kwboot: Marvell Dove UART booting
Pali Rohár
pali at kernel.org
Mon Feb 14 01:21:46 CET 2022
On Sunday 13 February 2022 16:08:51 Tony Dinh wrote:
> Hi Pali,
>
> On Sun, Feb 13, 2022 at 3:42 PM Pali Rohár <pali at kernel.org> wrote:
> >
> > On Sunday 13 February 2022 15:24:46 Tony Dinh wrote:
> > > Hi Pali,
> > >
> > > Looks different, but the BootROM did not start the image. Please see
> > > the log below.
> > >
> > > On Sun, Feb 13, 2022 at 2:48 PM Pali Rohár <pali at kernel.org> wrote:
> > > >
> > > > On Sunday 13 February 2022 14:45:07 Tony Dinh wrote:
> > > > > Hi Pali,
> > > > >
> > > > > Some compile errors, please see below.
> > > > >
> > > > > On Sun, Feb 13, 2022 at 8:16 AM Pali Rohár <pali at kernel.org> wrote:
> > > > > >
> > > > > > On Wednesday 22 December 2021 20:08:56 Tony Dinh wrote:
> > > > > > > *** Run kwboot
> > > > > > >
> > > > > > > # kwboot -t -p -B 115200 /dev/ttyUSB0 -D /localdisk/mtd0.t5335z
> > > > > > > Patching image boot signature to UART
> > > > > > > Aligning image header to Xmodem block size
> > > > > > > Waiting 2s and flushing tty
> > > > > > > Sending boot image header (512 bytes)...
> > > > > > > 25 % [.... ]
> > > > > > > Done
> > > > > > > Sending boot image data (607664 bytes)...
> > > > > > > 0 % [......................................................................]
> > > > > > > 1 % [......................................................................]
> > > > > > > 2 % [......................................................................]
> > > > > > > <snip>
> > > > > > > 95 % [......................................................................]
> > > > > > > 97 % [......................................................................]
> > > > > > > 98 % [.......................................................... ]
> > > > > > > Done
> > > > > > > Finishing transfer
> > > > > > > [Type Ctrl-\ + c to quit]
> > > > > > >
> > > > > > > *** Hung here! BootROM did not execute the image payload.
> > > > > > > ***
> > > > > > > *** The file mtd0.t5335z is a dd dump from the SPI flash mtd0 with
> > > > > > > *** this command:
> > > > > > > *** # dd if=/dev/mtd0 of=mtd0.t5335z bs=768k conv=sync
> > > > > > >
> > > > > > > <End log>
> > > > > > >
> > > > > > >
> > > > > > > - Pali's observation:
> > > > > > >
> > > > > > > It looks like Dove uses kwbimage v0 format with extensions, at
> > > > > > > least according to Function Spec. See 'Binary Code Extension' and
> > > > > > > 'Header Extension'. Currently kwboot and kwbimage supports v0 image only
> > > > > > > with one extension.
> > > > > >
> > > > > > I quickly looked at it. Could you try following patch?
> > > > > >
> > > > > > diff --git a/tools/kwbimage.h b/tools/kwbimage.h
> > > > > > index 74e5d87a4fef..15e83ececc76 100644
> > > > > > --- a/tools/kwbimage.h
> > > > > > +++ b/tools/kwbimage.h
> > > > > > @@ -61,14 +64,46 @@ struct ext_hdr_v0_reg {
> > > > > > uint32_t rdata;
> > > > > > } __packed;
> > > > > >
> > > > > > -#define EXT_HDR_V0_REG_COUNT ((0x1dc - 0x20) / sizeof(struct ext_hdr_v0_reg))
> > > > > > -
> > > > > > +/* Structure of the extension header, version 0 (Kirkwood, Dove) */
> > > > > > struct ext_hdr_v0 {
> > > > > > - uint32_t offset;
> > > > > > - uint8_t reserved[0x20 - sizeof(uint32_t)];
> > > > > > - struct ext_hdr_v0_reg rcfg[EXT_HDR_V0_REG_COUNT];
> > > > > > - uint8_t reserved2[7];
> > > > > > - uint8_t checksum;
> > > > > > + /*
> > > > > > + * Beware that extension header offsets specified in 88AP510 Functional
> > > > > > + * Specifications are relative to the start of the main header, not to
> > > > > > + * the start of the extension header itself.
