| Message ID | 1441052989-14603-1-git-send-email-kevin.smith@elecsyscorp.com |
|---|---|
| State | Deferred |
| Delegated to: | Stefan Roese |
| Headers | show |
Hi Kevin, (Added Luka to Cc, as the Marvell / MVEBU custodian) On 31.08.2015 22:30, Kevin Smith wrote: > On some processors such as Armada 38x, if the hardware- > configured boot mode fails, the CPU falls back to booting over > UART. When this happens the chip prints a failure message, waits > for the magic sequence and, when it is received, prints a > "(boot)" message, then sends a NAK to start the transfer. > > This breaks the current kwboot behavior because the xmodem > transfer only tries to read one character after the magic > sequence, looking for the NAK. Instead it gets the "(boot)" > text, and retries the magic sequence. The CPU thinks the > repeated sequence is part of the packet, stops NAKing, and one > side or another eventually times out. > > This patch adds support for a fallback mode which continues to > scan for a NAK in the characters received after the sequence, > printing out any non-NAK characters. This allows kwboot to skip > the "(boot)" message, find the NAK, and start the transfer > successfully. > > Signed-off-by: Kevin Smith <kevin.smith@elecsyscorp.com> I've not seen this "(boot)" yet. But the patch looks good. So: Reviewed-by: Stefan Roese <sr@denx.de> BTW: Please always keep Luka on MVEBU related patches on Cc. Thanks, Stefan
Hi Kevin, On 01.09.2015 14:52, Stefan Roese wrote: > Hi Kevin, > > (Added Luka to Cc, as the Marvell / MVEBU custodian) > > On 31.08.2015 22:30, Kevin Smith wrote: >> On some processors such as Armada 38x, if the hardware- >> configured boot mode fails, the CPU falls back to booting over >> UART. When this happens the chip prints a failure message, waits >> for the magic sequence and, when it is received, prints a >> "(boot)" message, then sends a NAK to start the transfer. >> >> This breaks the current kwboot behavior because the xmodem >> transfer only tries to read one character after the magic >> sequence, looking for the NAK. Instead it gets the "(boot)" >> text, and retries the magic sequence. The CPU thinks the >> repeated sequence is part of the packet, stops NAKing, and one >> side or another eventually times out. >> >> This patch adds support for a fallback mode which continues to >> scan for a NAK in the characters received after the sequence, >> printing out any non-NAK characters. This allows kwboot to skip >> the "(boot)" message, find the NAK, and start the transfer >> successfully. >> >> Signed-off-by: Kevin Smith <kevin.smith@elecsyscorp.com> > > I've not seen this "(boot)" yet. But the patch looks good. So: > > Reviewed-by: Stefan Roese <sr@denx.de> Is this patch still needed (or helpful)? If yes, please rebase on top of latest u-boot-marvell/master and send a new version so that I can pick it up. Thanks, Stefan
Hi Stefan, On 03/24/2016 04:29 AM, Stefan Roese wrote: > Hi Kevin, > > On 01.09.2015 14:52, Stefan Roese wrote: >> Hi Kevin, >> >> (Added Luka to Cc, as the Marvell / MVEBU custodian) >> >> On 31.08.2015 22:30, Kevin Smith wrote: >>> On some processors such as Armada 38x, if the hardware- >>> configured boot mode fails, the CPU falls back to booting over >>> UART. When this happens the chip prints a failure message, waits >>> for the magic sequence and, when it is received, prints a >>> "(boot)" message, then sends a NAK to start the transfer. >>> >>> This breaks the current kwboot behavior because the xmodem >>> transfer only tries to read one character after the magic >>> sequence, looking for the NAK. Instead it gets the "(boot)" >>> text, and retries the magic sequence. The CPU thinks the >>> repeated sequence is part of the packet, stops NAKing, and one >>> side or another