habanalabs: advanced FW loading
PATCH 1/1 habanalabs: advanced FW loading
From: Ofir Bitton <[email protected]>
Today driver is able to load a whole FW binary into a specific location on ASIC. We add support for loading sections from the same FW binary into different loactions.
--- drivers/misc/habanalabs/common/firmware_if.c | 21 ++++++++++++++++---- drivers/misc/habanalabs/common/habanalabs.h | 2 +- drivers/misc/habanalabs/gaudi/gaudi.c | 4 ++-- drivers/misc/habanalabs/goya/goya.c | 4 ++-- 4 files changed, 22 insertions(+), 9 deletions(-)
drivers/misc/habanalabs/common/firmware_if.c +17 -4
@@ -20,16 +20,18 @@* @hdev: pointer to hl_device structure. * @fw_name: the firmware image name * @dst: IO memory mapped address space to copy firmware to +* @src_offset: offset in src FW to copy from +* @size: amount of bytes to copy (0 to copy the whole binary) * * Copy fw code from firmware file to device memory. * * Return: 0 on success, non-zero for failure. */ int hl_fw_load_fw_to_device (struct hl_device *hdev,const char *fw_name, -void __iomem *dst) +void __iomem *dst,u32 src_offset,u32 size) {const struct firmware *fw; -const u64 *fw_data; +const void *fw_data;size_t fw_size;int rc;
@@ -57,9 +59,20 @@ int hl_fw_load_fw_to_device(struct hl_device *hdev, const char *fw_name, goto out; } - fw_data = (const u64 *) fw->data; +if (size - src_offset > fw_size) { +dev_err (hdev->dev, +"size to copy(%u) and offset(%u) are invalid\n" , + size, src_offset); + rc = -EINVAL ; + goto out; + } + +if (size) + fw_size = size; + + fw_data = (const void *) fw->data; -memcpy_toio (dst, fw_data, fw_size); +memcpy_toio (dst, fw_data + src_offset, fw_size); out:release_firmware (fw);
drivers/misc/habanalabs/common/habanalabs.h +1 -1
@@ -2028,7 +2028,7 @@ int hl_mmu_if_set_funcs(struct hl_device *hdev);void hl_mmu_v1_set_funcs (struct hl_device *hdev);int hl_fw_load_fw_to_device (struct hl_device *hdev,const char *fw_name, -void __iomem *dst); +void __iomem *dst,u32 src_offset,u32 size);int hl_fw_send_pci_access_msg (struct hl_device *hdev,u32 opcode);int hl_fw_send_cpu_message (struct hl_device *hdev,u32 hw_queue_id,u32 *msg,u16 len,u32 timeout,long *result);
drivers/misc/habanalabs/gaudi/gaudi.c +2 -2
@@ -3541,7 +3541,7 @@ static int gaudi_load_firmware_to_device(struct hl_device *hdev) dst = hdev->pcie_bar[HBM_BAR_ID ] +LINUX_FW_OFFSET ; -return hl_fw_load_fw_to_device (hdev,GAUDI_LINUX_FW_FILE , dst); +return hl_fw_load_fw_to_device (hdev,GAUDI_LINUX_FW_FILE , dst,0 ,0 ); }static int gaudi_load_boot_fit_to_device (struct hl_device *hdev)
@@ -3550,7 +3550,7 @@ static int gaudi_load_boot_fit_to_device(struct hl_device *hdev) dst = hdev->pcie_bar[SRAM_BAR_ID ] +BOOT_FIT_SRAM_OFFSET ; -return hl_fw_load_fw_to_device (hdev,GAUDI_BOOT_FIT_FILE , dst); +return hl_fw_load_fw_to_device (hdev,GAUDI_BOOT_FIT_FILE , dst,0 ,0 ); } static void gaudi_read_device_fw_version(struct hl_device *hdev,
drivers/misc/habanalabs/goya/goya.c +2 -2
@@ -2315,7 +2315,7 @@ static int goya_load_firmware_to_device(struct hl_device *hdev) dst = hdev->pcie_bar[DDR_BAR_ID ] +LINUX_FW_OFFSET ; -return hl_fw_load_fw_to_device (hdev,GOYA_LINUX_FW_FILE , dst); +return hl_fw_load_fw_to_device (hdev,GOYA_LINUX_FW_FILE , dst,0 ,0 ); }/*
@@ -2332,7 +2332,7 @@ static int goya_load_boot_fit_to_device(struct hl_device *hdev) dst = hdev->pcie_bar[SRAM_CFG_BAR_ID ] +BOOT_FIT_SRAM_OFFSET ; -return hl_fw_load_fw_to_device (hdev,GOYA_BOOT_FIT_FILE , dst); +return hl_fw_load_fw_to_device (hdev,GOYA_BOOT_FIT_FILE , dst,0 ,0 ); }/*
Signature · 1 line
2.17.1
PATCH 1/2 habanalabs: fetch security indication from FW
From: Ofir Bitton <[email protected]>
Add support for fetching security indication from FW. This indication is needed in order to skip unnecessary initializations done by FW.
