[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [[Bug 108996]] hw/dma.c: Fix conversion ioport_register
From: |
Andreas Färber |
Subject: |
Re: [Qemu-devel] [[Bug 108996]] hw/dma.c: Fix conversion ioport_register* to MemoryRegion |
Date: |
Fri, 14 Dec 2012 18:30:16 +0100 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:17.0) Gecko/17.0 Thunderbird/17.0 |
Am 14.12.2012 10:52, schrieb Julien Grall:
> The commit 582299336879504353e60c7937fbc70fea93f3da introduced a bug in
> dma emulation due to a bad conversion between ioport_register* and
> MemoryRegion.
>
> Cc: address@hidden
> Reported-by: Andreas Gustafsson <address@hidden>
> Signed-off-by: Julien Grall <address@hidden>
I had trouble following here, having handled the offending patch myself:
"Fix", "a bug" and "a bad conversion" is not really telling me what went
wrong and how the numbers are calculated correctly. Please suggest an
additional explanatory paragraph for the commit message (as a reply).
Formally the patch looks fine (modulo missing "of" or
s/conversion/converting/g in $subject).
>From what I gather, the cont region starts at base + 8 << dshift. Why is
the size in memory_region_init_io() 8 << d->dshift and not just 8 when
it previously looped over 0..7? Same question for the channel region.
Could be fixed as follow-up. More comments inline:
> ---
> hw/dma.c | 22 +++++++++++-----------
> 1 file changed, 11 insertions(+), 11 deletions(-)
>
> diff --git a/hw/dma.c b/hw/dma.c
> index c2d7b21..1b1d406 100644
> --- a/hw/dma.c
> +++ b/hw/dma.c
> @@ -200,7 +200,7 @@ static void write_cont(void *opaque, hwaddr nport,
> uint64_t data,
>
> iport = (nport >> d->dshift) & 0x0f;
> switch (iport) {
> - case 0x01: /* command */
> + case 0x00: /* command */
Since the shift is "reverted" above, we effectively have an 0x8 ->
0x8+0x1 -> 0x8+0x0 change, which looks correct.
This delta seems consistent for the other case changes ...
> if ((data != 0) && (data & CMD_NOT_SUPPORTED)) {
> dolog("command %"PRIx64" not supported\n", data);
> return;
> @@ -208,7 +208,7 @@ static void write_cont(void *opaque, hwaddr nport,
> uint64_t data,
> d->command = data;
> break;
>
> - case 0x02:
> + case 0x01:
> ichan = data & 3;
> if (data & 4) {
> d->status |= 1 << (ichan + 4);
> @@ -220,7 +220,7 @@ static void write_cont(void *opaque, hwaddr nport,
> uint64_t data,
> DMA_run();
> break;
>
> - case 0x03: /* single mask */
> + case 0x02: /* single mask */
> if (data & 4)
> d->mask |= 1 << (data & 3);
> else
> @@ -228,7 +228,7 @@ static void write_cont(void *opaque, hwaddr nport,
> uint64_t data,
> DMA_run();
> break;
>
> - case 0x04: /* mode */
> + case 0x03: /* mode */
> {
> ichan = data & 3;
> #ifdef DEBUG_DMA
> @@ -247,23 +247,23 @@ static void write_cont(void *opaque, hwaddr nport,
> uint64_t data,
> break;
> }
>
> - case 0x05: /* clear flip flop */
> + case 0x04: /* clear flip flop */
> d->flip_flop = 0;
> break;
>
> - case 0x06: /* reset */
> + case 0x05: /* reset */
> d->flip_flop = 0;
> d->mask = ~0;
> d->status = 0;
> d->command = 0;
> break;
>
> - case 0x07: /* clear mask for all channels */
> + case 0x06: /* clear mask for all channels */
> d->mask = 0;
> DMA_run();
> break;
>
> - case 0x08: /* write mask for all channels */
> + case 0x07: /* write mask for all channels */
> d->mask = data;
> DMA_run();
> break;
> @@ -288,11 +288,11 @@ static uint64_t read_cont(void *opaque, hwaddr nport,
> unsigned size)
>
> iport = (nport >> d->dshift) & 0x0f;
> switch (iport) {
> - case 0x08: /* status */
> + case 0x00: /* status */
> val = d->status;
> d->status &= 0xf0;
> break;
> - case 0x0f: /* mask */
> + case 0x01: /* mask */
> val = d->mask;
> break;
> default:
> @@ -467,7 +467,7 @@ void DMA_schedule(int nchan)
> static void dma_reset(void *opaque)
> {
> struct dma_cont *d = opaque;
> - write_cont(d, (0x06 << d->dshift), 0, 1);
> + write_cont(d, (0x05 << d->dshift), 0, 1);
... and for the (weird :)) reuse of the write_cont() callback function
from within the reset function.
> }
>
> static int dma_phony_handler (void *opaque, int nchan, int dma_pos, int
> dma_len)
Reviewed-by: Andreas Färber <address@hidden>
make check runs an fdc-test that passed okay. Can one of you add a test
case to avoid another regression here?
Regards,
Andreas
--
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg