Message-ID: <BN9PR11MB52769480729491394F4719728CD62@BN9PR11MB5276.namprd11.prod.outlook.com>
Date: Wed, 26 Jun 2024 01:53:23 +0000
From: "Tian, Kevin" <kevin.tian@...el.com>
To: Baolu Lu <baolu.lu@...ux.intel.com>, Joerg Roedel <joro@...tes.org>, "Liu,
 Yi L" <yi.l.liu@...el.com>, Jacob Pan <jacob.jun.pan@...ux.intel.com>
CC: Will Deacon <will@...nel.org>, Robin Murphy <robin.murphy@....com>,
	"iommu@...ts.linux.dev" <iommu@...ts.linux.dev>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: RE: [PATCH 2/2] iommu/vt-d: Remove hardware automatic ATS dependency

> From: Baolu Lu <baolu.lu@...ux.intel.com>
> Sent: Tuesday, June 25, 2024 4:40 PM
> 
> On 2024/6/25 10:32, Tian, Kevin wrote:
> >> From: Lu Baolu<baolu.lu@...ux.intel.com>
> >> Sent: Monday, June 24, 2024 1:25 PM
> >>
> >> If a device is listed in the SATC table with ATC_REQUIRED flag set, it
> >> indicates that the device has a functional requirement to enable its ATC
> >> (via the ATS capability) for device operation. However, when IOMMU is
> >> running in the legacy mode, ATS could be automatically supported by the
> >> hardware so that the OS has no need to support the ATS functionality.
> > hmm I don't think "has no need to support" matches...
> >
> >> This is a backward compatibility feature which enables older OSs. Since
> >> Linux VT-d implementation has already supported ATS features for a long
> >> time, there is no need to rely on this compatibility hardware. Remove it
> >> to make the driver future-proof.
> >>
> >> Signed-off-by: Lu Baolu<baolu.lu@...ux.intel.com>
> >> ---
> >>   drivers/iommu/intel/iommu.c | 9 +--------
> >>   1 file changed, 1 insertion(+), 8 deletions(-)
> >>
> >> diff --git a/drivers/iommu/intel/iommu.c b/drivers/iommu/intel/iommu.c
> >> index 07e394dfccc1..b63347c8bf5d 100644
> >> --- a/drivers/iommu/intel/iommu.c
> >> +++ b/drivers/iommu/intel/iommu.c
> >> @@ -3056,14 +3056,7 @@ static bool dmar_ats_supported(struct pci_dev
> >> *dev, struct intel_iommu *iommu)
> >>   	dev = pci_physfn(dev);
> >>   	satcu = dmar_find_matched_satc_unit(dev);
> >>   	if (satcu)
> >> -		/*
> >> -		 * This device supports ATS as it is in SATC table.
> >> -		 * When IOMMU is in legacy mode, enabling ATS is done
> >> -		 * automatically by HW for the device that requires
> >> -		 * ATS, hence OS should not enable this device ATS
> >> -		 * to avoid duplicated TLB invalidation.
> >> -		 */
> > ...what above comment tries to convey.
> >
> > If this comment is valid, it's not about whether the OS itself supports
> > ATS. instead it's a requirement for the OS to not manage ATS when
> > it's already managed by HW.
> >
> > Unless there is a way to disable hw management with this change...
> 
> This comment is not correct. The hardware automatic ATS is for older OS
> compatible purposes, where the ATS is not aware of by the OS yet, but
> ATS is functionally required for some SOC-integrated accelerators.
> 
> The HAS specification for those platforms states that OSs supporting ATS
> (so-called enlightened OSs) don't require automatic ATS anymore.
> 
>  From the iommu driver's point of view, automatic ATS is not part of the
> VT-d spec and also not enumerable, hence it should be transparent.
> 

I'm curious how automatic ATS can be disabled otherwise the old
comment still makes sense as you will have both HW and SW
managing ATS and then duplicated invalidations.

Is there a BIOS option to disable automatic ATS? Then the user will
need to know which kernel version supports ATS to decide.

Or is it automatically enabled/disabled based on whether the
IOMMU is in legacy or scalable mode? If yes then we may still want
to disable SW-managed ATS when the IOMMU is in legacy mode to
avoid duplicated invalidations.

btw ATS support was introduced long long time ago:

commit 93a23a7271dfb811b3adb72779054c3a24433112
Author: Yu Zhao <yu.zhao@...el.com>
Date:   Mon May 18 13:51:37 2009 +0800

    VT-d: support the device IOTLB

    Enable the device IOTLB (i.e. ATS) for both the bare metal and KVM
    environments.

    Signed-off-by: Yu Zhao <yu.zhao@...el.com>
    Signed-off-by: David Woodhouse <David.Woodhouse@...el.com>

while this hw-managed ATS was supported much later:

commit 97f2f2c5317f55ae3440733a090a96a251da222b
Author: Yian Chen <yian.chen@...el.com>
Date:   Tue Mar 1 10:01:59 2022 +0800

    iommu/vt-d: Enable ATS for the devices in SATC table

    Starting from Intel VT-d v3.2, Intel platform BIOS can provide additional
    SATC table structure. SATC table includes a list of SoC integrated devices
    that support ATC (Address translation cache).

    Enabling ATC (via ATS capability) can be a functional requirement for SATC
    device operation or optional to enhance device performance/functionality.
    This is determined by the bit of ATC_REQUIRED in SATC table. When IOMMU is
    working in scalable mode, software chooses to always enable ATS for every
    device in SATC table because Intel SoC devices in SATC table are trusted to
    use ATS.

    On the other hand, if IOMMU is in legacy mode, ATS of SATC capable devices
    can work transparently to software and be automatically enabled by IOMMU
    hardware. As the result, there is no need for software to enable ATS on
    these devices.

    This also removes dmar_find_matched_atsr_unit() helper as it becomes dead
    code now.

    Signed-off-by: Yian Chen <yian.chen@...el.com>
    Link: https://lore.kernel.org/r/20220222185416.1722611-1-yian.chen@intel.com
    Signed-off-by: Lu Baolu <baolu.lu@...ux.intel.com>
    Link: https://lore.kernel.org/r/20220301020159.633356-13-baolu.lu@linux.inte
    Signed-off-by: Joerg Roedel <jroedel@...e.de>

That history doesn't appear to support your argument that it's only
for old OS which is not aware of ATS...

Powered by blists - more mailing lists