On 07/02/2013 03:40 PM, Andreas Färber wrote:
Am 01.07.2013 23:34, schrieb Peter Maydell:
On 1 July 2013 21:52, Paolo Bonzini wrote:
Il 01/07/2013 18:26, Peter Maydell ha scritto:
Since the two largest cases are both "cpu_single_env must be TLS"
(ie (a) system emulation built with KVM support
Am 01.07.2013 23:34, schrieb Peter Maydell:
> On 1 July 2013 21:52, Paolo Bonzini wrote:
>> Il 01/07/2013 18:26, Peter Maydell ha scritto:
>>> Since the two largest cases are both "cpu_single_env must be TLS"
>>> (ie (a) system emulation built with KVM support and (b) linux-user),
>>> the set of t
On 1 July 2013 21:52, Paolo Bonzini wrote:
> Il 01/07/2013 18:26, Peter Maydell ha scritto:
>> Since the two largest cases are both "cpu_single_env must be TLS"
>> (ie (a) system emulation built with KVM support and (b) linux-user),
>> the set of targets which can be non-TLS is really really small
Il 01/07/2013 18:26, Peter Maydell ha scritto:
> > I think we should strive for one of these two:
> >
> > (1) all targets are TLS;
> >
> > (2) all targets are non-TLS if this is possible.
> >
> > Either maximizes the homogeneity across platforms.
>
> Since the two largest cases are both "cpu_singl
On 1 July 2013 17:21, Paolo Bonzini wrote:
> I think we should strive for one of these two:
>
> (1) all targets are TLS;
>
> (2) all targets are non-TLS if this is possible.
>
> Either maximizes the homogeneity across platforms.
Since the two largest cases are both "cpu_single_env must be TLS"
(i
Il 01/07/2013 13:05, Peter Maydell ha scritto:
>> > This patch does not make things worse than before. If anything, it's
>> > better because *more* targets have non-TLS semantics: namely non-KVM
>> > targets on Linux become non-TLS.
> That is making things worse! Non-TLS is the untested minority
>
On 1 July 2013 11:45, Paolo Bonzini wrote:
> Il 29/06/2013 12:55, Peter Maydell ha scritto:
>> We should just be consistent and always make this be
>> per-thread.
>
> If it's okay to make cpu_single_env accesses more expensive by a factor
> of 4 on TLS-deficient hosts (at least OpenBSD; do Darwin
Il 29/06/2013 12:55, Peter Maydell ha scritto:
> On 28 June 2013 19:26, Paolo Bonzini wrote:
>> The next patch will change qemu/tls.h to support more platforms, but at
>> some performance cost. Declare cpu_single_env directly instead of using
>> the tls.h abstractions.
>>
>> Signed-off-by: Paolo
Il 29/06/2013 01:53, Ed Maste ha scritto:
> On 28 June 2013 14:26, Paolo Bonzini wrote:
>>
>> +/* This is thread-local depending on __linux__ because:
>
> Is the comment perhaps unchanged from an earlier revision that used a
> different test? It seems odd to me to reference __linux__ here.
>
>>
On 28 June 2013 19:26, Paolo Bonzini wrote:
> The next patch will change qemu/tls.h to support more platforms, but at
> some performance cost. Declare cpu_single_env directly instead of using
> the tls.h abstractions.
>
> Signed-off-by: Paolo Bonzini
> ---
> exec.c | 10
On 28 June 2013 14:26, Paolo Bonzini wrote:
>
> +/* This is thread-local depending on __linux__ because:
Is the comment perhaps unchanged from an earlier revision that used a
different test? It seems odd to me to reference __linux__ here.
> + * - the only -user mode supporting multiple VCPU th
Paolo Bonzini writes:
> The next patch will change qemu/tls.h to support more platforms, but at
> some performance cost. Declare cpu_single_env directly instead of using
> the tls.h abstractions.
>
> Signed-off-by: Paolo Bonzini
Reviewed-by: Anthony Liguori
Regards,
Anthony Liguori
> ---
>
12 matches
Mail list logo
