habanalabs/gaudi: correct driver events numbering
Subsystems (2)
PATCH 1/3 habanalabs/gaudi: correct driver events numbering
From: Ofir Bitton <[email protected]>
Currently driver sends fc interrupt id to FW instead of using cpu interrupt id. We intend to fix that and keep backward compatibility by using the same interrupt values.
Signed-off-by Ofir Bitton <[email protected]>
Reviewed-by Oded Gabbay <[email protected]>
Signed-off-by Oded Gabbay <[email protected]>
--- drivers/misc/habanalabs/gaudi/gaudi.c | 12 ++++++++---- .../include/gaudi/gaudi_async_events.h | 8 ++++---- .../include/gaudi/gaudi_async_ids_map_extended.h | 16 ++++++++-------- 3 files changed, 20 insertions(+), 16 deletions(-)
drivers/misc/habanalabs/gaudi/gaudi.c +8 -4
@@ -3961,7 +3961,8 @@ static int gaudi_init_cpu_queues(struct hl_device *hdev, u32 cpu_timeout) mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR :le32_to_cpu (dyn_regs->gic_host_pi_upd_irq); -WREG32 (irq_handler_offset,GAUDI_EVENT_PI_UPDATE ); +WREG32 (irq_handler_offset, + gaudi_irq_map_table[GAUDI_EVENT_PI_UPDATE ].cpu_id); err = hl_poll_timeout( hdev,
@@ -4147,7 +4148,8 @@ static void gaudi_hw_fini(struct hl_device *hdev, bool hard_reset) mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR :le32_to_cpu (dyn_regs->gic_host_halt_irq); -WREG32 (irq_handler_offset,GAUDI_EVENT_HALT_MACHINE ); +WREG32 (irq_handler_offset, + 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);
@@ -4600,7 +4602,8 @@ static void gaudi_ring_doorbell(struct hl_device *hdev, u32 hw_queue_id, u32 pi) mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR :le32_to_cpu (dyn_regs->gic_host_pi_upd_irq); -WREG32 (irq_handler_offset,GAUDI_EVENT_PI_UPDATE ); +WREG32 (irq_handler_offset, + gaudi_irq_map_table[GAUDI_EVENT_PI_UPDATE ].cpu_id); } }
@@ -8989,7 +8992,8 @@ static void gaudi_enable_events_from_fw(struct hl_device *hdev) mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR :le32_to_cpu (dyn_regs->gic_host_ints_irq); -WREG32 (irq_handler_offset,GAUDI_EVENT_INTS_REGISTER ); +WREG32 (irq_handler_offset, + gaudi_irq_map_table[GAUDI_EVENT_INTS_REGISTER ].cpu_id); }static int gaudi_map_pll_idx_to_fw_idx (u32 pll_idx)
drivers/misc/habanalabs/include/gaudi/gaudi_async_events.h +4 -4
@@ -252,10 +252,6 @@ enum gaudi_async_event_id {GAUDI_EVENT_HBM3_SPI_0 =407 ,GAUDI_EVENT_HBM3_SPI_1 =408 ,GAUDI_EVENT_PSOC_GPIO_U16_0 =421 , -GAUDI_EVENT_PI_UPDATE =484 , -GAUDI_EVENT_HALT_MACHINE =485 , -GAUDI_EVENT_INTS_REGISTER =486 , -GAUDI_EVENT_SOFT_RESET =487 ,GAUDI_EVENT_RAZWI_OR_ADC =548 ,GAUDI_EVENT_TPC0_QM =572 , GAUDI_EVENT_TPC1_QM = 573,
@@ -303,6 +299,10 @@ enum gaudi_async_event_id {GAUDI_EVENT_NIC3_QP1 =619 ,GAUDI_EVENT_NIC4_QP0 =620 ,GAUDI_EVENT_NIC4_QP1 =621 , +GAUDI_EVENT_PI_UPDATE =635 , +GAUDI_EVENT_HALT_MACHINE =636 , +GAUDI_EVENT_INTS_REGISTER =637 , +GAUDI_EVENT_SOFT_RESET =638 ,GAUDI_EVENT_FW_ALIVE_S =645 ,GAUDI_EVENT_DEV_RESET_REQ =646 , GAUDI_EVENT_PKT_QUEUE_OUT_SYNC = 647,
drivers/misc/habanalabs/include/gaudi/gaudi_async_ids_map_extended.h +8 -8
@@ -508,10 +508,10 @@ static struct gaudi_async_events_ids_map gaudi_irq_map_table[] = { { .fc_id = 481, .cpu_id = 330, .valid = 0, .name = "" }, { .fc_id = 482, .cpu_id = 331, .valid = 0, .name = "" }, { .fc_id = 483, .cpu_id = 332, .valid = 0, .name = "" }, - { .fc_id = 484, .cpu_id = 333, .valid = 1, .name = "PI_UPDATE" }, - { .fc_id = 485, .cpu_id = 334, .valid = 1, .name = "HALT_MACHINE" }, - { .fc_id = 486, .cpu_id = 335, .valid = 1, .name = "INTS_REGISTER" }, - { .fc_id = 487, .cpu_id = 336, .valid = 1, .name = "SOFT_RESET" }, + { .fc_id = 484, .cpu_id = 333, .valid = 0, .name = "" }, + { .fc_id = 485, .cpu_id = 334, .valid = 0, .name = "" }, + { .fc_id = 486, .cpu_id = 335, .valid = 0, .name = "" }, + { .fc_id = 487, .cpu_id = 336, .valid = 0, .name = "" }, { .fc_id = 488, .cpu_id = 337, .valid = 0, .name = "" }, { .fc_id = 489, .cpu_id = 338, .valid = 0, .name = "" }, { .fc_id = 490, .cpu_id = 339, .valid = 0, .name = "" },
@@ -659,10 +659,10 @@ static struct gaudi_async_events_ids_map gaudi_irq_map_table[] = { { .fc_id = 632, .cpu_id = 481, .valid = 0, .name = "" }, { .fc_id = 633, .cpu_id = 482, .valid = 0, .name = "" }, { .fc_id = 634, .cpu_id = 483, .valid = 0, .name = "" }, - { .fc_id = 635, .cpu_id = 484, .valid = 0, .name = "" }, - { .fc_id = 636, .cpu_id = 485, .valid = 0, .name = "" }, - { .fc_id = 637, .cpu_id = 486, .valid = 0, .name = "" }, - { .fc_id = 638, .cpu_id = 487, .valid = 0, .name = "" }, + { .fc_id = 635, .cpu_id = 484, .valid = 1, .name = "PI_UPDATE" }, + { .fc_id = 636, .cpu_id = 485, .valid = 1, .name = "HALT_MACHINE" }, + { .fc_id = 637, .cpu_id = 486, .valid = 1, .name = "INTS_REGISTER" }, + { .fc_id = 638, .cpu_id = 487, .valid = 1, .name = "SOFT_RESET" }, { .fc_id = 639, .cpu_id = 488, .valid = 0, .name = "" }, { .fc_id = 640, .cpu_id = 489, .valid = 0, .name = "" }, { .fc_id = 641, .cpu_id = 490, .valid = 0, .name = "" },
Signature · 1 line
2.25.1
PATCH 2/3 habanalabs: fix typo
From: Zvika Yehudai <[email protected]>
fix a type in comment
Signed-off-by Zvika Yehudai <[email protected]>
Reviewed-by Oded Gabbay <[email protected]>
Signed-off-by Oded Gabbay <[email protected]>
--- drivers/misc/habanalabs/common/habanalabs.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/misc/habanalabs/common/habanalabs.h +1 -1
@@ -2119,7 +2119,7 @@ struct hl_mmu_funcs {* @kernel_queues: array of hl_hw_queue. * @cs_mirror_list: CS mirror list for TDR. * @cs_mirror_lock: protects cs_mirror_list. -* @kernel_cb_mgr: command buffer manager for creating/destroying/handling CGs. +* @kernel_cb_mgr: command buffer manager for creating/destroying/handling CBs. * @event_queue: event queue for IRQ from CPU-CP. * @dma_pool: DMA pool for small allocations. * @cpu_accessible_dma_mem: Host <-> CPU-CP shared memory CPU address.
Signature · 1 line
2.25.1
PATCH 3/3 debugfs: add skip_reset_on_timeout option
In reply to Oded GabbayFrom: Yuri Nudelman <[email protected]>
To be able to debug long-running CS better, without changing the userspace code, we are adding a new option through debugfs interface to skip the reset of the device in case of CS timeout.
Signed-off-by Yuri Nudelman <[email protected]>
Reviewed-by Oded Gabbay <[email protected]>
Signed-off-by Oded Gabbay <[email protected]>
--- Documentation/ABI/testing/debugfs-driver-habanalabs | 8 ++++++++ drivers/misc/habanalabs/common/command_submission.c | 1 + drivers/misc/habanalabs/common/debugfs.c | 5 +++++ drivers/misc/habanalabs/common/habanalabs.h | 3 +++ 4 files changed, 17 insertions(+)
Documentation/ABI/testing/debugfs-driver-habanalabs +8 -0
@@ -207,6 +207,14 @@ Contact: [email protected] Description: Sets the PCI power state. Valid values are "1" for D0 and "2" for D3Hot +What: /sys/kernel/debug/habanalabs/hl<n>/skip_reset_on_timeout +Date: Jun 2021 +KernelVersion: 5.13 +Contact: [email protected] +Description: Sets the skip reset on timeout option for the device. Value of + "0" means device will be reset in case some CS has timed out, + otherwise it will not be reset. + What: /sys/kernel/debug/habanalabs/hl<n>/stop_on_err Date: Mar 2020 KernelVersion: 5.6
drivers/misc/habanalabs/common/command_submission.c +1 -0
@@ -663,6 +663,7 @@ static int allocate_cs(struct hl_device *hdev, struct hl_ctx *ctx, cs->timestamp = !!(flags &HL_CS_FLAGS_TIMESTAMP ); cs->timeout_jiffies = timeout; cs->skip_reset_on_timeout = + hdev->skip_reset_on_timeout || !!(flags &HL_CS_FLAGS_SKIP_RESET_ON_TIMEOUT ); cs->submission_time_jiffies = jiffies;INIT_LIST_HEAD (&cs->job_list);
drivers/misc/habanalabs/common/debugfs.c +5 -0
@@ -1278,6 +1278,11 @@ void hl_debugfs_add_device(struct hl_device *hdev) dev_entry->root, &dev_entry->blob_desc); +debugfs_create_x8 ("skip_reset_on_timeout" , +0644 , + dev_entry->root, + &hdev->skip_reset_on_timeout); +for (i =0 , entry = dev_entry->entry_arr ; i < count ; i++, entry++) { debugfs_create_file(hl_debugfs_list[i].name, 0444,
drivers/misc/habanalabs/common/habanalabs.h +3 -0
@@ -2222,6 +2222,8 @@ struct hl_mmu_funcs {* @supports_staged_submission: true if staged submissions are supported * @curr_reset_cause: saves an enumerated reset cause when a hard reset is * 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. */ struct hl_device {struct pci_dev *pdev;
@@ -2337,6 +2339,7 @@ struct hl_device {u8 device_fini_pending;u8 supports_staged_submission;u8 curr_reset_cause; +u8 skip_reset_on_timeout;/* Parameters for bring-up */ u64 nic_ports_mask;
Signature · 1 line
2.25.1