Re: [PATCHv2 2/3] iommu/io-pgtable: Optimize partial walk flush for large scatter-gather list

Re: [PATCHv2 2/3] iommu/io-pgtable: Optimize partial walk flush for large scatter-gather list

[Doug Anderson]

Hi,

On Thu, Jun 17, 2021 at 7:51 PM Sai Prakash Ranjan
<saiprakash.ranjan@codeaurora.org> wrote:
>
> Currently for iommu_unmap() of large scatter-gather list with page size
> elements, the majority of time is spent in flushing of partial walks in
> __arm_lpae_unmap() which is a VA based TLB invalidation invalidating
> page-by-page on iommus like arm-smmu-v2 (TLBIVA) which do not support
> range based invalidations like on arm-smmu-v3.2.
>
> For example: to unmap a 32MB scatter-gather list with page size elements
> (8192 entries), there are 16->2MB buffer unmaps based on the pgsize (2MB
> for 4K granule) and each of 2MB will further result in 512 TLBIVAs (2MB/4K)
> resulting in a total of 8192 TLBIVAs (512*16) for 16->2MB causing a huge
> overhead.
>
> So instead use tlb_flush_all() callback (TLBIALL/TLBIASID) to invalidate
> the entire context for partial walk flush on select few platforms where
> cost of over-invalidation is less than unmap latency

It would probably be worth punching this description up a little bit.
Elsewhere you said in more detail why this over-invalidation is less
of a big deal for the Qualcomm SMMU. It's probably worth saying
something like that here, too. Like this bit paraphrased from your
other email:

On qcom impl, we have several performance improvements for TLB cache
invalidations in HW like wait-for-safe (for realtime clients such as
camera and display) and few others to allow for cache lookups/updates
when TLBI is in progress for the same context bank.


> using the newly
> introduced quirk IO_PGTABLE_QUIRK_TLB_INV_ALL. We also do this for
> non-strict mode given its all about over-invalidation saving time on
> individual unmaps and non-deterministic generally.

As per usual I'm mostly clueless, but I don't quite understand why you
want this new behavior for non-strict mode. To me it almost seems like
the opposite? Specifically, non-strict mode is already outside the
critical path today and so there's no need to optimize it. I'm
probably not explaining myself clearly, but I guess i'm thinking:

a) today for strict, unmap is in the critical path and it's important
to get it out of there. Getting it out of the critical path is so
important that we're willing to over-invalidate to speed up the
critical path.

b) today for non-strict, unmap is not in the critical path.

So I would almost expect your patch to _disable_ your new feature for
non-strict mappings, not auto-enable your new feature for non-strict
mappings.

If I'm babbling, feel free to ignore. ;-) Looking back, I guess Robin
was the one that suggested the behavior you're implementing, so it's
more likely he's right than I am. ;-)


-Doug

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Re: [PATCHv2 2/3] iommu/io-pgtable: Optimize partial walk flush for large scatter-gather list

[Robin Murphy]

On 2021-06-21 06:47, Sai Prakash Ranjan wrote:
> Hi,
> 
> On 2021-06-19 03:39, Doug Anderson wrote:
>> Hi,
>>
>> On Thu, Jun 17, 2021 at 7:51 PM Sai Prakash Ranjan
>> <saiprakash.ranjan@codeaurora.org> wrote:
>>>
>>> Currently for iommu_unmap() of large scatter-gather list with page size
>>> elements, the majority of time is spent in flushing of partial walks in
>>> __arm_lpae_unmap() which is a VA based TLB invalidation invalidating
>>> page-by-page on iommus like arm-smmu-v2 (TLBIVA) which do not support
>>> range based invalidations like on arm-smmu-v3.2.
>>>
>>> For example: to unmap a 32MB scatter-gather list with page size elements
>>> (8192 entries), there are 16->2MB buffer unmaps based on the pgsize (2MB
>>> for 4K granule) and each of 2MB will further result in 512 TLBIVAs 
>>> (2MB/4K)
>>> resulting in a total of 8192 TLBIVAs (512*16) for 16->2MB causing a huge
>>> overhead.
>>>
>>> So instead use tlb_flush_all() callback (TLBIALL/TLBIASID) to invalidate
>>> the entire context for partial walk flush on select few platforms where
>>> cost of over-invalidation is less than unmap latency
>>
>> It would probably be worth punching this description up a little bit.
>> Elsewhere you said in more detail why this over-invalidation is less
>> of a big deal for the Qualcomm SMMU. It's probably worth saying
>> something like that here, too. Like this bit paraphrased from your
>> other email:
>>
>> On qcom impl, we have several performance improvements for TLB cache
>> invalidations in HW like wait-for-safe (for realtime clients such as
>> camera and display) and few others to allow for cache lookups/updates
>> when TLBI is in progress for the same context bank.
>>
> 
> Sure will add this info as well in the next version.
> 
>>
>>> using the newly
>>> introduced quirk IO_PGTABLE_QUIRK_TLB_INV_ALL. We also do this for
>>> non-strict mode given its all about over-invalidation saving time on
>>> individual unmaps and non-deterministic generally.
>>
>> As per usual I'm mostly clueless, but I don't quite understand why you
>> want this new behavior for non-strict mode. To me it almost seems like
>> the opposite? Specifically, non-strict mode is already outside the
>> critical path today and so there's no need to optimize it. I'm
>> probably not explaining myself clearly, but I guess i'm thinking:
>>
>> a) today for strict, unmap is in the critical path and it's important
>> to get it out of there. Getting it out of the critical path is so
>> important that we're willing to over-invalidate to speed up the
>> critical path.
>>
>> b) today for non-strict, unmap is not in the critical path.
>>
>> So I would almost expect your patch to _disable_ your new feature for
>> non-strict mappings, not auto-enable your new feature for non-strict
>> mappings.
>>
>> If I'm babbling, feel free to ignore. ;-) Looking back, I guess Robin
>> was the one that suggested the behavior you're implementing, so it's
>> more likely he's right than I am. ;-)
>>
> 
> Thanks for taking a look. Non-strict mode is only for leaf entries and
> dma domains and this optimization is for non-leaf entries and is applicable
> for both, see __arm_lpae_unmap(). In other words, if you have 
> iommu.strict=0
> (non-strict mode) and try unmapping a large sg buffer as the problem 
> described
> in the commit text, you would still go via this path in unmap and see the
> delay without this patch. So what Robin suggested is that, let's do this
> unconditionally for all users with non-strict mode as opposed to only
> restricting it to implementation specific in case of strict mode.

Right, unmapping tables works out as a bit of a compromise for 
non-strict mode - we don't use a freelist to defer the freeing of 
pagetable pages, so we rely on making non-leaf invalidations 
synchronously to knock out walk caches which may be pointing to the page 
beforte we free it. We might actually be able to get away without that 
for non-strict unmaps, since partial walks pointing at freed memory 
probably aren't too much more hazardous than the equivalent leaf TLB 
entries while the IOVA region is held in the flush queue, but it 
certainly does matter for maps when we're knocking out a (presumably 
empty) table entry to put down a new block whose IOVA will be 
immediately live.

Robin.

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Re: [PATCHv2 1/3] iommu/io-pgtable: Add a quirk to use tlb_flush_all() for partial walk flush

[Robin Murphy]

On 2021-06-18 03:51, Sai Prakash Ranjan wrote:
> Add a quirk IO_PGTABLE_QUIRK_TLB_INV_ALL to invalidate entire context
> with tlb_flush_all() callback in partial walk flush to improve unmap
> performance on select few platforms where the cost of over-invalidation
> is less than the unmap latency.

I still think this doesn't belong anywhere near io-pgtable at all. It's 
a driver-internal decision how exactly it implements a non-leaf 
invalidation, and that may be more complex than a predetermined boolean 
decision. For example, I've just realised for SMMUv3 we can't invalidate 
multiple levels of table at once with a range command, since if we 
assume the whole thing is mapped at worst-case page granularity we may 
fail to invalidate any parts which are mapped as intermediate-level 
blocks. If invalidating a 1GB region (with 4KB granule) means having to 
fall back to 256K non-range commands, we may not want to invalidate by 
VA then, even though doing so for a 2MB region is still optimal.

It's also quite feasible that drivers might want to do this for leaf 
invalidations too - if you don't like issuing 512 commands to invalidate 
2MB, do you like issuing 511 commands to invalidate 2044KB? - and at 
that point the logic really has to be in the driver anyway.

Robin.

> Signed-off-by: Sai Prakash Ranjan <saiprakash.ranjan@codeaurora.org>
> ---
>   drivers/iommu/io-pgtable-arm.c | 3 ++-
>   include/linux/io-pgtable.h     | 5 +++++
>   2 files changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/iommu/io-pgtable-arm.c b/drivers/iommu/io-pgtable-arm.c
> index 87def58e79b5..5d362f2214bd 100644
> --- a/drivers/iommu/io-pgtable-arm.c
> +++ b/drivers/iommu/io-pgtable-arm.c
> @@ -768,7 +768,8 @@ arm_64_lpae_alloc_pgtable_s1(struct io_pgtable_cfg *cfg, void *cookie)
>   	if (cfg->quirks & ~(IO_PGTABLE_QUIRK_ARM_NS |
>   			    IO_PGTABLE_QUIRK_NON_STRICT |
>   			    IO_PGTABLE_QUIRK_ARM_TTBR1 |
> -			    IO_PGTABLE_QUIRK_ARM_OUTER_WBWA))
> +			    IO_PGTABLE_QUIRK_ARM_OUTER_WBWA |
> +			    IO_PGTABLE_QUIRK_TLB_INV_ALL))
>   		return NULL;
>   
>   	data = arm_lpae_alloc_pgtable(cfg);
> diff --git a/include/linux/io-pgtable.h b/include/linux/io-pgtable.h
> index 4d40dfa75b55..45441592a0e6 100644
> --- a/include/linux/io-pgtable.h
> +++ b/include/linux/io-pgtable.h
> @@ -82,6 +82,10 @@ struct io_pgtable_cfg {
>   	 *
>   	 * IO_PGTABLE_QUIRK_ARM_OUTER_WBWA: Override the outer-cacheability
>   	 *	attributes set in the TCR for a non-coherent page-table walker.
> +	 *
> +	 * IO_PGTABLE_QUIRK_TLB_INV_ALL: Use TLBIALL/TLBIASID to invalidate
> +	 *	entire context for partial walk flush to increase unmap
> +	 *	performance on select few platforms.
>   	 */
>   	#define IO_PGTABLE_QUIRK_ARM_NS		BIT(0)
>   	#define IO_PGTABLE_QUIRK_NO_PERMS	BIT(1)
> @@ -89,6 +93,7 @@ struct io_pgtable_cfg {
>   	#define IO_PGTABLE_QUIRK_NON_STRICT	BIT(4)
>   	#define IO_PGTABLE_QUIRK_ARM_TTBR1	BIT(5)
>   	#define IO_PGTABLE_QUIRK_ARM_OUTER_WBWA	BIT(6)
> +	#define IO_PGTABLE_QUIRK_TLB_INV_ALL	BIT(7)
>   	unsigned long			quirks;
>   	unsigned long			pgsize_bitmap;
>   	unsigned int			ias;
> 

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Re: [PATCHv2 1/3] iommu/io-pgtable: Add a quirk to use tlb_flush_all() for partial walk flush

[Robin Murphy]

On 2021-06-22 08:11, Sai Prakash Ranjan wrote:
> Hi Robin,
> 
> On 2021-06-21 21:15, Robin Murphy wrote:
>> On 2021-06-18 03:51, Sai Prakash Ranjan wrote:
>>> Add a quirk IO_PGTABLE_QUIRK_TLB_INV_ALL to invalidate entire context
>>> with tlb_flush_all() callback in partial walk flush to improve unmap
>>> performance on select few platforms where the cost of over-invalidation
>>> is less than the unmap latency.
>>
>> I still think this doesn't belong anywhere near io-pgtable at all.
>> It's a driver-internal decision how exactly it implements a non-leaf
>> invalidation, and that may be more complex than a predetermined
>> boolean decision. For example, I've just realised for SMMUv3 we can't
>> invalidate multiple levels of table at once with a range command,
>> since if we assume the whole thing is mapped at worst-case page
>> granularity we may fail to invalidate any parts which are mapped as
>> intermediate-level blocks. If invalidating a 1GB region (with 4KB
>> granule) means having to fall back to 256K non-range commands, we may
>> not want to invalidate by VA then, even though doing so for a 2MB
>> region is still optimal.
>>
>> It's also quite feasible that drivers might want to do this for leaf
>> invalidations too - if you don't like issuing 512 commands to
>> invalidate 2MB, do you like issuing 511 commands to invalidate 2044KB?
>> - and at that point the logic really has to be in the driver anyway.
>>
> 
> Ok I will move this to tlb_flush_walk() functions in the drivers. In the 
> previous
> v1 thread, you suggested to make the choice in iommu_get_dma_strict() test,
> I assume you meant the test in iommu_dma_init_domain() with a flag or 
> was it
> the leaf driver(ex:arm-smmu.c) test of iommu_get_dma_strict() in 
> init_domain?

Yes, I meant literally inside the same condition where we currently set 
"pgtbl_cfg.quirks |= IO_PGTABLE_QUIRK_NON_STRICT;" in 
arm_smmu_init_domain_context().

