Re: [RFC] proc: Add a new isolated /proc/pid/mempolicy type.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



[Cc linux-api - please do so for any patches making/updating
kernel<->user interfaces]


On Mon 26-09-22 17:10:33, hezhongkun wrote:
> From: Zhongkun He <hezhongkun.hzk@xxxxxxxxxxxxx>
> 
> /proc/pid/mempolicy can be used to check and adjust the userspace task's
> mempolicy dynamically.In many case, the application and the control plane
> are two separate systems. When the application is created, it doesn't know
> how to use memory, and it doesn't care. The control plane will decide the
> memory usage policy based on different reasons.In that case, we can
> dynamically adjust the mempolicy using /proc/pid/mempolicy interface.

Is there any reason to make it procfs interface rather than pidfd one?

[keeping the rest of the email for the linux-api reference]

> Format of input:
> ----------------
> <mode>[=<flags>][:<nodelist>]
> 
> Example
> -------
> set mempolicy:
>  $ echo "interleave=static:0-3" > /proc/27036/mempolicy
>  $ cat /proc/27036/mempolicy
>  interleave=static:0-3
> remove mempolicy:
> +  $ echo "default" > /proc/27036/mempolicy
> 
> The following 6 mempolicy mode types:
> "default" "prefer"  "bind" "interleave" "local" "prefer (many)"
> 
> The supported mode flags are:
> "static" "relative"
> 
> nodelist         For example:0-3 or 0,1,2,3
> 
> Signed-off-by: Zhongkun He <hezhongkun.hzk@xxxxxxxxxxxxx>
> ---
>  Documentation/filesystems/proc.rst |  40 +++++++++
>  fs/proc/base.c                     |   2 +
>  fs/proc/internal.h                 |   1 +
>  fs/proc/task_mmu.c                 | 129 +++++++++++++++++++++++++++++
>  include/linux/mempolicy.h          |   5 --
>  mm/mempolicy.c                     |   2 -
>  6 files changed, 172 insertions(+), 7 deletions(-)
> 
> diff --git a/Documentation/filesystems/proc.rst b/Documentation/filesystems/proc.rst
> index e7aafc82be99..fa7bc24c6a91 100644
> --- a/Documentation/filesystems/proc.rst
> +++ b/Documentation/filesystems/proc.rst
> @@ -47,6 +47,8 @@ fixes/update part 1.1  Stefani Seibold <stefani@xxxxxxxxxxx>    June 9 2009
>    3.10  /proc/<pid>/timerslack_ns - Task timerslack value
>    3.11	/proc/<pid>/patch_state - Livepatch patch operation state
>    3.12	/proc/<pid>/arch_status - Task architecture specific information
> +  3.13  /proc/<pid>/mempolicy & /proc/<pid>/task/<tid>/mempolicy- Adjust
> +                                                                the mempolicy
>  
>    4	Configuring procfs
>    4.1	Mount options
> @@ -2145,6 +2147,44 @@ AVX512_elapsed_ms
>    the task is unlikely an AVX512 user, but depends on the workload and the
>    scheduling scenario, it also could be a false negative mentioned above.
>  
> +3.13 /proc/<pid>/mempolicy & /proc/<pid>/task/<tid>/mempolicy- Adjust the mempolicy
> +-----------------------------------------------------------------------------------
> +When CONFIG_NUMA is enabled, these files can be used to check and adjust the current
> +mempolicy.Please note that the effectively <pid>,<tid> is from userspace programs.
> +
> +Format of input:
> +----------------
> +<mode>[=<flags>][:<nodelist>]
> +
> +Example
> +-------
> +set mempolicy:
> + $ echo "interleave=static:0-3" > /proc/27036/mempolicy
> + $ cat /proc/27036/mempolicy
> + interleave=static:0-3
> +
> +remove mempolicy:
> +  $ echo "default" > /proc/27036/mempolicy
> +
> +The following 6 mempolicy mode types are supported:
> +"default"         Default is converted to the NULL memory policy, any existing non-default policy
> +                  will simply be removed when "default" is specified.
> +"prefer"          The allocation should be attempted from the single node specified in the policy.
> +"bind"            Memory must come from the set of nodes specified by the policy.
> +"interleave"      Page allocations be interleaved across the nodes specified in the policy.
> +"local"           The memory is allocated on the node of the CPU that triggered the allocation.
> +"prefer (many)"   The allocation should be preferrably satisfied from the nodemask specified in the policy.
> +
> +The supported mode flags are:
> +
> +"static"          A nonempty nodemask specifies physical node IDs.
> +"relative"        A nonempty nodemask specifies node IDs that are relative
> +                  to the set of node IDs allowed by the thread's current cpuset.
> +
> +nodelist         For example: 0-3 or 0,1,2,3
> +
> +Please see: Documentation/admin-guide/mm/numa_memory_policy.rst  for descriptions of memory policy.
> +
>  Chapter 4: Configuring procfs
>  =============================
>  
> diff --git a/fs/proc/base.c b/fs/proc/base.c
> index 93f7e3d971e4..4dbe714b4e61 100644
> --- a/fs/proc/base.c
> +++ b/fs/proc/base.c
> @@ -3252,6 +3252,7 @@ static const struct pid_entry tgid_base_stuff[] = {
>  	REG("maps",       S_IRUGO, proc_pid_maps_operations),
>  #ifdef CONFIG_NUMA
>  	REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
> +	REG("mempolicy",  S_IRUGO|S_IWUSR, proc_mempolicy_operations),
>  #endif
>  	REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
>  	LNK("cwd",        proc_cwd_link),
> @@ -3600,6 +3601,7 @@ static const struct pid_entry tid_base_stuff[] = {
>  #endif
>  #ifdef CONFIG_NUMA
>  	REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
> +	REG("mempolicy",  S_IRUGO|S_IWUSR, proc_mempolicy_operations),
>  #endif
>  	REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
>  	LNK("cwd",       proc_cwd_link),
> diff --git a/fs/proc/internal.h b/fs/proc/internal.h
> index 06a80f78433d..33ffbd79db58 100644
> --- a/fs/proc/internal.h
> +++ b/fs/proc/internal.h
> @@ -300,6 +300,7 @@ extern const struct file_operations proc_pid_smaps_operations;
>  extern const struct file_operations proc_pid_smaps_rollup_operations;
>  extern const struct file_operations proc_clear_refs_operations;
>  extern const struct file_operations proc_pagemap_operations;
> +extern const struct file_operations proc_mempolicy_operations;
>  
>  extern unsigned long task_vsize(struct mm_struct *);
>  extern unsigned long task_statm(struct mm_struct *,
> diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
> index 4e0023643f8b..299276e19c52 100644
> --- a/fs/proc/task_mmu.c
> +++ b/fs/proc/task_mmu.c
> @@ -2003,4 +2003,133 @@ const struct file_operations proc_pid_numa_maps_operations = {
>  	.release	= proc_map_release,
>  };
>  
> +#define MPOLBUFLEN 64
> +/*
> + *Display task's  memory policy via /proc./
> + */
> +static ssize_t mempolicy_read(struct file *file, char __user *buf,
> +		size_t count, loff_t *ppos)
> +{
> +	struct task_struct *task = get_proc_task(file_inode(file));
> +	char buffer[MPOLBUFLEN];
> +	struct mempolicy *mpol;
> +	size_t len = 0;
> +
> +	if (!task)
> +		return -ESRCH;
> +
> +	task_lock(task);
> +	mpol = task->mempolicy;
> +	mpol_get(mpol);
> +	task_unlock(task);
> +
> +	if (!mpol || mpol->mode == MPOL_DEFAULT)
> +		goto out;
> +
> +	memset(buffer, 0, sizeof(buffer));
> +	mpol_to_str(buffer, sizeof(buffer), mpol);
> +	buffer[strlen(buffer)] = '\n';
> +	len = simple_read_from_buffer(buf, count, ppos, buffer, strlen(buffer));
> +
> +out:
> +	mpol_put(mpol);
> +	put_task_struct(task);
> +	return len;
> +}
> +
> +/*
> + *Update nodemask of mempolicy according to task->mems_allowed.
> + */
> +static int update_task_mpol(struct task_struct *task, struct mempolicy *mpol)
> +{
