ascend inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA
Signed-off-by: NZhou Guanghui <zhouguanghui1@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------------

Remove unnecessary params(pid, spg_id) of sp_unshare.
Signed-off-by: NZhou Guanghui <zhouguanghui1@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

The prot parameter is added to sp_group_add_task function. It indicates the
PROT_READ/PROT_WRITE permission of the task for this spg.
Signed-off-by: NPeng Wu <wupeng58@huawei.com>
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Signed-off-by: NZhou Guanghui <zhouguanghui1@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

In multi-group mode, we introduce struct sp_spg_stat which represents
statistics for an sp_group. It is the accumulation of all structs of
spg_proc_stat in an sp_group.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NZhou Guanghui <zhouguanghui1@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

In multi-group mode, we introduce struct spg_proc_stat which represents
statistics in an sp_group for a process. And struct sp_proc_stat
is the accumulation of all structs of spg_proc_stat for a process.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NZhou Guanghui <zhouguanghui1@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

When a process not in any sp group calls k2u_task, it will encounter a
null pointer visit problem in sp_init_proc_stat:

mm->sp_group_master is null, and visit mm->sp_group_master->sp_stat_id
is illegal.

To fix this, we initialize sp_group_master when call k2u_task.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NZhou Guanghui <zhouguanghui1@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

We are going to redesign the accounting subsystem of share pool.
We need to disambiguate the meaning of sp_stat_idr, as we will
introduce an struct which representing per-spg statistics.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NZhou Guanghui <zhouguanghui1@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

We are going to redesign the accounting subsystem of share pool.
First we need to disambiguate the meaning of sp_spg_stat, as it is
system-level not spg-level.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

The max num of process in a group is MAX_PROC_PER_GROUP.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

The max num of sp_group in the system is MAX_GROUP_FOR_SYSTEM.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

The share pool need to support multi-group mode for automotive platform,
it will enhance the system reliability and security.

The new multi-group mode could be enabled by boot command line. When
disabled, the share pool should only support single group mode by
default. When enabled, the task could be added to several groups
(at most 3k). At most 50k groups can be created in the whole system.

This patch also fixes the kabi problem for mm struct.
Signed-off-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

The situation below is not allowed:

int *result = mmap(ADDR, sizeof(int), PROT_READ | PROT_WRITE,
	MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, -1, 0);

As share pool uses an independent UVA allocation algorithm, it may
produce an address that is conflicted with user-specified address.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

Once sp group is created, the generated id will be freed in
sp_group_drop. Before that, we should call free_sp_group_id()
when error occurs.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

Free id at the end of sp_group_add_task when failed.
Benefits are below:
1. Less time to hold locks.
2. Avoid to forget freeing id in other error handling branches.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

Increase the value of sp_stat_sem when failed.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

Only root can enable pr_debug printing.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

Let memory compact to be configurable.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

We found a hungtask problem when do direct compact in
__alloc_pages_nodemask:

vmalloc_hugepage_user -> __vmalloc_node_range -> __vmalloc_area_node ->
sp_alloc_pages -> alloc_huge_page_node -> alloc_fresh_huge_page ->
__alloc_pages_nodemask.

Set PF_MEMALLOC then direct reclaim and direct compact won't be called.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

We found a concurrency problem of sp_group_add_task and sp_free which
lead to memory leak.

After process A calls __sp_free and vfs_fallocate but before calling
__sp_area_drop, process B is being added to the same group by a manager
process, the *dead* spa freed by sp_free may be mapped into process B
again, then do_mm_populate is called.

When sp_group_add_task is finished, this spa is dropped and can't be
seen in /proc/sharepool/spa_stat, but the memory of spa still reside in
the group. It can only be freed when the group is dead.

