Sascha Hauer <s.hauer@xxxxxxxxxxxxxx> writes: > Hi All, Hi Sascha, The short story : from my understanding, vc->desc_free() shouldn't be called with vd->lock held. The locking of callbacks called by virt-dma shouldn't be handled by virt-dma, but be a part of the dma driver's architecture rather than forced by virt-dma. The long story : This is what I remember and my understanding for the virt-dma : - virt_dma_chan->lock protects only : - all lists within virt_dma_chan - virt_dma_desc->node, which is on one of the above lists As a consequence of this statement, this lock shouldn't be held for other parts of the code, and therefore not while calling vc->desc_free(). I base the above statement on my "former" understanding, and that is a code I dealt with years ago, so I might be a little rusty here and there. The original code Russell wrote states in virt-dma.h : - in structure "struct_dma_desc" The "protected by vc.lock" is just for the "node" field - in structure "struct virt_dma_chan" The "protected by vc.lock" is just above the 4 lists (originally 3, and the I added one more) Now Peter might have more insight, as he has modified the code more recently that I did. Cheers. -- Robert PS: And yes, there are occurrences of list manipulations in virt-dma that should be protected by vc.lock as you pointed out.
