Merge tag 'drm-amdkfd-fixes-2017-11-26' of...

Merge tag 'drm-amdkfd-fixes-2017-11-26' of git://people.freedesktop.org/~gabbayo/linux into drm-fixes

This is amdkfd pull request for -rc2. It contains three small fixes to the
CIK SDMA code, compilation error fix in kfd_ioctl.h and fix to accessing
a pointer after it was released.

* tag 'drm-amdkfd-fixes-2017-11-26' of git://people.freedesktop.org/~gabbayo/linux:
  uapi: fix linux/kfd_ioctl.h userspace compilation errors
  drm/amdkfd: fix amdkfd use-after-free GP fault
  drm/amdkfd: Fix SDMA oversubsription handling
  drm/amdkfd: Fix SDMA ring buffer size calculation
  drm/amdgpu: Fix SDMA load/unload sequence on HWS disabled mode

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c

@@ -379,29 +379,50 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd)
 {
 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	struct cik_sdma_rlc_registers *m;
+	unsigned long end_jiffies;
 	uint32_t sdma_base_addr;
+	uint32_t data;
 
 	m = get_sdma_mqd(mqd);
 	sdma_base_addr = get_sdma_base_addr(m);
 
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_VIRTUAL_ADDR,
-			m->sdma_rlc_virtual_addr);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL,
+		m->sdma_rlc_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK));
 
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE,
-			m->sdma_rlc_rb_base);
+	end_jiffies = msecs_to_jiffies(2000) + jiffies;
+	while (true) {
+		data = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (data & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies))
+			return -ETIME;
+		usleep_range(500, 1000);
+	}
+	if (m->sdma_engine_id) {
+		data = RREG32(mmSDMA1_GFX_CONTEXT_CNTL);
+		data = REG_SET_FIELD(data, SDMA1_GFX_CONTEXT_CNTL,
+				RESUME_CTX, 0);
+		WREG32(mmSDMA1_GFX_CONTEXT_CNTL, data);
+	} else {
+		data = RREG32(mmSDMA0_GFX_CONTEXT_CNTL);
+		data = REG_SET_FIELD(data, SDMA0_GFX_CONTEXT_CNTL,
+				RESUME_CTX, 0);
+		WREG32(mmSDMA0_GFX_CONTEXT_CNTL, data);
+	}
 
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL,
+				m->sdma_rlc_doorbell);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_VIRTUAL_ADDR,
+				m->sdma_rlc_virtual_addr);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, m->sdma_rlc_rb_base);
 	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE_HI,
 			m->sdma_rlc_rb_base_hi);
-
 	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_LO,
 			m->sdma_rlc_rb_rptr_addr_lo);
-
 	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_HI,
 			m->sdma_rlc_rb_rptr_addr_hi);
-
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL,
-			m->sdma_rlc_doorbell);
-
 	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL,
 			m->sdma_rlc_rb_cntl);
 
@@ -574,9 +595,9 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
 	}
 
 	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, 0);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, 0);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL,
+		RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL) |
+		SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK);
 
 	return 0;
 }

drivers/gpu/drm/amd/amdkfd/kfd_module.c

@@ -24,6 +24,7 @@
 #include <linux/sched.h>
 #include <linux/moduleparam.h>
 #include <linux/device.h>
+#include <linux/printk.h>
 #include "kfd_priv.h"
 
 #define KFD_DRIVER_AUTHOR	"AMD Inc. and others"
@@ -132,7 +133,7 @@ static void __exit kfd_module_exit(void)
 	kfd_process_destroy_wq();
 	kfd_topology_shutdown();
 	kfd_chardev_exit();
-	dev_info(kfd_device, "Removed module\n");
+	pr_info("amdkfd: Removed module\n");
 }
 
 module_init(kfd_module_init);

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c

@@ -202,8 +202,8 @@ static int update_mqd_sdma(struct mqd_manager *mm, void *mqd,
 	struct cik_sdma_rlc_registers *m;
 
 	m = get_sdma_mqd(mqd);
-	m->sdma_rlc_rb_cntl = ffs(q->queue_size / sizeof(unsigned int)) <<
-			SDMA0_RLC0_RB_CNTL__RB_SIZE__SHIFT |
+	m->sdma_rlc_rb_cntl = (ffs(q->queue_size / sizeof(unsigned int)) - 1)
+			<< SDMA0_RLC0_RB_CNTL__RB_SIZE__SHIFT |
 			q->vmid << SDMA0_RLC0_RB_CNTL__RB_VMID__SHIFT |
 			1 << SDMA0_RLC0_RB_CNTL__RPTR_WRITEBACK_ENABLE__SHIFT |
 			6 << SDMA0_RLC0_RB_CNTL__RPTR_WRITEBACK_TIMER__SHIFT;

drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c

@@ -191,6 +191,24 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 
 	switch (type) {
 	case KFD_QUEUE_TYPE_SDMA:
+		if (dev->dqm->queue_count >=
+			CIK_SDMA_QUEUES_PER_ENGINE * CIK_SDMA_ENGINE_NUM) {
+			pr_err("Over-subscription is not allowed for SDMA.\n");
+			retval = -EPERM;
+			goto err_create_queue;
+		}
+
+		retval = create_cp_queue(pqm, dev, &q, properties, f, *qid);
+		if (retval != 0)
+			goto err_create_queue;
+		pqn->q = q;
+		pqn->kq = NULL;
+		retval = dev->dqm->ops.create_queue(dev->dqm, q, &pdd->qpd,
+						&q->properties.vmid);
+		pr_debug("DQM returned %d for create_queue\n", retval);
+		print_queue(q);
+		break;
+
 	case KFD_QUEUE_TYPE_COMPUTE:
 		/* check if there is over subscription */
 		if ((sched_policy == KFD_SCHED_POLICY_HWS_NO_OVERSUBSCRIPTION) &&

include/uapi/linux/kfd_ioctl.h

@@ -233,29 +233,29 @@ struct kfd_ioctl_wait_events_args {
 };
 
 struct kfd_ioctl_set_scratch_backing_va_args {
-	uint64_t va_addr;	/* to KFD */
-	uint32_t gpu_id;	/* to KFD */
-	uint32_t pad;
+	__u64 va_addr;	/* to KFD */
+	__u32 gpu_id;	/* to KFD */
+	__u32 pad;
 };
 
 struct kfd_ioctl_get_tile_config_args {
 	/* to KFD: pointer to tile array */
-	uint64_t tile_config_ptr;
+	__u64 tile_config_ptr;
 	/* to KFD: pointer to macro tile array */
-	uint64_t macro_tile_config_ptr;
+	__u64 macro_tile_config_ptr;
 	/* to KFD: array size allocated by user mode
 	 * from KFD: array size filled by kernel
 	 */
-	uint32_t num_tile_configs;
+	__u32 num_tile_configs;
 	/* to KFD: array size allocated by user mode
 	 * from KFD: array size filled by kernel
 	 */
-	uint32_t num_macro_tile_configs;
+	__u32 num_macro_tile_configs;
 
-	uint32_t gpu_id;		/* to KFD */
-	uint32_t gb_addr_config;	/* from KFD */
-	uint32_t num_banks;		/* from KFD */
-	uint32_t num_ranks;		/* from KFD */
+	__u32 gpu_id;		/* to KFD */
+	__u32 gb_addr_config;	/* from KFD */
+	__u32 num_banks;		/* from KFD */
+	__u32 num_ranks;		/* from KFD */
 	/* struct size can be extended later if needed
 	 * without breaking ABI compatibility
 	 */