habanalabs: Add CB IOCTL opcode to retrieve CB information
PATCH 1/1 habanalabs: Add CB IOCTL opcode to retrieve CB information
From: Tomer Tayar <[email protected]>
Add a new CB IOCTL opcode that enables a user to query about a CB and get its usage count.
--- .../misc/habanalabs/common/command_buffer.c | 38 +++++++++++++++++++ include/uapi/misc/habanalabs.h | 15 +++++++- 2 files changed, 51 insertions(+), 2 deletions(-)
drivers/misc/habanalabs/common/command_buffer.c +38 -0
@@ -375,12 +375,43 @@ int hl_cb_destroy(struct hl_device *hdev, struct hl_cb_mgr *mgr, u64 cb_handle)return rc; } +static int hl_cb_info (struct hl_device *hdev,struct hl_cb_mgr *mgr, +u64 cb_handle,u32 *usage_cnt) +{ +struct hl_cb *cb; +u32 handle; +int rc =0 ; + +/* The CB handle was given to user to do mmap, so need to shift it back +* to the value which was allocated by the IDR module. +*/ + cb_handle >>=PAGE_SHIFT ; + handle = (u32 ) cb_handle; + +spin_lock (&mgr->cb_lock); + + cb =idr_find (&mgr->cb_handles, handle); +if (!cb) { +dev_err (hdev->dev, +"CB info failed, no match to handle 0x%x\n" , handle); + rc = -EINVAL ; + goto out; + } + + *usage_cnt =atomic_read (&cb->cs_cnt); + +out: +spin_unlock (&mgr->cb_lock); +return rc; +} +int hl_cb_ioctl (struct hl_fpriv *hpriv,void *data) {union hl_cb_args *args = data;struct hl_device *hdev = hpriv->hdev;enum hl_device_status status;u64 handle =0 ; +u32 usage_cnt =0 ;int rc;if (!hl_device_operational (hdev, &status)) {
@@ -413,6 +444,13 @@ int hl_cb_ioctl(struct hl_fpriv *hpriv, void *data) args->in.cb_handle);break ; +case HL_CB_OP_INFO : + rc =hl_cb_info (hdev, &hpriv->cb_mgr, args->in.cb_handle, + &usage_cnt); +memset (args,0 ,sizeof (*args)); + args->out.usage_cnt = usage_cnt; +break ; +default : rc = -ENOTTY ;break ;
include/uapi/misc/habanalabs.h +13 -2
@@ -483,6 +483,8 @@ struct hl_info_args {#define HL_CB_OP_CREATE 0/* Opcode to destroy previously created command buffer */ #define HL_CB_OP_DESTROY 1 +/* Opcode to retrieve information about a command buffer */ +#define HL_CB_OP_INFO 2/* 2MB minus 32 bytes for 2xMSG_PROT */ #define HL_MAX_CB_SIZE (0x200000 - 32)
@@ -506,8 +508,17 @@ struct hl_cb_in { };struct hl_cb_out { -/* Handle of CB */ -__u64 cb_handle; +union { +/* Handle of CB */ +__u64 cb_handle; + +/* Information about CB */ +struct { +/* Usage count of CB */ +__u32 usage_cnt; +__u32 pad; + }; + }; };union hl_cb_args {
Signature · 1 line
2.17.1
PATCH 1/1 habanalabs: Modify the cs_cnt of a CB to be atomic
From: Tomer Tayar <[email protected]>
Modify the CS counter of a CB to be atomic, so no locking is required when it is being modified or read.
