udmabuf: Add support for mapping hugepages
Subsystems (2)
Vivek Kasireddy
maintainerPATCH 1/1 udmabuf: Add support for mapping hugepages
If the VMM's (Qemu) memory backend is backed up by memfd + Hugepages (hugetlbfs and not THP), we have to first find the hugepage(s) where the Guest allocations are located and then extract the regular 4k sized subpages from them.
--- drivers/dma-buf/udmabuf.c | 40 +++++++++++++++++++++++++++++++-------- 1 file changed, 32 insertions(+), 8 deletions(-)
drivers/dma-buf/udmabuf.c +32 -8
@@ -11,6 +11,7 @@#include <linux/shmem_fs.h> #include <linux/slab.h> #include <linux/udmabuf.h> +#include <linux/hugetlb.h> static const u32 list_limit =1024 ;/* udmabuf_create_list->count limit */ static const size_t size_limit_mb =64 ;/* total dmabuf size, in megabytes */
@@ -162,8 +163,9 @@ static long udmabuf_create(struct miscdevice *device,struct file *memfd =NULL ;struct udmabuf *ubuf;struct dma_buf *buf; -pgoff_t pgoff, pgcnt, pgidx, pgbuf =0 , pglimit; -struct page *page; +pgoff_t pgoff, pgcnt, pgidx, pgbuf =0 , pglimit, subpgoff; +struct page *page, *hpage =NULL ; +struct hstate *hpstate;int seals, ret = -EINVAL ;u32 i, flags;
@@ -194,7 +196,8 @@ static long udmabuf_create(struct miscdevice *device, memfd =fget (list[i].memfd);if (!memfd) goto err; -if (!shmem_mapping (file_inode (memfd)->i_mapping)) +if (!shmem_mapping (file_inode (memfd)->i_mapping) && + !is_file_hugepages (memfd)) goto err; seals =memfd_fcntl (memfd,F_GET_SEALS ,0 );if (seals == -EINVAL )
@@ -205,17 +208,38 @@ static long udmabuf_create(struct miscdevice *device, goto err; pgoff = list[i].offset >>PAGE_SHIFT ; pgcnt = list[i].size >>PAGE_SHIFT ; -for (pgidx =0 ; pgidx < pgcnt; pgidx++) { - page =shmem_read_mapping_page ( -file_inode (memfd)->i_mapping, pgoff + pgidx); -if (IS_ERR (page)) { - ret =PTR_ERR (page); +if (is_file_hugepages (memfd)) { + hpstate =hstate_file (memfd); + pgoff = list[i].offset >>huge_page_shift (hpstate); + subpgoff = (list[i].offset & + ~huge_page_mask (hpstate)) >>PAGE_SHIFT ; + hpage =find_get_page_flags ( +file_inode (memfd)->i_mapping, + pgoff,FGP_ACCESSED ); +if (IS_ERR (hpage)) { + ret =PTR_ERR (hpage); goto err; } + } +for (pgidx =0 ; pgidx < pgcnt; pgidx++) { +if (is_file_hugepages (memfd)) { + page = hpage + subpgoff + pgidx; +get_page (page); + }else { + page =shmem_read_mapping_page ( +file_inode (memfd)->i_mapping, + pgoff + pgidx); +if (IS_ERR (page)) { + ret =PTR_ERR (page); + goto err; + } + } ubuf->pages[pgbuf++] = page; }fput (memfd); memfd =NULL ; +if (hpage) +put_page (hpage); } exp_info.ops = &udmabuf_ops;
Signature · 1 line
2.30.2
Gerd Hoffmann
maintainerHi,
Quoted 7 lines
drivers/dma-buf/udmabuf.c +7
-0+if (is_file_hugepages (memfd)) { + hpstate =hstate_file (memfd); + pgoff = list[i].offset >>huge_page_shift (hpstate); + subpgoff = (list[i].offset & + ~huge_page_mask (hpstate)) >>PAGE_SHIFT ; + hpage = find_get_page_flags( + file_inode(memfd)->i_mapping,
drivers/dma-buf/udmabuf.c static long udmabuf_create(struct miscdevice *device, · line 227
Quoted 10 lines
drivers/dma-buf/udmabuf.c +8
-0+ pgoff, FGP_ACCESSED); +if (IS_ERR (hpage)) { + ret =PTR_ERR (hpage); goto err; } + } +for (pgidx =0 ; pgidx < pgcnt; pgidx++) { +if (is_file_hugepages (memfd)) { + page = hpage + subpgoff + pgidx; +get_page (page);
Hmm. Does this work in case the page range spans multiple huge pages?