> I am still a bit confused on where this flag would be? Should this be a 
> part
> of struct iommu_domain?

Well, if you were to rewrite the config with an alternative set of 
flush_ops at that point it would be implicit. For a flag, probably 
either in arm_smmu_domain or arm_smmu_impl. Maybe a flag would be less 
useful than generalising straight to a "maximum number of by-VA 
invalidations it's worth sending individually" threshold value? It's 
clear to me what overall shape and separation of responsibility is most 
logical, but beyond that I don't have a particularly strong opinion on 
the exact implementation; I've just been chucking ideas around :)

Cheers,
Robin.

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Re: [PATCHv2 1/3] iommu/io-pgtable: Add a quirk to use tlb_flush_all() for partial walk flush

[Robin Murphy]

On 2021-06-22 15:27, Sai Prakash Ranjan wrote:
> Hi Robin,
> 
> On 2021-06-22 17:41, Robin Murphy wrote:
>> On 2021-06-22 08:11, Sai Prakash Ranjan wrote:
>>> Hi Robin,
>>>
>>> On 2021-06-21 21:15, Robin Murphy wrote:
>>>> On 2021-06-18 03:51, Sai Prakash Ranjan wrote:
>>>>> Add a quirk IO_PGTABLE_QUIRK_TLB_INV_ALL to invalidate entire context
>>>>> with tlb_flush_all() callback in partial walk flush to improve unmap
>>>>> performance on select few platforms where the cost of 
>>>>> over-invalidation
>>>>> is less than the unmap latency.
>>>>
>>>> I still think this doesn't belong anywhere near io-pgtable at all.
>>>> It's a driver-internal decision how exactly it implements a non-leaf
>>>> invalidation, and that may be more complex than a predetermined
>>>> boolean decision. For example, I've just realised for SMMUv3 we can't
>>>> invalidate multiple levels of table at once with a range command,
>>>> since if we assume the whole thing is mapped at worst-case page
>>>> granularity we may fail to invalidate any parts which are mapped as
>>>> intermediate-level blocks. If invalidating a 1GB region (with 4KB
>>>> granule) means having to fall back to 256K non-range commands, we may
>>>> not want to invalidate by VA then, even though doing so for a 2MB
>>>> region is still optimal.
>>>>
>>>> It's also quite feasible that drivers might want to do this for leaf
>>>> invalidations too - if you don't like issuing 512 commands to
>>>> invalidate 2MB, do you like issuing 511 commands to invalidate 2044KB?
>>>> - and at that point the logic really has to be in the driver anyway.
>>>>
>>>
>>> Ok I will move this to tlb_flush_walk() functions in the drivers. In 
>>> the previous
>>> v1 thread, you suggested to make the choice in iommu_get_dma_strict() 
>>> test,
>>> I assume you meant the test in iommu_dma_init_domain() with a flag or 
>>> was it
>>> the leaf driver(ex:arm-smmu.c) test of iommu_get_dma_strict() in 
>>> init_domain?
>>
>> Yes, I meant literally inside the same condition where we currently
>> set "pgtbl_cfg.quirks |= IO_PGTABLE_QUIRK_NON_STRICT;" in
>> arm_smmu_init_domain_context().
>>
> 
> Ok got it, thanks.
> 
>>> I am still a bit confused on where this flag would be? Should this be 
>>> a part
>>> of struct iommu_domain?
>>
>> Well, if you were to rewrite the config with an alternative set of
>> flush_ops at that point it would be implicit. For a flag, probably
>> either in arm_smmu_domain or arm_smmu_impl. Maybe a flag would be less
>> useful than generalising straight to a "maximum number of by-VA
>> invalidations it's worth sending individually" threshold value?
> 
> But then we would still need some flag to make this implementation
> specific (qcom specific for now) and this threshold would just be
> another condition although it would have been useful if this was
> generic enough.

Well, for that approach I assume we could do something like special-case 
0, or if it's a mutable per-domain value maybe just initialise it to 
SIZE_MAX or whatever such that it would never be reached in practice. 
Whichever way, it was meant to be implied that anything at the domain 
level would still be subject to final adjustment by the init_context hook.

Robin.

>> It's clear to me what overall shape and separation of responsibility is
>> most logical, but beyond that I don't have a particularly strong
>> opinion on the exact implementation; I've just been chucking ideas
>> around :)
>>
> 
> Your ideas are very informative and useful :)
> 
> Thanks,
> Sai
> 

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel