This is the mail archive of the gdb-patches@sources.redhat.com mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Allow 'NULL' to i387_supply_fsave et. al.


Whoops, I seem to have let this go stale:

On Saturday, October 12, 2002, at 01:59 PM, Mark Kettenis wrote:

Klee Dienes <klee@apple.com> writes:

The following patch allows one to pass 'NULL' to i387_supply_fsave,
i387_supply_fxsave, and i387_supply_register, to mean "no register data
is available".  This is convenient for clients of i387-tdep.c that may
have gotten a "no register data available" response from the OS while
trying to fetch information, and causes the i387_supply_* functions to
have the same behavior as supply_register.

Makes sense to me, but perhaps you should take a look at
i386-linux-nat.c:dummy_sse_values().  On Linux, if the SSE registers
aren't available, they're filled with some non-zero dummy values.  I
think we should try to be consistent about this across platforms, so
perhaps calling i387_supply_fxsave() with NULL should do something
similar.


I'd actually argue that the i386/linux code should be changed to call supply_register (NULL), instead of setting up dummy values.  That way if we ever improve supply_register to distinguish "not available" from "set to zero", the correct information will be being provided by the i386 code.
Just FYI,

Registers can be in 4 states (this came up with a really long discussion to do with the register cache):

Invalid: yet to ask the target
Valid: target supplied value
Unavailable: target couldn't supply value, try again after next resume
DNE: target will never supply value (not supported on this ISA variant or target configuration)

At present the cache only represents three of those register states.

There should be a better way for the target to indicate this to the register cache.

enjoy,
Andrew

(I guess we can also toss in state 5 - modified, not yet written back :-).

Assuming your origional ChangeLog entry doesn't have these glaring
formatting problems, please go ahead and check this in.

It doesn't --- my mailer is "officially weird".  I'll check in the patch unless someone objects in the next day or two.




Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]