Message ID | 1436474552-31789-10-git-send-email-julien.grall@citrix.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On Thu, Jul 09, 2015 at 09:42:21PM +0100, Julien Grall wrote: > When Linux is using 64K page granularity, every page will be slipt in > multiple non-contiguous 4K MFN (page granularity of Xen). But you don't care about that on the Linux layer I think? As in, is there an SWIOTLB that does PFN to MFN and vice-versa translation? I thought that ARM guests are not exposed to the MFN<->PFN logic and trying to figure that out to not screw up the DMA engine on a PCIe device slurping up contingous MFNs which don't map to contingous PFNs? > > I'm not sure how to handle efficiently the check to know whether we can > merge 2 biovec with a such case. So for now, always says that biovec are > not mergeable. > > Signed-off-by: Julien Grall <julien.grall@citrix.com> > Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> > Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com> > Cc: David Vrabel <david.vrabel@citrix.com> > --- > Changes in v2: > - Remove the workaround and check if the Linux page granularity > is the same as Xen or not > --- > drivers/xen/biomerge.c | 7 +++++++ > 1 file changed, 7 insertions(+) > > diff --git a/drivers/xen/biomerge.c b/drivers/xen/biomerge.c > index 0edb91c..571567c 100644 > --- a/drivers/xen/biomerge.c > +++ b/drivers/xen/biomerge.c > @@ -6,10 +6,17 @@ > bool xen_biovec_phys_mergeable(const struct bio_vec *vec1, > const struct bio_vec *vec2) > { > +#if XEN_PAGE_SIZE == PAGE_SIZE > unsigned long mfn1 = pfn_to_mfn(page_to_pfn(vec1->bv_page)); > unsigned long mfn2 = pfn_to_mfn(page_to_pfn(vec2->bv_page)); > > return __BIOVEC_PHYS_MERGEABLE(vec1, vec2) && > ((mfn1 == mfn2) || ((mfn1+1) == mfn2)); > +#else > + /* XXX: bio_vec are not mergeable when using different page size in > + * Xen and Linux > + */ > + return 0; > +#endif > } > EXPORT_SYMBOL(xen_biovec_phys_mergeable); > -- > 2.1.4 >
Hi Konrad, On 10/07/2015 21:12, Konrad Rzeszutek Wilk wrote: > On Thu, Jul 09, 2015 at 09:42:21PM +0100, Julien Grall wrote: >> When Linux is using 64K page granularity, every page will be slipt in >> multiple non-contiguous 4K MFN (page granularity of Xen). > > But you don't care about that on the Linux layer I think? In general use case (i.e arch agnostic) we care about it. We don't want to merge 2 biovec if they are not living on the same MFNs. > As in, is there an SWIOTLB that does PFN to MFN and vice-versa > translation? > > I thought that ARM guests are not exposed to the MFN<->PFN logic > and trying to figure that out to not screw up the DMA engine > on a PCIe device slurping up contingous MFNs which don't map > to contingous PFNs? I will let these 2 questions for Stefano. He knows better than me swiotlb for ARM. So far, I skipped swiotlb implementation for 64KB page granularity as I'm not sure what to do when a page is split across multiple MFNs. Although I don't think this can happen with this specific series as: - The memory is a direct mapping so any Linux page is using contiguous MFNs. - Foreign mapping is using the 4KB of the Linux page. This is for an easier implementation. For the latter, I plan to work on using the Linux page to map multiple foreign gfn. I have to talk with Stefano about it how to handle it. Regards,
On Fri, 10 Jul 2015, Konrad Rzeszutek Wilk wrote: > On Thu, Jul 09, 2015 at 09:42:21PM +0100, Julien Grall wrote: > > When Linux is using 64K page granularity, every page will be slipt in > > multiple non-contiguous 4K MFN (page granularity of Xen). > > But you don't care about that on the Linux layer I think? > > As in, is there an SWIOTLB that does PFN to MFN and vice-versa > translation? > > I thought that ARM guests are not exposed to the MFN<->PFN logic > and trying to figure that out to not screw up the DMA engine > on a PCIe device slurping up contingous MFNs which don't map > to contingous PFNs? Dom0 is mapped 1:1, so pfn == mfn normally, however grant maps unavoidably screw up the 1:1, so the swiotlb jumps in to save the day when a foreign granted page is involved in a dma operation. Regarding xen_biovec_phys_mergeable, we could check that all the pfn == mfn and return true in that case. > > I'm not sure how to handle efficiently the check to know whether we can > > merge 2 biovec with a such case. So for now, always says that biovec are > > not mergeable. > > > > Signed-off-by: Julien Grall <julien.grall@citrix.com> > > Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> > > Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com> > > Cc: David Vrabel <david.vrabel@citrix.com> > > --- > > Changes in v2: > > - Remove the workaround and check if the Linux page granularity > > is the same as Xen or not > > --- > > drivers/xen/biomerge.c | 7 +++++++ > > 1 file changed, 7 insertions(+) > > > > diff --git a/drivers/xen/biomerge.c b/drivers/xen/biomerge.c > > index 0edb91c..571567c 100644 > > --- a/drivers/xen/biomerge.c > > +++ b/drivers/xen/biomerge.c > > @@ -6,10 +6,17 @@ > > bool xen_biovec_phys_mergeable(const struct bio_vec *vec1, > > const struct bio_vec *vec2) > > { > > +#if XEN_PAGE_SIZE == PAGE_SIZE > > unsigned long mfn1 = pfn_to_mfn(page_to_pfn(vec1->bv_page)); > > unsigned long mfn2 = pfn_to_mfn(page_to_pfn(vec2->bv_page)); > > > > return __BIOVEC_PHYS_MERGEABLE(vec1, vec2) && > > ((mfn1 == mfn2) || ((mfn1+1) == mfn2)); > > +#else > > + /* XXX: bio_vec are not mergeable when using different page size in > > + * Xen and Linux > > + */ > > + return 0; > > +#endif > > } > > EXPORT_SYMBOL(xen_biovec_phys_mergeable); > > -- > > 2.1.4 > > >
Hi Stefano, On 16/07/2015 16:33, Stefano Stabellini wrote: > On Fri, 10 Jul 2015, Konrad Rzeszutek Wilk wrote: >> On Thu, Jul 09, 2015 at 09:42:21PM +0100, Julien Grall wrote: >>> When Linux is using 64K page granularity, every page will be slipt in >>> multiple non-contiguous 4K MFN (page granularity of Xen). >> >> But you don't care about that on the Linux layer I think? >> >> As in, is there an SWIOTLB that does PFN to MFN and vice-versa >> translation? >> >> I thought that ARM guests are not exposed to the MFN<->PFN logic >> and trying to figure that out to not screw up the DMA engine >> on a PCIe device slurping up contingous MFNs which don't map >> to contingous PFNs? > > Dom0 is mapped 1:1, so pfn == mfn normally, however grant maps > unavoidably screw up the 1:1, so the swiotlb jumps in to save the day > when a foreign granted page is involved in a dma operation. > > Regarding xen_biovec_phys_mergeable, we could check that all the pfn == > mfn and return true in that case. I mentioned it in the commit message. Although, we would have to loop on every pfn which is slow on 64KB (16 times for every page). Given the biovec is called often, I don't think we can do a such things. Regards,
On Thu, Jul 16, 2015 at 05:15:41PM +0100, Julien Grall wrote: > Hi Stefano, > > On 16/07/2015 16:33, Stefano Stabellini wrote: > >On Fri, 10 Jul 2015, Konrad Rzeszutek Wilk wrote: > >>On Thu, Jul 09, 2015 at 09:42:21PM +0100, Julien Grall wrote: > >>>When Linux is using 64K page granularity, every page will be slipt in > >>>multiple non-contiguous 4K MFN (page granularity of Xen). > >> > >>But you don't care about that on the Linux layer I think? > >> > >>As in, is there an SWIOTLB that does PFN to MFN and vice-versa > >>translation? > >> > >>I thought that ARM guests are not exposed to the MFN<->PFN logic > >>and trying to figure that out to not screw up the DMA engine > >>on a PCIe device slurping up contingous MFNs which don't map > >>to contingous PFNs? > > > >Dom0 is mapped 1:1, so pfn == mfn normally, however grant maps > >unavoidably screw up the 1:1, so the swiotlb jumps in to save the day > >when a foreign granted page is involved in a dma operation. > > > >Regarding xen_biovec_phys_mergeable, we could check that all the pfn == > >mfn and return true in that case. > > I mentioned it in the commit message. Although, we would have to loop on > every pfn which is slow on 64KB (16 times for every page). Given the biovec > is called often, I don't think we can do a such things. OK - it would be good to have the gist of this email thread in the commit message. Thanks. > > Regards, > > -- > Julien Grall
On Thu, 16 Jul 2015, Julien Grall wrote: > Hi Stefano, > > On 16/07/2015 16:33, Stefano Stabellini wrote: > > On Fri, 10 Jul 2015, Konrad Rzeszutek Wilk wrote: > > > On Thu, Jul 09, 2015 at 09:42:21PM +0100, Julien Grall wrote: > > > > When Linux is using 64K page granularity, every page will be slipt in > > > > multiple non-contiguous 4K MFN (page granularity of Xen). > > > > > > But you don't care about that on the Linux layer I think? > > > > > > As in, is there an SWIOTLB that does PFN to MFN and vice-versa > > > translation? > > > > > > I thought that ARM guests are not exposed to the MFN<->PFN logic > > > and trying to figure that out to not screw up the DMA engine > > > on a PCIe device slurping up contingous MFNs which don't map > > > to contingous PFNs? > > > > Dom0 is mapped 1:1, so pfn == mfn normally, however grant maps > > unavoidably screw up the 1:1, so the swiotlb jumps in to save the day > > when a foreign granted page is involved in a dma operation. > > > > Regarding xen_biovec_phys_mergeable, we could check that all the pfn == > > mfn and return true in that case. > > I mentioned it in the commit message. Although, we would have to loop on every > pfn which is slow on 64KB (16 times for every page). Given the biovec is > called often, I don't think we can do a such things. We would have to run some benchmarks, but I think it would still be a win. We should write an ad-hoc __pfn_to_mfn translation function that operates on a range of pfns and simply checks whether an entry is present in that range. It should be just as fast as __pfn_to_mfn. I would definitely recommend it.
On 17/07/15 14:20, Stefano Stabellini wrote: > We would have to run some benchmarks, but I think it would still be a > win. We should write an ad-hoc __pfn_to_mfn translation function that > operates on a range of pfns and simply checks whether an entry is > present in that range. It should be just as fast as __pfn_to_mfn. I > would definitely recommend it. I'd like to see a basic support of 64KB support on Xen pushed in Linux upstream before looking to possible improvement in the code. Can we defer this as the follow-up of this series? Regards,
On Fri, 17 Jul 2015, Julien Grall wrote: > On 17/07/15 14:20, Stefano Stabellini wrote: > > We would have to run some benchmarks, but I think it would still be a > > win. We should write an ad-hoc __pfn_to_mfn translation function that > > operates on a range of pfns and simply checks whether an entry is > > present in that range. It should be just as fast as __pfn_to_mfn. I > > would definitely recommend it. > > I'd like to see a basic support of 64KB support on Xen pushed in Linux > upstream before looking to possible improvement in the code. Can we > defer this as the follow-up of this series? Yes, maybe add a TODO comment in the code.
On 17/07/15 15:45, Stefano Stabellini wrote: > On Fri, 17 Jul 2015, Julien Grall wrote: >> On 17/07/15 14:20, Stefano Stabellini wrote: >>> We would have to run some benchmarks, but I think it would still be a >>> win. We should write an ad-hoc __pfn_to_mfn translation function that >>> operates on a range of pfns and simply checks whether an entry is >>> present in that range. It should be just as fast as __pfn_to_mfn. I >>> would definitely recommend it. >> >> I'd like to see a basic support of 64KB support on Xen pushed in Linux >> upstream before looking to possible improvement in the code. Can we >> defer this as the follow-up of this series? > > Yes, maybe add a TODO comment in the code. Will do. Regards,
diff --git a/drivers/xen/biomerge.c b/drivers/xen/biomerge.c index 0edb91c..571567c 100644 --- a/drivers/xen/biomerge.c +++ b/drivers/xen/biomerge.c @@ -6,10 +6,17 @@ bool xen_biovec_phys_mergeable(const struct bio_vec *vec1, const struct bio_vec *vec2) { +#if XEN_PAGE_SIZE == PAGE_SIZE unsigned long mfn1 = pfn_to_mfn(page_to_pfn(vec1->bv_page)); unsigned long mfn2 = pfn_to_mfn(page_to_pfn(vec2->bv_page)); return __BIOVEC_PHYS_MERGEABLE(vec1, vec2) && ((mfn1 == mfn2) || ((mfn1+1) == mfn2)); +#else + /* XXX: bio_vec are not mergeable when using different page size in + * Xen and Linux + */ + return 0; +#endif } EXPORT_SYMBOL(xen_biovec_phys_mergeable);
When Linux is using 64K page granularity, every page will be slipt in multiple non-contiguous 4K MFN (page granularity of Xen). I'm not sure how to handle efficiently the check to know whether we can merge 2 biovec with a such case. So for now, always says that biovec are not mergeable. Signed-off-by: Julien Grall <julien.grall@citrix.com> Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com> Cc: David Vrabel <david.vrabel@citrix.com> --- Changes in v2: - Remove the workaround and check if the Linux page granularity is the same as Xen or not --- drivers/xen/biomerge.c | 7 +++++++ 1 file changed, 7 insertions(+)