From: Joerg Roedel <joro-zLv9SwRftAIdnm+yROfE0A@public.gmane.org>
To: Robin Murphy <robin.murphy-5wv7dgnIgG8@public.gmane.org>
Cc: Vincent.Wan-5C7GfCeVMHo@public.gmane.org,
iommu-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org,
Joerg Roedel <jroedel-l3A5Bk7waGM@public.gmane.org>,
linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
Subject: [PATCH 16/20 v2] iommu/amd: Optimize map_sg and unmap_sg
Date: Tue, 12 Jul 2016 15:30:42 +0200 [thread overview]
Message-ID: <20160712133042.GG12639@8bytes.org> (raw)
In-Reply-To: <5784D597.4010703-5wv7dgnIgG8@public.gmane.org>
On Tue, Jul 12, 2016 at 12:33:43PM +0100, Robin Murphy wrote:
> > + for_each_sg(sglist, s, nelems, i)
> > + npages += iommu_num_pages(sg_phys(s), s->length, PAGE_SIZE);
>
> This fails to account for the segment boundary mask[1]. Given a typical
> sglist from the block layer where the boundary mask is 64K, the first
> segment is 8k long, and subsequent segments are 64K long, those
> subsequent segments will end up with misaligned addresses which certain
> hardware may object to.
Yeah, right. It doesn't matter much on x86, as the smallest
boundary-mask I have seen is 4G, but to be correct it should be
accounted in. How does the attached patch look?
>
> > + address = dma_ops_alloc_iova(dev, dma_dom, npages, dma_mask);
>
> Since a typical dma_map_sg() call is likely to involve >128K worth of
> data, I wonder if it's worth going directly to a slow-path IOVA
> allocation...
Well, the allocator is the bottle-neck, so I try not to call it for
every sg-element. The global locks have been removed, but more
allocations/deallocations also mean that the per-cpu free-lists fill up
quicker and that we have to flush the IOTLBs more often, which costs
performance.
> [1]:http://article.gmane.org/gmane.linux.kernel.iommu/10553 - almost the
> 1-year anniversary of you making much the same comment to me :D
Touché ;-)
Here is the updated patch:
>From 88e1cc6c8e854a2bf55f972ddc5082a44760abe2 Mon Sep 17 00:00:00 2001
From: Joerg Roedel <jroedel-l3A5Bk7waGM@public.gmane.org>
Date: Wed, 6 Jul 2016 17:20:54 +0200
Subject: [PATCH 1/5] iommu/amd: Optimize map_sg and unmap_sg
Optimize these functions so that they need only one call
into the address alloctor. This also saves a couple of
io-tlb flushes in the unmap_sg path.
Signed-off-by: Joerg Roedel <jroedel-l3A5Bk7waGM@public.gmane.org>
---
drivers/iommu/amd_iommu.c | 112 +++++++++++++++++++++++++++++++++++-----------
1 file changed, 86 insertions(+), 26 deletions(-)
diff --git a/drivers/iommu/amd_iommu.c b/drivers/iommu/amd_iommu.c
index 2cd382e..203c50c 100644
--- a/drivers/iommu/amd_iommu.c
+++ b/drivers/iommu/amd_iommu.c
@@ -2395,50 +2395,110 @@ static void unmap_page(struct device *dev, dma_addr_t dma_addr, size_t size,
__unmap_single(domain->priv, dma_addr, size, dir);
}
+static int sg_num_pages(struct device *dev,
+ struct scatterlist *sglist,
+ int nelems)
+{
+ unsigned long mask, boundary_size;
+ struct scatterlist *s;
+ int i, npages = 0;
+
+ mask = dma_get_seg_boundary(dev);
+ boundary_size = mask + 1 ? ALIGN(mask + 1, PAGE_SIZE) >> PAGE_SHIFT :
+ 1UL << (BITS_PER_LONG - PAGE_SHIFT);
+
+ for_each_sg(sglist, s, nelems, i) {
+ int p, n;
+
+ s->dma_address = npages << PAGE_SHIFT;
+ p = npages % boundary_size;
+ n = iommu_num_pages(sg_phys(s), s->length, PAGE_SIZE);
+ if (p + n > boundary_size)
+ npages += boundary_size - p;
+ npages += n;
+ }
+
+ return npages;
+}
+
/*
* The exported map_sg function for dma_ops (handles scatter-gather
* lists).
