cxl: dport fixes from sashiko reports
Subsystems (2)
Dave Jiang
maintainerv4 cover cxl: dport fixes from sashiko reports
Couple dport fixes resulted from sashiko pre-existing issue reports. There are no depedencies between the two patches.
v4:
- Drop the unused dport output parameter in patch 2. (Sashiko, Alison)
- Add review tags from Alison and Ming.
v3:
- Add patch 3 to attach endpoints under the dport's port lock. (Ming)
- Add review tags from Ming and Gregory to patch 1.
v2:
- Addressed couple things from Jonathan. See patch for revlog.
Dave Jiang (3):
cxl/port: Clear cached dport pointers when a dport is removed cxl/core: Hold the dport host lock across dport lookup and use cxl/port: Attach endpoints to a dport under the lock that pins it
drivers/cxl/core/core.h | 6 +- drivers/cxl/core/pci.c | 8 +- drivers/cxl/core/port.c | 206 ++++++++++++++++++++++++------------- drivers/cxl/core/ras_rch.c | 12 ++- drivers/cxl/cxl.h | 22 +++- drivers/cxl/mem.c | 14 ++- drivers/cxl/pci.c | 11 +- 7 files changed, 186 insertions(+), 93 deletions(-)
base-commit: fd73f4a6659897191fa0d40695fe370925dd3780
Signature · 1 line
2.54.0
Dave Jiang
maintainerPATCH v4 1/3 cxl/port: Clear cached dport pointers when a dport is removed
Switch decoders cache dport pointers in cxlsd->target[], and nothing clears them when a dport is freed. Readers then dereference freed memory.
KASAN caught it under a cxl_test load/unload loop with concurrent sysfs reads (abbreviated).
Clear the matching slots from cxl_dport_remove(), walking the port's decoders the way cxl_port_update_decoder_targets() does on the add side. Scan all nr_targets slots, the target[] allocation size, and clear every hit.
---
Found by KASAN while stress-testing a separate dport locking fix, not from a failure report.
Reproducer: cxl_test, CONFIG_KASAN=y, 12 concurrent readers cat'ing /sys/bus/cxl/devices/*/target_list while cxl_test is unloaded and reloaded.
With the fix reverted the UAF lands ~1s into the first unload, on the trace above. With it applied, 36 cycles across two runs are clean: no KASAN report, no WARNING, lockdep never self-disables. clear_decoder_target() fired 2916 times during those runs, so the new path ran under the load.
KASAN only reports once per boot unless kasan_multi_shot is set (report_enabled(), mm/kasan/report.c), so the reverted-fix run cannot show a per-cycle count - one report is the ceiling, not the frequency.
target_list output is unaffected. Clearing a slot truncates the list at the first NULL, so that was the regression to watch for: 88 target_list files, no empty values before or after, and the multiset of values is unchanged once the ida-assigned decoder names are stripped.
Enumeration is unchanged too: 15 memdev, 12 port, 15 endpoint, 192 decoder.
--- drivers/cxl/core/port.c | 29 +++++++++++++++++++++++++++++ 1 file changed, 29 insertions(+)
drivers/cxl/core/port.c +29 -0
@@ -1089,11 +1089,40 @@ static void cond_cxl_root_unlock(struct cxl_port *port)device_unlock (&port->dev); } +static int clear_decoder_target (struct device *dev,void *data) +{ +struct cxl_dport *dport = data; +struct cxl_switch_decoder *cxlsd; + +if (!is_switch_decoder (dev)) +return 0 ; + + cxlsd =to_cxl_switch_decoder (dev); +guard (rwsem_write)(&cxl_rwsem.region); + +/* +* A dport can occupy more than one position of an interleave, so +* scan the whole target list rather than stopping at the first hit. +*/ +for (int i =0 ; i < cxlsd->nr_targets; i++) { +if (cxlsd->target[i] != dport) +continue ; + cxlsd->target[i] =NULL ; +dev_dbg (dev,"dport%d removed from target list, index %d\n" , + dport->port_id, i); + } + +return 0 ; +} +static void cxl_dport_remove (void *data) {struct cxl_dport *dport = data;struct cxl_port *port = dport->port; +/* counterpart of cxl_port_update_decoder_targets() */ +device_for_each_child (&port->dev, dport, clear_decoder_target); + port->nr_dports--;xa_erase (&port->dports, (unsigned long ) dport->dport_dev);put_device (dport->dport_dev);
Signature · 1 line
2.54.0
Dave Jiang
maintainerPATCH v4 2/3 cxl/core: Hold the dport host lock across dport lookup and use
In reply to Dave JiangCallers of cxl_pci_find_port() and cxl_mem_find_port() can be handed a 'struct cxl_dport' that is already freed, and all four dereference it without the lock that would prevent that.
