Hello Rohit,
On 11/2/18 1:06 PM, Rohit Kumar wrote:
On 11/2/2018 1:12 PM, Takashi Iwai wrote:
On Thu, 01 Nov 2018 13:38:49 +0100, Rohit kumar wrote:
Remove no_pcm check to invoke pcm_new() for backend dai-links too. This fixes crash in hdmi codec driver during hdmi_codec_startup() while accessing chmap_info struct. chmap_info struct memory is allocated in pcm_new() of hdmi codec driver which is not invoked in case of DPCM when hdmi codec driver is part of backend dai-link.
Below is the crash stack:
[ 61.635493] Unable to handle kernel NULL pointer dereference at virtual address 00000018 .. [ 61.666696] CM = 0, WnR = 1 [ 61.669778] user pgtable: 4k pages, 39-bit VAs, pgd = ffffffc0d6633000 [ 61.676526] [0000000000000018] *pgd=0000000153fc8003, *pud=0000000153fc8003, *pmd=0000000000000000 [ 61.685793] Internal error: Oops: 96000046 [#1] PREEMPT SMP [ 61.722955] CPU: 7 PID: 2238 Comm: aplay Not tainted 4.14.72 #21 .. [ 61.740269] PC is at hdmi_codec_startup+0x124/0x164 [ 61.745308] LR is at hdmi_codec_startup+0xe4/0x164
Signed-off-by: Rohit kumar rohitkr@codeaurora.org
Did you check whether all drivers have no side-effect by this change? The hdmi-codec isn't the only driver that has pcm_new ops, so we have to make sure that such a fundamental change wouldn't bring any regressions.
Below are the drivers calling pcm_new() other than hdmi codec driver. sound/soc/meson/axg-frddr.c sound/soc/meson/axg-toddr.c These two drivers are frontend DAI drivers and should not be impacted because of this.
Other than this, pcm_new() is called from sound/soc/stm/stm32_sai_sub.c I could not get much info about this driver. However, it is just adding kcontrols in pcm_new() which uses internal private structs in get()/put(). Olivier Moysan can too confirm on this.
First, i'm answering for Olivier: no regression identified for the SAI driver, it is not a DPCM driver.
Then i have a concern about the call of pcm_new for a no-PCM backend. Does it make sense? In DPCM concept, the backend is not linked to the PCM device...
Instead, I would suggest that you add protection in HDMI_codec on chmap_info pointer.
The drawback would be that the control is no more available...do you need it?
Regards Arnaud
Thanks, Rohit
thanks,
Takashi