1 files changed, 165 insertions, 121 deletions

diff --git a/drivers/gpu/drm/xe/xe_gt_pagefault.c b/drivers/gpu/drm/xe/xe_gt_pagefault.c
index 73f08f1924df..2606cd396df5 100644
--- a/drivers/gpu/drm/xe/xe_gt_pagefault.c
+++ b/drivers/gpu/drm/xe/xe_gt_pagefault.c
@@ -10,7 +10,6 @@
 
 #include <drm/drm_exec.h>
 #include <drm/drm_managed.h>
-#include <drm/ttm/ttm_execbuf_util.h>
 
 #include "abi/guc_actions_abi.h"
 #include "xe_bo.h"
@@ -19,8 +18,7 @@
 #include "xe_guc.h"
 #include "xe_guc_ct.h"
 #include "xe_migrate.h"
-#include "xe_pt.h"
-#include "xe_trace.h"
+#include "xe_trace_bo.h"
 #include "xe_vm.h"
 
 struct pagefault {
@@ -69,7 +67,7 @@ static bool access_is_atomic(enum access_type access_type)
 static bool vma_is_valid(struct xe_tile *tile, struct xe_vma *vma)
 {
 	return BIT(tile->id) & vma->tile_present &&
-		!(BIT(tile->id) & vma->usm.tile_invalidated);
+		!(BIT(tile->id) & vma->tile_invalidated);
 }
 
 static bool vma_matches(struct xe_vma *vma, u64 page_addr)
@@ -100,10 +98,9 @@ static int xe_pf_begin(struct drm_exec *exec, struct xe_vma *vma,
 {
 	struct xe_bo *bo = xe_vma_bo(vma);
 	struct xe_vm *vm = xe_vma_vm(vma);
-	unsigned int num_shared = 2; /* slots for bind + move */
 	int err;
 
-	err = xe_vm_prepare_vma(exec, vma, num_shared);
+	err = xe_vm_lock_vma(exec, vma);
 	if (err)
 		return err;
 
@@ -127,125 +124,122 @@ static int xe_pf_begin(struct drm_exec *exec, struct xe_vma *vma,
 	return 0;
 }
 
-static int handle_pagefault(struct xe_gt *gt, struct pagefault *pf)
+static int handle_vma_pagefault(struct xe_tile *tile, struct pagefault *pf,
+				struct xe_vma *vma)
 {
-	struct xe_device *xe = gt_to_xe(gt);
-	struct xe_tile *tile = gt_to_tile(gt);
+	struct xe_vm *vm = xe_vma_vm(vma);
 	struct drm_exec exec;
-	struct xe_vm *vm;
-	struct xe_vma *vma = NULL;
 	struct dma_fence *fence;
-	bool write_locked;
-	int ret = 0;
+	ktime_t end = 0;
+	int err;
 	bool atomic;
 
-	/* SW isn't expected to handle TRTT faults */
-	if (pf->trva_fault)
-		return -EFAULT;
-
-	/* ASID to VM */
-	mutex_lock(&xe->usm.lock);
-	vm = xa_load(&xe->usm.asid_to_vm, pf->asid);
-	if (vm)
-		xe_vm_get(vm);
-	mutex_unlock(&xe->usm.lock);
-	if (!vm || !xe_vm_in_fault_mode(vm))
-		return -EINVAL;
-
-retry_userptr:
-	/*
-	 * TODO: Avoid exclusive lock if VM doesn't have userptrs, or
-	 * start out read-locked?
-	 */
-	down_write(&vm->lock);
-	write_locked = true;
-	vma = lookup_vma(vm, pf->page_addr);
-	if (!vma) {
-		ret = -EINVAL;
-		goto unlock_vm;
-	}
-
-	if (!xe_vma_is_userptr(vma) ||
-	    !xe_vma_userptr_check_repin(to_userptr_vma(vma))) {
-		downgrade_write(&vm->lock);
-		write_locked = false;
-	}
-
 	trace_xe_vma_pagefault(vma);
-
 	atomic = access_is_atomic(pf->access_type);
 
 	/* Check if VMA is valid */
 	if (vma_is_valid(tile, vma) && !atomic)
-		goto unlock_vm;
-
-	/* TODO: Validate fault */
+		return 0;
 
-	if (xe_vma_is_userptr(vma) && write_locked) {
+retry_userptr:
+	if (xe_vma_is_userptr(vma) &&
+	    xe_vma_userptr_check_repin(to_userptr_vma(vma))) {
 		struct xe_userptr_vma *uvma = to_userptr_vma(vma);
 
-		spin_lock(&vm->userptr.invalidated_lock);
-		list_del_init(&uvma->userptr.invalidate_link);
-		spin_unlock(&vm->userptr.invalidated_lock);
-
-		ret = xe_vma_userptr_pin_pages(uvma);
-		if (ret)
-			goto unlock_vm;
-
-		downgrade_write(&vm->lock);
-		write_locked = false;
+		err = xe_vma_userptr_pin_pages(uvma);
+		if (err)
+			return err;
 	}
 
 	/* Lock VM and BOs dma-resv */
 	drm_exec_init(&exec, 0, 0);
 	drm_exec_until_all_locked(&exec) {
-		ret = xe_pf_begin(&exec, vma, atomic, tile->id);
+		err = xe_pf_begin(&exec, vma, atomic, tile->id);
 		drm_exec_retry_on_contention(&exec);
-		if (ret)
+		if (xe_vm_validate_should_retry(&exec, err, &end))
+			err = -EAGAIN;
+		if (err)
 			goto unlock_dma_resv;
-	}
 
-	/* Bind VMA only to the GT that has faulted */
-	trace_xe_vma_pf_bind(vma);
-	fence = __xe_pt_bind_vma(tile, vma, xe_tile_migrate_engine(tile), NULL, 0,
-				 vma->tile_present & BIT(tile->id));
-	if (IS_ERR(fence)) {
-		ret = PTR_ERR(fence);
-		goto unlock_dma_resv;
+		/* Bind VMA only to the GT that has faulted */
+		trace_xe_vma_pf_bind(vma);
+		fence = xe_vma_rebind(vm, vma, BIT(tile->id));
+		if (IS_ERR(fence)) {
+			err = PTR_ERR(fence);
+			if (xe_vm_validate_should_retry(&exec, err, &end))
+				err = -EAGAIN;
+			goto unlock_dma_resv;
+		}
 	}
 
-	/*
-	 * XXX: Should we drop the lock before waiting? This only helps if doing
-	 * GPU binds which is currently only done if we have to wait for more
-	 * than 10ms on a move.
-	 */
 	dma_fence_wait(fence, false);
 	dma_fence_put(fence);
-
-	if (xe_vma_is_userptr(vma))
-		ret = xe_vma_userptr_check_repin(to_userptr_vma(vma));
-	vma->usm.tile_invalidated &= ~BIT(tile->id);
+	vma->tile_invalidated &= ~BIT(tile->id);
 
 unlock_dma_resv:
 	drm_exec_fini(&exec);
-unlock_vm:
-	if (!ret)
-		vm->usm.last_fault_vma = vma;
-	if (write_locked)
-		up_write(&vm->lock);
-	else
-		up_read(&vm->lock);
-	if (ret == -EAGAIN)
+	if (err == -EAGAIN)
 		goto retry_userptr;
 
-	if (!ret) {
-		ret = xe_gt_tlb_invalidation_vma(gt, NULL, vma);
-		if (ret >= 0)
-			ret = 0;
+	return err;
+}
+
+static struct xe_vm *asid_to_vm(struct xe_device *xe, u32 asid)
+{
+	struct xe_vm *vm;
+
+	down_read(&xe->usm.lock);
+	vm = xa_load(&xe->usm.asid_to_vm, asid);
+	if (vm && xe_vm_in_fault_mode(vm))
+		xe_vm_get(vm);
+	else
+		vm = ERR_PTR(-EINVAL);
+	up_read(&xe->usm.lock);
+
+	return vm;
+}
+
+static int handle_pagefault(struct xe_gt *gt, struct pagefault *pf)
+{
+	struct xe_device *xe = gt_to_xe(gt);
+	struct xe_tile *tile = gt_to_tile(gt);
+	struct xe_vm *vm;
+	struct xe_vma *vma = NULL;
+	int err;
+
+	/* SW isn't expected to handle TRTT faults */
+	if (pf->trva_fault)
+		return -EFAULT;
+
+	vm = asid_to_vm(xe, pf->asid);
+	if (IS_ERR(vm))
+		return PTR_ERR(vm);
+
+	/*
+	 * TODO: Change to read lock? Using write lock for simplicity.
+	 */
+	down_write(&vm->lock);
+
+	if (xe_vm_is_closed(vm)) {
+		err = -ENOENT;
+		goto unlock_vm;
+	}
+
+	vma = lookup_vma(vm, pf->page_addr);
+	if (!vma) {
+		err = -EINVAL;
+		goto unlock_vm;
 	}
+
+	err = handle_vma_pagefault(tile, pf, vma);
+
+unlock_vm:
+	if (!err)
+		vm->usm.last_fault_vma = vma;
+	up_write(&vm->lock);
 	xe_vm_put(vm);
 
-	return ret;
+	return err;
 }
 
 static int send_pagefault_reply(struct xe_guc *guc,
@@ -285,9 +279,9 @@ static bool get_pagefault(struct pf_queue *pf_queue, struct pagefault *pf)
 	bool ret = false;
 
 	spin_lock_irq(&pf_queue->lock);
-	if (pf_queue->head != pf_queue->tail) {
+	if (pf_queue->tail != pf_queue->head) {
 		desc = (const struct xe_guc_pagefault_desc *)
-			(pf_queue->data + pf_queue->head);
+			(pf_queue->data + pf_queue->tail);
 
 		pf->fault_level = FIELD_GET(PFD_FAULT_LEVEL, desc->dw0);
 		pf->trva_fault = FIELD_GET(XE2_PFD_TRVA_FAULT, desc->dw0);
@@ -305,8 +299,8 @@ static bool get_pagefault(struct pf_queue *pf_queue, struct pagefault *pf)
 		pf->page_addr |= FIELD_GET(PFD_VIRTUAL_ADDR_LO, desc->dw2) <<
 			PFD_VIRTUAL_ADDR_LO_SHIFT;
 
-		pf_queue->head = (pf_queue->head + PF_MSG_LEN_DW) %
-			PF_QUEUE_NUM_DW;
+		pf_queue->tail = (pf_queue->tail + PF_MSG_LEN_DW) %
+			pf_queue->num_dw;
 		ret = true;
 	}
 	spin_unlock_irq(&pf_queue->lock);
@@ -318,7 +312,8 @@ static bool pf_queue_full(struct pf_queue *pf_queue)
 {
 	lockdep_assert_held(&pf_queue->lock);
 
-	return CIRC_SPACE(pf_queue->tail, pf_queue->head, PF_QUEUE_NUM_DW) <=
+	return CIRC_SPACE(pf_queue->head, pf_queue->tail,
+			  pf_queue->num_dw) <=
 		PF_MSG_LEN_DW;
 }
 
@@ -337,11 +332,17 @@ int xe_guc_pagefault_handler(struct xe_guc *guc, u32 *msg, u32 len)
 	asid = FIELD_GET(PFD_ASID, msg[1]);
 	pf_queue = gt->usm.pf_queue + (asid % NUM_PF_QUEUE);
 
+	/*
+	 * The below logic doesn't work unless PF_QUEUE_NUM_DW % PF_MSG_LEN_DW == 0
+	 */
+	xe_gt_assert(gt, !(pf_queue->num_dw % PF_MSG_LEN_DW));
+
 	spin_lock_irqsave(&pf_queue->lock, flags);
 	full = pf_queue_full(pf_queue);
 	if (!full) {
-		memcpy(pf_queue->data + pf_queue->tail, msg, len * sizeof(u32));
-		pf_queue->tail = (pf_queue->tail + len) % PF_QUEUE_NUM_DW;
+		memcpy(pf_queue->data + pf_queue->head, msg, len * sizeof(u32));
+		pf_queue->head = (pf_queue->head + len) %
+			pf_queue->num_dw;
 		queue_work(gt->usm.pf_wq, &pf_queue->worker);
 	} else {
 		drm_warn(&xe->drm, "PF Queue full, shouldn't be possible");
@@ -387,7 +388,7 @@ static void pf_queue_work_func(struct work_struct *w)
 		send_pagefault_reply(&gt->uc.guc, &reply);
 
 		if (time_after(jiffies, threshold) &&
-		    pf_queue->head != pf_queue->tail) {
+		    pf_queue->tail != pf_queue->head) {
 			queue_work(gt->usm.pf_wq, w);
 			break;
 		}
@@ -396,18 +397,59 @@ static void pf_queue_work_func(struct work_struct *w)
 
 static void acc_queue_work_func(struct work_struct *w);
 
+static void pagefault_fini(void *arg)
+{
+	struct xe_gt *gt = arg;
+	struct xe_device *xe = gt_to_xe(gt);
+
+	if (!xe->info.has_usm)
+		return;
+
+	destroy_workqueue(gt->usm.acc_wq);
+	destroy_workqueue(gt->usm.pf_wq);
+}
+
+static int xe_alloc_pf_queue(struct xe_gt *gt, struct pf_queue *pf_queue)
+{
+	struct xe_device *xe = gt_to_xe(gt);
+	xe_dss_mask_t all_dss;
+	int num_dss, num_eus;
+
+	bitmap_or(all_dss, gt->fuse_topo.g_dss_mask, gt->fuse_topo.c_dss_mask,
+		  XE_MAX_DSS_FUSE_BITS);
+
+	num_dss = bitmap_weight(all_dss, XE_MAX_DSS_FUSE_BITS);
+	num_eus = bitmap_weight(gt->fuse_topo.eu_mask_per_dss,
+				XE_MAX_EU_FUSE_BITS) * num_dss;
+
+	/* user can issue separate page faults per EU and per CS */
+	pf_queue->num_dw =
+		(num_eus + XE_NUM_HW_ENGINES) * PF_MSG_LEN_DW;
+
+	pf_queue->gt = gt;
+	pf_queue->data = devm_kcalloc(xe->drm.dev, pf_queue->num_dw,
+				      sizeof(u32), GFP_KERNEL);
+	if (!pf_queue->data)
+		return -ENOMEM;
+
+	spin_lock_init(&pf_queue->lock);
+	INIT_WORK(&pf_queue->worker, pf_queue_work_func);
+
+	return 0;
+}
+
 int xe_gt_pagefault_init(struct xe_gt *gt)
 {
 	struct xe_device *xe = gt_to_xe(gt);
-	int i;
+	int i, ret = 0;
 
 	if (!xe->info.has_usm)
 		return 0;
 
 	for (i = 0; i < NUM_PF_QUEUE; ++i) {
-		gt->usm.pf_queue[i].gt = gt;
-		spin_lock_init(&gt->usm.pf_queue[i].lock);
-		INIT_WORK(&gt->usm.pf_queue[i].worker, pf_queue_work_func);
+		ret = xe_alloc_pf_queue(gt, &gt->usm.pf_queue[i]);
+		if (ret)
+			return ret;
 	}
 	for (i = 0; i < NUM_ACC_QUEUE; ++i) {
 		gt->usm.acc_queue[i].gt = gt;
@@ -423,10 +465,12 @@ int xe_gt_pagefault_init(struct xe_gt *gt)
 	gt->usm.acc_wq = alloc_workqueue("xe_gt_access_counter_work_queue",
 					 WQ_UNBOUND | WQ_HIGHPRI,
 					 NUM_ACC_QUEUE);
-	if (!gt->usm.acc_wq)
+	if (!gt->usm.acc_wq) {
+		destroy_workqueue(gt->usm.pf_wq);
 		return -ENOMEM;
+	}
 
-	return 0;
+	return devm_add_action_or_reset(xe->drm.dev, pagefault_fini, gt);
 }
 
 void xe_gt_pagefault_reset(struct xe_gt *gt)
@@ -511,14 +555,9 @@ static int handle_acc(struct xe_gt *gt, struct acc *acc)
 	if (acc->access_type != ACC_TRIGGER)
 		return -EINVAL;
 
