| Message ID | 20220413032915.251254-2-liupeng256@huawei.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | hugetlb: Fix some incorrect behavior | expand |
On Wed, 13 Apr 2022 03:29:12 +0000 Peng Liu <liupeng256@huawei.com> wrote: > Certain systems are designed to have sparse/discontiguous nodes. In > this case, nr_online_nodes can not be used to walk through numa node. > Also, a valid node may be greater than nr_online_nodes. > > However, in hugetlb, it is assumed that nodes are contiguous. Recheck > all the places that use nr_online_nodes, and repair them one by one. > What are the runtime effects of this shortcoming?
On 2022/4/13 12:42, Andrew Morton wrote: > On Wed, 13 Apr 2022 03:29:12 +0000 Peng Liu<liupeng256@huawei.com> wrote: > >> Certain systems are designed to have sparse/discontiguous nodes. In >> this case, nr_online_nodes can not be used to walk through numa node. >> Also, a valid node may be greater than nr_online_nodes. >> >> However, in hugetlb, it is assumed that nodes are contiguous. Recheck >> all the places that use nr_online_nodes, and repair them one by one. >> > What are the runtime effects of this shortcoming? > . For sparse/discontiguous nodes, the current code may treat a valid node as invalid, and will fail to allocate all hugepages on a valid node that "nid >= nr_online_nodes". As David suggested: if (tmp >= nr_online_nodes) goto invalid; Just imagine node 0 and node 2 are online, and node 1 is offline. Assuming that "node < 2" is valid is wrong.
On 4/13/2022 11:29 AM, Peng Liu wrote: > Certain systems are designed to have sparse/discontiguous nodes. In > this case, nr_online_nodes can not be used to walk through numa node. > Also, a valid node may be greater than nr_online_nodes. > > However, in hugetlb, it is assumed that nodes are contiguous. Recheck > all the places that use nr_online_nodes, and repair them one by one. > > Suggested-by: David Hildenbrand <david@redhat.com> > Fixes: 4178158ef8ca ("hugetlbfs: fix issue of preallocation of gigantic pages can't work") > Fixes: b5389086ad7b ("hugetlbfs: extend the definition of hugepages parameter to support node allocation") > Fixes: e79ce9832316 ("hugetlbfs: fix a truncation issue in hugepages parameter") > Fixes: f9317f77a6e0 ("hugetlb: clean up potential spectre issue warnings") > Signed-off-by: Peng Liu <liupeng256@huawei.com> LGTM. Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
On Wed, 13 Apr 2022 14:27:54 +0800 "liupeng (DM)" <liupeng256@huawei.com> wrote: > > On 2022/4/13 12:42, Andrew Morton wrote: > > On Wed, 13 Apr 2022 03:29:12 +0000 Peng Liu<liupeng256@huawei.com> wrote: > > > >> Certain systems are designed to have sparse/discontiguous nodes. In > >> this case, nr_online_nodes can not be used to walk through numa node. > >> Also, a valid node may be greater than nr_online_nodes. > >> > >> However, in hugetlb, it is assumed that nodes are contiguous. Recheck > >> all the places that use nr_online_nodes, and repair them one by one. > >> > > What are the runtime effects of this shortcoming? > > . > > For sparse/discontiguous nodes, the current code may treat a valid node > as invalid, and will fail to allocate all hugepages on a valid node that > "nid >= nr_online_nodes". > > As David suggested: > if (tmp >= nr_online_nodes) > goto invalid; > > Just imagine node 0 and node 2 are online, and node 1 is offline. Assuming > that "node < 2" is valid is wrong. So do you think we should backport thtis fix into earlier kernel releases?
On 2022/4/14 6:04, Andrew Morton wrote: > On Wed, 13 Apr 2022 14:27:54 +0800 "liupeng (DM)"<liupeng256@huawei.com> wrote: > >> On 2022/4/13 12:42, Andrew Morton wrote: >>> On Wed, 13 Apr 2022 03:29:12 +0000 Peng Liu<liupeng256@huawei.com> wrote: >>> >>>> Certain systems are designed to have sparse/discontiguous nodes. In >>>> this case, nr_online_nodes can not be used to walk through numa node. >>>> Also, a valid node may be greater than nr_online_nodes. >>>> >>>> However, in hugetlb, it is assumed that nodes are contiguous. Recheck >>>> all the places that use nr_online_nodes, and repair them one by one. >>>> >>> What are the runtime effects of this shortcoming? >>> . >> For sparse/discontiguous nodes, the current code may treat a valid node >> as invalid, and will fail to allocate all hugepages on a valid node that >> "nid >= nr_online_nodes". >> >> As David suggested: >> if (tmp >= nr_online_nodes) >> goto invalid; >> >> Just imagine node 0 and node 2 are online, and node 1 is offline. Assuming >> that "node < 2" is valid is wrong. > So do you think we should backport thtis fix into earlier kernel releases? > . I think it is not an urgent bug, because: 1) Qemu does not support sparse node so far, although there are some sparse-node issues to make qemu support sparse node. 2) I don't find an actual normal machine that reports sparse-node and need to use hugepages so far.
On 4/12/22 20:29, Peng Liu wrote: > Certain systems are designed to have sparse/discontiguous nodes. In > this case, nr_online_nodes can not be used to walk through numa node. > Also, a valid node may be greater than nr_online_nodes. > > However, in hugetlb, it is assumed that nodes are contiguous. Recheck > all the places that use nr_online_nodes, and repair them one by one. > > Suggested-by: David Hildenbrand <david@redhat.com> > Fixes: 4178158ef8ca ("hugetlbfs: fix issue of preallocation of gigantic pages can't work") > Fixes: b5389086ad7b ("hugetlbfs: extend the definition of hugepages parameter to support node allocation") > Fixes: e79ce9832316 ("hugetlbfs: fix a truncation issue in hugepages parameter") > Fixes: f9317f77a6e0 ("hugetlb: clean up potential spectre issue warnings") > Signed-off-by: Peng Liu <liupeng256@huawei.com> > --- > mm/hugetlb.c | 12 ++++++------ > 1 file changed, 6 insertions(+), 6 deletions(-) Thank you! I am guessing that at one time nodes were contiguous at least at boot time. When that changed, hugetlb was not updated. :( Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
On Wed, 13 Apr 2022, Peng Liu wrote: >Certain systems are designed to have sparse/discontiguous nodes. In >this case, nr_online_nodes can not be used to walk through numa node. >Also, a valid node may be greater than nr_online_nodes. > >However, in hugetlb, it is assumed that nodes are contiguous. Recheck >all the places that use nr_online_nodes, and repair them one by one. > >Suggested-by: David Hildenbrand <david@redhat.com> >Fixes: 4178158ef8ca ("hugetlbfs: fix issue of preallocation of gigantic pages can't work") >Fixes: b5389086ad7b ("hugetlbfs: extend the definition of hugepages parameter to support node allocation") >Fixes: e79ce9832316 ("hugetlbfs: fix a truncation issue in hugepages parameter") >Fixes: f9317f77a6e0 ("hugetlb: clean up potential spectre issue warnings") >Signed-off-by: Peng Liu <liupeng256@huawei.com> >Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com> >Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com> Reviewed-by: Davidlohr Bueso <dave@stgolabs.net> ... but >--- > mm/hugetlb.c | 12 ++++++------ > 1 file changed, 6 insertions(+), 6 deletions(-) > >diff --git a/mm/hugetlb.c b/mm/hugetlb.c >index b34f50156f7e..5b5a2a5a742f 100644 >--- a/mm/hugetlb.c >+++ b/mm/hugetlb.c >@@ -2979,7 +2979,7 @@ int __alloc_bootmem_huge_page(struct hstate *h, int nid) > struct huge_bootmem_page *m = NULL; /* initialize for clang */ > int nr_nodes, node; > >- if (nid != NUMA_NO_NODE && nid >= nr_online_nodes) >+ if (nid != NUMA_NO_NODE && !node_online(nid)) afaict null_blk could also use this, actually the whole thing wants a helper - node_valid()?
