register_map *map = &dport->comp_map; + +if (!map->component_map.ras.valid) +dev_dbg (dport->dport_dev,"RAS registers not found\n" ); +else if (cxl_map_component_regs (map, &dport->regs.component, +BIT (CXL_CM_CAP_CAP_ID_RAS ))) +dev_dbg (dport->dport_dev,"Failed to map RAS capability.\n" ); + +if (dport->rch) +cxl_dport_map_rch_aer (dport); +} +static struct cxl_dport * __devm_cxl_add_dport(struct cxl_port *port, struct device *dport_dev, int port_id, resource_size_t component_reg_phys,
@@ -1007,6 +1038,8 @@ __devm_cxl_add_dport(struct cxl_port *port, struct device *dport_dev,if (rc)return ERR_PTR (rc); +cxl_dport_map_regs (dport); +cond_cxl_root_lock (port); rc =add_dport (port, dport);cond_cxl_root_unlock (port);
drivers/cxl/core/regs.c +1 -0
@@ -199,6 +199,7 @@ void __iomem *devm_cxl_iomap_block(struct device *dev, resource_size_t addr,return ret_val; } +EXPORT_SYMBOL_NS_GPL (devm_cxl_iomap_block,CXL ); int cxl_map_component_regs(const struct cxl_register_map *map, struct cxl_component_regs *regs,
drivers/cxl/cxl.h +12 -0
@@ -221,6 +221,14 @@ struct cxl_regs {struct_group_tagged (cxl_pmu_regs, pmu_regs, void __iomem *pmu; ); + +/* +* RCH downstream port specific RAS register +* @aer: CXL 3.0 8.2.1.1 RCH Downstream Port RCRB +*/ +struct_group_tagged (cxl_rch_regs, rch_regs, + void __iomem *dport_aer; + ); };struct cxl_reg_map {
@@ -273,6 +281,8 @@ void cxl_probe_component_regs(struct device *dev, void __iomem *base, struct cxl_component_reg_map *map);void cxl_probe_device_regs (struct device *dev,void __iomem *base,struct cxl_device_reg_map *map); +void __iomem *devm_cxl_iomap_block (struct device *dev,resource_size_t addr, +resource_size_t length);int cxl_map_component_regs (const struct cxl_register_map *map,struct cxl_component_regs *regs,unsigned long map_mask);
@@ -623,6 +633,7 @@ struct cxl_rcrb_info {* @rcrb: Data about the Root Complex Register Block layout * @rch: Indicate whether this dport was enumerated in RCH or VH mode * @port: reference to cxl_port that contains this downstream port +* @regs: Dport parsed register blocks */ struct cxl_dport {struct device *dport_dev;
@@ -631,6 +642,7 @@ struct cxl_dport {struct cxl_rcrb_info rcrb;bool rch;struct cxl_port *port; +struct cxl_regs regs; };/**
Signature · 1 line
2.34.1
PATCH v8 10/14 cxl/pci: Add RCH downstream port error logging
In reply to Terry BowmanRCH downstream port error logging is missing in the current CXL driver. The missing AER and RAS error logging is needed for communicating driver error details to userspace. Update the driver to include PCIe AER and CXL RAS error logging.
Add RCH downstream port error handling into the existing RCiEP handler. The downstream port error handler is added to the RCiEP error handler because the downstream port is implemented in a RCRB, is not PCI enumerable, and as a result is not directly accessible to the PCI AER root port driver. The AER root port driver calls the RCiEP handler for handling RCD errors and RCH downstream port protocol errors.
Update existing RCiEP correctable and uncorrectable handlers to also call the RCH handler. The RCH handler will read the RCH AER registers, check for error severity, and if an error exists will log using an existing kernel AER trace routine. The RCH handler will also log downstream port RAS errors if they exist.
