Re: [PATCH v8 2/7] iommu: Move global PASID allocation from SVA to core

[Jacob Pan <jacob.jun.pan@linux.intel.com>]

Hi Baolu,

On Tue, 13 Jun 2023 11:06:03 +0800, Baolu Lu <baolu.lu@linux.intel.com>
wrote:

> On 6/10/23 8:13 PM, Baolu Lu wrote:
> > On 6/3/23 2:22 AM, Jacob Pan wrote:  
> >> +ioasid_t iommu_alloc_global_pasid_dev(struct device *dev)
> >> +{
> >> +    int ret;
> >> +    ioasid_t max;
> >> +
> >> +    max = dev->iommu->max_pasids;
> >> +    /*
> >> +     * max_pasids is set up by vendor driver based on number of PASID 
> >> bits
> >> +     * supported but the IDA allocation is inclusive.
> >> +     */
> >> +    ret = ida_alloc_range(&iommu_global_pasid_ida, 
> >> IOMMU_FIRST_GLOBAL_PASID, max - 1, GFP_KERNEL);
> >> +    if (ret < 0)
> >> +        return IOMMU_PASID_INVALID;
> >> +
> >> +    return ret;
> >> +}
> >> +EXPORT_SYMBOL_GPL(iommu_alloc_global_pasid_dev);  
> > 
> > "dev->iommu->max_pasids == 0" indicates no pasid support on the device.
> > The code should return IOMMU_PASID_INVALID explicitly. Perhaps we can
> > make this function like this:
> > 
> > ioasid_t iommu_alloc_global_pasid_dev(struct device *dev)
> > {
> >          int ret;
> > 
> >          if (!dev->iommu->max_pasids)
> >                  return IOMMU_PASID_INVALID;
> > 
> >          /*
> >           * max_pasids is set up by vendor driver based on number of 
> > PASID bits
> >           * supported but the IDA allocation is inclusive.
> >           */
> >          ret = ida_alloc_range(&iommu_global_pasid_ida, 
> > IOMMU_FIRST_GLOBAL_PASID,
> >                                dev->iommu->max_pasids - 1, GFP_KERNEL);
> > 
> >          return ret < 0 ? IOMMU_PASID_INVALID : ret;
> > }
> > EXPORT_SYMBOL_GPL(iommu_alloc_global_pasid_dev);
> > 
> > Other change in this series looks good to me.
> > 
> > I hope I can queue this series including above change as part of my VT-d
> > update for v6.5 to Joerg if no objection.
> > 
> > Let's try to re-enable this key feature of Intel idxd driver in v6.5.  
> 
> This series didn't pass my test.
> 
> The first time when I run "idxd_ktest.sh -c 1 -t 1 -i 100 -m shared", it
> passed. But when I run it again, the idxd hardware operation resulted in
> timed-out issues.
> 
Hmm, not sure what happened. Are you using the out of tree idxd_ktest
kernel module or the dmaengine test sysfs?

e.g.
echo "Config params for DMA test"                   
echo $1 > /sys/module/dmatest/parameters/iterations 
#echo 1 > /sys/module/dmatest/parameters/noverify   
echo "" > /sys/module/dmatest/parameters/channel    
echo 1 > /sys/module/dmatest/parameters/run         
sleep 2                                             
echo 0 > /sys/module/dmatest/parameters/run         
echo "Completed!"                                   

It passed my test many iterations for shared
in-kernel DSA test, will get your tree and test again.


Thanks,

Jacob