Subject: Re: [PATCH] xen/x86: actually allocate legacy interrupts on PV guests
To: Stefano Stabellini <sstabellini@kernel.org>,
        Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
References: <alpine.DEB.2.10.1604201413070.23110@sstabellini-ThinkPad-X260>
 <57189880.6030102@suse.com>
 <alpine.DEB.2.10.1604211023130.23110@sstabellini-ThinkPad-X260>
Cc: boris.ostrovsky@oracle.com, linux-kernel@vger.kernel.org,
        david.vrabel@citrix.com, xen-devel@lists.xenproject.org
From: Juergen Gross <jgross@suse.com>
Message-ID: <572047D3.1000002@suse.com>
Date: Wed, 27 Apr 2016 07:02:11 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101
 Thunderbird/38.7.0
MIME-Version: 1.0
In-Reply-To: <alpine.DEB.2.10.1604211023130.23110@sstabellini-ThinkPad-X260>
Content-Type: text/plain; charset=windows-1252
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2386
Lines: 68

On 21/04/16 11:30, Stefano Stabellini wrote:
> On Thu, 21 Apr 2016, Juergen Gross wrote:
>> On 20/04/16 15:15, Stefano Stabellini wrote:
>>> b4ff8389ed14 is incomplete: relies on nr_legacy_irqs() to get the number
>>> of legacy interrupts when actually nr_legacy_irqs() returns 0 after
>>> probe_8259A(). Use NR_IRQS_LEGACY instead.
>>
>> Would you mind describing the resulting problem?
> 
> This is a good question. The symptom is:
> 
> ata_piix: probe of 0000:00:01.1 failed with error -22
> 
> 
>> With this commit message I'm absolutely not capable to decide whether
>> e.g. the other use of nr_legacy_irqs() in pci_xen_initial_domain() is
>> correct or not.
> 
> I looked at it but I couldn't really test that code because if I try to
> change the number of ioapics in the system using the "noapic" command
> line option (which actually changes the number if ioapics, not lapics),
> I get an error from Linux saying that noapic is not supported when
> running on Xen.
> 
> In my opinion having nr_legacy_irqs() calls in Xen code, which returns
> 0, is like playing with fire. I think it would be safer/saner to replace
> them all with NR_IRQS_LEGACY, simply because reading the code one would
> not expect that all those loops don't actually have any iterations.

I'm quite sure you should change both uses of nr_legacy_irqs() in
pci_xen_initial_domain().

Looking at xen_pcifront_enable_irq() I'm not really sure what is the
correct thing to do.

Adding Konrad as he might have a better insight.


Juergen

> 
> However I didn't make the change because I couldn't test it properly.
> 
> 
>>> Signed-off-by: Stefano Stabellini <sstabellini@kernel.org>
>>>
>>> diff --git a/arch/x86/pci/xen.c b/arch/x86/pci/xen.c
>>> index beac4df..349b8ce 100644
>>> --- a/arch/x86/pci/xen.c
>>> +++ b/arch/x86/pci/xen.c
>>> @@ -491,8 +491,11 @@ int __init pci_xen_initial_domain(void)
>>>  #endif
>>>  	__acpi_register_gsi = acpi_register_gsi_xen;
>>>  	__acpi_unregister_gsi = NULL;
>>> -	/* Pre-allocate legacy irqs */
>>> -	for (irq = 0; irq < nr_legacy_irqs(); irq++) {
>>> +	/*
>>> +	 * Pre-allocate the legacy IRQs.  Use NR_LEGACY_IRQS here
>>> +	 * because we don't have a PIC and thus nr_legacy_irqs() is zero.
>>> +	 */
>>> +	for (irq = 0; irq < NR_IRQS_LEGACY; irq++) {
>>>  		int trigger, polarity;
>>>  
>>>  		if (acpi_get_override_irq(irq, &trigger, &polarity) == -1)
>>>
>>
> 
>