Hi, Robin, Thanks very much for your comment, and sorry for the last
reply format.

On Tue, 2016-05-10 at 11:28 +0100, Robin Murphy wrote:

On 09/05/16 09:00, honghui.zhang-NuS5LvNUpcJWk0Htik3J/w@public.gmane.org wrote:
[...]

quoted

+static void *mtk_iommu_alloc_pgt(struct device *dev, size_t size, gfp_t gfp)
+{
+	dma_addr_t dma;
+	void *pages = alloc_pages_exact(size, gfp | __GFP_ZERO);
+
+	if (!pages)
+		return NULL;
+
+	dma = dma_map_single(dev, pages, size, DMA_TO_DEVICE);
+	if (dma_mapping_error(dev, dma))
+		goto out_free;
+	/*
+	 * We depend on the IOMMU being able to work with any physical
+	 * address directly, so if the DMA layer suggests otherwise by
+	 * translating or truncating them, that bodes very badly...
+	 */
+	if (dma != virt_to_phys(pages))
+		goto out_unmap;

Given that you've only got a single table to allocate, and at 4MB it has 
a fair chance of failing beyond early boot time, just use 
dma_alloc_coherent() - you don't need to care about the dma <-> phys 
relationship because you don't have multi-level tables to walk. That 
way, you can get rid of all the awkward streaming DMA stuff, and also 
benefit from CMA to avoid allocation failures.

The dma_alloc_coheret interface is good enough for me, thanks.

quoted

+	kmemleak_ignore(pages);
+	return pages;
+
+out_unmap:
+	dev_err(dev, "Cannot accommodate DMA translation for IOMMU page tables\n");
+	dma_unmap_single(dev, dma, size, DMA_TO_DEVICE);
+out_free:
+	free_pages_exact(pages, size);
+	return NULL;
+
+}
+
+static void mtk_iommu_free_pgt(struct device *dev, void *pages, size_t size)
+{
+	dma_unmap_single(dev, (dma_addr_t)virt_to_phys(pages),
+			 size, DMA_TO_DEVICE);
+	free_pages_exact(pages, size);
+}
+
+static int mtk_iommu_domain_finalise(struct mtk_iommu_data *data)
+{
+	struct mtk_iommu_domain *dom = data->m4u_dom;
+
+	spin_lock_init(&dom->pgtlock);
+
+	dom->pgt_va = mtk_iommu_alloc_pgt(data->dev,
+				dom->pgt_size, GFP_KERNEL);
+	if (!dom->pgt_va)
+		return -ENOMEM;
+
+	dom->pgt_pa = virt_to_phys(dom->pgt_va);
+
+	writel(dom->pgt_pa, data->base + REG_MMU_PT_BASE_ADDR);
+
+	dom->cookie = (void *)data;
+
+	return 0;
+}
+
+static struct iommu_domain *mtk_iommu_domain_alloc(unsigned type)
+{
+	struct mtk_iommu_domain *dom;
+
+	if (type != IOMMU_DOMAIN_UNMANAGED)
+		return NULL;
+
+	dom = kzalloc(sizeof(*dom), GFP_KERNEL);
+	if (!dom)
+		return NULL;
+
+	/*
+	 * MTK m4u support 4GB iova address space, and oly support 4K page
+	 * mapping. So the pagetable size should be exactly as 4M.
+	 */
+	dom->pgt_size = SZ_4M;

If the table size is fixed, then why bother having a variable at all?

I will follow your advise for next version.
thanks.

quoted

+	return &dom->domain;
+}
+
+static void mtk_iommu_domain_free(struct iommu_domain *domain)
+{
+	kfree(to_mtk_domain(domain));
+}
+

[...]

quoted

+static int mtk_iommu_map(struct iommu_domain *domain, unsigned long iova,
+			 phys_addr_t paddr, size_t size, int prot)
+{
+	struct mtk_iommu_domain *dom = to_mtk_domain(domain);
+	struct mtk_iommu_data *data = dom->cookie;
+	unsigned int page_num = size >> MTK_IOMMU_PAGE_SHIFT;

Since you only advertise a single page size, this will always be 1, so 
you could either get rid of the loop here...

I would prefer your following advise to modify the pgsize_bitmap, thanks
very much.

quoted

+	unsigned long flags;
+	unsigned int i;
+	u32 *pgt_base_iova;
+	u32 pabase = (u32)paddr;
+	int map_size = 0;
+
+	spin_lock_irqsave(&dom->pgtlock, flags);
+	pgt_base_iova = dom->pgt_va + (iova  >> MTK_IOMMU_PAGE_SHIFT);
+	for (i = 0; i < page_num; i++) {
+		pgt_base_iova[i] = pabase | F_DESC_VALID | F_DESC_NONSEC;
+		pabase += MTK_IOMMU_PAGE_SIZE;
+		map_size += MTK_IOMMU_PAGE_SIZE;
+	}
+	dma_sync_single_for_device(data->dev,
+			dom->pgt_pa + (iova >> MTK_IOMMU_PAGE_SHIFT),
+			(size >> MTK_IOMMU_PAGE_SHIFT) * sizeof(u32),
+			DMA_TO_DEVICE);
+	spin_unlock_irqrestore(&dom->pgtlock, flags);
+
+	mtk_iommu_tlb_flush_range(data, iova, size);
+
+	return map_size;
+}

[...]

quoted

+static struct iommu_ops mtk_iommu_ops = {
+	.domain_alloc	= mtk_iommu_domain_alloc,
+	.domain_free	= mtk_iommu_domain_free,
+	.attach_dev	= mtk_iommu_attach_device,
+	.detach_dev	= mtk_iommu_detach_device,
+	.map		= mtk_iommu_map,
+	.unmap		= mtk_iommu_unmap,
+	.map_sg		= default_iommu_map_sg,
+	.iova_to_phys	= mtk_iommu_iova_to_phys,
+	.add_device	= mtk_iommu_add_device,
+	.remove_device	= mtk_iommu_remove_device,
+	.device_group	= mtk_iommu_device_group,
+	.pgsize_bitmap	= MTK_IOMMU_PAGE_SIZE,
+};

...or perhaps advertise .pgsize_bitmap = ~0UL << MTK_IOMMU_PAGE_SHIFT 
here, so you actually can handle multiple entries at once for larger 
mappings - given how simple the page table format is that doesn't seem 
too unreasonable, especially since it should give you a big efficiency 
win in terms of TLB maintenance.

Robin.

--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html