On 1/10/19 10:55 AM, Stephan Gerhold wrote:
On Sun, Dec 16, 2018 at 07:49:56PM +0100, Stephan Gerhold wrote:
Right now, the machine devices are created early, before the SST context is initialized. This means that SST might not be fully initialized if sst_acpi_probe() fails later on (e.g. after sst_platform_get_resources() if the IRQ does not exist).
But that's a theoretical point here, isn't it. Your other patch solves the IRQ issue so do we have a real problem?
The reason why I am pushing back is that we've moved this code around several times and I am concerned about side effects - and none of the original developers are still around.
However, at least sst-mfld-platform assumes that sst_register_dsp() was called to set the (global) "sst" device pointer, which happens only later in sst_acpi_probe() when sst_context_init() is called. This may cause a NULL pointer dereference later when the ALSA device is first opened:
BUG: unable to handle kernel NULL pointer dereference at 0000000000000008 PGD 0 P4D 0 Oops: 0000 [#1] PREEMPT SMP PTI CPU: 0 PID: 790 Comm: pulseaudio Not tainted 4.20.0-rc6-mainline-00161-g6531e115b7ab #1 Hardware name: ASUSTeK COMPUTER INC. ME176C/ME176C, BIOS 5.6.5 09/16/2015 RIP: 0010:sst_handle_vb_timer+0x61/0x1b0 [snd_soc_sst_atom_hifi2_platform] Code: 44 24 04 e9 84 00 00 00 31 c9 c7 04 24 ff ff ff ff 66 89 4c 24 06 84 db 0f 84 90 00 00 00 48 8b 05 c4 23 01 00 be 01 00 00 00 <48> 8b 78 08 48 8b 40 10 48 8b 40 48 e8 2e 5e d7 f0 89 c3 85 c0 78 RSP: 0018:ffff9d968099fa30 EFLAGS: 00010202 RAX: 0000000000000000 RBX: 0000000000000001 RCX: 0000000000000000 RDX: 0000000080000001 RSI: 0000000000000001 RDI: 00000000ffffffff RBP: ffff968d33384618 R08: 0000000000000001 R09: 00000000000002e3 R10: ffff968d333a0800 R11: 0000000000000000 R12: ffff968d34bc7c00 R13: ffff968d333a3eb0 R14: 0000000000000001 R15: ffff968d333a08c0 FS: 00007f63a7e7b200(0000) GS:ffff968d37600000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 0000000000000008 CR3: 00000000297b8000 CR4: 00000000001006f0 Call Trace: sst_enable_ssp+0x24/0x40 [snd_soc_sst_atom_hifi2_platform] soc_pcm_open+0xeb/0x960 [snd_soc_core] ? __debugfs_create_file+0xcd/0x120 dpcm_be_dai_startup+0x183/0x3c0 [snd_soc_core] dpcm_fe_dai_open+0x10c/0xab0 [snd_soc_core] snd_pcm_open_substream+0x7f/0x140 [snd_pcm] snd_pcm_open+0xe6/0x220 [snd_pcm] ? wake_up_q+0x70/0x70 snd_pcm_playback_open+0x3d/0x70 [snd_pcm] chrdev_open+0xa3/0x1b0 ? cdev_put.part.0+0x20/0x20 do_dentry_open+0x12f/0x350 path_openat+0x2d1/0x14e0 ? inotify_handle_event+0x17b/0x1e0 do_filp_open+0x93/0x100 ? snd_card_file_remove+0x14b/0x170 [snd] ? __check_object_size+0x102/0x189 ? _raw_spin_unlock+0x12/0x30 do_sys_open+0x186/0x210 do_syscall_64+0x55/0x160 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x7f63a992f4c2 Code: 25 00 00 41 00 3d 00 00 41 00 74 4c 48 8d 05 85 70 0d 00 8b 00 85 c0 75 6d 89 f2 b8 01 01 00 00 48 89 fe bf 9c ff ff ff 0f 05 <48> 3d 00 f0 ff ff 0f 87 a2 00 00 00 48 8b 4c 24 28 64 48 33 0c 25 RSP: 002b:00007ffe71196b70 EFLAGS: 00000246 ORIG_RAX: 0000000000000101 RAX: ffffffffffffffda RBX: 0000000000080802 RCX: 00007f63a992f4c2 RDX: 0000000000080802 RSI: 00007ffe71196d20 RDI: 00000000ffffff9c RBP: 00000000ffffffff R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000246 R12: 00007ffe71196c00 R13: 0000000000000004 R14: 00007ffe71196d20 R15: 0000558466055a80 CR2: 0000000000000008 ---[ end trace 34534a02650ee26c ]---
This can be avoided if the machine device creation is delayed in sst_acpi_probe() until after sst_context_init(), when sst_register_dsp() is guaranteed to have already been called.
Signed-off-by: Stephan Gerhold stephan@gerhold.net
An other option to fix this would be to add proper NULL checks in the probe method of sst-mfld-platform and/or sst_enable_ssp(). Maybe this should be done additionally, but at least in my opinion there is not much point in registering the machine devices if they end up being broken anyway.
sound/soc/intel/atom/sst/sst_acpi.c | 24 ++++++++++++------------ 1 file changed, 12 insertions(+), 12 deletions(-)
diff --git a/sound/soc/intel/atom/sst/sst_acpi.c b/sound/soc/intel/atom/sst/sst_acpi.c index ac542535b9d5..493d32923815 100644 --- a/sound/soc/intel/atom/sst/sst_acpi.c +++ b/sound/soc/intel/atom/sst/sst_acpi.c @@ -345,6 +345,18 @@ static int sst_acpi_probe(struct platform_device *pdev) mach->mach_params.acpi_ipc_irq_index = pdata->res_info->acpi_ipc_irq_index;
- /* Fill sst platform data */
- ctx->pdata = pdata;
- strcpy(ctx->firmware_name, mach->fw_filename);
- ret = sst_platform_get_resources(ctx);
- if (ret)
return ret;- ret = sst_context_init(ctx);
- if (ret < 0)
return ret;- plat_dev = platform_device_register_data(dev, pdata->platform, -1, NULL, 0); if (IS_ERR(plat_dev)) {
@@ -365,18 +377,6 @@ static int sst_acpi_probe(struct platform_device *pdev) return PTR_ERR(mdev); }
- /* Fill sst platform data */
- ctx->pdata = pdata;
- strcpy(ctx->firmware_name, mach->fw_filename);
- ret = sst_platform_get_resources(ctx);
- if (ret)
return ret;- ret = sst_context_init(ctx);
- if (ret < 0)
return ret;- sst_configure_runtime_pm(ctx); platform_set_drvdata(pdev, ctx); return ret;
-- 2.20.0
Hi,
Mark's mail on the other thread ("ASoC: Intel: sst: Missing IRQ at index 5 on BYT-T device") just reminded me that this patch is still open.
With "ASoC: Intel: sst: Fallback to BYT-CR if IRQ 5 is missing" the initialization failure is solved for my device, and it does no longer run into this BUG. However, the NULL pointer dereference is still possible if another device has no or an invalid IRQ listed, causing SST initialization to fail.
This patch is one way to avoid it.
Let me know if I should re-send the patch (in case you cannot find it anymore). :)
Thanks, Stephan