Accessing the NVMe flash drive of the April 2016 MacBook 12" (A1534)

Accessing the NVMe flash drive of the April 2016 MacBook 12" (A1534)

[Carina Willbold]

Dear NVMe hackers,

I'm trying to access the built-in flash drive of the new April 2016 
MacBook 12" (A1534). Based on suggestions for older MacBooks on this 
list, I loaded the module nvme, and, since the flash drive reports an 
incorrect PCI device class, manually bound it with:

     echo 106b 2003 > /sys/bus/pci/drivers/nvme/new_id

The dmesg output is non-informative, it just reports that the probe 
failed. The flash drive reports under lspci -nn as:

     01:00.0 Mass storage controller [0180]: Apple Inc. Device 
[106b:2003] (rev 11)

Note that this is different from previous versions of the MacBook 
(where the PCI ID was 106b:2001). The flash drive is detected from the 
Grub shell and works just fine from Mac OS X.

I know a bit of C, but have zero experience debugging or programming 
drivers. I'm willing to learn! What should I try next?

Thank you for any advice :-).

LLAP
Carina

PS: I also posted this at 
http://unix.stackexchange.com/questions/283154/accessing-the-nvme-flash-drive-of-the-april-2016-macbook-12-a1534. 
However I believe that this list is better suited.

Accessing the NVMe flash drive of the April 2016 MacBook 12" (A1534)

[Carina Willbold]

Dear NVMe hackers,

Am 2016-05-15 22:19, schrieb Carina Willbold:
> I'm trying to access the built-in flash drive of the new April 2016
> MacBook 12" (A1534).

please forgive me for commiting a basic newbie mistake: I went with the
default Arch Linux kernel (4.5.1) instead of going for the newest one.
I'm sorry for the noise.

With 4.6.0-rc7, after binding using

     echo 106b 2003 > /sys/bus/pci/drivers/nvme/new_id

the flash disk *is* recognized:

     [  126.904767] nvme nvme0: pci function 0000:01:00.0
     [  127.117492]  nvme0n1: p1 p2 p3 p4

However actually trying to access one of the partitions (for instance, 
trying
to mount them or executing head -c 20 /dev/nvme0n1p1) results in 
timeouts:

     [  187.354937] nvme nvme0: I/O 0 QID 0 timeout, reset controller
     [  187.365069] nvme nvme0: I/O 0 QID 1 timeout, aborting
     [  200.322625] nvme nvme0: completing aborted command with status: 
0000
     [  200.322634] nvme nvme0: completing aborted command with status: 
0000
     [  269.366379] nvme nvme0: I/O 1 QID 1 timeout, reset controller
     [  308.027258] nvme nvme0: completing aborted command with status: 
0000
     [  308.027295] nvme nvme0: completing aborted command with status: 
0000
     [  308.027314] nvme nvme0: completing aborted command with status: 
0000

At one point printing the first few bytes of /dev/nvme0n1p1 actually 
succeeded
(giving the expected EFI/FAT32 headers).

I'm happy to provide any additional information or start hacking the 
driver code
(if given a starting point).

LLAP,
Carina

Accessing the NVMe flash drive of the April 2016 MacBook 12" (A1534) [SOLVED]

[Carina Willbold]

Dear NVMe hackers,

I promise that this is the last time I'll spam this list.

Am 2016-05-16 01:48, schrieb Carina Willbold:
> Am 2016-05-15 22:19, schrieb Carina Willbold:
>> I'm trying to access the built-in flash drive of the new April 2016
>> MacBook 12" (A1534).
[...]
> However actually trying to access one of the partitions (for 
> instance, trying
> to mount them or executing head -c 20 /dev/nvme0n1p1) results in 
> timeouts:
[...]

After a reboot, the timeouts vanished.

I can now happily confirm that accessing the flash drive works using 
kernel 4.6.0-rc7 and manually binding using:

     echo 106b 2003 > /sys/bus/pci/drivers/nvme/new_id

No controller resets or other IO errors occur even with (moderately) 
drive-intense activities like compiling and installing packages.

I'll submit a patch which will make the manual binding superfluous 
tomorrow.

LLAP,
Carina

Accessing the NVMe flash drive of the April 2016 MacBook 12" (A1534) [SOLVED]

[Keith Busch]

On Mon, May 16, 2016@04:59:51AM +0200, Carina Willbold wrote:
> >Am 2016-05-15 22:19, schrieb Carina Willbold:
> >However actually trying to access one of the partitions (for instance,
> >trying
> >to mount them or executing head -c 20 /dev/nvme0n1p1) results in timeouts:
> [...]
> 
> After a reboot, the timeouts vanished.
> 
> I can now happily confirm that accessing the flash drive works using kernel
> 4.6.0-rc7 and manually binding using:
> 
>     echo 106b 2003 > /sys/bus/pci/drivers/nvme/new_id
> 
> No controller resets or other IO errors occur even with (moderately)
> drive-intense activities like compiling and installing packages.
> 
> I'll submit a patch which will make the manual binding superfluous tomorrow.

It was starting to sound like you might need to copy the current Apple
driver quirk for the 0x2001 device, but if it's working without, you
can just append the VID/DID to nvme's pci_device_id table.

Accessing the NVMe flash drive of the April 2016 MacBook 12" (A1534) [SOLVED]

[Carina Willbold]

Dear Keith,

Am 2016-05-16 16:46, schrieb Keith Busch:
> On Mon, May 16, 2016@04:59:51AM +0200, Carina Willbold wrote:
[ Trying to get the flash disk on the new MacBook recognized ]
>> I can now happily confirm that accessing the flash drive works using 
>> kernel
>> 4.6.0-rc7 and manually binding using:
>>
>>     echo 106b 2003 > /sys/bus/pci/drivers/nvme/new_id
>>
>> No controller resets or other IO errors occur even with (moderately)
>> drive-intense activities like compiling and installing packages.
>>
>> I'll submit a patch which will make the manual binding superfluous 
>> tomorrow.
>
> It was starting to sound like you might need to copy the current 
> Apple
> driver quirk for the 0x2001 device, but if it's working without, you
> can just append the VID/DID to nvme's pci_device_id table.

yes, thought so too, but it's working without. I can confirm that the 
flash drive works fine and is automatically detected with the (trivial) 
patch given in the other mail. The patch is against 4.6, but it applies 
cleanly against the current linux-next and axboe/linux-block.

In my boot scripts, I had to include a short sleep because I was too 
lazy to properly wait for the device files to come into existence.

Thank you for your answer!

LLAP,
Carina