Re: [PATCH v3 00/11] CSI2RX support on J721E

From: Pratyush Yadav
Date: Thu Sep 09 2021 - 14:13:00 EST


Hi Tomi,

On 01/07/21 10:56AM, Tomi Valkeinen wrote:
> Hi Pratyush,
>
> On 24/06/2021 22:21, Pratyush Yadav wrote:
> > Hi,
> >
> > This series adds support for CSI2 capture on J721E. It includes some
> > fixes to the Cadence CSI2RX driver, adds runtime PM support to OV5640
> > driver, and finally adds the TI CSI2RX wrapper driver.
> >
> > This series used to include the DPHY and DMA engine patches as well, but
> > they have been split off to facilitate easier merging. Patch 3 is
> > build-dependent on the DPHY series [0].
> >
> > The DMA engine patch [1] can go in any order since that is only a run
> > time dependency. Things probably won't work without it but it will still
> > build fine.
> >
> > Tested on TI's J721E with OV5640 sensor.
>
> I applied these (csi-2 rx, phy, dma-engine) to linux-media/master, and added dts changes to add the csi2-rx. When sending the series, can you also push the branch you use for testing, as the posted patches do not include everything needed?
>
> Here are some notes from quick tests:
>
> Capture works, but the fps is ~28.98. I would expect it to be closer to 30. Are the clocks configured correctly?
>
> When I load the modules, I get:
>
> [ 237.322258] platform 4504000.csi-bridge: Fixing up cyclic dependency with 9-003c

I see this with CAL's OV5640 overlay as well. I think this is caused by
the endpoint nodes on csi-bridge and sensor pointing to each other. I
can't quite understand any bad implications of this warning since
everything seems to work correctly.

Should we model the connections between the sensor and CSI bridge
differently to not cause a cycle in the graph? Or can we just ignore
this warning since things seem to work fine despite it?

>
> I get a warning from DMA-API debug:
>
> [ 298.774236] ------------[ cut here ]------------
> [ 298.779109] DMA-API: ti-udma 31150000.dma-controller: mapping sg segment longer than device claims to support [len=1900544] [max=65536]
> [ 298.791331] WARNING: CPU: 1 PID: 605 at kernel/dma/debug.c:1172 debug_dma_map_sg+0x304/0x390
> [ 298.799764] Modules linked in: ov5640 j721e_csi2rx cdns_csi2rx cdns_dphy v4l2_fwnode v4l2_async tidss ti_tfp410 tc358767 display_connector cdns_mhdp8546 panel_simple
> drm_kms_helper drm drm_panel_orientation_quirks cfbfillrect cfbimgblt cfbcopyarea phy_j721e_wiz phy_cadence_torrent
> [ 298.824656] CPU: 1 PID: 605 Comm: cam-mplex.py Not tainted 5.13.0-rc4-00417-g3331992006e9 #3
> [ 298.833079] Hardware name: Texas Instruments K3 J721E SoC (DT)
> [ 298.838900] pstate: 60000005 (nZCv daif -PAN -UAO -TCO BTYPE=--)
> [ 298.844895] pc : debug_dma_map_sg+0x304/0x390
> [ 298.849245] lr : debug_dma_map_sg+0x304/0x390
> [ 298.853593] sp : ffff800014dcf730
> [ 298.856899] x29: ffff800014dcf730 x28: ffff00080154a880 x27: ffffffffffffffff
> [ 298.864032] x26: 0000000000000000 x25: 0000000000000002 x24: 0000000000000001
> [ 298.871164] x23: ffff80001163abe0 x22: 0000000000000000 x21: 0000000000000001
> [ 298.878295] x20: ffff000801fa3010 x19: ffff000807585300 x18: 0000000000000000
> [ 298.885426] x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000030
> [ 298.892558] x14: 6e61687420726567 x13: 6e6f6c20746e656d x12: ffff800011a91578
> [ 298.899689] x11: 00000000000c0000 x10: ffff8000116b18f8 x9 : ffff8000100eabe0
> [ 298.906820] x8 : ffff8000116598f8 x7 : ffff8000116b18f8 x6 : 0000000000000001
> [ 298.913951] x5 : 0000000000000001 x4 : 0000000000000001 x3 : ffff800011260000
> [ 298.921082] x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffff00080673b000
> [ 298.928214] Call trace:
> [ 298.930653] debug_dma_map_sg+0x304/0x390
> [ 298.934655] dma_map_sg_attrs+0x70/0xb0
> [ 298.938487] drm_gem_map_dma_buf+0x6c/0xf0 [drm]

I still can't reproduce this, and I think this might be why. I am saving
the stream to a file and then replaying on my PC. You seem to be sending
it to a display. The DMA warning looks to be coming from DRM side, not
CSI.

Anyway, I think this is not a big problem. The UDMA driver simply does
not populate max_segment_size, so dma_get_max_seg_size() returns 64 KiB
as default. The hardware can actually support virtually unlimited
segment size for TR mode and 2^27 bytes for packet mode. CSI uses TR
mode so it can certainly handle more than 64 KiB long segments. I am not
sure which mode DRM uses but 2^27 is still much bigger than 1900544.

Long story short, this warning has little to do with the CSI patches
here and can be safely ignored IMO.

> [ 298.943185] __map_dma_buf+0x28/0x80
> [ 298.946756] dma_buf_map_attachment+0xe4/0x220
> [ 298.951191] vb2_dc_map_dmabuf+0x3c/0x150
> [ 298.955194] __prepare_dmabuf+0x1dc/0x514
> [ 298.959197] __buf_prepare+0x1a0/0x25c
> [ 298.962938] vb2_core_qbuf+0x3d4/0x72c
> [ 298.966679] vb2_qbuf+0x9c/0xf4
> [ 298.969814] vb2_ioctl_qbuf+0x68/0x7c
> [ 298.973468] v4l_qbuf+0x54/0x70
> [ 298.976603] __video_do_ioctl+0x194/0x400
> [ 298.980603] video_usercopy+0x374/0xa14
> [ 298.984431] video_ioctl2+0x24/0x4c
> [ 298.987912] v4l2_ioctl+0x4c/0x70
> [ 298.991222] __arm64_sys_ioctl+0xb4/0xfc
> [ 298.995138] invoke_syscall+0x50/0x120
> [ 298.998885] el0_svc_common.constprop.0+0x68/0x104
> [ 299.003667] do_el0_svc+0x30/0x9c
> [ 299.006976] el0_svc+0x2c/0x54
> [ 299.010025] el0_sync_handler+0x1a8/0x1ac
> [ 299.014025] el0_sync+0x198/0x1c0
> [ 299.017333] irq event stamp: 98582
> [ 299.020727] hardirqs last enabled at (98581): [<ffff8000100ec2bc>] console_unlock+0x53c/0x6b4
> [ 299.029325] hardirqs last disabled at (98582): [<ffff800010be4bd4>] el1_dbg+0x24/0xa0
> [ 299.037144] softirqs last enabled at (98568): [<ffff800010010ba0>] __do_softirq+0x500/0x6bc
> [ 299.045565] softirqs last disabled at (98413): [<ffff80001005d504>] __irq_exit_rcu+0x1d4/0x1e0
> [ 299.054164] ---[ end trace bfe019acb2a9a04f ]---
>
> I get a warning from media graph walk:

Fixed this one. It was caused by the graph_mutex not being held.

>
> [ 299.066357] WARNING: CPU: 1 PID: 605 at drivers/media/mc/mc-entity.c:343 media_graph_walk_next+0x268/0x2cc
> [ 299.076005] Modules linked in: ov5640 j721e_csi2rx cdns_csi2rx cdns_dphy v4l2_fwnode v4l2_async tidss ti_tfp410 tc358767 display_connector cdns_mhdp8546 panel_simple
> drm_kms_helper drm drm_panel_orientation_quirks cfbfillrect cfbimgblt cfbcopyarea phy_j721e_wiz phy_cadence_torrent
> [ 299.100889] CPU: 1 PID: 605 Comm: cam-mplex.py Tainted: G W 5.13.0-rc4-00417-g3331992006e9 #3
> [ 299.110698] Hardware name: Texas Instruments K3 J721E SoC (DT)
> [ 299.116518] pstate: 60000005 (nZCv daif -PAN -UAO -TCO BTYPE=--)
> [ 299.122513] pc : media_graph_walk_next+0x268/0x2cc
> [ 299.127295] lr : media_graph_walk_next+0x264/0x2cc
> [ 299.132076] sp : ffff800014dcfa40
> [ 299.135382] x29: ffff800014dcfa40 x28: 0000000000000000 x27: 0000000040045612
> [ 299.142514] x26: 0000000000000001 x25: ffff800010d890f0 x24: ffff0008055c8148
> [ 299.149645] x23: ffff0008055c8148 x22: ffff80001182bc40 x21: ffff80001163e2e8
> [ 299.156776] x20: ffff80001182bbd0 x19: ffff0008055c8cb8 x18: 0000000000000000
> [ 299.163907] x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000028
> [ 299.171037] x14: 0000000000000002 x13: 0000000000007e6f x12: 0000000000000002
> [ 299.178169] x11: 0000000000040464 x10: 00000000916d3a5c x9 : ffff8000093110c0
> [ 299.185301] x8 : ffff000807583d88 x7 : 0000000000000000 x6 : ffff00080673b900
> [ 299.192431] x5 : 000000000000000a x4 : ffff000807583d80 x3 : ffff800011260000
> [ 299.199562] x2 : 00000000000000c0 x1 : 00000000000000c0 x0 : 0000000000000000
> [ 299.206693] Call trace:
> [ 299.209133] media_graph_walk_next+0x268/0x2cc
> [ 299.213568] ti_csi2rx_start_streaming+0xe0/0x5c8 [j721e_csi2rx]
> [ 299.219569] vb2_start_streaming+0x70/0x160
> [ 299.223745] vb2_core_streamon+0x9c/0x1a0
> [ 299.227745] vb2_ioctl_streamon+0x68/0xbc
> [ 299.231747] v4l_streamon+0x30/0x40
> [ 299.235230] __video_do_ioctl+0x194/0x400
> [ 299.239230] video_usercopy+0x374/0xa14
> [ 299.243058] video_ioctl2+0x24/0x4c
> [ 299.246539] v4l2_ioctl+0x4c/0x70
> [ 299.249847] __arm64_sys_ioctl+0xb4/0xfc
> [ 299.253764] invoke_syscall+0x50/0x120
> [ 299.257508] el0_svc_common.constprop.0+0x68/0x104
> [ 299.262291] do_el0_svc+0x30/0x9c
> [ 299.265599] el0_svc+0x2c/0x54
> [ 299.268648] el0_sync_handler+0x1a8/0x1ac
> [ 299.272647] el0_sync+0x198/0x1c0
> [ 299.275956] irq event stamp: 98754
> [ 299.279349] hardirqs last enabled at (98753): [<ffff800010bf216c>] _raw_spin_unlock_irqrestore+0x9c/0xc0
> [ 299.288900] hardirqs last disabled at (98754): [<ffff800010be4bd4>] el1_dbg+0x24/0xa0
> [ 299.296716] softirqs last enabled at (98606): [<ffff800010010ba0>] __do_softirq+0x500/0x6bc
> [ 299.305138] softirqs last disabled at (98585): [<ffff80001005d504>] __irq_exit_rcu+0x1d4/0x1e0
>
> Unloading the modules gives me:

Fixed this too. Both cdns-csi2rx and j721e-csi2rx share the same power
domain, but only j721e-csi2rx contains the power-domains property. If
you load cdns-csi2rx before j721e-csi2rx then the PD is not active and
register access causes an abort. Fixed by adding power-domains to the
csi-bridge node.

>
> ERROR: Unhandled External Abort received on 0x80000001 from S-EL1
> ERROR: exception reason=0 syndrome=0xbf000000
> Unhandled Exception from EL1
[...]

--
Regards,
Pratyush Yadav
Texas Instruments Inc.