--- drivers/misc/habanalabs/common/firmware_if.c | 60 ++++++++++++--- drivers/misc/habanalabs/common/habanalabs.h | 23 ++++-- drivers/misc/habanalabs/common/pci.c | 10 ++- drivers/misc/habanalabs/gaudi/gaudi.c | 15 ++-- drivers/misc/habanalabs/goya/goya.c | 17 +++-- .../misc/habanalabs/include/common/cpucp_if.h | 30 ++++++++ .../habanalabs/include/common/hl_boot_if.h | 74 +++++++++++++++++++ .../habanalabs/include/gaudi/gaudi_reg_map.h | 2 + .../habanalabs/include/goya/goya_reg_map.h | 2 + 9 files changed, 204 insertions(+), 29 deletions(-)
drivers/misc/habanalabs/common/firmware_if.c +51 -9
@@ -448,9 +448,10 @@ int hl_fw_cpucp_total_energy_get(struct hl_device *hdev, u64 *total_energy)return rc; } -static void fw_read_errors (struct hl_device *hdev,u32 boot_err0_reg) +static void fw_read_errors (struct hl_device *hdev,u32 boot_err0_reg, +u32 cpu_security_boot_status_reg) { -u32 err_val; +u32 err_val, security_val;/* Some of the firmware status codes are deprecated in newer f/w * versions. In those versions, the errors are reported
@@ -485,6 +486,11 @@ static void fw_read_errors(struct hl_device *hdev, u32 boot_err0_reg)if (err_val &CPU_BOOT_ERR0_NIC_FW_FAIL )dev_err (hdev->dev,"Device boot error - NIC F/W initialization failed\n" ); + + security_val =RREG32 (cpu_security_boot_status_reg); +if (security_val &CPU_BOOT_DEV_STS0_ENABLED ) +dev_info (hdev->dev,"Device security status %#x\n" , + security_val); }static void detect_cpu_boot_status (struct hl_device *hdev,u32 status)
@@ -537,10 +543,12 @@ static void detect_cpu_boot_status(struct hl_device *hdev, u32 status) } } -int hl_fw_read_preboot_ver (struct hl_device *hdev,u32 cpu_boot_status_reg, -u32 boot_err0_reg,u32 timeout) +int hl_fw_read_preboot_status (struct hl_device *hdev,u32 cpu_boot_status_reg, +u32 cpu_security_boot_status_reg,u32 boot_err0_reg, +u32 timeout) { -u32 status; +struct asic_fixed_properties *prop = &hdev->asic_prop; +u32 status, security_status;int rc;if (!hdev->cpu_enable)
@@ -570,19 +578,43 @@ int hl_fw_read_preboot_ver(struct hl_device *hdev, u32 cpu_boot_status_reg,if (rc) {dev_err (hdev->dev,"Failed to read preboot version\n" );detect_cpu_boot_status (hdev, status); -fw_read_errors (hdev, boot_err0_reg); +fw_read_errors (hdev, boot_err0_reg, + cpu_security_boot_status_reg);return -EIO ; } hdev->asic_funcs->read_device_fw_version (hdev,FW_COMP_PREBOOT ); + security_status =RREG32 (cpu_security_boot_status_reg); + +/* We read security status multiple times during boot: +* 1. preboot - we check if fw security feature is supported +* 2. boot cpu - we get boot cpu security status +* 3. FW application - we get FW application security status +* +* Preboot: +* Check security status bit (CPU_BOOT_DEV_STS0_ENABLED), if it is set +* check security enabled bit (CPU_BOOT_DEV_STS0_SECURITY_EN) +*/ +if (security_status &CPU_BOOT_DEV_STS0_ENABLED ) { + hdev->asic_prop.fw_security_status_valid =1 ; + prop->fw_security_disabled = + !(security_status &CPU_BOOT_DEV_STS0_SECURITY_EN ); + }else { + hdev->asic_prop.fw_security_status_valid =0 ; + prop->fw_security_disabled = true; + } + +dev_info (hdev->dev,"firmware-level security is %s\n" , + prop->fw_security_disabled ?"disabled" :"enabled" ); +return 0 ; }int hl_fw_init_cpu (struct hl_device *hdev,u32 cpu_boot_status_reg,u32 msg_to_cpu_reg,u32 cpu_msg_status_reg, -u32 boot_err0_reg,bool skip_bmc, -u32 cpu_timeout,u32 boot_fit_timeout) +u32 cpu_security_boot_status_reg,u32 boot_err0_reg, +bool skip_bmc,u32 cpu_timeout,u32 boot_fit_timeout) {u32 status;int rc;
@@ -652,6 +684,11 @@ int hl_fw_init_cpu(struct hl_device *hdev, u32 cpu_boot_status_reg,/* Read U-Boot version now in case we will later fail */ hdev->asic_funcs->read_device_fw_version (hdev,FW_COMP_UBOOT ); +/* Read boot_cpu security bits */ +if (hdev->asic_prop.fw_security_status_valid) + hdev->asic_prop.fw_boot_cpu_security_map = +RREG32 (cpu_security_boot_status_reg); +if (rc) {detect_cpu_boot_status (hdev, status); rc = -EIO ;
@@ -720,10 +757,15 @@ int hl_fw_init_cpu(struct hl_device *hdev, u32 cpu_boot_status_reg, goto out; } +/* Read FW application security bits */ +if (hdev->asic_prop.fw_security_status_valid) + hdev->asic_prop.fw_app_security_map = +RREG32 (cpu_security_boot_status_reg); +dev_info (hdev->dev,"Successfully loaded firmware to device\n" ); out: -fw_read_errors (hdev, boot_err0_reg); +fw_read_errors (hdev, boot_err0_reg, cpu_security_boot_status_reg);return rc; }
drivers/misc/habanalabs/common/habanalabs.h +18 -5
@@ -372,6 +372,12 @@ struct hl_mmu_properties {* @cb_pool_cb_size: size of each CB in the CB pool. * @max_pending_cs: maximum of concurrent pending command submissions * @max_queues: maximum amount of queues in the system +* @fw_boot_cpu_security_map: bitmap representation of boot cpu security status +* reported by FW, bit description can be found in +* CPU_BOOT_DEV_STS* +* @fw_app_security_map: bitmap representation of application security status +* reported by FW, bit description can be found in +* CPU_BOOT_DEV_STS* * @collective_first_sob: first sync object available for collective use * @collective_first_mon: first monitor available for collective use * @sync_stream_first_sob: first sync object available for sync stream use
@@ -382,6 +388,8 @@ struct hl_mmu_properties {* @completion_queues_count: number of completion queues. * @fw_security_disabled: true if security measures are disabled in firmware, * false otherwise +* @fw_security_status_valid: security status bits are valid and can be fetched +* from BOOT_DEV_STS0 */ struct asic_fixed_properties {struct hw_queue_properties *hw_queues_props;
@@ -426,6 +434,8 @@ struct asic_fixed_properties {u32 cb_pool_cb_size;u32 max_pending_cs;u32 max_queues; +u32 fw_boot_cpu_security_map; +u32 fw_app_security_map;u16 collective_first_sob;u16 collective_first_mon;u16 sync_stream_first_sob;
@@ -435,6 +445,7 @@ struct asic_fixed_properties {u8 tpc_enabled_mask;u8 completion_queues_count;u8 fw_security_disabled; +u8 fw_security_status_valid; };/**
