Re: [PATCH v8 1/2] cacheinfo: Allocate memory during CPU hotplug if not done from the primary CPU

From: Radu Rendec
Date: Fri Nov 29 2024 - 11:13:06 EST


On Wed, 2024-11-27 at 16:22 -0800, Ricardo Neri wrote:
> Commit 5944ce092b97 ("arch_topology: Build cacheinfo from primary CPU")
> adds functionality that architectures can use to optionally allocate and
> build cacheinfo early during boot. Commit 6539cffa9495 ("cacheinfo: Add
> arch specific early level initializer") lets secondary CPUs correct (and
> reallocate memory) cacheinfo data if needed.
>
> If the early build functionality is not used and cacheinfo does not need
> correction, memory for cacheinfo is never allocated. x86 does not use the
> early build functionality. Consequently, during the cacheinfo CPU hotplug
> callback, last_level_cache_is_valid() attempts to dereference a NULL
> pointer:
>
>      BUG: kernel NULL pointer dereference, address: 0000000000000100
>      #PF: supervisor read access in kernel mode
>      #PF: error_code(0x0000) - not present page
>      PGD 0 P4D 0
>      Oops: 0000 [#1] PREEPMT SMP NOPTI
>      CPU: 0 PID 19 Comm: cpuhp/0 Not tainted 6.4.0-rc2 #1
>      RIP: 0010: last_level_cache_is_valid+0x95/0xe0a
>
> Allocate memory for cacheinfo during the cacheinfo CPU hotplug callback if
> not done earlier.
>
> Moreover, before determining the validity of the last-level cache info,
> ensure that it has been allocated. Simply checking for non-zero
> cache_leaves() is not sufficient, as some architectures (e.g., Intel
> processors) have non-zero cache_leaves() before allocation.
>
> Dereferencing NULL cacheinfo can occur in update_per_cpu_data_slice_size().
> This function iterates over all online CPUs. However, a CPU may have come
> online recently, but its cacheinfo may not have been allocated yet.
>
> While here, remove an unnecessary indentation in allocate_cache_info().
>
> Reviewed-by: Andreas Herrmann <aherrmann@xxxxxxx>
> Reviewed-by: Nikolay Borisov <nik.borisov@xxxxxxxx>
> Reviewed-by: Radu Rendec <rrendec@xxxxxxxxxx>
> Reviewed-by: Sudeep Holla <sudeep.holla@xxxxxxx>
> Tested-by: Andreas Herrmann <aherrmann@xxxxxxx>
> Fixes: 6539cffa9495 ("cacheinfo: Add arch specific early level initializer")
> Signed-off-by: Ricardo Neri <ricardo.neri-calderon@xxxxxxxxxxxxxxx>
> ---
> Cc: Andreas Herrmann <aherrmann@xxxxxxxx>
> Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
> Cc: Chen Yu <yu.c.chen@xxxxxxxxx>
> Cc: Huang Ying <ying.huang@xxxxxxxxx>
> Cc: Len Brown <len.brown@xxxxxxxxx>
> Cc: Nikolay Borisov <nik.borisov@xxxxxxxx>
> Cc: Radu Rendec <rrendec@xxxxxxxxxx>
> Cc: Pierre Gondois <Pierre.Gondois@xxxxxxx>
> Cc: Pu Wen <puwen@xxxxxxxx>
> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@xxxxxxxxx>
> Cc: Sudeep Holla <sudeep.holla@xxxxxxx>
> Cc: Srinivas Pandruvada <srinivas.pandruvada@xxxxxxxxxxxxxxx>
> Cc: Will Deacon <will@xxxxxxxxxx>
> Cc: Zhang Rui <rui.zhang@xxxxxxxxx>
> Cc: linux-arm-kernel@xxxxxxxxxxxxxxxxxxx
> Cc: stable@xxxxxxxxxxxxxxx # 6.3+
> ---
> Change since v7:
>  * None
>
> Changes since v6:
>  * Merged patches 1 and 2 of v6 into one. (Borislav)
>  * Merged the history of patches 1 and 2ino this patch.
>  * Kept the Reviewed-by and Tested-by tags from the two merged patches.
>  * Fixed a formatting issue in allocate_cache_info(). (Borislav)
>
> Changes since v5:
>  * Fixed nonsensical subject (Nikolay).
>  * Added Reviewed-by and Tested-by tags from Andreas. Thanks!
>  * Added Reviewed-by tag from Nikolay. Thanks!
>
> Changes since v4:
>  * Combined checks for per_cpu_cacheinfo() and cache_leaves() in a single
>    line. (Sudeep)
>  * Added Reviewed-by tag from Sudeep. Thanks!
>
> Changes since v3:
>  * Added Reviewed-by tag from Radu and Sudeep. Thanks!
>
> Changes since v2:
>  * Introduced this patch.
>
> Changes since v1:
>  * N/A
>
> ---
> The motivation for commit 5944ce092b97 was to prevent a BUG splat in
> PREEMPT_RT kernels during memory allocation. This splat is not observed on
> x86 because the memory allocation for cacheinfo happens in
> detect_cache_attributes() from the cacheinfo CPU hotplug callback.
>
> The dereference of a NULL cacheinfo is not observed today because
> cache_leaves(cpu) is zero until after init_cache_level() is called
> (during the CPU hotplug callback). A subsequent changeset will set
> the number of cache leaves earlier and the NULL-pointer dereference
> will be observed.
> ---
>  drivers/base/cacheinfo.c | 11 +++++++----
>  1 file changed, 7 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/base/cacheinfo.c b/drivers/base/cacheinfo.c
> index 7a7609298e18..a1afc478e0e8 100644
> --- a/drivers/base/cacheinfo.c
> +++ b/drivers/base/cacheinfo.c
> @@ -58,7 +58,7 @@ bool last_level_cache_is_valid(unsigned int cpu)
>  {
>   struct cacheinfo *llc;
>  
> - if (!cache_leaves(cpu))
> + if (!cache_leaves(cpu) || !per_cpu_cacheinfo(cpu))
>   return false;
>  
>   llc = per_cpu_cacheinfo_idx(cpu, cache_leaves(cpu) - 1);
> @@ -466,8 +466,7 @@ int __weak populate_cache_leaves(unsigned int cpu)
>  static inline
>  int allocate_cache_info(int cpu)
>  {
> - per_cpu_cacheinfo(cpu) = kcalloc(cache_leaves(cpu),
> - sizeof(struct cacheinfo), GFP_ATOMIC);
> + per_cpu_cacheinfo(cpu) = kcalloc(cache_leaves(cpu), sizeof(struct cacheinfo), GFP_ATOMIC);
>   if (!per_cpu_cacheinfo(cpu)) {
>   cache_leaves(cpu) = 0;
>   return -ENOMEM;
> @@ -539,7 +538,11 @@ static inline int init_level_allocate_ci(unsigned int cpu)
>   */
>   ci_cacheinfo(cpu)->early_ci_levels = false;
>  
> - if (cache_leaves(cpu) <= early_leaves)
> + /*
> + * Some architectures (e.g., x86) do not use early initialization.
> + * Allocate memory now in such case.
> + */
> + if (cache_leaves(cpu) <= early_leaves && per_cpu_cacheinfo(cpu))
>   return 0;
>  
>   kfree(per_cpu_cacheinfo(cpu));

Since Borislav explicitly said you were not supposed to keep the tags,
I reviewed it again, and I'm OK to keep mine. Thanks!

Reviewed-by: Radu Rendec <rrendec@xxxxxxxxxx>

--
Best regards,
Radu