To fix the problem, we add a member is_dead in spa. We can access it
when spg->rw_lock is held. This may sound a little strange if not
realizing the life cycle of spa has a direct relation with sp group.
Suggested-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huwei.com>
Reviewed-by: N为珑 陈 <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

---------------------------------------------------

When mmput is called concurrently, the judgment of "mm_users == 2"
in sp_group_exit is not atomic with atomic_dec_and_test in mmput.
The judgment of "mm_users == 2" may never be valid. As a result,
mm leakage occurs.

For example, in a typical scenario, a process has two threads, with
the mmget is performed in sp_group_add_task. In this case, mm_users
is 3. When two threads exit at the same time, the judgment of
"mm_users == 2" fail.

Therefore, the judgment and atomic_dec_and_test are put in the spg
rw_lock to ensure the serialization of the whole process.
Signed-off-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NZhou Guanghui <zhouguanghui1@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------

Adding a function for getting node id, which can be used to alloc
share pool memory on a specified memory node.
Signed-off-by: NPeng Wu <wupeng58@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: Nchenweilong <chenweilong@huawei.com>
Reviewed-by: NTang Yizhou <tangyizhou@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: perf
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

We encounter a problem as follows:

[ 3057. 75094] share pool: task add group failed, current thread is killed
[ 3057. 75152] [ascend] [drv_buff] [buff_mv_pid_node_to_recycle_list 872] <rosnode:12273,12273> release empty list node pid 12273, group_id 1
[ 3057. 76380] [ascend] [ERROR] [drv_buff] [buff_req_ioctl_pid_add_group 443] <rosnode:12297,12297> pid add group failed, pid:12297, grp_id:1, ret -512
[ 3057. 76382] [ascend] [drv_buff] [buff_ioctl 841] <rosnode:12297,12297> buff_req_ioctl_handlers failed. ret:-512
[ 3057. 76452] Unable to handle kernel paging request at virtual address dead000000000108
[ 3057. 76454] Mem abort info:
[ 3057. 76456]   ESR = 0x96000044
[ 3057. 76457]   Exception class = DABT (current EL), IL = 32 bits
[ 3057. 76458]   SET = 0, FnV = 0
[ 3057. 76459]   EA = 0, S1PTW = 0
[ 3057. 76460] Data abort info:
[ 3057. 76461]   ISV = 0, ISS = 0x00000044
[ 3057. 76462]   CM = 0, WnR = 1
[ 3057. 76463] [dead000000000108] address between user and kernel address ranges
[ 3057. 76466] Internal error: Oops: 96000044 [#1] SMP
[ 3057. 76469] Process rosnode (pid: 12308, stack limit = 0x0000000012aa85df)
[ 3057. 76473] CPU: 10 PID: 12308 Comm: rosnode Tainted: P         C O      4.19.95-1.h1.AOS2.0.aarch64 #1
[ 3057. 76474] Hardware name: evb (DT)
[ 3057. 76476] pstate: 20400009 (nzCv daif +PAN -UAO)
[ 3057. 76483] pc : sp_group_exit+0x94/0x130
[ 3057. 76486] lr : sp_group_exit+0x48/0x130
[ 3057. 76486] sp : ffff00001a163c10
[ 3057. 76487] pmr_save: 000000e0
[ 3057. 76489] x29: ffff00001a163c10 x28: ffff800887e2a940
[ 3057. 76491] x27: 0000000000000000 x26: ffff800d8098ca40
[ 3057. 76492] x25: ffff80089a879168 x24: ffff00001a163dd0
[ 3057. 76494] x23: 0000000000000000 x22: 0000000000000002
[ 3057. 76495] x21: ffff800896e73088 x20: ffff80089a879100
[ 3057. 76496] x19: ffff800896e73000 x18: ffff7e002ca9a4f4
[ 3057. 76498] x17: 0000000000000001 x16: 0000000000000001
[ 3057. 76499] x15: 0400000000000000 x14: ffff800bd5d0d050
[ 3057. 76500] x13: 0000000000000001 x12: 0000000000000000
[ 3057. 76502] x11: 0000000000000000 x10: 00000000000009e0
[ 3057. 76503] x9 : ffff00001a163a90 x8 : ffff800887e2b380
[ 3057. 76505] x7 : 00000000000000b4 x6 : 0000001b5b9081bb
[ 3057. 76506] x5 : dead000000000100 x4 : dead000000000200
[ 3057. 76507] x3 : dead000000000100 x2 : dead000000000200
[ 3057. 76508] x1 : ffff800d81365400 x0 : ffff800896e73088
[ 3057. 76510] Call trace:
[ 3057. 76513]  sp_group_exit+0x94/0x130
[ 3057. 76517]  mmput+0x20/0x170
[ 3057. 76519]  do_exit+0x338/0xb38
[ 3057. 76520]  do_group_exit+0x3c/0xe8
[ 3057. 76522]  get_signal+0x14c/0x7d8
[ 3057. 76524]  do_signal+0x88/0x290
[ 3057. 76525]  do_notify_resume+0x150/0x3c8
[ 3057. 76528]  work_pending+0x8/0x10
[ 3057. 76530] Code: d2804004 f2fbd5a5 f2fbd5a4 aa1503e0 (f9000462)
[ 3057. 76534] [kbox] unable to set sctrl register, 				maybe the domain is not SD, continue
[ 3057. 76535] [kbox] catch die event on cpu 10
[ 3057. 76537] [kbox] catch die event, start logging
[ 3057. 76540] [kbox] die info:Oops:0044
[ 3057. 76540] [kbox] start to collect

If process A adds process B into an sp_group and B is killed at the
mean time, then the calling of sp_group_add_task for B is failed and

list_del(&mm->sp_node);

is executed. Notice there is also an execution of this code in
sp_group_exit for B, so mm->sp_node is double freed.

The addr of sp_node->next is LIST_POISON1, which is dead000000000108
in arm64.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NKefeng Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

Address range [MMAP_SHARE_POOL_START, MMAP_SHARE_POOL_16G_START) and
[MMAP_SHARE_POOL_16G_START, MMAP_SHARE_POOL_16G_START +
MMAP_SHARE_POOL_16G_SIZE) is always valid.
Signed-off-by: NPeng Wu <wupeng58@huawei.com>
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NKefeng Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: perf
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

We did an compaction experiment on Hi1951. First creating external
fragmentation, then echo 1 > /proc/sys/vm/compact_memory.

Fragmentation size | 0G   | 2G   | 4G  | 8G
Compaction time(s) | 0.07 | 0.85 | 1.7 | 3.4

Obviously, sysctl_compaction_handler has a big performance impact.

We optimize share pool compact procedure as follows:
1. At most one compact daemon is allowed.
2. Creating a compact daemon is only allowed when the last one is
finished at least *sysctl_sp_compact_interval* seconds ago.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: doc
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

Update function documentation, this also fix warnings:
Function parameter or member 'PARAM_NAME' not described in 'FUNC_NAME'
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: Kefeng Wang<wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

Fix warnings: symbol 'XX' was not declared. Should it be static?
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: Kefeng Wang<wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

Fix warnings: missing braces around initializer
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NKefeng Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

---------------------------------------------

During the u2k, the page is being migrated. Wait
until the migration is complete.
Signed-off-by: NZhou Guanghui <zhouguanghui1@huawei.com>
Reviewed-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NKefeng Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

1. Add parameter checking in sp_alloc.
2. Add variable initialization in sp_group_exit.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NKefeng Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

Now a process only calls sp_k2u(to task) but not in any sp group will
lead to memleak of struct sp_proc_stat, after the process exits.

We should decouple the release of struct sp_group from the release of
struct sp_proc_stat.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NKefeng Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

A task without adding to an sp group (such as only calls sp_k2u_to_task)
should be shown in /proc/sharepool/proc_stat correctly.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NKefeng Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

In AI training scene, the sp area range is specified by host user, it
will result in an address offset calculation error, so fix it to avoid
the offset overflow.

If the user needs to specify the sp area range, the user should reserve
the virtual memory space by themselvs, otherwise the sp area would be
used for none share pool process.
Signed-off-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NKefeng Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

We find there are two differnt errnos returned when do_mm_polulate is
called and current thread is being killed, which may confuse logger
readers.

1. do_mm_populate -> populate_vma_page_range -> __get_user_pages.
if fatal_signal_pending fails, return -ERESTARTSYS(-512).
2. __get_user_pages -> follow_hugetlb_page.
if fatal_signal_pending fails, return -EFAULT(-14).

So we call fatal_signal_pending after do_mm_populte fails in share pool
and give info about the thread killing event.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NKefeng  Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: perf
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

We decide to remove sp_mutex thoroughly and use sp_group_sem instead.

Moreover, we use down_read(&spg->rw_lock) instead of down_write(), this
increases parallelism of sp_add_group_task(), especially when there are
many normal pages in the target sp_group.

Test: First a process is added into sp_group 1, then it calls sp_alloc()
to get 3.5G normal page memory and 5.5G hugepage memory. Then 30
processes are created and added to sp_group 1 concurrently. The result
is as follows.

without this patch: 15.0s
with this patch:    4.5s
gain:               70%
Tested-by: NTang Yizhou <tangyizhou@huawei.com>
Signed-off-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NKefeng  Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

Let function names be general.
Rename buff_vzalloc_user to vzalloc_user_account.
Rename buff_vzalloc_hugepage_user to vzalloc_hugepage_user_account.

To support NUMA configuration, we also introduce parameter *node*,
which means NUMA node id.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NKefeng  Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

/proc/sharepool/spa_stat now can show the pid of applier process, which
helps to debug and check memleak.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NKefeng  Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

We encounter a call trace as follows:

[  243.545984] Call trace:
[  243.545984]  find_vma+0x90/0xa0
[  243.545985]  find_extend_vma+0x2c/0xd0
[  243.545985]  __get_user_pages+0x94/0x378
[  243.545985]  get_dump_page+0x50/0x80
[  243.545986]  elf_core_dump+0x560/0x8d8
[  243.545986]  do_coredump+0x508/0xe40
[  243.545986]  get_signal+0x130/0x788
[  243.545987]  do_signal+0x1d4/0x290
[  243.545987]  do_notify_resume+0x150/0x1c0
[  243.545988]  work_pending+0x8/0x10

elf_core_dump() doesn't hold mmap_sem because the other threads in the
same thread group are killed and blocked in exit_mm(), waiting for
calling coredump_finish().

However, share pool operations can modify the mm of any process in the
same share group and lead to concurrent problems when coredump happens.

Solution: in share pool operations, check whether coredump happened with
mm->core_state.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NKefeng  Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: perf
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------
According to kernel comment, printk_ratelimited is not recommended, we
should use pr_<level>_ratelimited instead.

This also helps to reduce cyclomatic complexity.

In addition, %pK is not useful when we want to know the virtual memory
address. We decide to use pr_debug() and %lx, only root users have the
permission to switch /sys/kernel/debug/dynamic_debug/control which meets
the security requirements.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NKefeng  Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

For the implementation of Linux, statistics of RSS has a maximum 64 pages
deviation (256KB) but the track of share pool are all precise. So the
calculation results may be negative and confuse people.

We decide to show zeros when the results are negative. It is still
imprecise, but maybe better.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NKefeng  Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

ascend inclusion
category: perf
bugzilla: https://gitee.com/openeuler/kernel/issues/I4EUVI
CVE: NA

-------------------------------------------------

sp_group_exit() and sp_group_post_exit() should be put together, and so
shall it be for free_sp_group() and sp_group_drop().

This helps the code to be more readable, and improves cache hit ratio.
Signed-off-by: NTang Yizhou <tangyizhou@huawei.com>
Reviewed-by: NDing Tianhong <dingtianhong@huawei.com>
Reviewed-by: NKefeng  Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>
Reviewed-by: NWeilong Chen <chenweilong@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>