Re: [PATCH] net: via-velocity.c fix sleep-with-spinlock bug during MTU change

["Séguier Régis" <rseguier@e-teleport.net>]

Francois Romieu a écrit :
> Séguier Régis <rseguier@e-teleport.net> :
> [...]
>   
>> But if I generate a paquet with a size more than around 3825, the driver 
>> don't send it and no paquet is send after, work again after an interface 
>> down/up and until no paquet > 3825.
>>     
>
> Does it qualify as a regression ?
>
>   
No,

I test with the version of the driver before patches and I have the same 
error.

I think error is in TX. (TX and RX equipments for the test have the same 
chip, I don't have other kind of chip available at the moment, that why 
I not sure)

If theses informations help :

When I transmit a paquet of ~ 4000 (for a mtu 4500),

When I reboot and PCIE-AER is compiled, I got this message :

Jun 23 18:28:31 apollo kernel: +------ PCI-Express Device Error ------+
Jun 23 18:28:31 apollo kernel: Error Severity^I^I: Uncorrected (Non-Fatal)
Jun 23 18:28:31 apollo kernel: PCIE Bus Error type^I: Transaction Layer
Jun 23 18:28:31 apollo kernel: Unsupported Request   ^I: Multiple
Jun 23 18:28:31 apollo kernel: Requester ID^I^I: 0300
Jun 23 18:28:31 apollo kernel: VendorID=1106h, DeviceID=3119h, Bus=03h, 
Device=00h, Function=00h
Jun 23 18:28:31 apollo kernel: TLB Header:
Jun 23 18:28:31 apollo kernel: 40000001 0018000c febffc7c 0000fdff
Jun 23 18:28:31 apollo kernel: Broadcast error_detected message
Jun 23 18:28:31 apollo kernel: Broadcast mmio_enabled message
Jun 23 18:28:31 apollo kernel: Broadcast resume message

When I reboot and "PCIE-AER and MSI" is compiled, I got this message :

Jun 23 18:35:59 apollo kernel: irq 163, desc: c0495dc8, depth: 1, count: 
0, unhandled: 0
Jun 23 18:35:59 apollo kernel: ->handle_irq():  c012d997, 
handle_bad_irq+0x0/0x199
Jun 23 18:35:59 apollo kernel: ->chip(): c0496b20, 0xc0496b20
Jun 23 18:35:59 apollo kernel: ->action(): 00000000
Jun 23 18:35:59 apollo kernel:   IRQ_DISABLED set
Jun 23 18:35:59 apollo kernel: unexpected IRQ trap at vector a3
-- 
Régis