MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII;
 format=flowed
Date: Mon, 14 May 2018 19:26:03 +0530
From: Govind Singh <govinds@codeaurora.org>
To: Bjorn Andersson <bjorn.andersson@linaro.org>
Cc: ath10k@lists.infradead.org, linux-wireless@vger.kernel.org
Subject: Re: [PATCH 07/12] ath10k: Add MSA handshake QMI mgs support
In-Reply-To: <20180511184331.GR2259@tuxbook-pro>
References: <1522042832-25975-1-git-send-email-govinds@codeaurora.org>
 <20180511184331.GR2259@tuxbook-pro>
Message-ID: <d5d978dd4c0845f98ca49fae68a62a2f@codeaurora.org> (sfid-20180514_155609_447022_EF63FC5C)
Sender: linux-wireless-owner@vger.kernel.org

Hi Bjorn,
Thanks for the review.


On 2018-05-12 00:13, Bjorn Andersson wrote:
> On Sun 25 Mar 22:40 PDT 2018, Govind Singh wrote:
> 
>> HOST allocates 2mb of region for modem and WCN3990
>> secure access and provides the address and access
>> control to target for secure access.
>> Add MSA handshake request/response messages.
>> 
>> Signed-off-by: Govind Singh <govinds@codeaurora.org>
>> ---
>>  drivers/net/wireless/ath/ath10k/qmi.c | 288 
>> +++++++++++++++++++++++++++++++++-
>>  drivers/net/wireless/ath/ath10k/qmi.h |  14 ++
>>  2 files changed, 300 insertions(+), 2 deletions(-)
>> 
>> diff --git a/drivers/net/wireless/ath/ath10k/qmi.c 
>> b/drivers/net/wireless/ath/ath10k/qmi.c
>> index bc80b8f..763b812 100644
>> --- a/drivers/net/wireless/ath/ath10k/qmi.c
>> +++ b/drivers/net/wireless/ath/ath10k/qmi.c
>> @@ -26,6 +26,8 @@
>>  #include <linux/string.h>
>>  #include <net/sock.h>
>>  #include <linux/soc/qcom/qmi.h>
>> +#include <linux/qcom_scm.h>
>> +#include <linux/of.h>
>>  #include "qmi.h"
>>  #include "qmi_svc_v01.h"
>> 
>> @@ -35,6 +37,240 @@
>>  static struct ath10k_qmi *qmi;
>> 
>>  static int
>> +ath10k_qmi_map_msa_permissions(struct ath10k_msa_mem_region_info 
>> *mem_region)
>> +{
>> +	struct qcom_scm_vmperm dst_perms[3];
>> +	unsigned int src_perms;
>> +	phys_addr_t addr;
>> +	u32 perm_count;
>> +	u32 size;
>> +	int ret;
>> +
>> +	addr = mem_region->reg_addr;
>> +	size = mem_region->size;
> 
> Skip the local variables.
> 

Sure, will do in next version.

>> +
>> +	src_perms = BIT(QCOM_SCM_VMID_HLOS);
>> +
>> +	dst_perms[0].vmid = QCOM_SCM_VMID_MSS_MSA;
>> +	dst_perms[0].perm = QCOM_SCM_PERM_RW;
>> +	dst_perms[1].vmid = QCOM_SCM_VMID_WLAN;
>> +	dst_perms[1].perm = QCOM_SCM_PERM_RW;
>> +
>> +	if (!mem_region->secure_flag) {
> 
> So with secure_flag equal to 0 we give less subsystems access to the
> data? Is this logic inverted?
> 

Yes with secure flag mean less privileges i,e: Copy engine hardware
can not access the region.

>> +		dst_perms[2].vmid = QCOM_SCM_VMID_WLAN_CE;
>> +		dst_perms[2].perm = QCOM_SCM_PERM_RW;
>> +		perm_count = 3;
>> +	} else {
>> +		dst_perms[2].vmid = 0;
>> +		dst_perms[2].perm = 0;
>> +		perm_count = 2;
> 
> If you set perm_count to 2 you don't need to clear vmid and perm of
> dst_perms[2].
> 

ok, will remove.

>> +	}
>> +
>> +	ret = qcom_scm_assign_mem(addr, size, &src_perms,
>> +				  dst_perms, perm_count);
>> +	if (ret < 0)
>> +		pr_err("msa map permission failed=%d\n", ret);
> 
> Use dev_err()
> 
>> +
>> +	return ret;
>> +}
>> +
>> +static int
>> +ath10k_qmi_unmap_msa_permissions(struct ath10k_msa_mem_region_info 
>> *mem_region)
>> +{
>> +	struct qcom_scm_vmperm dst_perms;
>> +	unsigned int src_perms;
>> +	phys_addr_t addr;
>> +	u32 size;
>> +	int ret;
>> +
>> +	addr = mem_region->reg_addr;
>> +	size = mem_region->size;
> 
> Skip the local variables.
> 

Sure, will do in next version.

>> +
>> +	src_perms = BIT(QCOM_SCM_VMID_MSS_MSA) | BIT(QCOM_SCM_VMID_WLAN);
>> +
>> +	if (!mem_region->secure_flag)
>> +		src_perms |= BIT(QCOM_SCM_VMID_WLAN_CE);
>> +
>> +	dst_perms.vmid = QCOM_SCM_VMID_HLOS;
>> +	dst_perms.perm = QCOM_SCM_PERM_RW;
>> +
>> +	ret = qcom_scm_assign_mem(addr, size, &src_perms, &dst_perms, 1);
>> +	if (ret < 0)
>> +		pr_err("msa unmap permission failed=%d\n", ret);
>> +
>> +	return ret;
>> +}
>> +

>> +static int
>> +	ath10k_qmi_msa_mem_info_send_sync_msg(struct ath10k_qmi *qmi)
>> +{
>> +	struct wlfw_msa_info_resp_msg_v01 *resp;
> 
> This is 40 bytes,
> 
Sure, will do in next version for all instances.

>> +	struct wlfw_msa_info_req_msg_v01 *req;
> 
> This is 6 bytes.
> 
> So just put them on the stack and skip the memory management.
> 

Sure, will do in next version.

>> +	struct qmi_txn txn;
>> +	int ret;
>> +	int i;
>> +
>> +	req = kzalloc(sizeof(*req), GFP_KERNEL);
>> +	if (!req)
>> +		return -ENOMEM;
>> +
>> +	resp = kzalloc(sizeof(*resp), GFP_KERNEL);
>> +	if (!resp) {
>> +		kfree(req);
>> +		return -ENOMEM;
>> +	}
>> +
>> +	req->msa_addr = qmi->msa_pa;
>> +	req->size = qmi->msa_mem_size;
>> +
>> +	ret = qmi_txn_init(&qmi->qmi_hdl, &txn,
>> +			   wlfw_msa_info_resp_msg_v01_ei, resp);
>> +	if (ret < 0) {
>> +		pr_err("fail to init txn for MSA mem info resp %d\n",
>> +		       ret);
> 
> Unless we have 2 billion outstanding transactions "ret" is going to be
> ENOMEM here, in which case there is already a printout telling you 
> about
> the problem.
> 

Sure, will remove this redundant logging.

>> +		goto out;
>> +	}
>> +
>> +	ret = qmi_send_request(&qmi->qmi_hdl, NULL, &txn,
>> +			       QMI_WLFW_MSA_INFO_REQ_V01,
>> +			       WLFW_MSA_INFO_REQ_MSG_V01_MAX_MSG_LEN,
>> +			       wlfw_msa_info_req_msg_v01_ei, req);
>> +	if (ret < 0) {
>> +		qmi_txn_cancel(&txn);
>> +		pr_err("fail to send MSA mem info req %d\n", ret);
> 
> Use dev_err().
> 
>> +		goto out;
>> +	}
>> +
>> +	ret = qmi_txn_wait(&txn, WLFW_TIMEOUT * HZ);
> 
> The timeout is in jiffies, but I doubt you intended to the timeout to 
> be
> 500 seconds either.
> 

yes, its pretty high. I will reduce this.

