Message-ID: <d28e6848-8fa4-c2c4-1140-7da5bcbe1287@linux.ibm.com>
Date:   Wed, 8 Dec 2021 11:20:03 -0500
From:   Matthew Rosato <mjrosato@...ux.ibm.com>
To:     Niklas Schnelle <schnelle@...ux.ibm.com>,
        Christian Borntraeger <borntraeger@...ux.ibm.com>,
        linux-s390@...r.kernel.org
Cc:     alex.williamson@...hat.com, cohuck@...hat.com,
        farman@...ux.ibm.com, pmorel@...ux.ibm.com, hca@...ux.ibm.com,
        gor@...ux.ibm.com, gerald.schaefer@...ux.ibm.com,
        agordeev@...ux.ibm.com, frankja@...ux.ibm.com, david@...hat.com,
        imbrenda@...ux.ibm.com, vneethv@...ux.ibm.com,
        oberpar@...ux.ibm.com, freude@...ux.ibm.com, thuth@...hat.com,
        pasic@...ux.ibm.com, kvm@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH 07/32] s390/pci: externalize the SIC operation controls
 and routine

On 12/8/21 10:59 AM, Niklas Schnelle wrote:
> On Wed, 2021-12-08 at 10:33 -0500, Matthew Rosato wrote:
>> On 12/8/21 8:53 AM, Niklas Schnelle wrote:
>>> On Wed, 2021-12-08 at 14:09 +0100, Christian Borntraeger wrote:
>>>> Am 07.12.21 um 21:57 schrieb Matthew Rosato:
>>>>> A subsequent patch will be issuing SIC from KVM -- export the necessary
>>>>> routine and make the operation control definitions available from a header.
>>>>> Because the routine will now be exported, let's swap the purpose of
>>>>> zpci_set_irq_ctrl and __zpci_set_irq_ctrl, leaving the latter as a static
>>>>> within pci_irq.c only for SIC calls that don't specify an iib.
>>>>
>>>> Maybe it would be simpler to export the __ version instead of renaming everything.
>>>> Whatever Niklas prefers.
>>>
>>> See below I think it's just not worth it having both variants at all.
>>>
>>>>> Signed-off-by: Matthew Rosato <mjrosato@...ux.ibm.com>
>>>>> ---
>>>>>     arch/s390/include/asm/pci_insn.h | 17 +++++++++--------
>>>>>     arch/s390/pci/pci_insn.c         |  3 ++-
>>>>>     arch/s390/pci/pci_irq.c          | 28 ++++++++++++++--------------
>>>>>     3 files changed, 25 insertions(+), 23 deletions(-)
>>>>>
>>>>> diff --git a/arch/s390/include/asm/pci_insn.h b/arch/s390/include/asm/pci_insn.h
>>>>> index 61cf9531f68f..5331082fa516 100644
>>>>> --- a/arch/s390/include/asm/pci_insn.h
>>>>> +++ b/arch/s390/include/asm/pci_insn.h
>>>>> @@ -98,6 +98,14 @@ struct zpci_fib {
>>>>>     	u32 gd;
>>>>>     } __packed __aligned(8);
>>>>>     
>>>>> +/* Set Interruption Controls Operation Controls  */
>>>>> +#define	SIC_IRQ_MODE_ALL		0
>>>>> +#define	SIC_IRQ_MODE_SINGLE		1
>>>>> +#define	SIC_IRQ_MODE_DIRECT		4
>>>>> +#define	SIC_IRQ_MODE_D_ALL		16
>>>>> +#define	SIC_IRQ_MODE_D_SINGLE		17
>>>>> +#define	SIC_IRQ_MODE_SET_CPU		18
>>>>> +
>>>>>     /* directed interruption information block */
>>>>>     struct zpci_diib {
>>>>>     	u32 : 1;
>>>>> @@ -134,13 +142,6 @@ int __zpci_store(u64 data, u64 req, u64 offset);
>>>>>     int zpci_store(const volatile void __iomem *addr, u64 data, unsigned long len);
>>>>>     int __zpci_store_block(const u64 *data, u64 req, u64 offset);
>>>>>     void zpci_barrier(void);
>>>>> -int __zpci_set_irq_ctrl(u16 ctl, u8 isc, union zpci_sic_iib *iib);
>>>>> -
>>>>> -static inline int zpci_set_irq_ctrl(u16 ctl, u8 isc)
>>>>> -{
>>>>> -	union zpci_sic_iib iib = {{0}};
>>>>> -
>>>>> -	return __zpci_set_irq_ctrl(ctl, isc, &iib);
>>>>> -}
>>>>> +int zpci_set_irq_ctrl(u16 ctl, u8 isc, union zpci_sic_iib *iib);
>>>
>>> Since the __zpci_set_irq_ctrl() was already non static/inline the above
>>> inline to non-inline change shouldn't make a performance difference.
>>>
>>> Looking at this makes me wonder though. Wouldn't it make sense to just
>>> have the zpci_set_irq_ctrl() function inline in the header. Its body is
>>> a single instruction inline asm plus a test_facility(). The latter by
>>> the way I think also looks rather out of place there considering we
>>> call zpci_set_irq_ctrl() in the interrupt handler and facilities can't
>>> go away so it's pretty silly to check for it on every single
>>> interrupt.. unless I'm totally missing something.
>>
>> This test_facility isn't new to this patch
> 
> Yeah I got that part, your patch just made me look.
> 
>> , it was added via
>>
>> commit 48070c73058be6de9c0d754d441ed7092dfc8f12
>> Author: Christian Borntraeger <borntraeger@...ibm.com>
>> Date:   Mon Oct 30 14:38:58 2017 +0100
>>
>>       s390/pci: do not require AIS facility
>>
>> It looks like in the past, we would not even initialize zpci at all if
>> AIS wasn't available.  With this, we initialize PCI but only do the SIC
>> when we have AIS, which makes sense.
> 
> Ah yes I guess that is the something I was missing. I was wondering why
> that wasn't just tested for during init.
> 
>>
>> So for this patch, the sane thing to do is probably just keep the
>> test_facility() in place and move to header, inline.
> 
> Yes sounds good.
> 
>>
>> Maybe there's a subsequent optimization to be made (setup a static key
>> like have_mio vs doing test_facility all the time?)
> 
> Yeah, looking again more closely at test_facilities() it's probably not
> that expensive either I'll do some tests. Maybe we can also just add a
> comment and a normal unlikely() macro since with this series KVM would
> also support AIS, correct?
AIS was already being set as a KVM facility / allowed as QEMU capability 
before this series, however there was a period of time where QEMU was 
disabling it (disabled in QEMU 3f2d07b3b01e, enabled again in QEMU 
a5c8617af691) which I suspect was the impetus for this kernel change; 
this means that there are older machines that won't have it, but moving 
forward we should be OK in the standard case.  Of course the kernel 
should still be able to tolerate the case where AIS is unavailable (old 
machine, intentionally forced off, etc), so maybe the unlikely indeed 
makes the most sense.

As far as a comment for the unlikely I could add something like 'some 
virtualized environments may have disabled the AIS facility'?

Powered by blists - more mailing lists