Re: [PATCH v4 08/10] lib/igt_drm_netlink: add event notify subscription and event wait support

["Koppuravuri, Ravi Kishore" <ravi.kishore.koppuravuri@intel.com>]

Hi Soham,

On 18-09-2026 20:33, Purkait, Soham wrote:
> Hi Ravi,
>
> On 03-09-2026 13:03, Ravi Kishore Koppuravuri wrote:
>> Added event notification support with subscribe for an error-notify 
>> event
>> and wait for the event notification functionalities.
>>
>> Cc: Purkait Soham <soham.purkait@intel.com>
>> Cc: Riana Tauro <riana.tauro@intel.com>
>> Cc: Gupta Anshuman <anshuman.gupta@intel.com>
>> Cc: Jadav Raag <raag.jadav@intel.com>
>> Cc: Nilawar Badal <badal.nilawar@intel.com>
>> Cc: Koujalagi Mallesh <mallesh.koujalagi@intel.com>
>> Cc: Harish Chegondi <harish.chegondi@intel.com>
>> Signed-off-by: Ravi Kishore Koppuravuri 
>> <ravi.kishore.koppuravuri@intel.com>
>> ---
>> v2:Handled SEQ_CHECK nl response using separate callback
>>     Added monotonic-clock deadline to wait for messages without 
>> exceeding the
>>     caller’s total timeout.
>>     Added support to receive events until it sees an event whose node 
>> ID and
>>     error ID match the requested values.
>> v4:Renamed the library functions with prefix "igt_drm_netlink_*" (Nikula
>>     Jani & Kamil)
>>     updated error event name from error-notify to error-report to align
>>     with the latest changes in drm-uapi/drm_ras.h
>> ---
>>   lib/igt_drm_netlink.c | 154 +++++++++++++++++++++++++++++++++++++++++-
>>   lib/igt_drm_netlink.h |   6 ++
>>   2 files changed, 157 insertions(+), 3 deletions(-)
>>
>> diff --git a/lib/igt_drm_netlink.c b/lib/igt_drm_netlink.c
>> index c0e3aff9b..6a3e475e8 100644
>> --- a/lib/igt_drm_netlink.c
>> +++ b/lib/igt_drm_netlink.c
>> @@ -6,6 +6,7 @@
>>   #include <errno.h>
>>   #include <poll.h>
>>   #include <stdint.h>
>> +#include <time.h>
>>     #include <netlink/genl/ctrl.h>
>>   #include <netlink/genl/genl.h>
>> @@ -28,6 +29,27 @@ static int ras_command_cb(struct nl_msg *msg, void 
>> *arg)
>>       switch (gnlh->cmd) {
>>       case DRM_RAS_CMD_SET_ERROR_THRESHOLD:
>>           break;
>> +    case DRM_RAS_CMD_ERROR_EVENT: {
>> +        struct nlattr *attrs[DRM_RAS_A_ERROR_EVENT_ATTRS_MAX + 1];
> Move this at the start of the function.
will separate the attrs for event and other commands. events specific 
attr will be needed only for DRM_RAS_CMD_ERROR_EVENT and so will create 
struct nlattr *attrs inside the switch case itself. And will move the 
attrs for all other commands to the beginning of the function.
>> +
>> +        ret = genlmsg_parse(nlh, 0, attrs,
>> +                    DRM_RAS_A_ERROR_EVENT_ATTRS_MAX, NULL);
>> +        if (ret < 0)
>> +            return NL_SKIP;
>> +
>> +        if (!attrs[DRM_RAS_A_ERROR_EVENT_ATTRS_NODE_ID] ||
>> +            !attrs[DRM_RAS_A_ERROR_EVENT_ATTRS_ERROR_ID] ||
>> +            !attrs[DRM_RAS_A_ERROR_EVENT_ATTRS_ERROR_VALUE])
>> +            return NL_SKIP;
>> +
>> +        ctx->event_node_id = 
>> nla_get_u32(attrs[DRM_RAS_A_ERROR_EVENT_ATTRS_NODE_ID]);
>> +        ctx->event_error_id = 
>> nla_get_u32(attrs[DRM_RAS_A_ERROR_EVENT_ATTRS_ERROR_ID]);
>> +        ctx->event_error_value =
>> + nla_get_u32(attrs[DRM_RAS_A_ERROR_EVENT_ATTRS_ERROR_VALUE]);
>> +        ctx->event_received = ctx->event_node_id == ctx->node_id &&
>> +                      ctx->event_error_id == ctx->error_id;
>> +        break;
>> +    }
