Re: [PATCH v6] swiotlb: Adjust SWIOTBL bounce buffer size for SEV guests.

From: Konrad Rzeszutek Wilk
Date: Tue Dec 01 2020 - 15:42:14 EST


On Tue, Nov 24, 2020 at 11:46:22PM +0000, Ashish Kalra wrote:
> Hello Konrad,
>
> On Mon, Nov 23, 2020 at 10:56:31PM +0000, Ashish Kalra wrote:
> > Hello Konrad,
> >
> > On Mon, Nov 23, 2020 at 12:56:32PM -0500, Konrad Rzeszutek Wilk wrote:
> > > On Mon, Nov 23, 2020 at 06:06:47PM +0100, Borislav Petkov wrote:
> > > > On Thu, Nov 19, 2020 at 09:42:05PM +0000, Ashish Kalra wrote:
> > > > > From: Ashish Kalra <ashish.kalra@xxxxxxx>
> > > > >
> > > > > For SEV, all DMA to and from guest has to use shared (un-encrypted) pages.
> > > > > SEV uses SWIOTLB to make this happen without requiring changes to device
> > > > > drivers. However, depending on workload being run, the default 64MB of
> > > > > SWIOTLB might not be enough and SWIOTLB may run out of buffers to use
> > > > > for DMA, resulting in I/O errors and/or performance degradation for
> > > > > high I/O workloads.
> > > > >
> > > > > Increase the default size of SWIOTLB for SEV guests using a minimum
> > > > > value of 128MB and a maximum value of 512MB, determining on amount
> > > > > of provisioned guest memory.
> > > >
> > > > That sentence needs massaging.
> > > >
> > > > > Using late_initcall() interface to invoke swiotlb_adjust() does not
> > > > > work as the size adjustment needs to be done before mem_encrypt_init()
> > > > > and reserve_crashkernel() which use the allocated SWIOTLB buffer size,
> > > > > hence calling it explicitly from setup_arch().
> > > >
> > > > "hence call it ... "
> > > >
> > > > >
> > > > > The SWIOTLB default size adjustment is added as an architecture specific
> > > >
> > > > "... is added... " needs to be "Add ..."
> > > >
> > > > > interface/callback to allow architectures such as those supporting memory
> > > > > encryption to adjust/expand SWIOTLB size for their use.
> > > > >
> > > > > v5 fixed build errors and warnings as
> > > > > Reported-by: kbuild test robot <lkp@xxxxxxxxx>
> > > > >
> > > > > Signed-off-by: Ashish Kalra <ashish.kalra@xxxxxxx>
> > > > > ---
> > > > > arch/x86/kernel/setup.c | 2 ++
> > > > > arch/x86/mm/mem_encrypt.c | 32 ++++++++++++++++++++++++++++++++
> > > > > include/linux/swiotlb.h | 6 ++++++
> > > > > kernel/dma/swiotlb.c | 24 ++++++++++++++++++++++++
> > > > > 4 files changed, 64 insertions(+)
> > > > >
> > > > > diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
> > > > > index 3511736fbc74..b073d58dd4a3 100644
> > > > > --- a/arch/x86/kernel/setup.c
> > > > > +++ b/arch/x86/kernel/setup.c
> > > > > @@ -1166,6 +1166,8 @@ void __init setup_arch(char **cmdline_p)
> > > > > if (boot_cpu_has(X86_FEATURE_GBPAGES))
> > > > > hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
> > > > >
> > > > > + swiotlb_adjust();
> > > > > +
> > > > > /*
> > > > > * Reserve memory for crash kernel after SRAT is parsed so that it
> > > > > * won't consume hotpluggable memory.
> > > > > diff --git a/arch/x86/mm/mem_encrypt.c b/arch/x86/mm/mem_encrypt.c
> > > > > index 3f248f0d0e07..c79a0d761db5 100644
> > > > > --- a/arch/x86/mm/mem_encrypt.c
> > > > > +++ b/arch/x86/mm/mem_encrypt.c
> > > > > @@ -490,6 +490,38 @@ static void print_mem_encrypt_feature_info(void)
> > > > > }
> > > > >
> > > > > /* Architecture __weak replacement functions */
> > > > > +unsigned long __init arch_swiotlb_adjust(unsigned long iotlb_default_size)
> > > > > +{
> > > > > + unsigned long size = 0;
> > > >
> > > > unsigned long size = iotlb_default_size;
> > > >
> > > > > +
> > > > > + /*
> > > > > + * For SEV, all DMA has to occur via shared/unencrypted pages.
> > > > > + * SEV uses SWOTLB to make this happen without changing device
> > > > > + * drivers. However, depending on the workload being run, the
> > > > > + * default 64MB of SWIOTLB may not be enough & SWIOTLB may
> > > > ^
> > > >
> > > > Use words pls, not "&".
> > > >
> > > >
> > > > > + * run out of buffers for DMA, resulting in I/O errors and/or
> > > > > + * performance degradation especially with high I/O workloads.
> > > > > + * Increase the default size of SWIOTLB for SEV guests using
> > > > > + * a minimum value of 128MB and a maximum value of 512MB,
> > > > > + * depending on amount of provisioned guest memory.
> > > > > + */
> > > > > + if (sev_active()) {
> > > > > + phys_addr_t total_mem = memblock_phys_mem_size();
> > > > > +
> > > > > + if (total_mem <= SZ_1G)
> > > > > + size = max(iotlb_default_size, (unsigned long) SZ_128M);
> > > > > + else if (total_mem <= SZ_4G)
> > > > > + size = max(iotlb_default_size, (unsigned long) SZ_256M);
> > >
> > > That is eating 128MB for 1GB, aka 12% of the guest memory allocated statically for this.
> > >
> > > And for guests that are 2GB, that is 12% until it gets to 3GB when it is 8%
> > > and then 6% at 4GB.
> > >
> > > I would prefer this to be based on your memory count, that is 6% of total
> > > memory. And then going forward we can allocate memory _after_ boot and then stich
> > > the late SWIOTLB pool and allocate on demand.
> > >
> > >
> > Ok.
> >
> > As i mentioned earlier, the patch was initially based on using a % of guest memory,
> > as below:
> >
> > +#define SEV_ADJUST_SWIOTLB_SIZE_PERCENT 5
> > +#define SEV_ADJUST_SWIOTLB_SIZE_MAX (1UL << 30)
> > ...
> > ...
> > + if (sev_active() && !io_tlb_nslabs) {
> > + unsigned long total_mem = get_num_physpages() <<
> > + PAGE_SHIFT;
> > +
> > + default_size = total_mem *
> > + SEV_ADJUST_SWIOTLB_SIZE_PERCENT / 100;
> > +
> > + default_size = ALIGN(default_size, 1 << IO_TLB_SHIFT);
> > +
> > + default_size = clamp_val(default_size, IO_TLB_DEFAULT_SIZE,
> > + SEV_ADJUST_SWIOTLB_SIZE_MAX);
> > + }
> >
> > So a similar logic can be applied here.
> >
> > >
> > > > > + else
> > > > > + size = max(iotlb_default_size, (unsigned long) SZ_512M);
> > > > > +
> > > > > + pr_info("SWIOTLB bounce buffer size adjusted to %luMB for SEV platform",
> > > >
> > > > just "... for SEV" - no need for "platform".
> > > >
> > > > ...
> > > >
> > > > > diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c
> > > > > index c19379fabd20..3be9a19ea0a5 100644
> > > > > --- a/kernel/dma/swiotlb.c
> > > > > +++ b/kernel/dma/swiotlb.c
> > > > > @@ -163,6 +163,30 @@ unsigned long swiotlb_size_or_default(void)
> > > > > return size ? size : (IO_TLB_DEFAULT_SIZE);
> > > > > }
> > > > >
> > > > > +unsigned long __init __weak arch_swiotlb_adjust(unsigned long size)
> > > > > +{
> > > > > + return 0;
> > > >
> > > > That, of course, needs to return size, not 0.
> > >
> > > This is not going to work for TDX. I think having a registration
> > > to SWIOTLB to have this function would be better going forward.
> > >
> > > As in there will be a swiotlb_register_adjuster() which AMD SEV
> > > code can call at start, also TDX can do it (and other platforms).
> > >
> >
> > The question is how does mem_encrypt_init() work ?
> >
> > That uses a similar logic as arch_swiotlb_adjust() as a "__weak"
> > function and i am sure it will also need to have added support for TDX,
> > can't both arch_swiotlb_adjust() and mem_encrypt_init() have specific
> > checks for active AMD/INTEL memory encryption technology and accordingly
> > perform actions, as mem_encrypt_init() currently checks for
> > sev_active().
> >
> > init/main.c:
> >
> > void __init __weak mem_encrypt_init(void) { }
> >
> > start_kernel()
> > {
> > ..
> > mem_encrypt_init();
> > ..
> > }
> >
> > arch/x86/mm/mem_encrypt.c:
> >
> > /* Architecture __weak replacement functions */
> >
> > void __init mem_encrypt_init(void)
> > {
> > if (!sme_me_mask)
> > return;
> >
> > /* Call into SWIOTLB to update the SWIOTLB DMA buffers */
> > swiotlb_update_mem_attributes();
> >
> > /*
> > * With SEV, we need to unroll the rep string I/O instructions.
> > */
> > if (sev_active())
> > static_branch_enable(&sev_enable_key);
> > ...
> > ...
> >
>
> Your thoughts on this ?

That looks quite sensible.
>
> Thanks,
> Ashish
>
> >
> > > >
> > > > > +}
> > > > > +
> > > > > +void __init swiotlb_adjust(void)
> > > > > +{
> > > > > + unsigned long size;
> > > > > +
> > > > > + /*
> > > > > + * If swiotlb parameter has not been specified, give a chance to
> > > > > + * architectures such as those supporting memory encryption to
> > > > > + * adjust/expand SWIOTLB size for their use.
> > > > > + */
> > > >
> > > > And when you preset the function-local argument "size" with the size
> > > > coming in as the size argument of arch_swiotlb_adjust()...
> > > >
> > > > > + if (!io_tlb_nslabs) {
> > > > > + size = arch_swiotlb_adjust(IO_TLB_DEFAULT_SIZE);
> > > > > + if (size) {
> > > >
> > > > ... you don't have to do if (size) here either but simply use size to
> > > > compute io_tlb_nslabs, I'd say.
> > > >
> > > > Thx.
> > > >
> > > > --
> > > > Regards/Gruss,
> > > > Boris.
> > > >
> > > > https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpeople.kernel.org%2Ftglx%2Fnotes-about-netiquette&amp;data=04%7C01%7CAshish.Kalra%40amd.com%7Cebd4a85f98f44bdfcb5408d88fd8dfac%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637417508926083910%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&amp;sdata=Ub9PjAPzhDWr7K2iQggTAXwgg4VbORxP%2F%2Fcg6gQreCc%3D&amp;reserved=0