Re: [v2,08/11] drm/xe/xe_ras: Add support for Uncorrectable Core-Compute errors

["Tauro, Riana" <riana.tauro@intel.com>]

On 3/6/2026 9:20 AM, Purkait, Soham wrote:
> Hi Riana,
>
> On 02-03-2026 15:52, Riana Tauro wrote:
>> Uncorrectable Core-Compute errors are classified into Global and Local
>> errors.
>>
>> Global error is an error that affects the entire device requiring a
>> reset. This type of error is not isolated. When an AER is reported and
>> error_detected is invoked return PCI_ERS_RESULT_NEED_RESET.
>>
>> A Local error is confined to a specific component or context like a
>> engine. These errors can be contained and recovered by resetting
>> only the affected part without distrupting the rest of the device.
>>
>> Upon detection of an Uncorrectable Local Core-Compute error, an AER is
>> generated and GuC is notified of the error. The KMD then sets
>> the context as non-runnable and initiates an engine reset.
>> (TODO: GuC <->KMD communication for the error).
>> Since the error is contained and recovered, PCI error handling
>> callback returns PCI_ERS_RESULT_RECOVERED.
>>
>> Signed-off-by: Riana Tauro <riana.tauro@intel.com>
>> ---
>> v2: add newline and fix log
>>      add bounds check (Mallesh)
>>      add ras specific enum (Raag)
>>      helper for sysctrl prepare command
>>      process all errors before deciding recovery action
>> ---
>>   drivers/gpu/drm/xe/xe_ras.c       | 139 ++++++++++++++++++++++++++++++
>>   drivers/gpu/drm/xe/xe_ras.h       |   3 +
>>   drivers/gpu/drm/xe/xe_ras_types.h |  16 ++++
>>   3 files changed, 158 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/xe/xe_ras.c b/drivers/gpu/drm/xe/xe_ras.c
>> index 3bef589082d7..61c01a4bfadb 100644
>> --- a/drivers/gpu/drm/xe/xe_ras.c
>> +++ b/drivers/gpu/drm/xe/xe_ras.c
>> @@ -4,7 +4,14 @@
>>    */
>>     #include "xe_device_types.h"
>> +#include "xe_printk.h"
>>   #include "xe_ras.h"
>> +#include "xe_ras_types.h"
>> +#include "xe_sysctrl_mailbox.h"
>> +#include "xe_sysctrl_mailbox_types.h"
>> +
>> +#define COMPUTE_ERROR_SEVERITY_MASK        GENMASK(26, 25)
>> +#define GLOBAL_UNCORR_ERROR            2
>>     /* Severity classification of detected errors */
>>   enum xe_ras_severity {
>> @@ -62,6 +69,138 @@ static inline const char *comp_to_str(struct 
>> xe_device *xe, u32 comp)
>>       return xe_ras_components[comp];
>>   }
>>   +static void log_ras_error(struct xe_device *xe, struct 
>> xe_ras_error_class *error_class)
>> +{
>> +    struct xe_ras_error_common common_info = error_class->common;
>> +    struct xe_ras_error_product product_info = error_class->product;
>> +    u8 tile = product_info.unit.tile;
>> +    u32 instance = product_info.unit.instance;
>> +    u32 cause = product_info.error_cause.cause;
>> +
>> +    xe_err(xe, "[RAS]: Tile%u, Instance %u, %s %s Error detected 
>> Cause: 0x%x\n",
>> +           tile, instance, severity_to_str(xe, common_info.severity),
>> +           comp_to_str(xe, common_info.component), cause);
>> +}
>> +
>> +static enum xe_ras_recovery_action handle_compute_errors(struct 
>> xe_device *xe,
>> +                             struct xe_ras_error_array *arr)
>> +{
>> +    struct xe_ras_compute_error *error_info = (struct 
>> xe_ras_compute_error *)arr->error_details;
>> +    u8 uncorr_type;
>> +
>> +    uncorr_type = FIELD_GET(COMPUTE_ERROR_SEVERITY_MASK, 
>> error_info->error_log_header);
>> +    log_ras_error(xe, &arr->error_class);
>> +
>> +    xe_err(xe, "[RAS]: Core Compute Error: timestamp %llu 
>> Uncorrected error type %u\n",
>> +           arr->timestamp, uncorr_type);
>> +
>> +    /* Request a RESET if error is global */
>> +    if (uncorr_type == GLOBAL_UNCORR_ERROR)
>> +        return XE_RAS_RECOVERY_ACTION_RESET;
>> +
>> +    /* Local errors are recovered using a engine reset */
>> +    return XE_RAS_RECOVERY_ACTION_RECOVERED;
>> +}
>> +
>> +static void xe_ras_prepare_sysctrl_command(struct 
>> xe_sysctrl_mailbox_command *command,
>
> You can drop prefix for static functions.

