habanalabs: missing mutex_unlock in process kill procedure
PATCH 1/7 habanalabs: missing mutex_unlock in process kill procedure
From: Ofir Bitton <[email protected]>
missing mutex unlock once driver is giving up killing user processes.
--- drivers/misc/habanalabs/common/device.c | 1 + 1 file changed, 1 insertion(+)
drivers/misc/habanalabs/common/device.c +1 -0
@@ -822,6 +822,7 @@ static int device_kill_open_processes(struct hl_device *hdev, u32 timeout) }else {dev_warn (hdev->dev,"Can't get task struct for PID so giving up on killing process\n" ); +mutex_unlock (&hdev->fpriv_list_lock);return -ETIME ; } }
Signature · 1 line
2.25.1
PATCH 2/7 habanalabs: rename cb_mmap to mmap
From: Zvika Yehudai <[email protected]>
This function will be used for more mmap operations than just mmaping CBs.
--- drivers/misc/habanalabs/common/command_buffer.c | 2 +- drivers/misc/habanalabs/common/habanalabs.h | 4 ++-- drivers/misc/habanalabs/gaudi/gaudi.c | 4 ++-- drivers/misc/habanalabs/goya/goya.c | 4 ++-- 4 files changed, 7 insertions(+), 7 deletions(-)
drivers/misc/habanalabs/common/command_buffer.c +1 -1
@@ -552,7 +552,7 @@ int hl_cb_mmap(struct hl_fpriv *hpriv, struct vm_area_struct *vma) vma->vm_private_data = cb; - rc = hdev->asic_funcs->cb_mmap (hdev, vma, cb->kernel_address, + rc = hdev->asic_funcs->mmap (hdev, vma, cb->kernel_address, cb->bus_address, cb->size);if (rc) {spin_lock (&cb->lock);
drivers/misc/habanalabs/common/habanalabs.h +2 -2
@@ -1024,7 +1024,7 @@ struct fw_load_mgr {* hw_fini and before CS rollback. * @suspend: handles IP specific H/W or SW changes for suspend. * @resume: handles IP specific H/W or SW changes for resume. -* @cb_mmap: maps a CB. +* @mmap: maps a memory. * @ring_doorbell: increment PI on a given QMAN. * @pqe_write: Write the PQ entry to the PQ. This is ASIC-specific * function because the PQs are located in different memory areas
@@ -1143,7 +1143,7 @@ struct hl_asic_funcs {void (*halt_engines)(struct hl_device *hdev,bool hard_reset);int (*suspend)(struct hl_device *hdev);int (*resume)(struct hl_device *hdev); -int (*cb_mmap)(struct hl_device *hdev,struct vm_area_struct *vma, +int (*mmap)(struct hl_device *hdev,struct vm_area_struct *vma,void *cpu_addr,dma_addr_t dma_addr,size_t size);void (*ring_doorbell)(struct hl_device *hdev,u32 hw_queue_id,u32 pi); void (*pqe_write)(struct hl_device *hdev, __le64 *pqe,
drivers/misc/habanalabs/gaudi/gaudi.c +2 -2
@@ -4299,7 +4299,7 @@ static int gaudi_resume(struct hl_device *hdev)return gaudi_init_iatu (hdev); } -static int gaudi_cb_mmap (struct hl_device *hdev,struct vm_area_struct *vma, +static int gaudi_mmap (struct hl_device *hdev,struct vm_area_struct *vma,void *cpu_addr,dma_addr_t dma_addr,size_t size) {int rc;
@@ -9313,7 +9313,7 @@ static const struct hl_asic_funcs gaudi_funcs = { .halt_engines = gaudi_halt_engines, .suspend = gaudi_suspend, .resume = gaudi_resume, - .cb_mmap = gaudi_cb_mmap, + .mmap = gaudi_mmap, .ring_doorbell = gaudi_ring_doorbell, .pqe_write = gaudi_pqe_write, .asic_dma_alloc_coherent = gaudi_dma_alloc_coherent,
drivers/misc/habanalabs/goya/goya.c +2 -2
@@ -2798,7 +2798,7 @@ int goya_resume(struct hl_device *hdev)return goya_init_iatu (hdev); } -static int goya_cb_mmap (struct hl_device *hdev,struct vm_area_struct *vma, +static int goya_mmap (struct hl_device *hdev,struct vm_area_struct *vma,void *cpu_addr,dma_addr_t dma_addr,size_t size) {int rc;
@@ -5584,7 +5584,7 @@ static const struct hl_asic_funcs goya_funcs = { .halt_engines = goya_halt_engines, .suspend = goya_suspend, .resume = goya_resume, - .cb_mmap = goya_cb_mmap, + .mmap = goya_mmap, .ring_doorbell = goya_ring_doorbell, .pqe_write = goya_pqe_write, .asic_dma_alloc_coherent = goya_dma_alloc_coherent,
Signature · 1 line
2.25.1
PATCH 3/7 habanalabs: fix nullifying of destroyed mmu pgt pool
In reply to Oded GabbayFrom: Tomer Tayar <[email protected]>
In case of host-resident MMU, when the page tables pool is destroyed, its pointer is not nullified correctly.
As a result, on a device fini which happens after a failing reset, the already destroyed pool is accessed, which leads to a kernel panic. The patch fixes the setting of the pool pointer to NULL.
--- drivers/misc/habanalabs/common/mmu/mmu_v1.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)
drivers/misc/habanalabs/common/mmu/mmu_v1.c +6 -6
@@ -470,13 +470,13 @@ static void hl_mmu_v1_fini(struct hl_device *hdev)if (!ZERO_OR_NULL_PTR (hdev->mmu_priv.hr.mmu_shadow_hop0)) {kvfree (hdev->mmu_priv.dr.mmu_shadow_hop0);gen_pool_destroy (hdev->mmu_priv.dr.mmu_pgt_pool); - } -/* Make sure that if we arrive here again without init was called we -* won't cause kernel panic. This can happen for example if we fail -* during hard reset code at certain points -*/ - hdev->mmu_priv.dr.mmu_shadow_hop0 =NULL ; +/* Make sure that if we arrive here again without init was +* called we won't cause kernel panic. This can happen for +* example if we fail during hard reset code at certain points +*/ + hdev->mmu_priv.dr.mmu_shadow_hop0 =NULL ; + } }/**
Signature · 1 line
2.25.1
PATCH 4/7 habanalabs: mark linux image as not loaded after hw_fini
In reply to Oded GabbayFrom: Tomer Tayar <[email protected]>
If hard reset fails after the call to hw_fini and before loading the linux image to the device, a subsequent call to hw_fini should communicate via COMMS (or MSG_TO_CPU regs for old FW versions). However, the driver still tries in this case to communicate via the GIC, and thus no hard reset is actually done.
To avoid that, the patch clears the linux_loaded flag after every call to hw_fini.
--- drivers/misc/habanalabs/common/device.c | 5 +++++ 1 file changed, 5 insertions(+)
drivers/misc/habanalabs/common/device.c +5 -0
@@ -1075,9 +1075,12 @@ int hl_device_reset(struct hl_device *hdev, u32 flags) hdev->asic_funcs->hw_fini (hdev, hard_reset);if (hard_reset) { + hdev->fw_loader.linux_loaded = false; +/* Release kernel context */ if (hdev->kernel_ctx &&hl_ctx_put (hdev->kernel_ctx) ==1 ) hdev->kernel_ctx =NULL ; +hl_vm_fini (hdev);hl_mmu_fini (hdev);hl_eq_reset (hdev, &hdev->event_queue);
@@ -1611,6 +1614,8 @@ void hl_device_fini(struct hl_device *hdev)/* Reset the H/W. It will be in idle state after this returns */ hdev->asic_funcs->hw_fini (hdev, true); + hdev->fw_loader.linux_loaded = false; +/* Release kernel context */ if ((hdev->kernel_ctx) && (hl_ctx_put (hdev->kernel_ctx) !=1 ))dev_err (hdev->dev,"kernel ctx is still alive\n" );
Signature · 1 line
2.25.1
PATCH 5/7 habanalabs: fix type of variable
In reply to Oded GabbayRecently, the size parameter in userptr structure was change to u64. As a result, we need to change the type of the local range_size in device_va_to_pa() to u64 to avoid overflow.