--- drivers/cxl/core/pci.c | 101 +++++++++++++++++++++++++++++++++++++++++ 1 file changed, 101 insertions(+)
drivers/cxl/core/pci.c +101 -0
@@ -5,6 +5,7 @@#include <linux/delay.h> #include <linux/pci.h> #include <linux/pci-doe.h> +#include <linux/aer.h> #include <cxlpci.h> #include <cxlmem.h> #include <cxl.h>
@@ -728,10 +729,107 @@ static bool cxl_handle_endpoint_ras(struct cxl_dev_state *cxlds)return __cxl_handle_ras (cxlds, cxlds->regs.ras); } +#ifdef CONFIG_PCIEAER_CXL + +static void cxl_handle_rdport_cor_ras (struct cxl_dev_state *cxlds, +struct cxl_dport *dport) +{ +return __cxl_handle_cor_ras (cxlds, dport->regs.ras); +} + +static bool cxl_handle_rdport_ras (struct cxl_dev_state *cxlds, +struct cxl_dport *dport) +{ +return __cxl_handle_ras (cxlds, dport->regs.ras); +} + +/* +* Copy the AER capability registers using 32 bit read accesses. +* This is necessary because RCRB AER capability is MMIO mapped. Clear the +* status after copying. +* +* @aer_base: base address of AER capability block in RCRB +* @aer_regs: destination for copying AER capability +*/ +static bool cxl_rch_get_aer_info (void __iomem *aer_base, +struct aer_capability_regs *aer_regs) +{ +int read_cnt =sizeof (struct aer_capability_regs ) /sizeof (u32); +u32 *aer_regs_buf = (u32 *)aer_regs; +int n; + +if (!aer_base) +return false; + +/* Use readl() to guarantee 32-bit accesses */ +for (n =0 ; n < read_cnt; n++) + aer_regs_buf[n] =readl (aer_base + n *sizeof (u32)); + +writel (aer_regs->uncor_status, aer_base +PCI_ERR_UNCOR_STATUS ); +writel (aer_regs->cor_status, aer_base +PCI_ERR_COR_STATUS ); + +return true; +} + +/* Get AER severity. Return false if there is no error. */ +static bool cxl_rch_get_aer_severity (struct aer_capability_regs *aer_regs, +int *severity) +{ +if (aer_regs->uncor_status & ~aer_regs->uncor_mask) { +if (aer_regs->uncor_status &PCI_ERR_ROOT_FATAL_RCV ) + *severity =AER_FATAL ; +else + *severity =AER_NONFATAL ; +return true; + } + +if (aer_regs->cor_status & ~aer_regs->cor_mask) { + *severity =AER_CORRECTABLE ; +return true; + } + +return false; +} + +static void cxl_handle_rdport_errors (struct cxl_dev_state *cxlds) +{ +struct pci_dev *pdev =to_pci_dev (cxlds->dev); +struct aer_capability_regs aer_regs; +struct cxl_dport *dport; +struct cxl_port *port; +int severity; + + port =cxl_pci_find_port (pdev, &dport); +if (!port) +return ; + +put_device (&port->dev); + +if (!cxl_rch_get_aer_info (dport->regs.dport_aer, &aer_regs)) +return ; + +if (!cxl_rch_get_aer_severity (&aer_regs, &severity)) +return ; + +pci_print_aer (pdev, severity, &aer_regs); + +if (severity ==AER_CORRECTABLE ) +cxl_handle_rdport_cor_ras (cxlds, dport); +else +cxl_handle_rdport_ras (cxlds, dport); +} + +#else +static void cxl_handle_rdport_errors (struct cxl_dev_state *cxlds) { } +#endif +void cxl_cor_error_detected (struct pci_dev *pdev) {struct cxl_dev_state *cxlds =pci_get_drvdata (pdev); +if (cxlds->rcd) +cxl_handle_rdport_errors (cxlds); +cxl_handle_endpoint_cor_ras (cxlds); }EXPORT_SYMBOL_NS_GPL (cxl_cor_error_detected,CXL );
@@ -744,6 +842,9 @@ pci_ers_result_t cxl_error_detected(struct pci_dev *pdev,struct device *dev = &cxlmd->dev;bool ue; +if (cxlds->rcd) +cxl_handle_rdport_errors (cxlds); +/* * A frozen channel indicates an impending reset which is fatal to * CXL.mem operation, and will likely crash the system. On the off
Signature · 1 line
2.34.1
PATCH v8 11/14 cxl/pci: Disable root port interrupts in RCH mode
In reply to Terry BowmanThe RCH root port contains root command AER registers that should not be enabled.[1] Disable these to prevent root port interrupts.
[1] CXL 3.0 - 12.2.1.1 RCH Downstream Port-detected Errors
--- drivers/cxl/core/core.h | 6 ++++++ drivers/cxl/core/pci.c | 29 +++++++++++++++++++++++++++++ drivers/cxl/core/port.c | 3 +++ 3 files changed, 38 insertions(+)
drivers/cxl/core/core.h +6 -0
@@ -87,4 +87,10 @@ enum cxl_poison_trace_type { CXL_POISON_TRACE_CLEAR, }; +#ifdef CONFIG_PCIEAER_CXL +void cxl_disable_rch_root_ints (struct cxl_dport *dport); +#else +static inline void cxl_disable_rch_root_ints (struct cxl_dport *dport) { }; +#endif +#endif /* __CXL_CORE_H__ */
drivers/cxl/core/pci.c +29 -0
@@ -819,6 +819,35 @@ static void cxl_handle_rdport_errors(struct cxl_dev_state *cxlds) cxl_handle_rdport_ras(cxlds, dport); } +void cxl_disable_rch_root_ints (struct cxl_dport *dport) +{ +void __iomem *aer_base = dport->regs.dport_aer; +struct pci_host_bridge *bridge; +u32 aer_cmd_mask, aer_cmd; + +if (!aer_base) +return ; + + bridge =to_pci_host_bridge (dport->dport_dev); + +/* +* Disable RCH root port command interrupts. +* CXL 3.0 12.2.1.1 - RCH Downstream Port-detected Errors +* +* This sequence may not be necessary. CXL spec states disabling +* the root cmd register's interrupts is required. But, PCI spec +* shows these are disabled by default on reset. +*/ +if (bridge->native_cxl_error) { + aer_cmd_mask = (PCI_ERR_ROOT_CMD_COR_EN | +PCI_ERR_ROOT_CMD_NONFATAL_EN | +PCI_ERR_ROOT_CMD_FATAL_EN ); + aer_cmd =readl (aer_base +PCI_ERR_ROOT_COMMAND ); + aer_cmd &= ~aer_cmd_mask; +writel (aer_cmd, aer_base +PCI_ERR_ROOT_COMMAND ); + } +} +#else static void cxl_handle_rdport_errors (struct cxl_dev_state *cxlds) { }#endif
drivers/cxl/core/port.c +3 -0
@@ -1040,6 +1040,9 @@ __devm_cxl_add_dport(struct cxl_port *port, struct device *dport_dev,cxl_dport_map_regs (dport); +if (dport->rch) +cxl_disable_rch_root_ints (dport); +cond_cxl_root_lock (port); rc =add_dport (port, dport);cond_cxl_root_unlock (port);
Signature · 1 line
2.34.1
PATCH v8 12/14 PCI/AER: Forward RCH downstream port-detected errors to the CXL.mem dev handler
In reply to Terry BowmanFrom: Robert Richter <[email protected]>
In Restricted CXL Device (RCD) mode a CXL device is exposed as an RCiEP, but CXL downstream and upstream ports are not enumerated and not visible in the PCIe hierarchy. [1] Protocol and link errors from these non-enumerated ports are signaled as internal AER errors, either Uncorrectable Internal Error (UIE) or Corrected Internal Errors (CIE) via an RCEC.
Restricted CXL host (RCH) downstream port-detected errors have the Requester ID of the RCEC set in the RCEC's AER Error Source ID register. A CXL handler must then inspect the error status in various CXL registers residing in the dport's component register space (CXL RAS capability) or the dport's RCRB (PCIe AER extended capability). [2]
Errors showing up in the RCEC's error handler must be handled and connected to the CXL subsystem. Implement this by forwarding the error to all CXL devices below the RCEC. Since the entire CXL device is controlled only using PCIe Configuration Space of device 0, function 0, only pass it there [3]. The error handling is limited to currently supported devices with the Memory Device class code set (CXL Type 3 Device, PCI_CLASS_MEMORY_CXL, 502h), handle downstream port errors in the device's cxl_pci driver. Support for other CXL Device Types (e.g. a CXL.cache Device) can be added later.
To handle downstream port errors in addition to errors directed to the CXL endpoint device, a handler must also inspect the CXL RAS and PCIe AER capabilities of the CXL downstream port the device is connected to.
Since CXL downstream port errors are signaled using internal errors, the handler requires those errors to be unmasked. This is subject of a follow-on patch.
The reason for choosing this implementation is that the AER service driver claims the RCEC device, but does not allow it to register a custom specific handler to support CXL. Connecting the RCEC hard-wired with a CXL handler does not work, as the CXL subsystem might not be present all the time. The alternative to add an implementation to the portdrv to allow the registration of a custom RCEC error handler isn't worth doing it as CXL would be its only user. Instead, just check for an CXL RCEC and pass it down to the connected CXL device's error handler. With this approach the code can entirely be implemented in the PCIe AER driver and is independent of the CXL subsystem. The CXL driver only provides the handler.
[1] CXL 3.0 spec: 9.11.8 CXL Devices Attached to an RCH
[2] CXL 3.0 spec, 12.2.1.1 RCH Downstream Port-detected Errors
[3] CXL 3.0 spec, 8.1.3 PCIe DVSEC for CXL Devices
--- drivers/pci/pcie/Kconfig | 12 +++++ drivers/pci/pcie/aer.c | 96 +++++++++++++++++++++++++++++++++++++++- 2 files changed, 106 insertions(+), 2 deletions(-)
drivers/pci/pcie/Kconfig +12 -0
@@ -49,6 +49,18 @@ config PCIEAER_INJECT gotten from: https://git.kernel.org/cgit/linux/kernel/git/gong.chen/aer-inject.git/ +config PCIEAER_CXL + bool "PCI Express CXL RAS support for Restricted Hosts (RCH)" + default y + depends on PCIEAER && CXL_PCI + help + Enables error handling of downstream ports of a CXL host + that is operating in RCD mode (Restricted CXL Host, RCH). + The downstream port reports AER errors to a given RCEC. + Errors are handled by the CXL memory device driver. + + If unsure, say Y. + # # PCI Express ECRC #
drivers/pci/pcie/aer.c +94 -2
114 lines not on this page · raw message
Signature · 1 line
2.34.1
PATCH v8 13/14 PCI/AER: Unmask RCEC internal errors to enable RCH downstream port error handling
In reply to Terry BowmanFrom: Robert Richter <[email protected]>
AER corrected and uncorrectable internal errors (CIE/UIE) are masked in their corresponding mask registers per default once in power-up state. [1][2] Enable internal errors for RCECs to receive CXL downstream port errors of Restricted CXL Hosts (RCHs).
[1] CXL 3.0 Spec, 12.2.1.1 - RCH Downstream Port Detected Errors
[2] PCIe Base Spec r6.0, 7.8.4.3 Uncorrectable Error Mask Register, 7.8.4.6 Correctable Error Mask Register
--- drivers/pci/pcie/aer.c | 57 ++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 57 insertions(+)
drivers/pci/pcie/aer.c +57 -0
@@ -948,6 +948,30 @@ static bool find_source_device(struct pci_dev *parent,#ifdef CONFIG_PCIEAER_CXL +/** +* pci_aer_unmask_internal_errors - unmask internal errors +* @dev: pointer to the pcie_dev data structure +* +* Unmasks internal errors in the Uncorrectable and Correctable Error +* Mask registers. +* +* Note: AER must be enabled and supported by the device which must be +* checked in advance, e.g. with pcie_aer_is_native(). +*/ +static void pci_aer_unmask_internal_errors (struct pci_dev *dev) +{ +int aer = dev->aer_cap; +u32 mask; + +pci_read_config_dword (dev, aer +PCI_ERR_UNCOR_MASK , &mask); + mask &= ~PCI_ERR_UNC_INTN ; +pci_write_config_dword (dev, aer +PCI_ERR_UNCOR_MASK , mask); + +pci_read_config_dword (dev, aer +PCI_ERR_COR_MASK , &mask); + mask &= ~PCI_ERR_COR_INTERNAL ; +pci_write_config_dword (dev, aer +PCI_ERR_COR_MASK , mask); +} +static bool is_cxl_mem_dev (struct pci_dev *dev) {/*
