Skip to content

L
linux
Switch branch/tag
  1. 14 Dec, 2005 1 commit
  3. 12 Dec, 2005 1 commit
    • Lothar Wassmann's avatar
      [ARM] 3201/1: PXA27x: Prevent hangup during resume due to inadvertedly... · 1ee9530a
      Lothar Wassmann authored 
      [ARM] 3201/1: PXA27x: Prevent hangup during resume due to inadvertedly enabling MBREQ (replaces: 3198/1)

Patch from Lothar Wassmann

The patch makes sure, that the ouptut functions of pins are restored
before restoring the Alternat Function settings, preventing pins from
being intermediately configured for undefined or unwanted alternate
functions.

Here is the original comment:
I've got a PXA270 system that uses GPIO80 as nCS4. This system did
hang on resume. Digging into the problem I found that the processor
stalled immediately when restoring the GAFR2_U register which restored
the alternate function for GPIO80. Since the GPDR registers were
restored after the GAFR registers, the offending GPIO was configured
as input at this point.
Thus the alternate function that was in effect after restoring the
GAFR was in fact the input function "MBREQ" instead of the output
function "nCS4". The "PXA27x Processor Family Developer's Manual"
(Footnote in Table 6-1 on page 6-3) states that:
"The MBREQ alternate function must not be enabled until the PSSR[RDH]
bit field is cleared. For more details, see Table 3-15, "PSSR Bit
Definitions" on page 3-71."

There is another note in the Developer's Manual (chapter 24.4.2
"GPIO operation as Alternate Function" on page 24-4)
stating that:
"Configuring a GPIO for an alternate function that is not defined for
it causes unpredictable results."

Since some GPIOs have no input function defined, and to prevent
inadvertedly programming the MBREQ function on some pin, the GAFR
registers should be restored after the GPDR registers have been
restored.

Additional provisions have to be made when the MBREQ function is
actually required. The corresponding GAFR bits should not be restored
with the regular GAFR restore, but must be set only after the PSSR
bits have been cleared.
Signed-off-by: default avatarLothar Wassmann <LW@KARO-electronics.de>
Signed-off-by: default avatarRussell King <rmk+kernel@arm.linux.org.uk>
      1ee9530a
  5. 10 Dec, 2005 1 commit
    • Nikola Valerjev's avatar
      [ARM] 3200/1: Singlestep over ARM BX and BLX instructions using ptrace fix · 22f975f4
      Nikola Valerjev authored 
      Patch from Nikola Valerjev

Single stepping an application using ptrace() fails over ARM instructions BX and BLX.

Steps to reproduce:

Compile and link the following files

main.c
-----
void foo();
int main() {
    foo();
    return 0;
}

foo.s
-----
	.text
	.globl foo
foo:
	BX LR

Using ptrace() functionality, run to main(), and start singlestepping.
Singlestep over \"BX LR\" instruction won\'t transfer the control back
to main, but run the code to completion.

This problems seems to be in the function get_branch_address() in
arch/arm/kernel/ptrace.c. The function doesn\'t seem to recognize BX
and BLX instructions as branches. BX and BLX instructions can be used
to convert from ARM to Thumb mode if the target address has the low
bit set. However, they are also perfectly legal in the ARM only mode.
Although other things in the kernel seem to indicate that only ARM
mode is accepted (and not Thumb), many compilers will generate BX
and BLX instructions even when generating ARM only code.
Signed-off-by: default avatarNikola Valerjev <nikola@ghs.com>
Signed-off-by: default avatarRussell King <rmk+kernel@arm.linux.org.uk>
      22f975f4
  7. 08 Dec, 2005 1 commit
  9. 05 Dec, 2005 3 commits
  11. 04 Dec, 2005 6 commits
  13. 03 Dec, 2005 13 commits
  15. 02 Dec, 2005 5 commits
  17. 01 Dec, 2005 9 commits
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7