Trouble with allocating memory on platform driver
Hello list, Trying to create a device driver for a raspberry pi zero (BCM2835) to control a gpio and an spi device. This eventually will be done in a FIQ interrupt handler. A simplified version of this batradio-driver can be found at: https://github.com/SietseAchterop/Batradio/blob/master/batradio_module/alloc... The problem is that modprobe-ing this driver directly yields the following in dmesg. Apr 4 16:50:23 raspberrypi kernel: [ 139.636850] batradio: loading out-of-tree module taints kernel. Apr 4 16:50:23 raspberrypi kernel: [ 139.653580] ------------[ cut here ]------------ Apr 4 16:50:23 raspberrypi kernel: [ 139.653650] WARNING: CPU: 0 PID: 1208 at include/linux/dma-mapping.h:516 init_module+0x164/0x1e0 [batradio] Apr 4 16:50:23 raspberrypi kernel: [ 139.653664] Modules linked in: batradio(O+) fuse rfcomm cmac bnep hci_uart btbcm serdev bluetooth ecdh_generic 8021q garp stp llc sg uas brcmfmac brcmutil sha256_generic cfg80211 rfkill snd_bcm2835(C) raspberrypi_hwmon snd_pcm hwmon snd_timer snd bcm2835_v4l2(C) bcm2835_codec(C) v4l2_mem2mem bcm2835_mmal_vchiq(C) v4l2_common videobuf2_vmalloc videobuf2_dma_contig videobuf2_memops videobuf2_v4l2 videobuf2_common videodev media vc_sm_cma(C) fixed uio_pdrv_genirq uio i2c_dev ip_tables x_tables ipv6 Apr 4 16:50:23 raspberrypi kernel: [ 139.653835] CPU: 0 PID: 1208 Comm: modprobe Tainted: G C O 4.19.108+ #1 Apr 4 16:50:23 raspberrypi kernel: [ 139.653844] Hardware name: BCM2835 Apr 4 16:50:23 raspberrypi kernel: [ 139.653897] [<c00184f8>] (unwind_backtrace) from [<c0015064>] (show_stack+0x20/0x24) Apr 4 16:50:23 raspberrypi kernel: [ 139.653931] [<c0015064>] (show_stack) from [<c0753394>] (dump_stack+0x20/0x28) Apr 4 16:50:23 raspberrypi kernel: [ 139.653970] [<c0753394>] (dump_stack) from [<c0025b58>] (__warn.part.3+0xb8/0xe0) Apr 4 16:50:23 raspberrypi kernel: [ 139.653998] [<c0025b58>] (__warn.part.3) from [<c0025cf8>] (warn_slowpath_null+0x50/0x5c) Apr 4 16:50:23 raspberrypi kernel: [ 139.654037] [<c0025cf8>] (warn_slowpath_null) from [<bf698164>] (init_module+0x164/0x1e0 [batradio]) Apr 4 16:50:23 raspberrypi kernel: [ 139.654110] [<bf698164>] (init_module [batradio]) from [<c000ae8c>] (do_one_initcall+0x4c/0x234) Apr 4 16:50:23 raspberrypi kernel: [ 139.654151] [<c000ae8c>] (do_one_initcall) from [<c00a7898>] (do_init_module+0x6c/0x1f8) Apr 4 16:50:23 raspberrypi kernel: [ 139.654181] [<c00a7898>] (do_init_module) from [<c00a9cdc>] (load_module+0x21ec/0x24cc) Apr 4 16:50:23 raspberrypi kernel: [ 139.654209] [<c00a9cdc>] (load_module) from [<c00aa1f8>] (sys_finit_module+0xcc/0xec) Apr 4 16:50:23 raspberrypi kernel: [ 139.654232] [<c00aa1f8>] (sys_finit_module) from [<c0009000>] (ret_fast_syscall+0x0/0x28) Apr 4 16:50:23 raspberrypi kernel: [ 139.654244] Exception stack(0xcc037fa8 to 0xcc037ff0) Apr 4 16:50:23 raspberrypi kernel: [ 139.654261] 7fa0: 93da3c00 00401760 00000003 0002d064 00000000 0002ec3c Apr 4 16:50:23 raspberrypi kernel: [ 139.654279] 7fc0: 93da3c00 00401760 00000000 0000017b 00402f10 00000000 00402e68 00000000 Apr 4 16:50:23 raspberrypi kernel: [ 139.654294] 7fe0: bed56338 bed56328 00022cb8 b6c26af0 Apr 4 16:50:23 raspberrypi kernel: [ 139.654307] ---[ end trace deb280ac68ce7734 ]--- Apr 4 16:50:23 raspberrypi kernel: [ 139.654330] platform batradio__.0: coherent DMA mask is unset Apr 4 16:50:23 raspberrypi kernel: [ 139.654344] platform batradio__.0: Couldn't allocate memory! The crash is at the very beginning of the init function of the module, here is the first part of that function: pdev = platform_device_register_simple("batradio__", 0, NULL, 0); if (IS_ERR(pdev)) return PTR_ERR(pdev); batradio_data = devm_kzalloc(&pdev->dev, sizeof(*batradio_data), GFP_KERNEL); if (!batradio_data) return -ENOMEM; batradio_data->fiq_base = dma_zalloc_coherent(&pdev->dev, FIQ_BUFFER_SIZE, &batradio_data->dma_handle, GFP_KERNEL); if (!batradio_data->fiq_base) { dev_err(&pdev->dev, "Couldn't allocate memory!\n"); return -ENOMEM; } Note that the crash seems to occur because of the call to dma_zalloc_coherent. The allocation does not work, FIQ_BUFFER_SIZE is 256*1024. and this is directly after reboot. Also note that I start the rpizero only with the console, so no X. And login with ssh. Why is this happening? What does "coherent DMA mask is unset" means. I do not intend to use dma, but I understood that allocating in this way is what I should do if I want to use this data also from my FIQ handler. Does it has to do with the use of platform_device_register_simple? Maybe the device structure isn't initialized properly for this? I actually use this function because I don't know which compatible value I have to use when using the devicetree, everything I tried failed. I also read the following in platform_device.h about platform_device_register_simple: In particular, when such drivers are built as modules, they can't be "hotplugged". Is this relevant. is using modprobe meant by this? Thanks in advance, Sietse
On 4/4/20 5:21 PM, s.achterop@rug.nl wrote:
Hello list,
Trying to create a device driver for a raspberry pi zero (BCM2835) to control a gpio and an spi device. This eventually will be done in a FIQ interrupt handler. A simplified version of this batradio-driver can be found at: https://github.com/SietseAchterop/Batradio/blob/master/batradio_module/alloc...
Regrettably no answers until now. I only can add that the same code fragments are used in an example that I found on https://bootlin.com/blog/fiq-handlers-in-the-arm-linux-kernel/ where it is contained in the probe function of the platform driver. Is there maybe a better place to ask for help here? Thanks, Sietse
On 4/15/20 11:29 AM, s.achterop@rug.nl wrote:
On 4/4/20 5:21 PM, s.achterop@rug.nl wrote:
Hello list,
Trying to create a device driver for a raspberry pi zero (BCM2835) to control a gpio and an spi device. This eventually will be done in a FIQ interrupt handler. A simplified version of this batradio-driver can be found at: https://github.com/SietseAchterop/Batradio/blob/master/batradio_module/alloc...
I found a working solution. I actually don't know what I am doing but it works. Before the call to dma_zalloc_coherent I added: pdev->dev.coherent_dma_mask = 0xffffffff; Now it works. Regards, Sietse
On Fri, 17 Apr 2020 10:47:33 +0200, "s.achterop@rug.nl" said:
I found a working solution. I actually don't know what I am doing but it works. Before the call to dma_zalloc_coherent I added:
pdev->dev.coherent_dma_mask = 0xffffffff;
DMA_BIT_MASK(32) rather than a hex constant would be a better choice.
Now it works.
Note that "works" and "is correct" are two different things. That should be a mask that tells how many bits of the value are actually used by the hardware. On most 32-bit systems, 32 will be correct, but there's other values in use (on S/390, it's either 31 or 64 bits, the PowerPC can have 44, and the old PNP stuff on x86 weirds out at the 24-bit line).
participants (2)
-
s.achterop＠rug.nl -
Valdis Klētnieks