@@ -2050,10 +2061,11 @@ int hl_fw_cpucp_total_energy_get(struct hl_device *hdev, u64 *total_energy);int hl_fw_init_cpu (struct hl_device *hdev,u32 cpu_boot_status_reg,u32 msg_to_cpu_reg,u32 cpu_msg_status_reg, -u32 boot_err0_reg,bool skip_bmc, -u32 cpu_timeout,u32 boot_fit_timeout); -int hl_fw_read_preboot_ver (struct hl_device *hdev,u32 cpu_boot_status_reg, -u32 boot_err0_reg,u32 timeout); +u32 cpu_security_boot_status_reg,u32 boot_err0_reg, +bool skip_bmc,u32 cpu_timeout,u32 boot_fit_timeout); +int hl_fw_read_preboot_status (struct hl_device *hdev,u32 cpu_boot_status_reg, +u32 cpu_security_boot_status_reg,u32 boot_err0_reg, +u32 timeout);int hl_pci_bars_map (struct hl_device *hdev,const char *const name[3 ],bool is_wc[3 ]);
@@ -2063,7 +2075,8 @@ int hl_pci_set_inbound_region(struct hl_device *hdev, u8 region,int hl_pci_set_outbound_region (struct hl_device *hdev,struct hl_outbound_pci_region *pci_region);int hl_pci_init (struct hl_device *hdev,u32 cpu_boot_status_reg, -u32 boot_err0_reg,u32 preboot_ver_timeout); +u32 cpu_security_boot_status_reg,u32 boot_err0_reg, +u32 preboot_ver_timeout);void hl_pci_fini (struct hl_device *hdev);long hl_get_frequency (struct hl_device *hdev,u32 pll_index,bool curr);
drivers/misc/habanalabs/common/pci.c +7 -3
@@ -339,6 +339,8 @@ static int hl_pci_set_dma_mask(struct hl_device *hdev)* hl_pci_init() - PCI initialization code. * @hdev: Pointer to hl_device structure. * @cpu_boot_status_reg: status register of the device's CPU +* @cpu_security_boot_status_reg: status register of device's CPU security +* configuration * @boot_err0_reg: boot error register of the device's CPU * @preboot_ver_timeout: how much to wait before bailing out on reading * the preboot version
@@ -348,7 +350,8 @@ static int hl_pci_set_dma_mask(struct hl_device *hdev)* Return: 0 on success, non-zero for failure. */ int hl_pci_init (struct hl_device *hdev,u32 cpu_boot_status_reg, -u32 boot_err0_reg,u32 preboot_ver_timeout) +u32 cpu_security_boot_status_reg,u32 boot_err0_reg, +u32 preboot_ver_timeout) {struct pci_dev *pdev = hdev->pdev;int rc;
@@ -384,8 +387,9 @@ int hl_pci_init(struct hl_device *hdev, u32 cpu_boot_status_reg,* version to determine whether we run with a security-enabled firmware * The check will be done in each ASIC's specific code */ - rc =hl_fw_read_preboot_ver (hdev, cpu_boot_status_reg, boot_err0_reg, - preboot_ver_timeout); + rc =hl_fw_read_preboot_status (hdev, cpu_boot_status_reg, + cpu_security_boot_status_reg, boot_err0_reg, + preboot_ver_timeout);if (rc) goto unmap_pci_bars;
drivers/misc/habanalabs/gaudi/gaudi.c +9 -6
@@ -514,6 +514,10 @@ static int gaudi_get_fixed_properties(struct hl_device *hdev) prop->sync_stream_first_mon + (num_sync_stream_queues *HL_RSVD_MONS ); +/* disable fw security for now, set it in a later stage */ + prop->fw_security_disabled = true; + prop->fw_security_status_valid = false; +return 0 ; }
@@ -638,13 +642,13 @@ static int gaudi_early_init(struct hl_device *hdev) prop->dram_pci_bar_size =pci_resource_len (pdev,HBM_BAR_ID ); rc =hl_pci_init (hdev, mmPSOC_GLOBAL_CONF_CPU_BOOT_STATUS, - mmCPU_BOOT_ERR0,GAUDI_BOOT_FIT_REQ_TIMEOUT_USEC ); + mmCPU_BOOT_DEV_STS0, mmCPU_BOOT_ERR0, +GAUDI_BOOT_FIT_REQ_TIMEOUT_USEC );if (rc) goto free_queue_props; -/* GAUDI Firmware does not yet support security */ - prop->fw_security_disabled = true; -dev_info (hdev->dev,"firmware-level security is disabled\n" ); +dev_info (hdev->dev,"firmware-level security is %s\n" , + hdev->asic_prop.fw_security_disabled ?"disabled" :"enabled" );return 0 ;
@@ -2319,7 +2323,6 @@ static void gaudi_init_golden_registers(struct hl_device *hdev)int tpc_id, i;gaudi_init_e2e (hdev); -gaudi_init_hbm_cred (hdev); hdev->asic_funcs->disable_clock_gating (hdev);
@@ -3608,7 +3611,7 @@ static int gaudi_init_cpu(struct hl_device *hdev) rc =hl_fw_init_cpu (hdev, mmPSOC_GLOBAL_CONF_CPU_BOOT_STATUS, mmPSOC_GLOBAL_CONF_KMD_MSG_TO_CPU, mmCPU_CMD_STATUS_TO_HOST, - mmCPU_BOOT_ERR0, + mmCPU_BOOT_DEV_STS0, mmCPU_BOOT_ERR0, !hdev->bmc_enable,GAUDI_CPU_TIMEOUT_USEC ,GAUDI_BOOT_FIT_REQ_TIMEOUT_USEC );
drivers/misc/habanalabs/goya/goya.c +11 -6
@@ -456,6 +456,10 @@ int goya_get_fixed_properties(struct hl_device *hdev) prop->max_pending_cs =GOYA_MAX_PENDING_CS ; +/* disable fw security for now, set it in a later stage */ + prop->fw_security_disabled = true; + prop->fw_security_status_valid = false; +return 0 ; }
@@ -601,14 +605,11 @@ static int goya_early_init(struct hl_device *hdev) prop->dram_pci_bar_size =pci_resource_len (pdev,DDR_BAR_ID ); rc =hl_pci_init (hdev, mmPSOC_GLOBAL_CONF_CPU_BOOT_STATUS, - mmCPU_BOOT_ERR0,GOYA_BOOT_FIT_REQ_TIMEOUT_USEC ); + mmCPU_BOOT_DEV_STS0, mmCPU_BOOT_ERR0, +GOYA_BOOT_FIT_REQ_TIMEOUT_USEC );if (rc) goto free_queue_props; -/* Goya Firmware does not support security */ - prop->fw_security_disabled = true; -dev_info (hdev->dev,"firmware-level security is disabled\n" ); -if (!hdev->pldm) { val =RREG32 (mmPSOC_GLOBAL_CONF_BOOT_STRAP_PINS);if (val &PSOC_GLOBAL_CONF_BOOT_STRAP_PINS_SRIOV_EN_MASK )
@@ -616,6 +617,9 @@ static int goya_early_init(struct hl_device *hdev) "PCI strap is not configured correctly, PCI bus errors may occur\n"); } +dev_info (hdev->dev,"firmware-level security is %s\n" , + hdev->asic_prop.fw_security_disabled ?"disabled" :"enabled" ); +return 0 ; free_queue_props:
@@ -2397,7 +2401,8 @@ static int goya_init_cpu(struct hl_device *hdev) rc =hl_fw_init_cpu (hdev, mmPSOC_GLOBAL_CONF_CPU_BOOT_STATUS, mmPSOC_GLOBAL_CONF_UBOOT_MAGIC, - mmCPU_CMD_STATUS_TO_HOST, mmCPU_BOOT_ERR0, + mmCPU_CMD_STATUS_TO_HOST, + mmCPU_BOOT_DEV_STS0, mmCPU_BOOT_ERR0, false,GOYA_CPU_TIMEOUT_USEC ,GOYA_BOOT_FIT_REQ_TIMEOUT_USEC );
drivers/misc/habanalabs/include/common/cpucp_if.h +30 -0
@@ -420,6 +420,29 @@ enum cpucp_card_types { cpucp_card_type_pmc }; +#define CPUCP_SEC_CONF_ENABLED_SHIFT 0 +#define CPUCP_SEC_CONF_ENABLED_MASK 0x00000001 + +#define CPUCP_SEC_CONF_FLASH_WP_SHIFT 1 +#define CPUCP_SEC_CONF_FLASH_WP_MASK 0x00000002 + +#define CPUCP_SEC_CONF_EEPROM_WP_SHIFT 2 +#define CPUCP_SEC_CONF_EEPROM_WP_MASK 0x00000004 + +/** +* struct cpucp_security_info - Security information. +* @config: configuration bit field +* @keys_num: number of stored keys +* @revoked_keys: revoked keys bit field +* @min_svn: minimal security version +*/ +struct cpucp_security_info { +__u8 config; +__u8 keys_num; +__u8 revoked_keys; +__u8 min_svn; +}; +/** * struct cpucp_info - Info from CpuCP that is necessary to the host's driver * @sensors: available sensors description.
@@ -435,6 +458,7 @@ enum cpucp_card_types {* @cpucp_version: CpuCP S/W version. * @dram_size: available DRAM size. * @card_name: card name that will be displayed in HWMON subsystem on the host +* @sec_info: security information */ struct cpucp_info {struct cpucp_sensor sensors[CPUCP_MAX_SENSORS ];
@@ -450,6 +474,12 @@ struct cpucp_info {__le32 reserved2;__le64 dram_size;char card_name[CARD_NAME_MAX_LEN ]; +__le64 reserved3; +__le64 reserved4; +__u8 reserved5; +__u8 pad[7 ]; +struct cpucp_security_info sec_info; +__le32 reserved6; };struct cpucp_mac_addr {
drivers/misc/habanalabs/include/common/hl_boot_if.h +74 -0
88 lines not on this page · raw message
drivers/misc/habanalabs/include/gaudi/gaudi_reg_map.h +2 -0
@@ -12,6 +12,8 @@* PSOC scratch-pad registers */ #define mmHW_STATE mmPSOC_GLOBAL_CONF_SCRATCHPAD_0 +#define mmCPU_BOOT_DEV_STS0 mmPSOC_GLOBAL_CONF_SCRATCHPAD_20 +#define mmCPU_BOOT_DEV_STS1 mmPSOC_GLOBAL_CONF_SCRATCHPAD_21#define mmFUSE_VER_OFFSET mmPSOC_GLOBAL_CONF_SCRATCHPAD_22#define mmCPU_CMD_STATUS_TO_HOST mmPSOC_GLOBAL_CONF_SCRATCHPAD_23#define mmCPU_BOOT_ERR0 mmPSOC_GLOBAL_CONF_SCRATCHPAD_24
drivers/misc/habanalabs/include/goya/goya_reg_map.h +2 -0
@@ -22,6 +22,8 @@#define mmCPU_CQ_BASE_ADDR_LOW mmPSOC_GLOBAL_CONF_SCRATCHPAD_8#define mmCPU_CQ_BASE_ADDR_HIGH mmPSOC_GLOBAL_CONF_SCRATCHPAD_9#define mmCPU_CQ_LENGTH mmPSOC_GLOBAL_CONF_SCRATCHPAD_10 +#define mmCPU_BOOT_DEV_STS0 mmPSOC_GLOBAL_CONF_SCRATCHPAD_20 +#define mmCPU_BOOT_DEV_STS1 mmPSOC_GLOBAL_CONF_SCRATCHPAD_21#define mmFUSE_VER_OFFSET mmPSOC_GLOBAL_CONF_SCRATCHPAD_22#define mmCPU_CMD_STATUS_TO_HOST mmPSOC_GLOBAL_CONF_SCRATCHPAD_23#define mmCPU_BOOT_ERR0 mmPSOC_GLOBAL_CONF_SCRATCHPAD_24
Signature · 1 line
2.17.1
PATCH 1/1 habanalabs: fix cs counters structure
In reply to Oded GabbayFrom: farah kassabri <[email protected]>
Fix cs counters structure in uapi to be one flat structure instead of two instances of the same other structure.
use atomic read/increment for context counters so we could use one structure for both aggregated and context counters.
--- .../habanalabs/common/command_submission.c | 18 +++-- drivers/misc/habanalabs/common/habanalabs.h | 73 ++++++++++--------- .../misc/habanalabs/common/habanalabs_ioctl.c | 35 ++++++--- drivers/misc/habanalabs/common/hw_queue.c | 5 +- drivers/misc/habanalabs/gaudi/gaudi.c | 4 +- include/uapi/misc/habanalabs.h | 35 +++++---- 6 files changed, 95 insertions(+), 75 deletions(-)
drivers/misc/habanalabs/common/command_submission.c +10 -8
@@ -462,7 +462,7 @@ static int allocate_cs(struct hl_device *hdev, struct hl_ctx *ctx,if (other && !completion_done (&other->completion)) {dev_dbg_ratelimited (hdev->dev,"Rejecting CS because of too many in-flights CS\n" ); - ctx->cs_counters.max_cs_in_flight_drop_cnt++; +atomic64_inc (&ctx->cs_counters.max_cs_in_flight_drop_cnt);atomic64_inc (&cntr->max_cs_in_flight_drop_cnt); rc = -EAGAIN ; goto free_fence;
@@ -720,7 +720,7 @@ static int cs_ioctl_default(struct hl_fpriv *hpriv, void __user *chunks, rc =validate_queue_index (hdev, chunk, &queue_type, &is_kernel_allocated_cb);if (rc) { - hpriv->ctx->cs_counters.parsing_drop_cnt++; +atomic64_inc (&hpriv->ctx->cs_counters.parsing_drop_cnt);atomic64_inc (&cntr->parsing_drop_cnt); goto free_cs_object; }
@@ -728,7 +728,8 @@ static int cs_ioctl_default(struct hl_fpriv *hpriv, void __user *chunks,if (is_kernel_allocated_cb) { cb =get_cb_from_cs_chunk (hdev, &hpriv->cb_mgr, chunk);if (!cb) { - hpriv->ctx->cs_counters.parsing_drop_cnt++; +atomic64_inc ( + &hpriv->ctx->cs_counters.parsing_drop_cnt);atomic64_inc (&cntr->parsing_drop_cnt); rc = -EINVAL ; goto free_cs_object;
@@ -743,7 +744,8 @@ static int cs_ioctl_default(struct hl_fpriv *hpriv, void __user *chunks, job =hl_cs_allocate_job (hdev, queue_type, is_kernel_allocated_cb);if (!job) { - hpriv->ctx->cs_counters.out_of_mem_drop_cnt++; +atomic64_inc ( + &hpriv->ctx->cs_counters.out_of_mem_drop_cnt);atomic64_inc (&cntr->out_of_mem_drop_cnt);dev_err (hdev->dev,"Failed to allocate a new job\n" ); rc = -ENOMEM ;
@@ -777,7 +779,7 @@ static int cs_ioctl_default(struct hl_fpriv *hpriv, void __user *chunks, rc =cs_parser (hpriv, job);if (rc) { - hpriv->ctx->cs_counters.parsing_drop_cnt++; +atomic64_inc (&hpriv->ctx->cs_counters.parsing_drop_cnt);atomic64_inc (&cntr->parsing_drop_cnt); dev_err(hdev->dev, "Failed to parse JOB %d.%llu.%d, err %d, rejecting the CS\n",
