Benjamin Herrenschmidt wrote:
;; >- out_le32(&md->control, (RUN|PAUSE|FLUSH|WAKE) << 16); /* stop dma */
;;
;; What about waiting for the DBDMA channel to actually stop ?
;; You should spin loop until the channel is actually stopped here.
;;
;; Note also that due to PCI write posting, doing an out_xxx, followed by a
;; udelay, may not work properly. You should do the out_xx, then a dummy
;; in_xxx from the same PCI bus to fush write buffers, then the udelay.
After applying the patch from velco, my mesh_init looks like this:
static void mesh_init(struct mesh_state *ms)
{
volatile struct mesh_regs *mr = ms->mesh;
volatile struct dbdma_regs *md = ms->dma;
udelay(100);
out_8(&mr->exception, 0xff); /* clear all exception bits */
out_8(&mr->error, 0xff); /* clear all error bits */
out_8(&mr->sequence, SEQ_RESETMESH);
/* flush write buffers.. How?
in_8(&mr->sequence); ?? */
udelay(10);
out_8(&mr->interrupt, 0xff); /* clear all interrupt bits */
out_8(&mr->intr_mask, INT_ERROR | INT_EXCEPTION | INT_CMDDONE);
out_8(&mr->source_id, ms->host->this_id);
out_8(&mr->sel_timeout, 25); /* 250ms */
out_8(&mr->sync_params, ASYNC_PARAMS); /* asynchronous initially */
/* stop dma */
out_le32(&md->control, (RUN|PAUSE|FLUSH|WAKE) << 16);
}
Since I have little idea what it is doing, I don't think I am
capable of incorporatng your suggestions into it... Could you
just tell me if the above code is dangerous to use (it could trash
my HD etc)?
regards,
abe
