linux-kernel - Re: [PATCH v2 11/11] firmware: qcom: scm: enable SHM bridge

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <1160e239-b227-411d-8d64-a23fde014dd5@quicinc.com>
Date:   Thu, 28 Sep 2023 10:10:51 -0700
From:   Elliot Berman <quic_eberman@...cinc.com>
To:     Bartosz Golaszewski <brgl@...ev.pl>,
        Andy Gross <agross@...nel.org>,
        "Bjorn Andersson" <andersson@...nel.org>,
        Konrad Dybcio <konrad.dybcio@...aro.org>,
        Maximilian Luz <luzmaximilian@...il.com>,
        Krzysztof Kozlowski <krzysztof.kozlowski@...aro.org>
CC:     <linux-kernel@...r.kernel.org>, <linux-arm-msm@...r.kernel.org>,
        <kernel@...cinc.com>,
        Bartosz Golaszewski <bartosz.golaszewski@...aro.org>
Subject: Re: [PATCH v2 11/11] firmware: qcom: scm: enable SHM bridge



On 9/28/2023 2:20 AM, Bartosz Golaszewski wrote:
> From: Bartosz Golaszewski <bartosz.golaszewski@...aro.org>
> 
> Extens the SCM memory allocator with using the SHM Bridge feature if
> available on the platform. This makes the trustzone only use dedicated
> buffers for SCM calls. We map the entire SCM genpool as a bridge.
> 
> Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@...aro.org>
> ---
>  drivers/firmware/qcom/qcom_scm-mem.c | 42 ++++++++++++++++++++++++++--
>  1 file changed, 39 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/firmware/qcom/qcom_scm-mem.c b/drivers/firmware/qcom/qcom_scm-mem.c
> index eafecbe23770..12b12b15f46f 100644
> --- a/drivers/firmware/qcom/qcom_scm-mem.c
> +++ b/drivers/firmware/qcom/qcom_scm-mem.c
> @@ -16,6 +16,8 @@
>  
>  #include "qcom_scm.h"
>  
> +#define QCOM_SHM_BRIDGE_NUM_VM_SHIFT 9
> +
>  static size_t qcom_scm_mem_pool_size = SZ_2M;
>  module_param_named(qcom_scm_mem_pool_size, qcom_scm_mem_pool_size,
>  		   ulong, 0400);
> @@ -108,8 +110,24 @@ phys_addr_t qcom_scm_mem_to_phys(void *vaddr)
>  	return chunk->paddr;
>  }
>  
> +static int qcom_scm_mem_shm_bridge_create(void)
> +{
> +	uint64_t pfn_and_ns_perm, ipfn_and_s_perm, size_and_flags, ns_perms;
> +
> +	ns_perms = (QCOM_SCM_PERM_WRITE | QCOM_SCM_PERM_READ);
> +	pfn_and_ns_perm = (u64)qcom_scm_mem.pbase | ns_perms;
> +	ipfn_and_s_perm = (u64)qcom_scm_mem.pbase | ns_perms;
> +	size_and_flags = qcom_scm_mem.size | (1 << QCOM_SHM_BRIDGE_NUM_VM_SHIFT);
> +
> +	return qcom_scm_create_shm_bridge(qcom_scm_mem.dev, pfn_and_ns_perm,
> +					  ipfn_and_s_perm, size_and_flags,
> +					  QCOM_SCM_VMID_HLOS);
> +}
> +
>  int qcom_scm_mem_enable(struct device *dev)
>  {
> +	int ret;
> +
>  	INIT_RADIX_TREE(&qcom_scm_mem.chunks, GFP_ATOMIC);
>  	spin_lock_init(&qcom_scm_mem.lock);
>  	qcom_scm_mem.dev = dev;
> @@ -128,7 +146,25 @@ int qcom_scm_mem_enable(struct device *dev)
>  
>  	gen_pool_set_algo(qcom_scm_mem.pool, gen_pool_best_fit, NULL);
>  
> -	return gen_pool_add_virt(qcom_scm_mem.pool,
> -				 (unsigned long)qcom_scm_mem.vbase,
> -				 qcom_scm_mem.pbase, qcom_scm_mem.size, -1);
> +	ret = gen_pool_add_virt(qcom_scm_mem.pool,
> +				(unsigned long)qcom_scm_mem.vbase,
> +				qcom_scm_mem.pbase, qcom_scm_mem.size, -1);
> +	if (ret)
> +		return ret;
> +
> +	ret = qcom_scm_enable_shm_bridge();
> +	if (ret) {
> +		if (ret == EOPNOTSUPP)
> +			dev_info(dev, "SHM Bridge not supported\n");
> +		else
> +			return ret;
> +	} else {
> +		ret = qcom_scm_mem_shm_bridge_create();
> +		if (ret)
> +			return ret;
> +
> +		dev_info(dev, "SHM Bridge enabled\n");

Do you need to add clean up (deletion) of the SHM bridge on driver remove?

One easy approach I could think: implemnet devm_qcom_scm_mem_shm_bridge_create
which calls qcom_scm_delete_shm_bridge on the clean up 
(qcom_scm_delete_shm_bridge implemented in downstream, not in this series).

> +	}
> +
> +	return 0;
>  }