Re: [PATCH v2 7/7] iommu/vt-d: Consolidate domain_init() to avoid duplication

[Lu Baolu <baolu.lu@linux.intel.com>]

Hi Alex,

On 7/19/19 7:16 AM, Alex Williamson wrote:
> On Wed, 12 Jun 2019 08:28:51 +0800
> Lu Baolu <baolu.lu@linux.intel.com> wrote:
> 
>> The domain_init() and md_domain_init() do almost the same job.
>> Consolidate them to avoid duplication.
>>
>> Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
>> ---
>>   drivers/iommu/intel-iommu.c | 123 +++++++++++-------------------------
>>   1 file changed, 36 insertions(+), 87 deletions(-)
>>
>> diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c
>> index 5215dcd535a1..b8c6cf1d5f90 100644
>> --- a/drivers/iommu/intel-iommu.c
>> +++ b/drivers/iommu/intel-iommu.c
>> @@ -1825,63 +1825,6 @@ static inline int guestwidth_to_adjustwidth(int gaw)
>>   	return agaw;
>>   }
>>   
>> -static int domain_init(struct dmar_domain *domain, struct intel_iommu *iommu,
>> -		       int guest_width)
>> -{
>> -	int adjust_width, agaw;
>> -	unsigned long sagaw;
>> -	int err;
>> -
>> -	init_iova_domain(&domain->iovad, VTD_PAGE_SIZE, IOVA_START_PFN);
>> -
>> -	err = init_iova_flush_queue(&domain->iovad,
>> -				    iommu_flush_iova, iova_entry_free);
>> -	if (err)
>> -		return err;
>> -
>> -	domain_reserve_special_ranges(domain);
>> -
>> -	/* calculate AGAW */
>> -	if (guest_width > cap_mgaw(iommu->cap))
>> -		guest_width = cap_mgaw(iommu->cap);
>> -	domain->gaw = guest_width;
>> -	adjust_width = guestwidth_to_adjustwidth(guest_width);
>> -	agaw = width_to_agaw(adjust_width);
>> -	sagaw = cap_sagaw(iommu->cap);
>> -	if (!test_bit(agaw, &sagaw)) {
>> -		/* hardware doesn't support it, choose a bigger one */
>> -		pr_debug("Hardware doesn't support agaw %d\n", agaw);
>> -		agaw = find_next_bit(&sagaw, 5, agaw);
>> -		if (agaw >= 5)
>> -			return -ENODEV;
>> -	}
>> -	domain->agaw = agaw;
>> -
>> -	if (ecap_coherent(iommu->ecap))
>> -		domain->iommu_coherency = 1;
>> -	else
>> -		domain->iommu_coherency = 0;
>> -
>> -	if (ecap_sc_support(iommu->ecap))
>> -		domain->iommu_snooping = 1;
>> -	else
>> -		domain->iommu_snooping = 0;
>> -
>> -	if (intel_iommu_superpage)
>> -		domain->iommu_superpage = fls(cap_super_page_val(iommu->cap));
>> -	else
>> -		domain->iommu_superpage = 0;
>> -
>> -	domain->nid = iommu->node;
>> -
>> -	/* always allocate the top pgd */
>> -	domain->pgd = (struct dma_pte *)alloc_pgtable_page(domain->nid);
>> -	if (!domain->pgd)
>> -		return -ENOMEM;
>> -	__iommu_flush_cache(iommu, domain->pgd, PAGE_SIZE);
>> -	return 0;
>> -}
>> -
>>   static void domain_exit(struct dmar_domain *domain)
>>   {
>>   	struct page *freelist;
>> @@ -2563,6 +2506,31 @@ static int get_last_alias(struct pci_dev *pdev, u16 alias, void *opaque)
>>   	return 0;
>>   }
>>   
>> +static int domain_init(struct dmar_domain *domain, int guest_width)
>> +{
>> +	int adjust_width;
>> +
>> +	init_iova_domain(&domain->iovad, VTD_PAGE_SIZE, IOVA_START_PFN);
>> +	domain_reserve_special_ranges(domain);
>> +
>> +	/* calculate AGAW */
>> +	domain->gaw = guest_width;
>> +	adjust_width = guestwidth_to_adjustwidth(guest_width);
>> +	domain->agaw = width_to_agaw(adjust_width);
> 
> 
> How do we justify that domain->agaw is nothing like it was previously
> here?  I spent some more time working on the failure to boot that I
> thought was caused by 4ec066c7b147, but there are so many breakages and
> fixes in Joerg's x86/vt-d branch that I think my bisect zero'd in on
> the wrong one.  Instead I cherry-picked every commit from Joerg's tree
> and matched Fixes patches to their original commit, which led me to
> this patch, mainline commit 123b2ffc376e.  The issue I'm seeing is that
> we call domain_context_mapping_one() and we are in this section:
> 
>          struct dma_pte *pgd = domain->pgd;
>          int agaw;
> 
>          context_set_domain_id(context, did);
> 
>          if (translation != CONTEXT_TT_PASS_THROUGH) {
>                  /*
>                   * Skip top levels of page tables for iommu which has
>                   * less agaw than default. Unnecessary for PT mode.
>                   */
>                  for (agaw = domain->agaw; agaw > iommu->agaw; agaw--) {
>                          ret = -ENOMEM;
>                          pgd = phys_to_virt(dma_pte_addr(pgd));
>                          if (!dma_pte_present(pgd))
>                                  goto out_unlock;
>                  }
> 
> Prior to this commit, we had domain->agaw=1 and iommu->agaw=1, so we
> don't enter the loop.  With this commit, we have domain->agaw=3,
> iommu->agaw=1 with pgd->val=0!
> 

iommu->agaw presents the level of page table which is by default
supported by the @iommu. domain->agaw presents the level of page
table which is used by the @domain.

agaw = 1: 3-level page table
agaw = 2: 4-level page table
agaw = 3: 5-level page table

The case here is that @iommu only supports 3-level page table, but the
@domain was set to use 5-level page table. So we must skip level 4 and 5
page tables of the @domain.

This code in the loop looks odd to me. It will always goto to unlock and
leave pgd->val==0. How about below change? (not tested yet!)

diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c
index 412f18aba501..98d6878cd29d 100644
--- a/drivers/iommu/intel-iommu.c
+++ b/drivers/iommu/intel-iommu.c
@@ -2020,11 +2020,15 @@ static int domain_context_mapping_one(struct 
dmar_domain *domain,
                          * Skip top levels of page tables for iommu 
which has
                          * less agaw than default. Unnecessary for PT mode.
                          */
-                       for (agaw = domain->agaw; agaw > iommu->agaw; 
agaw--) {
-                               ret = -ENOMEM;
-                               pgd = phys_to_virt(dma_pte_addr(pgd));
-                               if (!dma_pte_present(pgd))
-                                       goto out_unlock;
+                       while (iommu->agaw < domain->agaw) {
+                               struct dma_pte *pte;
+
+                               pte = domain->pgd;
+                               if (dma_pte_present(pte)) {
+                                       domain->pgd = 
phys_to_virt(dma_pte_addr(pte));
+                                       free_pgtable_page(pte);
+                               }
+                               domain->agaw--;
                         }

                         info = iommu_support_dev_iotlb(domain, iommu, 
bus, devfn);

> I don't really follow how the setting of these fields above is
> equivalent to what they were previously or if they're supposed to be
> updated lazily, but the current behavior is non-functional.  Commit
> 123b2ffc376e can be reverted with only a bit of offset, which brings
> Linus' tree back into working operation for me.  Should we revert or is
> there an obvious fix here?  Thanks,

This commit is not the root cause of this issue as far as I can see, it
only triggers above loop to get entered.

Best regards,
Baolu
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu