From: Bertrand Marquis <Bertrand.Marquis@arm.com>
To: Jan Beulich <jbeulich@suse.com>
Cc: "Rahul Singh" <Rahul.Singh@arm.com>,
"Andre Przywara" <Andre.Przywara@arm.com>,
"Stefano Stabellini" <sstabellini@kernel.org>,
"Julien Grall" <julien@xen.org>,
"Volodymyr Babchuk" <Volodymyr_Babchuk@epam.com>,
"Andrew Cooper" <andrew.cooper3@citrix.com>,
"George Dunlap" <george.dunlap@citrix.com>,
"Ian Jackson" <iwj@xenproject.org>, "Wei Liu" <wl@xen.org>,
"Paul Durrant" <paul@xen.org>,
"Roger Pau Monné" <roger.pau@citrix.com>,
"xen-devel@lists.xenproject.org" <xen-devel@lists.xenproject.org>
Subject: Re: [PATCH v5 08/11] xen/arm: Enable the existing x86 virtual PCI support for ARM.
Date: Mon, 11 Oct 2021 14:52:43 +0000 [thread overview]
Message-ID: <CC5D73DE-74EB-476A-965E-4467A0010439@arm.com> (raw)
In-Reply-To: <59c9e102-c710-64d3-2a1a-cc8dcbcceead@suse.com>
Hi Jan,
> On 11 Oct 2021, at 15:10, Jan Beulich <jbeulich@suse.com> wrote:
>
> On 11.10.2021 15:34, Bertrand Marquis wrote:
>>> On 11 Oct 2021, at 14:09, Jan Beulich <jbeulich@suse.com> wrote:
>>> On 11.10.2021 14:41, Bertrand Marquis wrote:
>>>>> On 7 Oct 2021, at 14:43, Jan Beulich <jbeulich@suse.com> wrote:
>>>>> On 06.10.2021 19:40, Rahul Singh wrote:
>>>>>> --- /dev/null
>>>>>> +++ b/xen/arch/arm/vpci.c
>>>>>> @@ -0,0 +1,102 @@
>>>>>> +/*
>>>>>> + * 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 <asm/mmio.h>
>>>>>> +
>>>>>> +#define REGISTER_OFFSET(addr) ( (addr) & 0x00000fff)
>>>>>
>>>>> Nit: Stray blank (like you had in an earlier version for MMCFG_BDF()).
>>>>> Also isn't this effectively part of the public interface (along with
>>>>> MMCFG_BDF()), alongside GUEST_VPCI_ECAM_{BASE,SIZE}?
>>>>
>>>> I will move that in the next version to xen/pci.h and rename itMMCFG_REG_OFFSET.
>>>> Would that be ok ?
>>>
>>> That would be okay and make sense when put next to MMCFG_BDF(), but
>>> it would not address my comment: That still wouldn't be the public
>>> interface. Otoh you only mimic hardware behavior, so perhaps the
>>> splitting of the address isn't as relevant to put there as would be
>>> GUEST_VPCI_ECAM_{BASE,SIZE}.
>>
>> Ok now I get what you wanted.
>>
>> You would actually like both MMCFG_BDF and MMCFG_REG_OFFSET to
>> be moved to arch-arm.h.
>>
>> Then I am not quite sure to follow why.
>> Those are not macros coming out of a way we have to define this but from
>> how it works in standard PCI.
>> The base and size are needed to know where the PCI bus will be.
>>
>> So why should those be needed in public headers ?
>
> Well, see my "Otoh ..." in the earlier reply. Keeping the two
> address splitting macros out of there is okay.
Ok.
>
>>>>>> --- a/xen/drivers/passthrough/pci.c
>>>>>> +++ b/xen/drivers/passthrough/pci.c
>>>>>> @@ -766,6 +766,24 @@ int pci_add_device(u16 seg, u8 bus, u8 devfn,
>>>>>> else
>>>>>> iommu_enable_device(pdev);
>>>>>
>>>>> Please note the context above; ...
>>>>>
>>>>>> +#ifdef CONFIG_ARM
>>>>>> + /*
>>>>>> + * On ARM PCI devices discovery will be done by Dom0. Add vpci handler when
>>>>>> + * Dom0 inform XEN to add the PCI devices in XEN.
