| Message ID | 7-v2-545d29711869+a76b5-iommu_pages_jgg@nvidia.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | iommu: Further abstract iommu-pages | expand |
On 2/15/25 01:07, Jason Gunthorpe wrote: > These are called in a lot of places and are not trivial. Move them to the > core module. > > Tidy some of the comments and function arguments, fold > __iommu_alloc_account() into its only caller, change > __iommu_free_account() into __iommu_free_page() to remove some > duplication. > > Signed-off-by: Jason Gunthorpe<jgg@nvidia.com> > --- > drivers/iommu/Makefile | 1 + > drivers/iommu/iommu-pages.c | 84 +++++++++++++++++++++++++++++ > drivers/iommu/iommu-pages.h | 103 ++---------------------------------- > 3 files changed, 90 insertions(+), 98 deletions(-) > create mode 100644 drivers/iommu/iommu-pages.c > > diff --git a/drivers/iommu/Makefile b/drivers/iommu/Makefile > index 5e5a83c6c2aae2..fe91d770abe16c 100644 > --- a/drivers/iommu/Makefile > +++ b/drivers/iommu/Makefile > @@ -1,6 +1,7 @@ > # SPDX-License-Identifier: GPL-2.0 > obj-y += amd/ intel/ arm/ iommufd/ riscv/ > obj-$(CONFIG_IOMMU_API) += iommu.o > +obj-$(CONFIG_IOMMU_SUPPORT) += iommu-pages.o > obj-$(CONFIG_IOMMU_API) += iommu-traces.o > obj-$(CONFIG_IOMMU_API) += iommu-sysfs.o > obj-$(CONFIG_IOMMU_DEBUGFS) += iommu-debugfs.o > diff --git a/drivers/iommu/iommu-pages.c b/drivers/iommu/iommu-pages.c > new file mode 100644 > index 00000000000000..dbf7205bb23dcc > --- /dev/null > +++ b/drivers/iommu/iommu-pages.c > @@ -0,0 +1,84 @@ > +// SPDX-License-Identifier: GPL-2.0-only > +/* > + * Copyright (c) 2024, Google LLC. > + * Pasha Tatashin<pasha.tatashin@soleen.com> > + */ > +#include "iommu-pages.h" > +#include <linux/gfp.h> > +#include <linux/mm.h> > + > +/** > + * iommu_alloc_pages_node - Allocate a zeroed page of a given order from > + * specific NUMA node > + * @nid: memory NUMA node id > + * @gfp: buddy allocator flags > + * @order: page order > + * > + * Returns the virtual address of the allocated page. The page must be > + * freed either by calling iommu_free_page() or via iommu_put_pages_list(). nit: ... by calling iommu_free_pages() ... and s/page/pages/g in above comments? > + */ > +void *iommu_alloc_pages_node(int nid, gfp_t gfp, unsigned int order) > +{ > + const unsigned long pgcnt = 1UL << order; > + struct page *page; > + > + page = alloc_pages_node(nid, gfp | __GFP_ZERO | __GFP_COMP, order); > + if (unlikely(!page)) > + return NULL; > + > + /* > + * All page allocations that should be reported to as "iommu-pagetables" > + * to userspace must use one of the functions below. This includes > + * allocations of page-tables and other per-iommu_domain configuration > + * structures. > + * > + * This is necessary for the proper accounting as IOMMU state can be > + * rather large, i.e. multiple gigabytes in size. > + */ > + mod_node_page_state(page_pgdat(page), NR_IOMMU_PAGES, pgcnt); > + mod_lruvec_page_state(page, NR_SECONDARY_PAGETABLE, pgcnt); > + > + return page_address(page); > +} > +EXPORT_SYMBOL_GPL(iommu_alloc_pages_node); > + > +static void __iommu_free_page(struct page *page) It's more readable if renaming it to __iommu_free_pages()? > +{ > + unsigned int order = folio_order(page_folio(page)); > + const unsigned long pgcnt = 1UL << order; > + > + mod_node_page_state(page_pgdat(page), NR_IOMMU_PAGES, -pgcnt); > + mod_lruvec_page_state(page, NR_SECONDARY_PAGETABLE, -pgcnt); > + put_page(page); > +} Thanks, baolu
On Sat, Feb 15, 2025 at 04:48:04PM +0800, Baolu Lu wrote: > > +#include <linux/gfp.h> > > +#include <linux/mm.h> > > + > > +/** > > + * iommu_alloc_pages_node - Allocate a zeroed page of a given order from > > + * specific NUMA node > > + * @nid: memory NUMA node id > > + * @gfp: buddy allocator flags > > + * @order: page order > > + * > > + * Returns the virtual address of the allocated page. The page must be > > + * freed either by calling iommu_free_page() or via iommu_put_pages_list(). > > nit: ... by calling iommu_free_pages() ... Got it > and > > s/page/pages/g in above comments? There is alot of historical confusion here because it was all designed around alloc_pages() which allocated a list of contiguous pages that could be subdivided. When this moved to GFP_COMP and later to folio_alloc() the subdivision is no longer possible. So it is not "pages" at all anymore, but a single "[compound] page". So the module name is called "iommu-pages" but aside from the free list functions everything else acts on a single [compound] page only. If you think about it too much it makes no sense but I didn't want to rename every function. I tried to keep it so that "iommu pages" was part of othe module name, and function designators, but the comments talk about a singular [compound] page > > +static void __iommu_free_page(struct page *page) > > It's more readable if renaming it to __iommu_free_pages()? Ah.. Well, it captures the module name but nothing it does acts on multiple things, since it is internal I used the other name How about I rename it later on to: static void __iommu_free_desc(struct ioptdesc *iopt) ? Thanks, Jason
On 2/19/25 04:19, Jason Gunthorpe wrote: > On Sat, Feb 15, 2025 at 04:48:04PM +0800, Baolu Lu wrote: > >>> +#include <linux/gfp.h> >>> +#include <linux/mm.h> >>> + >>> +/** >>> + * iommu_alloc_pages_node - Allocate a zeroed page of a given order from >>> + * specific NUMA node >>> + * @nid: memory NUMA node id >>> + * @gfp: buddy allocator flags >>> + * @order: page order >>> + * >>> + * Returns the virtual address of the allocated page. The page must be >>> + * freed either by calling iommu_free_page() or via iommu_put_pages_list(). >> nit: ... by calling iommu_free_pages() ... > Got it > >> and >> >> s/page/pages/g in above comments? > There is alot of historical confusion here because it was all designed > around alloc_pages() which allocated a list of contiguous pages that > could be subdivided. When this moved to GFP_COMP and later to > folio_alloc() the subdivision is no longer possible. So it is not > "pages" at all anymore, but a single "[compound] page". > > So the module name is called "iommu-pages" but aside from the free > list functions everything else acts on a single [compound] page only. > > If you think about it too much it makes no sense but I didn't want to > rename every function. I tried to keep it so that "iommu pages" was > part of othe module name, and function designators, but the comments > talk about a singular [compound] page > >>> +static void __iommu_free_page(struct page *page) >> It's more readable if renaming it to __iommu_free_pages()? > Ah.. Well, it captures the module name but nothing it does acts on > multiple things, since it is internal I used the other name > > How about I rename it later on to: > > static void __iommu_free_desc(struct ioptdesc *iopt) > > ? No problem. It actually depends on you. Since this is only used locally, the iommu drivers won't use it. So, anything that helps understand what it does is okay. :-) Thanks, baolu
diff --git a/drivers/iommu/Makefile b/drivers/iommu/Makefile index 5e5a83c6c2aae2..fe91d770abe16c 100644 --- a/drivers/iommu/Makefile +++ b/drivers/iommu/Makefile @@ -1,6 +1,7 @@ # SPDX-License-Identifier: GPL-2.0 obj-y += amd/ intel/ arm/ iommufd/ riscv/ obj-$(CONFIG_IOMMU_API) += iommu.o +obj-$(CONFIG_IOMMU_SUPPORT) += iommu-pages.o obj-$(CONFIG_IOMMU_API) += iommu-traces.o obj-$(CONFIG_IOMMU_API) += iommu-sysfs.o obj-$(CONFIG_IOMMU_DEBUGFS) += iommu-debugfs.o diff --git a/drivers/iommu/iommu-pages.c b/drivers/iommu/iommu-pages.c new file mode 100644 index 00000000000000..dbf7205bb23dcc --- /dev/null +++ b/drivers/iommu/iommu-pages.c @@ -0,0 +1,84 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * Copyright (c) 2024, Google LLC. + * Pasha Tatashin <pasha.tatashin@soleen.com> + */ +#include "iommu-pages.h" +#include <linux/gfp.h> +#include <linux/mm.h> + +/** + * iommu_alloc_pages_node - Allocate a zeroed page of a given order from + * specific NUMA node + * @nid: memory NUMA node id + * @gfp: buddy allocator flags + * @order: page order + * + * Returns the virtual address of the allocated page. The page must be + * freed either by calling iommu_free_page() or via iommu_put_pages_list(). + */ +void *iommu_alloc_pages_node(int nid, gfp_t gfp, unsigned int order) +{ + const unsigned long pgcnt = 1UL << order; + struct page *page; + + page = alloc_pages_node(nid, gfp | __GFP_ZERO | __GFP_COMP, order); + if (unlikely(!page)) + return NULL; + + /* + * All page allocations that should be reported to as "iommu-pagetables" + * to userspace must use one of the functions below. This includes + * allocations of page-tables and other per-iommu_domain configuration + * structures. + * + * This is necessary for the proper accounting as IOMMU state can be + * rather large, i.e. multiple gigabytes in size. + */ + mod_node_page_state(page_pgdat(page), NR_IOMMU_PAGES, pgcnt); + mod_lruvec_page_state(page, NR_SECONDARY_PAGETABLE, pgcnt); + + return page_address(page); +} +EXPORT_SYMBOL_GPL(iommu_alloc_pages_node); + +static void __iommu_free_page(struct page *page) +{ + unsigned int order = folio_order(page_folio(page)); + const unsigned long pgcnt = 1UL << order; + + mod_node_page_state(page_pgdat(page), NR_IOMMU_PAGES, -pgcnt); + mod_lruvec_page_state(page, NR_SECONDARY_PAGETABLE, -pgcnt); + put_page(page); +} + +/** + * iommu_free_pages - free pages + * @virt: virtual address of the page to be freed. + * + * The page must have have been allocated by iommu_alloc_pages_node() + */ +void iommu_free_pages(void *virt) +{ + if (!virt) + return; + __iommu_free_page(virt_to_page(virt)); +} +EXPORT_SYMBOL_GPL(iommu_free_pages); + +/** + * iommu_put_pages_list - free a list of pages. + * @head: the head of the lru list to be freed. + * + * Frees a list of pages allocated by iommu_alloc_pages_node(). + */ +void iommu_put_pages_list(struct list_head *head) +{ + while (!list_empty(head)) { + struct page *p = list_entry(head->prev, struct page, lru); + + list_del(&p->lru); + __iommu_free_page(p); + } +} +EXPORT_SYMBOL_GPL(iommu_put_pages_list); diff --git a/drivers/iommu/iommu-pages.h b/drivers/iommu/iommu-pages.h index fcd17b94f7b830..e3c35aa14ad716 100644 --- a/drivers/iommu/iommu-pages.h +++ b/drivers/iommu/iommu-pages.h @@ -7,67 +7,12 @@ #ifndef __IOMMU_PAGES_H #define __IOMMU_PAGES_H -#include <linux/vmstat.h> -#include <linux/gfp.h> -#include <linux/mm.h> +#include <linux/types.h> +#include <linux/topology.h> -/* - * All page allocations that should be reported to as "iommu-pagetables" to - * userspace must use one of the