As explained in commit 3bedbcc3aa ("arm: mvebu: a38x: serdes: Don't
overwrite read-only SAR PCIe registers") it is required to set Maximum Link
Width bits of PCIe Root Port Link Capabilities Register depending of number
of used serdes lanes. As this register is part of PCIe address space and
not serdes address space, move it into pci_mvebu.c driver.
Read number of PCIe lanes from DT property "num-lanes" which is used also
by other PCIe controller drivers in Linux kernel. If this property is
absent then it defaults to 1. This property needs to be set to 4 for every
mvebu board which use PEX_ROOT_COMPLEX_X4 or PEX_BUS_MODE_X4.
Enabling of PCIe port needs to be done afer all registers in PCIe address
space are properly configure. For this purpose use new mvebu-reset driver
(part of system-controller) and remove this code from serdes code.
Because some PCIe ports cannot be enabled individually, it is required to
first setup all PCIe ports and then enable them.
This change contains also all required "num-lanes" and "resets" DTS
properties, to make pci_mvebu.c driver work correctly.
Signed-off-by: Pali Rohár <pali@kernel.org>
The only user of board_pex_config() weak function is A385 controlcenterdc
board. It looks like that code in its board_pex_config() function needs to
be executed after PCIe link is up. Therefore put this code into
spl_board_init() function which is called after a38x serdes initialization,
and therefore it is after the serdes hws_pex_config() function finishes
(which is the state before this change).
With this change completely remove board_pex_config() function as it is not
used anymore.
Signed-off-by: Pali Rohár <pali@kernel.org>
This is a backport from Marvell U-Boot:
https://github.com/MarvellEmbeddedProcessors/u-boot-marvell
commit 381d029e7a ("fix: serdes: a38x, a39x: Improve USB3 electrical
configuration")
Improves electrical USB3 receiver jitter tolerance test:
- De-Emphasize force, in functional mode the transmitter should always
have 3.5db de-emphasize, so we are forcing it.
- After forcing De-Emphasize, choose 3.5db (After forcing, default is
6dB so need to change it to 3.5dB).
- Align90 set to 0x58 - this is the sample point in the receiver, after
the clock is recovered this sampler samples at the chosen value, usually
it is supposed to be 0x60(which is the center of the eye), but sometimes
after adding jitter and ISI the center of the eye can move slightly and
the sample point is not necessarily the exact center, and after
optimization (searching the middle of the eye manually) it was seen that
the center of the eye is actually 0x58 and not 0x60.
- FFE Res and FFE Cap set to 0xE & 0xF respectively: improves this
settings is adequate according to how the USB3 spec defines the
interconnect, thus improves USB3 jitter tolerance settings.
- Change the resolution of the DFE to 0x3 which is 6mV(highest
resolution) , this avoids the DFE to saturate and cease to work.
- HPF set to 0x3 which is 5Khz high pass filter, the function of the HPF
is to filter the low frequency patterns(below 5Khz) to make sure that
the signal is not a noise, the setting before was 0x1(205Khz), and the
change came since the USB3 CP0 pattern, that is used in the USB3 jitter
tolerance testing, is similar to PRBS15, which has 2^15=32768bits which
is 32768*200ps (200ps is one Unit interval in USB3(5Gbps)) = 6.5us,
which is in frequency terms: 152Khz. since the PRBS15 is a random
pattern and can theoretically have once in a while a pattern that will
be at frequency of 152Khz, hence the previous setting (205khz HPF) can
possibly filter this pattern which can cause to an error in the
receiver, thus this change to avoid such scenarios.
Signed-off-by: Stefan Eichenberger <eichest@gmail.com>
Signed-off-by: René Straub <rene.straub@netmodule.com>
Reviewed-by: Stefan Roese <sr@denx.de>
The electrical serdes configuration for USB3 expects an array as data
argument. For USB3 the second value is used (see data_arr_idx = USB3 =
1). However, because only one value is inside the array mv_seq_exec is
accessing an invalid element and the serdes is configured wrongly.
This wrong initialization is leading to an unreliable detection
mechanism for some USB3 devices. We were able to reproduce the issue
regularly with an LTE modem from Sierra Wireless (SM7455) where it was
not detected as USB3 device in 1/3 of all tests.
This commit fixes the issue by setting data_arr_idx to 0. This is the
same value as the original U-Boot from Marvell is using. There it is
called FIRST_CELL which is a define for 0.
See: https://github.com/MarvellEmbeddedProcessors/u-boot-marvell
commit 56f963ce4c ("fix: serdes: a38x, a39x: Fix USB3 serdes DB
initialization")
Signed-off-by: Stefan Eichenberger <eichest@gmail.com>
Signed-off-by: René Straub <rene.straub@netmodule.com>
Reviewed-by: Stefan Roese <sr@denx.de>
These are part of SOC_CONTROL_REG1 register, not PEX_CAPABILITIES_REG.
Signed-off-by: Pali Rohár <pali@kernel.org>
Reviewed-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Stefan Roese <sr@denx.de>
Remove unused PCIe functions from SerDes code. They are unused and are
duplicated either from generic PCIe code or from pci_mvebu.c.
Remove also unused PCIe macros from SerDes code. They are just obfuscated
variants of standards macros in include/pci.h or in pci_mvebu.c.
Signed-off-by: Pali Rohár <pali@kernel.org>
Reviewed-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Stefan Roese <sr@denx.de>
This code is trying to parse PCIe config space of PCIe card connected on
the other end of link and then is trying to force 5.0 GT/s speed via Target
Link Speed bits in PCIe Root Port Link Control 2 Register on the local part
of link if it sees that card supports 5.0 GT/s via Max Link Speed bits in
Link Capabilities Register.
The code is incorrect for more reasons:
- Accessing config space of an endpoint card cannot be done immediately.
If the PCIe link is not up, reading vendor/device ID registers will
return all ones.
- Parsing is incomplete, so it can cause issues even for working cards.
Moreover there is no need to force speed to 5.0 GT/s via Target Link Speed
bits on PCIe Root Port Link Control 2 Register. Hardware changes speed from
2.5 GT/s to 5.0 GT/s autonomously when it is supported.
Most importantly, this code does not change link speed at all, since
because after updating Target Link Speed bits on PCIe Root Port Link
Control 2 Register, it is required to retrain the link, and the code for
that is completely missing.
The code was probably needed for making buggy endpoint cards work. Such a
workaround, though, should be implemented via PCIe subsystem (via quirks,
for example), as buggy cards could also affect other PCIe controllers.
Note that this code is fully unrelated to a38x SerDes code and really
should not have been included in SerDes initialization. Usage of magic
constants without names and comments made this SerDes code hard to read and
understand.
Remove this PCIe application code from low level SerDes code. As this code
is configuring only 5.0 GT/s part, in the worst case, it could leave buggy
cards at the initial speed of 2.5 GT/s (if somehow before this change they
could have been "upgraded" to 5.0 GT/s speed even with missing link
retraining). Compliant cards which just need longer initialization should
work better after this change.
Signed-off-by: Pali Rohár <pali@kernel.org>
Reviewed-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Stefan Roese <sr@denx.de>
PCI device ID is part of the PCIe controller SoC / revision. For Root
Complex mode (which is the default and the only mode supported currently
by U-Boot and Linux kernel), it is PCI device ID of PCIe Root Port device.
If there is some issue with this device ID, it should be set / updated by
PCIe controller driver (pci_mvebu.c), as this register resides in address
space of the controller. It shouldn't be done in SerDes initialization
code.
In the worst case (a specific board for example) it could be done via
U-Boot's weak function board_pex_config().
But it should not be overwritten globally for all A38x devices.
Signed-off-by: Pali Rohár <pali@kernel.org>
Reviewed-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Stefan Roese <sr@denx.de>
Enabling Common Clock Configuration bit in PCIe Root Port Link Control
Register should not be done unconditionally. It is enabled by operating
system as part of ASPM. Also after enabling Common Clock Configuration it
is required to do more work, like retraining link. Some cards may be broken
due to this incomplete Common Clock Configuration and some cards are broken
and do not support ASPM at all.
Remove this incomplete code for Common Clock Configuration. It really
should not be done in SerDes code as it is not related to SerDes, but to
PCIe subsystem.
Signed-off-by: Pali Rohár <pali@kernel.org>
Reviewed-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Stefan Roese <sr@denx.de>
Device/Port Type bits of PCIe Root Port PCI Express Capabilities Register
are read-only SAR registers and are initialized according to current mode
configured by PCIe controller. Changing PCIe controller mode (from Root
Complex mode to Endpoint mode or the other way) is possible via PCI
Express Control Register (offset 0x41A00), bit 1 (ConfRoot Complex). This
has to be done in PCIe controller driver (in our case pci_mvebu.c). Note
that default mode is Root Complex.
