Re: [BUG] wifi: rtw88: Hard system freeze on RTL8821CE when power_save is enabled (LPS/ASPM conflict)

[Bitterblue Smith]

On 27/03/2026 01:52, LB F wrote:
> Hi Ping-Ke,
> 
> This is Oleksandr Havrylov again. Thank you for the ASPM/LPS Deep
> quirk and the rate validation patches — they are both working correctly
> (zero h2c timeouts, zero lps failures, zero mac80211 warnings).
> 
> However, I'm experiencing a different, separate bug that causes kernel
> oops and makes the system completely unresponsive, requiring a hard
> power-off. After disassembling the crash site, I believe I've found
> the root cause.
> 
> == Summary ==
> 
> When firmware sends a C2H_ADAPTIVITY (0x37) command to an RTL8821CE
> adapter, rtw_fw_adaptivity_result() dereferences rtwdev->chip->edcca_th
> without a NULL check. The RTL8821C chip_info (rtw8821c_hw_spec) does
> not define edcca_th, so the pointer is NULL, causing a kernel oops.
> 
> The crash occurs on the phy0 workqueue while holding rtwdev->mutex,
> which never gets released. This causes all subsequent processes that
> touch the network stack to hang in uninterruptible D-state, making
> the system completely unresponsive and requiring a hard power-off.
> 
> == Root cause analysis ==
> 
> rtw_fw_adaptivity_result() in fw.c (line ~282):
> 
>   static void rtw_fw_adaptivity_result(struct rtw_dev *rtwdev, u8 *payload,
>                                        u8 length)
>   {
>       const struct rtw_hw_reg_offset *edcca_th = rtwdev->chip->edcca_th;
>       ...
>       rtw_dbg(rtwdev, RTW_DBG_ADAPTIVITY, "Reg Setting: L2H %x H2L %x\n",
>               rtw_read32_mask(rtwdev, edcca_th[EDCCA_TH_L2H_IDX].hw_reg.addr,
>                                              ^^^^^^^^^ NULL dereference here
>                               edcca_th[EDCCA_TH_L2H_IDX].hw_reg.mask),
>               ...
> 
> The RTL8822C defines .edcca_th = rtw8822c_edcca_th in its chip_info,
> but RTL8821C does not set this field at all — it remains NULL.
> 
> I verified this by disassembling the compiled rtw_core.ko module:
> 
>   Crash RIP: rtw_fw_c2h_cmd_handle+0x127
>   Address:   0x1d527 = movl (%r12), %esi
> 
>   R12 is loaded at +0xe5 (0x1d4e5):
>     movq 0x140(%rax), %r12    ; rax = rtwdev->chip
>                               ; 0x140 = offset of edcca_th in rtw_chip_info
>                               ; R12 = chip->edcca_th = NULL for 8821c
> 
>   The function is entered via:
>     +0xd8 (0x1d4d8): cmpl $0x37, %ecx  ; c2h->id == C2H_ADAPTIVITY (0x37)
> 
>   With R12 = 0, the instruction at +0x127:
>     movl (%r12), %esi   ; reads from address 0x0 → NULL pointer dereference
> 
> I also confirmed that rtw8821c_hw_spec in the mainline kernel
> (torvalds/linux master, rtw8821c.c) does NOT set .edcca_th.
> 
> == Reproduction ==
> 
> The crash is highly reproducible: it occurred in 4 out of 7 recent
> boots. It happens during normal active usage with no specific trigger.
> 
>   boot  date/time of crash         uptime at crash
>   -5    2026-03-25 00:58:06        ~2 min
>   -4    2026-03-25 21:32:00        ~6h
>   -3    2026-03-26 00:28:14        ~2.5h
>   -1    2026-03-27 00:56:58        ~23.5h
> 
> Both ASPM and LPS Deep are disabled via the DMI quirk. The crash
> occurs every time with the same pattern and same RIP offset (+0x127).
> 
> == Crash pattern ==
> 
> Every crash follows the same sequence:
> 
>   1) Burst of 50-60 "unused phy status page" messages in ~1 second:
> 
>      rtw_8821ce 0000:13:00.0: unused phy status page (8)
>      rtw_8821ce 0000:13:00.0: unused phy status page (2)
>      ... (50+ more within same second)
> 

It looks like the firmware is not sending C2H_ADAPTIVITY (unexpected
for RTL8821CE), but rather you are getting garbage RX data. I am
curious what kind of garbage it is. Can you try this?


diff --git a/drivers/net/wireless/realtek/rtw88/rtw8821c.c b/drivers/net/wireless/realtek/rtw88/rtw8821c.c
index da67a6845fd5..aae246c2bc8e 100644
--- a/drivers/net/wireless/realtek/rtw88/rtw8821c.c
+++ b/drivers/net/wireless/realtek/rtw88/rtw8821c.c
@@ -665,6 +665,7 @@ static void query_phy_status(struct rtw_dev *rtwdev, u8 *phy_status,
 			     struct rtw_rx_pkt_stat *pkt_stat)
 {
 	u8 page;
+	u8 *rxdesc = phy_status - rtwdev->chip->rx_pkt_desc_sz - pkt_stat->shift;
 
 	page = *phy_status & 0xf;
 
@@ -677,6 +678,10 @@ static void query_phy_status(struct rtw_dev *rtwdev, u8 *phy_status,
 		break;
 	default:
 		rtw_warn(rtwdev, "unused phy status page (%d)\n", page);
+		print_hex_dump(KERN_INFO, "", DUMP_PREFIX_OFFSET, 4, 4,
+			       rxdesc, 56, true);
+		print_hex_dump(KERN_INFO, "", DUMP_PREFIX_OFFSET, 16, 1,
+			       rxdesc, 40, true);
 		return;
 	}
 }