Manorit Chawdhry [Tue, 17 Dec 2024 08:54:37 +0000 (14:24 +0530)]
drivers: firmware: ti_sci: Add DM_FLAG_PRE_RELOC to driver
Currently the driver relies on bootph flag to probe it during PRE_RELOC
stage but with the upcoming cleanup of v6.13, we don't have the bootph
property in the parent nodes anymore and ti_sci driver being one of the
parent nodes required during SPL stage would end up hampering the probe
model [0].
Add DM_FLAG_PRE_RELOC to ti_sci driver for mitigating this issue.
Bryan Brattlof [Tue, 10 Dec 2024 21:43:30 +0000 (15:43 -0600)]
arm: dts: k3-am62p-sk-binman: add SE security variant builds
The Texas Instruments Foundational Security (TIFS) firmware must match
the security level configured on the SoC. To boot Security Enforced (SE)
variants of the AM62Px, add another tiboot3 build which packages the
Security Enforced (SE) firmware variant for AM62Px SoCs.
Vaishnav Achath [Wed, 18 Dec 2024 13:13:41 +0000 (18:43 +0530)]
arm: dts: k3-j722s*: Add overrides specific to OSPI
OSPI Boot requires overrides specific to R5 and also
to use DMA in R5 SPL stage the DM_TIFS needs to be used.
Add the corresponding overrides for R5 SPL stage.
Vaishnav Achath [Wed, 18 Dec 2024 13:13:39 +0000 (18:43 +0530)]
arm: dts: k3-j722s-r5-evm: Fix DM2TIFS secproxy thread ID
Fix the DM2TIFS secureproxy thread ID as per the latest TISCI
documentation for J722S.
https://software-dl.ti.com/tisci/esd/latest/5_soc_doc/j722s/sec_proxy.html
Weijie Gao [Tue, 17 Dec 2024 08:40:03 +0000 (16:40 +0800)]
net: mediatek: fix usability with wget command
The wget command currently cannot work correctly with mtk_eth driver.
This patch fixed this by increase DMA ring size and invalidate ring data
after use.
Weijie Gao [Tue, 17 Dec 2024 08:39:41 +0000 (16:39 +0800)]
net: mediatek: fix sgmii selection for mt7622
Unlike other platforms, mt7622 has only one SGMII and it can be
attached to either gmac1 or gmac2. So the register field of the
sgmii selection differs from other platforms as newer platforms can
control each sgmii individually.
This patch adds a new capability for mt7622 to handle this case.
Weijie Gao [Tue, 17 Dec 2024 08:39:23 +0000 (16:39 +0800)]
net: mediatek: use correct register field for SGMII speed selection
The register field for SGMII speed selection is a 2-bit field with
value 0 for 1Gbps and 1 for 2.5Gbps (2/3 are reserved).
So it's necessary to set both bits instead of just setting/clearing
only the lower bit.
Weijie Gao [Tue, 17 Dec 2024 08:39:16 +0000 (16:39 +0800)]
clk: mediatek: mt7629: fix parent clock of some top clock muxes
According to the mt7629 programming guide, the CLK_TOP_F10M_REF_SEL
shares the same parent selection with CLK_TOP_IRRX_SEL, while the
present parent selection for CLK_TOP_F10M_REF_SEL is actually used
for CLK_TOP_SGMII_REF_1_SEL.
Picking up a series from Dan Carpenter and applying requested
changes for v2.
I had previously set CONFIG_64BIT for arm64. This patchset does the
same thing for sandbox and x86_64. (Mips and riscv were already
doing it). This CONFIG option is used in the Makefile to determine
if it's a 32 or 64 bit system for the CHECKER.
Makefile
1052 # the checker needs the correct machine size
1053 CHECKFLAGS += $(if $(CONFIG_64BIT),-m64,-m32)
Andrew Goodbody [Mon, 16 Dec 2024 18:07:35 +0000 (18:07 +0000)]
x86: select CONFIG_64BIT for X86_64
Select CONFIG_64BIT so that we pass the -m64 option (instead of -m32) to
static analysis tools.
Introduce CONFIG_SPL_64BIT and select it for architectures other than
x86 with 64 bit builds. Do not select it for x86 builds as x86 uses
a 32 bit SPL.
Ensure that when limits are set they use CONFIG_64BIT for U-Boot
proper and CONFIG_SPL_64BIT for SPL. This is to allow for the 32 bit
SPL build used by x86.
Andrew Goodbody [Mon, 16 Dec 2024 18:07:33 +0000 (18:07 +0000)]
sandbox: Correct guard around readq/writeq
In include/linux/io.h the declarations of ioread64 and iowrite64
which make use of readq/writeq are guarded with CONFIG_64BIT so
guard the sandbox declarations of readq and writeq also with
CONFIG_64BIT.
Evgeny Bachinin [Tue, 10 Dec 2024 22:39:57 +0000 (01:39 +0300)]
fdtdec: encapsulate dtb_dt_embedded() within
Patch keeps the access to dtb_dt_embedded() within fdtdec API,
by means of new API function introduction. This new function is a
common place for updating appropriate global_data fields for
OF_EMBED case.
Also, the consequence of the patch is movement of '___dtb_dt_*begin'
symbols' declaration from header file, because nobody used symbols
outside the lib/fdtdec.c.
Tom Rini [Thu, 12 Dec 2024 22:33:36 +0000 (16:33 -0600)]
Gitlab: Remove some "rules:when: always" lines
In commit 399f739be6b2 ("CI: allow jobs to be run in merge requests") we
added "rules:when: always" to many stages of the pipeline to allow for merge
requests to trigger a run. However based on current Gitlab
documentation, we should still be triggered on merge requests without
this. Furthermore the way we have things written today we always run all
stages of the CI rather than failing out early on problems, which is not
always useful. Remove these as we should still be fine with merge
requests triggering a run.
Tom Rini [Thu, 12 Dec 2024 20:14:59 +0000 (14:14 -0600)]
Gitlab: Make test.py stage only depend on binman et al testsuite
Our Gitlab pipeline is currently broken up in to several stages. This
was done with the thought process of "we should test tools and if
they're good test emulated targets and if they're good test real
hardware and if they're good test the world". However, in terms of that
first stage it only really matters that binman, et al are still
functional. And for a few years now Gitlab has had a "needs" keyword
that lets you refine pipeline dependencies. Use this to perform the
minor optimization of having test.py only require that tool testing job.
This will become more useful later when we add long running testsuites
that we do not want to block later jobs.
Simon Glass [Fri, 20 Dec 2024 03:59:27 +0000 (20:59 -0700)]
sandbox: Adjust configuration to hang on panic()
It is annoying to have sandbox enter a boot loop when an assertion
fails. Hang instead, since then the error message is only printed once
and Ctrl-C can be used to quit, as per normal.
J. Neuschäfer [Fri, 20 Dec 2024 10:37:54 +0000 (11:37 +0100)]
gpio: mpc8xxx: Preserve pre-init state of outputs
The mpc8xxx_gpio driver contains a workaround for certain chips
where the previously written state of outputs cannot be read back
from the GPIO data (GPDAT) register (MPC8572/MPC8536). This workaround
consists of tracking the state of GPDAT in a "shadow register" (i.e. a
software variable). The shadow register is initialized to zero.
This results in a problem w.r.t. outputs that are configured to a
high (1) state before U-Boot runs, but not touched by U-Boot itself:
Due to the zero-initialization, these GPIOs end up being set to zero,
the first time that any other output is set.
To avoid such issues initialize the GPDAT shadow register to the value
previously held by any outputs, if possible. On MPC8572/MPC8536 this
should make no difference, i.e. the shadow register should be
initialized to zero on these chips.
This patch has been tested on a MPC8314E-based board.
J. Neuschäfer [Fri, 20 Dec 2024 10:37:52 +0000 (11:37 +0100)]
powerpc: mpc83xx: Allow including initreg.h into multiple files
Globals defined in headers can result in multiple-definition errors
while linking, if they are visible beyond the current translation unit.
This hasn't been a problem for initreg.h so far, but would become a
problem in the next patch, where I use a constant from initreg.h in a
second C file.
This patchset changes the definition r0 etc. to %r0, so that the
assembler can check that registers are only used where expected, and
fixes the fallout.
