Hi, Svetly, Thanks for patch! Svetly Todorov <svetly.todorov@xxxxxxxxxxxx> writes: > --- > Adding documentation for the new MPOL_WEIGHTED_INTERLEAVE mode in the same > manpages that mention MPOL_INTERLEAVE; namely, mbind(2), set_mempolicy(2), > and get_mempolicy(2). > > Descriptions were based on the changes introduced in this patch: > https://lore.kernel.org/all/20240202170238.90004-4-gregory.price@xxxxxxxxxxxx/ > > Which was upstreamed to 6.9 here: > https://lore.kernel.org/linux-mm/20240313200532.34e4cff216acd3db8def4637@xxxxxxxxxxxxxxxxxxxx/ > > Signed-off-by: Svetly Todorov <svetly.todorov@xxxxxxxxxxxx> > --- > man2/get_mempolicy.2 | 8 ++++++-- > man2/mbind.2 | 13 +++++++++++++ > man2/set_mempolicy.2 | 13 +++++++++++++ > 3 files changed, 32 insertions(+), 2 deletions(-) > > diff --git a/man2/get_mempolicy.2 b/man2/get_mempolicy.2 > index 5248f04ba..03550331d 100644 > --- a/man2/get_mempolicy.2 > +++ b/man2/get_mempolicy.2 > @@ -137,7 +137,9 @@ specifies > but not > .BR MPOL_F_ADDR , > and the thread's current policy is > -.BR MPOL_INTERLEAVE , > +.BR MPOL_INTERLEAVE > +or > +.BR MPOL_WEIGHTED_INTERLEAVE , > then > .BR get_mempolicy () > will return in the location pointed to by a non-NULL > @@ -206,7 +208,9 @@ specified > but not > .B MPOL_F_ADDR > and the current thread policy is not > -.BR MPOL_INTERLEAVE . > +.BR MPOL_INTERLEAVE > +or > +.BR MPOL_WEIGHTED_INTERLEAVE . > Or, > .I flags > specified > diff --git a/man2/mbind.2 b/man2/mbind.2 > index b0e961f9c..cc994a01e 100644 > --- a/man2/mbind.2 > +++ b/man2/mbind.2 > @@ -105,6 +105,7 @@ argument must specify one of > .BR MPOL_DEFAULT , > .BR MPOL_BIND , > .BR MPOL_INTERLEAVE , > +.BR MPOL_WEIGHTED_INTERLEAVE , > .BR MPOL_PREFERRED , > or > .B MPOL_LOCAL > @@ -243,6 +244,18 @@ at least 1\ MB or bigger with a fairly uniform access pattern. > Accesses to a single page of the area will still be limited to > the memory bandwidth of a single node. > .TP > +.B MPOL_WEIGHTED_INTERLEAVE > +This mode interleaves page allocations across the nodes specified in > +.I nodemask > +according to the weights in /sys/kernel/mm/mempolicy/weighted_interleave. > +That is, if bits 0, 2, and 5 are raised in > +.IR nodemask , > +and the contents of /sys/kernel/mm/mempolicy/weighted_interleave/node0, > +/sys/.../node2, and /sys/.../node5 are 4, 7, and 9, respectively, then > +the first 4 pages in this region will be allocated on node 0; > +the next 7 will be on node 2; and the next 9 on node 5. > +Subsequent allocations will continue in this pattern. I prefer to describe the algorithm in more general words instead of reflecting the current implementation. Because we may change the implementation in the future. May be something like, the proportion of the number of pages allocated in node 0, 2, 5 will be 4:7:9, respectively. -- Best Regards, Huang, Ying > +.TP > .B MPOL_PREFERRED > This mode sets the preferred node for allocation. > The kernel will try to allocate pages from this > diff --git a/man2/set_mempolicy.2 b/man2/set_mempolicy.2 > index fc3ad9df8..a1b11b307 100644 > --- a/man2/set_mempolicy.2 > +++ b/man2/set_mempolicy.2 > @@ -63,6 +63,7 @@ argument must specify one of > .BR MPOL_DEFAULT , > .BR MPOL_BIND , > .BR MPOL_INTERLEAVE , > +.BR MPOL_WEIGHTED_INTERLEAVE , > .BR MPOL_PREFERRED , > or > .B MPOL_LOCAL > @@ -199,6 +200,18 @@ the memory bandwidth of a single node. > .\" To be effective the memory area should be fairly large, > .\" at least 1 MB or bigger. > .TP > +.B MPOL_WEIGHTED_INTERLEAVE > +This mode interleaves page allocations across the nodes specified in > +.I nodemask > +according to the weights in /sys/kernel/mm/mempolicy/weighted_interleave. > +That is, if bits 0, 2, and 5 are raised in > +.IR nodemask , > +and the contents of /sys/kernel/mm/mempolicy/weighted_interleave/node0, > +/sys/.../node2, and /sys/.../node5 are 4, 7, and 9, respectively, then > +the first 4 pages in this region will be allocated on node 0; > +the next 7 will be on node 2; and the next 9 on node 5. > +Subsequent allocations will continue in this pattern. > +.TP > .B MPOL_PREFERRED > This mode sets the preferred node for allocation. > The kernel will try to allocate pages from this node first > > --- > base-commit: a4ea5f76215dcf1d8d017eb65ab12919ccf09f3f > change-id: 20240313-weighted_interleave-e8141ed754f9 > > Best regards,
