Message-ID: <20181120164636.jcw7li2uaa3cmwc3@DESKTOP-E1NTVVP.localdomain>
Date:   Tue, 20 Nov 2018 16:46:37 +0000
From:   Brian Starkey <Brian.Starkey@....com>
To:     Liam Mark <lmark@...eaurora.org>
CC:     nd <nd@....com>, Sumit Semwal <sumit.semwal@...aro.org>,
        "linux-arm-kernel@...ts.infradead.org" 
        <linux-arm-kernel@...ts.infradead.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "devel@...verdev.osuosl.org" <devel@...verdev.osuosl.org>,
        Martijn Coenen <maco@...roid.com>,
        dri-devel <dri-devel@...ts.freedesktop.org>,
        John Stultz <john.stultz@...aro.org>,
        Todd Kjos <tkjos@...roid.com>,
        Arve Hjonnevag <arve@...roid.com>,
        "linaro-mm-sig@...ts.linaro.org" <linaro-mm-sig@...ts.linaro.org>,
        Laura Abbott <labbott@...hat.com>
Subject: Re: [RFC PATCH v2] android: ion: How to properly clean caches for
 uncached allocations

Hi Liam,

I'm missing a bit of context here, but I did read the v1 thread.
Please accept my apologies if I'm re-treading trodden ground.

I do know we're chasing nebulous ion "problems" on our end, which
certainly seem to be related to what you're trying to fix here.

On Thu, Nov 01, 2018 at 03:15:06PM -0700, Liam Mark wrote:
>Based on the suggestions from Laura I created a first draft for a change
>which will attempt to ensure that uncached mappings are only applied to
>ION memory who's cache lines have been cleaned.
>It does this by providing cached mappings (for uncached ION allocations)
>until the ION buffer is dma mapped and successfully cleaned, then it drops
>the userspace mappings and when pages are accessed they are faulted back
>in and uncached mappings are created.

If I understand right, there's no way to portably clean the cache of
the kernel mapping before we map the pages into userspace. Is that
right?

Alternatively, can we just make ion refuse to give userspace a
non-cached mapping for pages which are mapped in the kernel as cached?
Would userspace using the dma-buf sync ioctl around its accesses do
the "right thing" in that case?

Given that as you pointed out, the kernel does still have a cached
mapping to these pages, trying to give the CPU a non-cached mapping of
those same pages while preserving consistency seems fraught. Wouldn't
it be better to make sure all CPU mappings are cached, and have CPU
clients use the dma_buf_{begin,end}_cpu_access() hooks to get
consistency where needed?

>
>This change has the following potential disadvantages:
>- It assumes that userpace clients won't attempt to access the buffer
>while it is being mapped as we are removing the userpspace mappings at
>this point (though it is okay for them to have it mapped)
>- It assumes that kernel clients won't hold a kernel mapping to the buffer
>(ie dma_buf_kmap) while it is being dma-mapped. What should we do if there
>is a kernel mapping at the time of dma mapping, fail the mapping, warn?
>- There may be a performance penalty as a result of having to fault in the
>pages after removing the userspace mappings.

I wonder if the dma-buf sync ioctl might provide a way for userspace
to opt-in to when the zap/fault happens. Zap on (DMA_BUF_SYNC_WRITE |
DMA_BUF_SYNC_WRITE_END) and fault on (DMA_BUF_SYNC_READ |
DMA_BUF_SYNC_START)

>
>It passes basic testing involving reading writing and reading from
>uncached system heap allocations before and after dma mapping.
>
>Please let me know if this is heading in the right direction and if there
>are any concerns.
>
>Signed-off-by: Liam Mark <lmark at codeaurora.org>
>---
> drivers/staging/android/ion/ion.c | 146 +++++++++++++++++++++++++++++++++++++-
> drivers/staging/android/ion/ion.h |   9 +++
> 2 files changed, 152 insertions(+), 3 deletions(-)
>
>diff --git a/drivers/staging/android/ion/ion.c b/drivers/staging/android/ion/ion.c
>index 99073325b0c0..3dc0f5a265bf 100644
>--- a/drivers/staging/android/ion/ion.c
>+++ b/drivers/staging/android/ion/ion.c
>@@ -96,6 +96,7 @@ static struct ion_buffer *ion_buffer_create(struct ion_heap *heap,
> 	}
>
> 	INIT_LIST_HEAD(&buffer->attachments);
>+	INIT_LIST_HEAD(&buffer->vmas);
> 	mutex_init(&buffer->lock);
> 	mutex_lock(&dev->buffer_lock);
> 	ion_buffer_add(dev, buffer);
>@@ -117,6 +118,7 @@ void ion_buffer_destroy(struct ion_buffer *buffer)
> 		buffer->heap->ops->unmap_kernel(buffer->heap, buffer);
> 	}
> 	buffer->heap->ops->free(buffer);
>+	vfree(buffer->pages);
> 	kfree(buffer);
> }
>
>@@ -245,11 +247,29 @@ static void ion_dma_buf_detatch(struct dma_buf *dmabuf,
> 	kfree(a);
> }
>
>+static bool ion_buffer_uncached_clean(struct ion_buffer *buffer)
>+{
>+	return buffer->uncached_clean;
>+}