J. Neuschäfer [Thu, 12 Dec 2024 17:05:47 +0000 (18:05 +0100)]
powerpc: Introduce and enforce assembler checks on GPR usage
PowerPC general-purpose registers are historically specified as plain
numbers (0-31), which makes them hard to distinguish from immediates.
For this reason, include/ppc_asm.tmpl defines aliases named r0-r31.
This can still lead to uncaught mistakes if a register is used in place
of a number.
Instead of (e.g.) 5 use %r5, which will result in an assembler warning
if used as a number. Turn these warnings into errors by passing
`--fatal-warnings` to the assembler.
I verified with gazerbeam_defconfig (MPC83xx) and qemu-ppce500_defconfig
(MPC85xx) that this patch results in the same machine code.
J. Neuschäfer [Thu, 12 Dec 2024 17:05:46 +0000 (18:05 +0100)]
powerpc: Fix 0 vs. r0 confusion in X/D-form instructions
Instructions such as dcbi are in the X-form; they have RA and RB fields
and the effective address (EA) is computed as (RA|0)+(RB). In words,
this means that if RA is zero, the left-hand side of the addition is
zero, otherwise the corresponding GPR is used. r0 can never be used on
the left-hand side of a X-form instruction.
For D-form instructions such as addis, the Power ISA illustrates this in
the instruction pseudo-code:
if RA = 0 then RT <- EXTS(SI || 0x0000)
else RT <- (RA) + EXIS(SI || 0x0000)
In all of these cases, RA=0 indicates the value zero, not register r0.
I verified with gazerbeam_defconfig (MPC83xx) and qemu-ppce500_defconfig
(MPC85xx) that this patch results in the same machine code.
The LMB subsystem was used opportunistically for a number of years.
A while back Sughosh merged it with the EFI subsystem in order to have a
common allocator and avoid subsystems overwriting memory they shouldn't.
This is an initial cleanup of all the crud we gathered over the years.
There's no functional change expected from the patches as they just cleanup
some abstraction functions and rename a few variables to make more
sense.
I plan to make even bigger changes -- e.g I don't see the point of
having *_alloc() and *_reserve() versions of the functions since they
do the same thing and just cause confusion. lmb_alloc_addr_flags()
returning the base address on success makes little sense since we
already *request* the address on the function arguments, etc.
Since this patchset grew enough already, I'd like to get it in
before more refactoring happens.
It's worth noting that although some patches slightly increase the code
size due to an extra flags argument being carried around, the final
result is eventually smaller.
Ilias Apalodimas [Wed, 18 Dec 2024 07:02:37 +0000 (09:02 +0200)]
lmb: Rename _lmb_alloc_addr() to lmb_alloc_addr_flags()
lmb_alloc_addr_flags() is a wrapper for _lmb_alloc_addr() and it's the
only function using it. Rename _lmb_alloc_addr() to lmb_alloc_addr_flags()
and remove the wrapper.
Ilias Apalodimas [Wed, 18 Dec 2024 07:02:36 +0000 (09:02 +0200)]
lmb: Remove lmb_alloc_base_flags()
lmb_alloc_base() is just calling lmb_alloc_base_flags() with LMB_NONE.
There's not much we gain from this abstraction, so let's remove the
former add the flags argument to lmb_alloc_base() and make the code
a bit easier to follow.
Ilias Apalodimas [Wed, 18 Dec 2024 07:02:35 +0000 (09:02 +0200)]
lmb: Remove lmb_alloc_addr_flags()
lmb_alloc_addr() is just calling lmb_alloc_addr_flags() with LMB_NONE
There's not much we gain from this abstraction, so let's remove the
latter, add a flags argument to lmb_alloc_addr() and make the code a
bit easier to follow.
Ilias Apalodimas [Wed, 18 Dec 2024 07:02:33 +0000 (09:02 +0200)]
lmb: Rename free_mem to available_mem
free_mem is a misnomer. We never update it with the free memory for
LMB. Instead, it describes all available memory and is checked against
used_mem to decide whether an area is free or not.
So let's rename this field to better match its usage.
Ilias Apalodimas [Wed, 18 Dec 2024 07:02:32 +0000 (09:02 +0200)]
lmb: Remove lmb_reserve_flags()
lmb_reserve() is just calling lmb_reserve_flags() with LMB_NONE.