>> +	if (ret < 0)
>> +		goto out;
>> +
>> +	if (resp->resp.result != QMI_RESULT_SUCCESS_V01) {
>> +		pr_err("MSA mem info request rejected, result:%d error:%d\n",
>> +		       resp->resp.result, resp->resp.error);
>> +		ret = -resp->resp.result;
> 
> Future readers will expect this to be a errno, in particular as most
> other exit paths return errnos.
> 
>> +		goto out;
>> +	}
>> +
>> +	pr_debug("receive mem_region_info_len: %d\n",
>> +		 resp->mem_region_info_len);
>> +
>> +	if (resp->mem_region_info_len > QMI_WLFW_MAX_NUM_MEMORY_REGIONS_V01) 
>> {
>> +		pr_err("invalid memory region length received: %d\n",
>> +		       resp->mem_region_info_len);
>> +		ret = -EINVAL;
>> +		goto out;
>> +	}
>> +
>> +	qmi->nr_mem_region = resp->mem_region_info_len;
>> +	for (i = 0; i < resp->mem_region_info_len; i++) {
>> +		qmi->mem_region[i].reg_addr =
>> +			resp->mem_region_info[i].region_addr;
> 
> With the use of a local variable you should be able to avoid the line
> breaks here.
> 

Sure, will do in next version.


>> +
>> +	ret = qmi_send_request(&qmi->qmi_hdl, NULL, &txn,
>> +			       QMI_WLFW_MSA_READY_REQ_V01,
>> +			       WLFW_MSA_READY_REQ_MSG_V01_MAX_MSG_LEN,
>> +			       wlfw_msa_ready_req_msg_v01_ei, req);
>> +	if (ret < 0) {
>> +		qmi_txn_cancel(&txn);
>> +		pr_err("fail to send MSA mem ready req %d\n", ret);
>> +		goto out;
>> +	}
>> +
>> +	ret = qmi_txn_wait(&txn, WLFW_TIMEOUT * HZ);
> 
> As above.
> 
>> +	if (ret < 0)
>> +		goto out;
>> +
>> +	if (resp->resp.result != QMI_RESULT_SUCCESS_V01) {
>> +		pr_err("qmi MSA mem ready request rejected, result:%d error:%d\n",
>> +		       resp->resp.result, resp->resp.error);
> 
> Clean up the print statement, there's no reason to print
> resp->resp.result, it is 1. And use dev_err().
> 
>> +		ret = -resp->resp.result;
> 
> This will be interpreted by future readers as an errno.
> 
>> +	}
>> +
>> +	pr_debug("MSA mem ready request completed\n");
> 
> dev_dbg().
> 
>> +	kfree(resp);
>> +	kfree(req);
>> +	return 0;
>> +
>> +out:
>> +	kfree(resp);
>> +	kfree(req);
>> +	return ret;
>> +}
>> +
>> +static int
>>  ath10k_qmi_ind_register_send_sync_msg(struct ath10k_qmi *qmi)
>>  {
>>  	struct wlfw_ind_register_resp_msg_v01 *resp;
>> @@ -173,7 +409,27 @@ static void ath10k_qmi_event_server_arrive(struct 
>> work_struct *work)
>>  	if (ret)
>>  		return;
>> 
>> -	ath10k_qmi_ind_register_send_sync_msg(qmi);
>> +	ret = ath10k_qmi_ind_register_send_sync_msg(qmi);
>> +	if (ret)
>> +		return;
>> +
>> +	ret = ath10k_qmi_msa_mem_info_send_sync_msg(qmi);
>> +	if (ret)
>> +		return;
>> +
>> +	ret = ath10k_qmi_setup_msa_permissions(qmi);
>> +	if (ret)
>> +		return;
>> +
>> +	ret = ath10k_qmi_msa_ready_send_sync_msg(qmi);
>> +	if (ret)
>> +		goto err_setup_msa;
>> +
>> +	return;
>> +
>> +err_setup_msa:
>> +	ath10k_qmi_remove_msa_permissions(qmi);
>> +	return;
>>  }
>> 
>>  static void ath10k_qmi_event_server_exit(struct work_struct *work)
>> @@ -252,6 +508,34 @@ static int ath10k_alloc_qmi_resources(struct 
>> ath10k_qmi *qmi)
>>  	return ret;
>>  }
>> 
