Re: [PATCH v5 2/2] Add Intel LGM soc DMA support.

[Peter Ujfalusi]

On 18/08/2020 13.16, Peter Ujfalusi wrote:

...

>> +static void dma_issue_pending(struct dma_chan *chan)
>> +{
>> +	struct ldma_chan *c = to_ldma_chan(chan);
>> +	struct ldma_dev *d = to_ldma_dev(c->vchan.chan.device);
>> +	unsigned long flags;
>> +
>> +	if (d->ver == DMA_VER22) {
>> +		spin_lock_irqsave(&c->vchan.lock, flags);
>> +		if (vchan_issue_pending(&c->vchan)) {
>> +			struct virt_dma_desc *vdesc;
>> +
>> +			/* Get the next descriptor */
>> +			vdesc = vchan_next_desc(&c->vchan);
>> +			if (!vdesc) {
>> +				c->ds = NULL;
>> +				return;
>> +			}
>> +			list_del(&vdesc->node);
>> +			c->ds = to_lgm_dma_desc(vdesc);
> 
> you have set c->ds in dma_prep_slave_sg and the only way I can see that
> you will not leak memory is that the client must terminate_sync() after
> each transfer so that the synchronize callback is invoked between each
> prep_sg/issue_pending/competion.
> 
>> +			spin_unlock_irqrestore(&c->vchan.lock, flags);
>> +			ldma_chan_desc_hw_cfg(c, c->ds->desc_phys, c->ds->desc_cnt);
>> +			ldma_chan_irq_en(c);
>> +		}
> 
> If there is nothing pending, you will leave the spinlock wide open...

you leave it locked...

> 
>> +	}
>> +	ldma_chan_on(c);
>> +}

- Péter

Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki