[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CADnq5_OZMiK21niDFSUgptv2vtBj4vyyxBTafWnFwWuEV6VB7w@mail.gmail.com>
Date: Thu, 30 Jul 2020 00:31:44 -0400
From: Alex Deucher <alexdeucher@...il.com>
To: Logan Gunthorpe <logang@...tatee.com>
Cc: LKML <linux-kernel@...r.kernel.org>,
Linux PCI <linux-pci@...r.kernel.org>,
Bjorn Helgaas <bhelgaas@...gle.com>,
Christian König <christian.koenig@....com>,
Huang Rui <ray.huang@....com>
Subject: Re: [PATCH v2] PCI/P2PDMA: Allow P2PDMA on all AMD CPUs newer than
the Zen family
On Wed, Jul 29, 2020 at 7:18 PM Logan Gunthorpe <logang@...tatee.com> wrote:
>
> In order to avoid needing to add every new AMD CPU host bridge to the list
> every cycle, allow P2PDMA if the CPUs vendor is AMD and family is
> greater than 0x17 (Zen).
Might want to say "greater than or equal to" to clarify that all Zen
parts (Zen1, Zen2, etc.) support p2p. With that clarified,
Reviewed-by: Alex Deucher <alexander.deucher@....com>
>
> This should cut down a bunch of the churn adding to the list of allowed
> host bridges.
>
> Signed-off-by: Logan Gunthorpe <logang@...tatee.com>
> Cc: Bjorn Helgaas <bhelgaas@...gle.com>
> Cc: Alex Deucher <alexdeucher@...il.com>
> Cc: Christian König <christian.koenig@....com>
> Cc: Huang Rui <ray.huang@....com>
>
> ---
>
> Here's a reworked patch to enable P2PDMA on Zen2 (and in fact all
> subsequent Zen platforms).
>
> This should remove all the churn on the list for the AMD side. Still
> don't have a good solution for Intel.
>
> drivers/pci/p2pdma.c | 26 ++++++++++++++++++++------
> 1 file changed, 20 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/pci/p2pdma.c b/drivers/pci/p2pdma.c
> index e8e444eeb1cd..f1cab2c50595 100644
> --- a/drivers/pci/p2pdma.c
> +++ b/drivers/pci/p2pdma.c
> @@ -273,6 +273,24 @@ static void seq_buf_print_bus_devfn(struct seq_buf *buf, struct pci_dev *pdev)
> seq_buf_printf(buf, "%s;", pci_name(pdev));
> }
>
> +#ifdef CONFIG_X86
> +static bool cpu_supports_p2pdma(void)
> +{
> + struct cpuinfo_x86 *c = &cpu_data(0);
> +
> + /* Any AMD CPU who's family id is newer than Zen will support p2pdma */
> + if (c->x86_vendor == X86_VENDOR_AMD && c->x86 >= 0x17)
> + return true;
> +
> + return false;
> +}
> +#else
> +static bool cpu_supports_p2pdma(void)
> +{
> + return false;
> +}
> +#endif
> +
> static const struct pci_p2pdma_whitelist_entry {
> unsigned short vendor;
> unsigned short device;
> @@ -280,11 +298,6 @@ static const struct pci_p2pdma_whitelist_entry {
> REQ_SAME_HOST_BRIDGE = 1 << 0,
> } flags;
> } pci_p2pdma_whitelist[] = {
> - /* AMD ZEN */
> - {PCI_VENDOR_ID_AMD, 0x1450, 0},
> - {PCI_VENDOR_ID_AMD, 0x15d0, 0},
> - {PCI_VENDOR_ID_AMD, 0x1630, 0},
> -
> /* Intel Xeon E5/Core i7 */
> {PCI_VENDOR_ID_INTEL, 0x3c00, REQ_SAME_HOST_BRIDGE},
> {PCI_VENDOR_ID_INTEL, 0x3c01, REQ_SAME_HOST_BRIDGE},
> @@ -473,7 +486,8 @@ upstream_bridge_distance(struct pci_dev *provider, struct pci_dev *client,
> acs_redirects, acs_list);
>
> if (map_type == PCI_P2PDMA_MAP_THRU_HOST_BRIDGE) {
> - if (!host_bridge_whitelist(provider, client))
> + if (!cpu_supports_p2pdma() &&
> + !host_bridge_whitelist(provider, client))
> map_type = PCI_P2PDMA_MAP_NOT_SUPPORTED;
> }
>
>
> base-commit: 92ed301919932f777713b9172e525674157e983d
> --
> 2.20.1
Powered by blists - more mailing lists