functions below. This includes allocations of - * page-tables and other per-iommu_domain configuration structures. - * - * This is necessary for the proper accounting as IOMMU state can be rather - * large, i.e. multiple gigabytes in size. - */ - -/** - * __iommu_alloc_account - account for newly allocated page. - * @page: head struct page of the page. - * @order: order of the page - */ -static inline void __iommu_alloc_account(struct page *page, int order) -{ - const long pgcnt = 1l << order; - - mod_node_page_state(page_pgdat(page), NR_IOMMU_PAGES, pgcnt); - mod_lruvec_page_state(page, NR_SECONDARY_PAGETABLE, pgcnt); -} - -/** - * __iommu_free_account - account a page that is about to be freed. - * @page: head struct page of the page. - * @order: order of the page - */ -static inline void __iommu_free_account(struct page *page) -{ - unsigned int order = folio_order(page_folio(page)); - const long pgcnt = 1l << order; - - mod_node_page_state(page_pgdat(page), NR_IOMMU_PAGES, -pgcnt); - mod_lruvec_page_state(page, NR_SECONDARY_PAGETABLE, -pgcnt); -} - -/** - * iommu_alloc_pages_node - allocate a zeroed page of a given order from - * specific NUMA node. - * @nid: memory NUMA node id - * @gfp: buddy allocator flags - * @order: page order - * - * returns the virtual address of the allocated page - */ -static inline void *iommu_alloc_pages_node(int nid, gfp_t gfp, int order) -{ - struct page *page = - alloc_pages_node(nid, gfp | __GFP_ZERO | __GFP_COMP, order); - - if (unlikely(!page)) - return NULL; - - __iommu_alloc_account(page, order); - - return page_address(page); -} +void *iommu_alloc_pages_node(int nid, gfp_t gfp, unsigned int order); +void iommu_free_pages(void *virt); +void iommu_put_pages_list(struct list_head *head); /** * iommu_alloc_pages - allocate a zeroed page of a given order @@ -104,42 +49,4 @@ static inline void *iommu_alloc_page(gfp_t gfp) return iommu_alloc_pages_node(numa_node_id(), gfp, 0); } -/** - * iommu_free_pages - free pages - * @virt: virtual address of the page to be freed. - * - * The page must have have been allocated by iommu_alloc_pages_node() - */ -static inline void iommu_free_pages(void *virt) -{ - struct page *page; - - if (!virt) - return; - - page = virt_to_page(virt); - __iommu_free_account(page); - put_page(page); -} - -/** - * iommu_put_pages_list - free a list of pages. - * @page: the head of the lru list to be freed. - * - * There are no locking requirement for these pages, as they are going to be - * put on a free list as soon as refcount reaches 0. Pages are put on this LRU - * list once they are removed from the IOMMU page tables. However, they can - * still be access through debugfs. - */ -static inline void iommu_put_pages_list(struct list_head *page) -{ - while (!list_empty(page)) { - struct page *p = list_entry(page->prev, struct page, lru); - - list_del(&p->lru); - __iommu_free_account(p); - put_page(p); - } -} - #endif /* __IOMMU_PAGES_H */
These are called in a lot of places and are not trivial. Move them to the core module. Tidy some of the comments and function arguments, fold __iommu_alloc_account() into its only caller, change __iommu_free_account() into __iommu_free_page() to remove some duplication. Signed-off-by: Jason Gunthorpe <jgg@nvidia.com> --- drivers/iommu/Makefile | 1 + drivers/iommu/iommu-pages.c | 84 +++++++++++++++++++++++++++++ drivers/iommu/iommu-pages.h | 103 ++---------------------------------- 3 files changed, 90 insertions(+), 98 deletions(-) create mode 100644 drivers/iommu/iommu-pages.c