What is the difference between devm_kzalloc() and kzalloc() in linux driver programming

Question

I have found devm_kzalloc() and kzalloc() in device driver programmong. But I don't know when/where to use these functions. Can anyone please specify the importance of these functions and their usage.

Answer 1

kzalloc() allocates kernel memory like kmalloc(), but it also zero-initializes the allocated memory. devm_kzalloc() is managed kzalloc(). The memory allocated with managed functions is associated with the device. When the device is detached from the system or the driver for the device is unloaded, that memory is freed automatically. If multiple managed resources (memory or some other resource) were allocated for the device, the resource allocated last is freed first.

Managed resources are very helpful to ensure correct operation of the driver both for initialization failure at any point and for successful initialization followed by the device removal.

Please note that managed resources (whether it's memory or some other resource) are meant to be used in code responsible for the probing the device. They are generally a wrong choice for the code used for opening the device, as the device can be closed without being disconnected from the system. Closing the device requires freeing the resources manually, which defeats the purpose of managed resources.

The memory allocated with kzalloc() should be freed with kfree(). The memory allocated with devm_kzalloc() is freed automatically. It can be freed with devm_kfree(), but it's usually a sign that the managed memory allocation is not a good fit for the task.

Answer 2

In simple words devm_kzalloc() and kzalloc() both are used for memory allocation in device driver but the difference is if you allocate memory by kzalloc() than you have to free that memory when the life cycle of that device driver is ended or when it is unloaded from kernel but if you do the same with devm_kzalloc() you need not to worry about freeing memory,that memory is freed automatically by device library itself.

Both of them does the exactly the same thing but by using devm_kzalloc little overhead of freeing memory is released from programmers

Let explain you by giving example, first example by using kzalloc

static int pxa3xx_u2d_probe(struct platform_device *pdev)
{
    int err;
    u2d = kzalloc(sizeof(struct pxa3xx_u2d_ulpi), GFP_KERNEL);     1
    if (!u2d)
         return -ENOMEM;
    u2d->clk = clk_get(&pdev->dev, NULL);
    if (IS_ERR(u2d->clk)) {
        err = PTR_ERR(u2d->clk);                                    2
        goto err_free_mem;
    }
...
    return 0;
err_free_mem:
    kfree(u2d);
    return err;
}
static int pxa3xx_u2d_remove(struct platform_device *pdev)
{
    clk_put(u2d->clk);               
    kfree(u2d);                                                     3
    return 0;
}

In this example you can this in funtion pxa3xx_u2d_remove(), kfree(u2d)(line indicated by 3) is there to free memory allocated by u2d now see the same code by using devm_kzalloc()

static int pxa3xx_u2d_probe(struct platform_device *pdev)
{
    int err;
    u2d = devm_kzalloc(&pdev->dev, sizeof(struct pxa3xx_u2d_ulpi), GFP_KERNEL);
    if (!u2d)
        return -ENOMEM;
    u2d->clk = clk_get(&pdev->dev, NULL);
    if (IS_ERR(u2d->clk)) {
         err = PTR_ERR(u2d->clk);
         goto err_free_mem;
    }
...
    return 0;
err_free_mem:
    return err;
}
static int pxa3xx_u2d_remove(struct platform_device *pdev)
{
    clk_put(u2d->clk);
    return 0;
}

there is no kfree() to free function because the same is done by devm_kzalloc()