Transmission of short packets does not work good for XFI (GDM2) and
HSGMII (GDM3) interfaces. The issue can be solved with:
- padding of short packets to 60 bytes
- setting of PAD_EN bit in the corresponding REG_GDM_FWD_CFG(n)
register.
The issue should present for the lan switch (GDM1) as well, but it does
does not appear due to unknown reason.
This patch set PAD_EN bit for the used GDM.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
ARCH_DMA_MINALIGN is 64 for ARMv7a/ARMv8a architectures, but RX/TX
descriptors are 32 bytes long. So they may not be aligned on an
ARCH_DMA_MINALIGN boundary. In case of RX path, this may cause the
following problem
1) Assume that a packet has arrived and the EVEN rx descriptor has been
updated with the incoming data. The driver will invalidate and check
the corresponding rx descriptor.
2) Now suppose the next descriptor (ODD) has not yet completed.
Please note that all even descriptors starts on 64-byte boundary,
and the odd ones are NOT aligned on 64-byte boundary.
Inspecting even descriptor, we will read the entire CPU cache line
(64 bytes). So we read and sore in CPU cache also the next (odd)
descriptor.
3) Now suppose the next packet (for the odd rx descriptor) arrived
while the first packet was being processed. So we have new data
in memory but old data in cache.
4) After packet processing (in arht_eth_free_pkt() function) we will
cleanup the descriptor and put it back to rx queue.
This will call flush_dcache_range() function for the even descriptor,
so the odd one will be flushed as well (it is in the same cache line).
So the old data will be written to the next rx descriptor.
5) We get a freeze. The next descriptor is empty (so the driver is
waiting for packets), but the hardware will continue to receive
packets on other available descriptors. This will continue until
the last available rx descriptor is full. Then the hardware will
also freeze.
The problem will be solved if the previous descriptor will be put back
to the queue instead of the current one.
If the current descriptor is even (starts on a 64-byte boundary),
then putting the previous descriptor to the rx queue will affect
the previous cache line. To be 100% ok, we must make sure that the
previous and the one before the previous descriptor cannot be used
for receiving at this moment.
If the current descriptor is odd, then the previous descriptor is on
the same cache line. Both (current and previous) descriptors are not
currently in use, so issue will not arrise.
WARNING: The following restrictions on PKTBUFSRX must be held:
* PKTBUFSRX is even,
* PKTBUFSRX >= 4.
The bug appears on 32-bit airoha platform, but should be present on
64-bit as well.
The code was tested both on 32-bit and 64-bit airoha boards.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
The dma_map_single() function calls one of the functions
* invalidate_dcache_range(),
* flush_dcache_range().
Both of them expect that 'vaddr' is aligned to the ARCH_DMA_MINALIGN
boundary. Unfortunately, RX/TX descriptors are 32-byte long. Thus they
might not be aligned to the ARCH_DMA_MINALIGN boundary. Data flushing
(or invalidating) might do nothing in this case.
The same applies to dma_unmap_single() function.
In the TX path case the issue might prevent package transmission (filled
TX descriptor was not flushed).
To fix an issue a special wrappers for
* dma_map_single(),
* dma_unmap_single()
functions were created. The patch fix flushing/invalidatiog for the
RX path as well.
The bug appears on 32-bit airoha platform, but should be present on
64-bit as well.
The code was tested both on 32-bit and 64-bit airoha boards.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
Compatible device list must have a terminator. If terminator is missed
the u-boot driver subsystem will access random data placed after the
list in the memory.
The issue can be observed with the "dm compat" command.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
Romain Gantois <romain.gantois@bootlin.com> says:
This is version one of my series which enables Ethernet support on the BBGE
board. This requires three main changes:
- Describing the MAC<->PHY link and DP83867 PHY accurately in the device
tree
- Enabling the RGMII1 pinmux configuration
- Enabling the DP83867 driver
These changes are all applied in patch 2.
Patch 1 enables excluding the DP83867 driver from SPL. This is done to
avoid size issues when adding the DP83867 driver to the am335x-evm
defconfig.
Link: https://lore.kernel.org/r/20250626-bbge-ethernet-v1-0-5b544fb1898f@bootlin.com
Currently, the Ethernet interface of the BBGE board isn't properly
supported.
To support this interface, describe the MAC<->PHY RGMII link and the
DP83867 PHY in the BBGE device tree. Enable the DP83867 PHY driver in
U-Boot Proper and mux the relevant RGMII lines.
Signed-off-by: Romain Gantois <romain.gantois@bootlin.com>
Tested-by: Judith Mendez <jm@ti.com>
The DP83867 PHY driver is used by the BeagleBoneGreen Eco board, but adding
it to the am335x-evm defconfig causes SPL to overflow its size limits.
Add a separate option to enable this driver in SPL, so that it can be
enabled in U-Boot without adding unnecessary volume to SPL.
Signed-off-by: Romain Gantois <romain.gantois@bootlin.com>
Tested-by: Judith Mendez <jm@ti.com>
João Paulo Gonçalves <jpaulo.silvagoncalves@gmail.com> says:
In some use cases, board-specific device tree changes must not be overwritten
by system fixups. Although U-Boot provides ft_board_setup_ex() for this
purpose, it is currently only used on TI Keystone. This series makes
ft_board_setup_ex() a generic option, allowing its use by other architectures
and boards.
Additionally, considering that Toradex Verdin-AM62P hardware lifetime
guarantees are based on a 105°C junction temperature (while TI AM62Px supports
up to 125°C), this series implements necessary changes within TI K3 AM62P and
Toradex board code. These changes include exporting common fixup device Tree
functions used in TI K3 for board-code access and also fixup for AM62P thermal
zones to correctly reflect the number of CPU nodes according to the SoC part
number.
Link: https://lore.kernel.org/r/20250623-am62p-fdt-fixup-trip-points-v1-0-12355eb6a72f@toradex.com
While TI AM62P supports up to 125°C junction temperature, Tj, for
industrial and automotive parts, Toradex Verdin-AM62P hardware lifetime
guarantees consider a 105°C Tj. Fixup the device tree temperature
critical trip points to match the hardware specifications. The
implementation ensures the architecture code will not overwrite the
board specific changes by enabling CONFIG_OF_BOARD_SETUP_EXTENDED for
the Verdin-AM62P.
Signed-off-by: João Paulo Gonçalves <joao.goncalves@toradex.com>
Acked-by: Francesco Dolcini <francesco.dolcini@toradex.com>
TI AM62Px devices support CPU throttling based on thermal alerts.
However, the device tree assumes a 4-core configuration. Since the AM62P
also supports 2-core configurations, add a fixup to dynamically adjust
the cooling-device nodes within thermal zones based on the actual number
of CPU cores available.
Signed-off-by: João Paulo Gonçalves <joao.goncalves@toradex.com>
Avoid code duplication by making the common TI K3 device tree fixup
functions to be reusable by board-specific code.
Signed-off-by: João Paulo Gonçalves <joao.goncalves@toradex.com>
In some use cases, board-specific device tree changes must not be
overwritten by system fixups. Although U-Boot provides
ft_board_setup_ex() for this purpose, it is currently only used on TI
Keystone. Make ft_board_setup_ex() to be a generic option, allowing its
use by other architectures/boards. To maintain backward compatibility,
enable it by default on TI Keystone.
Signed-off-by: João Paulo Gonçalves <joao.goncalves@toradex.com>
scene_obj_find can return NULL but this is not checked for before
the return is dereferenced. Add a NULL check.
This issue was found by Smatch.
Also add a NULL check for str.
Signed-off-by: Andrew Goodbody <andrew.goodbody@linaro.org>
Reviewed-by: Tom Rini <trini@konsulko.com>
In the case where parse_sliteral returns an error then label_name
will not have been assigned to. In order to prevent evaluating
label_name in this case add a check for the return value of parse_sliteral.
Signed-off-by: Andrew Goodbody <andrew.goodbody@linaro.org>
It was discovered that env_do_env_set() currently suffer from a long
time of a possible out-of-bound access for the argv array handling.
The BUG is present in the function env_do_env_set() line:
name = argv[1];
where the function at this point assume the argv at index 1 is always
present and can't be NULL. Aside from the fact that it's always
better to validate argv entry with the argc variable, situation where
the argv[1] is NULL is actually possible and not an error condition.
A example of where an out-of-bound access is triggered is with the
command "askenv - Press ENTER to ...".
This is a common pattern for bootmenu entry to ask the user input after
a bootmenu command succeeded.
In the context of such command, the while loop before "name = argv[1];"
parse the "-" char as an option arg and increment the argv pointer by
one (to make the rest of the logic code ignore the option argv) and
decrement argc value.
The while loop logic is correct but at the "name = argv[1];" line, the
argv have only one element left (the "-" char) and accessing argv[1]
(aka the secong element from argv pointer) cause an out-of-bound access
(making the bootloader eventually crash with strchr searching in invalid
data)
To better handle this and prevent the out-of-bound access, actually
check the argv entry left (with the use of the argc variable) and exit
early before doing any kind of array access.
Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
Sometimes when building a platform such as imx28_btt3 we can get an
error such as:
gzip: fit-dtb.blob: warning: file timestamp out of range for gzip format
make[1]: *** [Makefile:1219: fit-dtb.blob.gz] Error 2
make[1]: *** Deleting file 'fit-dtb.blob.gz'
make: *** [Makefile:186: sub-make] Error 2
This is typically resolved by telling gzip to ignore the timestamp.
Signed-off-by: Tom Rini <trini@konsulko.com>
Commit 93b55636b0 ("tools: kwboot: Allow any baudrate on Linux") added a
tcgetattr() function to be used, instead of the libc's termios functions.
This was done to allow using the raw TCGETS2/TCSETS2 ioctls that allow to
support arbitrary baud rates.
This breaks the build for PowerPC, because that architecture does not have
a struct termios2 defined because the termios and ktermios are the same.
On PowerPC, the termios ioctl() emulates the TCGETS2/TCSETS*2 ioctls with
tcgetattr/tcsetattr using just the struct termios (that's as mentioned the
same than what is defined as struct termios2 in other architectures).
So there is no need to use the TCGETS2/TCSETS2 ioctls on that architecture
and just TCGETS/TCSETS ioctls with termios as defined by PowerPC is enough.
Fixes: 93b55636b0 ("tools: kwboot: Allow any baudrate on Linux")
Signed-off-by: Javier Martinez Canillas <javierm@redhat.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
The commit under fixes tag includes "pcie-cadence.h" using angle
brackets. Since the header file is not in standard include path change
it to double quotes to ensure proper inclusion and avoid build issues,
especially on older compilers.
Fixes: a4a0edc604 ("pci_endpoint: Add TI K3 Cadence PCIe Endpoint Controller driver")
Signed-off-by: Hrushikesh Salunke <h-salunke@ti.com>
Reviewed-by: Bryan Brattlof <bb@ti.com>
With the latest TIFS firmware, an additional virtual interrupt and
event is reserved for TIFS usage on am62x and am62ax devices.
Update the rm-cfg to reflect this new reservation.
Based on commit 87720385ab ("board: ti: rm-cfg: Update rm-cfg to
reflect new resource reservation").
Signed-off-by: Wadim Egorov <w.egorov@phytec.de>
EXT_DTB overrides the of_list_dirs that is passed to binman during
builds. This causes binman to not find the normal DTB paths and instead
it just is able to use the path for EXT_DTB only.
Add the default of_list_dirs that is used for normal binman builds to
EXT_DTB usecase as well.
Signed-off-by: Manorit Chawdhry <m-chawdhry@ti.com>
Reviewed-by: Aniket Limaye <a-limaye@ti.com>
The object= part matches against the label that the pkcs11 token uses
for that object, but in some cases, specifically with a Yubikey using
ykcs11, where the keys have been imported, the labels differ between the
private and public keys [1], making the object= matching useless. These
keys will have the same id however, so matching against that works for
both the private and public part.
[1]: https://github.com/Yubico/yubico-piv-tool/blob/master/doc/YKCS11/Functions_and_values.adoc#key-alias-per-slot-and-object-type
Signed-off-by: Tobias Olausson <tobias@eub.se>
The variable base is unsigned so >= 0 is always true. Fix this test
so that it is actually useful. The fix prevents the code from causing
a segfault in the case where Ctrl-w is pressed on a line consisting
only of spaces.
Fixes: dcc18ce0db ("cli: Implement delete-word in cread_line()")
Signed-off-by: Andrew Goodbody <andrew.goodbody@linaro.org>
This file does not need <asm/gpio.h> so remove it. This file does
however need <linux/types.h> so add that in. This also shows that
arch/arm/mach-rockchip/rk3399/rk3399.c and board/lg/star/star.c were
indirectly getting <asm/gpio.h> from here, so add <asm/gpio.h> to them.
Signed-off-by: Tom Rini <trini@konsulko.com>
Our last sync with the kernel was 5.1. Even that was a partial one
as some patches from 4.x kernels were already missing making the
transition to a modern kbuild infeasible.