On 2022/4/15 10:09, Davidlohr Bueso wrote: > On Wed, 13 Apr 2022, Peng Liu wrote: > >> Certain systems are designed to have sparse/discontiguous nodes. In >> this case, nr_online_nodes can not be used to walk through numa node. >> Also, a valid node may be greater than nr_online_nodes. >> >> However, in hugetlb, it is assumed that nodes are contiguous. Recheck >> all the places that use nr_online_nodes, and repair them one by one. >> >> Suggested-by: David Hildenbrand <david@redhat.com> >> Fixes: 4178158ef8ca ("hugetlbfs: fix issue of preallocation of >> gigantic pages can't work") >> Fixes: b5389086ad7b ("hugetlbfs: extend the definition of hugepages >> parameter to support node allocation") >> Fixes: e79ce9832316 ("hugetlbfs: fix a truncation issue in hugepages >> parameter") >> Fixes: f9317f77a6e0 ("hugetlb: clean up potential spectre issue >> warnings") >> Signed-off-by: Peng Liu <liupeng256@huawei.com> >> Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com> >> Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com> > > Reviewed-by: Davidlohr Bueso <dave@stgolabs.net> > > ... but > >> --- >> mm/hugetlb.c | 12 ++++++------ >> 1 file changed, 6 insertions(+), 6 deletions(-) >> >> diff --git a/mm/hugetlb.c b/mm/hugetlb.c >> index b34f50156f7e..5b5a2a5a742f 100644 >> --- a/mm/hugetlb.c >> +++ b/mm/hugetlb.c >> @@ -2979,7 +2979,7 @@ int __alloc_bootmem_huge_page(struct hstate *h, >> int nid) >> struct huge_bootmem_page *m = NULL; /* initialize for clang */ >> int nr_nodes, node; >> >> - if (nid != NUMA_NO_NODE && nid >= nr_online_nodes) >> + if (nid != NUMA_NO_NODE && !node_online(nid)) > > afaict null_blk could also use this, actually the whole thing wants a > helper - node_valid()? > This one should be unnecessary, and this patch looks has a bug, if a very nid passed to node_online(), it may crash, could you re-check it, see my changes below, 1) add tmp check against MAX_NUMNODES before node_online() check, and move it after get tmp in hugepages_setup() , this could cover both per-node alloc and normal alloc 2) due to for_each_online_node() usage, we can drop additional check of nid in __alloc_bootmem_huge_page() $ git diff diff --git a/mm/hugetlb.c b/mm/hugetlb.c index fb5a549169ce..5a3ddec181a0 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -2986,8 +2986,6 @@ int __alloc_bootmem_huge_page(struct hstate *h, int nid) struct huge_bootmem_page *m = NULL; /* initialize for clang */ int nr_nodes, node; - if (nid != NUMA_NO_NODE && nid >= nr_online_nodes) - return 0; /* do node specific alloc */ if (nid != NUMA_NO_NODE) { m = memblock_alloc_try_nid_raw(huge_page_size(h), huge_page_size(h), @@ -3095,7 +3093,7 @@ static void __init hugetlb_hstate_alloc_pages(struct hstate *h) } /* do node specific alloc */ - for (i = 0; i < nr_online_nodes; i++) { + for_each_online_node(i) { if (h->max_huge_pages_node[i] > 0) { hugetlb_hstate_alloc_pages_onenode(h, i); node_specific_alloc = true; @@ -4059,7 +4057,7 @@ static int __init hugetlb_init(void) default_hstate.max_huge_pages = default_hstate_max_huge_pages; - for (i = 0; i < nr_online_nodes; i++) + for_each_online_node(i) default_hstate.max_huge_pages_node[i] = default_hugepages_in_node[i]; } @@ -4168,15 +4166,15 @@ static int __init hugepages_setup(char *s) count = 0; if (sscanf(p, "%lu%n", &tmp, &count) != 1) goto invalid; + if (tmp > MAX_NUMNODES || !node_online(tmp)) + goto invalid; /* Parameter is node format */ if (p[count] == ':') { if (!hugetlb_node_alloc_supported()) { pr_warn("HugeTLB: architecture can't support node specific alloc, ignoring!\n"); return 0; } - if (tmp >= nr_online_nodes) - goto invalid; - node = array_index_nospec(tmp, nr_online_nodes); + node = array_index_nospec(tmp, MAX_NUMNODES); p += count + 1; /* Parse hugepages */ if (sscanf(p, "%lu%n", &tmp, &count) != 1) @@ -4304,7 +4302,7 @@ static int __init default_hugepagesz_setup(char *s) */ if (default_hstate_max_huge_pages) { default_hstate.max_huge_pages = default_hstate_max_huge_pages; - for (i = 0; i < nr_online_nodes; i++) + for_each_online_node(i) default_hstate.max_huge_pages_node[i] = default_hugepages_in_node[i]; if (hstate_is_gigantic(&default_hstate)) > .
On 2022/4/15 13:41, Kefeng Wang wrote: > > On 2022/4/15 10:09, Davidlohr Bueso wrote: >> On Wed, 13 Apr 2022, Peng Liu wrote: >> >>> Certain systems are designed to have sparse/discontiguous nodes. In >>> this case, nr_online_nodes can not be used to walk through numa node. >>> Also, a valid node may be greater than nr_online_nodes. >>> >>> However, in hugetlb, it is assumed that nodes are contiguous. Recheck >>> all the places that use nr_online_nodes, and repair them one by one. >>> >>> Suggested-by: David Hildenbrand <david@redhat.com> >>> Fixes: 4178158ef8ca ("hugetlbfs: fix issue of preallocation of >>> gigantic pages can't work") >>> Fixes: b5389086ad7b ("hugetlbfs: extend the definition of hugepages >>> parameter to support node allocation") >>> Fixes: e79ce9832316 ("hugetlbfs: fix a truncation issue in hugepages >>> parameter") >>> Fixes: f9317f77a6e0 ("hugetlb: clean up potential spectre issue >>> warnings") >>> Signed-off-by: Peng Liu <liupeng256@huawei.com> >>> Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com> >>> Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com> >> >> Reviewed-by: Davidlohr Bueso <dave@stgolabs.net> >> >> ... but >> >>> --- >>> mm/hugetlb.c | 12 ++++++------ >>> 1 file changed, 6 insertions(+), 6 deletions(-) >>> >>> diff --git a/mm/hugetlb.c b/mm/hugetlb.c >>> index b34f50156f7e..5b5a2a5a742f 100644 >>> --- a/mm/hugetlb.c >>> +++ b/mm/hugetlb.c >>> @@ -2979,7 +2979,7 @@ int __alloc_bootmem_huge_page(struct hstate >>> *h, int nid) >>> struct huge_bootmem_page *m = NULL; /* initialize for clang */ >>> int nr_nodes, node; >>> >>> - if (nid != NUMA_NO_NODE && nid >= nr_online_nodes) >>> + if (nid != NUMA_NO_NODE && !node_online(nid)) >> >> afaict null_blk could also use this, actually the whole thing wants a >> helper - node_valid()? >> > This one should be unnecessary, and this patch looks has a bug, > > if a very nid passed to node_online(), it may crash, could you > re-check it, > > see my changes below, > > 1) add tmp check against MAX_NUMNODES before node_online() check, > > and move it after get tmp in hugepages_setup() , this could cover > both per-node alloc and normal alloc > > 2) due to for_each_online_node() usage, we can drop additional check > of nid in __alloc_bootmem_huge_page() > > > $ git diff > diff --git a/mm/hugetlb.c b/mm/hugetlb.c > index fb5a549169ce..5a3ddec181a0 100644 > --- a/mm/hugetlb.c > +++ b/mm/hugetlb.c > @@ -2986,8 +2986,6 @@ int __alloc_bootmem_huge_page(struct hstate *h, > int nid) > struct huge_bootmem_page *m = NULL; /* initialize for clang */ > int nr_nodes, node; > > - if (nid != NUMA_NO_NODE && nid >= nr_online_nodes) > - return 0; > /* do node specific alloc */ > if (nid != NUMA_NO_NODE) { > m = memblock_alloc_try_nid_raw(huge_page_size(h), > huge_page_size(h), > @@ -3095,7 +3093,7 @@ static void __init > hugetlb_hstate_alloc_pages(struct hstate *h) > } > > /* do node specific alloc */ > - for (i = 0; i < nr_online_nodes; i++) { > + for_each_online_node(i) { > if (h->max_huge_pages_node[i] > 0) { > hugetlb_hstate_alloc_pages_onenode(h, i); > node_specific_alloc = true; > @@ -4059,7 +4057,7 @@ static int __init hugetlb_init(void) > default_hstate.max_huge_pages = > default_hstate_max_huge_pages; > > - for (i = 0; i < nr_online_nodes; i++) > + for_each_online_node(i) > default_hstate.max_huge_pages_node[i] = > default_hugepages_in_node[i]; > } > @@ -4168,15 +4166,15 @@ static int __init hugepages_setup(char *s) > count = 0; > if (sscanf(p, "%lu%n", &tmp, &count) != 1) > goto invalid; > + if (tmp > MAX_NUMNODES || !node_online(tmp)) > + goto invalid; > /* Parameter is node format */ > if (p[count] == ':') { > if (!hugetlb_node_alloc_supported()) { > pr_warn("HugeTLB: architecture can't > support node specific alloc, ignoring!