linux-arm-kernel.lists.infradead.org archive mirror
 help / color / mirror / Atom feed
From: Jason Gunthorpe <jgg@nvidia.com>
To: Michael Shavit <mshavit@google.com>
Cc: iommu@lists.linux.dev, linux-arm-kernel@lists.infradead.org,
	linux-kernel@vger.kernel.org, nicolinc@nvidia.com,
	tina.zhang@intel.com, jean-philippe@linaro.org, will@kernel.org,
	robin.murphy@arm.com, Dawei Li <set_pte_at@outlook.com>,
	Joerg Roedel <joro@8bytes.org>,
	"Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>,
	Lu Baolu <baolu.lu@linux.intel.com>,
	Mark Brown <broonie@kernel.org>
Subject: Re: [RFC PATCH v2 3/9] iommu/arm-smmu-v3: Issue invalidations commands to multiple SMMUs
Date: Tue, 22 Aug 2023 10:14:09 -0300	[thread overview]
Message-ID: <ZOS0ocsjy34N5s4l@nvidia.com> (raw)
In-Reply-To: <20230822185632.RFC.v2.3.I0f149f177e5478e28dc3223c2d10729d8f28d53a@changeid>

On Tue, Aug 22, 2023 at 06:56:59PM +0800, Michael Shavit wrote:
> Assume that devices in the smmu_domain->domain list that belong to the
> same SMMU are adjacent to each other in the list.
> Batch TLB/ATC invalidation commands for an smmu_domain by the SMMU
> devices that the domain is installed to.
> 
> Signed-off-by: Michael Shavit <mshavit@google.com>
> ---
> 
> Changes in v2:
> - Moved the ARM_SMMU_FEAT_BTM changes into a new prepatory commit
> 
>  .../iommu/arm/arm-smmu-v3/arm-smmu-v3-sva.c   |   6 +-
>  drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c   | 134 +++++++++++++-----
>  drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h   |   2 +-
>  3 files changed, 104 insertions(+), 38 deletions(-)
> 
> diff --git a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-sva.c b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-sva.c
> index 53f65a89a55f9..fe88a7880ad57 100644
> --- a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-sva.c
> +++ b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3-sva.c
> @@ -112,7 +112,7 @@ arm_smmu_share_asid(struct mm_struct *mm, u16 asid)
>  	arm_smmu_write_ctx_desc_devices(smmu_domain, 0, cd);
>  
>  	/* Invalidate TLB entries previously associated with that context */
> -	arm_smmu_tlb_inv_asid(smmu, asid);
> +	arm_smmu_tlb_inv_asid(smmu_domain, asid);
>  
>  	xa_erase(&arm_smmu_asid_xa, asid);
>  	return NULL;
> @@ -252,7 +252,7 @@ static void arm_smmu_mm_release(struct mmu_notifier *mn, struct mm_struct *mm)
>  	 */
>  	arm_smmu_write_ctx_desc_devices(smmu_domain, mm->pasid, &quiet_cd);
>  
> -	arm_smmu_tlb_inv_asid(smmu_domain->smmu, smmu_mn->cd->asid);
> +	arm_smmu_tlb_inv_asid(smmu_domain, smmu_mn->cd->asid);
>  	arm_smmu_atc_inv_domain(smmu_domain, mm->pasid, 0, 0);
>  
>  	smmu_mn->cleared = true;
> @@ -340,7 +340,7 @@ static void arm_smmu_mmu_notifier_put(struct arm_smmu_mmu_notifier *smmu_mn)
>  	 * new TLB entry can have been formed.
>  	 */
>  	if (!smmu_mn->cleared) {
> -		arm_smmu_tlb_inv_asid(smmu_domain->smmu, cd->asid);
> +		arm_smmu_tlb_inv_asid(smmu_domain, cd->asid);
>  		arm_smmu_atc_inv_domain(smmu_domain, mm->pasid, 0, 0);
>  	}
>  
> diff --git a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c
> index db4df9d6aef10..1d072fd38a2d6 100644
> --- a/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c
> +++ b/drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c
> @@ -960,15 +960,28 @@ static int arm_smmu_page_response(struct device *dev,
>  }
>  
>  /* Context descriptor manipulation functions */
> -void arm_smmu_tlb_inv_asid(struct arm_smmu_device *smmu, u16 asid)
> +void arm_smmu_tlb_inv_asid(struct arm_smmu_domain *smmu_domain, u16 asid)
>  {
> +	struct arm_smmu_device *smmu = NULL;
> +	struct arm_smmu_master *master;
>  	struct arm_smmu_cmdq_ent cmd = {
> -		.opcode	= smmu->features & ARM_SMMU_FEAT_E2H ?
> -			CMDQ_OP_TLBI_EL2_ASID : CMDQ_OP_TLBI_NH_ASID,
>  		.tlbi.asid = asid,
>  	};
> +	unsigned long flags;
>  
> -	arm_smmu_cmdq_issue_cmd_with_sync(smmu, &cmd);
> +	spin_lock_irqsave(&smmu_domain->devices_lock, flags);
> +	list_for_each_entry(master, &smmu_domain->devices,
> +			    domain_head) {
> +		if (!smmu)
> +			smmu = master->smmu;
> +		if (smmu != master->smmu ||
> +		    list_is_last(&master->domain_head, &smmu_domain->devices)) {

Finding the end of the list seems too complicated, just:

	struct arm_smmu_device *invalidated_smmu = NULL;
	list_for_each_entry(master, &smmu_domain->devices,
			    domain_head) {
             if (master->smmu == invalidated_smmu)
	          continue;
             cmd.opcode = smmu->features & ARM_SMMU_FEAT_E2H ?
			CMDQ_OP_TLBI_EL2_ASID : CMDQ_OP_TLBI_NH_ASID,
             arm_smmu_cmdq_issue_cmd_with_sync(master->smmu, &cmd);
	     invalidated_smmu = master->smmu;
        }

> @@ -1839,28 +1851,56 @@ int arm_smmu_atc_inv_domain(struct arm_smmu_domain *smmu_domain, int ssid,
>  	arm_smmu_atc_inv_to_cmd(ssid, iova, size, &cmd);
>  
>  	cmds.num = 0;
> -
>  	spin_lock_irqsave(&smmu_domain->devices_lock, flags);
>  	list_for_each_entry(master, &smmu_domain->devices, domain_head) {
>  		if (!master->ats_enabled)
>  			continue;
> +		if (!smmu)
> +			smmu = master->smmu;
> +		if (smmu != master->smmu ||
> +		    list_is_last(&master->domain_head, &smmu_domain->devices)) {
> +			ret = arm_smmu_cmdq_batch_submit(smmu, &cmds);
> +			if (ret)
> +				break;
> +			cmds.num = 0;
> +		}
>  
>  		for (i = 0; i < master->num_streams; i++) {
>  			cmd.atc.sid = master->streams[i].id;
> -			arm_smmu_cmdq_batch_add(smmu_domain->smmu, &cmds, &cmd);
> +			arm_smmu_cmdq_batch_add(smmu, &cmds, &cmd);
>  		}
>  	}

Doesn't the IOTLB invalidate have to come before the ATC invalidate?

So again, use the pattern as above?

>  	spin_unlock_irqrestore(&smmu_domain->devices_lock, flags);
>  
> -	return arm_smmu_cmdq_batch_submit(smmu_domain->smmu, &cmds);
> +	return ret;
> +}
> +
> +static void arm_smmu_tlb_inv_vmid(struct arm_smmu_domain *smmu_domain)
> +{
> +	struct arm_smmu_device *smmu = NULL;
> +	struct arm_smmu_master *master;
> +	struct arm_smmu_cmdq_ent cmd = {
> +		.opcode = CMDQ_OP_TLBI_S12_VMALL,
> +		.tlbi.vmid = smmu_domain->s2_cfg.vmid,
> +	};
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&smmu_domain->devices_lock, flags);
> +	list_for_each_entry(master, &smmu_domain->devices,
> +			    domain_head) {
> +		if (!smmu)
> +			smmu = master->smmu;
> +		if (smmu != master->smmu ||
> +		    list_is_last(&master->domain_head, &smmu_domain->devices))
> +			arm_smmu_cmdq_issue_cmd_with_sync(smmu, &cmd);
> +	}
> +	spin_unlock_irqrestore(&smmu_domain->devices_lock, flags);
>  }

I count three of these, so a macro helper is probably a good
idea. Something approx like:

static struct arm_smmu_master *smmu_next_entry(struct arm_smmu_master *pos,
					       struct arm_smmu_domain *domain)
{
	struct arm_smmu *smmu = pos->smmu;

	do {
		pos = list_next_entry(pos, domain_head);
	} while (!list_entry_is_head(pos, domain->devices, domain_head) &&
		 pos->smmu == smmu);
	return pos;
}

#define for_each_smmu(pos, domain, smmu)                                       \
	for (pos = list_first_entry((domain)->devices, struct arm_smmu_master, \
				    domain_head),                              \
	    smmu = (pos)->smmu;                                                \
	     !list_entry_is_head(pos, (domain)->devices, domain_head);         \
	     pos = smmu_next_entry(pos, domain), smmu = (pos)->smmu)

> @@ -1949,21 +1987,36 @@ static void arm_smmu_tlb_inv_range_domain(unsigned long iova, size_t size,
>  					  size_t granule, bool leaf,
>  					  struct arm_smmu_domain *smmu_domain)
>  {
> +	struct arm_smmu_device *smmu = NULL;
> +	struct arm_smmu_master *master;
>  	struct arm_smmu_cmdq_ent cmd = {
>  		.tlbi = {
>  			.leaf	= leaf,
>  		},
>  	};
> +	unsigned long flags;
>  
> -	if (smmu_domain->stage == ARM_SMMU_DOMAIN_S1) {
> -		cmd.opcode	= smmu_domain->smmu->features & ARM_SMMU_FEAT_E2H ?
> -				  CMDQ_OP_TLBI_EL2_VA : CMDQ_OP_TLBI_NH_VA;
> -		cmd.tlbi.asid	= smmu_domain->cd.asid;
> -	} else {
> -		cmd.opcode	= CMDQ_OP_TLBI_S2_IPA;
> -		cmd.tlbi.vmid	= smmu_domain->s2_cfg.vmid;
> +	spin_lock_irqsave(&smmu_domain->devices_lock, flags);
> +	list_for_each_entry(master, &smmu_domain->devices, domain_head) {
> +		if (!smmu)
> +			smmu = master->smmu;
> +		if (smmu != master->smmu ||
> +		    list_is_last(&master->domain_head, &smmu_domain->devices)) {
> +			if (smmu_domain->stage == ARM_SMMU_DOMAIN_S1) {
> +				cmd.opcode = smmu->features &
> +							     ARM_SMMU_FEAT_E2H ?
> +						     CMDQ_OP_TLBI_EL2_VA :
> +						     CMDQ_OP_TLBI_NH_VA;
> +				cmd.tlbi.asid = smmu_domain->cd.asid;
> +			} else {
> +				cmd.opcode = CMDQ_OP_TLBI_S2_IPA;
> +				cmd.tlbi.vmid = smmu_domain->s2_cfg.vmid;
> +			}

These calculations based on smmu domain shouldn't be in the loop, the
smmu_domain doesn't change.

> -	__arm_smmu_tlb_inv_range(&cmd, iova, size, granule, smmu_domain);
> +	spin_lock_irqsave(&smmu_domain->devices_lock, flags);
> +	list_for_each_entry(master, &smmu_domain->devices, domain_head) {
> +		if (!smmu)
> +			smmu = master->smmu;
> +		if (smmu != master->smmu ||
> +		    list_is_last(&master->domain_head, &smmu_domain->devices)) {
> +			if (skip_btm_capable_devices &&
> +			    smmu->features & ARM_SMMU_FEAT_BTM)
> +				continue;
> +			cmd.opcode = smmu->features & ARM_SMMU_FEAT_E2H ?
> +					     CMDQ_OP_TLBI_EL2_VA :
> +					     CMDQ_OP_TLBI_NH_VA;

There are 3 places doing this if, maybe it should be in a wrapper of
__arm_smmu_tlb_inv_range ?

Jason

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

  reply	other threads:[~2023-08-22 13:16 UTC|newest]

Thread overview: 15+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-08-22 10:56 [RFC PATCH v2 0/9] Install domain onto multiple smmus Michael Shavit
2023-08-22 10:56 ` [RFC PATCH v2 1/9] iommu/arm-smmu-v3: group attached devices by smmu Michael Shavit
2023-08-22 12:49   ` Jason Gunthorpe
2023-08-22 10:56 ` [RFC PATCH v2 2/9] iommu/arm-smmu-v3-sva: Move SVA optimization into arm_smmu_tlb_inv_range_asid Michael Shavit
2023-08-22 10:56 ` [RFC PATCH v2 3/9] iommu/arm-smmu-v3: Issue invalidations commands to multiple SMMUs Michael Shavit
2023-08-22 13:14   ` Jason Gunthorpe [this message]
2023-08-22 10:57 ` [RFC PATCH v2 4/9] iommu/arm-smmu-v3-sva: Allocate new ASID from installed_smmus Michael Shavit
2023-08-22 13:19   ` Jason Gunthorpe
2023-08-23  7:26     ` Michael Shavit
2023-08-22 10:57 ` [RFC PATCH v2 5/9] iommu/arm-smmu-v3: Alloc vmid from global pool Michael Shavit
2023-08-22 10:57 ` [RFC PATCH v2 6/9] iommu/arm-smmu-v3: check smmu compatibility on attach Michael Shavit
2023-08-22 10:57 ` [RFC PATCH v2 7/9] iommu/arm-smmu-v3: Add arm_smmu_device as a parameter to domain_finalise Michael Shavit
2023-08-22 10:57 ` [RFC PATCH v2 8/9] iommu/arm-smmu-v3: check for domain initialization using pgtbl_ops Michael Shavit
2023-08-22 10:57 ` [RFC PATCH v2 9/9] iommu/arm-smmu-v3: allow multi-SMMU domain installs Michael Shavit
2023-08-23  2:42 ` [RFC PATCH v2 0/9] Install domain onto multiple smmus Baolu Lu

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=ZOS0ocsjy34N5s4l@nvidia.com \
    --to=jgg@nvidia.com \
    --cc=baolu.lu@linux.intel.com \
    --cc=broonie@kernel.org \
    --cc=iommu@lists.linux.dev \
    --cc=jean-philippe@linaro.org \
    --cc=joro@8bytes.org \
    --cc=kirill.shutemov@linux.intel.com \
    --cc=linux-arm-kernel@lists.infradead.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mshavit@google.com \
    --cc=nicolinc@nvidia.com \
    --cc=robin.murphy@arm.com \
    --cc=set_pte_at@outlook.com \
    --cc=tina.zhang@intel.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).