On 2024/8/27 0:46, Alexander Duyck wrote:
> On Mon, Aug 26, 2024 at 5:46 AM Yunsheng Lin <linyunsheng@xxxxxxxxxx> wrote:
>>
>> Currently there is one 'struct page_frag' for every 'struct
>> sock' and 'struct task_struct', we are about to replace the
>> 'struct page_frag' with 'struct page_frag_cache' for them.
>> Before begin the replacing, we need to ensure the size of
>> 'struct page_frag_cache' is not bigger than the size of
>> 'struct page_frag', as there may be tens of thousands of
>> 'struct sock' and 'struct task_struct' instances in the
>> system.
>>
>> By or'ing the page order & pfmemalloc with lower bits of
>> 'va' instead of using 'u16' or 'u32' for page size and 'u8'
>> for pfmemalloc, we are able to avoid 3 or 5 bytes space waste.
>> And page address & pfmemalloc & order is unchanged for the
>> same page in the same 'page_frag_cache' instance, it makes
>> sense to fit them together.
>>
>> After this patch, the size of 'struct page_frag_cache' should be
>> the same as the size of 'struct page_frag'.
>>
>> CC: Alexander Duyck <alexander.duyck@xxxxxxxxx>
>> Signed-off-by: Yunsheng Lin <linyunsheng@xxxxxxxxxx>
>> ---
>> include/linux/mm_types_task.h | 19 ++++++-----
>> include/linux/page_frag_cache.h | 60 +++++++++++++++++++++++++++++++--
>> mm/page_frag_cache.c | 51 +++++++++++++++-------------
>> 3 files changed, 97 insertions(+), 33 deletions(-)
>>
>> diff --git a/include/linux/mm_types_task.h b/include/linux/mm_types_task.h
>> index cdc1e3696439..a8635460e027 100644
>> --- a/include/linux/mm_types_task.h
>> +++ b/include/linux/mm_types_task.h
>> @@ -50,18 +50,21 @@ struct page_frag {
>> #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
>> #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
>> struct page_frag_cache {
>> - void *va;
>> -#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
>> + /* encoded_page consists of the virtual address, pfmemalloc bit and order
>> + * of a page.
>> + */
>> + unsigned long encoded_page;
>> +
>> + /* we maintain a pagecount bias, so that we dont dirty cache line
>> + * containing page->_refcount every time we allocate a fragment.
>> + */
>> +#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) && (BITS_PER_LONG <= 32)
>> __u16 offset;
>> - __u16 size;
>> + __u16 pagecnt_bias;
>> #else
>> __u32 offset;
>> + __u32 pagecnt_bias;
>> #endif
>> - /* we maintain a pagecount bias, so that we dont dirty cache line
>> - * containing page->_refcount every time we allocate a fragment.
>> - */
>> - unsigned int pagecnt_bias;
>> - bool pfmemalloc;
>> };
>>
>> /* Track pages that require TLB flushes */
>> diff --git a/include/linux/page_frag_cache.h b/include/linux/page_frag_cache.h
>> index 0a52f7a179c8..372d6ed7e20a 100644
>> --- a/include/linux/page_frag_cache.h
>> +++ b/include/linux/page_frag_cache.h
>> @@ -3,18 +3,74 @@
>> #ifndef _LINUX_PAGE_FRAG_CACHE_H
>> #define _LINUX_PAGE_FRAG_CACHE_H
>>
>> +#include <linux/bits.h>
>> +#include <linux/build_bug.h>
>> #include <linux/log2.h>
>> +#include <linux/mm.h>
>> #include <linux/mm_types_task.h>
>> #include <linux/types.h>
>>
>> +#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
>> +/* Use a full byte here to enable assembler optimization as the shift
>> + * operation is usually expecting a byte.
>> + */
>> +#define PAGE_FRAG_CACHE_ORDER_MASK GENMASK(7, 0)
>> +#define PAGE_FRAG_CACHE_PFMEMALLOC_SHIFT 8
>> +#define PAGE_FRAG_CACHE_PFMEMALLOC_BIT BIT(PAGE_FRAG_CACHE_PFMEMALLOC_SHIFT)
>> +#else
>> +/* Compiler should be able to figure out we don't read things as any value
>> + * ANDed with 0 is 0.
>> + */
>> +#define PAGE_FRAG_CACHE_ORDER_MASK 0
>> +#define PAGE_FRAG_CACHE_PFMEMALLOC_SHIFT 0
>> +#define PAGE_FRAG_CACHE_PFMEMALLOC_BIT BIT(PAGE_FRAG_CACHE_PFMEMALLOC_SHIFT)
>> +#endif
>> +
>> +static inline unsigned long page_frag_encode_page(struct page *page,
>> + unsigned int order,
>> + bool pfmemalloc)
>> +{
>> + BUILD_BUG_ON(PAGE_FRAG_CACHE_MAX_ORDER > PAGE_FRAG_CACHE_ORDER_MASK);
>> + BUILD_BUG_ON(PAGE_FRAG_CACHE_PFMEMALLOC_BIT >= PAGE_SIZE);
>> +
>> + return (unsigned long)page_address(page) |
>> + (order & PAGE_FRAG_CACHE_ORDER_MASK) |
>> + ((unsigned long)pfmemalloc << PAGE_FRAG_CACHE_PFMEMALLOC_SHIFT);
>> +}
>> +
>> +static inline unsigned long page_frag_encoded_page_order(unsigned long encoded_page)
>> +{
>> + return encoded_page & PAGE_FRAG_CACHE_ORDER_MASK;
>> +}
>> +
>> +static inline bool page_frag_encoded_page_pfmemalloc(unsigned long encoded_page)
>> +{
>> + return !!(encoded_page & PAGE_FRAG_CACHE_PFMEMALLOC_BIT);
>> +}
>> +
>> +static inline void *page_frag_encoded_page_address(unsigned long encoded_page)
>> +{
>> + return (void *)(encoded_page & PAGE_MASK);
>> +}
>> +
>> +static inline struct page *page_frag_encoded_page_ptr(unsigned long encoded_page)
>> +{
>> + return virt_to_page((void *)encoded_page);
>> +}
>> +
>> static inline void page_frag_cache_init(struct page_frag_cache *nc)
>> {
>> - nc->va = NULL;
>> + nc->encoded_page = 0;
>> }
>>
>> static inline bool page_frag_cache_is_pfmemalloc(struct page_frag_cache *nc)
>> {
>> - return !!nc->pfmemalloc;
>> + return page_frag_encoded_page_pfmemalloc(nc->encoded_page);
>> +}
>> +
>> +static inline unsigned int page_frag_cache_page_size(unsigned long encoded_page)
>> +{
>> + return PAGE_SIZE << page_frag_encoded_page_order(encoded_page);
>> }
>>
>> void page_frag_cache_drain(struct page_frag_cache *nc);
>
> So how many of these additions are actually needed outside of the
> page_frag_cache.c file? I'm just wondering if we could look at moving
At least page_frag_cache_is_pfmemalloc(), page_frag_encoded_page_order(),
page_frag_encoded_page_ptr(), page_frag_encoded_page_address() are needed
out of the page_frag_cache.c file for now, which are used mostly in
__page_frag_cache_commit() and __page_frag_alloc_refill_probe_align() for
debugging and performance reason, see patch 7 & 10.
The only left one is page_frag_encode_page(), I am not sure if it makes
much sense to move it to page_frag_cache.c while the rest of them are in
.h file.
> most of these into the c file itself instead of making them accessible
> to all callers as I don't believe we currently have anyone looking
> into the size of the frag cache or anything like that and I would
> prefer to avoid exposing such functionality if possible. As the
> non-order0 allocation problem with this has pointed out people will
> exploit any interface exposed even if unintentionally.
>
> I would want to move the size/order logic as well as splitting out the
> virtual address as we shouldn't be allowing the user to look at that
> without going through an allocation function.
I am generally agreed with the above argument if there are ways to do
that without sacrificing the above mentioned debugging and performance.
