On 22-11-17 17:20, Stefan Wahren wrote:
Hi,
Stefan Wahren <stefan.wahren@xxxxxxxx> hat am 12. November 2017 um 17:50 geschrieben:
Hi,
i discovered a random oops during probe of brcmfmac on Raspberry Pi 3 in yesterdays kernelci run for net-next [1]. I need to point out there is no DT entry for the wifi chip on Raspberry Pi 3 in the lack of a driver for the necessary GPIO expander. So the "HT Avail timeout" is expected.
I was also able to trigger this oops by calling "modprobe brcmfmac" on my Raspberry Pi 3 with latest linux-next.
Any help to fix this is appreciated.
[1] - https://storage.kernelci.org/net-next/master/v4.14-rc8-2221-ga8a6f1e4ea78/arm64/defconfig+kselftest/lab-baylibre/boot-bcm2837-rpi-3-b.txt
the issue still persists in linux-next-20171122:
[ 817.659632] brcmfmac: brcmf_fw_map_chip_to_name: using brcm/brcmfmac43430-sdio.bin for chip 0x00a9a6(43430) rev 0x000001
[ 817.672739] brcmfmac mmc1:0001:1: Direct firmware load for brcm/brcmfmac43430-sdio.txt failed with error -2
[ 818.708435] brcmfmac: brcmf_sdio_htclk: HT Avail timeout (1000000): clkctl 0x50
[ 819.722103] brcmfmac: brcmf_sdio_htclk: HT Avail timeout (1000000): clkctl 0x50
[ 819.755321] Unable to handle kernel paging request at virtual address 726f6674616cd8
[ 819.763255] Mem abort info:
[ 819.766131] ESR = 0x96000004
[ 819.769240] Exception class = DABT (current EL), IL = 32 bits
[ 819.775290] SET = 0, FnV = 0
[ 819.778416] EA = 0, S1PTW = 0
[ 819.781605] Data abort info:
[ 819.784552] ISV = 0, ISS = 0x00000004
[ 819.788495] CM = 0, WnR = 0
[ 819.791532] [00726f6674616cd8] address between user and kernel address ranges
[ 819.798805] Internal error: Oops: 96000004 [#1] PREEMPT SMP
[ 819.804469] Modules linked in: brcmfmac cfg80211 brcmutil vc4 cec drm_kms_helper drm bcm2835_rng rfkill rng_core smsc95xx usbnet bcm2835_dma crc32_ce pwm_bcm2835 i2c_bcm2835 ip_tables x_tables ipv6 [last unloaded: cfg80211]
[ 819.824687] CPU: 3 PID: 92 Comm: kworker/3:1 Not tainted 4.14.0-next-20171122 #1
[ 819.832207] Hardware name: Raspberry Pi 3 Model B Rev 1.2 (DT)
[ 819.838152] Workqueue: events request_firmware_work_func
[ 819.843561] task: ffff80003833aa00 task.stack: ffff000009498000
[ 819.849583] pstate: 60000005 (nZCv daif -PAN -UAO)
[ 819.854460] pc : mutex_lock+0xc/0x40
[ 819.858101] lr : device_release_driver_internal+0x38/0x1f0
[ 819.863676] sp : ffff00000949bbf0
[ 819.867045] x29: ffff00000949bbf0 x28: 0000000000000000
[ 819.872451] x27: ffff00000931bd20 x26: ffff000008f9fbb0
[ 819.877856] x25: ffff80003833aa00 x24: ffff800036e20f80
[ 819.883261] x23: 0000000000000000 x22: ffff800039fd3d00
[ 819.888665] x21: 0000000000000000 x20: 6d726f6674616cd8
[ 819.894068] x19: 6d726f6674616c78 x18: 0000000000000000
[ 819.899473] x17: 0000ffffa52ba930 x16: ffff000008256b30
[ 819.904877] x15: 0000000000000000 x14: 0000000000000400
[ 819.910282] x13: 0000000000000400 x12: 0000000000000000
[ 819.915685] x11: 0000000000000000 x10: 0000000000000001
[ 819.921087] x9 : 0000000000000001 x8 : ffff800032fdec08
[ 819.926493] x7 : ffff800032d7c158 x6 : 0000000000000000
[ 819.931895] x5 : ffff800032d7c288 x4 : 0000000000000000
[ 819.937299] x3 : ffff80003833aa00 x2 : 0000000000000000
[ 819.942705] x1 : ffff80003833aa00 x0 : 6d726f6674616cd8
[ 819.948112] Process kworker/3:1 (pid: 92, stack limit = 0xffff000009498000)
[ 819.955188] Call trace:
[ 819.957681] mutex_lock+0xc/0x40
[ 819.960967] device_release_driver+0x14/0x20
[ 819.965450] brcmf_sdio_firmware_callback+0x160/0x578 [brcmfmac]
[ 819.971689] brcmf_fw_request_nvram_done+0x258/0x5e8 [brcmfmac]
[ 819.977717] request_firmware_work_func+0x34/0x60
[ 819.982510] process_one_work+0x1cc/0x328
[ 819.986592] worker_thread+0x50/0x450
[ 819.990320] kthread+0xf8/0x128
[ 819.993516] ret_from_fork+0x10/0x18
[ 819.997154] Code: d65f03c0 d5384101 d2800002 f9800011 (c85ffc03)
[ 820.003351] ---[ end trace f4bff35241327a18 ]---
seems like a use-after-free. We do a device_release_driver() twice. Once
for sdio func #1 and for sdio func #2. This was introduced by:
commit 7a51461fc2da82a6c565a3ee65c41c197f28225d
Author: Arend Van Spriel <arend.vanspriel@xxxxxxxxxxxx>
Date: Mon Jun 12 12:47:34 2017 +0100
brcmfmac: unbind all devices upon failure in firmware callback
What we do is:
device_release_driver(dev);
device_release_driver(&sdiodev->func[2]->dev);
with the assumption that dev == &sdiodev->func[1]->dev. I wonder if that
is always true. The error print did not make it in your log. Maybe we
should make it explicit here (see below).
Regards,
Arend
---
diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.c
b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.c
index 613caca..0fda9a4 100644
--- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.c
+++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.c
@@ -4096,7 +4096,7 @@ static void brcmf_sdio_firmware_callback(struct
device *dev, int err,
sdio_release_host(sdiodev->func[1]);
fail:
brcmf_dbg(TRACE, "failed: dev=%s, err=%d\n", dev_name(dev), err);
- device_release_driver(dev);
+ device_release_driver(&sdiodev->func[1]->dev);
device_release_driver(&sdiodev->func[2]->dev);
}