Maximum Link Speed bits of PCIe Root Port Link Capabilities Register are
platform specific and overwriting them does not make sense. They are set by
PCIe controller according to current SerDes configuration. For A38x it is
5.0 GT/s if SerDes supports appropriate speed.
Maximum Link Width bits of PCIe Root Port Link Capabilities Register are
read-only SAR registers, but unfortunately if this is not set correctly
here, then access PCI config space of the endpoint card behind this Root
Port does not work.
Signed-off-by: Pali Rohár <pali@kernel.org>
Reviewed-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Stefan Roese <sr@denx.de>
Add comments to understand what this magic code is doing.
Signed-off-by: Pali Rohár <pali@kernel.org>
Reviewed-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Stefan Roese <sr@denx.de>
SoC Control 1 Register (offset 0x18204) is already defined by macro
SOC_CONTROL_REG1.
Use macro SOC_CONTROL_REG1 instead of macro SOC_CTRL_REG in ctrl_pex.c
code and remove the other definition.
Signed-off-by: Pali Rohár <pali@kernel.org>
Reviewed-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Stefan Roese <sr@denx.de>
Replace magic register offsets by macros to make code more readable.
Add comments about what this code is doing.
Signed-off-by: Pali Rohár <pali@kernel.org>
Reviewed-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Stefan Roese <sr@denx.de>
Config option ARMADA_39X is never set so remove all dead code hidden under
ifdef CONFIG_ARMADA_39X blocks.
Also remove useless checks for CONFIG_ARMADA_38X define as this macro is
always defined for a38x code path.
Signed-off-by: Pali Rohár <pali@kernel.org>
Reviewed-by: Stefan Roese <sr@denx.de>
Hardware testing of a board using the Armada 385 has shown that an
impedance threshold setting of 0x7 performs better in an eye-diagram
test than with Marvell's recommended value 0x6.
As other boards may still perform better with Marvell's reccomended value,
a configuration option is added with a default value of 0x6.
Signed-off-by: Joshua Scott <joshua.scott@alliedtelesis.co.nz>
Reviewed-by: Stefan Roese <sr@denx.de>
When running USB compliance tests on our Armada-385 hardware platforms
we have seen some eye mask violations. Marvell's internal documentation
says: Based on silicon test results, it is recommended to change the
impedance calibration threshold setting to 0x6 prior to calibration.
Port changes from Marvell's u-boot fork[1] to address this.
[1] - https://github.com/MarvellEmbeddedProcessors/u-boot-marvell/commit/a6221551
Signed-off-by: Chris Packham <judge.packham@gmail.com>
Reviewed-by: Stefan Roese <sr@denx.de>
Update the RTC (Read Timing Control) values for PCIe memory wrappers
following an ERRATA (ERRATA# TDB). This means the PCIe accesses will
used slower memory Read Timing, to allow more efficient energy
consumption, in order to lower the minimum VDD of the memory. Will lead
to more robust memory when voltage drop occurs (VDDSEG)
The code is based on changes from Marvell's U-Boot, specifically:
20cd270407eb608a7c8dc4af19ae2b
Signed-off-by: Chris Packham <chris.packham@alliedtelesis.co.nz>
Signed-off-by: Chris Packham <judge.packham@gmail.com>
Reviewed-by: Stefan Roese <sr@denx.de>
Reduce Auto Voltage Scaling VDD limit when core frequency is lower than
1600MHz. This reduces core voltage level from 1.25V to 1.15V, which
saves power.
The code is taken from Marvell's U-Boot 2013.01 revision 18.06.
Reviewed-by: Chris Packham <judge.packham@gmail.com>
Tested-by: Chris Packham <judge.packham@gmail.com>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: Stefan Roese <sr@denx.de>
Marvell have release a B0 revision of the Armada-385 SoC. This fixes a
hardware errata enabling RGMII to work when the Ethernet voltage is
configured to 3.3V.
Signed-off-by: Chris Packham <judge.packham@gmail.com>
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Stefan Roese <sr@denx.de>
Some QCA988x based modules presence is not detected by the SERDES lanes,
so force this detection which will trigger the LTSSM state machine to
negotiate link.
An example of such a card is WLE900VX.
