Fix: VeraRubin host BMC displays boot order in a different format by hakhondzadeh · Pull Request #112 · NVIDIA/libredfish

hakhondzadeh · 2026-07-24T19:59:07Z

Redfish path redfish/v1/Systems/System_0 for a Vera Rubin host bmc, show boot orders in this format:
"BootOrder": [ "Boot0019: Ubuntu", "Boot0008: UEFI HTTPv4 (MAC:AC3AE2634C82)",
While previously, we saw them in this format:

 "BootOrder": [
            "Boot0002",
            "Boot0001",
            "Boot0000"
        ],

This PR fixes this issue by removing the suffix of the boot order. Also, when patching a boot order, the name is added to keep the formatting consistent.

Unit tests were also added.

Vera Rubin host BMC firmware reports BootOrder strings like "Boot0019: Ubuntu" while BootOption resources live at bare Ids. Strip the display-name suffix before GET and preserve firmware format when reordering boot options.

Add boot_order_prepend_first helper with unit tests for the PATCH path, simplify vera_rubin boot-order reordering, and fix stale GB200 error text.

krish-nvidia · 2026-07-24T20:49:26Z

Could we handle the Vera Rubin boot order quirk directly in set_boot_order_dpu_first()? I don't want vendor specific things bleeding out into the standard model (as well as vendor specific unit tests). I have a reference implementation below that you can drop in which follows how we set the boot order for other vendors.

fn set_boot_order_dpu_first<'a>(
    &'a self,
    boot_interface: crate::BootInterfaceRef<'a>,
) -> crate::RedfishFuture<'a, Result<Option<String>, RedfishError>> {
    Box::pin(async move {
        let address = crate::resolve_boot_interface_mac(self, boot_interface).await?;
        let mac_address = address.replace(':', "").to_uppercase();
        let boot_option_name =
            format!("{} (MAC:{})", BootOptionName::Http.to_string(), mac_address);
        let system = self.s.get_system().await?;
        let boot_options_id =
            system
                .boot
                .boot_options
                .clone()
                .ok_or_else(|| RedfishError::MissingKey {
                    key: "boot.boot_options".to_string(),
                    url: system.odata.odata_id.clone(),
                })?;
        let boot_options: Vec<BootOption> = self
            .get_collection(boot_options_id)
            .await
            .and_then(|collection| collection.try_get::<BootOption>())?
            .members;
        let target = boot_options
            .iter()
            .find(|option| option.display_name.starts_with(&boot_option_name))
            .ok_or_else(|| RedfishError::GenericError {
                error: format!("Could not find boot option matching {boot_option_name}"),
            })?;
        let mut boot_order = system.boot.boot_order;
        let target_index = boot_order
            .iter()
            .position(|entry| {
                entry
                    .split_once(": ")
                    .map_or(entry.as_str(), |(reference, _)| reference)
                    == target.boot_option_reference
            })
            .ok_or_else(|| RedfishError::GenericError {
                error: format!(
                    "Boot option {} is not present in BootOrder",
                    target.boot_option_reference
                ),
            })?;
        // Preserve the exact entry returned by Vera Rubin firmware.
        let target_entry = boot_order.remove(target_index);
        boot_order.insert(0, target_entry);
        self.change_boot_order(boot_order).await?;
        Ok(None)
    })
}

async fn get_expected_and_actual_first_boot_option(
    &self,
    boot_interface_mac: &str,
) -> Result<(Option<String>, Option<String>), RedfishError> {
    let mac_address = boot_interface_mac.replace(':', "").to_uppercase();
    let boot_option_name =
        format!("{} (MAC:{})", BootOptionName::Http.to_string(), mac_address);
    let boot_order = self.s.get_system().await?.boot.boot_order;
    let actual_first_boot_option = if let Some(first) = boot_order.first() {
        let reference = first
            .split_once(": ")
            .map_or(first.as_str(), |(reference, _)| reference);
        Some(self.s.get_boot_option(reference).await?.display_name)
    } else {
        None
    };
    let mut expected_first_boot_option = None;
    for entry in &boot_order {
        let reference = entry
            .split_once(": ")
            .map_or(entry.as_str(), |(reference, _)| reference);
        let option = self.s.get_boot_option(reference).await?;
        if option.display_name.starts_with(&boot_option_name) {
            expected_first_boot_option = Some(option.display_name);
            break;
        }
    }
    Ok((
        expected_first_boot_option,
        actual_first_boot_option,
    ))
}

krish-nvidia · 2026-07-24T21:04:56Z

Also, I don't see a fix for this issue other than changing the model in the error message?

Details: BMC vendor does not support this operation: GB200 doesn't have Systems EthernetInterface BlueField_NIC_0_Port_0.

Checked the VR BMC and it does exist:

{
    "@odata.id": "/redfish/v1/Systems/System_0/EthernetInterfaces",
    "@odata.type": "#EthernetInterfaceCollection.EthernetInterfaceCollection",
    "Description": "Collection of EthernetInterfaces of the host",
    "Members": [
        {
            "@odata.id": "/redfish/v1/Systems/System_0/EthernetInterfaces/BlueField_NIC_0_Port_0"
        },

bcavnvidia · 2026-07-24T21:20:33Z

/redfish/v1/Systems/System_0/EthernetInterfaces/BlueField_NIC_0_Port_0
Also, I don't see a fix for this issue other than changing the model in the error message?

Details: BMC vendor does not support this operation: GB200 doesn't have Systems EthernetInterface BlueField_NIC_0_Port_0.

Checked the VR BMC and it does exist:

{
    "@odata.id": "/redfish/v1/Systems/System_0/EthernetInterfaces",
    "@odata.type": "#EthernetInterfaceCollection.EthernetInterfaceCollection",
    "Description": "Collection of EthernetInterfaces of the host",
    "Members": [
        {
            "@odata.id": "/redfish/v1/Systems/System_0/EthernetInterfaces/BlueField_NIC_0_Port_0"
        },

~~Seems like just minor clean-up. I think this is just house-keeping. Current version of libredfish only had the boot-order issue, not NIC location issue.~~ Nvm, I see what you meant. It's currently short-circuiting to error, but the VR has the that path.

hakhondzadeh · 2026-07-24T21:29:54Z

/redfish/v1/Systems/System_0/EthernetInterfaces/BlueField_NIC_0_Port_0
Also, I don't see a fix for this issue other than changing the model in the error message?
Details: BMC vendor does not support this operation: GB200 doesn't have Systems EthernetInterface BlueField_NIC_0_Port_0.
Checked the VR BMC and it does exist:
{
    "@odata.id": "/redfish/v1/Systems/System_0/EthernetInterfaces",
    "@odata.type": "#EthernetInterfaceCollection.EthernetInterfaceCollection",
    "Description": "Collection of EthernetInterfaces of the host",
    "Members": [
        {
            "@odata.id": "/redfish/v1/Systems/System_0/EthernetInterfaces/BlueField_NIC_0_Port_0"
        },
Seems like just minor clean-up. I think this is just house-keeping. Current version of libredfish only had the boot-order issue, not NIC location issue.

There is no functional NIC fix in this PR; just fixed the string and changed GB200 to VR so to avoid confusion later. Implementing real System EthernetInterface support for Vera Rubin is separate work

hakhondzadeh · 2026-07-24T21:40:29Z

Could we handle the Vera Rubin boot order quirk directly in set_boot_order_dpu_first()? I don't want vendor specific things bleeding out into the standard model (as well as vendor specific unit tests). I have a reference implementation below that you can drop in which follows how we set the boot order for other vendors.

fn set_boot_order_dpu_first<'a>(
    &'a self,
    boot_interface: crate::BootInterfaceRef<'a>,
) -> crate::RedfishFuture<'a, Result<Option<String>, RedfishError>> {
    Box::pin(async move {
        let address = crate::resolve_boot_interface_mac(self, boot_interface).await?;
        let mac_address = address.replace(':', "").to_uppercase();
        let boot_option_name =
            format!("{} (MAC:{})", BootOptionName::Http.to_string(), mac_address);
        let system = self.s.get_system().await?;
        let boot_options_id =
            system
                .boot
                .boot_options
                .clone()
                .ok_or_else(|| RedfishError::MissingKey {
                    key: "boot.boot_options".to_string(),
                    url: system.odata.odata_id.clone(),
                })?;
        let boot_options: Vec<BootOption> = self
            .get_collection(boot_options_id)
            .await
            .and_then(|collection| collection.try_get::<BootOption>())?
            .members;
        let target = boot_options
            .iter()
            .find(|option| option.display_name.starts_with(&boot_option_name))
            .ok_or_else(|| RedfishError::GenericError {
                error: format!("Could not find boot option matching {boot_option_name}"),
            })?;
        let mut boot_order = system.boot.boot_order;
        let target_index = boot_order
            .iter()
            .position(|entry| {
                entry
                    .split_once(": ")
                    .map_or(entry.as_str(), |(reference, _)| reference)
                    == target.boot_option_reference
            })
            .ok_or_else(|| RedfishError::GenericError {
                error: format!(
                    "Boot option {} is not present in BootOrder",
                    target.boot_option_reference
                ),
            })?;
        // Preserve the exact entry returned by Vera Rubin firmware.
        let target_entry = boot_order.remove(target_index);
        boot_order.insert(0, target_entry);
        self.change_boot_order(boot_order).await?;
        Ok(None)
    })
}

async fn get_expected_and_actual_first_boot_option(
    &self,
    boot_interface_mac: &str,
) -> Result<(Option<String>, Option<String>), RedfishError> {
    let mac_address = boot_interface_mac.replace(':', "").to_uppercase();
    let boot_option_name =
        format!("{} (MAC:{})", BootOptionName::Http.to_string(), mac_address);
    let boot_order = self.s.get_system().await?.boot.boot_order;
    let actual_first_boot_option = if let Some(first) = boot_order.first() {
        let reference = first
            .split_once(": ")
            .map_or(first.as_str(), |(reference, _)| reference);
        Some(self.s.get_boot_option(reference).await?.display_name)
    } else {
        None
    };
    let mut expected_first_boot_option = None;
    for entry in &boot_order {
        let reference = entry
            .split_once(": ")
            .map_or(entry.as_str(), |(reference, _)| reference);
        let option = self.s.get_boot_option(reference).await?;
        if option.display_name.starts_with(&boot_option_name) {
            expected_first_boot_option = Some(option.display_name);
            break;
        }
    }
    Ok((
        expected_first_boot_option,
        actual_first_boot_option,
    ))
}

Helper functions was intentionally added to model/boot.rs because it’s not really Vera Rubin–specific. Newer firmware on other GPU platforms could easily do the same thing. When that happens, a shared helper avoids copying the same parsing into every nvidia_*.rs file.
However, I changed to your approach. If another platform starts emitting "Boot####: DisplayName" in BootOrder, we can promote more of the reorder logic into boot.rs then. For now, with only VR showing this on the hosts we debugged, keeping promote in the vendor module works.

bcavnvidia · 2026-07-24T21:43:26Z

+                .members;
+            let target = boot_options
+                .iter()
+                .find(|option| option.display_name.starts_with(&boot_option_name))


I think you might need exact match. Live data looked like this:

...: UEFI HTTPv4 (MAC:XXXXXXXXXXXX) ...: UEFI HTTPv4 (MAC:XXXXXXXXXXXX) 2

Oh true, do you know what the 2 option is? I'm assuming we just need to pick one without the 2.

fixed, thanks

krish-nvidia · 2026-07-24T21:48:46Z

There is no functional NIC fix in this PR; just fixed the string and changed GB200 to VR so to avoid confusion later. Implementing real System EthernetInterface support for Vera Rubin is separate work

Correct me if I'm wrong, but the host won't move past the set-boot-order state because it first calls is_boot_order_setup(), which calls resolve_boot_interface_mac(), which then calls Vera Rubin's get_system_ethernet_interface() implementation. That currently returns NotSupported. The fix should be to delegate to the standard implementation instead, since I verified that the standard Redfish path exists and works:

fn get_system_ethernet_interfaces<'a>(
    &'a self,
) -> crate::RedfishFuture<'a, Result<Vec<String>, RedfishError>> {
    Box::pin(async move { self.s.get_system_ethernet_interfaces().await })
}
fn get_system_ethernet_interface<'a>(
    &'a self,
    id: &'a str,
) -> crate::RedfishFuture<'a, Result<crate::EthernetInterface, RedfishError>> {
    Box::pin(async move { self.s.get_system_ethernet_interface(id).await })
}

bcavnvidia · 2026-07-24T21:54:13Z

There is no functional NIC fix in this PR; just fixed the string and changed GB200 to VR so to avoid confusion later. Implementing real System EthernetInterface support for Vera Rubin is separate work

Correct me if I'm wrong, but the host won't move past the set-boot-order state because it first calls is_boot_order_setup(), which calls resolve_boot_interface_mac(), which then calls Vera Rubin's get_system_ethernet_interface() implementation. That currently returns NotSupported. The fix should be to delegate to the standard implementation instead, since I verified that the standard Redfish path exists and works:
fn get_system_ethernet_interfaces<'a>(
    &'a self,
) -> crate::RedfishFuture<'a, Result<Vec<String>, RedfishError>> {
    Box::pin(async move { self.s.get_system_ethernet_interfaces().await })
}
fn get_system_ethernet_interface<'a>(
    &'a self,
    id: &'a str,
) -> crate::RedfishFuture<'a, Result<crate::EthernetInterface, RedfishError>> {
    Box::pin(async move { self.s.get_system_ethernet_interface(id).await })
}

I think the changes here avoid the get_system_ethernet_interface call #110

EDIT:
Traced:

(core) e3705ccc0+ with libredfish v0.45+: guaranteed to send Mac or Pair once.
VeraRubin then resolves either directly from the supplied MAC without calling get_system_ethernet_interface().

The fix should be to delegate to the standard implementation instead, since I verified that the standard Redfish path exists and works

Still fine to do it for sure, but it seems no reason to lump it into this PR. Though, we're going to need a version bump either way.

krish-nvidia · 2026-07-24T22:05:03Z

I think the changes here avoid the get_system_ethernet_interface call #110

Lesson learned to update my local libredfish checkout before starting to review new changes 🤦‍♂️ Thanks for the explanation!

hakhondzadeh · 2026-07-24T22:32:37Z

There is no functional NIC fix in this PR; just fixed the string and changed GB200 to VR so to avoid confusion later. Implementing real System EthernetInterface support for Vera Rubin is separate work

Correct me if I'm wrong, but the host won't move past the set-boot-order state because it first calls is_boot_order_setup(), which calls resolve_boot_interface_mac(), which then calls Vera Rubin's get_system_ethernet_interface() implementation. That currently returns NotSupported. The fix should be to delegate to the standard implementation instead, since I verified that the standard Redfish path exists and works:
fn get_system_ethernet_interfaces<'a>(
    &'a self,
) -> crate::RedfishFuture<'a, Result<Vec<String>, RedfishError>> {
    Box::pin(async move { self.s.get_system_ethernet_interfaces().await })
}
fn get_system_ethernet_interface<'a>(
    &'a self,
    id: &'a str,
) -> crate::RedfishFuture<'a, Result<crate::EthernetInterface, RedfishError>> {
    Box::pin(async move { self.s.get_system_ethernet_interface(id).await })
}

This PR fixes the boot URL. NIC interfaces can be done in a separate PR. thanks for pointing it out

hakhondzadeh added 2 commits July 24, 2026 11:11

refine(vera_rubin): extract boot_order_prepend_first and add patch tests

513c6ba

Add boot_order_prepend_first helper with unit tests for the PATCH path, simplify vera_rubin boot-order reordering, and fix stale GB200 error text.

hakhondzadeh changed the title ~~Vr boot order~~ Fix: VeraRubin host BMC displays boot order in a different format Jul 24, 2026

hakhondzadeh requested review from bcavnvidia and krish-nvidia July 24, 2026 20:00

Fix: move some helper function from boot.rs to vera_rubin.rs

bbfcecf

bcavnvidia reviewed Jul 24, 2026

View reviewed changes

fix: match boot string instead of prefix match

24a34b0

hakhondzadeh requested a review from bcavnvidia July 24, 2026 22:33

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Fix: VeraRubin host BMC displays boot order in a different format#112

Fix: VeraRubin host BMC displays boot order in a different format#112
hakhondzadeh wants to merge 4 commits into
NVIDIA:mainfrom
hakhondzadeh:VR_bootOrder

hakhondzadeh commented Jul 24, 2026 •

edited

Loading

Uh oh!

krish-nvidia commented Jul 24, 2026 •

edited

Loading

Uh oh!

krish-nvidia commented Jul 24, 2026

Uh oh!

bcavnvidia commented Jul 24, 2026 •

edited

Loading

Uh oh!

hakhondzadeh commented Jul 24, 2026 •

edited

Loading

Uh oh!

hakhondzadeh commented Jul 24, 2026

Uh oh!

bcavnvidia Jul 24, 2026

Uh oh!

krish-nvidia Jul 24, 2026

Uh oh!

hakhondzadeh Jul 24, 2026

Uh oh!

krish-nvidia commented Jul 24, 2026

Uh oh!

bcavnvidia commented Jul 24, 2026 •

edited

Loading

Uh oh!

krish-nvidia commented Jul 24, 2026

Uh oh!

hakhondzadeh commented Jul 24, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

hakhondzadeh commented Jul 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

krish-nvidia commented Jul 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

krish-nvidia commented Jul 24, 2026

Uh oh!

bcavnvidia commented Jul 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

hakhondzadeh commented Jul 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

hakhondzadeh commented Jul 24, 2026

Uh oh!

bcavnvidia Jul 24, 2026

Choose a reason for hiding this comment

Uh oh!

krish-nvidia Jul 24, 2026

Choose a reason for hiding this comment

Uh oh!

hakhondzadeh Jul 24, 2026

Choose a reason for hiding this comment

Uh oh!

krish-nvidia commented Jul 24, 2026

Uh oh!

bcavnvidia commented Jul 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

krish-nvidia commented Jul 24, 2026

Uh oh!

hakhondzadeh commented Jul 24, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

hakhondzadeh commented Jul 24, 2026 •

edited

Loading

krish-nvidia commented Jul 24, 2026 •

edited

Loading

bcavnvidia commented Jul 24, 2026 •

edited

Loading

hakhondzadeh commented Jul 24, 2026 •

edited

Loading

bcavnvidia commented Jul 24, 2026 •

edited

Loading