Sure will fix this

Thanks
Riana

>
> Thanks,
> Soham
>
>> +                       u32 cmd_mask, void *request, size_t request_len,
>> +                       void *response, size_t response_len)
>> +{
>> +    struct xe_sysctrl_mailbox_app_msg_hdr hdr = {0};
>> +    u32 req_hdr;
>> +
>> +    req_hdr = FIELD_PREP(APP_HDR_GROUP_ID_MASK, 
>> XE_SYSCTRL_GROUP_GFSP) |
>> +          FIELD_PREP(APP_HDR_COMMAND_MASK, cmd_mask);
>> +
>> +    hdr.data = req_hdr;
>> +    command->header = hdr;
>> +    command->data_in = request;
>> +    command->data_in_len = request_len;
>> +    command->data_out = response;
>> +    command->data_out_len = response_len;
>> +}
>> +
>> +/**
>> + * xe_ras_process_errors - Process and contain hardware errors
>> + * @xe: xe device instance
>> + *
>> + * Get error details from system controller and return recovery
>> + * method. Called only from PCI error handling.
>> + *
>> + * Returns: recovery action to be taken
>> + */
>> +enum xe_ras_recovery_action xe_ras_process_errors(struct xe_device *xe)
>> +{
>> +    struct xe_sysctrl_mailbox_command command = {0};
>> +    struct xe_ras_get_error_response response;
>> +    enum xe_ras_recovery_action final_action;
>> +    size_t rlen;
>> +    int ret;
>> +
>> +    /* Default action */
>> +    final_action = XE_RAS_RECOVERY_ACTION_RECOVERED;
>> +
>> +    if (!xe->info.has_sysctrl)
>> +        return XE_RAS_RECOVERY_ACTION_RESET;
>> +
>> +    xe_ras_prepare_sysctrl_command(&command, 
>> XE_SYSCTRL_CMD_GET_SOC_ERROR, NULL, 0,
>> +                       &response, sizeof(response));
>> +
>> +    do {
>> +        memset(&response, 0, sizeof(response));
>> +        rlen = 0;
>> +
>> +        ret = xe_sysctrl_send_command(xe, &command, &rlen);
>> +        if (ret || !rlen) {
>> +            xe_err(xe, "[RAS]: Sysctrl error ret %d\n", ret);
>> +            goto err;
>> +        }
>> +
>> +        if (rlen != sizeof(response)) {
>> +            xe_err(xe, "[RAS]: Sysctrl response does not match 
>> len!!\n");
>> +            goto err;
>> +        }
>> +
>> +        if (response.num_errors > XE_RAS_NUM_ERROR_ARR) {
>> +            xe_err(xe, "[RAS]: Number of errors out of bound (%d)\n",
>> +                   XE_RAS_NUM_ERROR_ARR);
>> +            goto err;
>> +        }
>> +
>> +        for (int i = 0; i < response.num_errors; i++) {
>> +            struct xe_ras_error_array arr = response.error_arr[i];
>> +            enum xe_ras_recovery_action action;
>> +            struct xe_ras_error_class error_class;
>> +            u8 component;
>> +
>> +            error_class = arr.error_class;
>> +            component = error_class.common.component;
>> +
>> +            switch (component) {
>> +            case XE_RAS_COMPONENT_CORE_COMPUTE:
>> +                action = handle_compute_errors(xe, &arr);
>> +                break;
>> +            default:
>> +                xe_err(xe, "[RAS]: Unknown error component %u\n", 
>> component);
>> +                break;
>> +            }
>> +
>> +            /*
>> +             * Retain the highest severity action. Process and log 
>> all errors
>> +             * and then take appropriate recovery action
>> +             */
>> +            if (action > final_action)
>> +                final_action = action;
>> +        }
>> +
>> +    } while (response.additional_errors);
>> +
>> +    return final_action;
>> +
>> +err:
>> +    return XE_RAS_RECOVERY_ACTION_RESET;
>> +}
>> +
>>   #ifdef CONFIG_PCIEAER
>>   static void aer_unmask_and_downgrade_internal_error(struct 
>> xe_device *xe)
>>   {
>> diff --git a/drivers/gpu/drm/xe/xe_ras.h b/drivers/gpu/drm/xe/xe_ras.h
>> index 14cb973603e7..e191ab80080c 100644
>> --- a/drivers/gpu/drm/xe/xe_ras.h
>> +++ b/drivers/gpu/drm/xe/xe_ras.h
>> @@ -6,8 +6,11 @@
>>   #ifndef _XE_RAS_H_
>>   #define _XE_RAS_H_
>>   +#include "xe_ras_types.h"
>> +
>>   struct xe_device;
>>     void xe_ras_init(struct xe_device *xe);
>> +enum xe_ras_recovery_action  xe_ras_process_errors(struct xe_device 
>> *xe);
>>     #endif
>> diff --git a/drivers/gpu/drm/xe/xe_ras_types.h 
>> b/drivers/gpu/drm/xe/xe_ras_types.h
>> index 676755732ef6..221d07efd84c 100644
>> --- a/drivers/gpu/drm/xe/xe_ras_types.h
>> +++ b/drivers/gpu/drm/xe/xe_ras_types.h
>> @@ -11,6 +11,22 @@
>>   #define XE_RAS_NUM_ERROR_ARR        3
>>   #define XE_RAS_MAX_ERROR_DETAILS    16
>>   +/**
>> + * enum xe_ras_recovery_action - RAS recovery actions
>> + *
>> + * @XE_RAS_RECOVERY_ACTION_RECOVERED: Error recovered
>> + * @XE_RAS_RECOVERY_ACTION_RESET: Requires reset
>> + * @XE_RAS_RECOVERY_ACTION_DISCONNECT: Requires disconnect
>> + *
>> + * This enum defines the possible recovery actions that can be taken 
>> in response
>> + * to RAS errors.
>> + */
>> +enum xe_ras_recovery_action {
>> +    XE_RAS_RECOVERY_ACTION_RECOVERED = 0,
>> +    XE_RAS_RECOVERY_ACTION_RESET,
>> +    XE_RAS_RECOVERY_ACTION_DISCONNECT
>> +};
>> +
>>   /**
>>    * struct xe_ras_error_common - Common RAS error class
>>    *