--- .../habanalabs/common/command_submission.c | 29 +++++-------------- drivers/misc/habanalabs/common/debugfs.c | 2 +- drivers/misc/habanalabs/common/habanalabs.h | 4 +-- drivers/misc/habanalabs/gaudi/gaudi.c | 10 +++---- drivers/misc/habanalabs/goya/goya.c | 4 +-- 5 files changed, 17 insertions(+), 32 deletions(-)
drivers/misc/habanalabs/common/command_submission.c +7 -22
@@ -233,10 +233,7 @@ static int cs_parser(struct hl_fpriv *hpriv, struct hl_cs_job *job) job->patched_cb = parser.patched_cb; job->job_cb_size = parser.patched_cb_size; job->contains_dma_pkt = parser.contains_dma_pkt; - -spin_lock (&job->patched_cb->lock); - job->patched_cb->cs_cnt++; -spin_unlock (&job->patched_cb->lock); +atomic_inc (&job->patched_cb->cs_cnt); }/*
@@ -244,9 +241,7 @@ static int cs_parser(struct hl_fpriv *hpriv, struct hl_cs_job *job)* original CB anymore because it was already parsed and * won't be accessed again for this CS */ -spin_lock (&job->user_cb->lock); - job->user_cb->cs_cnt--; -spin_unlock (&job->user_cb->lock); +atomic_dec (&job->user_cb->cs_cnt);hl_cb_put (job->user_cb); job->user_cb =NULL ; }else if (!rc) {
@@ -268,10 +263,7 @@ static void complete_job(struct hl_device *hdev, struct hl_cs_job *job)* created, so we need to check it's not NULL */ if (job->patched_cb) { -spin_lock (&job->patched_cb->lock); - job->patched_cb->cs_cnt--; -spin_unlock (&job->patched_cb->lock); - +atomic_dec (&job->patched_cb->cs_cnt);hl_cb_put (job->patched_cb); } }
@@ -284,10 +276,7 @@ static void complete_job(struct hl_device *hdev, struct hl_cs_job *job)if (job->is_kernel_allocated_cb && ((job->queue_type ==QUEUE_TYPE_HW && hdev->mmu_enable) || job->queue_type ==QUEUE_TYPE_INT )) { -spin_lock (&job->user_cb->lock); - job->user_cb->cs_cnt--; -spin_unlock (&job->user_cb->lock); - +atomic_dec (&job->user_cb->cs_cnt);hl_cb_put (job->user_cb); }
@@ -680,9 +669,7 @@ static struct hl_cb *get_cb_from_cs_chunk(struct hl_device *hdev, goto release_cb; } -spin_lock (&cb->lock); - cb->cs_cnt++; -spin_unlock (&cb->lock); +atomic_inc (&cb->cs_cnt);return cb;
@@ -936,9 +923,7 @@ static int cs_ioctl_default(struct hl_fpriv *hpriv, void __user *chunks, goto put_cs; release_cb: -spin_lock (&cb->lock); - cb->cs_cnt--; -spin_unlock (&cb->lock); +atomic_dec (&cb->cs_cnt);hl_cb_put (cb); free_cs_object:cs_rollback (hdev, cs);
@@ -1133,7 +1118,7 @@ static int cs_ioctl_signal_wait_create_jobs(struct hl_device *hdev, job->id =0 ; job->cs = cs; job->user_cb = cb; - job->user_cb->cs_cnt++; +atomic_inc (&job->user_cb->cs_cnt); job->user_cb_size = cb_size; job->hw_queue_id = q_idx;
drivers/misc/habanalabs/common/debugfs.c +1 -1
@@ -116,7 +116,7 @@ static int command_buffers_show(struct seq_file *s, void *data)" %03llu %d 0x%08x %d %d %d\n" , cb->id, cb->ctx->asid, cb->size,kref_read (&cb->refcount), - cb->mmap, cb->cs_cnt); + cb->mmap, atomic_read(&cb->cs_cnt)); }spin_unlock (&dev_entry->cb_spinlock);
drivers/misc/habanalabs/common/habanalabs.h +2 -2
@@ -532,7 +532,7 @@ struct hl_cb_mgr {* @refcount: reference counter for usage of the CB. * @hdev: pointer to device this CB belongs to. * @ctx: pointer to the CB owner's context. -* @lock: spinlock to protect mmap/cs flows. +* @lock: spinlock to protect mmap flows. * @debugfs_list: node in debugfs list of command buffers. * @pool_list: node in pool list of command buffers. * @va_block_list: list of virtual addresses blocks of the CB if it is mapped to
@@ -561,7 +561,7 @@ struct hl_cb {dma_addr_t bus_address;u32 mmap_size;u32 size; -u32 cs_cnt; +atomic_t cs_cnt;u8 mmap;u8 is_pool;u8 is_internal;
drivers/misc/habanalabs/gaudi/gaudi.c +5 -5
@@ -838,7 +838,7 @@ static int _gaudi_init_tpc_mem(struct hl_device *hdev, job->id =0 ; job->user_cb = cb; - job->user_cb->cs_cnt++; +atomic_inc (&job->user_cb->cs_cnt); job->user_cb_size = cb_size; job->hw_queue_id =GAUDI_QUEUE_ID_DMA_0_0 ; job->patched_cb = job->user_cb;
@@ -861,7 +861,7 @@ static int _gaudi_init_tpc_mem(struct hl_device *hdev,hl_userptr_delete_list (hdev, &job->userptr_list);hl_debugfs_remove_job (hdev, job);kfree (job); - cb->cs_cnt--; +atomic_dec (&cb->cs_cnt); release_cb:hl_cb_put (cb);
@@ -1248,7 +1248,7 @@ static int gaudi_collective_wait_create_job(struct hl_device *hdev, job->id =0 ; job->cs = cs; job->user_cb = cb; - job->user_cb->cs_cnt++; +atomic_inc (&job->user_cb->cs_cnt); job->user_cb_size = cb_size; job->hw_queue_id = queue_id;
@@ -5570,7 +5570,7 @@ static int gaudi_memset_device_memory(struct hl_device *hdev, u64 addr, job->id =0 ; job->user_cb = cb; - job->user_cb->cs_cnt++; +atomic_inc (&job->user_cb->cs_cnt); job->user_cb_size = cb_size; job->hw_queue_id =GAUDI_QUEUE_ID_DMA_0_0 ; job->patched_cb = job->user_cb;
@@ -5581,7 +5581,7 @@ static int gaudi_memset_device_memory(struct hl_device *hdev, u64 addr, rc =gaudi_send_job_on_qman0 (hdev, job);hl_debugfs_remove_job (hdev, job);kfree (job); - cb->cs_cnt--; +atomic_dec (&cb->cs_cnt);/* Verify DMA is OK */ err_cause =RREG32 (mmDMA0_CORE_ERR_CAUSE);
drivers/misc/habanalabs/goya/goya.c +2 -2
@@ -4811,7 +4811,7 @@ static int goya_memset_device_memory(struct hl_device *hdev, u64 addr, u64 size, job->id =0 ; job->user_cb = cb; - job->user_cb->cs_cnt++; +atomic_inc (&job->user_cb->cs_cnt); job->user_cb_size = cb_size; job->hw_queue_id =GOYA_QUEUE_ID_DMA_0 ; job->patched_cb = job->user_cb;
@@ -4823,7 +4823,7 @@ static int goya_memset_device_memory(struct hl_device *hdev, u64 addr, u64 size,hl_debugfs_remove_job (hdev, job);kfree (job); - cb->cs_cnt--; +atomic_dec (&cb->cs_cnt); release_cb:hl_cb_put (cb);
Signature · 1 line
2.17.1
PATCH 1/1 habanalabs: Add mask for CS type bits in CS flags
In reply to Oded GabbayFrom: Tomer Tayar <[email protected]>
hl_cs_sanity_checks() extracts the CS type bits of the CS flags, by masking out the non-type bits.
To save the need for updating the function whenever new bits for non-type flags are added, add an explicit mask for the CS type bits.