There's not much we gain from this abstraction.
So let's remove the latter, add the flags argument to lmb_reserve()
and make the code a bit easier to follow.
Ilias Apalodimas [Wed, 18 Dec 2024 07:02:31 +0000 (09:02 +0200)]
lmb: Move enum lmb_flags to a u32
LMB flags is not an enum anymore. It's currently used as a bitmask
in various places of our code. So make it a u32 which is more
appropriate when dealing with masks.
Cleanup the LMB code a bit, after fixing the false positive error
messages. No functional change. This series depends on [1] (which is
"lmb: Fix reserving the same region multiple times").
Sam Protsenko [Wed, 11 Dec 2024 02:25:50 +0000 (20:25 -0600)]
lmb: Improve kernel-doc comments
Fix warnings from kernel-doc script. Improve and unify overall style of
kernel-doc comments in lmb source files. Move all kernel-doc comments
for public functions into the header, as recommended in U-Boot
documentation [1]:
Non-trivial functions should have a comment which describes what
they do. If it is an exported function, put the comment in the
header file so the API is in one place. If it is a static function,
put it in the C file.
This also takes care of existing duplication. While at it, do a bit of
cosmetic cleanups as well.
Sam Protsenko [Wed, 11 Dec 2024 02:25:47 +0000 (20:25 -0600)]
lmb: Fix flags data type in lmb_add_region_flags()
rgnflags variable in lmb_add_region_flags() has incorrect type: it's
declared as phys_size_t when it should be enum lmb_flags. That
copy-paste mistake was firstly introduced in commit 59c0ea5df33f ("lmb:
Add support of flags for no-map properties"), and then copied further
into commit ed17a33fed29 ("lmb: make LMB memory map persistent and
global"). Fix it by using the correct type to match struct lmb_region
field.
mtd: spi-nor: Fix the spi_nor_read() when config SPI_STACKED_PARALLEL is enabled
Update the spi_nor_read() function based on the config SPI_FLASH_BAR
and update the length and bank calculation by spliting the memory of
16MB size banks only when the address width is 3byte.
Fix the read issue for 4byte address width by passing the entire
length to the read function.
Marek Vasut [Fri, 20 Dec 2024 00:48:42 +0000 (01:48 +0100)]
net: rswitch: Implement C22 to C45 access
Add support for mapping C22 register access to C45-only PHYs.
This is mainly useful for 'mii info' command, which performs
C22 only access to determine PHY ID and link state and does
not work well with this driver so far.
Marek Vasut [Fri, 20 Dec 2024 00:48:41 +0000 (01:48 +0100)]
net: rswitch: Add PHY C22 access support
Implement C22 PHY access support in addition to C45 PHY access
support which is already present. This is used for PHYs which
do not support C45 access or which are C22 only.
The C22 access can be recognized when devad is set to -1 or
0xffffffff hex, which also matches MDIO_DEVAD_NONE macro. Test
for this special devad value and if it is set this way, perform
C22 access, otherwise perform C45 access.
Marek Vasut [Fri, 20 Dec 2024 00:48:40 +0000 (01:48 +0100)]
net: rswitch: Fold MPSM C45 setting into rswitch_mii_access_c45()
The Set Station Management Mode : Clause 45 setting of MFF bit in MPSM
register can be done in rswitch_mii_access_c45() once, instead of this
being done before each rswitch_mii_access_c45() call. Deduplicate the
bit setting into rswitch_mii_access_c45(). No functional change.
Marek Vasut [Sun, 15 Dec 2024 23:57:49 +0000 (00:57 +0100)]
net: rswitch: Replace enum rswitch_reg with plain macros
Replace enum rswitch_reg with plain #define REGISTER OFFSET macros.
The enum rswitch_reg was not referenced anywhere, so there was no
benefit of keeping it around. Include register block labels. Turn
all register offsets into lowercase hex values. No functional change.
Rename EATDQDC to EATDQDCR, GWTRC to GWTRCR, GWDCC to GWDCCR, FWPC0
to FWPC, FWPBFC to FWPBFCR, FWPBFCSDC to FWPBFCSDCR because there
are both register names which used to be part of this enum and also
macros with the same name, each used for slightly different purpose.
Make sure there is no collission.
Marek Vasut [Fri, 20 Dec 2024 00:02:16 +0000 (01:02 +0100)]
arm64: renesas: Enable CR52 remoteproc on R-Car R8A779G0 V4H
Enable remoteproc command and APMU remoteproc driver to start Cortex-R52
cores from U-Boot command line. Code on the Cortex-R52 #0 can be started
as follows, code on other cores can be started by passing the correct ID
to 'rproc load' and 'rproc start' to select the core:
Describe APMU controller as a remoteproc device capable of starting
the Cortex-R52 cores in Renesas R8A779G0 V4H SoC DT. The APMU IP is
in fact a power management unit capable of additional operations, but
those are not used by U-Boot so far.
Add R-Car Gen4 APMU controller remoteproc driver capable of starting
the Cortex-R52 cores in Renesas R8A779G0 V4H/V4M SoC. The APMU IP is
in fact a power management unit capable of additional operations, but
those are not used by U-Boot so far.
This requires slight adjustment to the SPL entry point code, as that
is being executed on the Cortex-R52 #0 and the Cortex-R52 #0 enters an
endless loop once it starts the rest of the SPL on Cortex-A76 core.
The endless loop now checks for content of APMU CRBARP registers and
tests whether valid VLD_BARP and BAREN_VALID bits are set, if so, the
Cortex-R52 core exits the endless loop and jumps to address started
in CRBARP[31:18] register in ARM mode, which is a trampoline code to
jump to the final entry point.
The trampoline code is in place to avoid limitation of CRBARP[31:18]
address field, which limits the core start address to memory addresses
aligned to 0x40000 or 256 kiB . The trampoline is placed at 0x40000
aligned address and jumps to the final entry point, which can be at
an address with arbitrary alignment at instruction granularity.
Marek Vasut [Sat, 14 Dec 2024 22:45:52 +0000 (23:45 +0100)]
arm64: dts: renesas: Drop OF_UPSTREAM conversion remnant
This DTC_FLAGS assignment is no longer necessary as all R-Car Gen2/Gen3/Gen4
platforms have been converted to OF_UPSTREAM and matching DTC_FLAGS assignment
is present in dts/upstream/src/arm64/Makefile . Drop the remnant.
Marek Vasut [Sat, 14 Dec 2024 22:45:11 +0000 (23:45 +0100)]
arm64: renesas: Align configuration headers
Align R-Car Gen2/Gen3/Gen4 configuration header file to look
basically the same way across these three SoC generations.
There are subtle difference between the remaining bits in
those files across SoC generations, but the common bits are
now aligned. There is not much left in those headers either,
most of the configuration is now converted to Kconfig.
Specifically for R-Car Gen3, GIC registers have been moved
to architecture specific header file rcar-gen3-base.h , the
rest of the changes here are comment changes and indentation
changes.
Add support for building U-Boot SPL for Renesas R-Car Gen4 R8A779G0 V4H SoC.
The SPL initializes the DBSC5 DRAM controller, RT-VRAM and loads and starts
U-Boot proper on the Cortex-A76 core.
The SoC BootROM can not boot the CA76 core directly, instead the SPL starts
on the CR52 core which immediately brings up the CA76 core, which in turn
starts executing the actual SPL. This is achieved by placing a tiny bit of
precompiled Aarch32 code at the very beginning of the SPL. The code consists
of some 32 instructions, uses APMU to configure CA76 start address to offset
0x80 Bytes from start of the SPL, and uses APMU to start the CA76 core. The
code parts the CR52 core in an endless loop once the CA76 core got started.
The 32 instructions are completely arbitrary number, so is the offset 0x80
Bytes from start of SPL, because 0x80 = 128 decimal and 128 / 4 bytes per
instruction is 32 instructions. The 32 instructions turned out to be enough
to started the CA76 and 0x80 is nicely aligned.
Once the SPL completes hardware initialization, the SPL loads U-Boot proper.
The u-boot.itb proper fitImage contains 64bit build on u-boot-nodtb.bin and
a DT for R8A779G0 V4H White Hawk board and is generated by binman. The
u-boot.itb is loaded from SPI NOR offset 0x80000.