>>>>>> + */
>>>>>> + ret = vpci_add_handlers(pdev);
>>>>>> + if ( ret )
>>>>>> + {
>>>>>> + printk(XENLOG_ERR "setup of vPCI failed: %d\n", ret);
>>>>>> + pci_cleanup_msi(pdev);
>>>>>> + ret = iommu_remove_device(pdev);
>>>>>> + if ( pdev->domain )
>>>>>> + list_del(&pdev->domain_list);
>>>>>> + free_pdev(pseg, pdev);
>>>>>
>>>>> ... you unconditionally undo the if() part of the earlier conditional;
>>>>> is there any guarantee that the other path can't ever be taken, now
>>>>> and forever? If it can't be taken now (which I think is the case as
>>>>> long as Dom0 wouldn't report the same device twice), then at least some
>>>>> precaution wants taking. Maybe moving your addition into that if()
>>>>> could be an option.
>>>>>
>>>>> Furthermore I continue to wonder whether this ordering is indeed
>>>>> preferable over doing software setup before hardware arrangements. This
>>>>> would address the above issue as well as long as vpci_add_handlers() is
>>>>> idempotent. And it would likely simplify error cleanup.
>>>>
>>>> I agree with you so I will move this code block before iommu part.
>>>>
>>>> But digging deeper into this, I would have 2 questions:
>>>>
>>>> - msi_cleanup was done there after a request from Stefano, but is not
>>>> done in case or iommu error, is there an issue to solve here ?
>>>
>>> Maybe, but I'm not sure. This very much depends on what a domain
>>> could in principle do with a partly set-up device. Plus let's
>>> not forget that we're talking of only Dom0 here (for now at least,
>>> i.e. not considering the dom0less case).
>>>
>>> But I'd also like to further defer to Stefano.
>>
>> Ok, I must admit I do not really see at that stage why doing an MSI cleanup
>> could be needed so I will wait for Stefano to know if I need to keep this when
>> moving the block up (at the end it is theoretical right now as this is empty).
>>
>>>
>>>> Same could also go for the free_pdev ?
>>>
>>> I think it's wrong to free_pdev() here. We want to internally keep
>>> record of the device, even if the device ends up unusable. The only
>>> place where free_pdev() ought to be called is imo pci_remove_device().
>>
>> Ok.
>>
>>>
>>>> - cleanup code was exactly the same as pci_remove_device code.
>>>> Should the question about the path also be checked there ?
>>>
>>> I'm sorry, but I'm afraid I don't see what "the path" refers to
>>> here. You can't mean the conditional in pci_add_device() selecting
>>> between iommu_add_device() and iommu_enable_device(), as "remove"
>>> can only mean "remove", never "disable".
>>
>> I will try to explain: when we just enable we do not add an entry in the list but
>> we still remove an entry from the list every time (as the condition becomes
>> always true because pdev->domain is at the end always set)
>
> Well, that anomaly is what I did point out in my review remarks to
> Rahul. We shouldn't remove an entry from the list if we didn't add
> one. But quite likely, if we don't free the pdev, we shouldn't be
> removing the list entry in either case.
This problem will not exist anymore when I will move the code up but I will add to adapt the error case in iommu to also remove the vpci handlers.
To be coherent I will do the same in the pci_remove_device code.
I will do all those in the v6 of the serie.
Thanks a lot for the answers.
Cheers
Bertrand
>
> Jan
>
next prev parent reply other threads:[~2021-10-11 14:53 UTC|newest]
Thread overview: 190+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-10-06 17:40 [PATCH v5 00/11] PCI devices passthrough on Arm Rahul Singh
2021-10-06 17:40 ` [PATCH v5 01/11] xen/arm: xc_domain_ioport_permission(..) not supported on ARM Rahul Singh
2021-10-11 11:47 ` Roger Pau Monné
2021-10-11 12:11 ` Bertrand Marquis
2021-10-11 13:20 ` Roger Pau Monné
2021-10-11 13:40 ` Bertrand Marquis
2021-10-11 13:57 ` Roger Pau Monné
2021-10-11 14:16 ` Bertrand Marquis
2021-10-11 16:32 ` Roger Pau Monné
2021-10-11 17:11 ` Bertrand Marquis
2021-10-12 8:29 ` Jan Beulich
2021-10-12 8:41 ` Bertrand Marquis
2021-10-12 9:32 ` Jan Beulich
2021-10-12 9:38 ` Oleksandr Andrushchenko
2021-10-12 10:01 ` Jan Beulich
2021-10-12 10:06 ` Oleksandr Andrushchenko
2021-10-12 10:20 ` Jan Beulich
2021-10-12 10:41 ` Bertrand Marquis
2021-10-12 10:44 ` Jan Beulich
2021-10-12 14:53 ` Ian Jackson
2021-10-12 16:15 ` Bertrand Marquis
2021-10-12 16:29 ` Ian Jackson
2021-10-12 20:42 ` Stefano Stabellini
2021-10-13 8:07 ` Roger Pau Monné
2021-10-13 11:52 ` Ian Jackson
2021-10-13 8:02 ` Roger Pau Monné
2021-10-13 12:02 ` Ian Jackson
2021-10-12 9:40 ` Bertrand Marquis
2021-10-12 10:03 ` Jan Beulich
2021-10-11 14:16 ` Oleksandr Andrushchenko
2021-10-06 17:40 ` [PATCH v5 02/11] xen/arm: Add PHYSDEVOP_pci_device_(*add/remove) support for ARM Rahul Singh
2021-10-07 0:05 ` Stefano Stabellini
2021-10-07 12:58 ` Jan Beulich
2021-10-21 9:28 ` xen/arm: Missing appropriate locking for the IOMMU (WAS Re: [PATCH v5 02/11] xen/arm: Add PHYSDEVOP_pci_device_(*add/remove) support for ARM) Julien Grall
2021-10-21 13:15 ` Bertrand Marquis
2021-10-21 13:47 ` Julien Grall
2021-10-21 13:52 ` Bertrand Marquis
2021-10-06 17:40 ` [PATCH v5 03/11] xen/arm: Add cmdline boot option "pci-passthrough = <boolean>" Rahul Singh
2021-10-07 8:27 ` Jan Beulich
2021-10-07 8:32 ` Rahul Singh
2021-10-07 12:59 ` Jan Beulich
2021-10-06 17:40 ` [PATCH v5 04/11] xen/arm: PCI host bridge discovery within XEN on ARM Rahul Singh
2021-10-06 17:40 ` [PATCH v5 05/11] xen/arm: Add support for Xilinx ZynqMP PCI host controller Rahul Singh
2021-10-06 17:40 ` [PATCH v5 06/11] xen/arm: Implement pci access functions Rahul Singh
2021-10-06 17:40 ` [PATCH v5 07/11] xen/domctl: Introduce XEN_DOMCTL_CDF_vpci flag Rahul Singh
2021-10-07 13:08 ` Jan Beulich
2021-10-08 18:06 ` Andrew Cooper
2021-10-08 21:12 ` Julien Grall
2021-10-08 21:46 ` Stefano Stabellini
2021-10-11 9:24 ` Julien Grall
2021-10-11 11:29 ` Michal Orzel
2021-10-11 11:35 ` Jan Beulich
2021-10-11 13:17 ` Roger Pau Monné
2021-10-11 9:48 ` Ian Jackson
2021-10-11 9:27 ` Roger Pau Monné
2021-10-11 12:06 ` Michal Orzel
2021-10-12 10:38 ` Michal Orzel
2021-10-13 8:30 ` Roger Pau Monné
2021-10-13 9:36 ` Bertrand Marquis
2021-10-13 10:56 ` Roger Pau Monné
