WIP: Obtain Host Panic/Boot Failure Messages

Cargo.toml

@@ -139,7 +139,7 @@ apob = { git = "https://github.com/oxidecomputer/apob", default-features = false
 # for the migration.
 attest-data = { git = "https://github.com/oxidecomputer/dice-util", default-features = false, version = "0.4.0", rev = "a0811d06c75c757a6e12c91ed6ea81fde137ba43" }
 dice-mfg-msgs = { git = "https://github.com/oxidecomputer/dice-util", default-features = false, version = "0.2.1", rev = "a0811d06c75c757a6e12c91ed6ea81fde137ba43" }
-gateway-messages = { git = "https://github.com/oxidecomputer/management-gateway-service", default-features = false, features = ["smoltcp"] }
+gateway-messages = { git = "https://github.com/oxidecomputer/management-gateway-service", default-features = false, features = ["smoltcp"], branch = "james/host-fails" }
 gateway-ereport-messages = { git = "https://github.com/oxidecomputer/management-gateway-service", default-features = false }
 gimlet-inspector-protocol = { git = "https://github.com/oxidecomputer/gimlet-inspector-protocol", version = "0.1.0" }
 hif = { git = "https://github.com/oxidecomputer/hif", default-features = false }

app/cosmo/base.toml

@@ -132,7 +132,7 @@ notifications = ["i2c1-irq", "i2c2-irq", "i2c3-irq", "i2c4-irq"]
 [tasks.packrat]
 name = "task-packrat"
 priority = 1
-stacksize = 1400
+stacksize = 1600
 start = true
 task-slots = ["jefe"]
 features = ["cosmo", "ereport"]
@@ -258,7 +258,7 @@ features = ["stm32h753", "usart6", "baud_rate_3M", "hardware_flow_control", "vla
 uses = ["usart6", "dbgmcu"]
 interrupts = {"usart6.irq" = "usart-irq"}
 priority = 9
-max-sizes = {flash = 74000, ram = 65536}
+max-sizes = {flash = 75000, ram = 65536}
 stacksize = 5400
 start = true
 task-slots = ["sys", { cpu_seq = "cosmo_seq" }, "hf", "control_plane_agent", "net", "packrat", "i2c_driver", { spi_driver = "spi2_driver" }, "sprot", "auxflash"]

app/gimlet/base.toml

@@ -125,7 +125,7 @@ notifications = ["i2c1-irq", "jefe-state-change"]
 [tasks.packrat]
 name = "task-packrat"
 priority = 1
-stacksize = 1400
+stacksize = 1504
 start = true
 task-slots = ["jefe"]
 features = ["gimlet", "ereport"]
@@ -243,7 +243,7 @@ features = ["stm32h753", "uart7", "baud_rate_3M", "hardware_flow_control", "vlan
 uses = ["uart7", "dbgmcu"]
 interrupts = {"uart7.irq" = "usart-irq"}
 priority = 8
-max-sizes = {flash = 72000, ram = 65536}
+max-sizes = {flash = 74000, ram = 65536}
 stacksize = 5376
 start = true
 task-slots = ["sys", { cpu_seq = "gimlet_seq" }, "hf", "control_plane_agent", "net", "packrat", "i2c_driver", { spi_driver = "spi2_driver" }, "sprot"]

app/grapefruit/standalone.toml

@@ -11,7 +11,7 @@ interrupts = {"uart8.irq" = "usart-irq"}
 
 # Ereport stuff
 [tasks.packrat]
-stacksize = 1400
+stacksize = 1504
 task-slots = ["jefe"]
 features = ["ereport"]
 notifications = ["task-faulted"]

idl/packrat.idol

@@ -144,5 +144,84 @@ Interface(
             ),
             idempotent: true,
         ),
+        "write_host_bootfail": (
+            doc: "Write a host's boot failure message and return the index of this failure",
+            args: {
+                "reason": "u8",
+            },
+            leases: {
+                "data": (type: "[u8]", read: true),
+            },
+            reply: Result(
+                ok: "HostInfoWriteOutput",
+                err: ServerDeath,
+            ),
+        ),
+        "read_first_host_bootfail_fragment": (
+            doc: "Read a portion of the host's boot failure message",
+            args: {
+            },
+            leases: {
+                "data": (type: "[u8]", write: true),
+            },
+            reply: Result(
+                ok: "HostBootfailReadOutput",
+                err: CLike("HostInfoReadError"),
+            ),
+            idempotent: true,
+        ),
+        "read_host_bootfail_fragment": (
+            doc: "Read a portion of the host's boot failure message",
+            args: {
+                "request": "HostInfoRequest",
+            },
+            leases: {
+                "data": (type: "[u8]", write: true),
+            },
+            reply: Result(
+                ok: "HostBootfailReadOutput",
+                err: CLike("HostInfoReadError"),
+            ),
+            idempotent: true,
+        ),
+        "write_host_panic": (
+            doc: "Write a host's panic message and return the index of this panic",
+            args: {
+            },
+            leases: {
+                "data": (type: "[u8]", read: true),
+            },
+            reply: Result(
+                ok: "HostInfoWriteOutput",
+                err: ServerDeath,
+            ),
+        ),
+        "read_first_host_panic_fragment": (
+            doc: "Read a portion of the host's panic message",
+            args: {
+            },
+            leases: {
+                "data": (type: "[u8]", write: true),
+            },
+            reply: Result(
+                ok: "HostPanicReadOutput",
+                err: CLike("HostInfoReadError"),
+            ),
+            idempotent: true,
+        ),
+        "read_host_panic_fragment": (
+            doc: "Read a portion of the host's panic message",
+            args: {
+                "request": "HostInfoRequest",
+            },
+            leases: {
+                "data": (type: "[u8]", write: true),
+            },
+            reply: Result(
+                ok: "HostPanicReadOutput",
+                err: CLike("HostInfoReadError"),
+            ),
+            idempotent: true,
+        ),
     },
 )

task/control-plane-agent/src/mgs_compute_sled.rs

@@ -17,7 +17,8 @@ use gateway_messages::sp_impl::{
 use gateway_messages::{
     ApobComponentAction, ComponentAction, ComponentActionResponse,
     ComponentDetails, ComponentUpdatePrepare, DiscoverResponse, DumpSegment,
-    DumpTask, GpioToggleCount, Header, IgnitionCommand, IgnitionState,
+    DumpTask, GpioToggleCount, Header, HostBootfailPayloadData,
+    HostInfoRequest, HostPanicPayloadData, IgnitionCommand, IgnitionState,
     LastPostCode, Message, MessageKind, MgsError, MgsRequest, MgsResponse,
     PmbusStatus, PostCode, PowerRailName, PowerState, PowerStateTransition,
     RotBootInfo, RotRequest, RotResponse, SERIAL_CONSOLE_IDLE_TIMEOUT,
@@ -1261,6 +1262,101 @@ impl SpHandler for MgsHandler {
         self.host_flash_update.get_hash(slot)
     }
 
+    fn get_host_panic_payload(
+        &mut self,
+        request: Option<HostInfoRequest>,
+        len: u32,
+        trailing_tx_buf: &mut [u8],
+    ) -> Result<HostPanicPayloadData, SpError> {
+        let max_len_usize = len as usize;
+        let max_len_usize = max_len_usize.min(trailing_tx_buf.len());
+        let dest = &mut trailing_tx_buf[..max_len_usize];
+
+        let res = if let Some(req) = request {
+            self.common.packrat().read_host_panic_fragment(
+                task_packrat_api::HostInfoRequest {
+                    offset: req.offset,
+                    index: req.index,
+                },
+                dest,
+            )
+        } else {
+            self.common.packrat().read_first_host_panic_fragment(dest)
+        };
+
+        let info = res.map_err(|e| {
+            SpError::HostPanic(match e {
+                task_packrat_api::HostInfoReadError::NoHostInfo => {
+                    gateway_messages::HostPanicError::NoHostInfo
+                }
+                task_packrat_api::HostInfoReadError::InvalidOffset => {
+                    gateway_messages::HostPanicError::InvalidOffset
+                }
+                task_packrat_api::HostInfoReadError::InvalidIndex => {
+                    gateway_messages::HostPanicError::InvalidIndex
+                }
+                task_packrat_api::HostInfoReadError::ServerRestarted => {
+                    gateway_messages::HostPanicError::ServerRestarted
+                }
+            })
+        })?;
+
+        Ok(HostPanicPayloadData {
+            index: info.index,
+            len: info.read,
+            total_len: info.total_len as u32,
+        })
+    }
+
+    fn get_host_bootfail_payload(
+        &mut self,
+        request: Option<HostInfoRequest>,
+        len: u32,
+        trailing_tx_buf: &mut [u8],
+    ) -> Result<HostBootfailPayloadData, SpError> {
+        let max_len_usize = len as usize;
+        let max_len_usize = max_len_usize.min(trailing_tx_buf.len());
+        let dest = &mut trailing_tx_buf[..max_len_usize];
+
+        let res = if let Some(req) = request {
+            self.common.packrat().read_host_bootfail_fragment(
+                task_packrat_api::HostInfoRequest {
+                    offset: req.offset,
+                    index: req.index,
+                },
+                dest,
+            )
+        } else {
+            self.common
+                .packrat()
+                .read_first_host_bootfail_fragment(dest)
+        };
+
+        let info = res.map_err(|e| {
+            SpError::HostBootfail(match e {
+                task_packrat_api::HostInfoReadError::NoHostInfo => {
+                    gateway_messages::HostBootfailError::NoHostInfo
+                }
+                task_packrat_api::HostInfoReadError::InvalidOffset => {
+                    gateway_messages::HostBootfailError::InvalidOffset
+                }
+                task_packrat_api::HostInfoReadError::InvalidIndex => {
+                    gateway_messages::HostBootfailError::InvalidIndex
+                }
+                task_packrat_api::HostInfoReadError::ServerRestarted => {
+                    gateway_messages::HostBootfailError::ServerRestarted
+                }
+            })
+        })?;
+
+        Ok(HostBootfailPayloadData {
+            index: info.index,
+            len: info.read,
+            total_len: info.total_len as u32,
+            reason: info.reason,
+        })
+    }
+
     fn get_pmbus_status(
         &mut self,
         rail: &PowerRailName,

task/control-plane-agent/src/mgs_minibar.rs

@@ -17,6 +17,7 @@ use gateway_messages::sp_impl::{
 use gateway_messages::{
     ComponentAction, ComponentActionResponse, ComponentDetails,
     ComponentUpdatePrepare, DiscoverResponse, DumpSegment, DumpTask,
+    HostBootfailPayloadData, HostInfoRequest, HostPanicPayloadData,
     IgnitionCommand, IgnitionState, MgsError, MgsRequest, MgsResponse,
     PmbusStatus, PowerRailName, PowerState, PowerStateTransition, RotBootInfo,
     RotRequest, RotResponse, SensorRequest, SensorResponse, SpComponent,
@@ -695,4 +696,22 @@ impl SpHandler for MgsHandler {
     ) -> Result<PmbusStatus, SpError> {
         self.common.get_pmbus_status(rail)
     }
+
+    fn get_host_bootfail_payload(
+        &mut self,
+        _request: Option<HostInfoRequest>,
+        _len: u32,
+        _trailing_tx_buf: &mut [u8],
+    ) -> Result<HostBootfailPayloadData, SpError> {
+        Err(SpError::RequestUnsupportedForSp)
+    }
+
+    fn get_host_panic_payload(
+        &mut self,
+        _request: Option<HostInfoRequest>,
+        _len: u32,
+        _trailing_tx_buf: &mut [u8],
+    ) -> Result<HostPanicPayloadData, SpError> {
+        Err(SpError::RequestUnsupportedForSp)
+    }
 }

task/control-plane-agent/src/mgs_psc.rs

@@ -14,6 +14,7 @@ use gateway_messages::sp_impl::{
 use gateway_messages::{
     ComponentAction, ComponentActionResponse, ComponentDetails,
     ComponentUpdatePrepare, DiscoverResponse, DumpSegment, DumpTask,
+    HostBootfailPayloadData, HostInfoRequest, HostPanicPayloadData,
     IgnitionCommand, IgnitionState, MgsError, MgsRequest, MgsResponse,
     PmbusStatus, PowerRailName, PowerState, PowerStateTransition, RotBootInfo,
     RotRequest, RotResponse, SensorRequest, SensorResponse, SpComponent,
@@ -710,4 +711,22 @@ impl SpHandler for MgsHandler {
     ) -> Result<PmbusStatus, SpError> {
         self.common.get_pmbus_status(rail)
     }
+
+    fn get_host_bootfail_payload(
+        &mut self,
+        _request: Option<HostInfoRequest>,
+        _len: u32,
+        _trailing_tx_buf: &mut [u8],
+    ) -> Result<HostBootfailPayloadData, SpError> {
+        Err(SpError::RequestUnsupportedForSp)
+    }
+
+    fn get_host_panic_payload(
+        &mut self,
+        _request: Option<HostInfoRequest>,
+        _len: u32,
+        _trailing_tx_buf: &mut [u8],
+    ) -> Result<HostPanicPayloadData, SpError> {
+        Err(SpError::RequestUnsupportedForSp)
+    }
 }

task/control-plane-agent/src/mgs_sidecar.rs

@@ -21,14 +21,14 @@ use gateway_messages::sp_impl::{
 use gateway_messages::{
     ComponentAction, ComponentActionResponse, ComponentDetails,
     ComponentUpdatePrepare, DiscoverResponse, DumpSegment, DumpTask,
-    EcdsaSha2Nistp256Challenge, IgnitionCommand, IgnitionState, MgsError,
-    MgsRequest, MgsResponse, MonorailComponentAction,
-    MonorailComponentActionResponse, MonorailError as GwMonorailError,
-    PcieRegisterRead, PmbusStatus, PowerRailName, PowerState,
-    PowerStateTransition, RotBootInfo, RotRequest, RotResponse, SensorRequest,
-    SensorResponse, SpComponent, SpError, SpStateV2, SpUpdatePrepare,
-    UnlockChallenge, UnlockResponse, UpdateChunk, UpdateId, UpdateStatus,
-    ignition,
+    EcdsaSha2Nistp256Challenge, HostBootfailPayloadData, HostInfoRequest,
+    HostPanicPayloadData, IgnitionCommand, IgnitionState, MgsError, MgsRequest,
+    MgsResponse, MonorailComponentAction, MonorailComponentActionResponse,
+    MonorailError as GwMonorailError, PcieRegisterRead, PmbusStatus,
+    PowerRailName, PowerState, PowerStateTransition, RotBootInfo, RotRequest,
+    RotResponse, SensorRequest, SensorResponse, SpComponent, SpError,
+    SpStateV2, SpUpdatePrepare, UnlockChallenge, UnlockResponse, UpdateChunk,
+    UpdateId, UpdateStatus, ignition,
 };
 use host_sp_messages::HostStartupOptions;
 use idol_runtime::{Leased, RequestError};
@@ -1236,6 +1236,24 @@ impl SpHandler for MgsHandler {
     ) -> Result<PmbusStatus, SpError> {
         self.common.get_pmbus_status(rail)
     }
+
+    fn get_host_bootfail_payload(
+        &mut self,
+        _request: Option<HostInfoRequest>,
+        _len: u32,
+        _trailing_tx_buf: &mut [u8],
+    ) -> Result<HostBootfailPayloadData, SpError> {
+        Err(SpError::RequestUnsupportedForSp)
+    }
+
+    fn get_host_panic_payload(
+        &mut self,
+        _request: Option<HostInfoRequest>,
+        _len: u32,
+        _trailing_tx_buf: &mut [u8],
+    ) -> Result<HostPanicPayloadData, SpError> {
+        Err(SpError::RequestUnsupportedForSp)
+    }
 }
 
 fn get_ecdsa_challenge() -> Result<EcdsaSha2Nistp256Challenge, SpError> {

task/host-sp-comms/Cargo.toml

@@ -48,6 +48,10 @@ task-sensor-api = { path = "../../task/sensor-api", optional = true, features =
 ksz8463 = { path = "../../drv/ksz8463", optional = true }
 drv-sprot-api = { path = "../../drv/sprot-api"}
 
+# ereports deps
+ereports = { path = "../../lib/ereports", features = ["ereporter-macro"] }
+microcbor = { path = "../../lib/microcbor" }
+
 [build-dependencies]
 build-util.path = "../../build/util"
 build-i2c = { path = "../../build/i2c", optional = true }