Re: [dpdk-dev] 17.11.4 patches review and test

[Yongseok Koh <yskoh@mellanox.com>]

> On Aug 22, 2018, at 5:19 PM, Yongseok Koh <yskoh@mellanox.com> wrote:
> 
> On Tue, Aug 21, 2018 at 12:07:49PM +0200, Alejandro Lucero wrote:
>> Hi Yonngseok,
>> 
>> There is a patchset aimed at 17.11.x:
>> 
>> https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpatches.dpdk.org%2Fcover%2F42741%2F&amp;data=02%7C01%7Cyskoh%40mellanox.com%7Cb1b0e3eff71c499ff3fb08d6088e1ede%7Ca652971c7d2e4d9ba6a4d149256f461b%7C0%7C0%7C636705803846548967&amp;sdata=8f12c1IuUe4mw2EaTZ18vVTuLTjXOD2cSe%2B%2B7f6OFfk%3D&amp;reserved=0
>> 
>> It was not accepted for master because the memory code has changed a lot
>> since 17.11, and I'm working on another patchset for adjusting to the those
>> changes.
>> 
>> I wonder if there is any issue with adding this patchset to stable 17.11.4.
>> Note that this makes unlikely a known limitation with emulated IOMMU inside
>> VMs.
> 
> This patchset seems quite large for stable release and need to be well verified
> before GA. In -rc1 stage, we don't usually take such a large patchset as people
> have already started verification. And we don't usually release -rc2. If you're
> trying to solve a very critical issue with this patchset, I have to release -rc2
> and ask people to verify again. How critical is your issue?

Looks like you have sent a mail to stable@dpdk.org in July.
http://mails.dpdk.org/archives/stable/2018-July/008589.html

I don't know why some of emails to stable@dpdk.org haven't arrived at my mailbox.
I'm still trying to figure out the reason with IT depart in my company and
re-subscribed to the mailing list as well.

My apologies for that.

Yongseok

> For the patchset,
> - "mem: add function for checking memsegs IOVAs addresses"
>  This is adding a new API, so I don't expect any API/ABI breakage, but want to
>  double-confirm with Thomas. Thomas?
> 
> - "bus/pci: use IOVAs check when setting IOVA mode"
>  All the patches got ack except for this one but from looking at the threads in
>  dev mailing list, it looks okay. I have a question though.
> 
>> @@ -640,13 +643,17 @@
>> {
>>        struct rte_pci_device *dev = NULL;
>>        struct rte_pci_driver *drv = NULL;
>> +       int iommu_dma_mask_check_done = 0;
>> 
>>        FOREACH_DRIVER_ON_PCIBUS(drv) {
>>                FOREACH_DEVICE_ON_PCIBUS(dev) {
>>                        if (!rte_pci_match(drv, dev))
>>                                continue;
>> -                       if (!pci_one_device_iommu_support_va(dev))
>> -                               return false;
>> +                       if (!iommu_dma_mask_check_done) {
>> +                               if (pci_one_device_iommu_support_va(dev) < 0)
> 
> pci_one_device_iommu_support_va() returns true/false(1/0), then why do you
> expect to see a negative return value in case of failure?
> 
>> +                                       return false;
>> +                               iommu_dma_mask_check_done  = 1;
>> +                       }
>>                }
>>        }
>>        return true;