* [PATCH v3 1/2] PCI: Add pci=noaer_recovery kernel boot option
2026-10-22 16:58 [PATCH v3 0/2] PCI: Add boot options to disable DPC and AER recovery Yury Murashka
@ 2026-10-22 16:58 ` Yury Murashka
2026-10-22 16:58 ` [PATCH v3 2/2] PCI: Add pci=nodpc " Yury Murashka
1 sibling, 0 replies; 3+ messages in thread
From: Yury Murashka @ 2026-10-22 16:58 UTC (permalink / raw)
To: bhelgaas
Cc: rafael, lenb, corbet, skhan, rdunlap, mahesh, oohall, linux-pci,
linux-acpi, linux-doc, linux-kernel, linuxppc-dev
On large modular systems with a complex PCIe topology, allow userspace to
inspect uncorrectable AER errors before Linux attempts recovery.
Add pci=noaer_recovery to retain AER logging while skipping recovery for
uncorrectable errors. Check the option before dispatching either CXL RCH
internal errors or ordinary PCIe errors so CXL driver error callbacks
cannot bypass it. Also skip recovery for errors reported through GHES,
while retaining error logging and releasing error buffers and device
references.
Correctable errors continue to be handled normally. Skipping recovery
avoids its clearing of uncorrectable error status, but does not suppress
Root Port interrupt acknowledgment or change firmware error handling.
The option is AER-specific and does not disable DPC recovery.
Signed-off-by: Yury Murashka <yurypm@arista.com>
---
Changes in v3:
- Check the option before both CXL RCH and ordinary PCIe error dispatch.
- Preserve logging, correctable handling, and current GHES cleanup.
- Clarify the AER-only scope and rebase onto PCI next.
v2: https://lore.kernel.org/linux-pci/20260709185429.627968-2-yurypm@arista.com/
Review: https://lore.kernel.org/linux-pci/20260709190616.DF0CF1F000E9@smtp.kernel.org/
Documentation/admin-guide/kernel-parameters.txt | 5 +++++
drivers/pci/pci.c | 2 ++
drivers/pci/pci.h | 2 ++
drivers/pci/pcie/aer.c | 13 +++++++++++--
4 files changed, 20 insertions(+), 2 deletions(-)
diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 68647ff4b..6047efd3c 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -5118,6 +5118,11 @@ Kernel parameters
noaer [PCIE] If the PCIEAER kernel config parameter is
enabled, this kernel boot option can be used to
disable the use of PCIE advanced error reporting.
+ noaer_recovery [PCIE] Disable Linux AER recovery for
+ uncorrectable errors, including CXL RCH errors,
+ while retaining AER logging. Correctable errors
+ are handled normally. Requires CONFIG_PCIEAER.
+ Does not disable DPC or firmware error recovery.
nodomains [PCI] Disable support for multiple PCI
root domains (aka PCI segments, in ACPI-speak).
nommconf [X86] Disable use of MMCONFIG for PCI
diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
index a62af7320..9ca8d3fbb 100644
--- a/drivers/pci/pci.c
+++ b/drivers/pci/pci.c
@@ -6840,6 +6840,8 @@ static int __init pci_setup(char *str)
pcie_ats_disabled = true;
} else if (!strcmp(str, "noaer")) {
pci_no_aer();
+ } else if (!strcmp(str, "noaer_recovery")) {
+ pci_no_aer_recovery();
} else if (!strcmp(str, "earlydump")) {
pci_early_dump = true;
} else if (!strncmp(str, "realloc=", 8)) {
diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h
index 77efa4c5c..ccc6a678c 100644
--- a/drivers/pci/pci.h
+++ b/drivers/pci/pci.h
@@ -1337,6 +1337,7 @@ static inline void of_pci_remove_host_bridge_node(struct pci_host_bridge *bridge
#ifdef CONFIG_PCIEAER
void pci_no_aer(void);
+void pci_no_aer_recovery(void);
void pci_aer_init(struct pci_dev *dev);
void pci_aer_exit(struct pci_dev *dev);
extern const struct attribute_group aer_stats_attr_group;
@@ -1348,6 +1349,7 @@ void pci_save_aer_state(struct pci_dev *dev);
void pci_restore_aer_state(struct pci_dev *dev);
#else
static inline void pci_no_aer(void) { }
+static inline void pci_no_aer_recovery(void) { }
static inline void pci_aer_init(struct pci_dev *d) { }
static inline void pci_aer_exit(struct pci_dev *d) { }
static inline void pci_aer_clear_fatal_status(struct pci_dev *dev) { }
diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
index c376c3adb..c41e5e13f 100644
--- a/drivers/pci/pcie/aer.c
+++ b/drivers/pci/pcie/aer.c
@@ -142,6 +142,7 @@ struct aer_info {
PCI_ERR_ROOT_MULTI_UNCOR_RCV)
static bool pcie_aer_disable;
+static bool pcie_aer_recovery_disable;
static pci_ers_result_t aer_root_reset(struct pci_dev *dev);
void pci_no_aer(void)
@@ -149,6 +150,11 @@ void pci_no_aer(void)
pcie_aer_disable = true;
}
+void pci_no_aer_recovery(void)
+{
+ pcie_aer_recovery_disable = true;
+}
+
bool pci_aer_available(void)
{
return !pcie_aer_disable && pci_msi_enabled();
@@ -1338,6 +1344,9 @@ static void pci_aer_handle_error(struct pci_dev *dev, struct aer_err_info *info)
static void handle_error_source(struct pci_dev *dev, struct aer_err_info *info)
{
+ if (pcie_aer_recovery_disable && info->severity != AER_CORRECTABLE)
+ return;
+
cxl_rch_handle_error(dev, info);
pci_aer_handle_error(dev, info);
}
@@ -1383,15 +1392,15 @@ static void aer_recover_work_func(struct work_struct *work)
ghes_estatus_pool_region_free((unsigned long)entry.regs,
sizeof(struct aer_capability_regs));
- if (pdev) {
+ if (pdev && !pcie_aer_recovery_disable) {
if (entry.severity == AER_NONFATAL)
pcie_do_recovery(pdev, pci_channel_io_normal,
aer_root_reset);
else if (entry.severity == AER_FATAL)
pcie_do_recovery(pdev, pci_channel_io_frozen,
aer_root_reset);
- pci_dev_put(pdev);
}
+ pci_dev_put(pdev);
}
}
--
2.51.0
^ permalink raw reply related [flat|nested] 3+ messages in thread* [PATCH v3 2/2] PCI: Add pci=nodpc kernel boot option
2026-10-22 16:58 [PATCH v3 0/2] PCI: Add boot options to disable DPC and AER recovery Yury Murashka
2026-10-22 16:58 ` [PATCH v3 1/2] PCI: Add pci=noaer_recovery kernel boot option Yury Murashka
@ 2026-10-22 16:58 ` Yury Murashka
1 sibling, 0 replies; 3+ messages in thread
From: Yury Murashka @ 2026-10-22 16:58 UTC (permalink / raw)
To: bhelgaas
Cc: rafael, lenb, corbet, skhan, rdunlap, mahesh, oohall, linux-pci,
linux-acpi, linux-doc, linux-kernel, linuxppc-dev
Add pci=nodpc to let Linux opt out of Downstream Port Containment (DPC)
and Error Disconnect Recover (EDR) support on systems where userspace
manages PCIe errors.
Skip DPC initialization, state operations, hotplug recovery checks, and
port service registration when disabled. Expose pci_dpc_available() so
ACPI negotiation and PCIe port service discovery honor the same policy.
Do not advertise EDR support or request DPC control through _OSC, and
prevent native-port overrides from reintroducing DPC ownership.
Skip EDR notifier installation before the DPC-enabling _DSM call. Merely
skipping the native DPC driver would allow EDR to enable hardware DPC
while the handler lacks the capability information needed to recover.
The option disables Linux participation in DPC and EDR; it does not turn
off firmware-owned DPC hardware. AER reporting and recovery remain
independent and can be controlled with their own boot options.
Signed-off-by: Yury Murashka <yurypm@arista.com>
---
Changes in v3:
- Add pci_dpc_available() and use it for ACPI DPC/EDR negotiation.
- Skip EDR notifier registration and its DPC-enabling _DSM call.
- Gate DPC service discovery and prevent forced native DPC ownership.
- Clarify the firmware boundary and rebase onto PCI next.
v2: https://lore.kernel.org/linux-pci/20260709185429.627968-3-yurypm@arista.com/
Review: https://lore.kernel.org/linux-pci/20260709190730.E19A11F000E9@smtp.kernel.org/
.../admin-guide/kernel-parameters.txt | 4 ++++
drivers/acpi/pci_root.c | 9 +++++--
drivers/pci/pci.c | 2 ++
drivers/pci/pci.h | 2 ++
drivers/pci/pcie/dpc.c | 24 ++++++++++++++++---
drivers/pci/pcie/edr.c | 5 +++-
drivers/pci/pcie/portdrv.c | 3 ++-
include/linux/pci.h | 6 +++++
8 files changed, 48 insertions(+), 7 deletions(-)
diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 6047efd3c..f98486d4d 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -5115,6 +5115,10 @@ Kernel parameters
through ports 0xC000-0xCFFF).
See http://wiki.osdev.org/PCI for more info
on the configuration access mechanisms.
+ nodpc [PCIE] Disable Linux DPC and EDR support. Linux
+ neither requests DPC control from firmware nor
+ enables DPC through EDR. Firmware-owned DPC
+ settings are not changed. Requires CONFIG_PCIE_DPC.
noaer [PCIE] If the PCIEAER kernel config parameter is
enabled, this kernel boot option can be used to
disable the use of PCIE advanced error reporting.
diff --git a/drivers/acpi/pci_root.c b/drivers/acpi/pci_root.c
index a1a9c0601..6bca8206c 100644
--- a/drivers/acpi/pci_root.c
+++ b/drivers/acpi/pci_root.c
@@ -438,7 +438,7 @@ static u32 calculate_support(void)
support |= OSC_PCI_ASPM_SUPPORT | OSC_PCI_CLOCK_PM_SUPPORT;
if (pci_msi_enabled())
support |= OSC_PCI_MSI_SUPPORT;
- if (IS_ENABLED(CONFIG_PCIE_EDR))
+ if (IS_ENABLED(CONFIG_PCIE_EDR) && pci_dpc_available())
support |= OSC_PCI_EDR_SUPPORT;
return support;
@@ -517,7 +517,7 @@ static u32 calculate_control(void)
* OSC_PCI_EXPRESS_DPC_CONTROL indicates the OS supports both DPC
* and EDR.
*/
- if (IS_ENABLED(CONFIG_PCIE_DPC) && IS_ENABLED(CONFIG_PCIE_EDR))
+ if (pci_dpc_available() && IS_ENABLED(CONFIG_PCIE_EDR))
control |= OSC_PCI_EXPRESS_DPC_CONTROL;
return control;
@@ -1064,6 +1064,11 @@ struct pci_bus *acpi_pci_root_create(struct acpi_pci_root *root,
else if (pcie_ports_dpc_native)
override = OSC_PCI_EXPRESS_DPC_CONTROL & ~ctrl;
+ if (!pci_dpc_available()) {
+ ctrl &= ~OSC_PCI_EXPRESS_DPC_CONTROL;
+ override &= ~OSC_PCI_EXPRESS_DPC_CONTROL;
+ }
+
if (override) {
decode_osc_control(root, pcie_ports_native ?
"OS forcing control (\"pcie_ports=native\") of" :
diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
index 9ca8d3fbb..c2a6aacc5 100644
--- a/drivers/pci/pci.c
+++ b/drivers/pci/pci.c
@@ -6838,6 +6838,8 @@ static int __init pci_setup(char *str)
} else if (!strncmp(str, "noats", 5)) {
pr_info("PCIe: ATS is disabled\n");
pcie_ats_disabled = true;
+ } else if (!strcmp(str, "nodpc")) {
+ pci_no_dpc();
} else if (!strcmp(str, "noaer")) {
pci_no_aer();
} else if (!strcmp(str, "noaer_recovery")) {
diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h
index ccc6a678c..ae8fc06e5 100644
--- a/drivers/pci/pci.h
+++ b/drivers/pci/pci.h
@@ -934,6 +934,7 @@ struct rcec_ea {
#endif
#ifdef CONFIG_PCIE_DPC
+void pci_no_dpc(void);
void pci_save_dpc_state(struct pci_dev *dev);
void pci_restore_dpc_state(struct pci_dev *dev);
void pci_dpc_init(struct pci_dev *pdev);
@@ -942,6 +943,7 @@ pci_ers_result_t dpc_reset_link(struct pci_dev *pdev);
bool pci_dpc_recovered(struct pci_dev *pdev);
unsigned int dpc_tlp_log_len(struct pci_dev *dev);
#else
+static inline void pci_no_dpc(void) { }
static inline void pci_save_dpc_state(struct pci_dev *dev) { }
static inline void pci_restore_dpc_state(struct pci_dev *dev) { }
static inline void pci_dpc_init(struct pci_dev *pdev) { }
diff --git a/drivers/pci/pcie/dpc.c b/drivers/pci/pcie/dpc.c
index 2b779bd1d..fbafc3ca0 100644
--- a/drivers/pci/pcie/dpc.c
+++ b/drivers/pci/pcie/dpc.c
@@ -43,12 +43,24 @@ static const char * const rp_pio_error_string[] = {
"Memory Request Completion Timeout", /* Bit Position 18 */
};
+static bool pcie_dpc_disable;
+
+void pci_no_dpc(void)
+{
+ pcie_dpc_disable = true;
+}
+
+bool pci_dpc_available(void)
+{
+ return !pcie_dpc_disable;
+}
+
void pci_save_dpc_state(struct pci_dev *dev)
{
struct pci_cap_saved_state *save_state;
u16 *cap;
- if (!pci_is_pcie(dev))
+ if (!pci_dpc_available() || !pci_is_pcie(dev))
return;
save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_DPC);
@@ -64,7 +76,7 @@ void pci_restore_dpc_state(struct pci_dev *dev)
struct pci_cap_saved_state *save_state;
u16 *cap;
- if (!pci_is_pcie(dev))
+ if (!pci_dpc_available() || !pci_is_pcie(dev))
return;
save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_DPC);
@@ -104,7 +116,7 @@ bool pci_dpc_recovered(struct pci_dev *pdev)
{
struct pci_host_bridge *host;
- if (!pdev->dpc_cap)
+ if (!pci_dpc_available() || !pdev->dpc_cap)
return false;
/*
@@ -404,6 +416,9 @@ void pci_dpc_init(struct pci_dev *pdev)
{
u16 cap;
+ if (!pci_dpc_available())
+ return;
+
pdev->dpc_cap = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_DPC);
if (!pdev->dpc_cap)
return;
@@ -532,5 +547,8 @@ static struct pcie_port_service_driver dpcdriver = {
int __init pcie_dpc_init(void)
{
+ if (!pci_dpc_available())
+ return 0;
+
return pcie_port_service_register(&dpcdriver);
}
diff --git a/drivers/pci/pcie/edr.c b/drivers/pci/pcie/edr.c
index e86298dbb..8567dd1d5 100644
--- a/drivers/pci/pcie/edr.c
+++ b/drivers/pci/pcie/edr.c
@@ -223,6 +223,9 @@ void pci_acpi_add_edr_notifier(struct pci_dev *pdev)
struct acpi_device *adev = ACPI_COMPANION(&pdev->dev);
acpi_status status;
+ if (!pci_dpc_available())
+ return;
+
if (!adev) {
pci_dbg(pdev, "No valid ACPI node, skipping EDR init\n");
return;
@@ -246,7 +249,7 @@ void pci_acpi_remove_edr_notifier(struct pci_dev *pdev)
{
struct acpi_device *adev = ACPI_COMPANION(&pdev->dev);
- if (!adev)
+ if (!pci_dpc_available() || !adev)
return;
acpi_remove_notify_handler(adev->handle, ACPI_SYSTEM_NOTIFY,
diff --git a/drivers/pci/pcie/portdrv.c b/drivers/pci/pcie/portdrv.c
index 32fc623dd..aa1c57e33 100644
--- a/drivers/pci/pcie/portdrv.c
+++ b/drivers/pci/pcie/portdrv.c
@@ -262,7 +262,8 @@ static int get_port_device_capability(struct pci_dev *dev)
* With dpc-native, allow Linux to use DPC even if it doesn't have
* permission to use AER.
*/
- if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_DPC) &&
+ if (pci_dpc_available() &&
+ pci_find_ext_capability(dev, PCI_EXT_CAP_ID_DPC) &&
pci_aer_available() &&
(pcie_ports_dpc_native || host->native_aer))
services |= PCIE_PORT_SERVICE_DPC;
diff --git a/include/linux/pci.h b/include/linux/pci.h
index f3e9c6c9c..dd667223b 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -1984,6 +1984,12 @@ bool pci_aer_available(void);
static inline bool pci_aer_available(void) { return false; }
#endif
+#ifdef CONFIG_PCIE_DPC
+bool pci_dpc_available(void);
+#else
+static inline bool pci_dpc_available(void) { return false; }
+#endif
+
bool pci_ats_disabled(void);
#define PCIE_PTM_CONTEXT_UPDATE_AUTO 0
--
2.51.0
^ permalink raw reply related [flat|nested] 3+ messages in thread