dmaengine: ep93xx: Treat STALL and NFB IRQs same way

Debugging ALSA hangups it was found that EP9302 (latest E2 rev.) DMA controller
sometimes asserts STALL interrupt instead of NFB interrupt. Simply ignoring the
difference and simply acting upon the amount of data we still have to transfer
seems to work fine. This somehow sounds similar to M2M issue which is already
dealt with in the driver, when the controller asserts DONE interrupt too early.

The issue is not documented in Cirrus Logic erratas for EP93XX, but original
Cirrus DMA driver from 2003 (not based on DMA API) did the similar handling
of STALL interrupt. In-tree driver (6d831c65) did it also, before conversion to
DMA engine API.

Signed-off-by: Alexander Sverdlin <alexander.sverdlin@gmail.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
This commit is contained in:
Alexander Sverdlin 2016-01-07 23:05:48 +01:00 committed by Vinod Koul
parent c9784a4673
commit 94901e1b24
1 changed files with 15 additions and 13 deletions

View File

@ -421,23 +421,25 @@ static int m2p_hw_interrupt(struct ep93xx_dma_chan *edmac)
desc->size);
}
switch (irq_status & (M2P_INTERRUPT_STALL | M2P_INTERRUPT_NFB)) {
case M2P_INTERRUPT_STALL:
/* Disable interrupts */
control = readl(edmac->regs + M2P_CONTROL);
control &= ~(M2P_CONTROL_STALLINT | M2P_CONTROL_NFBINT);
m2p_set_control(edmac, control);
return INTERRUPT_DONE;
case M2P_INTERRUPT_NFB:
if (ep93xx_dma_advance_active(edmac))
m2p_fill_desc(edmac);
/*
* Even latest E2 silicon revision sometimes assert STALL interrupt
* instead of NFB. Therefore we treat them equally, basing on the
* amount of data we still have to transfer.
*/
if (!(irq_status & (M2P_INTERRUPT_STALL | M2P_INTERRUPT_NFB)))
return INTERRUPT_UNKNOWN;
if (ep93xx_dma_advance_active(edmac)) {
m2p_fill_desc(edmac);
return INTERRUPT_NEXT_BUFFER;
}
return INTERRUPT_UNKNOWN;
/* Disable interrupts */
control = readl(edmac->regs + M2P_CONTROL);
control &= ~(M2P_CONTROL_STALLINT | M2P_CONTROL_NFBINT);
m2p_set_control(edmac, control);
return INTERRUPT_DONE;
}
/*