--- drivers/misc/habanalabs/common/debugfs.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/misc/habanalabs/common/debugfs.c +1 -2
@@ -491,11 +491,10 @@ static int device_va_to_pa(struct hl_device *hdev, u64 virt_addr, u32 size,struct hl_vm_phys_pg_pack *phys_pg_pack;struct hl_ctx *ctx = hdev->compute_ctx;struct hl_vm_hash_node *hnode; +u64 end_address, range_size;struct hl_userptr *userptr;enum vm_type *vm_type;bool valid = false; -u64 end_address; -u32 range_size;int i, rc =0 ;if (!ctx) {
Signature · 1 line
2.25.1
PATCH 6/7 habanalabs: add asic property of host dma offset
In reply to Oded GabbayEach ASIC can have a different offset to add to a host dma address, to enable the ASIC to access that host memory.
The usage for this can be common code so add this to the asic property structure.
--- drivers/misc/habanalabs/common/habanalabs.h | 3 +++ drivers/misc/habanalabs/gaudi/gaudi.c | 1 + drivers/misc/habanalabs/goya/goya.c | 1 + 3 files changed, 5 insertions(+)
drivers/misc/habanalabs/common/habanalabs.h +3 -0
@@ -434,6 +434,8 @@ struct hl_hints_range {* the device's MMU. * @dram_hints_align_mask: dram va hint addresses alignment mask which is used * for hints validity check. +* device_dma_offset_for_host_access: the offset to add to host DMA addresses +* to enable the device to access them. * @mmu_pgt_size: MMU page tables total size. * @mmu_pte_size: PTE size in MMU page tables. * @mmu_hop_table_size: MMU hop table size.
@@ -527,6 +529,7 @@ struct asic_fixed_properties {u64 cb_va_start_addr;u64 cb_va_end_addr;u64 dram_hints_align_mask; +u64 device_dma_offset_for_host_access;u32 mmu_pgt_size;u32 mmu_pte_size;u32 mmu_hop_table_size;
drivers/misc/habanalabs/gaudi/gaudi.c +1 -0
@@ -537,6 +537,7 @@ static int gaudi_set_fixed_properties(struct hl_device *hdev)get_collective_mode (hdev, i); } + prop->device_dma_offset_for_host_access =HOST_PHYS_BASE ; prop->completion_queues_count =NUMBER_OF_CMPLT_QUEUES ; prop->collective_first_sob =0 ; prop->collective_first_mon =0 ;
drivers/misc/habanalabs/goya/goya.c +1 -0
@@ -389,6 +389,7 @@ int goya_set_fixed_properties(struct hl_device *hdev) prop->hw_queues_props[i].cb_alloc_flags =CB_ALLOC_USER ; } + prop->device_dma_offset_for_host_access =HOST_PHYS_BASE ; prop->completion_queues_count =NUMBER_OF_CMPLT_QUEUES ; prop->dram_base_address =DRAM_PHYS_BASE ;
Signature · 1 line
2.25.1
PATCH 7/7 habanalabs: set dma max segment size
In reply to Oded GabbayThis is required from any device that is capable to perform DMA.
--- drivers/misc/habanalabs/common/pci/pci.c | 2 ++ 1 file changed, 2 insertions(+)
drivers/misc/habanalabs/common/pci/pci.c +2 -0
@@ -436,6 +436,8 @@ int hl_pci_init(struct hl_device *hdev) goto unmap_pci_bars; } +dma_set_max_seg_size (&pdev->dev,U32_MAX ); +return 0 ; unmap_pci_bars:
Signature · 1 line
2.25.1