On 2/24/20 4:07 AM, Maxime Ripard wrote:
static void vc4_hdmi_encoder_enable(struct drm_encoder *encoder) { struct drm_display_mode *mode = &encoder->crtc->state->adjusted_mode; @@ -1314,6 +1438,92 @@ static int vc4_hdmi_init_resources(struct vc4_hdmi *vc4_hdmi) return 0; }
This function fails on my Raspberry Pi 4 running patched 5.6-rc6. The errors printed to syslog are as follows:
[ 15.167559] vc4-drm gpu: [drm] *ERROR* fbdev: Failed to setup generic emulation (ret=-22)
[ 46.116273] WARNING: CPU: 2 PID: 1057 at drivers/gpu/drm/vc4/vc4_hdmi_phy.c:414 vc5_hdmi_phy_init+0x7b4/0x2078 [vc4]
[ 47.127798] Timeout waiting for VC4_HDMI_SCHEDULER_CONTROL_HDMI_ACTIVE
[ 47.127865] WARNING: CPU: 1 PID: 1057 at drivers/gpu/drm/vc4/vc4_hdmi.c:652 vc4_hdmi_encoder_enable+0x1518/0x1e10 [vc4]
[ 47.128353] WARNING: CPU: 1 PID: 1057 at drivers/gpu/drm/vc4/vc4_hdmi.c:671 vc4_hdmi_encoder_enable+0x18c8/0x1e10 [vc4]
-----------------------------------------------------------------
Backtrace:
[ 46.116373] pc : vc5_hdmi_phy_init+0x7b4/0x2078 [vc4]
[ 46.116386] lr : vc4_hdmi_encoder_enable+0x1cc/0x1e10 [vc4]
[ 46.116440] vc5_hdmi_phy_init+0x7b4/0x2078 [vc4]
[ 46.116451] vc4_hdmi_encoder_enable+0x1cc/0x1e10 [vc4]
[ 46.116497] vc4_atomic_complete_commit+0x3f0/0x530 [vc4]
[ 46.116508] vc4_atomic_commit+0x1d8/0x1f8 [vc4]
The specific offending conditional (before the warning on line 652) under vc4_hdmi_encoder_enable() in drm/vc4/vc4_hdmi.c:
645 if (vc4_encoder->hdmi_monitor) {
646 HDMI_WRITE(HDMI_SCHEDULER_CONTROL,
647 HDMI_READ(HDMI_SCHEDULER_CONTROL) |
648 VC4_HDMI_SCHEDULER_CONTROL_MODE_HDMI);
649 650 ret = wait_for(HDMI_READ(HDMI_SCHEDULER_CONTROL) &
651 VC4_HDMI_SCHEDULER_CONTROL_HDMI_ACTIVE, 1000);
652 WARN_ONCE(ret, "Timeout waiting for "
653 "VC4_HDMI_SCHEDULER_CONTROL_HDMI_ACTIVE\n");
Which causes vc4_hdmi_encoder_enable() to fail.
The failure of vc5_hdmi_phy_init() earlier left the phy inactive, causing the encoder enabling function above to fail.
The offending code around line 414 in vc4_hdmi_phy.c, under vc5_hdmi_phy_init() reads
411 HDMI_WRITE(HDMI_TX_PHY_TMDS_CLK_WORD_SEL, word_sel);
412 413 HDMI_WRITE(HDMI_TX_PHY_CTL_3,
414 VC4_SET_FIELD(phy_get_cp_current(vco_freq),
415 VC4_HDMI_TX_PHY_CTL_3_ICP) |
416 VC4_SET_FIELD(1, VC4_HDMI_TX_PHY_CTL_3_CP) |
417 VC4_SET_FIELD(1, VC4_HDMI_TX_PHY_CTL_3_CP1) |
418 VC4_SET_FIELD(3, VC4_HDMI_TX_PHY_CTL_3_CZ) |
419 VC4_SET_FIELD(4, VC4_HDMI_TX_PHY_CTL_3_RP) |
420 VC4_SET_FIELD(6, VC4_HDMI_TX_PHY_CTL_3_RZ));
As the hdmi-related timeout occurs 30 seconds after the drm failure, I'm bound to believe that the timeout occurs due to the drm setup failure leaving nothing for the phy functions to act on.
Earlier in the syslog an error potentially related to the VC4_SET_FIELD(phy_get_cp_current(vco_freq) failure:
[ 3.729745] raspberrypi-clk raspberrypi-clk: Missing firmware node
[ 3.743915] raspberrypi-clk: probe of raspberrypi-clk failed with error -2
I thought the patch series added the firmware node? Perhaps the bcm2835 clock stub in the bcm2835 common dts is not being imported in the bcm2711 devicetree?
The result is a connected hdmi monitor has signal from the simple framebuffer until the failed modeswitch, upon which it loses signal.
I'm looking around for these possibilities in the kernel tree, though I thought the code worked as it was?
I did have to disable CONFIG_DRM_VC4_HDMI_CEC in Kconfig to get the patches to compile as another user in linux-arm-kernel discovered the CEC code relied on removed functions (Jian-Hong Pan).
I still hope these patches can be cleaned up/fixed to make the 5.7 merge window.
Daniel Rodriguez