2021-10-13 12:11 ` Bertrand Marquis
2021-10-13 12:57 ` Jan Beulich
2021-10-13 20:41 ` Stefano Stabellini
2021-10-14 6:23 ` Jan Beulich
2021-10-14 7:53 ` Bertrand Marquis
2021-10-13 14:28 ` Roger Pau Monné
2021-10-13 20:53 ` Stefano Stabellini
2021-10-13 23:21 ` Stefano Stabellini
2021-10-12 21:48 ` Stefano Stabellini
2021-10-13 6:18 ` Jan Beulich
2021-10-13 7:11 ` Michal Orzel
2021-10-06 17:40 ` [PATCH v5 08/11] xen/arm: Enable the existing x86 virtual PCI support for ARM Rahul Singh
2021-10-07 13:43 ` Jan Beulich
2021-10-11 12:41 ` Bertrand Marquis
2021-10-11 13:09 ` Jan Beulich
2021-10-11 13:34 ` Bertrand Marquis
2021-10-11 14:10 ` Jan Beulich
2021-10-11 14:52 ` Bertrand Marquis [this message]
2021-10-11 18:18 ` Stefano Stabellini
2021-10-12 8:04 ` Jan Beulich
2021-10-12 21:37 ` Stefano Stabellini
2021-10-13 6:10 ` Jan Beulich
2021-10-13 10:02 ` Bertrand Marquis
2021-10-13 12:21 ` Jan Beulich
2021-10-12 15:04 ` Julien Grall
2021-10-12 16:12 ` Bertrand Marquis
2021-10-12 16:20 ` Julien Grall
2021-10-12 17:50 ` Bertrand Marquis
2021-10-11 10:51 ` Roger Pau Monné
2021-10-11 16:12 ` Bertrand Marquis
2021-10-11 16:20 ` Oleksandr Andrushchenko
2021-10-11 16:43 ` Roger Pau Monné
2021-10-11 17:15 ` Bertrand Marquis
2021-10-11 18:30 ` Oleksandr Andrushchenko
2021-10-11 19:27 ` Stefano Stabellini
2021-10-12 5:34 ` Oleksandr Andrushchenko
2021-10-12 7:44 ` Bertrand Marquis
2021-10-12 14:32 ` Julien Grall
2021-10-12 14:34 ` Bertrand Marquis
2021-10-13 8:45 ` Roger Pau Monné
2021-10-13 9:48 ` Bertrand Marquis
2021-10-13 10:33 ` Roger Pau Monné
2021-10-13 13:00 ` Jan Beulich
2021-10-13 14:51 ` Oleksandr Andrushchenko
2021-10-13 15:15 ` Jan Beulich
2021-10-13 19:27 ` Stefano Stabellini
2021-10-14 6:33 ` Jan Beulich
2021-10-14 7:53 ` Bertrand Marquis
2021-10-14 9:03 ` Bertrand Marquis
2021-10-14 9:24 ` Jan Beulich
2021-10-06 17:40 ` [PATCH v5 09/11] xen/arm: Transitional change to build HAS_VPCI on ARM Rahul Singh
2021-10-11 11:43 ` Roger Pau Monné
2021-10-11 12:15 ` Bertrand Marquis
2021-10-12 1:32 ` Stefano Stabellini
2021-10-06 17:40 ` [PATCH v5 10/11] arm/libxl: Emulated PCI device tree node in libxl Rahul Singh
2021-10-06 18:01 ` Julien Grall
2021-10-07 0:26 ` Stefano Stabellini
2021-10-07 15:31 ` Rahul Singh
2021-10-07 10:53 ` Ian Jackson
2021-10-07 15:29 ` Rahul Singh
2021-10-07 16:11 ` Ian Jackson
2021-10-11 13:46 ` Roger Pau Monné
2021-10-14 17:16 ` Bertrand Marquis
2021-10-14 14:49 ` [PATCH v6 0/3] PCI devices passthrough on Arm Bertrand Marquis
2021-10-14 14:49 ` [PATCH v6 1/3] xen/vpci: Move ecam access functions to common code Bertrand Marquis
2021-10-14 16:06 ` Jan Beulich
2021-10-14 17:09 ` Bertrand Marquis
2021-10-15 6:29 ` Jan Beulich
2021-10-15 7:37 ` Bertrand Marquis
2021-10-15 8:13 ` Jan Beulich
2021-10-15 8:20 ` Bertrand Marquis
2021-10-15 8:24 ` Jan Beulich
2021-10-15 9:49 ` Roger Pau Monné