In order to install this setup on an existing R8A779G0 V4H White Hawk board,
build using r8a779g0_whitehawk_defconfig, generate SPI NOR image flash.bin
and write flash.bin to SPI NOR offset 0x0 . Finally, configure board MD pin
switches according to the R8A779G0 V4H White Hawk board documentation for
40 MHz SPI NOR boot using DMA and restart the board.
Marek Vasut [Sat, 21 Dec 2024 21:46:36 +0000 (22:46 +0100)]
arm64: renesas: Make bottom 128 MiB of DRAM available in EL3
In case U-Boot runs in EL3, which is the highest privilege level on ARM64,
there can be no firmware running that would restrict access to the bottom
128 MiB of DRAM. In fact, it is likely that U-Boot would have to load that
firmware into those bottom 128 MiB of DRAM and start that firmware.
Make those bottom 128 MiB of DRAM available in case U-Boot runs in EL3 to
allow loading the firmware to that area.
Marek Vasut [Thu, 12 Dec 2024 13:38:27 +0000 (14:38 +0100)]
arm64: renesas: Split R-Car Gen3 and Gen4 common board code
Split common board code for R-Car Gen3 and Gen4 into separate files.
The R-Car Gen3 board code contains fixups specific to TFA which are
no longer required on R-Car Gen4, keep those fixups in its own file
so they would not interfere with Gen4.
All R-Car Gen4 board files are copies of one another at this point.
Deduplicate them into single board/renesas/rcar-common/gen4-common.c
and remove all the duplicates. The one exception is R-Car V3U Falcon
board, which enables RWDT reset in board_init(), conditionally build
RWDT enablement in board_init() in the new common code for V3U.
Marek Vasut [Thu, 12 Dec 2024 13:36:09 +0000 (14:36 +0100)]
arm64: renesas: Make stub PSCI implementation available on 64bit R-Car SoCs
Make the R-Car V3U stub PSCI implementation available on 64bit R-Car SoCs.
This implementation is useful during early board bring up, where it can
supplant missing fully-featured PSCI implementation. Note that this PSCI
implementation is very basic and offers only SoC reset functionality. It
is unable to enable or disable secondary CPU cores nor does it offer any
suspend/resume functionality.
Describe DBSC5 DRAM controller and RT-VRAM configuration interface
as two new DT nodes in R-Car Gen4 R8A779G0 U-Boot DT extras file.
This node is used by the U-Boot SPL for R8A779G0 SoC, where the
DBSC5 and RT-VRAM drivers bind to these nodes and bring up the
DRAM controller and RT-VRAM settings respectively, so U-Boot
proper can be loaded into DRAM and started on Cortex A76 core.
Marek Vasut [Thu, 12 Dec 2024 13:34:30 +0000 (14:34 +0100)]
ram: renesas: Add Renesas R-Car Gen4 DBSC5 driver
Add Renesas R-Car Gen4 DBSC5 DRAM controller driver. This driver is currently
capable of bringing LPDDR5 DRAM on Renesas R-Car V4H Whitehawk board. Further
boards can be supported by supplying board specific DRAM configuration data
via dbsc5_get_board_data(). Support for R-Car V4M is not implemented, however
the driver is already mostly prepared to support this SoC.
Marek Vasut [Sat, 14 Dec 2024 22:47:43 +0000 (23:47 +0100)]
pinctrl: renesas: Convert to IS_ENABLED() macro
Use the IS_ENABLED() macro to reduce amount of #ifdef use in the driver
and improve code coverage. With IS_ENABLED() macro, the code is compiled
and then optimized out, which prevents bitrot.
In case no PFC table matches the SoC in use, do not probe the driver
and instead exit with -ENODEV. This should never happen under normal
conditions, because this would mean the driver DT compatible string
match happened, but the list in probe() cannot match the model listed
in match data associated with the compatible string on which the match
did happen.
Legacy TCP stack is bad. Here are some of the known issues:
* tcp packet from other connection can break a current one
* tcp send sequence always starts from zero
* bad tcp options processing
* strange assumptions on packet size for selective acknowledge
* tcp interface assumes one of the two scenarios:
- data downloading from remote host to a board
- request-response exchange with a small packets
so it's not possible to upload large amount of data from the
board to remote host.
* wget test generate bad tcp stream, test should fail but it passes instead
This series of patches fixes all of the above issues.
The benefits:
* A lot of bug was fixed
* Better and more reliable TCP state machine
* Tcp clients becomes smaller/simpler
* Data uploading was fixed (now it's possible to transmit a huge amount of
data from the board to remote host)
Modification was verified with
* firmware downloading via u-boot wget command
* fastboot over tcp
* netcat linux client using test netcat implementation (not included
to this patch series)
* Firefox/Chrome/Edge using test web-server implementation (not included
to this patch series)
[trini: snip]
WARNING: The v16 patch series does NOT fix lib/efi_selftest/efi_selftest_http.c
issue. It looks like the efi_selftest_http test is wrong by itself. The
following issues were detected during efi_selftest_http test study:
* The test should fail with HTTP status code 404 because:
* nowday most web-servers requires the presence of "HOST:" request header
* wget does not support sending "HOST:" request header
* web-server of "http://example.com/" site does NOT provide "default server"
configuration, so it answer 404 on any request without "HOST:" header.
* The test states that:
* test send HTTP HEAD request to a server,
* then test send HTTP GET request to a server,
* reads the actual bytes sent by the server and compare it with
the value from "Contents-Length:" responce header of the HEAD request
But actually it
* does not send HTTP HEAD request, only a single HTTP GET request
is performed
* the test reads the responce twice from the same request. It looks
very suspictiuos
Some driver implements it's own network packet pool, so PKTBUFSRX is zero.
This results in zero-size TCP receive window, so data transfer doesn't
work. Avoid it by setting a reasonable fallback value.
test/cmd/wget: replace bogus response with an actual response from the HTTP server
According to HTTP/1.0 standard the HTTP reply consist of
* Status Line + CRLF
* Zero or more Response Header Fields (each ended with CRLF)
* CRLF on new line (Response Header Fields end marker)
* Optional Entity Body.
Thus in response headers we state:
Content-Length = 30
but actual transferred file data is:
"\r\n<html><body>Hi</body></html>\r\n".
This is 32 bytes of data.
So we get and check for correctness 32 bytes of data, but
* The response we are used is incorrect, real server will
set Content-Length to 32.
* default_wget_info->hdr_cont_len will be set to wrong
value 30 (used for efi http booting).
Fix an issue by:
* replace bogus response with an actual response from the HTTP server
* format response to show HTTP response structure
* recalculate md5sum as transferred file data has been changed.
The server response was captured with the commands
net/tcp: improve tcp framework, use better state machine
Changes:
* Fix initial send sequence always zero issue
* Use state machine close to RFC 9293. This should make TCP
transfers more reliable (now we can upload a huge array
of data from the board to external server)
* Improve TCP framework a lot. This should make tcp client
code much more simple.
* rewrite wget with new tcp stack
* rewrite fastboot_tcp with new tcp stack
It's quite hard to fix the initial send sequence (ISS) issue
with the separate patch. A naive attempt to fix an issue
inside the tcp_set_tcp_header() function will break tcp packet
retransmit logic in wget and other clients.
Example:
Wget stores tcp_seq_num value before tcp_set_tcp_header() will
be called and (on failure) retransmit the packet with the stored
tcp_seq_num value. Thus:
* the same ISS must allways be used (current case)
* or tcp clients needs to generate a proper ISS when
required.
A proper ISS fix will require a big redesing comparable with
a this one.
net/tcp: add connection info to tcp_stream structure
Changes:
* Avoid use net_server_ip in tcp code, use tcp_stream data instead
* Ignore packets from other connections if connection already created.
This prevents us from connection break caused by other tcp stream.
Current code assume that all (except last) packets are of the same size.
This is definitely wrong. Replace SACK code with a new one, that does
not rely on this assumption. Also this code uses less memory.
This includes various patches towards implementing the VBE abrec
bootmeth in U-Boot. It mostly focuses on SPL tweaks and adjusting what
fatures are available in VPL.
Simon Glass [Thu, 19 Dec 2024 18:29:00 +0000 (11:29 -0700)]
spl: Drop use of uintptr_t
U-Boot uses ulong for addresses. It is confusing to use uintptr_t in a
few places, since it makes people wonder if the types are compatible.
Change the few occurences in SPL to use ulong