@@ -787,7 +789,7 @@ static int cs_ioctl_default(struct hl_fpriv *hpriv, void __user *chunks, }if (int_queues_only) { - hpriv->ctx->cs_counters.parsing_drop_cnt++; +atomic64_inc (&hpriv->ctx->cs_counters.parsing_drop_cnt);atomic64_inc (&cntr->parsing_drop_cnt); dev_err(hdev->dev, "Reject CS %d.%llu because only internal queues jobs are present\n",
@@ -880,7 +882,7 @@ static int cs_ioctl_signal_wait_create_jobs(struct hl_device *hdev, job =hl_cs_allocate_job (hdev, q_type, true);if (!job) { - ctx->cs_counters.out_of_mem_drop_cnt++; +atomic64_inc (&ctx->cs_counters.out_of_mem_drop_cnt);atomic64_inc (&cntr->out_of_mem_drop_cnt);dev_err (hdev->dev,"Failed to allocate a new job\n" );return -ENOMEM ;
@@ -894,7 +896,7 @@ static int cs_ioctl_signal_wait_create_jobs(struct hl_device *hdev, cb =hl_cb_kernel_create (hdev, cb_size, q_type ==QUEUE_TYPE_HW && hdev->mmu_enable);if (!cb) { - ctx->cs_counters.out_of_mem_drop_cnt++; +atomic64_inc (&ctx->cs_counters.out_of_mem_drop_cnt);atomic64_inc (&cntr->out_of_mem_drop_cnt);kfree (job);return -EFAULT ;
drivers/misc/habanalabs/common/habanalabs.h +37 -36
97 lines not on this page · raw message
drivers/misc/habanalabs/common/habanalabs_ioctl.c +23 -12
@@ -315,7 +315,7 @@ static int clk_throttle_info(struct hl_fpriv *hpriv, struct hl_info_args *args)static int cs_counters_info (struct hl_fpriv *hpriv,struct hl_info_args *args) {void __user *out = (void __user *) (uintptr_t ) args->return_pointer; -struct hl_info_cs_counters cs_counters = { {0 } }; +struct hl_info_cs_counters cs_counters = {0 };struct hl_device *hdev = hpriv->hdev;struct hl_cs_counters_atomic *cntr;u32 max_size = args->return_size;
@@ -325,23 +325,34 @@ static int cs_counters_info(struct hl_fpriv *hpriv, struct hl_info_args *args)if ((!max_size) || (!out))return -EINVAL ; -memcpy (&cs_counters.cs_counters, &hdev->aggregated_cs_counters, -sizeof (struct hl_cs_counters )); - - cs_counters.cs_counters.out_of_mem_drop_cnt = + cs_counters.total_out_of_mem_drop_cnt =atomic64_read (&cntr->out_of_mem_drop_cnt); - cs_counters.cs_counters.parsing_drop_cnt = + cs_counters.total_parsing_drop_cnt =atomic64_read (&cntr->parsing_drop_cnt); - cs_counters.cs_counters.queue_full_drop_cnt = + cs_counters.total_queue_full_drop_cnt =atomic64_read (&cntr->queue_full_drop_cnt); - cs_counters.cs_counters.device_in_reset_drop_cnt = + cs_counters.total_device_in_reset_drop_cnt =atomic64_read (&cntr->device_in_reset_drop_cnt); - cs_counters.cs_counters.max_cs_in_flight_drop_cnt = + cs_counters.total_max_cs_in_flight_drop_cnt =atomic64_read (&cntr->max_cs_in_flight_drop_cnt); -if (hpriv->ctx) -memcpy (&cs_counters.ctx_cs_counters, &hpriv->ctx->cs_counters, -sizeof (struct hl_cs_counters )); +if (hpriv->ctx) { + cs_counters.ctx_out_of_mem_drop_cnt = +atomic64_read ( + &hpriv->ctx->cs_counters.out_of_mem_drop_cnt); + cs_counters.ctx_parsing_drop_cnt = +atomic64_read ( + &hpriv->ctx->cs_counters.parsing_drop_cnt); + cs_counters.ctx_queue_full_drop_cnt = +atomic64_read ( + &hpriv->ctx->cs_counters.queue_full_drop_cnt); + cs_counters.ctx_device_in_reset_drop_cnt = +atomic64_read ( + &hpriv->ctx->cs_counters.device_in_reset_drop_cnt); + cs_counters.ctx_max_cs_in_flight_drop_cnt = +atomic64_read ( + &hpriv->ctx->cs_counters.max_cs_in_flight_drop_cnt); + }return copy_to_user (out, &cs_counters,min ((size_t ) max_size,sizeof (cs_counters))) ? -EFAULT :0 ;
drivers/misc/habanalabs/common/hw_queue.c +3 -2
@@ -524,7 +524,7 @@ int hl_hw_queue_schedule_cs(struct hl_cs *cs) hdev->asic_funcs->hw_queues_lock (hdev);if (hl_device_disabled_or_in_reset (hdev)) { - ctx->cs_counters.device_in_reset_drop_cnt++; +atomic64_inc (&ctx->cs_counters.device_in_reset_drop_cnt);atomic64_inc (&cntr->device_in_reset_drop_cnt);dev_err (hdev->dev,"device is disabled or in reset, CS rejected!\n" );
@@ -558,7 +558,8 @@ int hl_hw_queue_schedule_cs(struct hl_cs *cs) }if (rc) { - ctx->cs_counters.queue_full_drop_cnt++; +atomic64_inc ( + &ctx->cs_counters.queue_full_drop_cnt);atomic64_inc (&cntr->queue_full_drop_cnt); goto unroll_cq_resv; }
drivers/misc/habanalabs/gaudi/gaudi.c +2 -2
@@ -1141,7 +1141,7 @@ static int gaudi_collective_wait_create_job(struct hl_device *hdev, hw_queue_prop = &hdev->asic_prop.hw_queues_props[queue_id]; job =hl_cs_allocate_job (hdev, hw_queue_prop->type, true);if (!job) { - ctx->cs_counters.out_of_mem_drop_cnt++; +atomic64_inc (&ctx->cs_counters.out_of_mem_drop_cnt);atomic64_inc (&cntr->out_of_mem_drop_cnt);dev_err (hdev->dev,"Failed to allocate a new job\n" );return -ENOMEM ;
@@ -1151,7 +1151,7 @@ static int gaudi_collective_wait_create_job(struct hl_device *hdev, cb =hl_cb_kernel_create (hdev, cb_size, hdev->mmu_enable && !patched_cb);if (!cb) { - ctx->cs_counters.out_of_mem_drop_cnt++; +atomic64_inc (&ctx->cs_counters.out_of_mem_drop_cnt);atomic64_inc (&cntr->out_of_mem_drop_cnt);kfree (job);return -EFAULT ;
include/uapi/misc/habanalabs.h +20 -15
@@ -407,23 +407,28 @@ struct hl_info_sync_manager {/** * struct hl_info_cs_counters - command submission counters -* @out_of_mem_drop_cnt: dropped due to memory allocation issue -* @parsing_drop_cnt: dropped due to error in packet parsing -* @queue_full_drop_cnt: dropped due to queue full -* @device_in_reset_drop_cnt: dropped due to device in reset -* @max_cs_in_flight_drop_cnt: dropped due to maximum CS in-flight +* @total_out_of_mem_drop_cnt: total dropped due to memory allocation issue +* @ctx_out_of_mem_drop_cnt: context dropped due to memory allocation issue +* @total_parsing_drop_cnt: total dropped due to error in packet parsing +* @ctx_parsing_drop_cnt: context dropped due to error in packet parsing +* @total_queue_full_drop_cnt: total dropped due to queue full +* @ctx_queue_full_drop_cnt: context dropped due to queue full +* @total_device_in_reset_drop_cnt: total dropped due to device in reset +* @ctx_device_in_reset_drop_cnt: context dropped due to device in reset +* @total_max_cs_in_flight_drop_cnt: total dropped due to maximum CS in-flight +* @ctx_max_cs_in_flight_drop_cnt: context dropped due to maximum CS in-flight */ -struct hl_cs_counters { -__u64 out_of_mem_drop_cnt; -__u64 parsing_drop_cnt; -__u64 queue_full_drop_cnt; -__u64 device_in_reset_drop_cnt; -__u64 max_cs_in_flight_drop_cnt; -}; -struct hl_info_cs_counters { -struct hl_cs_counters cs_counters; -struct hl_cs_counters ctx_cs_counters; +__u64 total_out_of_mem_drop_cnt; +__u64 ctx_out_of_mem_drop_cnt; +__u64 total_parsing_drop_cnt; +__u64 ctx_parsing_drop_cnt; +__u64 total_queue_full_drop_cnt; +__u64 ctx_queue_full_drop_cnt; +__u64 total_device_in_reset_drop_cnt; +__u64 ctx_device_in_reset_drop_cnt; +__u64 total_max_cs_in_flight_drop_cnt; +__u64 ctx_max_cs_in_flight_drop_cnt; };enum gaudi_dcores {
Signature · 1 line
2.17.1
PATCH 1/1 habanalabs/gaudi: remove unreachable code
In reply to Oded GabbayFrom: Ofir Bitton <[email protected]>
Remove unreachable code in gaudi collective flow.
--- drivers/misc/habanalabs/gaudi/gaudi.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/misc/habanalabs/gaudi/gaudi.c +1 -3
@@ -1224,10 +1224,8 @@ static int gaudi_collective_wait_create_jobs(struct hl_device *hdev,if (collective_engine_id ==GAUDI_ENGINE_ID_DMA_5 ) collective_queue =GAUDI_QUEUE_ID_DMA_5_0 + stream; -else if (collective_engine_id ==GAUDI_ENGINE_ID_TPC_7 ) - collective_queue =GAUDI_QUEUE_ID_TPC_7_0 + stream;else -return -EINVAL ; + collective_queue =GAUDI_QUEUE_ID_TPC_7_0 + stream; num_jobs =NUMBER_OF_SOBS_IN_GRP +1 ; nic_queue =GAUDI_QUEUE_ID_NIC_0_0 + stream;
Signature · 1 line
2.17.1
PATCH 1/1 habanalabs: make sure cs type is valid in cs_ioctl_signal_wait
In reply to Oded GabbayAlthough we get a valid cs type from the callee, in case new values will be added in the future, it is best to check the expected values in that function.
--- drivers/misc/habanalabs/common/command_submission.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/misc/habanalabs/common/command_submission.c +3 -1
@@ -1070,9 +1070,11 @@ static int cs_ioctl_signal_wait(struct hl_fpriv *hpriv, enum hl_cs_type cs_type,if (cs_type ==CS_TYPE_WAIT || cs_type ==CS_TYPE_SIGNAL ) rc =cs_ioctl_signal_wait_create_jobs (hdev, ctx, cs, q_type, q_idx); -else +else if (cs_type ==CS_TYPE_COLLECTIVE_WAIT ) rc = hdev->asic_funcs->collective_wait_create_jobs (hdev, ctx, cs, q_idx, collective_engine_id); +else + rc = -EINVAL ;if (rc) goto free_cs_object;
Signature · 1 line
2.17.1
PATCH 2/2 habanalabs/gaudi: add support for FW security
In reply to Oded GabbayFrom: Ofir Bitton <[email protected]>
Skip relevant HW configurations once FW security is enabled because these configurations are being performed by FW.