take care,
Gerd
Vivek Kasireddy
maintainerPATCH 1/1 udmabuf: Add support for mapping hugepages (v2)
If the VMM's (Qemu) memory backend is backed up by memfd + Hugepages (hugetlbfs and not THP), we have to first find the hugepage(s) where the Guest allocations are located and then extract the regular 4k sized subpages from them.
v2: Ensure that the subpage offsets are calculated correctly when the range of subpage allocations cuts across multiple hugepages.
--- drivers/dma-buf/udmabuf.c | 46 ++++++++++++++++++++++++++++++++------- 1 file changed, 38 insertions(+), 8 deletions(-)
drivers/dma-buf/udmabuf.c +38 -8
@@ -11,6 +11,7 @@#include <linux/shmem_fs.h> #include <linux/slab.h> #include <linux/udmabuf.h> +#include <linux/hugetlb.h> static const u32 list_limit =1024 ;/* udmabuf_create_list->count limit */ static const size_t size_limit_mb =64 ;/* total dmabuf size, in megabytes */
@@ -162,8 +163,10 @@ static long udmabuf_create(struct miscdevice *device,struct file *memfd =NULL ;struct udmabuf *ubuf;struct dma_buf *buf; -pgoff_t pgoff, pgcnt, pgidx, pgbuf =0 , pglimit; -struct page *page; +pgoff_t pgoff, pgcnt, pgidx, pgbuf =0 , pglimit, subpgoff; +uint32_t maxsubpgs; +struct page *page, *hpage =NULL ; +struct hstate *hpstate;int seals, ret = -EINVAL ;u32 i, flags;
@@ -194,7 +197,8 @@ static long udmabuf_create(struct miscdevice *device, memfd =fget (list[i].memfd);if (!memfd) goto err; -if (!shmem_mapping (file_inode (memfd)->i_mapping)) +if (!shmem_mapping (file_inode (memfd)->i_mapping) && + !is_file_hugepages (memfd)) goto err; seals =memfd_fcntl (memfd,F_GET_SEALS ,0 );if (seals == -EINVAL )
@@ -205,12 +209,38 @@ static long udmabuf_create(struct miscdevice *device, goto err; pgoff = list[i].offset >>PAGE_SHIFT ; pgcnt = list[i].size >>PAGE_SHIFT ; +if (is_file_hugepages (memfd)) { + hpstate =hstate_file (memfd); + pgoff = list[i].offset >>huge_page_shift (hpstate); + subpgoff = (list[i].offset & + ~huge_page_mask (hpstate)) >>PAGE_SHIFT ; + maxsubpgs =huge_page_size (hpstate) >>PAGE_SHIFT ; + }for (pgidx =0 ; pgidx < pgcnt; pgidx++) { - page =shmem_read_mapping_page ( -file_inode (memfd)->i_mapping, pgoff + pgidx); -if (IS_ERR (page)) { - ret =PTR_ERR (page); - goto err; +if (is_file_hugepages (memfd)) { + hpage =find_get_page_flags ( +file_inode (memfd)->i_mapping, + pgoff,FGP_ACCESSED ); +if (IS_ERR (hpage)) { + ret =PTR_ERR (hpage); + goto err; + } + + page = hpage + (subpgoff % maxsubpgs); +get_page (page); +put_page (hpage); + + subpgoff++; +if (subpgoff % maxsubpgs ==0 ) + pgoff++; + }else { + page =shmem_read_mapping_page ( +file_inode (memfd)->i_mapping, + pgoff + pgidx); +if (IS_ERR (page)) { + ret =PTR_ERR (page); + goto err; + } } ubuf->pages[pgbuf++] = page; }
Signature · 1 line
2.30.2
Gerd Hoffmann
maintainerRe: [PATCH] udmabuf: Add support for mapping hugepages (v2)
Hi,
drivers/dma-buf/udmabuf.c static long udmabuf_create(struct miscdevice *device, · line 231
Quoted 13 lines
drivers/dma-buf/udmabuf.c +12
-0for (pgidx =0 ; pgidx < pgcnt; pgidx++) { +if (is_file_hugepages (memfd)) { + hpage =find_get_page_flags ( +file_inode (memfd)->i_mapping, + pgoff,FGP_ACCESSED ); +if (IS_ERR (hpage)) { + ret =PTR_ERR (hpage); + goto err; + } + + page = hpage + (subpgoff % maxsubpgs); +get_page (page); +put_page (hpage);
if (hpage && subpgoff == maxsubpgs) {
put_page(hpage);
hpage = NULL;
}
if (!hpage) {
hpage = find_get_page_flags(...)
[ ... ]
}Only lookup the huge page when you cross a hugepage border.
take care,
Gerd
Vivek Kasireddy
maintainerPATCH 1/1 udmabuf: Add support for mapping hugepages (v3)
If the VMM's (Qemu) memory backend is backed up by memfd + Hugepages (hugetlbfs and not THP), we have to first find the hugepage(s) where the Guest allocations are located and then extract the regular 4k sized subpages from them.
v2: Ensure that the subpage and hugepage offsets are calculated correctly when the range of subpage allocations cuts across multiple hugepages.
v3: Instead of repeatedly looking up the hugepage for each subpage, only do it when the subpage allocation crosses over into a different hugepage. (suggested by Gerd and DW)
--- drivers/dma-buf/udmabuf.c | 51 +++++++++++++++++++++++++++++++++------ 1 file changed, 44 insertions(+), 7 deletions(-)
drivers/dma-buf/udmabuf.c +44 -7
@@ -11,6 +11,7 @@#include <linux/shmem_fs.h> #include <linux/slab.h> #include <linux/udmabuf.h> +#include <linux/hugetlb.h> static const u32 list_limit =1024 ;/* udmabuf_create_list->count limit */ static const size_t size_limit_mb =64 ;/* total dmabuf size, in megabytes */
@@ -163,7 +164,9 @@ static long udmabuf_create(struct miscdevice *device,struct udmabuf *ubuf;struct dma_buf *buf;pgoff_t pgoff, pgcnt, pgidx, pgbuf =0 , pglimit; -struct page *page; +struct page *page, *hpage =NULL ; +pgoff_t subpgoff, maxsubpgs; +struct hstate *hpstate;int seals, ret = -EINVAL ;u32 i, flags;
@@ -194,7 +197,8 @@ static long udmabuf_create(struct miscdevice *device, memfd =fget (list[i].memfd);if (!memfd) goto err; -if (!shmem_mapping (file_inode (memfd)->i_mapping)) +if (!shmem_mapping (file_inode (memfd)->i_mapping) && + !is_file_hugepages (memfd)) goto err; seals =memfd_fcntl (memfd,F_GET_SEALS ,0 );if (seals == -EINVAL )
@@ -205,17 +209,50 @@ static long udmabuf_create(struct miscdevice *device, goto err; pgoff = list[i].offset >>PAGE_SHIFT ; pgcnt = list[i].size >>PAGE_SHIFT ; +if (is_file_hugepages (memfd)) { + hpstate =hstate_file (memfd); + pgoff = list[i].offset >>huge_page_shift (hpstate); + subpgoff = (list[i].offset & + ~huge_page_mask (hpstate)) >>PAGE_SHIFT ; + maxsubpgs =huge_page_size (hpstate) >>PAGE_SHIFT ; + }for (pgidx =0 ; pgidx < pgcnt; pgidx++) { - page =shmem_read_mapping_page ( -file_inode (memfd)->i_mapping, pgoff + pgidx); -if (IS_ERR (page)) { - ret =PTR_ERR (page); - goto err; +if (is_file_hugepages (memfd)) { +if (!hpage) { + hpage =find_get_page_flags ( +file_inode (memfd)->i_mapping, + pgoff,FGP_ACCESSED ); +if (IS_ERR (hpage)) { + ret =PTR_ERR (hpage); + goto err; + } + } + page = hpage + subpgoff; +get_page (page); + subpgoff++; +if (subpgoff == maxsubpgs) { +put_page (hpage); + hpage =NULL ; + subpgoff =0 ; + pgoff++; + } + }else { + page =shmem_read_mapping_page ( +file_inode (memfd)->i_mapping, + pgoff + pgidx); +if (IS_ERR (page)) { + ret =PTR_ERR (page); + goto err; + } } ubuf->pages[pgbuf++] = page; }fput (memfd); memfd =NULL ; +if (hpage) { +put_page (hpage); + hpage =NULL ; + } } exp_info.ops = &udmabuf_ops;
Signature · 1 line
2.30.2
Re: [PATCH] udmabuf: Add support for mapping hugepages (v3)
I see the number of entries in the list often exceeds list_limit currently hardcoded to 1024 for full HD scanout resource (== 1920*1080*4 bytes). Can we include a change to increase it to something like 4096 or higher in this patch?
Quoted 111 lines · Vivek Kasireddy
If the VMM's (Qemu) memory backend is backed up by memfd + Hugepages (hugetlbfs and not THP), we have to first find the hugepage(s) where the Guest allocations are located and then extract the regular 4k sized subpages from them.
v2: Ensure that the subpage and hugepage offsets are calculated correctly when the range of subpage allocations cuts across multiple hugepages.
v3: Instead of repeatedly looking up the hugepage for each subpage, only do it when the subpage allocation crosses over into a different hugepage. (suggested by Gerd and DW)
Cc Gerd Hoffmann <[email protected]>Signed-off-by Vivek Kasireddy <[email protected]>Signed-off-by Dongwon Kim <[email protected]>--- drivers/dma-buf/udmabuf.c | 51 +++++++++++++++++++++++++++++++++------ 1 file changed, 44 insertions(+), 7 deletions(-)drivers/dma-buf/udmabuf.c +44
-7@@ -11,6 +11,7 @@#include <linux/shmem_fs.h> #include <linux/slab.h> #include <linux/udmabuf.h> +#include <linux/hugetlb.h> static const u32 list_limit =1024 ;/* udmabuf_create_list->count limit */ static const size_t size_limit_mb =64 ;/* total dmabuf size, in megabytes */ @@ -163,7 +164,9 @@ static long udmabuf_create(struct miscdevice *device,struct udmabuf *ubuf;struct dma_buf *buf;pgoff_t pgoff, pgcnt, pgidx, pgbuf =0 , pglimit; -struct page *page; +struct page *page, *hpage =NULL ; +pgoff_t subpgoff, maxsubpgs; +struct hstate *hpstate;int seals, ret = -EINVAL ;u32 i, flags;@@ -194,7 +197,8 @@ static long udmabuf_create(struct miscdevice *device, memfd =fget (list[i].memfd);if (!memfd) goto err; -if (!shmem_mapping (file_inode (memfd)->i_mapping)) +if (!shmem_mapping (file_inode (memfd)->i_mapping) && + !is_file_hugepages (memfd)) goto err; seals =memfd_fcntl (memfd,F_GET_SEALS ,0 );if (seals == -EINVAL )@@ -205,17 +209,50 @@ static long udmabuf_create(struct miscdevice *device, goto err; pgoff = list[i].offset >>PAGE_SHIFT ; pgcnt = list[i].size >>PAGE_SHIFT ; +if (is_file_hugepages (memfd)) { + hpstate =hstate_file (memfd); + pgoff = list[i].offset >>huge_page_shift (hpstate); + subpgoff = (list[i].offset & + ~huge_page_mask (hpstate)) >>PAGE_SHIFT ; + maxsubpgs =huge_page_size (hpstate) >>PAGE_SHIFT ; + }for (pgidx =0 ; pgidx < pgcnt; pgidx++) { - page =shmem_read_mapping_page ( -file_inode (memfd)->i_mapping, pgoff + pgidx); -if (IS_ERR (page)) { - ret =PTR_ERR (page); - goto err; +if (is_file_hugepages (memfd)) { +if (!hpage) { + hpage =find_get_page_flags ( +file_inode (memfd)->i_mapping, + pgoff,FGP_ACCESSED ); +if (IS_ERR (hpage)) { + ret =PTR_ERR (hpage); + goto err; + } + } + page = hpage + subpgoff; +get_page (page); + subpgoff++; +if (subpgoff == maxsubpgs) { +put_page (hpage); + hpage =NULL ; + subpgoff =0 ; + pgoff++; + } + }else { + page =shmem_read_mapping_page ( +file_inode (memfd)->i_mapping, + pgoff + pgidx); +if (IS_ERR (page)) { + ret =PTR_ERR (page); + goto err; + } } ubuf->pages[pgbuf++] = page; }fput (memfd); memfd =NULL ; +if (hpage) { +put_page (hpage); + hpage =NULL ; + } } exp_info.ops = &udmabuf_ops;Signature · 1 line
2.30.2
Gerd Hoffmann
maintainerRe: [PATCH] udmabuf: Add support for mapping hugepages (v3)
Quoted 4 lines · Dongwon Kim
I see the number of entries in the list often exceeds
list_limitcurrently hardcoded to 1024 for full HD scanout resource (== 1920*1080*4 bytes). Can we include a change to increase it to something like 4096 or higher in this patch?
Separate patch please (and maybe turn limits into module parameters while being at it).
take care,
Gerd
Gerd Hoffmann
maintainerRe: [PATCH] udmabuf: Add support for mapping hugepages (v3)
In reply to Vivek KasireddyHi,
drivers/dma-buf/udmabuf.c static long udmabuf_create(struct miscdevice *device, · line 222
Quoted 1 line
drivers/dma-buf/udmabuf.c +1
-0+ hpage = find_get_page_flags(
checkpatch doesn't like that:
-:76: CHECK:OPEN_ENDED_LINE: Lines should not end with a '('
drivers/dma-buf/udmabuf.c static long udmabuf_create(struct miscdevice *device, · line 240
Quoted 1 line
drivers/dma-buf/udmabuf.c +1
-0+ page = shmem_read_mapping_page(
Same here.
take care,
Gerd
Vivek Kasireddy
maintainerPATCH 1/1 udmabuf: Add support for mapping hugepages (v4)
If the VMM's (Qemu) memory backend is backed up by memfd + Hugepages (hugetlbfs and not THP), we have to first find the hugepage(s) where the Guest allocations are located and then extract the regular 4k sized subpages from them.
v2: Ensure that the subpage and hugepage offsets are calculated correctly when the range of subpage allocations cuts across multiple hugepages.
v3: Instead of repeatedly looking up the hugepage for each subpage, only do it when the subpage allocation crosses over into a different hugepage. (suggested by Gerd and DW)
v4: Fix the following warning identified by checkpatch:
CHECK:OPEN_ENDED_LINE: Lines should not end with a '('
--- drivers/dma-buf/udmabuf.c | 50 +++++++++++++++++++++++++++++++++------ 1 file changed, 43 insertions(+), 7 deletions(-)
drivers/dma-buf/udmabuf.c +43 -7
@@ -11,6 +11,7 @@#include <linux/shmem_fs.h> #include <linux/slab.h> #include <linux/udmabuf.h> +#include <linux/hugetlb.h> static const u32 list_limit =1024 ;/* udmabuf_create_list->count limit */ static const size_t size_limit_mb =64 ;/* total dmabuf size, in megabytes */
@@ -160,10 +161,13 @@ static long udmabuf_create(struct miscdevice *device, {DEFINE_DMA_BUF_EXPORT_INFO (exp_info);struct file *memfd =NULL ; +struct address_space *mapping =NULL ;struct udmabuf *ubuf;struct dma_buf *buf;pgoff_t pgoff, pgcnt, pgidx, pgbuf =0 , pglimit; -struct page *page; +struct page *page, *hpage =NULL ; +pgoff_t subpgoff, maxsubpgs; +struct hstate *hpstate;int seals, ret = -EINVAL ;u32 i, flags;
@@ -194,7 +198,8 @@ static long udmabuf_create(struct miscdevice *device, memfd =fget (list[i].memfd);if (!memfd) goto err; -if (!shmem_mapping (file_inode (memfd)->i_mapping)) + mapping =file_inode (memfd)->i_mapping; +if (!shmem_mapping (mapping) && !is_file_hugepages (memfd)) goto err; seals =memfd_fcntl (memfd,F_GET_SEALS ,0 );if (seals == -EINVAL )
@@ -205,17 +210,48 @@ static long udmabuf_create(struct miscdevice *device, goto err; pgoff = list[i].offset >>PAGE_SHIFT ; pgcnt = list[i].size >>PAGE_SHIFT ; +if (is_file_hugepages (memfd)) { + hpstate =hstate_file (memfd); + pgoff = list[i].offset >>huge_page_shift (hpstate); + subpgoff = (list[i].offset & + ~huge_page_mask (hpstate)) >>PAGE_SHIFT ; + maxsubpgs =huge_page_size (hpstate) >>PAGE_SHIFT ; + }for (pgidx =0 ; pgidx < pgcnt; pgidx++) { - page =shmem_read_mapping_page ( -file_inode (memfd)->i_mapping, pgoff + pgidx); -if (IS_ERR (page)) { - ret =PTR_ERR (page); - goto err; +if (is_file_hugepages (memfd)) { +if (!hpage) { + hpage =find_get_page_flags (mapping, + pgoff,FGP_ACCESSED ); +if (IS_ERR (hpage)) { + ret =PTR_ERR (hpage); + goto err; + } + } + page = hpage + subpgoff; +get_page (page); + subpgoff++; +if (subpgoff == maxsubpgs) { +put_page (hpage); + hpage =NULL ; + subpgoff =0 ; + pgoff++; + } + }else { + page =shmem_read_mapping_page (mapping, + pgoff + pgidx); +if (IS_ERR (page)) { + ret =PTR_ERR (page); + goto err; + } } ubuf->pages[pgbuf++] = page; }fput (memfd); memfd =NULL ; +if (hpage) { +put_page (hpage); + hpage =NULL ; + } } exp_info.ops = &udmabuf_ops;
Signature · 1 line
2.30.2
Gerd Hoffmann
maintainerRe: [PATCH] udmabuf: Add support for mapping hugepages (v4)
Quoted 19 lines · Vivek Kasireddy
If the VMM's (Qemu) memory backend is backed up by memfd + Hugepages (hugetlbfs and not THP), we have to first find the hugepage(s) where the Guest allocations are located and then extract the regular 4k sized subpages from them.
v2: Ensure that the subpage and hugepage offsets are calculated correctly when the range of subpage allocations cuts across multiple hugepages.
v3: Instead of repeatedly looking up the hugepage for each subpage, only do it when the subpage allocation crosses over into a different hugepage. (suggested by Gerd and DW)
v4: Fix the following warning identified by checkpatch:
CHECK:OPEN_ENDED_LINE: Lines should not end with a '('Cc Gerd Hoffmann <[email protected]>Signed-off-by Vivek Kasireddy <[email protected]>Signed-off-by Dongwon Kim <[email protected]>
drivers/dma-buf/udmabuf.c static long udmabuf_create(struct miscdevice *device, · line 224
Quoted 2 lines
drivers/dma-buf/udmabuf.c +2
-0+ hpage =find_get_page_flags (mapping, + pgoff,FGP_ACCESSED );
checkpatch is still not happy. Fixed that up & pushed to drm-misc-next.
thanks,
Gerd