nit: The function name sounds like a verb to me - as in "calling this
will clean the buffer". I feel ion_buffer_is_uncached_clean() would
read better.

Thanks,
-Brian

>+
>+/* expect buffer->lock to be already taken */
>+static void ion_buffer_zap_mappings(struct ion_buffer *buffer)
>+{
>+	struct ion_vma_list *vma_list;
>+
>+	list_for_each_entry(vma_list, &buffer->vmas, list) {
>+		struct vm_area_struct *vma = vma_list->vma;
>+
>+		zap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start);
>+	}
>+}
>+
> static struct sg_table *ion_map_dma_buf(struct dma_buf_attachment *attachment,
> 					enum dma_data_direction direction)
> {
> 	struct ion_dma_buf_attachment *a = attachment->priv;
> 	struct sg_table *table;
>+	struct ion_buffer *buffer = attachment->dmabuf->priv;
>
> 	table = a->table;
>
>@@ -257,6 +277,19 @@ static struct sg_table *ion_map_dma_buf(struct dma_buf_attachment *attachment,
> 			direction))
> 		return ERR_PTR(-ENOMEM);
>
>+	if (!ion_buffer_cached(buffer)) {
>+		mutex_lock(&buffer->lock);
>+		if (!ion_buffer_uncached_clean(buffer)) {
>+			ion_buffer_zap_mappings(buffer);
>+			if (buffer->kmap_cnt > 0) {
>+				pr_warn_once("%s: buffer still mapped in the kernel\n",
>+					     __func__);
>+			}
>+			buffer->uncached_clean = true;
>+		}
>+		mutex_unlock(&buffer->lock);
>+	}
>+
> 	return table;
> }
>
>@@ -267,6 +300,94 @@ static void ion_unmap_dma_buf(struct dma_buf_attachment *attachment,
> 	dma_unmap_sg(attachment->dev, table->sgl, table->nents, direction);
> }
>
>+static void __ion_vm_open(struct vm_area_struct *vma, bool lock)
>+{
>+	struct ion_buffer *buffer = vma->vm_private_data;
>+	struct ion_vma_list *vma_list;
>+
>+	vma_list = kmalloc(sizeof(*vma_list), GFP_KERNEL);
>+	if (!vma_list)
>+		return;
>+	vma_list->vma = vma;
>+
>+	if (lock)
>+		mutex_lock(&buffer->lock);
>+	list_add(&vma_list->list, &buffer->vmas);
>+	if (lock)
>+		mutex_unlock(&buffer->lock);
>+}
>+
>+static void ion_vm_open(struct vm_area_struct *vma)
>+{
>+	__ion_vm_open(vma, true);
>+}
>+
>+static void ion_vm_close(struct vm_area_struct *vma)
>+{
>+	struct ion_buffer *buffer = vma->vm_private_data;
>+	struct ion_vma_list *vma_list, *tmp;
>+
>+	mutex_lock(&buffer->lock);
>+	list_for_each_entry_safe(vma_list, tmp, &buffer->vmas, list) {
>+		if (vma_list->vma != vma)
>+			continue;
>+		list_del(&vma_list->list);
>+		kfree(vma_list);
>+		break;
>+	}
>+	mutex_unlock(&buffer->lock);
>+}
>+
>+static int ion_vm_fault(struct vm_fault *vmf)
>+{
>+	struct vm_area_struct *vma = vmf->vma;
>+	struct ion_buffer *buffer = vma->vm_private_data;
>+	unsigned long pfn;
>+	int ret;
>+
>+	mutex_lock(&buffer->lock);
>+	if (!buffer->pages || !buffer->pages[vmf->pgoff]) {
>+		mutex_unlock(&buffer->lock);
>+		return VM_FAULT_ERROR;
>+	}
>+
>+	vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
>+	pfn = page_to_pfn(buffer->pages[vmf->pgoff]);
>+	ret = vm_insert_pfn(vma, vmf->address, pfn);
>+	mutex_unlock(&buffer->lock);
