Re: [PATCH net-next v4 02/13] net: wwan: t7xx: Add control DMA interface

[Andy Shevchenko <andriy.shevchenko@linux.intel.com>]

On Thu, Jan 13, 2022 at 06:06:16PM -0700, Ricardo Martinez wrote:
> From: Haijun Liu <haijun.liu@mediatek.com>
> 
> Cross Layer DMA (CLDMA) Hardware interface (HIF) enables the control
> path of Host-Modem data transfers. CLDMA HIF layer provides a common
> interface to the Port Layer.
> 
> CLDMA manages 8 independent RX/TX physical channels with data flow
> control in HW queues. CLDMA uses ring buffers of General Packet
> Descriptors (GPD) for TX/RX. GPDs can represent multiple or single
> data buffers (DB).
> 
> CLDMA HIF initializes GPD rings, registers ISR handlers for CLDMA
> interrupts, and initializes CLDMA HW registers.
> 
> CLDMA TX flow:
> 1. Port Layer write
> 2. Get DB address
> 3. Configure GPD
> 4. Triggering processing via HW register write
> 
> CLDMA RX flow:
> 1. CLDMA HW sends a RX "done" to host
> 2. Driver starts thread to safely read GPD
> 3. DB is sent to Port layer
> 4. Create a new buffer for GPD ring

...

> + * Copyright (c) 2021, Intel Corporation.

2021-2022 (in all files)?

...

> +#include <linux/dev_printk.h>
> +#include <linux/device.h>

I believe the former is guaranteed to be included in the latter.
The rest of the headers in this file looks good to me.

...

> +	if (dma_mapping_error(md_ctrl->dev, req->mapped_buff)) {

> +		dev_err(md_ctrl->dev, "DMA mapping failed\n");

Can we first free resources and then print messages?

Printing messages is a slow path and freeing resources first is a good
(micro-)optimization.

> +		dev_kfree_skb_any(req->skb);
> +		req->skb = NULL;
> +		req->mapped_buff = 0;
> +		return -ENOMEM;
> +	}

...

> +static int t7xx_cldma_rx_ring_init(struct cldma_ctrl *md_ctrl, struct cldma_ring *ring)
> +{
> +	struct cldma_request *req, *first_req = NULL;
> +	struct cldma_rgpd *prev_gpd, *gpd = NULL;
> +	int i;
> +
> +	for (i = 0; i < ring->length; i++) {
> +		req = t7xx_alloc_rx_request(md_ctrl, ring->pkt_size);
> +		if (!req) {
> +			t7xx_cldma_ring_free(md_ctrl, ring, DMA_FROM_DEVICE);
> +			return -ENOMEM;
> +		}
> +
> +		gpd = req->gpd;
> +		t7xx_cldma_rgpd_set_data_ptr(gpd, req->mapped_buff);
> +		gpd->data_allow_len = cpu_to_le16(ring->pkt_size);
> +		gpd->gpd_flags = GPD_FLAGS_IOC | GPD_FLAGS_HWO;
> +
> +		if (i)
> +			t7xx_cldma_rgpd_set_next_ptr(prev_gpd, req->gpd_addr);
> +		else
> +			first_req = req;
> +
> +		INIT_LIST_HEAD(&req->entry);
> +		list_add_tail(&req->entry, &ring->gpd_ring);
> +		prev_gpd = gpd;
> +	}

> +	if (first_req)

At which circumstances it is not defined? Only when ring->length == 0, right?

> +		t7xx_cldma_rgpd_set_next_ptr(gpd, first_req->gpd_addr);

Looking into this, perhaps it makes sense to refactor this way:

1. Define list head pointer on the stack
2. Iterate over the ring->length and add elements to that list
3. Iterate over the list and set the DMA links
   (t7xx_cldma_rgpd_set_next_ptr() calls)
4. Add this list to the main one.

> +	return 0;
> +}

...

> +static int t7xx_cldma_tx_ring_init(struct cldma_ctrl *md_ctrl, struct cldma_ring *ring)
> +{
> +	struct cldma_request *req, *first_req = NULL;
> +	struct cldma_tgpd *tgpd, *prev_gpd;
> +	int i;
> +
> +	for (i = 0; i < ring->length; i++) {
> +		req = t7xx_alloc_tx_request(md_ctrl);
> +		if (!req) {
> +			t7xx_cldma_ring_free(md_ctrl, ring, DMA_TO_DEVICE);
> +			return -ENOMEM;
> +		}
> +
> +		tgpd = req->gpd;
> +		tgpd->gpd_flags = GPD_FLAGS_IOC;
> +
> +		if (!first_req)
> +			first_req = req;
> +		else
> +			t7xx_cldma_tgpd_set_next_ptr(prev_gpd, req->gpd_addr);
> +
> +		INIT_LIST_HEAD(&req->entry);
> +		list_add_tail(&req->entry, &ring->gpd_ring);
> +		prev_gpd = tgpd;
> +	}
> +
> +	if (first_req)
> +		t7xx_cldma_tgpd_set_next_ptr(tgpd, first_req->gpd_addr);
> +
> +	return 0;

Ditto.

> +}


-- 
With Best Regards,
Andy Shevchenko