>>       case DRM_RAS_CMD_GET_ERROR_COUNTER: {
>>           struct nlattr *attrs[DRM_RAS_A_ERROR_COUNTER_ATTRS_MAX + 1];
>>   @@ -61,13 +83,20 @@ static int ras_command_cb(struct nl_msg *msg, 
>> void *arg)
>>           return NL_SKIP;
>>       }
>>   -    ctx->reply_received = true;
>> -    if (ctx->ack_received)
>> -        ctx->cmd_done = true;
>> +    if (gnlh->cmd != DRM_RAS_CMD_ERROR_EVENT) {
>> +        ctx->reply_received = true;
>> +        if (ctx->ack_received)
>> +            ctx->cmd_done = true;
>> +    }
>>         return NL_OK;
>>   }
>>   +static int ras_multicast_seq_check_cb(struct nl_msg *msg, void *arg)
> What is the use of this callback ?
libnl checks the seq no of responses against the requests and if it does 
not match then the response will be rejected. For 
DRM_RAS_CMD_ERROR_EVENT, we are not sending any request but listening 
for the broadcasts from kernel, which will not have any matching 
sequence number to compare against. This function is registered in 
register_multicast_callbacks() to just return NL_OK to skip that check.
>> +{
>> +    return NL_OK;
>> +}
>> +
>>   static int ras_error_cb(struct sockaddr_nl *nla, struct nlmsgerr 
>> *err, void *arg)
>>   {
>>       struct app_context *ctx = arg;
>> @@ -121,6 +150,23 @@ static int register_callbacks(struct nl_cb *cb, 
>> struct app_context *ctx)
>>       return 0;
>>   }
>>   +static int register_multicast_callbacks(struct nl_cb *cb,
>> +                    struct app_context *ctx)
>> +{
>> +    int ret;
>> +
>> +    ret = nl_cb_set(cb, NL_CB_VALID, NL_CB_CUSTOM, ras_command_cb, 
>> ctx);
>> +    if (ret < 0)
>> +        return ret;
>> +
>> +    ret = nl_cb_set(cb, NL_CB_SEQ_CHECK, NL_CB_CUSTOM,
>> +            ras_multicast_seq_check_cb, ctx);
>> +    if (ret < 0)
>> +        return ret;
>> +
>> +    return 0;
>> +}
>> +
>>   static int send_and_recv_nl_msg(struct app_context *ctx,
>>                   struct nl_cb *cb,
>>                   struct nl_msg *msg)
>> @@ -263,6 +309,10 @@ static void reset_drm_ras_netlink_context(struct 
>> app_context *ctx)
>>       ctx->ack_received = false;
>>       ctx->response_required = false;
>>       ctx->error_threshold = 0;
>> +    ctx->event_node_id = UINT32_MAX;
>> +    ctx->event_error_id = UINT32_MAX;
>> +    ctx->event_error_value = 0;
>> +    ctx->event_received = false;
>>       ctx->family_id = -1;
>>   }
>>   @@ -407,3 +457,101 @@ int 
>> igt_drm_netlink_set_error_threshold(struct app_context *ctx)
>>         return 0;
>>   }
>> +
>> +int igt_drm_netlink_subscribe_error_report(struct app_context *ctx, 
>> const char *notify_group_name)
>> +{
>> +    int grp_id;
>> +    int ret;
>> +
>> +    if (!ctx || !ctx->sock || ctx->family_id < 0 ||
>> +        !notify_group_name || notify_group_name[0] == '\0')
>> +        return -EINVAL;
>> +
>> +    grp_id = genl_ctrl_resolve_grp(ctx->sock,
>> +                       DRM_RAS_FAMILY_NAME,
>> +                       notify_group_name);
>> +    if (grp_id < 0)
>> +        return grp_id;
>> +
>> +    ret = nl_socket_add_membership(ctx->sock, grp_id);
> Before this it is better to have a check for duplicate-subscription.

Even though multiple subscriptions are registered for the same group, 
kernel broadcasts only single message and it does not lead to any error.

But it is safer to add a guard to check if already subscribed to the 
same multicast group or not. will add it.

