LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: John Garry <john.garry@huawei.com>
To: Will Deacon <will@kernel.org>
Cc: <joro@8bytes.org>, <robin.murphy@arm.com>,
	<baolu.lu@linux.intel.com>, <iommu@lists.linux-foundation.org>,
	<linuxarm@huawei.com>, <thierry.reding@gmail.com>,
	<airlied@linux.ie>, <daniel@ffwll.ch>, <jonathanh@nvidia.com>,
	<sakari.ailus@linux.intel.com>, <bingbu.cao@intel.com>,
	<tian.shu.qiu@intel.com>, <mchehab@kernel.org>,
	<gregkh@linuxfoundation.org>, <digetx@gmail.com>,
	<mst@redhat.com>, <jasowang@redhat.com>,
	<linux-kernel@vger.kernel.org>, <chenxiang66@hisilicon.com>
Subject: Re: [PATCH v4 2/6] iova: Allow rcache range upper limit to be flexible
Date: Mon, 2 Aug 2021 16:23:00 +0100	[thread overview]
Message-ID: <27bb22cf-db64-0aa5-215f-2adf06b6455d@huawei.com> (raw)
In-Reply-To: <20210802150153.GC28735@willie-the-truck>

On 02/08/2021 16:01, Will Deacon wrote:
> On Wed, Jul 14, 2021 at 06:36:39PM +0800, John Garry wrote:
>> Some LLDs may request DMA mappings whose IOVA length exceeds that of the
>> current rcache upper limit.
> 
> What's an LLD?
> 

low-level driver

maybe I'll stick with simply "drivers"

>> This means that allocations for those IOVAs will never be cached, and
>> always must be allocated and freed from the RB tree per DMA mapping cycle.
>> This has a significant effect on performance, more so since commit
>> 4e89dce72521 ("iommu/iova: Retry from last rb tree node if iova search
>> fails"), as discussed at [0].
>>
>> As a first step towards allowing the rcache range upper limit be
>> configured, hold this value in the IOVA rcache structure, and allocate
>> the rcaches separately.
>>
>> [0] https://lore.kernel.org/linux-iommu/20210129092120.1482-1-thunder.leizhen@huawei.com/
>>
>> Signed-off-by: John Garry <john.garry@huawei.com>
>> ---
>>   drivers/iommu/dma-iommu.c |  2 +-
>>   drivers/iommu/iova.c      | 23 +++++++++++++++++------
>>   include/linux/iova.h      |  4 ++--
>>   3 files changed, 20 insertions(+), 9 deletions(-)
>>
>> diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
>> index 98ba927aee1a..4772278aa5da 100644
>> --- a/drivers/iommu/dma-iommu.c
>> +++ b/drivers/iommu/dma-iommu.c
>> @@ -434,7 +434,7 @@ static dma_addr_t iommu_dma_alloc_iova(struct iommu_domain *domain,
>>   	 * rounding up anything cacheable to make sure that can't happen. The
>>   	 * order of the unadjusted size will still match upon freeing.
>>   	 */
>> -	if (iova_len < (1 << (IOVA_RANGE_CACHE_MAX_SIZE - 1)))
>> +	if (iova_len < (1 << (iovad->rcache_max_size - 1)))
>>   		iova_len = roundup_pow_of_two(iova_len);
>>   
>>   	dma_limit = min_not_zero(dma_limit, dev->bus_dma_limit);
>> diff --git a/drivers/iommu/iova.c b/drivers/iommu/iova.c
>> index b6cf5f16123b..07ce73fdd8c1 100644
>> --- a/drivers/iommu/iova.c
>> +++ b/drivers/iommu/iova.c
>> @@ -15,6 +15,8 @@
>>   /* The anchor node sits above the top of the usable address space */
>>   #define IOVA_ANCHOR	~0UL
>>   
>> +#define IOVA_RANGE_CACHE_MAX_SIZE 6    /* log of max cached IOVA range size (in pages) */
> 
> Is that the same as an 'order'? i.e. IOVA_RANGE_CACHE_MAX_ORDER?

Yeah, that may be better. I was just using the same name as before.

> 
>> +
>>   static bool iova_rcache_insert(struct iova_domain *iovad,
>>   			       unsigned long pfn,
>>   			       unsigned long size);
>> @@ -881,7 +883,14 @@ static void init_iova_rcaches(struct iova_domain *iovad)
>>   	unsigned int cpu;
>>   	int i;
>>   
>> -	for (i = 0; i < IOVA_RANGE_CACHE_MAX_SIZE; ++i) {
>> +	iovad->rcache_max_size = IOVA_RANGE_CACHE_MAX_SIZE;
>> +
>> +	iovad->rcaches = kcalloc(iovad->rcache_max_size,
>> +				 sizeof(*iovad->rcaches), GFP_KERNEL);
>> +	if (!iovad->rcaches)
>> +		return;
> 
> Returning quietly here doesn't seem like the right thing to do. At least, I
> don't think the rest of the functions here are checking rcaches against
> NULL.
> 

For sure, but that is what other code which can fail here already does, 
like:

static void init_iova_rcaches(struct iova_domain *iovad)
{
	...

	for (i = 0; i < IOVA_RANGE_CACHE_MAX_SIZE; ++i) {
		...

		rcache->cpu_rcaches = __alloc_percpu(sizeof(*cpu_rcache), 
cache_line_size());
		if (WARN_ON(!rcache->cpu_rcaches))
			continue;
}

and that is not safe either.

This issue was raised a while ago. I don't mind trying to fix it - a 
slightly painful part is that it touches a few subsystems.

Thanks,
John

  reply	other threads:[~2021-08-02 15:23 UTC|newest]

Thread overview: 17+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-07-14 10:36 [PATCH v4 0/6] iommu: Allow IOVA rcache range be configured John Garry
2021-07-14 10:36 ` [PATCH v4 1/6] iommu: Refactor iommu_group_store_type() John Garry
2021-08-02 14:46   ` Will Deacon
2021-07-14 10:36 ` [PATCH v4 2/6] iova: Allow rcache range upper limit to be flexible John Garry
2021-08-02 15:01   ` Will Deacon
2021-08-02 15:23     ` John Garry [this message]
2021-08-02 16:09       ` Robin Murphy
2021-07-14 10:36 ` [PATCH v4 3/6] iommu: Allow iommu_change_dev_def_domain() realloc default domain for same type John Garry
2021-07-14 10:36 ` [PATCH v4 4/6] iommu: Allow max opt DMA len be set for a group via sysfs John Garry
2021-07-14 10:36 ` [PATCH v4 5/6] iova: Add iova_len argument to init_iova_domain() John Garry
2021-08-02 15:06   ` Will Deacon
2021-08-02 16:06     ` John Garry
2021-08-02 16:16       ` Robin Murphy
2021-08-02 16:40         ` John Garry
2021-08-02 17:18           ` John Garry
2021-09-21  8:48         ` John Garry
2021-07-14 10:36 ` [PATCH v4 6/6] dma-iommu: Pass iova len for IOVA domain init John Garry

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=27bb22cf-db64-0aa5-215f-2adf06b6455d@huawei.com \
    --to=john.garry@huawei.com \
    --cc=airlied@linux.ie \
    --cc=baolu.lu@linux.intel.com \
    --cc=bingbu.cao@intel.com \
    --cc=chenxiang66@hisilicon.com \
    --cc=daniel@ffwll.ch \
    --cc=digetx@gmail.com \
    --cc=gregkh@linuxfoundation.org \
    --cc=iommu@lists.linux-foundation.org \
    --cc=jasowang@redhat.com \
    --cc=jonathanh@nvidia.com \
    --cc=joro@8bytes.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linuxarm@huawei.com \
    --cc=mchehab@kernel.org \
    --cc=mst@redhat.com \
    --cc=robin.murphy@arm.com \
    --cc=sakari.ailus@linux.intel.com \
    --cc=thierry.reding@gmail.com \
    --cc=tian.shu.qiu@intel.com \
    --cc=will@kernel.org \
    --subject='Re: [PATCH v4 2/6] iova: Allow rcache range upper limit to be flexible' \
    /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

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).