> +	nodemask_t tsk_allowed;
> +	struct mempolicy *old = NULL;
> +	int err = 0;
> +
> +	task_lock(task);
> +	local_irq_disable();
> +	old = task->mempolicy;
> +
> +	if (mpol)
> +		nodes_and(tsk_allowed, task->mems_allowed, mpol->w.user_nodemask);
> +	else
> +		nodes_clear(tsk_allowed);
> +
> +	if (!nodes_empty(tsk_allowed)) {
> +		task->mempolicy = mpol;
> +		mpol_rebind_task(task, &tsk_allowed);
> +	} else if (!mpol || mpol->mode == MPOL_LOCAL) {
> +		/*default (pol==NULL), clear the old mpol;
> +		 *local memory policies are not a subject of any remapping.
> +		 */
> +		task->mempolicy = mpol;
> +	} else {
> +		/*tsk_allowed is empty.*/
> +		err = -EINVAL;
> +	}
> +
> +	if (!err && mpol && mpol->mode == MPOL_INTERLEAVE)
> +		task->il_prev = MAX_NUMNODES-1;
> +
> +	local_irq_enable();
> +	task_unlock(task);
> +
> +	/*If successful, release old policy,
> +	 * otherwise keep old and release mpol.
> +	 */
> +	if (err)
> +		mpol_put(mpol);
> +	else
> +		mpol_put(old);
> +
> +	return err;
> +}
> +
> +/*
> + *Modify task's memory policy via /proc.
> + */
> +static ssize_t mempolicy_write(struct file *file, const char __user *buf,
> +		size_t count, loff_t *ppos)
> +{
> +	char buffer[MPOLBUFLEN];
> +	struct mempolicy *mpol = NULL;
> +	struct task_struct *task;
> +	int err = 0;
> +
> +	task = get_proc_task(file_inode(file));
> +
> +	if (!task)
> +		return -ESRCH;
> +
> +	/*we can only change the user's mempolicy*/
> +	if (task->flags & PF_KTHREAD || is_global_init(task)) {
> +		err = -EPERM;
> +		goto out;
> +	}
> +
> +	memset(buffer, 0, sizeof(buffer));
> +	if (count > sizeof(buffer) - 1)
> +		count = sizeof(buffer) - 1;
> +	if (copy_from_user(buffer, buf, count)) {
> +		err = -EFAULT;
> +		goto out;
> +	}
> +
> +	err = mpol_parse_str(strstrip(buffer), &mpol);
> +	if (err) {
> +		err = -EINVAL;
> +		goto out;
> +	}
> +	err = update_task_mpol(task, mpol);
> +out:
> +	put_task_struct(task);
> +	return err < 0 ? err : count;
> +}
> +
> +const struct file_operations proc_mempolicy_operations = {
> +	.read		= mempolicy_read,
> +	.write		= mempolicy_write,
> +	.llseek		= default_llseek,
> +};
> +
>  #endif /* CONFIG_NUMA */
> diff --git a/include/linux/mempolicy.h b/include/linux/mempolicy.h
> index 668389b4b53d..a08f66972e6b 100644
> --- a/include/linux/mempolicy.h
> +++ b/include/linux/mempolicy.h
> @@ -172,10 +172,7 @@ int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
>  		     const nodemask_t *to, int flags);
>  
>  
> -#ifdef CONFIG_TMPFS
>  extern int mpol_parse_str(char *str, struct mempolicy **mpol);
> -#endif
> -
>  extern void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol);
>  
>  /* Check if a vma is migratable */
> @@ -277,12 +274,10 @@ static inline void check_highest_zone(int k)
>  {
>  }
>  
> -#ifdef CONFIG_TMPFS
>  static inline int mpol_parse_str(char *str, struct mempolicy **mpol)
>  {
>  	return 1;	/* error */
>  }
> -#endif
>  
>  static inline int mpol_misplaced(struct page *page, struct vm_area_struct *vma,
>  				 unsigned long address)
> diff --git a/mm/mempolicy.c b/mm/mempolicy.c
> index b73d3248d976..a1ae6412e3ae 100644
> --- a/mm/mempolicy.c
> +++ b/mm/mempolicy.c
> @@ -2958,7 +2958,6 @@ static const char * const policy_modes[] =
>  };
>  
>  
> -#ifdef CONFIG_TMPFS
>  /**
>   * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option.
>   * @str:  string containing mempolicy to parse
> @@ -3091,7 +3090,6 @@ int mpol_parse_str(char *str, struct mempolicy **mpol)
>  		*mpol = new;
>  	return err;
>  }
> -#endif /* CONFIG_TMPFS */
>  
>  /**
>   * mpol_to_str - format a mempolicy structure for printing
> -- 
> 2.25.1

-- 
Michal Hocko
SUSE Labs




[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux