2009-12-15 12:25:23

by Csaba Henk

Subject: [PATCH 1/1] PCI: Handle case when no pci device can provide cache line size hint

Prior to this patch, if pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, ...)
returns 0 for all dev, pci_cache_line_size ends up set to zero
(instead of pci_dfl_cache_line_size).

This patch ensures the pci_cache_line_size = pci_dfl_cache_line_size
setting in the above scenario.

This happens in case of a kvm-88 guest (where, consequently, the rtl8139
NIC failed to initialize).

Signed-off-by: Csaba Henk <[email protected]>
---
drivers/pci/quirks.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
index 7cfa7c3..f70f4e2 100644
--- a/drivers/pci/quirks.c
+++ b/drivers/pci/quirks.c
@@ -2629,7 +2629,7 @@ static int __init pci_apply_final_quirks(void)
if (!pci_cache_line_size) {
printk(KERN_DEBUG "PCI: CLS %u bytes, default %u\n",
cls << 2, pci_dfl_cache_line_size << 2);
- pci_cache_line_size = cls;
+ pci_cache_line_size = cls ? cls : pci_dfl_cache_line_size;
}

return 0;
--
1.6.5.6

2009-12-16 01:46:48

by Tejun Heo

Subject: Re: [PATCH 1/1] PCI: Handle case when no pci device can provide cache line size hint

Hello,

On 12/15/2009 09:25 PM, Csaba Henk wrote:
> Prior to this patch, if pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, ...)
> returns 0 for all dev, pci_cache_line_size ends up set to zero
> (instead of pci_dfl_cache_line_size).
>
> This patch ensures the pci_cache_line_size = pci_dfl_cache_line_size
> setting in the above scenario.
>
> This happens in case of a kvm-88 guest (where, consequently, the rtl8139
> NIC failed to initialize).
>
> Signed-off-by: Csaba Henk <[email protected]>
> ---
> drivers/pci/quirks.c | 2 +-
> 1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
> index 7cfa7c3..f70f4e2 100644
> --- a/drivers/pci/quirks.c
> +++ b/drivers/pci/quirks.c
> @@ -2629,7 +2629,7 @@ static int __init pci_apply_final_quirks(void)
> if (!pci_cache_line_size) {
> printk(KERN_DEBUG "PCI: CLS %u bytes, default %u\n",
> cls << 2, pci_dfl_cache_line_size << 2);
> - pci_cache_line_size = cls;
> + pci_cache_line_size = cls ? cls : pci_dfl_cache_line_size;

Oh, so, all CLS values are zero? I've never thought of that.

Acked-by: Tejun Heo <[email protected]>

Thanks.

--
tejun

2009-12-16 19:22:38

by Jesse Barnes

Subject: Re: [PATCH 1/1] PCI: Handle case when no pci device can provide cache line size hint

On Tue, 15 Dec 2009 17:55:25 +0530
Csaba Henk <[email protected]> wrote:

> Prior to this patch, if pci_read_config_byte(dev,
> PCI_CACHE_LINE_SIZE, ...) returns 0 for all dev, pci_cache_line_size
> ends up set to zero (instead of pci_dfl_cache_line_size).
>
> This patch ensures the pci_cache_line_size = pci_dfl_cache_line_size
> setting in the above scenario.
>
> This happens in case of a kvm-88 guest (where, consequently, the
> rtl8139 NIC failed to initialize).
>
> Signed-off-by: Csaba Henk <[email protected]>
> ---
> drivers/pci/quirks.c | 2 +-
> 1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
> index 7cfa7c3..f70f4e2 100644
> --- a/drivers/pci/quirks.c
> +++ b/drivers/pci/quirks.c
> @@ -2629,7 +2629,7 @@ static int __init pci_apply_final_quirks(void)
> if (!pci_cache_line_size) {
> printk(KERN_DEBUG "PCI: CLS %u bytes, default %u\n",
> cls << 2, pci_dfl_cache_line_size << 2);
> - pci_cache_line_size = cls;
> + pci_cache_line_size = cls ? cls :
> pci_dfl_cache_line_size; }
>
> return 0;

Applied, thanks.

--
Jesse Barnes, Intel Open Source Technology Center