--- drivers/misc/habanalabs/common/command_submission.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/misc/habanalabs/common/command_submission.c +4 -2
@@ -11,6 +11,9 @@#include <linux/uaccess.h> #include <linux/slab.h> +#define HL_CS_FLAGS_TYPE_MASK (HL_CS_FLAGS_SIGNAL | HL_CS_FLAGS_WAIT | \ + HL_CS_FLAGS_COLLECTIVE_WAIT) +/** * enum hl_cs_wait_status - cs wait status * @CS_WAIT_STATUS_BUSY: cs was not completed yet
@@ -737,8 +740,7 @@ static int hl_cs_sanity_checks(struct hl_fpriv *hpriv, union hl_cs_args *args)return -EBUSY ; } - cs_type_flags = args->in.cs_flags & - ~(HL_CS_FLAGS_FORCE_RESTORE |HL_CS_FLAGS_TIMESTAMP ); + cs_type_flags = args->in.cs_flags &HL_CS_FLAGS_TYPE_MASK ;if (unlikely (cs_type_flags && !is_power_of_2 (cs_type_flags))) { dev_err(hdev->dev,
Signature · 1 line
2.17.1
PATCH 1/1 habanalabs: change messages to debug level
In reply to Oded GabbaySome messages should be changed to debug mode as we want to keep minimal prints during normal operation of the device.
--- drivers/misc/habanalabs/common/firmware_if.c | 4 ++-- drivers/misc/habanalabs/gaudi/gaudi.c | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-)
drivers/misc/habanalabs/common/firmware_if.c +2 -2
@@ -526,7 +526,7 @@ static void fw_read_errors(struct hl_device *hdev, u32 boot_err0_reg, 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" , +dev_dbg (hdev->dev,"Device security status %#x\n" , security_val); }
@@ -806,7 +806,7 @@ int hl_fw_init_cpu(struct hl_device *hdev, u32 cpu_boot_status_reg, hdev->asic_prop.hard_reset_done_by_fw = true; } -dev_info (hdev->dev,"Firmware hard-reset is %s\n" , +dev_dbg (hdev->dev,"Firmware hard-reset is %s\n" , hdev->asic_prop.hard_reset_done_by_fw ?"enabled" :"disabled" );dev_info (hdev->dev,"Successfully loaded firmware to device\n" );
drivers/misc/habanalabs/gaudi/gaudi.c +1 -1
@@ -1732,7 +1732,7 @@ static int gaudi_enable_msi_single(struct hl_device *hdev) {int rc, irq; -dev_info (hdev->dev,"Working in single MSI IRQ mode\n" ); +dev_dbg (hdev->dev,"Working in single MSI IRQ mode\n" ); irq =gaudi_pci_irq_vector (hdev,0 , false); rc = request_irq(irq, gaudi_irq_handler_single, 0,
Signature · 1 line
2.17.1
PATCH 1/1 habanalabs: free host huge va_range if not used
In reply to Oded GabbayFrom: Ofir Bitton <[email protected]>
If huge range is not valid, driver uses the host range also for huge page allocations, but driver never frees its allocation. This introduces a memory leak every time a user closes its context.
--- drivers/misc/habanalabs/common/memory.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/misc/habanalabs/common/memory.c +3 -1
@@ -1761,6 +1761,7 @@ static int vm_ctx_init_with_ranges(struct hl_ctx *ctx, goto clear_host_va_range; } }else { +kfree (ctx->va_range[HL_VA_RANGE_TYPE_HOST_HUGE ]); ctx->va_range[HL_VA_RANGE_TYPE_HOST_HUGE ] = ctx->va_range[HL_VA_RANGE_TYPE_HOST ]; }
@@ -1906,9 +1907,10 @@ void hl_vm_ctx_fini(struct hl_ctx *ctx)spin_unlock (&vm->idr_lock);va_range_fini (hdev, ctx->va_range[HL_VA_RANGE_TYPE_DRAM ]); +va_range_fini (hdev, ctx->va_range[HL_VA_RANGE_TYPE_HOST ]); +if (hdev->pmmu_huge_range)va_range_fini (hdev, ctx->va_range[HL_VA_RANGE_TYPE_HOST_HUGE ]); -va_range_fini (hdev, ctx->va_range[HL_VA_RANGE_TYPE_HOST ]);mutex_destroy (&ctx->mem_hash_lock);hl_mmu_ctx_fini (ctx);
Signature · 1 line
2.17.1