From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id EDCA1EE49A0 for ; Wed, 23 Aug 2023 12:51:30 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 9A40410E41F; Wed, 23 Aug 2023 12:51:30 +0000 (UTC) Received: from mgamail.intel.com (mgamail.intel.com [134.134.136.126]) by gabe.freedesktop.org (Postfix) with ESMTPS id 32DFE10E41F for ; Wed, 23 Aug 2023 12:51:28 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1692795088; x=1724331088; h=message-id:date:mime-version:subject:to:references:from: in-reply-to:content-transfer-encoding; bh=ZyldREa/ozCGh8I4WS61ZwtK37lHOECvVhlWQXy0hi4=; b=AdR89wB2jn2L0mieXeZ0mZcsYnDRoPlOSBlLRAwUM+kqgKmaTvade87z jRkgVsA9MFeeucH8SqjVE+RI6ibul6zZpFgUju0LMAFKOYOfMhDZDNuYs C758OsLXMij2LUUDDvquAvo9KQu763G2RawIohuXNFYAn4te4vh2fO2jv q+E87idn91LvwuX7O53sj78n7NAY6JWNBC67eO/eJLXH46VG0NIjgOA+i ijlF78OWVJvhIO03lEUr6cF11VEoRWLVerL2RK+yJjkPHiwFGyl/7A3tq viUBx0VdydrDkiooauM9FS0rxUTNLDr+FPFvA9fDGDJV26lccvhQHXH9k Q==; X-IronPort-AV: E=McAfee;i="6600,9927,10810"; a="359136380" X-IronPort-AV: E=Sophos;i="6.01,195,1684825200"; d="scan'208";a="359136380" Received: from orsmga002.jf.intel.com ([10.7.209.21]) by orsmga106.jf.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 23 Aug 2023 05:51:27 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10810"; a="736654058" X-IronPort-AV: E=Sophos;i="6.01,195,1684825200"; d="scan'208";a="736654058" Received: from noors55x-mobl1.gar.corp.intel.com (HELO [10.249.254.82]) ([10.249.254.82]) by orsmga002-auth.jf.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 23 Aug 2023 05:51:26 -0700 Message-ID: Date: Wed, 23 Aug 2023 14:51:23 +0200 MIME-Version: 1.0 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101 Thunderbird/102.13.0 Content-Language: en-US To: Matthew Brost , intel-xe@lists.freedesktop.org References: <20230817043148.740495-1-matthew.brost@intel.com> <20230817043148.740495-4-matthew.brost@intel.com> From: =?UTF-8?Q?Thomas_Hellstr=c3=b6m?= In-Reply-To: <20230817043148.740495-4-matthew.brost@intel.com> Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit Subject: Re: [Intel-xe] [PATCH 3/3] drm/xe: Fix array of binds X-BeenThere: intel-xe@lists.freedesktop.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Intel Xe graphics driver List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: intel-xe-bounces@lists.freedesktop.org Sender: "Intel-xe" Hi, Matt, In addition to Rodrigo's comment, one minor thing below. Will need to apply this series to get a better understanding what the changes actually do, and to what extent we can unwind. /Thomas On 8/17/23 06:31, Matthew Brost wrote: > If multiple bind ops in an array of binds touch the same address range > invalid GPUVA operations are generated as each GPUVA operation is > generated based on the orignal GPUVA state. To fix this, after each > GPUVA operations is generated, commit the GPUVA operation updating the > GPUVA state so subsequent bind ops can see a current GPUVA state. > > Signed-off-by: Matthew Brost > --- > drivers/gpu/drm/xe/xe_vm.c | 418 +++++++++++++++++++------------------ > 1 file changed, 212 insertions(+), 206 deletions(-) > > diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c > index bd20840616ca..2452e24fbc81 100644 > --- a/drivers/gpu/drm/xe/xe_vm.c > +++ b/drivers/gpu/drm/xe/xe_vm.c > @@ -2426,24 +2426,73 @@ static u64 