On Sun, 20 Jan 2008 21:37:41 -0700 Matthew Wilcox <matthew@xxxxxx> wrote: > On Mon, Jan 21, 2008 at 01:08:58PM +0900, FUJITA Tomonori wrote: > > On Sun, 20 Jan 2008 09:40:11 -0700 > > Matthew Wilcox <matthew@xxxxxx> wrote: > > > Longer-term, I want to allow low-level drivers to allocate the > > > sense_buffer themselves so they can DMA directly into it (ie grown-up dma > > > mapping, rather than this quaint x86 __GFP_DMA). This patch doesn't get > > > > Yeah, I think that the approach is one of candidates. > > > > If we go with it, I think that the major issue is that LLDs don't know > > when they can reclaim sense_buffer from scsi-ml; scsi-ml uses > > sense_buffer after scmd->scsi_done. > > The midlayer would call a function in the scsi_host_template to free the > command. The sense_buffer would be freed at the same time. Yeah, it would work though I'm a bit concerned about adding another phase to the scsi_cmnd interaction between the midlayer and LLDs. Another possible option is that putting some sense_buffer information to the host template and the midlayer allocates sense buffers for LLDs. LLDs can ask for them when it's necessary. One disadvantage of this is the many LLDs have sense buffer in their own data structure so it's not fit well for these LLDs, I think. - To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
