Re: [PATCH 2/3] scsi: ufs: initialize max_lu_supported while booting

[Bart Van Assche]

On 2020-01-10 10:36, Bean Huo wrote:
> +static int ufshcd_read_geometry_desc(struct ufs_hba *hba, u8 *buf, u32 size)
> +{
> +	return ufshcd_read_desc(hba, QUERY_DESC_IDN_GEOMETRY, 0, buf, size);
> +}

The declaration of this function is longer than its body. Do we really
need this function? Has it been considered to inline this function into
its caller?

> +static int ufshcd_init_device_param(struct ufs_hba *hba)
> +{
> +	int err;
> +	size_t buff_len;
> +	u8 *desc_buf;
> +
> +	buff_len = QUERY_DESC_MAX_SIZE;
> +	desc_buf = kmalloc(buff_len, GFP_KERNEL);
> +	if (!desc_buf) {
> +		err = -ENOMEM;
> +		goto out;
> +	}

Has it been considered to use hba->desc_size.geom_desc instead of
QUERY_DESC_MAX_SIZE?

> +	err = ufshcd_read_geometry_desc(hba, desc_buf,
> +			hba->desc_size.geom_desc);
> +	if (err) {
> +		dev_err(hba->dev, "%s: Failed reading Geometry Desc. err = %d\n",
> +			__func__, err);
> +		goto out;
> +	}
> +
> +	if (desc_buf[GEOMETRY_DESC_PARAM_MAX_NUM_LUN] == 1)
> +		hba->dev_info.max_lu_supported = 32;
> +	else if (desc_buf[GEOMETRY_DESC_PARAM_MAX_NUM_LUN] == 0)
> +		hba->dev_info.max_lu_supported = 8;

Can it happen that GEOMETRY_DESC_PARAM_MAX_NUM_LUN >=
hba->desc_size.geom_desc and hence that the above code reads
uninitialized data?

> @@ -7016,13 +7052,22 @@ static int ufshcd_probe_hba(struct ufs_hba *hba)
>  
>  	/*
>  	 * If we are in error handling context or in power management callbacks
> -	 * context, no need to scan the host
> +	 * context, no need to scan the host and to reinitialize the parameters
>  	 */
>  	if (!ufshcd_eh_in_progress(hba) && !hba->pm_op_in_progress) {
>  		bool flag;
>  
>  		/* clear any previous UFS device information */
>  		memset(&hba->dev_info, 0, sizeof(hba->dev_info));
> +		/* Init parameters according to UFS relevant descriptors */
> +		ret = ufshcd_init_device_param(hba);
> +		if (ret) {
> +			dev_err(hba->dev,
> +				"%s: Init of device param failed. err = %d\n",
> +				__func__, ret);
> +			goto out;
> +		}
> +
>  		if (!ufshcd_query_flag_retry(hba, UPIU_QUERY_OPCODE_READ_FLAG,
>  				QUERY_FLAG_IDN_PWR_ON_WPE, &flag))
>  			hba->dev_info.f_power_on_wp_en = flag;

The context check in ufshcd_probe_hba() looks ugly to me. Has it been
considered to move all code that is controlled by the if-statement with
the context check into ufshcd_async_scan()?

Thanks,

Bart.