A dport is kzalloc()'d into a devres group on its host device, so it dies when that host unbinds. The host is &port->dev for a switch port and port->uport_dev for the root port.
Both helpers return the dport through an output parameter. The caller's port reference pins struct cxl_port, not the dport allocation, and the lookup itself is a bare xa_load() under no lock.
Add cxl_port_dport_host() to name the device that pins a dport, and cxl_pci_find_dport() / cxl_mem_find_dport() to look one up with that device's lock held and asserted. Convert the four callers to use them and bail if the dport is gone. cxl_port_add_dport() already follows this rule.
cxl_mem_probe() chose its endpoint devm host from dport->rch, which meant dereferencing the dport before taking any lock. Use cxl_port_dport_host():
rch dports only ever land on the root port, so it selects the same device without needing the dport.
With every caller converted, the dport output parameter of cxl_pci_find_port() and cxl_mem_find_port() is unused. Drop it.
---
v4:
- Drop the now unused dport output parameter. (Sashiko, Alison)
Found by review, not from a failure report.
Untested: two of the four converted sites add a new lock acquisition, and neither executed under test.
cxl_pci_setup_regs() new guard() sits behind is_cxl_restricted(),
which needs a PCI_EXP_TYPE_RC_END device
cxl_handle_rdport_errors() needs an RCH dport AER error; cxl_test has no
injection path
Only cxl_pci_setup_regs() introduces a lock order lockdep has not already
seen: device_lock(&pdev->dev) from probe, then the dport host lock.
cxl_handle_rdport_errors() nests its new host lock inside the device_lock(&cxlmd->dev) that core/ras.c:274 and :298 hold, but cxl_mem_probe() takes that same pair in that same order and ran clean under PROVE_LOCKING, so the edge itself is covered. What is untested there is the path, not the ordering.
Reaching either needs a cxl-type3 attached directly to a host bridge so pci_pcie_type() is PCI_EXP_TYPE_RC_END, which run_qemu.sh does not generate.
Tested in QEMU with cxl_test, PROVE_LOCKING and KASAN on, 8 module load/unload cycles with concurrent sysfs readers. No lockdep splat, no device_lock_assert() firing, enumeration unchanged (15 memdev, 12 port, 192 decoder before and after).
cxl_mem_find_dport() ran for 15 memdevs on each of 9 loads, covering both cxl_port_dport_host() branches. The rch branch is covered too: "cxl_mem
mem2: endpoint9 added to root3" shows the new selection picking the same parent dport->rch did.
--- drivers/cxl/core/core.h | 6 +--- drivers/cxl/core/pci.c | 8 ++++- drivers/cxl/core/port.c | 62 ++++++++++++++++++++++++++------------ drivers/cxl/core/ras_rch.c | 12 +++++++- drivers/cxl/cxl.h | 22 +++++++++++--- drivers/cxl/mem.c | 14 ++++++--- drivers/cxl/pci.c | 11 ++++--- 7 files changed, 95 insertions(+), 40 deletions(-)
drivers/cxl/core/core.h +1 -5
@@ -178,11 +178,7 @@ static inline struct device *port_to_host(struct cxl_port *port)static inline struct device *dport_to_host (struct cxl_dport *dport) { -struct cxl_port *port = dport->port; - -if (is_cxl_root (port)) -return port->uport_dev; -return &port->dev; +return cxl_port_dport_host (dport->port); }#ifdef CONFIG_CXL_RAS void cxl_ras_init (void );
drivers/cxl/core/pci.c +7 -1
@@ -734,10 +734,16 @@ int cxl_pci_setup_regs(struct pci_dev *pdev, enum cxl_regloc_type type,if (rc && type ==CXL_REGLOC_RBI_COMPONENT &&is_cxl_restricted (pdev)) {struct cxl_dport *dport; struct cxl_port *port __free(put_cxl_port) = -cxl_pci_find_port (pdev, &dport); +cxl_pci_find_port (pdev);if (!port)return -EPROBE_DEFER ; +guard (device)(cxl_port_dport_host (port)); + + dport =cxl_pci_find_dport (pdev, port); +if (!dport) +return -EPROBE_DEFER ; + rc =cxl_rcrb_get_comp_regs (pdev, map, dport);if (rc)return rc;
drivers/cxl/core/port.c +43 -19
108 lines not on this page · raw message
drivers/cxl/core/ras_rch.c +11 -1
@@ -118,10 +118,20 @@ void cxl_handle_rdport_errors(struct cxl_dev_state *cxlds)int severity; struct cxl_port *port __free(put_cxl_port) = -cxl_pci_find_port (pdev, &dport); +cxl_pci_find_port (pdev);if (!port)return ; +/* +* The dport may have been freed since the lookup above, so take the +* lock that pins it and find it again. +*/ +guard (device)(cxl_port_dport_host (port)); + + dport =cxl_pci_find_dport (pdev, port); +if (!dport) +return ; +if (!cxl_rch_get_aer_info (dport->regs.dport_aer, &aer_regs))return ;
drivers/cxl/cxl.h +18 -4
@@ -701,6 +701,18 @@ static inline bool is_cxl_root(struct cxl_port *port)return port->uport_dev == port->dev.parent; } +/* +* Return the owner device of the devm allocations of @port's dports. They +* are freed when that device unbinds, so hold device_lock() on the returned +* device across any lookup and use of a 'struct cxl_dport'. +*/ +static inline struct device *cxl_port_dport_host (struct cxl_port *port) +{ +if (is_cxl_root (port)) +return port->uport_dev; +return &port->dev; +} +/* Address translation functions exported to cxl_translate test module only */ int cxl_validate_translation_params (u8 eiw,u16 eig,int pos);u64 cxl_calculate_hpa_offset (u64 dpa_offset,int pos,u8 eiw,u16 eig);
@@ -747,10 +759,12 @@ DEFINE_FREE(put_cxl_dax_region, struct cxl_dax_region *, if (!IS_ERR_OR_NULL(_T)int devm_cxl_enumerate_ports (struct cxl_memdev *cxlmd);void cxl_bus_rescan (void );void cxl_bus_drain (void ); -struct cxl_port *cxl_pci_find_port (struct pci_dev *pdev, -struct cxl_dport **dport); -struct cxl_port *cxl_mem_find_port (struct cxl_memdev *cxlmd, -struct cxl_dport **dport); +struct cxl_dport *cxl_pci_find_dport (struct pci_dev *pdev, +struct cxl_port *port); +struct cxl_dport *cxl_mem_find_dport (struct cxl_memdev *cxlmd, +struct cxl_port *port); +struct cxl_port *cxl_pci_find_port (struct pci_dev *pdev); +struct cxl_port *cxl_mem_find_port (struct cxl_memdev *cxlmd);bool schedule_cxl_memdev_detach (struct cxl_memdev *cxlmd); struct cxl_dport *devm_cxl_add_dport(struct cxl_port *port,
drivers/cxl/mem.c +9 -5
@@ -133,7 +133,7 @@ static int cxl_mem_probe(struct device *dev)return rc; struct cxl_port *parent_port __free(put_cxl_port) = -cxl_mem_find_port (cxlmd, &dport); +cxl_mem_find_port (cxlmd);if (!parent_port) {dev_err (dev,"CXL port topology not found\n" );return -ENXIO ;
@@ -148,10 +148,7 @@ static int cxl_mem_probe(struct device *dev) } } -if (dport->rch) - endpoint_parent = parent_port->uport_dev; -else - endpoint_parent = &parent_port->dev; + endpoint_parent =cxl_port_dport_host (parent_port); scoped_guard(device, endpoint_parent) {if (!endpoint_parent->driver) {
@@ -160,6 +157,13 @@ static int cxl_mem_probe(struct device *dev)return -ENXIO ; } +/* @dport is only valid while @endpoint_parent stays bound */ + dport =cxl_mem_find_dport (cxlmd, parent_port); +if (!dport) { +dev_err (dev,"CXL topology dport not found\n" ); +return -ENXIO ; + } + rc =devm_cxl_add_endpoint (endpoint_parent, cxlmd, dport);if (rc)return rc;
drivers/cxl/pci.c +6 -5
@@ -706,22 +706,23 @@ static ssize_t rcd_pcie_cap_emit(struct device *dev, u16 offset, char *buf, sizestruct device *root_dev;struct cxl_dport *dport; struct cxl_port *root __free(put_cxl_port) = -cxl_mem_find_port (cxlmd, &dport); +cxl_mem_find_port (cxlmd);if (!root)return -ENXIO ; - root_dev = root->uport_dev; + root_dev =cxl_port_dport_host (root);if (!root_dev)return -ENXIO ; -if (!dport->regs.rcd_pcie_cap) -return -ENXIO ; -guard (device)(root_dev);if (!root_dev->driver)return -ENXIO ; + dport =cxl_mem_find_dport (cxlmd, root); +if (!dport || !dport->regs.rcd_pcie_cap) +return -ENXIO ; +switch (width) {case 2 : return sysfs_emit(buf, "%#x\n",
Signature · 1 line
2.54.0
[email protected]
AI reviewRe: [PATCH v4 2/3] cxl/core: Hold the dport host lock across dport lookup and use
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [High] cxl/core: AB-BA deadlock in RCH topologies due to child-to-parent lock inversion --
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] cxl/core: AB-BA deadlock in RCH topologies due to child-to-parent lock inversion --
commit 689adbac9a3f5c00d58408df886ec90fd00ca46d
Author: Dave Jiang <[email protected]>
cxl/core: Hold the dport host lock across dport lookup and use
Callers of cxl_pci_find_port() and cxl_mem_find_port() can be handed a struct cxl_dport that is already freed, and all four dereference it without the lock that would prevent that.
A dport is kzalloc()'d into a devres group on its host device, so it dies when that host unbinds. The host is &port->dev for a switch port and port->uport_dev for the root port.
Quoted 5 lines
drivers/cxl/core/pci.c +0
-0
drivers/cxl/core/pci.c int cxl_pci_setup_regs(struct pci_dev *pdev, enum cxl_regloc_type type, · line 741
Quoted 9 lines
drivers/cxl/core/pci.c +2
-1@@ -734,10 +734,16 @@ int cxl_pci_setup_regs(struct pci_dev *pdev, enum cxl_regloc_type type,if (rc && type ==CXL_REGLOC_RBI_COMPONENT &&is_cxl_restricted (pdev)) {struct cxl_dport *dport; struct cxl_port *port __free(put_cxl_port) = -cxl_pci_find_port (pdev, &dport); +cxl_pci_find_port (pdev);if (!port)return -EPROBE_DEFER ; +guard (device)(cxl_port_dport_host (port));
[Severity: High]
Does acquiring the ancestor host bridge lock here while the driver core holds the descendant (pdev) device lock cause an AB-BA deadlock?
If concurrent device removal locks the parent device and then iterates to lock children, could it deadlock against this path in cxl_pci_setup_regs() during device probe?
Quoted 8 lines
drivers/cxl/core/pci.c +5
-0+ + dport =cxl_pci_find_dport (pdev, port); +if (!dport) +return -EPROBE_DEFER ; + rc =cxl_rcrb_get_comp_regs (pdev, map, dport);if (rc)return rc;
[ ... ]
Quoted 8 lines
drivers/cxl/core/ras_rch.c +0
-0@@ -118,10 +118,20 @@ void cxl_handle_rdport_errors(struct cxl_dev_state *cxlds)int severity; struct cxl_port *port __free(put_cxl_port) =
drivers/cxl/core/ras_rch.c void cxl_handle_rdport_errors(struct cxl_dev_state *cxlds) · line 129
Quoted 10 lines
drivers/cxl/core/ras_rch.c +6
-1-cxl_pci_find_port (pdev, &dport); +cxl_pci_find_port (pdev);if (!port)return ; +/* +* The dport may have been freed since the lookup above, so take the +* lock that pins it and find it again. +*/ +guard (device)(cxl_port_dport_host (port));
[Severity: High]
Could this also lead to an AB-BA deadlock on Restricted CXL Host (RCH) topologies?
This acquires the ancestor host bridge lock while the caller holds the descendant (&cxlmd->dev) device lock. Concurrent device hot-remove paths typically lock from parent to child, which seems like it would deadlock against this AER handling path.
Quoted 8 lines
drivers/cxl/core/ras_rch.c +5
-0+ + dport =cxl_pci_find_dport (pdev, port); +if (!dport) +return ; +if (!cxl_rch_get_aer_info (dport->regs.dport_aer, &aer_regs))return ;
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=2
Dave Jiang
maintainerRe: [PATCH v4 2/3] cxl/core: Hold the dport host lock across dport lookup and use
Quoted 36 lines · On 10/1/26 9:02 AM, [email protected] wrote:
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] cxl/core: AB-BA deadlock in RCH topologies due to child-to-parent lock inversion --commit
689adbac9a3f5c00d58408df886ec90fd00ca46d
Author: Dave Jiang <[email protected]>cxl/core: Hold the dport host lock across dport lookup and use
Callers of
cxl_pci_find_port()andcxl_mem_find_port()can be handed a structcxl_dportthat is already freed, and all four dereference it without the lock that would prevent that.A dport is kzalloc()'d into a devres group on its host device, so it dies when that host unbinds. The host is
&port->devfor a switch port andport->uport_devfor the root port.Quoted 5 lines
drivers/cxl/core/pci.c +0
-0drivers/cxl/core/pci.c int cxl_pci_setup_regs(struct pci_dev *pdev, enum cxl_regloc_type type, · line 741
Quoted 9 lines
drivers/cxl/core/pci.c +2
-1@@ -734,10 +734,16 @@ int cxl_pci_setup_regs(struct pci_dev *pdev, enum cxl_regloc_type type,if (rc && type ==CXL_REGLOC_RBI_COMPONENT &&is_cxl_restricted (pdev)) {struct cxl_dport *dport; struct cxl_port *port __free(put_cxl_port) = -cxl_pci_find_port (pdev, &dport); +cxl_pci_find_port (pdev);if (!port)return -EPROBE_DEFER ; +guard (device)(cxl_port_dport_host (port));[Severity: High]
Does acquiring the ancestor host bridge lock here while the driver core holds the descendant (pdev) device lock cause an AB-BA deadlock?