Signed-off-by: Rabeeh Khoury <rabeeh@solid-run.com>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Tested-by: Chris Packham <judge.packham@gmail.com>
Tested-by: Mario Six <mario.six@gdsys.cc>
Signed-off-by: Stefan Roese <sr@denx.de>
Move sys_env_device_rev_get() from the ddr training code to
sys_env_lib.c (which currently resides with the serdes code). This
brings sys_env_device_rev_get() into line with sys_env_device_id_get()
and sys_env_model_get().
Signed-off-by: Chris Packham <judge.packham@gmail.com>
Signed-off-by: Stefan Roese <sr@denx.de>
PEX_CFG_DIRECT_ACCESS was defined in ddr3_hws_hw_training_def.h despite
only being used in the serdes code. Move this definition to ctrl_pex.h
where all the other PEX defines are. Also remove the duplicate
definition of PEX_DEVICE_AND_VENDOR_ID which is already defined in
ctrl_pex.h.
Signed-off-by: Chris Packham <judge.packham@gmail.com>
Signed-off-by: Stefan Roese <sr@denx.de>
When U-Boot started using SPDX tags we were among the early adopters and
there weren't a lot of other examples to borrow from. So we picked the
area of the file that usually had a full license text and replaced it
with an appropriate SPDX-License-Identifier: entry. Since then, the
Linux Kernel has adopted SPDX tags and they place it as the very first
line in a file (except where shebangs are used, then it's second line)
and with slightly different comment styles than us.
In part due to community overlap, in part due to better tag visibility
and in part for other minor reasons, switch over to that style.
This commit changes all instances where we have a single declared
license in the tag as both the before and after are identical in tag
contents. There's also a few places where I found we did not have a tag
and have introduced one.
Signed-off-by: Tom Rini <trini@konsulko.com>
- Fix spelling error of SERDES_VERSION
- Remove superfluous definition of this macro
- Remove unnecessary include of i2c.h
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Stefan Roese <sr@denx.de>
Allow boards to do some initialization when PCIe comes up.
Signed-off-by: Dirk Eibach <dirk.eibach@gdsys.cc>
Signed-off-by: Mario Six <mario.six@gdsys.cc>
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: Stefan Roese <sr@denx.de>
88F6820 is a specific Armada-38x chip that is used on the DB-88F6820-GP
board. Rather than having DB_88F6820_GP and TARGET_DB_88F6820_GP which
selects the former. Rename DB_88F6820_GP to 88F6820 so that other boards
using the 88F6820 can be added.
Stefan:
Change 88F6820 for clearfog as well.
Signed-off-by: Chris Packham <judge.packham@gmail.com>
Signed-off-by: Stefan Roese <sr@denx.de>
Ensure appropriate error messages are generated. Previously all errors
indicated that the serdes was already in use. Now appropriate error
messages are given.
Signed-off-by: Chris Packham <judge.packham@gmail.com>
Signed-off-by: Stefan Roese <sr@denx.de>
Instead of allocating space in the driver for the serdes
specification table, just allow the board file to set a pointer
to it. Also, allow the board to only specify the lanes that are
used instead of including unused lanes.
Signed-off-by: Kevin Smith <kevin.smith@elecsyscorp.com>
Acked-by: Stefan Roese <sr@denx.de>
Cc: Dirk Eibach <eibach@gdsys.de>
Cc: Luka Perkov <luka.perkov@sartura.hr>
Functions that do not modify the pointer passed should declare it
as const.
Signed-off-by: Kevin Smith <kevin.smith@elecsyscorp.com>
Acked-by: Stefan Roese <sr@denx.de>
Cc: Dirk Eibach <eibach@gdsys.de>
Cc: Luka Perkov <luka.perkov@sartura.hr>
A lot of extra configuration information was left over in the
Marvell serdes and DDR3 initialization code for boards that
U-boot does not support. Remove this extra config information,
and the concept of fixing up board topologies with information
loaded from an EEPROM. If this needs to be done, it should be
handled in the board file, not in core code.
Signed-off-by: Kevin Smith <kevin.smith@elecsyscorp.com>
Acked-by: Stefan Roese <sr@denx.de>
Cc: Dirk Eibach <eibach@gdsys.de>
Cc: Luka Perkov <luka.perkov@sartura.hr>
This code is ported from the Marvell bin_hdr code into mainline
SPL U-Boot. It needs to be executed very early so that the devices
connected to the serdes PHY are configured correctly.
Signed-off-by: Stefan Roese <sr@denx.de>