>+	if (ret)
>+		return VM_FAULT_ERROR;
>+
>+	return VM_FAULT_NOPAGE;
>+}
>+
>+static const struct vm_operations_struct ion_vma_ops = {
>+	.open = ion_vm_open,
>+	.close = ion_vm_close,
>+	.fault = ion_vm_fault,
>+};
>+
>+static int ion_init_fault_pages(struct ion_buffer *buffer)
>+{
>+	int num_pages = PAGE_ALIGN(buffer->size) / PAGE_SIZE;
>+	struct scatterlist *sg;
>+	int i, j, k = 0;
>+	struct sg_table *table = buffer->sg_table;
>+
>+	buffer->pages = vmalloc(sizeof(struct page *) * num_pages);
>+	if (!buffer->pages)
>+		return -ENOMEM;
>+
>+	for_each_sg(table->sgl, sg, table->nents, i) {
>+		struct page *page = sg_page(sg);
>+
>+		for (j = 0; j < sg->length / PAGE_SIZE; j++)
>+			buffer->pages[k++] = page++;
>+	}
>+
>+	return 0;
>+}
>+
> static int ion_mmap(struct dma_buf *dmabuf, struct vm_area_struct *vma)
> {
> 	struct ion_buffer *buffer = dmabuf->priv;
>@@ -278,12 +399,31 @@ static int ion_mmap(struct dma_buf *dmabuf, struct vm_area_struct *vma)
> 		return -EINVAL;
> 	}
>
>-	if (!(buffer->flags & ION_FLAG_CACHED))
>-		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
>-
> 	mutex_lock(&buffer->lock);
>+
>+	if (!ion_buffer_cached(buffer)) {
>+		if (!ion_buffer_uncached_clean(buffer)) {
>+			if (!buffer->pages)
>+				ret = ion_init_fault_pages(buffer);
>+
>+			if (ret)
>+				goto end;
>+
>+			vma->vm_private_data = buffer;
>+			vma->vm_ops = &ion_vma_ops;
>+			vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND |
>+					 VM_DONTDUMP;
>+			__ion_vm_open(vma, false);
>+		} else {
>+			vma->vm_page_prot =
>+				pgprot_writecombine(vma->vm_page_prot);
>+		}
>+	}
>+
> 	/* now map it to userspace */
> 	ret = buffer->heap->ops->map_user(buffer->heap, buffer, vma);
>+
>+end:
> 	mutex_unlock(&buffer->lock);
>
> 	if (ret)
>diff --git a/drivers/staging/android/ion/ion.h b/drivers/staging/android/ion/ion.h
>index c006fc1e5a16..438c9f4fa125 100644
>--- a/drivers/staging/android/ion/ion.h
>+++ b/drivers/staging/android/ion/ion.h
>@@ -44,6 +44,11 @@ struct ion_platform_heap {
> 	void *priv;
> };
>
>+struct ion_vma_list {
>+	struct list_head list;
>+	struct vm_area_struct *vma;
>+};
>+
> /**
>  * struct ion_buffer - metadata for a particular buffer
>  * @ref:		reference count
>@@ -59,6 +64,7 @@ struct ion_platform_heap {
>  * @kmap_cnt:		number of times the buffer is mapped to the kernel
>  * @vaddr:		the kernel mapping if kmap_cnt is not zero
>  * @sg_table:		the sg table for the buffer if dmap_cnt is not zero
>+ * @vmas:		list of vma's mapping for uncached buffer
>  */
> struct ion_buffer {
> 	union {
>@@ -76,6 +82,9 @@ struct ion_buffer {
> 	void *vaddr;
> 	struct sg_table *sg_table;
> 	struct list_head attachments;
>+	struct list_head vmas;
>+	struct page **pages;
>+	bool uncached_clean;
> };
>
> void ion_buffer_destroy(struct ion_buffer *buffer);
>-- 
>1.9.1
>
>
>Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
>a Linux Foundation Collaborative Project
>

Powered by blists - more mailing lists