Re: [PATCH 2/6] firmware: qcom: scm: Fix missing read barrier in qcom_scm_get_tzmem_pool()

[Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>]

On 19/11/2024 19:33, Krzysztof Kozlowski wrote:
> Commit 2e4955167ec5 ("firmware: qcom: scm: Fix __scm and waitq
> completion variable initialization") introduced a write barrier in probe
> function to store global '__scm' variable.  We all known barriers are
> paired (see memory-barriers.txt: "Note that write barriers should
> normally be paired with read or address-dependency barriers"), therefore
> accessing it from concurrent contexts requires read barrier.  Previous
> commit added such barrier in qcom_scm_is_available(), so let's use that
> directly.
> 
> Lack of this read barrier can result in fetching stale '__scm' variable
> value, NULL, and dereferencing it.
> 
> Fixes: ca61d6836e6f ("firmware: qcom: scm: fix a NULL-pointer dereference")
> Fixes: 449d0d84bcd8 ("firmware: qcom: scm: smc: switch to using the SCM allocator")
> Cc: <stable@vger.kernel.org>
> Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
> ---
>  drivers/firmware/qcom/qcom_scm.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/firmware/qcom/qcom_scm.c b/drivers/firmware/qcom/qcom_scm.c
> index 246d672e8f7f0e2a326a03a5af40cd434a665e67..5d91b8e22844608f35432f1ba9c08d477d4ff762 100644
> --- a/drivers/firmware/qcom/qcom_scm.c
> +++ b/drivers/firmware/qcom/qcom_scm.c
> @@ -217,7 +217,10 @@ static DEFINE_SPINLOCK(scm_query_lock);
>  
>  struct qcom_tzmem_pool *qcom_scm_get_tzmem_pool(void)
>  {
> -	return __scm ? __scm->mempool : NULL;
> +	if (!qcom_scm_is_available())
> +		return NULL;
> +
> +	return __scm->mempool;

I mentioned in commit msg that previous commit adds barrier in
qcom_scm_get_tzmem_pool(), so to be clear:
This depends on previous commit, because that barrier in
qcom_scm_is_available() solves the control dependency here, assuming the
minimal guarantee #1 ("On any given CPU, dependent memory accesses will
be issued in order, with respect to itself.")

If this is inlined by compiler it will be:

scm = READ_ONCE(__scm);
barrier()
if (scm)
	return scm->mempool;
else
	return NULL;

Which should be even safer than standard guarantee above (according to
my understanding).

Best regards,
Krzysztof