From: John Garry <john.garry@huawei.com>
To: <joro@8bytes.org>, <will@kernel.org>, <robin.murphy@arm.com>,
<hch@lst.de>, <m.szyprowski@samsung.com>
Cc: <iommu@lists.linux-foundation.org>, <baolu.lu@linux.intel.com>,
<linux-kernel@vger.kernel.org>, <linux-scsi@vger.kernel.org>,
<linuxarm@huawei.com>
Subject: Re: [PATCH v2 00/15] dma mapping/iommu: Allow IOMMU IOVA rcache range to be configured
Date: Thu, 20 May 2021 09:28:36 +0100 [thread overview]
Message-ID: <42aff9e5-096c-f47d-60f5-08cd39152c1c@huawei.com> (raw)
In-Reply-To: <1620656249-68890-1-git-send-email-john.garry@huawei.com>
On 10/05/2021 15:17, John Garry wrote:
Hi Robin, guys,
A friendly reminder on this one...
Thanks
> For streaming DMA mappings involving an IOMMU and whose IOVA len regularly
> exceeds the IOVA rcache upper limit (meaning that they are not cached),
> performance can be reduced.
>
> This is much more pronounced from commit 4e89dce72521 ("iommu/iova: Retry
> from last rb tree node if iova search fails"), as discussed at [0].
>
> IOVAs which cannot be cached are highly involved in the IOVA aging issue,
> as discussed at [1].
>
> This series allows the IOVA rcache range be configured, so that we may
> cache all IOVAs per domain, thus improving performance.
>
> A new IOMMU group sysfs file is added - max_opt_dma_size - which is used
> indirectly to configure the IOVA rcache range:
> /sys/kernel/iommu_groups/X/max_opt_dma_size
>
> This file is updated same as how the IOMMU group default domain type is
> updated, i.e. must unbind the only device in the group first. However, the
> IOMMU default domain is reallocated in the device driver reprobe, and not
> immediately.
>
> In addition, we keep (from v1 series) the DMA mapping API to allow DMA max
> optimised size be set from a LLDD. How it works is a lot different. When
> the LLDD calls this during probe, once the value is successfully recorded, we
> return -EDEFER_PROBE. In the reprobe, the IOMM group default domain is
> reallocated, and the new IOVA domain rcache upper limit is set according
> to that DMA max optimised size. As such, we don't operate on a live IOMMU
> domain.
>
> Note that the DMA mapping API frontend is not strictly required, but saves
> the LLDD calling IOMMU APIs directly, that being not preferred.
>
> Some figures for storage scenario:
> v5.13-rc1 baseline: 1200K IOPS
> With series: 1800K IOPS
>
> All above are for IOMMU strict mode. Non-strict mode gives ~1800K IOPS in
> all scenarios.
>
> Patch breakdown:
> 1-11: Add support for setting DMA max optimised size via sysfs
> 12-15: Add support for setting DMA max optimised size from LLDD
>
> [0] https://lore.kernel.org/linux-iommu/20210129092120.1482-1-thunder.leizhen@huawei.com/
> [1] https://lore.kernel.org/linux-iommu/1607538189-237944-1-git-send-email-john.garry@huawei.com/
>
> Differences to v1:
> - Many
> - Change method to not operate on a 'live' IOMMU domain:
> - rather, force device driver to be re-probed once
> dma_max_opt_size is set, and reconfig a new IOMMU group then
> - Add iommu sysfs max_dma_opt_size file, and allow updating same as how
> group type is changed
>
> John Garry (15):
> iommu: Reactor iommu_group_store_type()
> iova: Allow rcache range upper limit to be flexible
> iommu: Allow max opt DMA len be set for a group via sysfs
> iommu: Add iommu_group_get_max_opt_dma_size()
> iova: Add iova_domain_len_is_cached()
> iommu: Allow iommu_change_dev_def_domain() realloc default domain for
> same type
> iommu: Add iommu_realloc_dev_group()
> dma-iommu: Add iommu_reconfig_dev_group_dma()
> iova: Add init_iova_domain_ext()
> dma-iommu: Use init_iova_domain_ext() for IOVA domain init
> dma-iommu: Reconfig group domain
> iommu: Add iommu_set_dev_dma_opt_size()
> dma-mapping: Add dma_set_max_opt_size()
> dma-iommu: Add iommu_dma_set_opt_size()
> scsi: hisi_sas: Set max optimal DMA size for v3 hw
>
> drivers/iommu/dma-iommu.c | 51 +++++-
> drivers/iommu/iommu.c | 231 +++++++++++++++++++------
> drivers/iommu/iova.c | 61 +++++--
> drivers/scsi/hisi_sas/hisi_sas_v3_hw.c | 5 +
> include/linux/dma-iommu.h | 4 +
> include/linux/dma-map-ops.h | 1 +
> include/linux/dma-mapping.h | 8 +
> include/linux/iommu.h | 19 ++
> include/linux/iova.h | 21 ++-
> kernel/dma/mapping.c | 11 ++
> 10 files changed, 344 insertions(+), 68 deletions(-)
>
prev parent reply other threads:[~2021-05-20 8:29 UTC|newest]
Thread overview: 19+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-05-10 14:17 [PATCH v2 00/15] dma mapping/iommu: Allow IOMMU IOVA rcache range to be configured John Garry
2021-05-10 14:17 ` [PATCH v2 01/15] iommu: Reactor iommu_group_store_type() John Garry
2021-05-10 14:17 ` [PATCH v2 02/15] iova: Allow rcache range upper limit to be flexible John Garry
2021-05-10 14:17 ` [PATCH v2 03/15] iommu: Allow max opt DMA len be set for a group via sysfs John Garry
2021-05-10 14:17 ` [PATCH v2 04/15] iommu: Add iommu_group_get_max_opt_dma_size() John Garry
2021-05-10 14:17 ` [PATCH v2 05/15] iova: Add iova_domain_len_is_cached() John Garry
2021-05-10 14:17 ` [PATCH v2 06/15] iommu: Allow iommu_change_dev_def_domain() realloc default domain for same type John Garry
2021-05-10 14:17 ` [PATCH v2 07/15] iommu: Add iommu_realloc_dev_group() John Garry
2021-05-10 14:17 ` [PATCH v2 08/15] dma-iommu: Add iommu_reconfig_dev_group_dma() John Garry
2021-05-10 14:17 ` [PATCH v2 09/15] iova: Add init_iova_domain_ext() John Garry
2021-05-10 17:50 ` kernel test robot
2021-05-10 17:50 ` [RFC PATCH] iova: __init_iova_domain can be static kernel test robot
2021-05-10 14:17 ` [PATCH v2 10/15] dma-iommu: Use init_iova_domain_ext() for IOVA domain init John Garry
2021-05-10 14:17 ` [PATCH v2 11/15] dma-iommu: Reconfig group domain John Garry
2021-05-10 14:17 ` [PATCH v2 12/15] iommu: Add iommu_set_dev_dma_opt_size() John Garry
2021-05-10 14:17 ` [PATCH v2 13/15] dma-mapping: Add dma_set_max_opt_size() John Garry
2021-05-10 14:17 ` [PATCH v2 14/15] dma-iommu: Add iommu_dma_set_opt_size() John Garry
2021-05-10 14:17 ` [PATCH v2 15/15] scsi: hisi_sas: Set max optimal DMA size for v3 hw John Garry
2021-05-20 8:28 ` John Garry [this message]
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=42aff9e5-096c-f47d-60f5-08cd39152c1c@huawei.com \
--to=john.garry@huawei.com \
--cc=baolu.lu@linux.intel.com \
--cc=hch@lst.de \
--cc=iommu@lists.linux-foundation.org \
--cc=joro@8bytes.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-scsi@vger.kernel.org \
--cc=linuxarm@huawei.com \
--cc=m.szyprowski@samsung.com \
--cc=robin.murphy@arm.com \
--cc=will@kernel.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).