eventually times out. >>> >>> This patch adds support for a fallback mode which continues to >>> scan for a NAK in the characters received after the sequence, >>> printing out any non-NAK characters. This allows kwboot to skip >>> the "(boot)" message, find the NAK, and start the transfer >>> successfully. >>> >>> Signed-off-by: Kevin Smith <kevin.smith@elecsyscorp.com> >> >> I've not seen this "(boot)" yet. But the patch looks good. So: >> >> Reviewed-by: Stefan Roese <sr@denx.de> > > Is this patch still needed (or helpful)? If yes, please rebase > on top of latest u-boot-marvell/master and send a new version > so that I can pick it up. > > Thanks, > Stefan > I don't think this is needed in most cases. I will leave it out to avoid cluttering up the program. If that changes, I will resubmit. Thank you, Kevin
diff --git a/tools/kwboot.c b/tools/kwboot.c index af7a6ee..6b58fc9 100644 --- a/tools/kwboot.c +++ b/tools/kwboot.c @@ -71,6 +71,7 @@ struct kwboot_block { #define KWBOOT_BLK_RSP_TIMEO 1000 /* ms */ static int kwboot_verbose; +static int support_fallback; static int msg_req_delay = KWBOOT_MSG_REQ_DELAY; static int msg_rsp_timeo = KWBOOT_MSG_RSP_TIMEO; @@ -300,7 +301,15 @@ kwboot_bootmsg(int tty, void *msg) continue; } - rc = kwboot_tty_recv(tty, &c, 1, msg_rsp_timeo); + while (1) { + rc = kwboot_tty_recv(tty, &c, 1, msg_rsp_timeo); + if (support_fallback && rc == 0 && c != NAK) { + printf("%c", c); + fflush(stdout); + continue; + } + break; + } kwboot_spinner(); @@ -657,7 +666,7 @@ static void kwboot_usage(FILE *stream, char *progname) { fprintf(stream, - "Usage: %s [-d | -a | -q <req-delay> | -s <resp-timeo> | -b <image> | -D <image> ] [ -t ] [-B <baud> ] <TTY>\n", + "Usage: %s [-f] [-d | -a | -q <req-delay> | -s <resp-timeo> | -b <image> | -D <image> ] [ -t ] [-B <baud> ] <TTY>\n", progname); fprintf(stream, "\n"); fprintf(stream, @@ -667,6 +676,7 @@ kwboot_usage(FILE *stream, char *progname) " -D <image>: boot <image> without preamble (Dove)\n"); fprintf(stream, " -d: enter debug mode\n"); fprintf(stream, " -a: use timings for Armada XP\n"); + fprintf(stream, " -f: support UART fallback after failed boot\n"); fprintf(stream, " -q <req-delay>: use specific request-delay\n"); fprintf(stream, " -s <resp-timeo>: use specific response-timeout\n"); fprintf(stream, "\n"); @@ -701,7 +711,7 @@ main(int argc, char **argv) kwboot_verbose = isatty(STDOUT_FILENO); do { - int c = getopt(argc, argv, "hb:ptaB:dD:q:s:"); + int c = getopt(argc, argv, "hb:ptfaB:dD:q:s:"); if (c < 0) break; @@ -728,6 +738,10 @@ main(int argc, char **argv) term = 1; break; + case 'f': + support_fallback = 1; + break; + case 'a': msg_req_delay = KWBOOT_MSG_REQ_DELAY_AXP; msg_rsp_timeo = KWBOOT_MSG_RSP_TIMEO_AXP;
On some processors such as Armada 38x, if the hardware- configured boot mode fails, the CPU falls back to booting over UART. When this happens the chip prints a failure message, waits for the magic sequence and, when it is received, prints a "(boot)" message, then sends a NAK to start the transfer. This breaks the current kwboot behavior because the xmodem transfer only tries to read one character after the magic sequence, looking for the NAK. Instead it gets the "(boot)" text, and retries the magic sequence. The CPU thinks the repeated sequence is part of the packet, stops NAKing, and one side or another eventually times out. This patch adds support for a fallback mode which continues to scan for a NAK in the characters received after the sequence, printing out any non-NAK characters. This allows kwboot to skip the "(boot)" message, find the NAK, and start the transfer successfully. Signed-off-by: Kevin Smith <kevin.smith@elecsyscorp.com> --- tools/kwboot.c | 20 +++++++++++++++++--- 1 file changed, 17 insertions(+), 3 deletions(-)