xe_vma_set_pte_size(struct xe_vma *vma, u64 size) > return SZ_4K; > } > > -/* > - * Parse operations list and create any resources needed for the operations > - * prior to fully committing to the operations. This setup can fail. > - */ > +static int xe_vma_op_commit(struct xe_vm *vm, struct xe_vma_op *op) > +{ > + int err = 0; > + > + lockdep_assert_held_write(&vm->lock); > + > + switch (op->base.op) { > + case DRM_GPUVA_OP_MAP: > + err |= xe_vm_insert_vma(vm, op->map.vma); > + if (!err) > + op->flags |= XE_VMA_OP_COMMITTED; > + break; > + case DRM_GPUVA_OP_REMAP: > + prep_vma_destroy(vm, gpuva_to_vma(op->base.remap.unmap->va), > + true); > + op->flags |= XE_VMA_OP_COMMITTED; > + > + if (op->remap.prev) { > + err |= xe_vm_insert_vma(vm, op->remap.prev); > + if (!err) > + op->flags |= XE_VMA_OP_PREV_COMMITTED; > + if (!err && op->remap.skip_prev) > + op->remap.prev = NULL; > + } > + if (op->remap.next) { > + err |= xe_vm_insert_vma(vm, op->remap.next); > + if (!err) > + op->flags |= XE_VMA_OP_NEXT_COMMITTED; > + if (!err && op->remap.skip_next) > + op->remap.next = NULL; > + } > + > + /* Adjust for partial unbind after removin VMA from VM */ > + if (!err) { > + op->base.remap.unmap->va->va.addr = op->remap.start; > + op->base.remap.unmap->va->va.range = op->remap.range; > + } > + break; > + case DRM_GPUVA_OP_UNMAP: > + prep_vma_destroy(vm, gpuva_to_vma(op->base.unmap.va), true); > + op->flags |= XE_VMA_OP_COMMITTED; > + break; > + case DRM_GPUVA_OP_PREFETCH: > + op->flags |= XE_VMA_OP_COMMITTED; > + break; > + default: > + XE_WARN_ON("NOT POSSIBLE"); > + } > + > + return err; > +} > + > + > static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q, > - struct drm_gpuva_ops **ops, int num_ops_list, > + struct drm_gpuva_ops *ops, > struct xe_sync_entry *syncs, u32 num_syncs, > - struct list_head *ops_list, bool async) > + struct list_head *ops_list, bool last, > + bool async) > { > struct xe_vma_op *last_op = NULL; > - struct list_head *async_list = NULL; > struct async_op_fence *fence = NULL; > - int err, i; > + struct drm_gpuva_op *__op; > + int err = 0; > > lockdep_assert_held_write(&vm->lock); > - XE_WARN_ON(num_ops_list > 1 && !async); > > - if (num_syncs && async) { > + if (last && num_syncs && async) { > u64 seqno; > > fence = kmalloc(sizeof(*fence), GFP_KERNEL); > @@ -2462,145 +2511,145 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q, > } > } > > - for (i = 0; i < num_ops_list; ++i) { > - struct drm_gpuva_ops *__ops = ops[i]; > - struct drm_gpuva_op *__op; > + drm_gpuva_for_each_op(__op, ops) { > + struct xe_vma_op *op = gpuva_op_to_vma_op(__op); > + bool first = list_empty(ops_list); > > - drm_gpuva_for_each_op(__op, __ops) { > - struct xe_vma_op *op = gpuva_op_to_vma_op(__op); > - bool first = !async_list; > + XE_WARN_ON(!first && !async); > + > + INIT_LIST_HEAD(&op->link); > + list_add_tail(&op->link, ops_list); > > - XE_WARN_ON(!first && !async); > + if (first) { > + op->flags |= XE_VMA_OP_FIRST; > + op->num_syncs = num_syncs; > + op->syncs = syncs; > + } > > - INIT_LIST_HEAD(&op->link); > - if (first) > - async_list = ops_list; > - list_add_tail(&op->link, async_list); > + op->q = q; > + > + switch (op->base.op) { > + case DRM_GPUVA_OP_MAP: > + { > + struct xe_vma *vma; > > - if (first) { > - op->flags |= XE_VMA_OP_FIRST; > - op->num_syncs = num_syncs; > - op->syncs = syncs; > + vma = new_vma(vm, &op->base.map, > + op->tile_mask, op->map.read_only, > + op->map.is_null); > + if (IS_ERR(vma)) { > + err = PTR_ERR(vma); > + goto free_fence; > } > > - op->q = q; > + op->map.vma = vma; > + break; > + } > + case DRM_GPUVA_OP_REMAP: > + { > + struct xe_vma *old = > + gpuva_to_vma(op->base.remap.unmap->va); > > - switch (op->base.op) { > - case DRM_GPUVA_OP_MAP: > - { > - struct xe_vma *vma; > + op->remap.start = xe_vma_start(old); > + op->remap.range = xe_vma_size(old); > > - vma = new_vma(vm, &op->base.map, > - op->tile_mask, op->map.read_only, > - op->map.is_null); > + if (op->base.remap.prev) { > + struct xe_vma *vma; > + bool read_only = > + op->base.remap.unmap->va->flags & > + XE_VMA_READ_ONLY; > + bool is_null = > + op->base.remap.unmap->va->flags & > + DRM_GPUVA_SPARSE; > + > + vma = new_vma(vm, op->base.remap.prev, > + op->tile_mask, read_only, > + is_null); > if (IS_ERR(vma)) { > err = PTR_ERR(vma); > goto free_fence; > } > > - op->map.vma = vma; > - break; > + op->remap.prev = vma; > + > + /* > + * Userptr creates a new SG mapping so > + * we must also rebind. > + */ > + op->remap.skip_prev = !xe_vma_is_userptr(old) && > + IS_ALIGNED(xe_vma_end(vma), > + xe_vma_max_pte_size(old)); > + if (op->remap.skip_prev) { > + xe_vma_set_pte_size(vma, xe_vma_max_pte_size(old)); > + op->remap.range -= > + xe_vma_end(vma) - > + xe_vma_start(old); > + op->remap.start = xe_vma_end(vma); > + } > } > - case DRM_GPUVA_OP_REMAP: > - { > - struct xe_vma *old = > - gpuva_to_vma(op->base.remap.unmap->va); > - > - op->remap.start = xe_vma_start(old); > - op->remap.range = xe_vma_size(old); > - > - if (op->base.remap.prev) { > - struct xe_vma *vma; > - bool read_only = > - op->base.remap.unmap->va->flags & > - XE_VMA_READ_ONLY; > - bool is_null = > - op->base.remap.unmap->va->flags & > - DRM_GPUVA_SPARSE; > - > - vma = new_vma(vm, op->base.remap.prev, > - op->tile_mask, read_only, > - is_null); > - if (IS_ERR(vma)) { > - err = PTR_ERR(vma); > - goto free_fence; > - } > - > - op->remap.prev = vma; > - > - /* > - * Userptr creates a new SG mapping so > - * we must also rebind. > - */ > - op->remap.skip_prev = !xe_vma_is_userptr(old) && > - IS_ALIGNED(xe_vma_end(vma), > - xe_vma_max_pte_size(old)); > - if (op->remap.skip_prev) { > - xe_vma_set_pte_size(vma, xe_vma_max_pte_size(old)); > - op->remap.range -= > - xe_vma_end(vma) - > - xe_vma_start(old); > - op->remap.start = xe_vma_end(vma); > - } > + > + if (op->base.remap.next) { > + struct xe_vma *vma; > + bool read_only = > + op->base.remap.unmap->va->flags & > + XE_VMA_READ_ONLY; > + > + bool is_null = > + op->base.remap.unmap->va->flags & > + DRM_GPUVA_SPARSE; > + > + vma = new_vma(vm, op->base.remap.next, > + op->tile_mask, read_only, > + is_null); > + if (IS_ERR(vma)) { > + err = PTR_ERR(vma); > + goto free_fence; > } > > - if (op->base.remap.next) { > - struct xe_vma *vma; > - bool read_only = > - op->base.remap.unmap->va->flags & > - XE_VMA_READ_ONLY; > - > - bool is_null = > - op->base.remap.unmap->va->flags & > - DRM_GPUVA_SPARSE; > - > - vma = new_vma(vm, op->base.remap.next, > - op->tile_mask, read_only, > - is_null); > - if (IS_ERR(vma)) { > - err = PTR_ERR(vma); > - goto free_fence; > - } > - > - op->remap.next = vma; > - > - /* > - * Userptr creates a new SG mapping so > - * we must also rebind. > - */ > - op->remap.skip_next = !xe_vma_is_userptr(old) && > - IS_ALIGNED(xe_vma_start(vma), > - xe_vma_max_pte_size(old)); > - if (op->remap.skip_next) { > - xe_vma_set_pte_size(vma, xe_vma_max_pte_size(old)); > - op->remap.range -= > - xe_vma_end(old) - > - xe_vma_start(vma); > - } > + op->remap.next = vma; > + > + /* > + * Userptr creates a new SG mapping so > + * we must also rebind. > + */ > + op->remap.skip_next = !xe_vma_is_userptr(old) && > + IS_ALIGNED(xe_vma_start(vma), > + xe_vma_max_pte_size(old)); > + if (op->remap.skip_next) { > + xe_vma_set_pte_size(vma, xe_vma_max_pte_size(old)); > + op->remap.range -= > + xe_vma_end(old) - > + xe_vma_start(vma); > } > - break; > - } > - case DRM_GPUVA_OP_UNMAP: > - case DRM_GPUVA_OP_PREFETCH: > - /* Nothing to do */ > - break; > - default: > - XE_WARN_ON("NOT POSSIBLE"); > } > - > - last_op = op; > + break; > + } > + case DRM_GPUVA_OP_UNMAP: > + case DRM_GPUVA_OP_PREFETCH: > + /* Nothing to do */ > + break; > + default: > + XE_WARN_ON("NOT POSSIBLE"); > } > > - last_op->ops = __ops; > + last_op = op; > + > + err = xe_vma_op_commit(vm, op); > + if (err) > + goto free_fence; > } > > - if (!last_op) > - return -ENODATA; > + /* FIXME: Unhandled corner case */ > + XE_WARN_ON(!last_op && last && !list_empty(ops_list)); > > - last_op->flags |= XE_VMA_OP_LAST; > - last_op->num_syncs = num_syncs; > - last_op->syncs = syncs; > - last_op->fence = fence; > + if (!last_op) > + goto free_fence; > + last_op->ops = ops; > + if (last) { > + last_op->flags |= XE_VMA_OP_LAST; > + last_op->num_syncs = num_syncs; > + last_op->syncs = syncs; > + last_op->fence = fence; > + } > > return 0; > > @@ -2609,58 +2658,6 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q, > return err; > } > > -static int xe_vma_op_commit(struct xe_vm *vm, struct xe_vma_op *op) > -{ > - int err = 0; > - > - lockdep_assert_held_write(&vm->lock); > - > - switch (op->base.op) { > - case DRM_GPUVA_OP_MAP: > - err |= xe_vm_insert_vma(vm, op->map.vma); > - if (!err) > - op->flags |= XE_VMA_OP_COMMITTED; > - break; > - case DRM_GPUVA_OP_REMAP: > - prep_vma_destroy(vm, gpuva_to_vma(op->base.remap.unmap->va), > - true); > - op->flags |= XE_VMA_OP_COMMITTED; > - > - if (op->remap.prev) { > - err |= xe_vm_insert_vma(vm, op->remap.prev); > - if (!err) > - op->flags |= XE_VMA_OP_PREV_COMMITTED; > - if (!err && op->remap.skip_prev) > - op->remap.prev = NULL; > - } > - if (op->remap.next) { > - err |= xe_vm_insert_vma(vm, op->remap.next); > - if (!err) > - op->flags |= XE_VMA_OP_NEXT_COMMITTED; > - if (!err && op->remap.skip_next) > - op->remap.next = NULL; > - } > - > - /* Adjust for partial unbind after removin VMA from VM */ > - if (!err) { > - op->base.remap.unmap->va->va.addr = op->remap.start; > - op->base.remap.unmap->va->va.range = op->remap.range; > - } > - break; > - case DRM_GPUVA_OP_UNMAP: > - prep_vma_destroy(vm, gpuva_to_vma(op->base.unmap.va), true); > - op->flags |= XE_VMA_OP_COMMITTED; > - break; > - case DRM_GPUVA_OP_PREFETCH: > - op->flags |= XE_VMA_OP_COMMITTED; > - break; > - default: > - XE_WARN_ON("NOT POSSIBLE"); > - } > - > - return err; > -} > - > static int __xe_vma_op_execute(struct xe_vm *vm, struct xe_vma *vma, > struct xe_vma_op *op) > { > @@ -2878,11 +2875,13 @@ static void xe_vma_op_unwind(struct xe_vm *vm, struct xe_vma_op *op, > { > struct xe_vma *vma = gpuva_to_vma(op->base.unmap.va); > > - down_read(&vm->userptr.notifier_lock); > - vma->gpuva.flags &= ~XE_VMA_DESTROYED; > - up_read(&vm->userptr.notifier_lock); > - if (post_commit) > - xe_vm_insert_vma(vm, vma); > + if (vma) { > + down_read(&vm->userptr.notifier_lock); > + vma->gpuva.flags &= ~XE_VMA_DESTROYED; > + up_read(&vm->userptr.notifier_lock); > + if (post_commit) > + xe_vm_insert_vma(vm, vma); > + } > break; > } > case DRM_GPUVA_OP_REMAP: > @@ -2897,11 +2896,13 @@ static void xe_vma_op_unwind(struct xe_vm *vm, struct xe_vma_op *op, > prep_vma_destroy(vm, op->remap.next, next_post_commit); > xe_vma_destroy_unlocked(op->remap.next); > } > - down_read(&vm->userptr.notifier_lock); > - vma->gpuva.flags &= ~XE_VMA_DESTROYED; > - up_read(&vm->userptr.notifier_lock); > - if (post_commit) > - xe_vm_insert_vma(vm, vma); > + if (vma) { > + down_read(&vm->userptr.notifier_lock); > + vma->gpuva.flags &= ~XE_VMA_DESTROYED; > + up_read(&vm->userptr.notifier_lock); > + if (post_commit) > + xe_vm_insert_vma(vm, vma); > + } > break; > } > case DRM_GPUVA_OP_PREFETCH: > @@ -2990,20 +2991,16 @@ static void xe_vma_op_work_func(struct work_struct *w) > } > } > > -static int vm_bind_ioctl_ops_commit(struct xe_vm *vm, > - struct list_head *ops_list, bool async) > +static int vm_bind_ioctl_ops_execute(struct xe_vm *vm, > + struct list_head *ops_list, bool async) > { > struct xe_vma_op *op, *last_op, *next; > int err; > > lockdep_assert_held_write(&vm->lock); > > - list_for_each_entry(op, ops_list, link) { > + list_for_each_entry(op, ops_list, link) > last_op = op; Use list_last_entry() ? > - err = xe_vma_op_commit(vm, op); > - if (err) > - goto unwind; > - } > > if (!async) { > err = xe_vma_op_execute(vm, last_op); > @@ -3042,28 +3039,29 @@ static int vm_bind_ioctl_ops_commit(struct xe_vm *vm, > return err; > } > > -/* > - * Unwind operations list, called after a failure of vm_bind_ioctl_ops_create or > - * vm_bind_ioctl_ops_parse. > - */ > static void vm_bind_ioctl_ops_unwind(struct xe_vm *vm, > struct drm_gpuva_ops **ops, > int num_ops_list) > { > int i; > > - for (i = 0; i < num_ops_list; ++i) { > + for (i = num_ops_list - 1; i; ++i) { > struct drm_gpuva_ops *__ops = ops[i]; > struct drm_gpuva_op *__op; > > if (!__ops) > continue; > > - drm_gpuva_for_each_op(__op, __ops) { > + drm_gpuva_for_each_op_reverse(__op, __ops) { > struct xe_vma_op *op = gpuva_op_to_vma_op(__op); > > - xe_vma_op_unwind(vm, op, false, false, false); > + xe_vma_op_unwind(vm, op, > + op->flags & XE_VMA_OP_COMMITTED, > + op->flags & XE_VMA_OP_PREV_COMMITTED, > + op->flags & XE_VMA_OP_NEXT_COMMITTED); > } > + > + drm_gpuva_ops_free(&vm->mgr, __ops); > } > } > > @@ -3384,14 +3382,22 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) > ops[i] = NULL; > goto unwind_ops; > } > + > + err = vm_bind_ioctl_ops_parse(vm, q, ops[i], syncs, num_syncs, > + &ops_list, > + i == args->num_binds - 1, > + async); > + if (err) > + goto unwind_ops; > } > > - err = vm_bind_ioctl_ops_parse(vm, q, ops, args->num_binds, > - syncs, num_syncs, &ops_list, async); > - if (err) > + /* Nothing to do */ > + if (list_empty(&ops_list)) { > + err = -ENODATA; > goto unwind_ops; > + } > > - err = vm_bind_ioctl_ops_commit(vm, &ops_list, async); > + err = vm_bind_ioctl_ops_execute(vm, &ops_list, async); > up_write(&vm->lock); > > for (i = 0; i < args->num_binds; ++i)