Re: [PATCH 0/5] Improve support for regcache_map_entry with variable register base

[John Baldwin <jhb@FreeBSD.org>]

On 7/7/22 5:58 PM, John Baldwin wrote:
> When I added support for TLS to the ARM and AArch64 architectures on
> FreeBSD, I had to construct register maps and associated register sets
> on the fly in both the callback for processing register core dump
> notes and in the native target routines to fetch and store registers.
> 
> The reason I had to create these on the fly is that the register
> number was not known at compile time, so the register map and register
> set could not be constant structures shared among the -tdep and -nat
> code the way other register sets were.
> 
> This series aims to rectify that.  The first patch adds new variants
> of regcache::collect_regset and regcache::supply_regset that accept an
> additional 'regbase' argument.  When these functions iterate over
> regcache map entries, the effective register number for the entry is
> computed by adding the value in the entry to 'regbase' permittting the
> regcache map entries to hold relative register numbers for a block of
> registers.
> 
> The rest of the series then makes use of this to use a single, shared
> register map and register set for TLS on FreeBSD/ARM and
> FreeBSD/AArch64.  Using this requires custom supply and collect regset
> routines for the TLS regsets that extract the base register number
> from the gdbarch's tdep and invoke the new regcache class methods, so
> patch 2 updates the wrappers for dealing with regsets in fbsd-nat to
> always invoke the regset routines from the regset instead of directly
> calling the default functions.  Patch 3 is another change to fbsd-nat
> to cope with the fact that regcache_map_supplies() needs to pass in
> the relative register number in the wrapper routines rather than the
> absolute register number.
> 
> Patches 4 and 5 are the updates to the ARM and AArch64 FreeBSD
> architectures and targets.
> 
> I have not done the work to make use of this in the Linux AArch64
> architecture, though I think it would apply to at least the TLS and
> MTE register sets there in a similar fashion.  SVE is harder because
> the register sizes change (though SVE might be able to make use of
> this if register_size() returns the right value by using a register
> size of 0 in the relevant register cache map entries).
> 
> John Baldwin (5):
>    regcache: Add collect/supply_regset variants that accept a register
>      base.
>    fbsd-nat: Use regset supply/collect methods.
>    fbsd-nat: Pass an optional register base to the register set helpers.
>    arm-fbsd: Use a static regset for the TLS register set.
>    aarch64-fbsd: Use a static regset for the TLS register set.
> 
>   gdb/aarch64-fbsd-nat.c  | 36 +++------------------------
>   gdb/aarch64-fbsd-tdep.c | 54 ++++++++++++++++++++++++++++-------------
>   gdb/aarch64-fbsd-tdep.h |  1 +
>   gdb/arm-fbsd-nat.c      | 34 +++-----------------------
>   gdb/arm-fbsd-tdep.c     | 52 +++++++++++++++++++++++++++------------
>   gdb/arm-fbsd-tdep.h     |  1 +
>   gdb/fbsd-nat.c          | 42 +++++++++++++++++---------------
>   gdb/fbsd-nat.h          | 49 +++++++++++++++++++++----------------
>   gdb/regcache.c          | 28 ++++++++++++++++++---
>   gdb/regcache.h          | 24 ++++++++++++++++--
>   10 files changed, 181 insertions(+), 140 deletions(-)

Ping.

-- 
John Baldwin