In 73545f75b6 "ahci: wait longer for link" I increased the
timeout to 40ms based on the observed behaviour of a WD disk on a
Cubietruck. Since then Karsten Merker and myself have both
observed timeouts with HGST disks (Karsten on Cubietruck, me on
Cubieboard2). Increasing the timeout to ~175ms fixes this, so go
to 200ms for a bit of headroom.
Signed-off-by: Ian Campbell <ijc@hellion.org.uk>
Cc: Karsten Merker <merker@debian.org>
Acked-by: Hans de Goede <hdegoede@redhat.com>
Commit 2faf5fb82e introduced a regression that causes a data
abort when running scsi init followed by scsi reset.
There are 2 problems with the original commit
1) ALLOC_CACHE_ALIGN_BUFFER() allocates memory on the stack but is
assigned to ataid[port] and used by other functions.
2) The function ata_scsiop_inquiry() tries to free memory which was
never allocated on the heap.
Fix these problems by using tmpid as a temporary cache aligned buffer.
Allocate memory separately for ataid[port] and re-use it if required.
Fixes: 2faf5fb82e (ahci: Fix cache align error messages)
Reported-by: Eli Nidam <elini@marvell.com>
Signed-off-by: Roger Quadros <rogerq@ti.com>
I have observed timeouts on a cubietruck.
The increase to 40ms is completely arbitrary and Works For Me(tm). I
couldn't find a good reference for how long you are supposed to wait,
although googling around it seems like tens of ms rather than single
digits is more common. I don't think there is any harm in waiting a bit
longer.
Signed-off-by: Ian Campbell <ijc@hellion.org.uk>
Align the ATA ID buffer to the cache-line boundary. This gets rid
of the below error mesages on ARM v7 platforms.
scanning bus for devices...
ERROR: v7_dcache_inval_range - start address is not aligned - 0xfee48618
ERROR: v7_dcache_inval_range - stop address is not aligned - 0xfee48818
CC: Aneesh V <aneesh@ti.com>
Signed-off-by: Roger Quadros <rogerq@ti.com>
If malloc() fails, we don't want to continue in ahci_init() and
ahci_init_one(). Also print a more informative error message on
malloc() failures.
CC: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: Roger Quadros <rogerq@ti.com>
libata already has similar functions as implemented in the ahci code.
Refactor the code to use the libata variants and remove the dependency on
ata.h. Convert some defines to use the version from libata.h. Also, remove
some unnecessary memset's of bss data.
This is a step toward hopefully merging ahci.c and dw_ahsata.c which are
essentially the same driver.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Reviewed-by: Tom Rini <trini@ti.com>
Based on Linux libata code, most drives are less than 10 sec, but some
need up to 20 sec.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Reviewed-by: Tom Rini <trini@ti.com>
Some Intel SSDs can send a COMINIT after the initial COMRESET. This causes
the link to go down and we need to re-initialize the link.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Move the link bring-up handling to a separate weak function in order to
allow platforms to override it. This is needed on highbank platform which
needs special phy handling.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
gcc 4.7 will generate unaligned accesses to local char arrays, so make
them static to avoid that.
Signed-off-by: Rob Herring <rob.herring@calxeda.com>
Reviewed-by: Tom Rini <trini@ti.com>
The AHCI driver was incorrectly using the Capabilities register NP (number
of ports) field to determine which ports to activate. This commit changes
it to correctly use the PORTS_IMPL register as a port map.
Signed-off-by: Richard Gibbs <richard.gibbs@calxeda.com>
Reviewed-by: Tom Rini <trini@ti.com>
'bool' is defined in random places. This patch consolidates them into a
single header file include/linux/types.h, using stdbool.h introduced in C99.
All other #define, typedef and enum are removed. They are all consistent with
true = 1, false = 0.
Replace FALSE, False with false. Replace TRUE, True with true.
Skip *.py, *.php, lib/* files.
Signed-off-by: York Sun <yorksun@freescale.com>
Currently, this driver uses a 28bit interface to AHCI, this
limits the number of blocks addressable to 2^28, or the max
disk size to 512(2^28) or about 137GB. This change allows
supporting drives up to about 2TB.
Testing this is a bit difficult. There is test code that
can be inserted into U-Boot that will write test patterns
into certain unused blocks. These patterns can be manually
checked using 'dd' after boot. Another way is to confirm the
original error that exposed this bug is fixed. IOW: see if
AU (Auto Update) will now work on the drive. Also, check
that there are no warning messages from the 'cgpt' utility.
Signed-off-by: Walter Murphy <wmurphy@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Writes in u-boot are so rare, and the logic to know when is
the last write and do a flush only there is sufficiently
difficult. Just do a flush after every write. This incurs,
usually, one extra flush when the rare writes do happen.
Signed-off-by: Marc Jones <marc.jones@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Add HDD handling to the SSD-only AHCI driver, by separately dealing with
spin-up and link-up.
Signed-off-by: Marc Jones <marc.jones@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Note: These are timeout values and not delay values, so the event being
timed out will complete whenever it is actually ready, with a
measurement granularity of 1 millisecond, up till the timeout value.
Therefore, there is no effect on SSD booting.
The values were determined by instrumenting the code and measuring the
actual time taken by several different models of HDD for each of the
parameters and then adding 50% more for the spinup value and just
doubling the command timeout value.
Signed-off-by: Walter Murphy <wmurphy@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Exynos5 automatically performs DMA when the SATA controller executes
commands. This adds the necessary dcache-to-memory flush &
invalidation calls to allow the DMA to properly function.
Signed-off-by: Taylor Hutt <thutt@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Update the assignment of various physical memory buffers used by the
SATA controller to explicitly be denoted as physical addresses.
The memory is identity-mapped, so these function calls are a nop, but
they provide good semantic documentation for any maintainers.
The return value of virt_to_phys() is 'unsigned long'. On machines
where sizeof(unsigned long) != sizeof(pointer), a cast through
(uintptr_t) is needed to appease the compiler due to the potential of
losing the upper 32 bits of the address.
In compilation this scenario, a physical address could be 64-bits, yet
the C pointer environment only allows 32-bit addresses; the constraint
is that pointers cannot address more than 4Gb of memory and if
virt_to_phys() ever returns an out-of-range value for the physical
address, there are issues with emmory mapping which must be solved.
However, since the memory is identify mappeed, there is no problem
introducing the cast: the original pointer will reside in 32-bits, so
the physical address will also be within in 32-bits.
Signed-off-by: Taylor Hutt <thutt@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
This fixes a spelling error in a message which can be output to the
console.
Signed-off-by: Taylor Hutt <thutt@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
This cleanup replaces the hardcoded use of '20', which represents the
number of bytes in the FIS, with sizeof(fis).
Signed-off-by: Taylor Hutt <thutt@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
In the structure returned by the ATA identify device command, there are two
fields which describe the device capacity. One is a 32 bit data type which
reports the number of sectors as a 28 bit LBA, and the other is a 64 bit data
type which is for a 48 bit LBA. If the device doesn't support 48 bit LBAs,
the small value is the only value with the correct size. If it supports more,
if the number of sectors is small enough to fit into 28 bits, both fields
reflect the correct value. If it's too large, the smaller field has 28 bits of
1s, 0xfffffff, and the other field has the correct value.
The AHCI driver is implemented by attaching to the generic SCSI code and
translating on the fly between SCSI binary data structures and AHCI data
structures. It responds to requests to execute specific SCSI commands by
executing the equivalent AHCI commands and then crafting a response which
matches what a SCSI disk would send.
The AHCI driver now considers both fields and chooses the correct one when
implementing both the SCSI READ CAPACITY (10) and READ CAPACITY (16) commands.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
The "scsi write" command requires support from underlying driver.
This CL enables SCSI_WRITE10 in AHCI driver.
Tested in U-Boot console, try to i/o with sector #64:
scsi read 1000 40 1
md.b 1000 200 # check if things are not 0xcc
mw.b 1000 cc 200 # try to fill with 0xcc
scsi write 1000 40 1
mw.b 1000 0 200 # fill with zero
md.b 1000 200 # should be all 0
scsi read 1000 40 1
md.b 1000 200 # should be all 0xcc
Signed-off-by: Hung-Te Lin <hungte@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
This command doesn't really do anything when talking to a SATA device, and
sending it confuses some of them. This change makes sending the command
optional, and defaults to not. The situations where it should be sent are not
the common case.
With the standard SSD in the machine, here are some times with the option
turned off:
1. 8277
2. 8273
3. 8050
And turned on:
1. 8303
2. 8155
3. 8276
Sending that command seems to have no meaningful effect on performance.
This fixes problems with an SSD marked Toshiba NV6424, Taiwan 11159AE P
and TC58NVG5D2FTA10.
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: Taylor Hutt <thutt@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
- print the correct speed
- print all the AHCI capability flags
(information taken from Linux kernel driver)
- clean up some comments
For example, this might show the following string:
AHCI 0001.0300 32 slots 6 ports 6 Gbps 0x3 impl SATA mode
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
Commit-Ready: Stefan Reinauer <reinauer@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Tested-by: Stefan Reinauer <reinauer@chromium.org>
- remove unused ssleep macro
- add some useful debugging information
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
The existing code waits a whole second for the AHCI controller to reset.
Instead, let's poll the status register to see if the reset has
succeeded and return earlier if possible. This brings down the time for
AHCI probing from 1s to 20ms.
Signed-off-by: Stefan Reinauer <reinauer@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
With an Intel AHCI controller, the driver does not operate properly
if the requested amount of blocks to read exceeds 255.
It is probably possible to specify 0 as the block count and the driver
will read 256 blocks, but it was decided to limit the number of blocks
read at once to 128 (it should be a power of 2 for the optimal
performance of solid state drives).
Signed-off-by: Vadim Bendebury <vbendeb@chromium.org>
Signed-off-by: Simon Glass <sjg@chromium.org>
Fix:
ahci.c: In function 'ata_scsiop_read10':
ahci.c:564:6: warning: variable 'lba' set but not used [-Wunused-but-set-variable]
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
ahci.c: In function 'ahci_port_start':
ahci.c:401: warning: format '%x' expects type 'unsigned int', but argument 2 has
type 'struct ahci_cmd_hdr *'
Signed-off-by: Marek Vasut <marek.vasut@gmail.com>
Cc: Wolfgang Denk <wd@denx.de>
Cc: Simon Glass <sjg@chromium.org>
Cc: Mike Frysinger <vapier@gentoo.org>
ahci.c: In function 'ata_scsiop_read_capacity10':
ahci.c:616: warning: dereferencing type-punned pointer will break strict-aliasing rules
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
The code assumes that the pci bus address and the virtual
address used to access a region are the same, but they might
not be. Fix this assumption.
Signed-off-by: Becky Bruce <beckyb@kernel.crashing.org>