> > > > > > + */
> > > > > > + uint32_t offset; /* 0x0-0x3 */
> > > > > > + uint8_t rsvd1[8]; /* 0x4-0xB */
> > > > > > + uint32_t ddrinitdelay; /* 0xC-0xF */
> > > > > > + uint32_t match_addr; /* 0x10-0x13 */
> > > > > > + uint32_t match_mask; /* 0x14-0x17 */
> > > > > > + uint32_t match_value; /* 0x18-0x1B */
> > > > > > + uint8_t ddrwritetype; /* 0x1C */
> > > > > > + uint8_t ddrresetmpp; /* 0x1D */
> > > > > > + uint8_t ddrclkenmpp; /* 0x1E */
> > > > > > + uint8_t ddrmppdelay; /* 0x1F */
> > > > > > + struct ext_hdr_v0_reg rcfg[55]; /* 0x20-0x1D7 */
> > > > > > + uint8_t rsvd2[7]; /* 0x1D8-0x1DE */
> > > > > > + uint8_t checksum; /* 0x1DF */
> > > > > > +} __packed;
> > > > > > +
> > > > > > +/* Structure of the binary code header, version 0 (Dove) */
> > > > > > +struct binext_hdr_v0 {
> > > > > > + uint32_t match_addr; /* 0x00-0x03 */
> > > > > > + uint32_t match_mask; /* 0x04-0x07 */
> > > > > > + uint32_t match_value; /* 0x08-0x0B */
> > > > > > + uint32_t offset; /* 0x0C-0x0F */
> > > > > > + uint32_t destaddr; /* 0x10-0x13 */
> > > > > > + uint32_t size; /* 0x14-0x17 */
> > > > > > + uint32_t execaddr; /* 0x18-0x1B */
> > > > > > + uint32_t param1; /* 0x1C-0x1F */
> > > > > > + uint32_t param2; /* 0x20-0x23 */
> > > > > > + uint32_t param3; /* 0x24-0x27 */
> > > > > > + uint32_t param4; /* 0x28-0x2B */
> > > > > > + uint8_t params; /* 0x2C */
> > > > > > + uint8_t rsvd1; /* 0x2D */
> > > > > > + uint8_t rsvd2; /* 0x2E */
> > > > > > + uint8_t checksum; /* 0x2F */
> > > > > > + uint8_t code[2000]; /* 0x30-0x7FF */
> > > > > > } __packed;
> > > > > >
> > > > > > /* Structure of the main header, version 1 (Armada 370/XP/375/38x/39x) */
> > > > > > @@ -213,8 +248,20 @@ static inline size_t kwbheader_size(const void *header)
> > > > > > if (kwbimage_version(header) == 0) {
> > > > > > const struct main_hdr_v0 *hdr = header;
> > > > > >
> > > > > > + /*
> > > > > > + * First extension header starts immediately after the main
> > > > > > + * header without any padding. Between extension headers is
> > > > > > + * 0x20 byte padding. There is no padding after the last
> > > > > > + * extension header. First binary code header starts immediately
> > > > > > + * after the last extension header (or immediately after the
> > > > > > + * main header if there is no extension header) without any
> > > > > > + * padding. There is no padding between binary code headers and
> > > > > > + * neither after the last binary code header.
> > > > > > + */
> > > > > > return sizeof(*hdr) +
> > > > > > - hdr->ext ? sizeof(struct ext_hdr_v0) : 0;
> > > > > > + hdr->ext * sizeof(struct ext_hdr_v0) +
> > > > > > + ((hdr->ext > 1) ? (hdr->ext * 0x20) : 0) +
> > > > > > + hdr->binext * sizeof(struct binext_hdr_v0);
> > > > > > } else {
> > > > > > const struct main_hdr_v1 *hdr = header;
> > > > > >
> > > > > >
> > > > > >
> > > > > > It fixes kwbheader_size() function to returns correct size of the image
> > > > > > header (with all v0 extensions), so it could help kwboot to convert
> > > > > > image with non-UART sign to UART version and send it over UART.