--- drivers/misc/habanalabs/gaudi/gaudi.c | 123 ++++++++++++------ .../misc/habanalabs/gaudi/gaudi_security.c | 83 ++++++------ 2 files changed, 128 insertions(+), 78 deletions(-)
drivers/misc/habanalabs/gaudi/gaudi.c +82 -41
@@ -1410,7 +1410,8 @@ static int gaudi_alloc_cpu_accessible_dma_mem(struct hl_device *hdev) hdev->cpu_pci_msb_addr =GAUDI_CPU_PCI_MSB_ADDR (hdev->cpu_accessible_dma_address); -GAUDI_PCI_TO_CPU_ADDR (hdev->cpu_accessible_dma_address); +if (hdev->asic_prop.fw_security_disabled) +GAUDI_PCI_TO_CPU_ADDR (hdev->cpu_accessible_dma_address); free_dma_mem_arr:for (j =0 ; j < i ; j++)
@@ -1564,8 +1565,9 @@ static int gaudi_sw_init(struct hl_device *hdev) free_cpu_accessible_dma_pool:gen_pool_destroy (hdev->cpu_accessible_dma_pool); free_cpu_dma_mem: -GAUDI_CPU_TO_PCI_ADDR (hdev->cpu_accessible_dma_address, - hdev->cpu_pci_msb_addr); +if (hdev->asic_prop.fw_security_disabled) +GAUDI_CPU_TO_PCI_ADDR (hdev->cpu_accessible_dma_address, + hdev->cpu_pci_msb_addr); hdev->asic_funcs->asic_dma_free_coherent(hdev, HL_CPU_ACCESSIBLE_MEM_SIZE, hdev->cpu_accessible_dma_mem,
@@ -1585,8 +1587,10 @@ static int gaudi_sw_fini(struct hl_device *hdev)gen_pool_destroy (hdev->cpu_accessible_dma_pool); -GAUDI_CPU_TO_PCI_ADDR (hdev->cpu_accessible_dma_address, +if (hdev->asic_prop.fw_security_disabled) +GAUDI_CPU_TO_PCI_ADDR (hdev->cpu_accessible_dma_address, hdev->cpu_pci_msb_addr); + hdev->asic_funcs->asic_dma_free_coherent(hdev, HL_CPU_ACCESSIBLE_MEM_SIZE, hdev->cpu_accessible_dma_mem,
@@ -1772,6 +1776,14 @@ static void gaudi_init_scrambler_sram(struct hl_device *hdev) {struct gaudi_device *gaudi = hdev->asic_specific; +if (!hdev->asic_prop.fw_security_disabled) +return ; + +if (hdev->asic_prop.fw_security_status_valid && + (hdev->asic_prop.fw_app_security_map & +CPU_BOOT_DEV_STS0_SRAM_SCR_EN )) +return ; +if (gaudi->hw_cap_initialized &HW_CAP_SRAM_SCRAMBLER )return ;
@@ -1836,6 +1848,14 @@ static void gaudi_init_scrambler_hbm(struct hl_device *hdev) {struct gaudi_device *gaudi = hdev->asic_specific; +if (!hdev->asic_prop.fw_security_disabled) +return ; + +if (hdev->asic_prop.fw_security_status_valid && + (hdev->asic_prop.fw_boot_cpu_security_map & +CPU_BOOT_DEV_STS0_DRAM_SCR_EN )) +return ; +if (gaudi->hw_cap_initialized &HW_CAP_HBM_SCRAMBLER )return ;
@@ -1898,6 +1918,14 @@ static void gaudi_init_scrambler_hbm(struct hl_device *hdev)static void gaudi_init_e2e (struct hl_device *hdev) { +if (!hdev->asic_prop.fw_security_disabled) +return ; + +if (hdev->asic_prop.fw_security_status_valid && + (hdev->asic_prop.fw_boot_cpu_security_map & +CPU_BOOT_DEV_STS0_E2E_CRED_EN )) +return ; +WREG32 (mmSIF_RTR_CTRL_0_E2E_HBM_WR_SIZE,247 >>3 );WREG32 (mmSIF_RTR_CTRL_0_E2E_HBM_RD_SIZE,785 >>3 );WREG32 (mmSIF_RTR_CTRL_0_E2E_PCI_WR_SIZE,49 );
@@ -2265,6 +2293,14 @@ static void gaudi_init_hbm_cred(struct hl_device *hdev) {uint32_t hbm0_wr, hbm1_wr, hbm0_rd, hbm1_rd; +if (!hdev->asic_prop.fw_security_disabled) +return ; + +if (hdev->asic_prop.fw_security_status_valid && + (hdev->asic_prop.fw_boot_cpu_security_map & +CPU_BOOT_DEV_STS0_HBM_CRED_EN )) +return ; + hbm0_wr =0x33333333 ; hbm0_rd =0x77777777 ; hbm1_wr =0x55555555 ;
@@ -3606,7 +3642,8 @@ static int gaudi_init_cpu(struct hl_device *hdev)* The device CPU works with 40 bits addresses. * This register sets the extension to 50 bits. */ -WREG32 (mmCPU_IF_CPU_MSB_ADDR, hdev->cpu_pci_msb_addr); +if (hdev->asic_prop.fw_security_disabled) +WREG32 (mmCPU_IF_CPU_MSB_ADDR, hdev->cpu_pci_msb_addr); rc = hl_fw_init_cpu(hdev, mmPSOC_GLOBAL_CONF_CPU_BOOT_STATUS, mmPSOC_GLOBAL_CONF_KMD_MSG_TO_CPU,
@@ -3691,17 +3728,19 @@ static void gaudi_pre_hw_init(struct hl_device *hdev)/* Perform read from the device to make sure device is up */ RREG32 (mmPCIE_DBI_DEVICE_ID_VENDOR_ID_REG); -/* Set the access through PCI bars (Linux driver only) as -* secured -*/ -WREG32 (mmPCIE_WRAP_LBW_PROT_OVR, - (PCIE_WRAP_LBW_PROT_OVR_RD_EN_MASK | -PCIE_WRAP_LBW_PROT_OVR_WR_EN_MASK )); +if (hdev->asic_prop.fw_security_disabled) { +/* Set the access through PCI bars (Linux driver only) as +* secured +*/ +WREG32 (mmPCIE_WRAP_LBW_PROT_OVR, + (PCIE_WRAP_LBW_PROT_OVR_RD_EN_MASK | +PCIE_WRAP_LBW_PROT_OVR_WR_EN_MASK )); -/* Perform read to flush the waiting writes to ensure -* configuration was set in the device -*/ -RREG32 (mmPCIE_WRAP_LBW_PROT_OVR); +/* Perform read to flush the waiting writes to ensure +* configuration was set in the device +*/ +RREG32 (mmPCIE_WRAP_LBW_PROT_OVR); + }/* * Let's mark in the H/W that we have reached this point. We check
@@ -3710,32 +3749,33 @@ static void gaudi_pre_hw_init(struct hl_device *hdev)* cleared by the H/W upon H/W reset */ WREG32 (mmHW_STATE,HL_DEVICE_HW_STATE_DIRTY ); +if (hdev->asic_prop.fw_security_disabled) { +/* Configure the reset registers. Must be done as early as +* possible in case we fail during H/W initialization +*/ +WREG32 (mmPSOC_GLOBAL_CONF_SOFT_RST_CFG_H, + (CFG_RST_H_DMA_MASK | +CFG_RST_H_MME_MASK | +CFG_RST_H_SM_MASK | +CFG_RST_H_TPC_7_MASK )); -/* Configure the reset registers. Must be done as early as possible -* in case we fail during H/W initialization -*/ -WREG32 (mmPSOC_GLOBAL_CONF_SOFT_RST_CFG_H, - (CFG_RST_H_DMA_MASK | -CFG_RST_H_MME_MASK | -CFG_RST_H_SM_MASK | -CFG_RST_H_TPC_7_MASK )); - -WREG32 (mmPSOC_GLOBAL_CONF_SOFT_RST_CFG_L,CFG_RST_L_TPC_MASK ); - -WREG32 (mmPSOC_GLOBAL_CONF_SW_ALL_RST_CFG_H, - (CFG_RST_H_HBM_MASK | -CFG_RST_H_TPC_7_MASK | -CFG_RST_H_NIC_MASK | -CFG_RST_H_SM_MASK | -CFG_RST_H_DMA_MASK | -CFG_RST_H_MME_MASK | -CFG_RST_H_CPU_MASK | -CFG_RST_H_MMU_MASK )); - -WREG32 (mmPSOC_GLOBAL_CONF_SW_ALL_RST_CFG_L, - (CFG_RST_L_IF_MASK | -CFG_RST_L_PSOC_MASK | -CFG_RST_L_TPC_MASK )); +WREG32 (mmPSOC_GLOBAL_CONF_SOFT_RST_CFG_L,CFG_RST_L_TPC_MASK ); + +WREG32 (mmPSOC_GLOBAL_CONF_SW_ALL_RST_CFG_H, + (CFG_RST_H_HBM_MASK | +CFG_RST_H_TPC_7_MASK | +CFG_RST_H_NIC_MASK | +CFG_RST_H_SM_MASK | +CFG_RST_H_DMA_MASK | +CFG_RST_H_MME_MASK | +CFG_RST_H_CPU_MASK | +CFG_RST_H_MMU_MASK )); + +WREG32 (mmPSOC_GLOBAL_CONF_SW_ALL_RST_CFG_L, + (CFG_RST_L_IF_MASK | +CFG_RST_L_PSOC_MASK | +CFG_RST_L_TPC_MASK )); + } }static int gaudi_hw_init (struct hl_device *hdev)
@@ -3851,7 +3891,8 @@ static void gaudi_hw_fini(struct hl_device *hdev, bool hard_reset)WREG32 (mmPSOC_GLOBAL_CONF_BOOT_STRAP_PINS, boot_strap & ~0x2 );/* Restart BTL/BLR upon hard-reset */ -WREG32 (mmPSOC_GLOBAL_CONF_BOOT_SEQ_RE_START,1 ); +if (hdev->asic_prop.fw_security_disabled) +WREG32 (mmPSOC_GLOBAL_CONF_BOOT_SEQ_RE_START,1 );WREG32 (mmPSOC_GLOBAL_CONF_SW_ALL_RST,1 <<PSOC_GLOBAL_CONF_SW_ALL_RST_IND_SHIFT );
drivers/misc/habanalabs/gaudi/gaudi_security.c +46 -37
109 lines not on this page · raw message
Signature · 1 line
2.17.1
PATCH 1/1 habanalabs/gaudi: monitor device memory usage
In reply to Oded GabbayIn GAUDI we don't have an MMU towards the HBM device memory. Therefore, the user access that memory directly through physical address (via the different engines) without the need to go through the driver to allocate/free memory on the HBM.
For system monitoring purposes, the driver will keep track of the HBM usage. This can be done as long as the user accurately reports the allocations and releases of HBM memory, through the existing MEMORY IOCTL uapi.
--- drivers/misc/habanalabs/common/memory.c | 56 ++++++++++++++++++++++--- 1 file changed, 50 insertions(+), 6 deletions(-)
drivers/misc/habanalabs/common/memory.c +50 -6
@@ -1236,18 +1236,35 @@ int hl_mem_ioctl(struct hl_fpriv *hpriv, void *data)switch (args->in.op) {case HL_MEM_OP_ALLOC : -if (!hdev->dram_supports_virtual_memory) { -dev_err (hdev->dev,"DRAM alloc is not supported\n" ); - rc = -EINVAL ; - goto out; - } -if (args->in.alloc.mem_size ==0 ) {dev_err (hdev->dev,"alloc size must be larger than 0\n" ); rc = -EINVAL ; goto out; } + +/* If DRAM does not support virtual memory the driver won't +* handle the allocation/freeing of that memory. However, for +* system administration/monitoring purposes, the driver will +* keep track of the amount of DRAM memory that is allocated +* and freed by the user. Because this code totally relies on +* the user's input, the driver can't ensure the validity +* of this accounting. +*/ +if (!hdev->dram_supports_virtual_memory) { +atomic64_add (args->in.alloc.mem_size, + &ctx->dram_phys_mem); +atomic64_add (args->in.alloc.mem_size, + &hdev->dram_used_mem); + +dev_dbg (hdev->dev,"DRAM alloc is not supported\n" ); + rc =0 ; + +memset (args,0 ,sizeof (*args)); + args->out.handle =0 ; + goto out; + } + rc =alloc_device_memory (ctx, &args->in, &handle);memset (args,0 ,sizeof (*args));
@@ -1255,6 +1272,26 @@ int hl_mem_ioctl(struct hl_fpriv *hpriv, void *data)break ;case HL_MEM_OP_FREE : +/* If DRAM does not support virtual memory the driver won't +* handle the allocation/freeing of that memory. However, for +* system administration/monitoring purposes, the driver will +* keep track of the amount of DRAM memory that is allocated +* and freed by the user. Because this code totally relies on +* the user's input, the driver can't ensure the validity +* of this accounting. +*/ +if (!hdev->dram_supports_virtual_memory) { +atomic64_sub (args->in.alloc.mem_size, + &ctx->dram_phys_mem); +atomic64_sub (args->in.alloc.mem_size, + &hdev->dram_used_mem); + +dev_dbg (hdev->dev,"DRAM alloc is not supported\n" ); + rc =0 ; + + goto out; + } + rc =free_device_memory (ctx, args->in.free.handle);break ;
@@ -1773,6 +1810,13 @@ void hl_vm_ctx_fini(struct hl_ctx *ctx)mutex_destroy (&ctx->mem_hash_lock);hl_mmu_ctx_fini (ctx); + +/* In this case we need to clear the global accounting of DRAM usage +* because the user notifies us on allocations. If the user is no more, +* all DRAM is available +*/ +if (!ctx->hdev->dram_supports_virtual_memory) +atomic64_set (&ctx->hdev->dram_used_mem,0 ); }/*
Signature · 1 line
2.17.1
PATCH 1/1 habanalabs: initialize variable before use
In reply to Oded GabbayGCC 7.3.1 20180303 (Red Hat 7.3.1-5) complains that collective_engine_id might be used uninitialized.
--- drivers/misc/habanalabs/common/command_submission.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/misc/habanalabs/common/command_submission.c +1 -1
@@ -942,7 +942,7 @@ static int cs_ioctl_signal_wait(struct hl_fpriv *hpriv, enum hl_cs_type cs_type,struct hl_cs_compl *sig_waitcs_cmpl;struct hl_cs *cs;enum hl_queue_type q_type; -u32 size_to_copy, q_idx, collective_engine_id; +u32 size_to_copy, q_idx, collective_engine_id =0 ;u64 signal_seq;int rc;
Signature · 1 line
2.17.1