\n"); > return 0; > } > - if (tmp >= nr_online_nodes) > - goto invalid; > - node = array_index_nospec(tmp, nr_online_nodes); > + node = array_index_nospec(tmp, MAX_NUMNODES); > p += count + 1; > /* Parse hugepages */ > if (sscanf(p, "%lu%n", &tmp, &count) != 1) > @@ -4304,7 +4302,7 @@ static int __init default_hugepagesz_setup(char *s) > */ > if (default_hstate_max_huge_pages) { > default_hstate.max_huge_pages = > default_hstate_max_huge_pages; > - for (i = 0; i < nr_online_nodes; i++) > + for_each_online_node(i) > default_hstate.max_huge_pages_node[i] = > default_hugepages_in_node[i]; > if (hstate_is_gigantic(&default_hstate)) > > Yes, node_online is not a safe function which will cause panic if a very big number nid is received. So, this patch needs to be modified. Thanks.
On 2022/4/15 13:41, Kefeng Wang wrote: > > On 2022/4/15 10:09, Davidlohr Bueso wrote: >> On Wed, 13 Apr 2022, Peng Liu wrote: >> >>> Certain systems are designed to have sparse/discontiguous nodes. In >>> this case, nr_online_nodes can not be used to walk through numa node. >>> Also, a valid node may be greater than nr_online_nodes. >>> >>> However, in hugetlb, it is assumed that nodes are contiguous. Recheck >>> all the places that use nr_online_nodes, and repair them one by one. >>> >>> Suggested-by: David Hildenbrand <david@redhat.com> >>> Fixes: 4178158ef8ca ("hugetlbfs: fix issue of preallocation of >>> gigantic pages can't work") >>> Fixes: b5389086ad7b ("hugetlbfs: extend the definition of hugepages >>> parameter to support node allocation") >>> Fixes: e79ce9832316 ("hugetlbfs: fix a truncation issue in hugepages >>> parameter") >>> Fixes: f9317f77a6e0 ("hugetlb: clean up potential spectre issue >>> warnings") >>> Signed-off-by: Peng Liu <liupeng256@huawei.com> >>> Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com> >>> Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com> >> >> Reviewed-by: Davidlohr Bueso <dave@stgolabs.net> >> >> ... but >> >>> --- >>> mm/hugetlb.c | 12 ++++++------ >>> 1 file changed, 6 insertions(+), 6 deletions(-) >>> >>> diff --git a/mm/hugetlb.c b/mm/hugetlb.c >>> index b34f50156f7e..5b5a2a5a742f 100644 >>> --- a/mm/hugetlb.c >>> +++ b/mm/hugetlb.c >>> @@ -2979,7 +2979,7 @@ int __alloc_bootmem_huge_page(struct hstate >>> *h, int nid) >>> struct huge_bootmem_page *m = NULL; /* initialize for clang */ >>> int nr_nodes, node; >>> >>> - if (nid != NUMA_NO_NODE && nid >= nr_online_nodes) >>> + if (nid != NUMA_NO_NODE && !node_online(nid)) >> >> afaict null_blk could also use this, actually the whole thing wants a >> helper - node_valid()? >> > This one should be unnecessary, and this patch looks has a bug, > > if a very nid passed to node_online(), it may crash, could you > re-check it, > > see my changes below, > > 1) add tmp check against MAX_NUMNODES before node_online() check, > > and move it after get tmp in hugepages_setup() , this could cover > both per-node alloc and normal alloc sorry，for normal alloc, tmp is the number of huge pages, we don't need the movement, only add tmp >= MAX_NUMNODES is ok > > 2) due to for_each_online_node() usage, we can drop additional check > of nid in __alloc_bootmem_huge_page() >