> > > > >
> > > > > Applied the patch, and make tools gave this error:
> > > > >
> > > > > tools/kwbimage.h:255:13: error: ‘const struct main_hdr_v0’ has no
> > > > > member named ‘binext’
> > > > > 255 | hdr->binext * sizeof(struct binext_hdr_v0);
> > > > >
> > > > > Thanks,
> > > > > Tony
> > > >
> > > > Ah, I have not sent whole patch...
> > > >
> > > > In tools/kwbimage.h for struct main_hdr_v0 { ... } replace
> > > >
> > > > uint16_t rsvd2;
> > > >
> > > > by
> > > >
> > > > uint8_t rsvd2;
> > > > uint8_t binext;
> > > >
> > > >
> > >
> > > Fixed that, and then I ran it the same way before. In the serial
> > > console, tell BootROM to set the boot device to UART, and then exit,
> > > run kwboot.
> > >
> > > <BEGIN log>
> > >
> > > Bootstrap 2.33>x 0x0E
> > >
> > > Terminating...
> > > Thanks for using picocom
> > >
> > > ./kwboot-2022/kwboot.dove -t -p -B 115200 /dev/ttyUSB0 -D ./t5335z/mtd0.t5335z
> > > kwboot version 2022.04-rc1-00360-g162c22bfbc-dirty
> > > Patching image boot signature to UART
> > > Sending boot image header (3072 bytes)...
> > > 4 % [........................ ]
> > > Done
> > > Sending boot image data (605104 bytes)...
> > > 0 % [......................................................................]
> > > <snip>
> > > 97 % [......................................................................]
> > > 99 % [...................................... ]
> > > Done
> > > Finishing transfer
> > > [Type Ctrl-\ + c to quit]
> > > <END log>
> > >
> > > It hung here. And I have to recycle power to boot again.
> > >
> > > Observation: in the previous run with unpatched kwboot, the header
> > > size was 512 bytes, and the image was 607664 bytes. So this time
> > > kwboot sends the extension header, too. Not sure if 3072 bytes is
> > > correct.
> >
> > If I parsed that image (file mtd0.t5335z) correctly it has:
> > * main header (0x20)
> > * ext header (0x1E0)
> > * 0x20 padding (0x20)
> > * ext header (0x1E0)
> > * bin header (0x800)
> > * image data (0x93bb0)
> >
> > So it is: 3072 bytes for headers + 605104 bytes for image data, which
> > now matches.
> >
> > Plus in main header is stored that image data starts at offset 0xc00
> > which is 3072 bytes.
> >
> > Now kwboot showed just that is patching UART signature and no more
> > alignment, so it should send headers and data without adding any
> > additional padding or alignment.
> >
> > If it still does not work, I do not have an idea...
> >
> > It would be required to take U-Boot source code for that board and build
> > UART image from sources. Maybe UART version needs to be compiled
> > differently, like it was in past also for Armada builds.
>
> To double check I dumped the images.
>
> ./kwboot-2022/dumpimage -T kwbimage -p -1 ./t5335z/mtd0.t5335z -o
> /tmp/mtd0.t5335z.cfg
> -rw-r--r-- 1 root root 1155 Feb 13 15:58 mtd0.t5335z.cfg
This cfg file is incomplete as kwbimage does not read those new
structures.
There is also bin header for '-p 1' but again kwbimage cannot extract it
yet.
(Btw, I have WIP patches for this stuff, if you are interesting I can send it)
> ./kwboot-2022/dumpimage -T kwbimage -p 0 ./t5335z/mtd0.t5335z -o
> /tmp/mtd0.t5335z.0.img
> -rw------- 1 root root 605100 Feb 13 15:58 mtd0.t5335z.0.img
>
> ./kwboot-2022/mkimage -l ./t5335z/mtd0.t5335z
> Image Type: MVEBU Boot from spi Image
> Image version:0
> Data Size: 605100 Bytes = 590.92 KiB = 0.58 MiB
> Load Address: 00600000
> Entry Point: 00690000
>
> So it is checked out. The image is 605100 with 4 bytes checksum =
> 605104. Did I read it correctly? i.e. the dump does not include the
> checksum?
Yes!
> Seems I got to get to work on that stock UART image!
>
> Thanks,
> Tony
More information about the U-Boot
mailing list