Re: [PATCH v8 2/5] xen/arm: Enable the existing x86 virtual PCI support for ARM

[Jan Beulich <jbeulich@xxxxxxxx>]

On 18.10.2021 16:37, Oleksandr Andrushchenko wrote:
> 
> 
> On 18.10.21 17:19, Jan Beulich wrote:
>> On 18.10.2021 16:07, Oleksandr Andrushchenko wrote:
>>> On 18.10.21 13:29, Jan Beulich wrote:
>>>> On 18.10.2021 12:11, Bertrand Marquis wrote:
>>>>>> On 18 Oct 2021, at 08:47, Jan Beulich <jbeulich@xxxxxxxx> wrote:
>>>>>> On 15.10.2021 18:51, Bertrand Marquis wrote:
>>>>>>> --- /dev/null
>>>>>>> +++ b/xen/arch/arm/vpci.c
>>>>>>> @@ -0,0 +1,77 @@
>>>>>>> +/*
>>>>>>> + * xen/arch/arm/vpci.c
>>>>>>> + *
>>>>>>> + * This program is free software; you can redistribute it and/or modify
>>>>>>> + * it under the terms of the GNU General Public License as published by
>>>>>>> + * the Free Software Foundation; either version 2 of the License, or
>>>>>>> + * (at your option) any later version.
>>>>>>> + *
>>>>>>> + * This program is distributed in the hope that it will be useful,
>>>>>>> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
>>>>>>> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
>>>>>>> + * GNU General Public License for more details.
>>>>>>> + */
>>>>>>> +#include <xen/sched.h>
>>>>>>> +#include <xen/vpci.h>
>>>>>>> +
>>>>>>> +#include <asm/mmio.h>
>>>>>>> +
>>>>>>> +static int vpci_mmio_read(struct vcpu *v, mmio_info_t *info,
>>>>>>> +                          register_t *r, void *p)
>>>>>>> +{
>>>>>>> +    pci_sbdf_t sbdf;
>>>>>>> +    /* data is needed to prevent a pointer cast on 32bit */
>>>>>>> +    unsigned long data;
>>>>>>> +
>>>>>>> +    /* We ignore segment part and always handle segment 0 */
>>>>>>> +    sbdf.sbdf = VPCI_ECAM_BDF(info->gpa);
>>>>>>> +
>>>>>>> +    if ( vpci_ecam_read(sbdf, ECAM_REG_OFFSET(info->gpa),
>>>>>>> +                        1U << info->dabt.size, &data) )
>>>>>>> +    {
>>>>>> Here it is quite clear that the SBDF you pass into vpci_ecam_read() is
>>>>>> the virtual one. The function then calls vpci_read(), which in turn
>>>>>> will call vpci_read_hw() in a number of situations (first and foremost
>>>>>> whenever pci_get_pdev_by_domain() returns NULL). That function as well
>>>>>> as pci_get_pdev_by_domain() use the passed in SBDF as if it was a
>>>>>> physical one; I'm unable to spot any translation. Yet I do recall
>>>>>> seeing assignment of a virtual device and function number somewhere
>>>>>> (perhaps another of the related series), so the model also doesn't
>>>>>> look to assume 1:1 mapping of SBDF.
>>>>> This question was answered by Oleksandr I think.
>>>> Yes; I continue to be sure though that I saw devfn allocation logic in
>>>> one of the many patches that are related here. And I'm relatively sure
>>>> that there no adjustment to logic here was made (but since it's hard
>>>> to pick the right patch out of the huge pile without knowing its title,
>>>> I can't reasonably go check).
>>> Offtop: I was somehow dropped from the Cc..
>>>
>>> Please see:
>>>
>>> [PATCH v3 10/11] vpci: Add initial support for virtual PCI bus topology
>>> [PATCH v3 11/11] xen/arm: Translate virtual PCI bus topology for guests
>>>
>>> [1] 
>>> https://urldefense.com/v3/__https://patchwork.kernel.org/project/xen-devel/list/?series=555481__;!!GF_29dbcQIUBPA!mM8A39p8nk4UMK3YeKMMW9ua9BHj1UOWzaQcyx7G46YPdudxMpD3huqZfih0Uc8S-GyWXD_mPg$
>>>  [patchwork[.]kernel[.]org]
>> Ah yes, this way I can find them in my mailbox. And indeed - no translation
>> from virtual to physical SBDF on the config space read/write paths afaics.
> There are translations for both read and write [2] such as:
> 
> +    /*
> +     * For the passed through devices we need to map their virtual SBDF
> +     * to the physical PCI device being passed through.
> +     */
> +    if ( priv->is_virt_ecam && !pci_translate_virtual_device(v->domain, 
> &sbdf) )
> +            return 1;
> +

Oh, that's before you even call vpci_read(). I would have expected this
to live in common vPCI code ...

Jan