On Sat, 16 Apr 2022 09:21:45 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > > On 2022/4/15 13:41, Kefeng Wang wrote: > > > > On 2022/4/15 10:09, Davidlohr Bueso wrote: > >> On Wed, 13 Apr 2022, Peng Liu wrote: > >> > >>> Certain systems are designed to have sparse/discontiguous nodes. In > >>> this case, nr_online_nodes can not be used to walk through numa node. > >>> Also, a valid node may be greater than nr_online_nodes. > >>> > >>> However, in hugetlb, it is assumed that nodes are contiguous. Recheck > >>> all the places that use nr_online_nodes, and repair them one by one. > >>> > >>> Suggested-by: David Hildenbrand <david@redhat.com> > >>> Fixes: 4178158ef8ca ("hugetlbfs: fix issue of preallocation of > >>> gigantic pages can't work") > >>> Fixes: b5389086ad7b ("hugetlbfs: extend the definition of hugepages > >>> parameter to support node allocation") > >>> Fixes: e79ce9832316 ("hugetlbfs: fix a truncation issue in hugepages > >>> parameter") > >>> Fixes: f9317f77a6e0 ("hugetlb: clean up potential spectre issue > >>> warnings") > >>> Signed-off-by: Peng Liu <liupeng256@huawei.com> > >>> Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com> > >>> Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com> > >> > >> Reviewed-by: Davidlohr Bueso <dave@stgolabs.net> > >> > >> ... but > >> > >>> --- > >>> mm/hugetlb.c | 12 ++++++------ > >>> 1 file changed, 6 insertions(+), 6 deletions(-) > >>> > >>> diff --git a/mm/hugetlb.c b/mm/hugetlb.c > >>> index b34f50156f7e..5b5a2a5a742f 100644 > >>> --- a/mm/hugetlb.c > >>> +++ b/mm/hugetlb.c > >>> @@ -2979,7 +2979,7 @@ int __alloc_bootmem_huge_page(struct hstate > >>> *h, int nid) > >>> struct huge_bootmem_page *m = NULL; /* initialize for clang */ > >>> int nr_nodes, node; > >>> > >>> - if (nid != NUMA_NO_NODE && nid >= nr_online_nodes) > >>> + if (nid != NUMA_NO_NODE && !node_online(nid)) > >> > >> afaict null_blk could also use this, actually the whole thing wants a > >> helper - node_valid()? > >> > > This one should be unnecessary, and this patch looks has a bug, > > > > if a very nid passed to node_online(), it may crash, could you > > re-check it, > > > > see my changes below, > > > > 1) add tmp check against MAX_NUMNODES before node_online() check, > > > > and move it after get tmp in hugepages_setup() , this could cover > > both per-node alloc and normal alloc > > sorry，for normal alloc, tmp is the number of huge pages, we don't need > the movement, only add tmp >= MAX_NUMNODES is ok > Does the v4 patch address the issues which were raised in this thread? --- a/mm/hugetlb.c~hugetlb-fix-wrong-use-of-nr_online_nodes-v4 +++ a/mm/hugetlb.c @@ -2986,8 +2986,6 @@ int __alloc_bootmem_huge_page(struct hst struct huge_bootmem_page *m = NULL; /* initialize for clang */ int nr_nodes, node; - if (nid != NUMA_NO_NODE && !node_online(nid)) - return 0; /* do node specific alloc */ if (nid != NUMA_NO_NODE) { m = memblock_alloc_try_nid_raw(huge_page_size(h), huge_page_size(h), @@ -4174,7 +4172,7 @@ static int __init hugepages_setup(char * pr_warn("HugeTLB: architecture can't support node specific alloc, ignoring!\n"); return 0; } - if (!node_online(tmp)) + if (tmp >= MAX_NUMNODES || !node_online(tmp)) goto invalid; node = array_index_nospec(tmp, MAX_NUMNODES); p += count + 1;
On 2022/4/19 12:40, Andrew Morton wrote: > On Sat, 16 Apr 2022 09:21:45 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > >> On 2022/4/15 13:41, Kefeng Wang wrote: >>> On 2022/4/15 10:09, Davidlohr Bueso wrote: >>>> On Wed, 13 Apr 2022, Peng Liu wrote: >>>> >>>>> Certain systems are designed to have sparse/discontiguous nodes. In >>>>> this case, nr_online_nodes can not be used to walk through numa node. >>>>> Also, a valid node may be greater than nr_online_nodes. >>>>> >>>>> However, in hugetlb, it is assumed that nodes are contiguous. Recheck >>>>> all the places that use nr_online_nodes, and repair them one by one. >>>>> >>>>> Suggested-by: David Hildenbrand <david@redhat.com> >>>>> Fixes: 4178158ef8ca ("hugetlbfs: fix issue of preallocation of >>>>> gigantic pages can't work") >>>>> Fixes: b5389086ad7b ("hugetlbfs: extend the definition of hugepages >>>>> parameter to support node allocation") >>>>> Fixes: e79ce9832316 ("hugetlbfs: fix a truncation issue in hugepages >>>>> parameter") >>>>> Fixes: f9317f77a6e0 ("hugetlb: clean up potential spectre issue >>>>> warnings") >>>>> Signed-off-by: Peng Liu <liupeng256@huawei.com> >>>>> Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com> >>>>> Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com> >>>> Reviewed-by: Davidlohr Bueso <dave@stgolabs.net> >>>> >>>> ... but >>>> >>>>> --- >>>>> mm/hugetlb.c | 12 ++++++------ >>>>> 1 file changed, 6 insertions(+), 6 deletions(-) >>>>> >>>>> diff --git a/mm/hugetlb.c b/mm/hugetlb.c >>>>> index b34f50156f7e..5b5a2a5a742f 100644 >>>>> --- a/mm/hugetlb.c >>>>> +++ b/mm/hugetlb.c >>>>> @@ -2979,7 +2979,7 @@ int __alloc_bootmem_huge_page(struct hstate >>>>> *h, int nid) >>>>> struct huge_bootmem_page *m = NULL; /* initialize for clang */ >>>>> int nr_nodes, node; >>>>> >>>>> - if (nid != NUMA_NO_NODE && nid >= nr_online_nodes) >>>>> + if (nid != NUMA_NO_NODE && !node_online(nid)) >>>> afaict null_blk could also use this, actually the whole thing wants a >>>> helper - node_valid()? >>>> >>> This one should be unnecessary, and this patch looks has a bug, >>> >>> if a very nid passed to node_online(), it may crash, could you >>> re-check it, >>> >>> see my changes below, >>> >>> 1) add tmp check against MAX_NUMNODES before node_online() check, >>> >>> and move it after get tmp in hugepages_setup() , this could cover >>> both per-node alloc and normal alloc >> sorry，for normal alloc, tmp is the number of huge pages, we don't need >> the movement, only add tmp >= MAX_NUMNODES is ok >> > Does the v4 patch address the issues which were raised in this thread? Yes， v4 has fix this. > > > --- a/mm/hugetlb.c~hugetlb-fix-wrong-use-of-nr_online_nodes-v4 > +++ a/mm/hugetlb.c > @@ -2986,8 +2986,6 @@ int __alloc_bootmem_huge_page(struct hst > struct huge_bootmem_page *m = NULL; /* initialize for clang */ > int nr_nodes, node; > > - if (nid != NUMA_NO_NODE && !node_online(nid)) > - return 0; > /* do node specific alloc */ > if (nid != NUMA_NO_NODE) { > m = memblock_alloc_try_nid_raw(huge_page_size(h), huge_page_size(h), > @@ -4174,7 +4172,7 @@ static int __init hugepages_setup(char * > pr_warn("HugeTLB: architecture can't support node specific alloc, ignoring!\n"); > return 0; > } > - if (!node_online(tmp)) > + if (tmp >= MAX_NUMNODES || !node_online(tmp)) > goto invalid; > node = array_index_nospec(tmp, MAX_NUMNODES); > p += count + 1; > _ > > .
diff --git a/mm/hugetlb.c b/mm/hugetlb.c index b34f50156f7e..5b5a2a5a742f 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -2979,7 +2979,7 @@ int __alloc_bootmem_huge_page(struct hstate *h, int nid) struct huge_bootmem_page *m = NULL; /* initialize for clang */ int nr_nodes, node; - if (nid != NUMA_NO_NODE && nid >= nr_online_nodes) + if (nid != NUMA_NO_NODE && !node_online(nid)) return 0; /* do node specific alloc */ if (nid != NUMA_NO_NODE) { @@ -3088,7 +3088,7 @@ static void __init hugetlb_hstate_alloc_pages(struct hstate *h) } /* do node specific alloc */ - for (i = 0; i < nr_online_nodes; i++) { + for_each_online_node(i) { if (h->max_huge_pages_node[i] > 0) { hugetlb_hstate_alloc_pages_onenode(h, i); node_specific_alloc = true; @@ -4049,7 +4049,7 @@ static int __init hugetlb_init(void) default_hstate.max_huge_pages = default_hstate_max_huge_pages; - for (i = 0; i < nr_online_nodes; i++) + for_each_online_node(i) default_hstate.max_huge_pages_node[i] = default_hugepages_in_node[i]; } @@ -4164,9 +4164,9 @@ static int __init hugepages_setup(char *s) pr_warn("HugeTLB: architecture can't support node specific alloc, ignoring!\n"); return 0; } - if (tmp >= nr_online_nodes) + if (!node_online(tmp)) goto invalid; - node = array_index_nospec(tmp, nr_online_nodes); + node = array_index_nospec(tmp, MAX_NUMNODES); p += count + 1; /* Parse hugepages */ if (sscanf(p, "%lu%n", &tmp, &count) != 1) @@ -4294,7 +4294,7 @@ static int __init default_hugepagesz_setup(char *s) */ if (default_hstate_max_huge_pages) { default_hstate.max_huge_pages = default_hstate_max_huge_pages; - for (i = 0; i < nr_online_nodes; i++) + for_each_online_node(i) default_hstate.max_huge_pages_node[i] = default_hugepages_in_node[i]; if (hstate_is_gigantic(&default_hstate))
Certain systems are designed to have sparse/discontiguous nodes. In this case, nr_online_nodes can not be used to walk through numa node. Also, a valid node may be greater than nr_online_nodes. However, in hugetlb, it is assumed that nodes are contiguous. Recheck all the places that use nr_online_nodes, and repair them one by one. Suggested-by: David Hildenbrand <david@redhat.com> Fixes: 4178158ef8ca ("hugetlbfs: fix issue of preallocation of gigantic pages can't work") Fixes: b5389086ad7b ("hugetlbfs: extend the definition of hugepages parameter to support node allocation") Fixes: e79ce9832316 ("hugetlbfs: fix a truncation issue in hugepages parameter") Fixes: f9317f77a6e0 ("hugetlb: clean up potential spectre issue warnings") Signed-off-by: Peng Liu <liupeng256@huawei.com> --- mm/hugetlb.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)