dmaengine: pl330: fix bug that cause start the same descs in cyclic
authorAddy Ke <addy.ke@rock-chips.com>
Mon, 8 Dec 2014 11:28:20 +0000 (19:28 +0800)
committerVinod Koul <vinod.koul@intel.com>
Wed, 11 Feb 2015 00:20:19 +0000 (16:20 -0800)
commit0091b9d6c1ef2caab6cb3b6c0aa75f9948307856
tree5da9067503338a84fce68273cbdc80b365b5df9a
parent6d3a7d9e3ada345948f72564ce638c412ccd8c4a
dmaengine: pl330: fix bug that cause start the same descs in cyclic

This bug will cause NULL pointer after commit dfac17, and cause
wrong package in I2S DMA transfer before commit dfac17.

Tested on RK3288-pinky2 board.

Detail:
I2S DMA transfer(sound/core/pcm_dmaengine.c):
dmaengine_pcm_prepare_and_submit -->
dmaengine_prep_dma_cyclic -->
pl330_prep_dma_cyclic -->
the case:
1. pl330_submit_req(desc0): thrd->req[0].desc = desc0, thrd->lstenq = 0
2. pl330_submit_req(desc1): thrd->req[1].desc = desc1, thrd->lstenq = 1
3. _start(desc0) by submit_req: thrd->req_running = 0
   because: idx = 1 - thrd->lstenq = 0
4. pl330_update(desc0 OK): thrd->req[0].desc = NULL, desc0 to req_done list
   because: idx = active = thrd->req_running = 0
5. _start(desc1) by pl330_update: thrd->req_running = 1
   because:
   idx = 1 - thrd->lstenq = 0, but thrd->req[0].desc == NULL,
   so:
   idx = thrd->lstenq = 1
6. pl330_submit_req(desc2): thrd->req[0].desc = desc2, thrd->lstenq = 0
7. _start(desc1) by submit_req: thrd->req_running = 1
   because: idx = 1 - thrd->lstenq = 1
   Note: _start started the same descs
         _start should start desc2 here, NOT desc1

8. pl330_update(desc1 OK): thrd->req[1].desc = NULL, desc1 to req_done list
   because: idx = active = thrd->req_running = 1
9. _start(desc2) by pl330_update : thrd->req_running = 0
   because: idx = 1 - thrd->lstenq = 0
10.pl330_update(desc1 OK, NOT desc2): thrd->req[0].desc = NULL,
   desc2 to req_done list
   because: idx = active = thrd->req_running = 0

11.pl330_submit_req(desc3): thrd->req[0].desc = desc3, thrd->lstenq = 0
12.pl330_submit_req(desc4): thrd->req[1].desc = desc4, thrd->lstenq = 1
13._start(desc3) by submit_req: thrd->req_running = 0
   because: idx = 1 - thrd->lstenq = 0
14.pl330_update(desc2 OK NOT desc3): thrd->req[0].desc = NULL
   desc3 to req_done list
   because: idx = active = thrd->req_running = 0
15._start(desc4) by pl330_update: thrd->req_running = 1
   because:
   idx = 1 - thrd->lstenq = 0, but thrd->req[0].desc == NULL,
   so:
   idx = thrd->lstenq = 1
16.pl330_submit_req(desc5): thrd->req[0].desc = desc5, thrd->lstenq = 0
17._start(desc4) by submit_req: thrd->req_running = 1
   because: idx = 1 - thrd->lstenq = 1
18.pl330_update(desc3 OK NOT desc4): thrd->req[1].desc = NULL
   desc4 to req_done list
   because: idx = active = thrd->req_running = 1
19._start(desc4) by pl330_update: thrd->req_running = 0
   because:
   idx = 1 - thrd->lstenq = 1, but thrd->req[1].desc == NULL,
   so:
   idx = thrd->lstenq = 0
20.pl330_update(desc4 OK): thrd->req[0].desc = NULL, desc5 to req_done list
   because: idx = active = thrd->req_running = 0
21.pl330_update(desc4 OK):
   1) before commit dfac17(set req_running -1 in pl330_update/mark_free()):
      because: active = -1, abort
      result: desc0-desc5's callback are all called,
      but step 10 and step 18 go wrong.
   2) before commit dfac17:
      idx = active = thrd->req_runnig = 0 -->
      descdone = thrd->req[0] = NULL -->
      list_add_tail(&descdone->rqd, &pl330->req_done); -->
      got NULL pointer!!!

Signed-off-by: Addy Ke <addy.ke@rock-chips.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
drivers/dma/pl330.c