@@ -1027,7 +1051,39 @@ static void cxl_rch_handle_error(struct pci_dev *dev, struct aer_err_info *info)pcie_walk_rcec (dev, cxl_rch_handle_error_iter, info); } +static int handles_cxl_error_iter (struct pci_dev *dev,void *data) +{ +bool *handles_cxl = data; + +if (!*handles_cxl) + *handles_cxl =is_cxl_mem_dev (dev) &&cxl_error_is_native (dev); + +/* Non-zero terminates iteration */ +return *handles_cxl; +} + +static bool handles_cxl_errors (struct pci_dev *rcec) +{ +bool handles_cxl = false; + +if (pci_pcie_type (rcec) ==PCI_EXP_TYPE_RC_EC && +pcie_aer_is_native (rcec)) +pcie_walk_rcec (rcec, handles_cxl_error_iter, &handles_cxl); + +return handles_cxl; +} + +static void cxl_rch_enable_rcec (struct pci_dev *rcec) +{ +if (!handles_cxl_errors (rcec)) +return ; + +pci_aer_unmask_internal_errors (rcec); +pci_info (rcec,"CXL: Internal errors unmasked" ); +} +#else +static inline void cxl_rch_enable_rcec (struct pci_dev *dev) { }static inline void cxl_rch_handle_error (struct pci_dev *dev,struct aer_err_info *info) { }#endif
@@ -1428,6 +1484,7 @@ static int aer_probe(struct pcie_device *dev)return status; } +cxl_rch_enable_rcec (port);aer_enable_rootport (rpc);pci_info (port,"enabled with IRQ %d\n" , dev->irq);return 0 ;
Signature · 1 line
2.34.1
PATCH v8 14/14 cxl/core/regs: Rename phys_addr in cxl_map_component_regs()
In reply to Terry BowmanFrom: Robert Richter <[email protected]>
Trivial change that renames variable phys_addr in cxl_map_component_regs() to shorten its length to keep the 80 char size limit for the line and also for consistency between the different paths.
--- drivers/cxl/core/regs.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/cxl/core/regs.c +3 -3
@@ -217,16 +217,16 @@ int cxl_map_component_regs(const struct cxl_register_map *map,for (i =0 ; i <ARRAY_SIZE (mapinfo); i++) {struct mapinfo *mi = &mapinfo[i]; -resource_size_t phys_addr; +resource_size_t addr;resource_size_t length;if (!mi->rmap->valid)continue ;if (!test_bit (mi->rmap->id, &map_mask))continue ; - phys_addr = map->resource + mi->rmap->offset; + addr = map->resource + mi->rmap->offset; length = mi->rmap->size; - *(mi->addr) =devm_cxl_iomap_block (dev, phys_addr, length); + *(mi->addr) =devm_cxl_iomap_block (dev, addr, length);if (!*(mi->addr))return -ENOMEM ; }
Signature · 1 line
2.34.1
Quoted 14 lines · Terry Bowman
This patchset enables CXL RCH error handling. This is necessary because RCH downstream port protocol error handling is implemented uniquely and not currently supported. These patches address the following:
* Discovery and mapping of RCH downstream port AER registers.
* AER portdrv changes to support CXL RCH protocol errors.
* Interrupt setup specific to RCH mode: enabling RCEC internal errors and disabling root port interrupts.
Changes in V8:
- Rebased onto: commit0c0df63177e3 ("Merge branch 'for-6.5/cxl-rch-eh' into for-6.5/cxl")
Quoted 16 lines
- cxl/port: Pre-initialize component register mappings
- Added patch to pre-initialize component register mappings.
- cxl/pci: Remove Component Register base address from
- Separated removal of Component Register base address in structcxl_dev_stateto not break functionality.
- cxl/hdm: Use stored Component Register mappings to map HDM decoder capability
- Implemented a less strict check indevm_cxl_setup_hdm(), be tolerant if HDM decoder registers are not implemented.
- cxl/pci: Map RCH downstream AER registers for logging protocol errors
- Fixed uninitialized access ofmap->devincxl_dport_map_regs().
- PCI/AER: Forward RCH downstream port-detected errors to the CXL.mem dev handler
- Fix typo in patch description.
- cxl/pci: Update CXL error logging to use RAS register address
- Fix typo in patch description.
Hey Terry, thank you for turning this around so quickly. I will have a look and see if this is suitable as a follow-up merge for v6.5 next week since it would be nice to finish this off this cycle, but no promises.