Philipp Tomsich [Wed, 21 Mar 2018 11:18:58 +0000 (12:18 +0100)]
sunxi: improve throughput in the sunxi_mmc driver
Throughput tests have shown the sunxi_mmc driver to take over 10s to
read 10MB from a fast eMMC device due to excessive delays in polling
loops.
This commit restructures the main polling loops to use get_timer(...)
to determine whether a (millisecond) timeout has expired. We choose
not to use the wait_bit function, as we don't need interruptability
with ctrl-c and have at least one case where two bits (one for an
error condition and another one for completion) need to be read and
using wait_bit would have not added to the clarity.
The observed speedup in testing on a A31 is greater than 10x (e.g. a
10MB write decreases from 9.302s to 0.884s).
Signed-off-by: Philipp Tomsich <philipp.tomsich@theobroma-systems.com> Signed-off-by: Maxime Ripard <maxime.ripard@bootlin.com> Tested-by: Mylène Josserand <mylene.josserand@bootlin.com> Acked-by: Jagan Teki <jagan@openedev.com> Reviewed-by: Tom Rini <trini@konsulko.com>
Stefan Mavrodiev [Tue, 27 Mar 2018 13:57:23 +0000 (16:57 +0300)]
sunxi: mmc: Fix phase delays
U-boot driver for sunxi-mmc uses PLL6, unlike linux kernel where
PLL5 is used, with clock rates respectively 600MHz and 768MHz.
Thus there are different phase degree steps - 24 for the kernel and
30 for u-boot.
In the kernel driver the phase is set 90 deg for output and 120 for
sample. Dividing by 30 will result values 3 and 4. Those are the
values set in the u-boot driver.
However, the condition defining delays is wrong. MMC core driver
requests clock of 52MHz, sunxi-driver sets clock of 50MHz, but
phase is set 30 deg for output and 120 deg for sample.
Apparently this works for most cards.
On A20-SOM204-EVB-eMMC there is eMMC card (KLMAG2GEND) which complains
about it. Maybe there is other boards with similar problem?
So the fix is to match delays for both u-boot and kernel.
Signed-off-by: Stefan Mavrodiev <stefan@olimex.com> Acked-by: Maxime Ripard <maxime.ripard@bootlin.com> Reviewed-by: Jagan Teki <jagan@openedev.com>
Marek Vasut [Sat, 21 Apr 2018 14:19:56 +0000 (16:19 +0200)]
ARM: rmobile: Update E2 Silk
The E2 Silk port was broken since some time. This patch updates
the E2 Silk port to use modern frameworks, DM, DT probing, SPL
for the preloading and puts it on par with the M2 Porter board.
Signed-off-by: Marek Vasut <marek.vasut+renesas@gmail.com> Cc: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
---
NOTE: The port is missing support for I2C1 for DA9063 reset, since the
I2C driver needs to be converted to DM and DT probing. That's not
an issue for this patch though, since the reset was broken on Silk
since forever.
Marek Vasut [Sat, 21 Apr 2018 15:27:11 +0000 (17:27 +0200)]
mmc: sh_mmcif: Add DM and DT probing support
Add MMC DM and DT probing support into the SH MMCIF driver.
This patch abstracts out the common bits of the send command
and set ios functions, so they can be used both by DM and non
DM setups and adds the DM probe support.
Neil Armstrong [Thu, 12 Apr 2018 08:03:19 +0000 (10:03 +0200)]
reset: fix bulk API when DM_RESET is disabled
In the commit "reset: Add get/assert/deassert/release for bulk of reset signals"
the disabled reset_release_bulk() and reset_get_bulk() used the wrong
struct clk_bulk instead of struct reset_ctl_bulk.
Fixes: 0c28233903b5 ("reset: Add get/assert/deassert/release for bulk of reset signals") Reported-by: Jean-Jacques Hiblot <jjhiblot@ti.com> Signed-off-by: Neil Armstrong <narmstrong@baylibre.com> Reviewed-by: Simon Glass <sjg@chromium.org>
commit b4d956f6bc0f ("bootm: Align cache flush end address correctly")
aligns the end address of the cache flush operation to a cache-line size to
ensure lower-layers in the code accept the range provided and flush.
A similar action should be taken for the begin address of a cache flush
operation. The load address may not be aligned to a cache-line boundary, so
ensure the passed address is aligned.
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Reported-by: Breno Matheus Lima <brenomatheus@gmail.com> Suggested-by: Tom Rini <trini@konsulko.com> Cc: Simon Glass <sjg@chromium.org> Tested-by: Breno Lima <breno.lima@nxp.com> Reviewed-by: Simon Glass <sjg@chromium.org>
Marek Vasut [Tue, 17 Apr 2018 12:13:11 +0000 (14:13 +0200)]
ARM: rmobile: Update M2 Koelsch
The M2 Koelsch port was broken since some time. This patch updates
the M2 Koelsch port to use modern frameworks, DM, DT probing, SPL
for the preloading and puts it on par with the M2 Porter board.
Signed-off-by: Marek Vasut <marek.vasut+renesas@gmail.com> Cc: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
configs: socfpga: disable EFI and ISO partition types
None of the SoCFPGA platforms will support EFI/ISO partition types that
is needed for DISTRO_DEFAULTS. SoCFPGA bootroom will only support 0xa2
partition type.
This is needed to help limit the size of the SPL to within the 64k limit
that is required for SoCFPGA.
Add code to look for a reset manager property. Specifically, look for the
reset-names of 'i2c'. A reset property is an optional feature, so only print
out a warning and do not fail if a reset property is not present.
If a reset property is discovered, then use it to deassert, thus bringing the
IP out of reset.
Matt Pelland [Mon, 16 Apr 2018 14:08:18 +0000 (10:08 -0400)]
mmc: mv_sdhci: zero out sdhci_host structure
The mv_sdhci driver was not zeroing the sdhci_host structure it
allocates causing random access violations in parts of the mmc core
where the "ops" member pointers are checked and called if not NULL.
Signed-off-by: Matt Pelland <mpelland@starry.com> Reviewed-by: Stefan Roese <sr@denx.de> Signed-off-by: Stefan Roese <sr@denx.de>
env: mmc/fat/ext4: undefined reference to `mmc_initialize'
For CONFIG_ENV_FAT_INTERFACE != 'mmc' a link error
env/fat.c:93: undefined reference to `mmc_initialize'
occurs if CONFIG_MMC_SUPPORT is not enabled.
Fixes: 26862b4a40c3 ("env: mmc/fat/ext4: make sure that the MMC sub-system
is initialized before using it") Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Kconfig: Enlarge default SYS_MALLOC_F_LEN for AM33XX
Since commit 8e14ba7bd524 ("gpio: omap_gpio: Add DM_FLAG_PRE_RELOC
flag") omap GPIO gets bound before relocation. Unfortunately due to
this, on at least the beaglebone black, the pre-relocation memory pool
gets exhausted before probing the serial port. This then causes u-boot
to panic as CONFIG_REQUIRE_SERIAL_CONSOLE is set...
Resolve this by resizing the default size of the pre-relocation malloc
pool for AM335X platforms.
Alexander Graf [Thu, 12 Apr 2018 07:58:44 +0000 (09:58 +0200)]
part: Disable CONFIG_SPL_ISO_PARTITION by default
We enabled CONFIG_ISO_PARTITION by default for distro boot, so that U-Boot
could load distro images that usually get shipped as iso images. These images
usually come with a board agnostic boot environment.
However, there is very little point in having ISO support enabled (for anyone
really) in SPL, as the whole idea of SPL is to load U-Boot proper which again
is board specific. So the fact that we enable ISO support in U-Boot proper does
not mean at all that we want ISO support in U-Boot SPL.
