Re: [PATCH 00/16] The new slab memory controller

[Bharata B Rao <bharata@linux.ibm.com>]

On Mon, Dec 09, 2019 at 02:47:52PM +0530, Bharata B Rao wrote:
> Hi,
> 
> I see the below crash during early boot when I try this patchset on
> PowerPC host. I am on new_slab.rfc.v5.3 branch.
> 
> BUG: Unable to handle kernel data access at 0x81030236d1814578
> Faulting instruction address: 0xc0000000002cc314
> Oops: Kernel access of bad area, sig: 11 [#1]
> LE PAGE_SIZE=64K MMU=Hash SMP NR_CPUS=2048 NUMA PowerNV
> Modules linked in: ip_tables x_tables autofs4 sr_mod cdrom usbhid bnx2x crct10dif_vpmsum crct10dif_common mdio libcrc32c crc32c_vpmsum
> CPU: 31 PID: 1752 Comm: keyboard-setup. Not tainted 5.3.0-g9bd85fd72a0c #155
> NIP:  c0000000002cc314 LR: c0000000002cc2e8 CTR: 0000000000000000
> REGS: c000001e40f378b0 TRAP: 0380   Not tainted  (5.3.0-g9bd85fd72a0c)
> MSR:  900000010280b033 <SF,HV,VEC,VSX,EE,FP,ME,IR,DR,RI,LE,TM[E]>  CR: 44022224  XER: 00000000
> CFAR: c0000000002c6ad4 IRQMASK: 1 
> GPR00: c0000000000b8a40 c000001e40f37b40 c000000000ed9600 0000000000000000 
> GPR04: 0000000000000023 0000000000000010 c000001e40f37b24 c000001e3cba3400 
> GPR08: 0000000000000020 81030218815f4578 0000001e50220000 0000000000000030 
> GPR12: 0000000000002200 c000001fff774d80 0000000000000000 00000001072600d8 
> GPR16: 0000000000000000 c0000000000bbaac 0000000000000000 0000000000000000 
> GPR20: c000001e40f37c48 0000000000000001 0000000000000000 c000001e3cba3400 
> GPR24: c000001e40f37dd8 0000000000000000 c000000000fa0d58 0000000000000000 
> GPR28: c000001e3a080080 c000001e32da0100 0000000000000118 0000000000000010 
> NIP [c0000000002cc314] __mod_memcg_state+0x58/0xd0
> LR [c0000000002cc2e8] __mod_memcg_state+0x2c/0xd0
> Call Trace:
> [c000001e40f37b90] [c0000000000b8a40] account_kernel_stack+0xa4/0xe4
> [c000001e40f37bd0] [c0000000000ba4a4] copy_process+0x2b4/0x16f0
> [c000001e40f37cf0] [c0000000000bbaac] _do_fork+0x9c/0x3e4
> [c000001e40f37db0] [c0000000000bc030] sys_clone+0x74/0xa8
> [c000001e40f37e20] [c00000000000bb34] ppc_clone+0x8/0xc
> Instruction dump:
> 4bffa7e9 2fa30000 409e007c 395efffb 3d000020 2b8a0001 409d0008 39000020 
> e93d0718 e94d0028 7bde1f24 7d29f214 <7ca9502a> 7fff2a14 7fe9fe76 7d27fa78 
> 
> Looks like page->mem_cgroup_vec is allocated but not yet initialized
> with memcg pointers when we try to access them.
> 
> I did get past the crash by initializing the pointers like this
> in account_kernel_stack(),

The above is not an accurate description of the hack I showed below.
Essentially I am making sure that I get to the memcg corresponding
to task_struct_cachep object in the page.

But that still doesn't explain why we don't hit this problem on x86.

> but I am pretty sure that this is not the
> place to do this:
> 
> diff --git a/kernel/fork.c b/kernel/fork.c
> index 541fd805fb88..be21419feae2 100644
> --- a/kernel/fork.c
> +++ b/kernel/fork.c
> @@ -380,13 +380,26 @@ static void account_kernel_stack(struct task_struct *tsk, int account)
>                  * All stack pages are in the same zone and belong to the
>                  * same memcg.
>                  */
> -               struct page *first_page = virt_to_page(stack);
> +               struct page *first_page = virt_to_head_page((stack));
> +               unsigned long off;
> +               struct mem_cgroup_ptr *memcg_ptr;
> +               struct mem_cgroup *memcg;
>  
>                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
>                                     THREAD_SIZE / 1024 * account);
>  
> -               mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
> +               if (!first_page->mem_cgroup_vec)
> +                       return;
> +               off = obj_to_index(task_struct_cachep, first_page, stack);
> +               memcg_ptr = first_page->mem_cgroup_vec[off];
> +               if (!memcg_ptr)
> +                       return;
> +               rcu_read_lock();
> +               memcg = memcg_ptr->memcg;
> +               if (memcg)
> +                       __mod_memcg_state(memcg, MEMCG_KERNEL_STACK_KB,
>                                      account * (THREAD_SIZE / 1024));
> +               rcu_read_unlock();
>         }
>  }
> 
> Regards,
> Bharata.