Message ID | 20240429155053.72509-1-roger.pau@citrix.com (mailing list archive) |
---|---|
State | Accepted |
Commit | e4faf06dcc7fb1b2da4cdb40e1d984e934b4e122 |
Headers | show |
Series | xen/x86: add extra pages to unpopulated-alloc if available | expand |
On 29.04.24 17:50, Roger Pau Monne wrote: > Commit 262fc47ac174 ('xen/balloon: don't use PV mode extra memory for zone > device allocations') removed the addition of the extra memory ranges to the > unpopulated range allocator, using those only for the balloon driver. > > This forces the unpopulated allocator to attach hotplug ranges even when spare > memory (as part of the extra memory ranges) is available. Furthermore, on PVH > domains it defeats the purpose of commit 38620fc4e893 ('x86/xen: attempt to > inflate the memory balloon on PVH'), as extra memory ranges would only be > used to map foreign memory if the kernel is built without XEN_UNPOPULATED_ALLOC > support. > > Fix this by adding a helpers that adds the extra memory ranges to the list of > unpopulated pages, and zeroes the ranges so they are not also consumed by the > balloon driver. > > This should have been part of 38620fc4e893, hence the fixes tag. > > Note the current logic relies on unpopulated_init() (and hence > arch_xen_unpopulated_init()) always being called ahead of balloon_init(), so > that the extra memory regions are consumed by arch_xen_unpopulated_init(). > > Fixes: 38620fc4e893 ('x86/xen: attempt to inflate the memory balloon on PVH') > Signed-off-by: Roger Pau Monné <roger.pau@citrix.com> Reviewed-by: Juergen Gross <jgross@suse.com> Juergen
diff --git a/arch/x86/xen/enlighten.c b/arch/x86/xen/enlighten.c index a01ca255b0c6..b88722dfc4f8 100644 --- a/arch/x86/xen/enlighten.c +++ b/arch/x86/xen/enlighten.c @@ -382,3 +382,36 @@ void __init xen_add_extra_mem(unsigned long start_pfn, unsigned long n_pfns) memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns)); } + +#ifdef CONFIG_XEN_UNPOPULATED_ALLOC +int __init arch_xen_unpopulated_init(struct resource **res) +{ + unsigned int i; + + if (!xen_domain()) + return -ENODEV; + + /* Must be set strictly before calling xen_free_unpopulated_pages(). */ + *res = &iomem_resource; + + /* + * Initialize with pages from the extra memory regions (see + * arch/x86/xen/setup.c). + */ + for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) { + unsigned int j; + + for (j = 0; j < xen_extra_mem[i].n_pfns; j++) { + struct page *pg = + pfn_to_page(xen_extra_mem[i].start_pfn + j); + + xen_free_unpopulated_pages(1, &pg); + } + + /* Zero so region is not also added to the balloon driver. */ + xen_extra_mem[i].n_pfns = 0; + } + + return 0; +} +#endif
Commit 262fc47ac174 ('xen/balloon: don't use PV mode extra memory for zone device allocations') removed the addition of the extra memory ranges to the unpopulated range allocator, using those only for the balloon driver. This forces the unpopulated allocator to attach hotplug ranges even when spare memory (as part of the extra memory ranges) is available. Furthermore, on PVH domains it defeats the purpose of commit 38620fc4e893 ('x86/xen: attempt to inflate the memory balloon on PVH'), as extra memory ranges would only be used to map foreign memory if the kernel is built without XEN_UNPOPULATED_ALLOC support. Fix this by adding a helpers that adds the extra memory ranges to the list of unpopulated pages, and zeroes the ranges so they are not also consumed by the balloon driver. This should have been part of 38620fc4e893, hence the fixes tag. Note the current logic relies on unpopulated_init() (and hence arch_xen_unpopulated_init()) always being called ahead of balloon_init(), so that the extra memory regions are consumed by arch_xen_unpopulated_init(). Fixes: 38620fc4e893 ('x86/xen: attempt to inflate the memory balloon on PVH') Signed-off-by: Roger Pau Monné <roger.pau@citrix.com> --- There's a lot of duplication between the unpopulated allocator and the balloon driver. I feel like the balloon driver should request any extra memory it needs to the unpopulated allocator, so that the current helpers provided by the XEN_BALLOON_MEMORY_HOTPLUG option could be replaced with wrappers around the unpopulated handlers. However this is much more work than strictly required here, and won't be suitable for backport IMO. Hence the more contained fix presented in this patch. --- arch/x86/xen/enlighten.c | 33 +++++++++++++++++++++++++++++++++ 1 file changed, 33 insertions(+)