>> +    if (ret < 0)
>> +        return ret;
>
> libnl error codes (NLE_*) are not the same as standard errno/-errno 
> codes.
Will convert libnl error codes to standard error codes.
>
> Thanks,
> Soham
>> +
>> +    igt_debug("Subscribed to DRM RAS multicast group '%s' (id=%d).\n",
>> +          notify_group_name, grp_id);
>> +
>> +    return 0;
>> +}
>> +
>> +int igt_drm_netlink_wait_for_error_report_event(struct app_context 
>> *ctx, int timeout_ms)
>> +{
>> +    struct nl_cb *cb;
>> +    struct pollfd pfd;
>> +    struct timespec start, now;
>> +    int remaining_ms;
>> +    int ret;
>> +
>> +    if (!ctx || !ctx->sock || ctx->family_id < 0)
>> +        return -EINVAL;
>> +
>> +    if (timeout_ms < -1)
>> +        return -EINVAL;
>> +
>> +    cb = nl_cb_alloc(NL_CB_DEFAULT);
>> +    if (!cb)
>> +        return -ENOMEM;
>> +
>> +    ret = register_multicast_callbacks(cb, ctx);
>> +    if (ret < 0) {
>> +        nl_cb_put(cb);
>> +        return ret;
>> +    }
>> +
>> +    ctx->event_received = false;
>> +    clock_gettime(CLOCK_MONOTONIC, &start);
>> +
>> +    pfd.fd = nl_socket_get_fd(ctx->sock);
>> +    pfd.events = POLLIN;
>> +    pfd.revents = 0;
>> +
>> +    while (!ctx->event_received) {
>> +        if (timeout_ms == -1) {
>> +            remaining_ms = -1;
>> +        } else {
>> +            clock_gettime(CLOCK_MONOTONIC, &now);
>> +            remaining_ms = timeout_ms -
>> +                       (int)((now.tv_sec - start.tv_sec) * 1000 +
>> +                       (now.tv_nsec - start.tv_nsec) / 1000000);
>> +            if (remaining_ms <= 0) {
>> +                ret = -ETIMEDOUT;
>> +                break;
>> +            }
>> +        }
>> +
>> +        ret = poll(&pfd, 1, remaining_ms);
>> +        if (ret == 0) {
>> +            ret = -ETIMEDOUT;
>> +            break;
>> +        }
>> +
>> +        if (ret < 0) {
>> +            ret = -errno;
>> +            break;
>> +        }
>> +
>> +        ret = nl_recvmsgs(ctx->sock, cb);
>> +        if (ret < 0)
>> +            break;
>> +    }
>> +    nl_cb_put(cb);
>> +    if (ret < 0)
>> +        return ret;
>> +
>> +    if (!ctx->event_received)
>> +        return -ENOMSG;
>> +
>> +    igt_debug("Received error-notify event: node_id=%u error_id=%u 
>> value=%u\n",
>> +          ctx->event_node_id, ctx->event_error_id, 
>> ctx->event_error_value);
>> +
>> +    return 0;
>> +}
>> diff --git a/lib/igt_drm_netlink.h b/lib/igt_drm_netlink.h
>> index 9376e3dee..ebc575da6 100644
>> --- a/lib/igt_drm_netlink.h
>> +++ b/lib/igt_drm_netlink.h
>> @@ -19,6 +19,10 @@ struct app_context {
>>       bool ack_received;
>>       bool response_required;
>>       uint32_t error_threshold;
>> +    uint32_t event_node_id;
>> +    uint32_t event_error_id;
>> +    uint32_t event_error_value;
>> +    bool event_received;
>>       int family_id;
>>   };
>>   @@ -27,6 +31,8 @@ int igt_drm_netlink_init(struct app_context *ctx);
>>   int igt_drm_netlink_get_error_counter(struct app_context *ctx);
>>   int igt_drm_netlink_get_error_threshold(struct app_context *ctx);
>>   int igt_drm_netlink_set_error_threshold(struct app_context *ctx);
>> +int igt_drm_netlink_subscribe_error_report(struct app_context *ctx, 
>> const char *notify_group_name);
>> +int igt_drm_netlink_wait_for_error_report_event(struct app_context 
>> *ctx, int timeout_ms);
>>     #endif /* IGT_DRM_NETLINK_H */