>> +static int ath10k_qmi_setup_msa_resources(struct device *dev,
>> +					  struct ath10k_qmi *qmi)
>> +{
>> +	int ret;
>> +
>> +	ret = of_property_read_u32(dev->of_node, "qcom,wlan-msa-memory",
>> +				   &qmi->msa_mem_size);
>> +
>> +	if (ret || qmi->msa_mem_size == 0) {
>> +		pr_err("fail to get MSA memory size: %u, ret: %d\n",
>> +		       qmi->msa_mem_size, ret);
> 
> Use dev_err() and drop the msa_mem_size and ret from the printout, it
> doesn't add any value.
> 
>> +		return -ENOMEM;
>> +	}
>> +
>> +	qmi->msa_va = dmam_alloc_coherent(dev, qmi->msa_mem_size,
>> +					  &qmi->msa_pa, GFP_KERNEL);
> 
> The third parameter is supposed to be a dma_addr_t, not a phys_addr_t.
> But it looks like you can just change the type.
> 
>> +	if (!qmi->msa_va) {
>> +		pr_err("dma alloc failed for MSA\n");
>> +		return -ENOMEM;
>> +	}
>> +
>> +	pr_debug("MSA pa: %pa, MSA va: 0x%p\n",
>> +		 &qmi->msa_pa,
>> +		 qmi->msa_va);
> 
> dev_dbg() and the appropriate types are %pad and %pK.
> 
>> +
>> +	return 0;
>> +}
>> +
>>  static void ath10k_remove_qmi_resources(struct ath10k_qmi *qmi)
>>  {
>>  	cancel_work_sync(&qmi->work_svc_arrive);
>> @@ -272,7 +556,7 @@ static int ath10k_qmi_probe(struct platform_device 
>> *pdev)
>> 
>>  	qmi->pdev = pdev;
>>  	platform_set_drvdata(pdev, qmi);
>> -
> 
> This newline was a nice divider between the "chunks" of statements, 
> like
> paragraphs in a book...
> 
>> +	ath10k_qmi_setup_msa_resources(&pdev->dev, qmi);
>>  	ret = ath10k_alloc_qmi_resources(qmi);
>>  	if (ret < 0)
>>  		goto err;
>> diff --git a/drivers/net/wireless/ath/ath10k/qmi.h 
>> b/drivers/net/wireless/ath/ath10k/qmi.h
>> index 7697d24..47af020 100644
>> --- a/drivers/net/wireless/ath/ath10k/qmi.h
>> +++ b/drivers/net/wireless/ath/ath10k/qmi.h
>> @@ -16,6 +16,8 @@
>>  #ifndef _QMI_H_
>>  #define _QMI_H_
>> 
>> +#define MAX_NUM_MEMORY_REGIONS			2
>> +
>>  enum ath10k_qmi_driver_event_type {
>>  	ATH10K_QMI_EVENT_SERVER_ARRIVE,
>>  	ATH10K_QMI_EVENT_SERVER_EXIT,
>> @@ -23,6 +25,12 @@ enum ath10k_qmi_driver_event_type {
>>  	ATH10K_QMI_EVENT_MAX,
>>  };
>> 
>> +struct ath10k_msa_mem_region_info {
> 
> Afaict this struct is only used in qmi.c, so move it there.
> 
>> +	u64 reg_addr;
> 
> This is your driver-local copy of the data being received in a
> wlfw_msa_info_resp_msg_v01, use kernel native types instead.
> 
> I.e. make this phys_addr_t,
> 

Sure, will do in next version.

>> +	u32 size;
> 
> this is a size_t
> 
>> +	u8 secure_flag;
> 
> And based on how you use this, just make it bool secure (and invert the
> logic?)
> 

Sure, will do in next version.

>> +};
>> +
>>  struct ath10k_qmi {
>>  	struct platform_device *pdev;
>>  	struct qmi_handle qmi_hdl;
>> @@ -33,5 +41,11 @@ struct ath10k_qmi {
>>  	struct work_struct work_svc_exit;
>>  	struct workqueue_struct *event_wq;
>>  	spinlock_t event_lock; /* spinlock for fw ready status*/
>> +	u32 nr_mem_region;
>> +	struct ath10k_msa_mem_region_info
>> +		mem_region[MAX_NUM_MEMORY_REGIONS];
>> +	phys_addr_t msa_pa;
>> +	u32 msa_mem_size;
>> +	void *msa_va;
>>  };
> 
> Regards,
> Bjorn

BR,
Govind