*/
static int map_sg(struct device *dev, struct scatterlist *sglist,
- int nelems, enum dma_data_direction dir,
+ int nelems, enum dma_data_direction direction,
struct dma_attrs *attrs)
{
+ int mapped_pages = 0, npages = 0, prot = 0, i;
struct protection_domain *domain;
- int i;
+ struct dma_ops_domain *dma_dom;
struct scatterlist *s;
- phys_addr_t paddr;
- int mapped_elems = 0;
+ unsigned long address;
u64 dma_mask;
domain = get_domain(dev);
if (IS_ERR(domain))
return 0;
+ dma_dom = domain->priv;
dma_mask = *dev->dma_mask;
+ npages = sg_num_pages(dev, sglist, nelems);
+
+ address = dma_ops_alloc_iova(dev, dma_dom, npages, dma_mask);
+ if (address == DMA_ERROR_CODE)
+ goto out_err;
+
+ prot = dir2prot(direction);
+
+ /* Map all sg entries */
for_each_sg(sglist, s, nelems, i) {
- paddr = sg_phys(s);
+ int j, pages = iommu_num_pages(sg_phys(s), s->length, PAGE_SIZE);
+
+ for (j = 0; j < pages; ++j) {
+ unsigned long bus_addr, phys_addr;
+ int ret;
- s->dma_address = __map_single(dev, domain->priv,
- paddr, s->length, dir, dma_mask);
+ bus_addr = address + s->dma_address + (j << PAGE_SHIFT);
+ phys_addr = (sg_phys(s) & PAGE_MASK) + (j << PAGE_SHIFT);
+ ret = iommu_map_page(domain, bus_addr, phys_addr, PAGE_SIZE, prot, GFP_ATOMIC);
+ if (ret)
+ goto out_unmap;
- if (s->dma_address) {
- s->dma_length = s->length;
- mapped_elems++;
- } else
- goto unmap;
+ mapped_pages += 1;
+ }
}
- return mapped_elems;
+ /* Everything is mapped - write the right values into s->dma_address */
+ for_each_sg(sglist, s, nelems, i) {
+ s->dma_address += address + s->offset;
+ s->dma_length = s->length;
+ }
+
+ return nelems;
+
+out_unmap:
+ pr_err("%s: IOMMU mapping error in map_sg (io-pages: %d)\n",
+ dev_name(dev), npages);
+
+ for_each_sg(sglist, s, nelems, i) {
+ int j, pages = iommu_num_pages(sg_phys(s), s->length, PAGE_SIZE);
+
+ for (j = 0; j < pages; ++j) {
+ unsigned long bus_addr;
-unmap:
- for_each_sg(sglist, s, mapped_elems, i) {
- if (s->dma_address)
- __unmap_single(domain->priv, s->dma_address,
- s->dma_length, dir);
- s->dma_address = s->dma_length = 0;
+ bus_addr = address + s->dma_address + (j << PAGE_SHIFT);
+ iommu_unmap_page(domain, bus_addr, PAGE_SIZE);
+
+ if (--mapped_pages)
+ goto out_free_iova;
+ }
}
+out_free_iova:
+ free_iova_fast(&dma_dom->iovad, address, npages);
+
+out_err:
return 0;
}
@@ -2451,18 +2511,18 @@ static void unmap_sg(struct device *dev, struct scatterlist *sglist,
struct dma_attrs *attrs)
{
struct protection_domain *domain;
- struct scatterlist *s;
- int i;
+ unsigned long startaddr;
+ int npages = 2;
domain = get_domain(dev);
if (IS_ERR(domain))
return;
- for_each_sg(sglist, s, nelems, i) {
- __unmap_single(domain->priv, s->dma_address,
- s->dma_length, dir);
- s->dma_address = s->dma_length = 0;
- }
+ startaddr = sg_dma_address(sglist) & PAGE_MASK;
+ dma_dom = domain->priv;
+ npages = sg_num_pages(dev, sglist, nelems);
+
+ __unmap_single(domain->priv, startaddr, npages << PAGE_SHIFT, dir);
}
/*
--
2.6.6
next prev parent reply other threads:[~2016-07-12 13:30 UTC|newest]
Thread overview: 33+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-07-08 11:44 [PATCH 00/20] iommu/amd: Use generic IOVA allocator Joerg Roedel
[not found] ` <1467978311-28322-1-git-send-email-joro-zLv9SwRftAIdnm+yROfE0A@public.gmane.org>
2016-07-08 11:44 ` [PATCH 01/20] iommu: Add apply_dm_region call-back to iommu-ops Joerg Roedel
2016-07-08 11:44 ` [PATCH 02/20] iommu/amd: Select IOMMU_IOVA for AMD IOMMU Joerg Roedel
2016-07-08 11:44 ` [PATCH 03/20] iommu/amd: Allocate iova_domain for dma_ops_domain Joerg Roedel
2016-07-08 11:44 ` [PATCH 04/20] iommu/amd: Create a list of reserved iova addresses Joerg Roedel
2016-07-08 11:44 ` [PATCH 05/20] iommu/amd: Implement apply_dm_region call-back Joerg Roedel
2016-07-08 11:44 ` [PATCH 07/20] iommu/amd: Remove special mapping code for dma_ops path Joerg Roedel
[not found] ` <1467978311-28322-8-git-send-email-joro-zLv9SwRftAIdnm+yROfE0A@public.gmane.org>
2016-07-12 10:55 ` Robin Murphy
2016-07-12 11:08 ` Joerg Roedel
2016-07-12 11:42 ` Robin Murphy
[not found] ` <5784D7C3.4010104-5wv7dgnIgG8@public.gmane.org>
2016-07-12 11:48 ` Joerg Roedel
2016-07-08 11:44 ` [PATCH 08/20] iommu/amd: Make use of the generic IOVA allocator Joerg Roedel
2016-07-08 11:45 ` [PATCH 09/20] iommu/amd: Remove other remains of old address allocator Joerg Roedel
2016-07-08 11:45 ` [PATCH 10/20] iommu/amd: Remove align-parameter from __map_single() Joerg Roedel
2016-07-08 11:45 ` [PATCH 11/20] iommu/amd: Set up data structures for flush queue Joerg Roedel
2016-07-08 11:45 ` [PATCH 12/20] iommu/amd: Allow NULL pointer parameter for domain_flush_complete() Joerg Roedel
2016-07-08 11:45 ` [PATCH 13/20] iommu/amd: Implement flush queue Joerg Roedel
2016-07-08 11:45 ` [PATCH 14/20] iommu/amd: Implement timeout to flush unmap queues Joerg Roedel
2016-07-08 11:45 ` [PATCH 15/20] iommu/amd: Introduce dir2prot() helper Joerg Roedel
2016-07-08 11:45 ` [PATCH 17/20] iommu/amd: Use dev_data->domain in get_domain() Joerg Roedel
2016-07-08 11:44 ` [PATCH 06/20] iommu/amd: Pass gfp-flags to iommu_map_page() Joerg Roedel
2016-07-08 11:45 ` [PATCH 16/20] iommu/amd: Optimize map_sg and unmap_sg Joerg Roedel
2016-07-12 11:33 ` Robin Murphy
[not found] ` <5784D597.4010703-5wv7dgnIgG8@public.gmane.org>
2016-07-12 13:30 ` Joerg Roedel [this message]
2016-07-12 15:34 ` [PATCH 16/20 v2] " Robin Murphy
[not found] ` <57850DF8.9040507-5wv7dgnIgG8@public.gmane.org>
2016-07-13 10:27 ` Joerg Roedel
2016-07-08 11:45 ` [PATCH 18/20] iommu/amd: Handle IOMMU_DOMAIN_DMA in ops->domain_free call-back Joerg Roedel
2016-07-08 11:45 ` [PATCH 19/20] iommu/amd: Flush iova queue before releasing dma_ops_domain Joerg Roedel
2016-07-08 11:45 ` [PATCH 20/20] iommu/amd: Use container_of to get dma_ops_domain Joerg Roedel
2016-07-12 9:03 ` [PATCH 00/20] iommu/amd: Use generic IOVA allocator Wan Zongshun
2016-07-12 10:55 ` Joerg Roedel
[not found] ` <20160712105533.GE12639-zLv9SwRftAIdnm+yROfE0A@public.gmane.org>
2016-07-13 9:44 ` Wan Zongshun
[not found] ` <57860D78.5090409-6ukY98dZOFrYtjvyW6yDsg@public.gmane.org>
2016-07-13 9:51 ` Joerg Roedel
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20160712133042.GG12639@8bytes.org \
--to=joro-zlv9swrftaidnm+yrofe0a@public.gmane.org \
--cc=Vincent.Wan-5C7GfCeVMHo@public.gmane.org \
--cc=iommu-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org \
--cc=jroedel-l3A5Bk7waGM@public.gmane.org \
--cc=linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org \
--cc=robin.murphy-5wv7dgnIgG8@public.gmane.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).