We are so out of sync now, that tracking the patches and backporting
them one by one makes little sense and it's going to take ages.
This is an attempt to sync up Makefile[.lib/.kbuild].
Unfortunately due to sheer amount of patches this is not easy to review,
but that's what we decided during a community call.
One of the biggest changes is get rid of partial linking entirely and
build .a archives isntead of .o.
We diaviate from the kernel on that. Instead of calling a custom script
to create the archive symbol table, we call ar with rcTP (isntead of
rcSTP) since we want a resulting archive that's sauble with the linker.
The only affected platforms are PPC ones. Unfortunately I don't have any
of them around to test, but the objdump of the resulting files --
arch/powerpc/lib/built-in.[oa] looks identical.
Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Since we are updating our kbuild system to 5.1, add the relevant
clang changes from upstream
Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Pull request net-next-20250708
lwip:
- Call sys_check_timeouts() and schedule() on RX to fix an issue on
boards with a watchdog and simplify the code
- Enable IP_FRAG and IP_REASSEMBLY
- Add support for setting the TFTP blocksize at runtime
- Fix DNS initialization in wget
- Add the sntp command
- Move code from net/lwip/${cmd}.c to cmd/lwip/${cmd}.c
In DDR mode, even bytes are read using DMA, while the remaining odd
bytes are read using STIG mode. However, the data is not correctly
transferred into the flash read data lower register because the
supplementary byte of the STIG opcode is not being written to the
opcode extension register, resulting in incorrect data being read.
To resolve this issue, when using STIG transactions, the corresponding
supplementary byte of any STIG opcode must be defined in the Opcode
Extension Register (Lower). Issue has been observed on the Macronix
MX66UM2G45G flashes.
Signed-off-by: Prasad Kummari <prasad.kummari@amd.com>
Signed-off-by: Venkatesh Yadav Abbarapu <venkatesh.abbarapu@amd.com>
Link: https://lore.kernel.org/r/20250702053953.640046-1-venkatesh.abbarapu@amd.com
Signed-off-by: Michal Simek <michal.simek@amd.com>
tshsl_ns is the clock delay for chip select deassert. This is the delay in
master reference clocks for the length that the master mode chip select
outputs are de-asserted between transactions.
The minimum delay is always SCLK period to ensure the chip select is never
re-asserted within one SCLK period.
That is why tshsl_ns delay should be at least one sclk_ns value. If it is
less than sclk_ns, set it equal to sclk_ns.
Signed-off-by: Ashok Reddy Soma <ashok.reddy.soma@amd.com>
Signed-off-by: Venkatesh Yadav Abbarapu <venkatesh.abbarapu@amd.com>
Link: https://lore.kernel.org/r/20250702065717.3871435-1-venkatesh.abbarapu@amd.com
Signed-off-by: Michal Simek <michal.simek@amd.com>
Starting from 'commit <8077d296adff> ("spi: cadence-quadspi: Use STIG
mode for all ops with small payload") the utilization of STIG mode
has been implemented for read and write operations involving less
than 8 bytes of data.
However, following this commit, encountering timeout issues occurs when
writing odd bytes of data in DDR mode, as indicated below:
"jedec_spi_nor flash@0: flash operation timed out
SF: 3 bytes @ 0x0 Written: ERROR -110"
To resolve this issue, the number of bytes to write has been updated
specifically for DDR mode.
Signed-off-by: Tejas Bhumkar <tejas.arvind.bhumkar@amd.com>
Signed-off-by: Venkatesh Yadav Abbarapu <venkatesh.abbarapu@amd.com>
Link: https://lore.kernel.org/r/20250704040444.671604-1-venkatesh.abbarapu@amd.com
Signed-off-by: Michal Simek <michal.simek@amd.com>
Legacy SPI flash devices used a 24-bit (3-byte) addressing scheme,
limiting the addressable memory to 16 MB. To support larger densities
(256 Mbit and higher), extended addressing schemes, such as 32-bit
(4-byte) addressing, were introduced. If the flash density exceeds
16 MB and CONFIG_SPI_FLASH_BAR is disabled, the device will use a
4-byte addressing mode.
Signed-off-by: Prasad Kummari <prasad.kummari@amd.com>
Signed-off-by: Venkatesh Yadav Abbarapu <venkatesh.abbarapu@amd.com>
Link: https://lore.kernel.org/r/20250707043738.795179-1-venkatesh.abbarapu@amd.com
Signed-off-by: Michal Simek <michal.simek@amd.com>