2021-10-14 23:47 ` Stefano Stabellini
2021-10-15 7:44 ` Roger Pau Monné
2021-10-15 7:53 ` Bertrand Marquis
2021-10-15 9:53 ` Roger Pau Monné
2021-10-15 10:12 ` Bertrand Marquis
2021-10-15 10:14 ` Jan Beulich
2021-10-14 14:49 ` [PATCH v6 2/3] xen/arm: Enable the existing x86 virtual PCI support for ARM Bertrand Marquis
2021-10-14 23:49 ` Stefano Stabellini
2021-10-15 6:40 ` Jan Beulich
2021-10-15 9:59 ` Ian Jackson
2021-10-15 10:10 ` Bertrand Marquis
2021-10-15 8:00 ` Jan Beulich
2021-10-15 10:09 ` Bertrand Marquis
2021-10-15 10:14 ` Ian Jackson
2021-10-15 10:18 ` Jan Beulich
2021-10-15 11:35 ` Roger Pau Monné
2021-10-15 12:13 ` Bertrand Marquis
2021-10-15 12:18 ` Jan Beulich
2021-10-15 12:28 ` Bertrand Marquis
2021-10-15 13:00 ` Jan Beulich
2021-10-15 13:10 ` Bertrand Marquis
2021-10-15 10:38 ` Jan Beulich
2021-10-15 8:32 ` Roger Pau Monné
2021-10-15 8:42 ` Michal Orzel
2021-10-15 9:52 ` Bertrand Marquis
2021-10-15 10:13 ` Luca Fancellu
2021-10-15 10:17 ` Bertrand Marquis
2021-10-15 10:19 ` Roger Pau Monné
2021-10-15 10:31 ` Bertrand Marquis
2021-10-15 10:24 ` Jan Beulich
2021-10-15 10:33 ` Bertrand Marquis
2021-10-15 10:41 ` Jan Beulich
2021-10-15 10:48 ` Bertrand Marquis
2021-10-15 10:51 ` Jan Beulich
2021-10-15 11:08 ` Bertrand Marquis
2021-10-15 13:47 ` Roger Pau Monné
2021-10-15 14:00 ` Luca Fancellu
2021-10-15 14:32 ` Roger Pau Monné
2021-10-14 14:49 ` [PATCH v6 3/3] arm/libxl: Emulated PCI device tree node in libxl Bertrand Marquis
2021-10-14 17:54 ` [PATCH v6 3/3] arm/libxl: Emulated PCI device tree node in libxl [and 1 more messages] Ian Jackson
2021-10-14 23:50 ` Stefano Stabellini
2021-10-15 7:28 ` Julien Grall
2021-10-15 7:41 ` Michal Orzel
2021-10-15 9:01 ` Julien Grall
2021-10-15 10:02 ` Ian Jackson
2021-10-15 10:58 ` Michal Orzel
2021-10-15 11:04 ` Michal Orzel
2021-10-15 11:46 ` Ian Jackson
2021-10-15 11:53 ` Michal Orzel
2021-10-15 12:10 ` Julien Grall
2021-10-15 12:14 ` Ian Jackson
2021-10-15 12:13 ` Ian Jackson
2021-10-12 15:03 ` [PATCH v5 10/11] arm/libxl: Emulated PCI device tree node in libxl Ian Jackson
2021-10-06 17:40 ` [PATCH v5 11/11] xen/arm: Add linux,pci-domain property for hwdom if not available Rahul Singh
2021-10-13 20:54 ` Stefano Stabellini
2021-10-07 19:54 ` [PATCH v5 00/11] PCI devices passthrough on Arm Stefano Stabellini
2021-10-07 21:29 ` Rahul Singh
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CC5D73DE-74EB-476A-965E-4467A0010439@arm.com \
--to=bertrand.marquis@arm.com \
--cc=Andre.Przywara@arm.com \
--cc=Rahul.Singh@arm.com \
--cc=Volodymyr_Babchuk@epam.com \
--cc=andrew.cooper3@citrix.com \
--cc=george.dunlap@citrix.com \
--cc=iwj@xenproject.org \
--cc=jbeulich@suse.com \
--cc=julien@xen.org \
--cc=paul@xen.org \
--cc=roger.pau@citrix.com \
--cc=sstabellini@kernel.org \
--cc=wl@xen.org \
--cc=xen-devel@lists.xenproject.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).