[PATCH v2 0/7] Allow dynamic allocation of software IO TLB bounce buffers

Petr Tesařík petr at tesarici.cz
Wed Apr 26 13:16:09 UTC 2023


On Wed, 26 Apr 2023 14:53:52 +0200
Greg Kroah-Hartman <gregkh at linuxfoundation.org> wrote:

> On Wed, Apr 26, 2023 at 02:44:39PM +0200, Petr Tesařík wrote:
> > Hi Greg,
> > 
> > On Wed, 26 Apr 2023 14:26:36 +0200
> > Greg Kroah-Hartman <gregkh at linuxfoundation.org> wrote:
> >   
> > > On Wed, Apr 26, 2023 at 02:15:20PM +0200, Petr Tesařík wrote:  
> > > > Hi,
> > > > 
> > > > On Wed, 19 Apr 2023 12:03:52 +0200
> > > > Petr Tesarik <petrtesarik at huaweicloud.com> wrote:
> > > >     
> > > > > From: Petr Tesarik <petr.tesarik.ext at huawei.com>
> > > > > 
> > > > > The goal of my work is to provide more flexibility in the sizing of
> > > > > SWIOTLB.
> > > > > 
> > > > > The software IO TLB was designed with these assumptions:
> > > > > 
> > > > > 1. It would not be used much, especially on 64-bit systems.
> > > > > 2. A small fixed memory area (64 MiB by default) is sufficient to
> > > > >    handle the few cases which require a bounce buffer.
> > > > > 3. 64 MiB is little enough that it has no impact on the rest of the
> > > > >    system.
> > > > > 
> > > > > First, if SEV is active, all DMA must be done through shared
> > > > > unencrypted pages, and SWIOTLB is used to make this happen without
> > > > > changing device drivers. The software IO TLB size is increased to
> > > > > 6% of total memory in sev_setup_arch(), but that is more of an
> > > > > approximation. The actual requirements may vary depending on the
> > > > > amount of I/O and which drivers are used. These factors may not be
> > > > > know at boot time, i.e. when SWIOTLB is allocated.
> > > > > 
> > > > > Second, other colleagues have noticed that they can reliably get
> > > > > rid of occasional OOM kills on an Arm embedded device by reducing
> > > > > the SWIOTLB size. This can be achieved with a kernel parameter, but
> > > > > determining the right value puts additional burden on pre-release
> > > > > testing, which could be avoided if SWIOTLB is allocated small and
> > > > > grows only when necessary.    
> > > > 
> > > > Now that merging into 6.4 has begun, what about this patch series? I'm
> > > > eager to get some feedback (positive or negative) and respin the next
> > > > version.    
> > > 
> > > It's the merge window, we can't add new things that haven't been in
> > > linux-next already.  
> > 
> > This is understood. I'm not asking for immediate inclusion.
> >   
> > >   Please resubmit it after -rc1 is out.  
> > 
> > If you can believe that rebasing to -rc1 will be enough, then I will
> > also try to believe I'm lucky. ;-)
> > 
> > The kind of feedback I really want to get is e.g. about the extra
> > per-device DMA-specific fields. If they cannot be added to struct
> > device, then I'd rather start discussing an interim solution, because
> > getting all existing DMA fields out of that struct will take a lot of
> > time...  
> 
> I thought the goal was to get them out of the device and into the bus
> instead right?  Or was it the other way around?  I can't remember
> anymore, sorry...

You remember it almost right. The goal is to get them out of struct
device into the bus (or platform device, or whatever holds struct
device). But I'd like to keep this task decoupled from the dynamic
swiotlb.

Thanks,
Petr T


More information about the dri-devel mailing list