Re: [REGRESSION] Re: [PATCH 6.1 033/219] memcg: drop kmem.limit_in_bytes

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

 



On 9/21/2023 9:52 AM, Michal Hocko wrote:
> On Wed 20-09-23 14:46:52, Shakeel Butt wrote:
>> On Wed, Sep 20, 2023 at 1:08 PM Michal Hocko <mhocko@xxxxxxxx> wrote:
>>>
>> [...]
>>>> have a strong opinion against it. Also just to be clear we are not
>>>> talking about full revert of 58056f77502f but just the returning of
>>>> EOPNOTSUPP, right?
>>>
>>> If we allow the limit to be set without returning a failure then we
>>> still have options 2 and 3 on how to deal with that. One of them is to
>>> enforce the limit.
>>>
>>
>> Option 3 is a partial revert of 58056f77502f where we keep the no
>> limit enforcement and remove the EOPNOTSUPP return on write. Let's go
>> with option 3. In addition, let's add pr_warn_once on the read of
>> kmem.limit_in_bytes as well.
> 
> How about this?
> --- 

I'm OK with this approach. You're missing this in the patch below:

// static struct cftype mem_cgroup_legacy_files[] = {

+       {
+               .name = "kmem.limit_in_bytes",
+               .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
+               .write = mem_cgroup_write,
+               .read_u64 = mem_cgroup_read_u64,
+       },


Thanks,
Jeremi

>>From 81ae0797d8da1b9cfbf357b4be4787a5bbf46bb4 Mon Sep 17 00:00:00 2001
> From: Michal Hocko <mhocko@xxxxxxxx>
> Date: Thu, 21 Sep 2023 09:38:29 +0200
> Subject: [PATCH] mm, memcg: reconsider kmem.limit_in_bytes deprecation
> 
> This reverts commits 86327e8eb94c ("memcg: drop kmem.limit_in_bytes")
> and partially reverts 58056f77502f ("memcg, kmem: further deprecate
> kmem.limit_in_bytes") which have incrementally removed support for the
> kernel memory accounting hard limit. Unfortunately it has turned out
> that there is still userspace depending on the existence of
> memory.kmem.limit_in_bytes [1]. The underlying functionality is not
> really required but the non-existent file just confuses the userspace
> which fails in the result. The patch to fix this on the userspace side
> has been submitted but it is hard to predict how it will propagate
> through the maze of 3rd party consumers of the software.
> 
> Now, reverting alone 86327e8eb94c is not an option because there is
> another set of userspace which cannot cope with ENOTSUPP returned when
> writing to the file. Therefore we have to go and revisit 58056f77502f
> as well. There are two ways to go ahead. Either we give up on the
> deprecation and fully revert 58056f77502f as well or we can keep
> kmem.limit_in_bytes but make the write a noop and warn about the fact.
> This should work for both known breaking workloads which depend on the
> existence but do not depend on the hard limit enforcement.
> 
> [1] http://lkml.kernel.org/r/20230920081101.GA12096@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
> Fixes: 86327e8eb94c ("memcg: drop kmem.limit_in_bytes")
> Fixes: 58056f77502f ("memcg, kmem: further deprecate kmem.limit_in_bytes")
> Signed-off-by: Michal Hocko <mhocko@xxxxxxxx>
> ---
>  Documentation/admin-guide/cgroup-v1/memory.rst |  7 +++++++
>  mm/memcontrol.c                                | 12 ++++++++++++
>  2 files changed, 19 insertions(+)
> 
> diff --git a/Documentation/admin-guide/cgroup-v1/memory.rst b/Documentation/admin-guide/cgroup-v1/memory.rst
> index 5f502bf68fbc..ff456871bf4b 100644
> --- a/Documentation/admin-guide/cgroup-v1/memory.rst
> +++ b/Documentation/admin-guide/cgroup-v1/memory.rst
> @@ -92,6 +92,13 @@ Brief summary of control files.
>   memory.oom_control		     set/show oom controls.
>   memory.numa_stat		     show the number of memory usage per numa
>  				     node
> + memory.kmem.limit_in_bytes          Deprecated knob to set and read the kernel
> +                                     memory hard limit. Kernel hard limit is not
> +                                     supported since 5.16. Writing any value to
> +                                     do file will not have any effect same as if
> +                                     nokmem kernel parameter was specified.
> +                                     Kernel memory is still charged and reported
> +                                     by memory.kmem.usage_in_bytes.
>   memory.kmem.usage_in_bytes          show current kernel memory allocation
>   memory.kmem.failcnt                 show the number of kernel memory usage
>  				     hits limits
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index a4d3282493b6..ac7f14b2338d 100644
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -3097,6 +3097,7 @@ static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg,
>  static int obj_cgroup_charge_pages(struct obj_cgroup *objcg, gfp_t gfp,
>  				   unsigned int nr_pages)
>  {
> +	struct page_counter *counter;
>  	struct mem_cgroup *memcg;
>  	int ret;
>  
> @@ -3107,6 +3108,10 @@ static int obj_cgroup_charge_pages(struct obj_cgroup *objcg, gfp_t gfp,
>  		goto out;
>  
>  	memcg_account_kmem(memcg, nr_pages);
> +
> +	/* There is no way to set up kmem hard limit so this operation cannot fail */
> +	if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
> +		WARN_ON(!page_counter_try_charge(&memcg->kmem, nr_pages, &counter));
>  out:
>  	css_put(&memcg->css);
>  
> @@ -3867,6 +3872,13 @@ static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
>  		case _MEMSWAP:
>  			ret = mem_cgroup_resize_max(memcg, nr_pages, true);
>  			break;
> +		case _KMEM:
> +			pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. "
> +				     "Writing any value to this file has no effect. "
> +				     "Please report your usecase to linux-mm@xxxxxxxxx if you "
> +				     "depend on this functionality.\n");
> +			ret = 0;
> +			break;
>  		case _TCP:
>  			ret = memcg_update_tcp_max(memcg, nr_pages);
>  			break;




[Index of Archives]     [Linux Kernel]     [Kernel Development Newbies]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite Hiking]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux