On 12/11/19 1:03 PM, Linus Torvalds wrote:
> On Wed, Dec 11, 2019 at 11:34 AM Jens Axboe <axboe@xxxxxxxxx> wrote:
>>
>> I can't tell a difference in the results, there's no discernable
>> difference between NOT calling mark_page_accessed() or calling it.
>> Behavior seems about the same, in terms of pre and post page cache full,
>> and kswapd still churns a lot once the page cache is filled up.
>
> Yeah, that sounds like a bug. I'm sure the RWF_UNCACHED flag fixes it
> when you do the IO that way, but it seems to be a bug relardless.
Hard to disagree with that.
> Does /proc/meminfo have everything inactive for file data (ie the
> "Active(file)" line is basically zero?).
$ cat /proc/meminfo | grep -i active
Active: 134136 kB
Inactive: 28683916 kB
Active(anon): 97064 kB
Inactive(anon): 4 kB
Active(file): 37072 kB
Inactive(file): 28683912 kB
This is after a run with RWF_NOACCESS.
> Maybe pages got activated other ways (eg a problem with the workingset
> code)? You said "See patch below", but there wasn't any.
Oops, now below.
>
> That said, it's also entirely possible that even with everything in
> the inactive list, we might try to shrink other things first for
> whatever odd reason..
>
> The fact that you see that xas_create() so prominently would imply
> perhaps add_to_swap_cache(), which certainly implies that the page
> shrinking isn't hitting the file pages...
That's presumably misleading, as it's just lookups. But yes,
confusing...
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 5ea5fc167524..b2ecc66f5bd5 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -316,6 +316,7 @@ enum rw_hint {
#define IOCB_WRITE (1 << 6)
#define IOCB_NOWAIT (1 << 7)
#define IOCB_UNCACHED (1 << 8)
+#define IOCB_NOACCESS (1 << 9)
struct kiocb {
struct file *ki_filp;
@@ -3423,6 +3424,8 @@ static inline int kiocb_set_rw_flags(struct kiocb *ki, rwf_t flags)
ki->ki_flags |= IOCB_APPEND;
if (flags & RWF_UNCACHED)
ki->ki_flags |= IOCB_UNCACHED;
+ if (flags & RWF_NOACCESS)
+ ki->ki_flags |= IOCB_NOACCESS;
return 0;
}
diff --git a/include/uapi/linux/fs.h b/include/uapi/linux/fs.h
index 357ebb0e0c5d..f20f0048d5c5 100644
--- a/include/uapi/linux/fs.h
+++ b/include/uapi/linux/fs.h
@@ -302,8 +302,10 @@ typedef int __bitwise __kernel_rwf_t;
/* drop cache after reading or writing data */
#define RWF_UNCACHED ((__force __kernel_rwf_t)0x00000040)
+#define RWF_NOACCESS ((__force __kernel_rwf_t)0x00000080)
+
/* mask of flags supported by the kernel */
#define RWF_SUPPORTED (RWF_HIPRI | RWF_DSYNC | RWF_SYNC | RWF_NOWAIT |\
- RWF_APPEND | RWF_UNCACHED)
+ RWF_APPEND | RWF_UNCACHED | RWF_NOACCESS)
#endif /* _UAPI_LINUX_FS_H */
diff --git a/mm/filemap.c b/mm/filemap.c
index 4dadd1a4ca7c..c37b0e221a8a 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -2058,7 +2058,7 @@ static ssize_t generic_file_buffered_read(struct kiocb *iocb,
if (iocb->ki_flags & IOCB_NOWAIT)
goto would_block;
/* UNCACHED implies no read-ahead */
- if (iocb->ki_flags & IOCB_UNCACHED)
+ if (iocb->ki_flags & (IOCB_UNCACHED|IOCB_NOACCESS))
goto no_cached_page;
page_cache_sync_readahead(mapping,
ra, filp,
@@ -2144,7 +2144,8 @@ static ssize_t generic_file_buffered_read(struct kiocb *iocb,
* When a sequential read accesses a page several times,
* only mark it as accessed the first time.
*/
- if (prev_index != index || offset != prev_offset)
+ if ((prev_index != index || offset != prev_offset) &&
+ !(iocb->ki_flags & IOCB_NOACCESS))
mark_page_accessed(page);
prev_index = index;
--
Jens Axboe