Quoted 3 lines
If concurrent device removal locks the parent device and then iterates to lock children, could it deadlock against this path in
cxl_pci_setup_regs()during device probe?
No. The lock taken is not the host bridge and not an ancestor of pdev.
These paths only run for an RCH, where cxl_pci_find_port() returns the CXL root port. cxl_port_dport_host() of a root port is root->uport_dev, the ACPI0017 platform device. The host bridge is only the dport device.
On x86 the PCI root bus is created with a NULL parent, so the RCiEP and its memdev live under /sys/devices/pci0000:xx, and ACPI0017 lives under the platform bus. A parent-to-child walk from either tree never reaches the other.
The order is also not new. cxl_mem_probe() already locks root->uport_dev for an RCH endpoint, and cxl_mem probes synchronously from cxl_pci_probe(). That establishes:
pdev -> cxlmd -> ACPI0017
cxl_pci_setup_regs() adds pdev -> ACPI0017, which that chain already implies.
Nothing takes these locks in the reverse order:
- delete_endpoint() runs from cxlmd devm teardown: cxlmd -> ACPI0017.
- unregister_port() of the endpoint takes only the endpoint's own lock.
- detach_memdev() locks cxlmd from cxl_bus_wq, and nothing waits on that work under the ACPI0017 lock.
- PCI hot-remove goes pdev -> cxlmd -> ACPI0017.
Quoted 35 lines
Quoted 9 lines
drivers/cxl/core/pci.c +5
-0+ + dport =cxl_pci_find_dport (pdev, port); +if (!dport) +return -EPROBE_DEFER ; + rc =cxl_rcrb_get_comp_regs (pdev, map, dport);if (rc)return rc;[ ... ]
Quoted 8 lines
drivers/cxl/core/ras_rch.c +0
-0@@ -118,10 +118,20 @@ void cxl_handle_rdport_errors(struct cxl_dev_state *cxlds)int severity; struct cxl_port *port __free(put_cxl_port) =drivers/cxl/core/ras_rch.c void cxl_handle_rdport_errors(struct cxl_dev_state *cxlds) · line 129
Quoted 10 lines
drivers/cxl/core/ras_rch.c +6
-1-cxl_pci_find_port (pdev, &dport); +cxl_pci_find_port (pdev);if (!port)return ; +/* +* The dport may have been freed since the lookup above, so take the +* lock that pins it and find it again. +*/ +guard (device)(cxl_port_dport_host (port));[Severity: High]
Could this also lead to an AB-BA deadlock on Restricted CXL Host (RCH) topologies?
Quoted 4 lines
This acquires the ancestor host bridge lock while the caller holds the descendant (
&cxlmd->dev) device lock. Concurrent device hot-remove paths typically lock from parent to child, which seems like it would deadlock against this AER handling path.
Same answer. This is cxlmd -> ACPI0017, the pair cxl_mem_probe() takes in the same order. Hot-remove takes them in that order too.
Quoted 10 lines
Quoted 9 lines
drivers/cxl/core/ras_rch.c +5
-0+ + dport =cxl_pci_find_dport (pdev, port); +if (!dport) +return ; +if (!cxl_rch_get_aer_info (dport->regs.dport_aer, &aer_regs))return ;
Dave Jiang
maintainerPATCH v4 3/3 cxl/port: Attach endpoints to a dport under the lock that pins it
In reply to Dave Jiangdevm_cxl_enumerate_ports() and add_port_attach_ep() look up or create a dport under its port's device lock, drop the lock, then pass the dport to cxl_add_ep(). add_ep() reads dport->port before taking the lock, and cxl_gpf_port_setup() reads and writes the dport with no lock held.
A concurrent cxl_detach_ep() for another memdev below the same port can free the dport in that window: it calls del_dports() under the port lock when the last endpoint leaves. The enumerating memdev then dereferences freed memory.
Attach the endpoint and set up GPF while the port lock from the lookup is still held. Make add_ep() assert that lock instead of taking it. Under the lock, a dport found in port->dports is still live, since removal erases it from the xarray before freeing it.
---
Found by review, not from a failure report.
Tested in QEMU with cxl_test, PROVE_LOCKING and KASAN on: 6 module load/unload cycles, 12 concurrent target_list readers, and each cycle 20 rounds of every memdev unbinding and rebinding in parallel. No KASAN report, no WARNING, no lockdep splat; debug_locks stays 1. Both converted paths run under this load, so the new device_lock_assert() in add_ep() never fired.
Enumeration is unchanged against the same test at patch 2: 11 memdev, 8 port, 11 endpoint, 158 decoder after every load.
The races produce transient -EBUSY cxl_mem probe failures (424 with this patch, 407 without) from the pending detach_work check in cxl_mem_probe(). Memdevs left unbound rebind on the first retry.
--- drivers/cxl/core/port.c | 115 ++++++++++++++++++++++------------------ 1 file changed, 62 insertions(+), 53 deletions(-)
drivers/cxl/core/port.c +62 -53
@@ -1349,7 +1349,7 @@ static int add_ep(struct cxl_ep *new) {struct cxl_port *port = new->dport->port; -guard (device)(&port->dev); +device_lock_assert (&port->dev);if (port->dead)return -ENXIO ;
@@ -1365,6 +1365,9 @@ static int add_ep(struct cxl_ep *new)* Intermediate CXL ports are scanned based on the arrival of endpoints. * When those endpoints depart the port can be destroyed once all * endpoints that care about that port have been removed. +* +* Context: Caller must hold the device lock of @dport->port, which pins +* @dport, across the lookup of @dport and this call. */ static int cxl_add_ep (struct cxl_dport *dport,struct device *ep_dev) {
@@ -1695,13 +1698,15 @@ static struct cxl_dport *probe_dport(struct cxl_port *port,return drv->add_dport (port, dport_dev); } -static struct cxl_dport *devm_cxl_create_port (struct device *ep_dev, -struct cxl_port *parent_port, -struct cxl_dport *parent_dport, -struct device *uport_dev, -struct device *dport_dev) +static int devm_cxl_create_port (struct device *ep_dev, +struct cxl_port *parent_port, +struct cxl_dport *parent_dport, +struct device *uport_dev, +struct device *dport_dev) {resource_size_t component_reg_phys; +struct cxl_dport *dport; +int rc;device_lock_assert (&parent_port->dev);if (!parent_port->dev.driver) {
@@ -1718,7 +1723,7 @@ static struct cxl_dport *devm_cxl_create_port(struct device *ep_dev, port =devm_cxl_add_port (&parent_port->dev, uport_dev, component_reg_phys, parent_dport);if (IS_ERR (port)) -return ERR_CAST (port); +return PTR_ERR (port);/* * retry to make sure a port is found. a port device
@@ -1726,7 +1731,7 @@ static struct cxl_dport *devm_cxl_create_port(struct device *ep_dev,*/ port =find_cxl_port_by_uport (uport_dev);if (!port) -return ERR_PTR (-ENODEV ); +return -ENODEV ;dev_dbg (ep_dev,"created port %s:%s\n" ,dev_name (&port->dev),dev_name (port->uport_dev));
@@ -1735,11 +1740,25 @@ static struct cxl_dport *devm_cxl_create_port(struct device *ep_dev,* Port was created before right before this function is * called. Signal the caller to deal with it. */ -return ERR_PTR (-EAGAIN ); +return -EAGAIN ; } +/* @dport is only valid while @port's lock is held */ guard (device)(&port->dev); -return probe_dport (port, dport_dev); + dport =probe_dport (port, dport_dev); +if (IS_ERR (dport)) +return PTR_ERR (dport); + + rc =cxl_add_ep (dport, ep_dev); +if (rc == -EBUSY ) { +/* +* "can't" happen, but this error code means +* something to the caller, so translate it. +*/ + rc = -ENXIO ; + } + +return rc; } static int add_port_attach_ep(struct cxl_memdev *cxlmd,
@@ -1747,7 +1766,7 @@ static int add_port_attach_ep(struct cxl_memdev *cxlmd, struct device *dport_dev) {struct device *dparent =grandparent (dport_dev); -struct cxl_dport *dport, *parent_dport; +struct cxl_dport *parent_dport;int rc;if (is_cxl_host_bridge (dparent)) {
@@ -1776,33 +1795,23 @@ static int add_port_attach_ep(struct cxl_memdev *cxlmd,return PTR_ERR (parent_dport); } - dport =devm_cxl_create_port (&cxlmd->dev, parent_port, - parent_dport, uport_dev, - dport_dev); -if (IS_ERR (dport)) { -/* Port or dport already exists, restart iteration */ -if (PTR_ERR (dport) == -EAGAIN ||PTR_ERR (dport) == -EBUSY ) -return 0 ; -return PTR_ERR (dport); - } + rc =devm_cxl_create_port (&cxlmd->dev, parent_port, + parent_dport, uport_dev, dport_dev); } - rc =cxl_add_ep (dport, &cxlmd->dev); -if (rc == -EBUSY ) { -/* -* "can't" happen, but this error code means -* something to the caller, so translate it. -*/ - rc = -ENXIO ; - } +/* Port or dport already exists, restart iteration */ +if (rc == -EAGAIN || rc == -EBUSY ) +return 0 ;return rc; } -static struct cxl_dport *find_or_add_dport (struct cxl_port *port, -struct device *dport_dev) +static int find_or_add_dport_attach_ep (struct cxl_memdev *cxlmd, +struct cxl_port *port, +struct device *dport_dev) {struct cxl_dport *dport; +int rc;/* * The port is already visible in CXL hierarchy, but it may still
@@ -1818,13 +1827,28 @@ static struct cxl_dport *find_or_add_dport(struct cxl_port *port,if (!dport) { dport =probe_dport (port, dport_dev);if (IS_ERR (dport)) -return dport; +return PTR_ERR (dport);/* New dport added, restart iteration */ -return ERR_PTR (-EAGAIN ); +return -EAGAIN ; } -return dport; +/* @dport is only valid while @port's lock is held */ + rc =cxl_add_ep (dport, &cxlmd->dev); + +/* +* If the endpoint already exists in the port's list, +* that's ok, it was added on a previous pass. +* Otherwise, retry in add_port_attach_ep() after taking +* the parent_port lock as the current port may be being +* reaped. +*/ +if (rc && rc != -EBUSY ) +return rc; + +cxl_gpf_port_setup (dport); + +return 0 ; }int devm_cxl_enumerate_ports (struct cxl_memdev *cxlmd)
@@ -1853,7 +1877,6 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)for (iter = dev; iter; iter =grandparent (iter)) {struct device *dport_dev =grandparent (iter);struct device *uport_dev; -struct cxl_dport *dport;if (is_cxl_host_bridge (dport_dev))return 0 ;
@@ -1880,27 +1903,13 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)* RP port enumerated by cxl_acpi without dport will * have the dport added here. */ - dport =find_or_add_dport (port, dport_dev); -if (IS_ERR (dport)) { -if (PTR_ERR (dport) == -EAGAIN ) - goto retry; -return PTR_ERR (dport); - } - - rc =cxl_add_ep (dport, &cxlmd->dev); - -/* -* If the endpoint already exists in the port's list, -* that's ok, it was added on a previous pass. -* Otherwise, retry in add_port_attach_ep() after taking -* the parent_port lock as the current port may be being -* reaped. -*/ -if (rc && rc != -EBUSY ) + rc =find_or_add_dport_attach_ep (cxlmd, port, + dport_dev); +if (rc == -EAGAIN ) + goto retry; +if (rc)return rc; -cxl_gpf_port_setup (dport); -/* Any more ports to add between this one and the root? */ if (!dev_is_cxl_root_child (&port->dev))continue ;
Signature · 1 line
2.54.0
Re: [PATCH v4 3/3] cxl/port: Attach endpoints to a dport under the lock that pins it
Quoted 15 lines · Dave Jiang
devm_cxl_enumerate_ports()andadd_port_attach_ep()look up or create a dport under its port's device lock, drop the lock, then pass the dport tocxl_add_ep().add_ep()readsdport->portbefore taking the lock, andcxl_gpf_port_setup()reads and writes the dport with no lock held.A concurrent
cxl_detach_ep()for another memdev below the same port can free the dport in that window: it callsdel_dports()under the port lock when the last endpoint leaves. The enumerating memdev then dereferences freed memory.Attach the endpoint and set up GPF while the port lock from the lookup is still held. Make
add_ep()assert that lock instead of taking it. Under the lock, a dport found inport->dportsis still live, since removal erases it from the xarray before freeing it.
Quoted 7 lines
Fixes de516b40116e ("cxl/port: Record dport in endpoint references")Reported-by Li Ming <[email protected]>Assisted-by LLMSigned-off-by Dave Jiang <[email protected]>Reviewed-by Li Ming <[email protected]>Reviewed-by Alison Schofield <[email protected]>
second look, see inline...
Quoted 34 lines
---
Found by review, not from a failure report.Tested in QEMU with
cxl_test,PROVE_LOCKINGand KASAN on: 6 module load/unload cycles, 12 concurrenttarget_listreaders, and each cycle 20 rounds of every memdev unbinding and rebinding in parallel. No KASAN report, no WARNING, no lockdep splat;debug_locksstays 1. Both converted paths run under this load, so the newdevice_lock_assert()inadd_ep()never fired.Enumeration is unchanged against the same test at patch 2: 11 memdev, 8 port, 11 endpoint, 158 decoder after every load.
The races produce transient
-EBUSYcxl_memprobe failures (424 with this patch, 407 without) from the pendingdetach_workcheck incxl_mem_probe(). Memdevs left unbound rebind on the first retry.--- drivers/cxl/core/port.c | 115 ++++++++++++++++++++++------------------ 1 file changed, 62 insertions(+), 53 deletions(-)drivers/cxl/core/port.c +1
-1@@ -1349,7 +1349,7 @@ static int add_ep(struct cxl_ep *new) {struct cxl_port *port = new->dport->port; -guard (device)(&port->dev); +device_lock_assert (&port->dev);if (port->dead)return -ENXIO ;
drivers/cxl/core/port.c static int add_ep(struct cxl_ep *new) · line 1371
Quoted 7 lines
drivers/cxl/core/port.c +3
-0@@ -1365,6 +1365,9 @@ static int add_ep(struct cxl_ep *new)* Intermediate CXL ports are scanned based on the arrival of endpoints. * When those endpoints depart the port can be destroyed once all * endpoints that care about that port have been removed. +* +* Context: Caller must hold the device lock of @dport->port, which pins +* @dport, across the lookup of @dport and this call. */
I found this a bit confusing, since patch 2 establishes that root port dports are pinned by uport_dev, rather than &port->dev. As written this sounds like a general locking rule, but it's only true here because cxl_add_ep() is never called with a root port. (and that took me a bit to figure out)
How about appending that to the above kernel-doc comment like this:
* Context: Caller must hold the device lock of @dport->port, which pins
* @dport, across the lookup of @dport and this call. @dport->port * is never the CXL root port, whose dports are pinned by its * uport_dev instead.
Maybe you can do that upon applying -
Alison Schofield
maintainerQuoted 2 lines · Dave Jiang
Couple dport fixes resulted from sashiko pre-existing issue reports. There are no depedencies between the two patches.
Hi Dave,
Above is stale. Now 3 depends on 2. Doesn't matter at this point ;) I did come up w a new review comment on Patch3, but I think you may be able to fix it up upon applying if you agree.
-- Alison
Quoted 31 lines
v4:
- Drop the unused dport output parameter in patch 2. (Sashiko, Alison)
- Add review tags from Alison and Ming.v3:
- Add patch 3 to attach endpoints under the dport's port lock. (Ming)
- Add review tags from Ming and Gregory to patch 1.v2:
- Addressed couple things from Jonathan. See patch for revlog.Dave Jiang (3):
cxl/port: Clear cached dport pointers when a dport is removed cxl/core: Hold the dport host lock across dport lookup and use cxl/port: Attach endpoints to a dport under the lock that pins itdrivers/cxl/core/core.h | 6 +- drivers/cxl/core/pci.c | 8 +- drivers/cxl/core/port.c | 206 ++++++++++++++++++++++++------------- drivers/cxl/core/ras_rch.c | 12 ++- drivers/cxl/cxl.h | 22 +++- drivers/cxl/mem.c | 14 ++- drivers/cxl/pci.c | 11 +- 7 files changed, 186 insertions(+), 93 deletions(-)base-commit:
fd73f4a6659897191fa0d40695fe370925dd3780Signature · 1 line
2.54.0
Dave Jiang
maintainerQuoted 29 lines · Dave Jiang
Couple dport fixes resulted from sashiko pre-existing issue reports. There are no depedencies between the two patches.
v4:
- Drop the unused dport output parameter in patch 2. (Sashiko, Alison)
- Add review tags from Alison and Ming.v3:
- Add patch 3 to attach endpoints under the dport's port lock. (Ming)
- Add review tags from Ming and Gregory to patch 1.v2:
- Addressed couple things from Jonathan. See patch for revlog.Dave Jiang (3):
cxl/port: Clear cached dport pointers when a dport is removed cxl/core: Hold the dport host lock across dport lookup and use cxl/port: Attach endpoints to a dport under the lock that pins itdrivers/cxl/core/core.h | 6 +- drivers/cxl/core/pci.c | 8 +- drivers/cxl/core/port.c | 206 ++++++++++++++++++++++++------------- drivers/cxl/core/ras_rch.c | 12 ++- drivers/cxl/cxl.h | 22 +++- drivers/cxl/mem.c | 14 ++- drivers/cxl/pci.c | 11 +- 7 files changed, 186 insertions(+), 93 deletions(-)
Quoted 1 line
base-commit:
fd73f4a6659897191fa0d40695fe370925dd3780
Applied to cxl/nextbcda00c78d07b72a03ebd958019560d23450