Hence, let's remove the Kconfig dependency. Along the way, let's also clean up
all those default configs that disabled SPL ISO support.
Simon Glass [Thu, 12 Apr 2018 20:04:40 +0000 (14:04 -0600)]
x86: Rename coreboot-x86 to coreboot
We only use coreboot as a target on x86 platforms, since on ARM platforms
U-Boot always runs as the primary boot loader. Rename the coreboot-x86
platform to reflect this.
Signed-off-by: Simon Glass <sjg@chromium.org> Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Bin Meng [Thu, 12 Apr 2018 05:02:22 +0000 (22:02 -0700)]
bootvx: x86: Assign bootaddr based on kernel memory base
On VxWorks x86 its bootline address is at a pre-defined offset @
0x1200. If 'bootaddr' is not passed via environment variable, we
assign its value based on the kernel memory base address.
Bin Meng [Thu, 12 Apr 2018 05:02:19 +0000 (22:02 -0700)]
bootvx: x86: Make VxWorks EFI console driver happy
When booting from EFI BIOS, VxWorks bootloader stores the EFI GOP
framebuffer info at a pre-defined offset @ 0x6100. When VxWorks
kernel boots up, its EFI console driver tries to find such a block
and if the signature matches, the framebuffer information will be
used to initialize the driver.
However it is not necessary to prepare an EFI environment for
VxWorks's EFI console driver to function (eg: EFI loader in
U-Boot). If U-Boot has already initialized the graphics card and
set it to a VESA mode that is compatible with EFI GOP, we can
simply prepare such a block for VxWorks.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com> Reviewed-by: Simon Glass <sjg@chromium.org>
Bin Meng [Thu, 12 Apr 2018 05:02:18 +0000 (22:02 -0700)]
pci: video: Only print out when everything is OK
If video initialization fails, the "Video:" output message will be
mixed with the next console log. Change to print out such message
only when everything is OK, which improves the boot log readability.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com> Reviewed-by: Simon Glass <sjg@chromium.org>
Bin Meng [Thu, 12 Apr 2018 05:02:15 +0000 (22:02 -0700)]
bios: vesa: Guard setting vesa mode with CONFIG_FRAMEBUFFER_SET_VESA_MODE
If CONFIG_FRAMEBUFFER_SET_VESA_MODE is not set, don't switch
graphics card to VESA mode. This applies to both native mode
and emulator mode of running the VGA BIOS.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com> Reviewed-by: Simon Glass <sjg@chromium.org>
Bin Meng [Thu, 12 Apr 2018 05:02:12 +0000 (22:02 -0700)]
elf: Clean up the ELF header file
Fix various style violations in elf.h
- use correct comment format if the comment fits in just one line
- remove the ending period for the one-line comment
- use tab for the indention instead of space
- put the opening brace at the same line of a typedef/union
- remove <name> in a 'typedef struct' for consistency
Signed-off-by: Bin Meng <bmeng.cn@gmail.com> Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Bin Meng [Thu, 12 Apr 2018 05:02:08 +0000 (22:02 -0700)]
bootvx: x86: Explicitly clear the bootloader image size
VxWorks bootloader stores its size at a pre-defined offset @ 0x5004.
Later when VxWorks kernel boots up and system memory information is
retrieved from the E820 table, the bootloader size will be subtracted
from the total system memory size to calculate the size of available
memory for the OS.
Explicitly clear the bootloader image size otherwise if memory
at this offset happens to contain some garbage data, the final
available memory size for the kernel is insane.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com> Reviewed-by: Simon Glass <sjg@chromium.org>
Bin Meng [Thu, 12 Apr 2018 05:02:07 +0000 (22:02 -0700)]
bootvx: x86: Prepare e820 related stuff from the given kernel memory base address
At present two environment variables 'e820data'/'e820info' are required
to boot a VxWorks x86 kernel, but this is superfluous. The offset of
these two tables are actually at a fixed offset from the kernel memory
base address and we can provide the kernel memory base address to U-Boot
via only one variable 'vx_phys_mem_base'.
Note as it name indicates, the physical address should be provided.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com> Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Ivan Gorinov [Fri, 6 Apr 2018 21:43:04 +0000 (14:43 -0700)]
x86: Add 64-bit memory-mapped I/O functions
Add readq() and writeq() definitions for x86.
Please note: in 32-bit code readq/writeq will generate two 32-bit
memory access instructions instead of one atomic 64-bit operation.
Signed-off-by: Ivan Gorinov <ivan.gorinov@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Each imx image is created by a separate sub-make and during this process
the mkimage config file is run though cpp.
The cpp output is to the same file no matter what imx image is being
created.
This means if two imx images are generated in parallel they will attempt
to independently produce the same pre-processed mkimage config file at
the same time.
Avoid the problem by making the pre-processed config file name unique
based on the imx image it will be used in. This way each image will
create a unique config file and they won't clobber each other when run
in parallel.
This should fixed the build bug referenced in b5b0e4e3 ("imximage:
Remove failure when no IVT offset is found").
Cc: Breno Lima <breno.lima@nxp.com> Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com> Cc: Fabio Estevam <fabio.estevam@nxp.com> Signed-off-by: Trent Piepho <tpiepho@impinj.com> Tested-by: Fabio Estevam <fabio.estevam@nxp.com>
Working with HAB on the i.MX7 we've encountered a case where a board that
successfully authenticates u-boot when booting Linux via OPTEE subsequently
fails to properly bring up the RTC.
The RTC registers live in the low-power block of the Secure Non-Volatile
Storage (SNVS) block.
The root cause of the error has been traced to the HAB handing off the
SNVS-RTC in a state where HPCOMR::NPSWA_EN = 0 in other words where the
Non-Privileged Software Access Enable bit is zero. In ordinary
circumstances this is OK since we typically do not run in TZ mode, however
when we boot via HAB and enablng TrustZone, it is required to set
HPCOMR::NPSWA_EN = 1 in order for the upstream Linux driver to have
sufficient permissions to manipulate the SNVS-LP block.
On our reference board it is the difference between Linux doing this:
root@imx7s-warp-mbl:~# dmesg | grep rtc
snvs_rtc_enable read 0x00000000 from SNVS_LPLR @ 0x00000034
snvs_rtc_enable read 0x00000021 from SNVS_LPCR @ 0x00000038
snvs_rtc_enable read 0x00000000 from SNVS_HPLR @ 0x00000000
snvs_rtc_enable read 0x80002100 from SNVS_HPCOMR @ 0x00000004
snvs_rtc 30370000.snvs:snvs-rtc-lp: rtc core: registered 30370000.snvs:snvs-rtc-lp as rtc0
snvs_rtc 30370000.snvs:snvs-rtc-lp: setting system clock to2018-04-01 00:51:04 UTC (1522543864)
and doing this:
root@imx7s-warp-mbl:~# dmesg | grep rtc
snvs_rtc_enable read 0x00000000 from SNVS_LPLR @ 0x00000034
snvs_rtc_enable read 0x00000020 from SNVS_LPCR @ 0x00000038
snvs_rtc_enable read 0x00000001 from SNVS_HPLR @ 0x00000000
snvs_rtc_enable read 0x00002020 from SNVS_HPCOMR @ 0x00000004
snvs_rtc 30370000.snvs:snvs-rtc-lp: failed to enable rtc -110
snvs_rtc: probe of 30370000.snvs:snvs-rtc-lp failed with error -110
hctosys: unable to open rtc device (rtc0)
Note bit 1 of LPCR is not set in the second case and is set in the first
case and that bit 31 of HPCOMR is set in the second case but not in the
first.
Setting NPSWA_EN in HPCOMR allows us to boot through enabling TrustZone
and continue onto the kernel. The kernel then has the necessary permissions
to set LPCR::SRTC_ENV (RTC enable in the LP command register) whereas in
contrast - in the failing case the non-privileged kernel cannot do so.
This patch adds a simple init_snvs() call which sets the permission-bit
called from soc.c for the i.MX7. It may be possible, safe and desirable to
perform this on other i.MX processors but for now this is only tested on
i.MX7 as working.
Bryan O'Donoghue [Mon, 26 Mar 2018 14:36:46 +0000 (15:36 +0100)]
imx: hab: Provide hab_auth_img_or_fail command
This patch adds hab_auth_img_or_fail() a command line function that
encapsulates a common usage of authenticate and failover, namely if
authenticate image fails, then drop to BootROM USB recovery mode.
For secure-boot systems, this type of locked down behavior is important to
ensure no unsigned images can be run.
It's possible to script this logic but, when done over and over again the
environment starts get very complex and repetitive, reducing that script
repetition down to a command line function makes sense.
Bryan O'Donoghue [Mon, 26 Mar 2018 14:36:45 +0000 (15:36 +0100)]
imximage: Encase majority of header in __ASSEMBLY__ declaration
Subsequent patches will want to include imageimage.h but in doing so
include it on an assembly compile path causing a range of compile errors.
Fix the errors pre-emptively by encasing the majority of the declarations
in imximage.h inside an ifdef __ASSEMBLY__ block.
Bryan O'Donoghue [Mon, 26 Mar 2018 14:27:34 +0000 (15:27 +0100)]
warp7: Set u-boot serial# based on OTP value
u-boot has a standard "serial#" environment variable that is suitable
for storing the iSerial number we will supply via the USB device
descriptor. serial# is automatically picked up by the disk subsystem in
u-boot - thus providing a handy unique identifier in /dev/disk/by-id as
detailed below.
Storing the hardware serial identifier in serial# means we can change the
serial# if we want before USB enumeration - thus making iSerial automatic
via OTP but overridable if necessary.
This patch reads the defined OTP fuse and sets environment variable
"serial#" to the value read.
With this patch in place the USB mass storage device will appear in
/dev/disk/by-id with a unique name based on the OTP value. For example
Bryan O'Donoghue [Mon, 26 Mar 2018 14:27:33 +0000 (15:27 +0100)]
imx: mx7: Add comment to describe OTP TESTER registers
The tester registers provide a unique chip-level identifier which
get_board_serial() returns in a "struct tag_serialnr".
This patch documents the properties of the registers; in summary.
31:0 OCOTP_TESTER0 (most significant)
- FSL-wide unique, encoded LOT ID STD II/SJC CHALLENGE/ Unique ID
OCOTP_TESTER1 (least significant)
31:24
- The X-coordinate of the die location on the wafer/SJC CHALLENGE/ Unique
ID
23:16
- The Y-coordinate of the die location on the wafer/SJC CHALLENGE/ Unique
ID
15:11
- The wafer number of the wafer on which the device was fabricated/SJC
CHALLENGE/ Unique ID
10:0
- FSL-wide unique, encoded LOT ID STD II/SJC CHALLENGE/ Unique ID
The 64 bits of data generate a unique serial number per-chip.
Marek Vasut [Fri, 30 Mar 2018 01:04:43 +0000 (03:04 +0200)]
ARM: mx6: ddr: Add write leveling correction code
When the DDR calibration is enabled, a situation may happen that it
will fail on a few select boards out of a whole production lot. In
particular, after the first write leveling stage, the MPWLDECTRLx
registers will contain a value 0x1nn , for nn usually being 0x7f or
slightly lower.
What this means is that the HW write leveling detected that the DQS
rising edge on one or more bundles arrives slightly _after_ CLK and
therefore when the DDR DRAM samples CLK on the DQS rising edge, the
CLK signal is already high (cfr. AN4467 rev2 Figure 7 on page 18).
The HW write leveling then ends up adding almost an entire cycle (thus
the 0x17f) to the DQS delay, which indeed aligns it, but also triggers
subsequent calibration failure in DQS gating due to this massive offset.
There are two observations here:
- If the MPWLDECTRLx value is corrected from 0x17f to 0x0 , then the
DQS gating passes, the entire calibration passes as well and the
DRAM is perfectly stable even under massive load.
- When using the NXP DRAM calibrator for iMX6/7, the value 0x17f or so
in MPWLDECTRx register is not there, but it is replaced by 0x0 as one
would expect.
Someone from NXP finally explains why, quoting [1]:
"
Having said all that, the DDR Stress Test does something that we
do not advertise to the users. The Stress Test iself looks at the
values of the MPWLDECTRL0/1 fields before reporting results, and
if it sees any filed with a value greater than 200/256 delay
(reported as half-cycle = 0x1 and ABS_OFFSET > 0x48), the DDR
Stress test will reset the Write Leveling delay for this lane
to 0x000 and not report it in the log.
The reason that the DDR Stress test does this is because a delay
of more than 78% a clock cycle means that the DQS edge is arriving
within the JEDEC tolerence of 25% of the clock edge. In most cases,
DQS is arriving < 5% tCK of the SDCLK edge in the early case, and
it does not make sense to delay the DQS strobe almost a full clock
cycle and add extra latency to each Write burst just to make the
two edges align exactly. In this case, we are guilty of making a
decision for the customer without telling them we are doing it so
that we don't have to provide the above explanation to every customer.
They don't need to know it.
"
This patch adds the correction described above, that is if the MPWLDECTRx
value is over 0x148, the value is corrected back to 0x0.
[1] https://community.nxp.com/thread/456246
Signed-off-by: Marek Vasut <marex@denx.de> Cc: Stefano Babic <sbabic@denx.de> Reviewed-by: Fabio Estevam <fabio.estevam@nxp.com> Reviewed-by: Eric Nelson <eric@nelint.com> Reviewed-by: Stefano Babic <sbabic@denx.de>
Rasmus Villemoes [Fri, 23 Mar 2018 11:08:03 +0000 (12:08 +0100)]
tools/imximage: use 0x prefix in HAB Blocks line
The u-boot-ivt.img.log file contains 0x prefixes in the HAB Blocks line,
while the SPL.log does not. For consistency, and to make it easier to
extract and put into a .csf file for use with NXP's code signing tool,
add 0x prefixes here.
Rasmus Villemoes [Fri, 23 Mar 2018 11:08:02 +0000 (12:08 +0100)]
Makefile: always preserve output for images that can contain HAB Blocks
The current makefile logic disables creation of the
SPL.log/u-boot-ivt.img.log etc. files when V=1 is given on the command
line, the rationale presumably being that the user wants and gets the
information on the console.
However, from general principles, I don't think a higher V= level
should affect which build artifacts get generated (and certainly
shouldn't produce fewer). Concretely, it's also a problem that when
doing a V=1 build in a terminal, the relevant HAB blocks lines easily
drown in all the other V=1 output.
Moreover, build systems such as Yocto by default pass V=1, so in that
case the information gets hidden away in the do_compile log file, making
it nigh impossible to create a recipe for creating signed U-boot images
- I don't want to disable V=1, because having verbose output in the log
file is valuable when things go wrong, but OTOH trying to go digging in
the do_compile log file (and getting exactly the right lines) is not
pleasant to even think about.
So change the logic so that for V=0, the mkimage output is redirected
to MKIMAGEOUTPUT (which is also the current behaviour), while for any
other value of V, we _additionally_ write the information to make's
stdout, whatever that might be.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk> Tested-by: Breno Lima <breno.lima@nxp.com>
projects / u-boot.git / log