-	/* ASID to VM */
-	mutex_lock(&xe->usm.lock);
-	vm = xa_load(&xe->usm.asid_to_vm, acc->asid);
-	if (vm)
-		xe_vm_get(vm);
-	mutex_unlock(&xe->usm.lock);
-	if (!vm || !xe_vm_in_fault_mode(vm))
-		return -EINVAL;
+	vm = asid_to_vm(xe, acc->asid);
+	if (IS_ERR(vm))
+		return PTR_ERR(vm);
 
 	down_read(&vm->lock);
 
@@ -562,9 +601,9 @@ static bool get_acc(struct acc_queue *acc_queue, struct acc *acc)
 	bool ret = false;
 
 	spin_lock(&acc_queue->lock);
-	if (acc_queue->head != acc_queue->tail) {
+	if (acc_queue->tail != acc_queue->head) {
 		desc = (const struct xe_guc_acc_desc *)
-			(acc_queue->data + acc_queue->head);
+			(acc_queue->data + acc_queue->tail);
 
 		acc->granularity = FIELD_GET(ACC_GRANULARITY, desc->dw2);
 		acc->sub_granularity = FIELD_GET(ACC_SUBG_HI, desc->dw1) << 31 |
@@ -577,7 +616,7 @@ static bool get_acc(struct acc_queue *acc_queue, struct acc *acc)
 		acc->va_range_base = make_u64(desc->dw3 & ACC_VIRTUAL_ADDR_RANGE_HI,
 					      desc->dw2 & ACC_VIRTUAL_ADDR_RANGE_LO);
 
-		acc_queue->head = (acc_queue->head + ACC_MSG_LEN_DW) %
+		acc_queue->tail = (acc_queue->tail + ACC_MSG_LEN_DW) %
 				  ACC_QUEUE_NUM_DW;
 		ret = true;
 	}
@@ -605,7 +644,7 @@ static void acc_queue_work_func(struct work_struct *w)
 		}
 
 		if (time_after(jiffies, threshold) &&
-		    acc_queue->head != acc_queue->tail) {
+		    acc_queue->tail != acc_queue->head) {
 			queue_work(gt->usm.acc_wq, w);
 			break;
 		}
@@ -616,7 +655,7 @@ static bool acc_queue_full(struct acc_queue *acc_queue)
 {
 	lockdep_assert_held(&acc_queue->lock);
 
-	return CIRC_SPACE(acc_queue->tail, acc_queue->head, ACC_QUEUE_NUM_DW) <=
+	return CIRC_SPACE(acc_queue->head, acc_queue->tail, ACC_QUEUE_NUM_DW) <=
 		ACC_MSG_LEN_DW;
 }
 
@@ -627,6 +666,11 @@ int xe_guc_access_counter_notify_handler(struct xe_guc *guc, u32 *msg, u32 len)
 	u32 asid;
 	bool full;
 
+	/*
+	 * The below logic doesn't work unless ACC_QUEUE_NUM_DW % ACC_MSG_LEN_DW == 0
+	 */
+	BUILD_BUG_ON(ACC_QUEUE_NUM_DW % ACC_MSG_LEN_DW);
+
 	if (unlikely(len != ACC_MSG_LEN_DW))
 		return -EPROTO;
 
@@ -636,9 +680,9 @@ int xe_guc_access_counter_notify_handler(struct xe_guc *guc, u32 *msg, u32 len)
 	spin_lock(&acc_queue->lock);
 	full = acc_queue_full(acc_queue);
 	if (!full) {
-		memcpy(acc_queue->data + acc_queue->tail, msg,
+		memcpy(acc_queue->data + acc_queue->head, msg,
 		       len * sizeof(u32));
-		acc_queue->tail = (acc_queue->tail + len) % ACC_QUEUE_NUM_DW;
+		acc_queue->head = (acc_queue->head + len) % ACC_QUEUE_NUM_DW;
 		queue_work(gt->usm.acc_wq, &acc_queue->worker);
 	} else {
 		drm_warn(&gt_to_xe(gt)->drm, "ACC Queue full, dropping ACC");