habanalabs/gaudi: refactor hard-reset related code
authorKoby Elbaz <kelbaz@habana.ai>
Tue, 15 Jun 2021 14:07:02 +0000 (17:07 +0300)
committerOded Gabbay <ogabbay@kernel.org>
Mon, 21 Jun 2021 07:21:51 +0000 (10:21 +0300)
There is code related to hard-reset, which is done in gaudi specific
code. However, this code can be used by future ASICs and therefore it
is better to move it to the common code section.

Signed-off-by: Koby Elbaz <kelbaz@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
drivers/misc/habanalabs/common/firmware_if.c
drivers/misc/habanalabs/common/habanalabs.h
drivers/misc/habanalabs/gaudi/gaudi.c
drivers/misc/habanalabs/gaudi/gaudiP.h

index ce87053..2e4d04e 100644 (file)
@@ -972,6 +972,47 @@ int hl_fw_cpucp_power_get(struct hl_device *hdev, u64 *power)
        return rc;
 }
 
+void hl_fw_ask_hard_reset_without_linux(struct hl_device *hdev)
+{
+       struct static_fw_load_mgr *static_loader =
+                       &hdev->fw_loader.static_loader;
+       int rc;
+
+       if (hdev->asic_prop.dynamic_fw_load) {
+               rc = hl_fw_dynamic_send_protocol_cmd(hdev, &hdev->fw_loader,
+                               COMMS_RST_DEV, 0, false,
+                               hdev->fw_loader.cpu_timeout);
+               if (rc)
+                       dev_warn(hdev->dev, "Failed sending COMMS_RST_DEV\n");
+       } else {
+               WREG32(static_loader->kmd_msg_to_cpu_reg, KMD_MSG_RST_DEV);
+       }
+}
+
+void hl_fw_ask_halt_machine_without_linux(struct hl_device *hdev)
+{
+       struct static_fw_load_mgr *static_loader =
+                       &hdev->fw_loader.static_loader;
+       int rc;
+
+       if (hdev->device_cpu_is_halted)
+               return;
+
+       /* Stop device CPU to make sure nothing bad happens */
+       if (hdev->asic_prop.dynamic_fw_load) {
+               rc = hl_fw_dynamic_send_protocol_cmd(hdev, &hdev->fw_loader,
+                               COMMS_GOTO_WFE, 0, true,
+                               hdev->fw_loader.cpu_timeout);
+               if (rc)
+                       dev_warn(hdev->dev, "Failed sending COMMS_GOTO_WFE\n");
+       } else {
+               WREG32(static_loader->kmd_msg_to_cpu_reg, KMD_MSG_GOTO_WFE);
+               msleep(static_loader->cpu_reset_wait_msec);
+       }
+
+       hdev->device_cpu_is_halted = true;
+}
+
 static void detect_cpu_boot_status(struct hl_device *hdev, u32 status)
 {
        /* Some of the status codes below are deprecated in newer f/w
index 98aa852..6b3cdd7 100644 (file)
@@ -894,6 +894,7 @@ struct pci_mem_region {
  * @preboot_version_offset_reg: SRAM offset to preboot version register
  * @boot_fit_version_offset_reg: SRAM offset to boot fit version register
  * @sram_offset_mask: mask for getting offset into the SRAM
+ * @cpu_reset_wait_msec: used when setting WFE via kmd_msg_to_cpu_reg
  */
 struct static_fw_load_mgr {
        u64 preboot_version_max_off;
@@ -908,6 +909,7 @@ struct static_fw_load_mgr {
        u32 preboot_version_offset_reg;
        u32 boot_fit_version_offset_reg;
        u32 sram_offset_mask;
+       u32 cpu_reset_wait_msec;
 };
 
 /**
@@ -2199,6 +2201,10 @@ struct hl_mmu_funcs {
  *                    triggered, and cleared after it is shared with preboot.
  * @skip_reset_on_timeout: Skip device reset if CS has timed out, wait for it to
  *                         complete instead.
+ * @device_cpu_is_halted: Flag to indicate whether the device CPU was already
+ *                        halted. We can't halt it again because the COMMS
+ *                        protocol will throw an error. Relevant only for
+ *                        cases where Linux was not loaded to device CPU
  */
 struct hl_device {
        struct pci_dev                  *pdev;
@@ -2315,6 +2321,7 @@ struct hl_device {
        u8                              supports_staged_submission;
        u8                              curr_reset_cause;
        u8                              skip_reset_on_timeout;
+       u8                              device_cpu_is_halted;
 
        /* Parameters for bring-up */
        u64                             nic_ports_mask;
@@ -2596,6 +2603,8 @@ int get_used_pll_index(struct hl_device *hdev, u32 input_pll_index,
 int hl_fw_cpucp_pll_info_get(struct hl_device *hdev, u32 pll_index,
                u16 *pll_freq_arr);
 int hl_fw_cpucp_power_get(struct hl_device *hdev, u64 *power);
+void hl_fw_ask_hard_reset_without_linux(struct hl_device *hdev);
+void hl_fw_ask_halt_machine_without_linux(struct hl_device *hdev);
 int hl_fw_init_cpu(struct hl_device *hdev);
 int hl_fw_read_preboot_status(struct hl_device *hdev, u32 cpu_boot_status_reg,
                                u32 sts_boot_dev_sts0_reg,
index 82d5613..aa8a0ca 100644 (file)
@@ -1934,45 +1934,6 @@ static void gaudi_disable_msi(struct hl_device *hdev)
        gaudi->hw_cap_initialized &= ~HW_CAP_MSI;
 }
 
-static void gaudi_ask_hard_reset_without_linux(struct hl_device *hdev)
-{
-       int rc;
-
-       if (hdev->asic_prop.dynamic_fw_load) {
-               rc = hl_fw_dynamic_send_protocol_cmd(hdev, &hdev->fw_loader,
-                               COMMS_RST_DEV, 0, false,
-                               hdev->fw_loader.cpu_timeout);
-               if (rc)
-                       dev_warn(hdev->dev, "Failed sending COMMS_RST_DEV\n");
-       } else {
-               WREG32(mmPSOC_GLOBAL_CONF_KMD_MSG_TO_CPU, KMD_MSG_RST_DEV);
-       }
-}
-
-static void gaudi_ask_halt_machine_without_linux(struct hl_device *hdev)
-{
-       struct gaudi_device *gaudi = hdev->asic_specific;
-       int rc;
-
-       if (gaudi && gaudi->device_cpu_is_halted)
-               return;
-
-       /* Stop device CPU to make sure nothing bad happens */
-       if (hdev->asic_prop.dynamic_fw_load) {
-               rc = hl_fw_dynamic_send_protocol_cmd(hdev, &hdev->fw_loader,
-                               COMMS_GOTO_WFE, 0, true,
-                               hdev->fw_loader.cpu_timeout);
-               if (rc)
-                       dev_warn(hdev->dev, "Failed sending COMMS_GOTO_WFE\n");
-       } else {
-               WREG32(mmPSOC_GLOBAL_CONF_KMD_MSG_TO_CPU, KMD_MSG_GOTO_WFE);
-               msleep(GAUDI_CPU_RESET_WAIT_MSEC);
-       }
-
-       if (gaudi)
-               gaudi->device_cpu_is_halted = true;
-}
-
 static void gaudi_init_scrambler_sram(struct hl_device *hdev)
 {
        struct gaudi_device *gaudi = hdev->asic_specific;
@@ -3859,6 +3820,9 @@ static void gaudi_init_static_firmware_loader(struct hl_device *hdev)
        static_loader->preboot_version_offset_reg = mmPREBOOT_VER_OFFSET;
        static_loader->boot_fit_version_offset_reg = mmUBOOT_VER_OFFSET;
        static_loader->sram_offset_mask = ~(lower_32_bits(SRAM_BASE_ADDR));
+       static_loader->cpu_reset_wait_msec = hdev->pldm ?
+                       GAUDI_PLDM_RESET_WAIT_MSEC :
+                       GAUDI_CPU_RESET_WAIT_MSEC;
 }
 
 static void gaudi_init_firmware_loader(struct hl_device *hdev)
@@ -4151,9 +4115,9 @@ static void gaudi_hw_fini(struct hl_device *hdev, bool hard_reset)
                        gaudi_irq_map_table[GAUDI_EVENT_HALT_MACHINE].cpu_id);
        } else {
                if (hdev->asic_prop.hard_reset_done_by_fw)
-                       gaudi_ask_hard_reset_without_linux(hdev);
+                       hl_fw_ask_hard_reset_without_linux(hdev);
                else
-                       gaudi_ask_halt_machine_without_linux(hdev);
+                       hl_fw_ask_halt_machine_without_linux(hdev);
        }
 
        if (driver_performs_reset) {
@@ -4228,7 +4192,7 @@ static void gaudi_hw_fini(struct hl_device *hdev, bool hard_reset)
 
                memset(gaudi->events_stat, 0, sizeof(gaudi->events_stat));
 
-               gaudi->device_cpu_is_halted = false;
+               hdev->device_cpu_is_halted = false;
        }
 }
 
index b23336a..957bf37 100644 (file)
@@ -315,10 +315,6 @@ struct gaudi_internal_qman_info {
  *                  Multi MSI is possible only with IOMMU enabled.
  * @mmu_cache_inv_pi: PI for MMU cache invalidation flow. The H/W expects an
  *                    8-bit value so use u8.
- * @device_cpu_is_halted: Flag to indicate whether the device CPU was already
- *                        halted. We can't halt it again because the COMMS
- *                        protocol will throw an error. Relevant only for
- *                        cases where Linux was not loaded to device CPU
  */
 struct gaudi_device {
        int (*cpucp_info_get)(struct hl_device *hdev);
@@ -340,7 +336,6 @@ struct gaudi_device {
        u32                             hw_cap_initialized;
        u8                              multi_msi_mode;
        u8                              mmu_cache_inv_pi;
-       u8                              device_cpu_is_halted;
 };
 
 void gaudi_init_security(struct hl_device *hdev);