scsi: gvp11: add module parameter for DMA transfer bit mask
PATCH 1/1 scsi: gvp11: add module parameter for DMA transfer bit mask
Commit bfaa4a0ce1bb ("scsi: gvp11: Remove unused gvp11_setup() function") removed the unused gvp11_setup() which had allowed to override the default DMA transfer mask defined for GVP II SCSI boards on Amiga. There now is no way to set a non-default DMA mask on these boards.
Introduce a module parameter to allow users to override the default DMA mask for the gvp11 driver.
--- drivers/scsi/gvp11.c | 3 +++ 1 file changed, 3 insertions(+)
drivers/scsi/gvp11.c +3 -0
@@ -50,6 +50,9 @@ static irqreturn_t gvp11_intr(int irq, void *data)static int gvp11_xfer_mask =0 ; +module_param (gvp11_xfer_mask, int,0444 ); +MODULE_PARM_DESC (gvp11_xfer_mask,"DMA mask (0xff000000 == 24 bit DMA)" ); +static int dma_setup (struct scsi_cmnd *cmd,int dir_in) {struct scsi_pointer *scsi_pointer =WD33C93_scsi_pointer (cmd);
Signature · 1 line
2.17.1
Quoted 2 lines · Michael Schmitz
SCSI boards on Amiga. There now is no way to set a non-default DMA mask on these boards.
It might help to mention here in which cases the default mask is actually wrong.
drivers/scsi/gvp11.c static irqreturn_t gvp11_intr(int irq, void *data) · line 55
Quoted 3 lines
drivers/scsi/gvp11.c +3
-0+module_param (gvp11_xfer_mask, int,0444 ); +MODULE_PARM_DESC (gvp11_xfer_mask,"DMA mask (0xff000000 == 24 bit DMA)" ); +
I think the comment is the wrong way round, it should be 0x00ffffff in this case, which also matches the default mask for ZORRO_PROD_GVP_SERIES_II, in the match table:
static struct zorro_device_id gvp11_zorro_tbl[] = {
{ ZORRO_PROD_GVP_COMBO_030_R3_SCSI, ~0x00ffffff },
{ ZORRO_PROD_GVP_SERIES_II, ~0x00ffffff },
{ ZORRO_PROD_GVP_GFORCE_030_SCSI, ~0x01ffffff },
{ ZORRO_PROD_GVP_A530_SCSI, ~0x01ffffff },
{ ZORRO_PROD_GVP_COMBO_030_R4_SCSI, ~0x01ffffff },
{ ZORRO_PROD_GVP_A1291, ~0x07ffffff },
{ ZORRO_PROD_GVP_GFORCE_040_SCSI_1, ~0x07ffffff },
{ 0 }
};Arnd
Hi Arnd,
thanks for your comments!
Quoted 5 lines · Arnd Bergmann
Quoted 2 lines · Michael Schmitz
SCSI boards on Amiga. There now is no way to set a non-default DMA mask on these boards.
It might help to mention here in which cases the default mask is actually wrong.
All I have is:
Probably it's needed on A2000 with an accelerator card and GVP II SCSI, to prevent DMA to RAM banks that do not support fast DMA cycles.
from Geert's reply. I can add that. It just did sound a shade speculative...
Quoted 8 lines
drivers/scsi/gvp11.c static irqreturn_t gvp11_intr(int irq, void *data) · line 55
Quoted 4 lines
drivers/scsi/gvp11.c +3
-0+module_param (gvp11_xfer_mask, int,0444 ); +MODULE_PARM_DESC (gvp11_xfer_mask,"DMA mask (0xff000000 == 24 bit DMA)" ); +I think the comment is the wrong way round, it should be
0x00ffffffin this case, which also matches the default mask forZORRO_PROD_GVP_SERIES_II, in the match table:
Quoted 10 lines
static struct zorro_device_id gvp11_zorro_tbl[] = {
{ ZORRO_PROD_GVP_COMBO_030_R3_SCSI, ~0x00ffffff },
{ ZORRO_PROD_GVP_SERIES_II, ~0x00ffffff },
{ ZORRO_PROD_GVP_GFORCE_030_SCSI, ~0x01ffffff },
{ ZORRO_PROD_GVP_A530_SCSI, ~0x01ffffff },
{ ZORRO_PROD_GVP_COMBO_030_R4_SCSI, ~0x01ffffff },
{ ZORRO_PROD_GVP_A1291, ~0x07ffffff },
{ ZORRO_PROD_GVP_GFORCE_040_SCSI_1, ~0x07ffffff },
{ 0 }
};gvp11_xfer_mask works inverse to what you'd expect (and inverse to what a DMA mask usually is defined as). DMA can _not_ be used if (address & gvp11_xfer_mask) isn't zero. See code in dma_setup() for details.
All those definitions have a '~' prefix, for that very reason.
I agree it isn't intuitive, and caused a little head scratching when preparing this patch. But I believe it is correct.
Now you could argue to shift the bit mask inversion to gvp11_probe() or even dma_setup() instead to rule out such confusion in future, but that would be an actual code change and would benefit from testing on at least one of these boards IMO. Not sure how easy that will be.
Cheers,
Michael
Quoted 1 line
Arnd
Arnd Bergmann
Acked-byQuoted 25 lines · Michael Schmitz
Quoted 8 lines
drivers/scsi/gvp11.c static irqreturn_t gvp11_intr(int irq, void *data) · line 55
Quoted 4 lines
drivers/scsi/gvp11.c +3
-0+module_param (gvp11_xfer_mask, int,0444 ); +MODULE_PARM_DESC (gvp11_xfer_mask,"DMA mask (0xff000000 == 24 bit DMA)" ); +I think the comment is the wrong way round, it should be
0x00ffffffin this case, which also matches the default mask forZORRO_PROD_GVP_SERIES_II, in the match table:Quoted 10 lines
static struct zorro_device_id gvp11_zorro_tbl[] = { { ZORRO_PROD_GVP_COMBO_030_R3_SCSI, ~0x00ffffff }, { ZORRO_PROD_GVP_SERIES_II, ~0x00ffffff }, { ZORRO_PROD_GVP_GFORCE_030_SCSI, ~0x01ffffff }, { ZORRO_PROD_GVP_A530_SCSI, ~0x01ffffff }, { ZORRO_PROD_GVP_COMBO_030_R4_SCSI, ~0x01ffffff }, { ZORRO_PROD_GVP_A1291, ~0x07ffffff }, { ZORRO_PROD_GVP_GFORCE_040_SCSI_1, ~0x07ffffff }, { 0 } };
gvp11_xfer_maskworks inverse to what you'd expect (and inverse to what a DMA mask usually is defined as). DMA can _not_ be used if (address &gvp11_xfer_mask) isn't zero. See code indma_setup()for details.All those definitions have a '~' prefix, for that very reason.
Quoted 7 lines
I agree it isn't intuitive, and caused a little head scratching when preparing this patch. But I believe it is correct.
Now you could argue to shift the bit mask inversion to
gvp11_probe()or evendma_setup()instead to rule out such confusion in future, but that would be an actual code change and would benefit from testing on at least one of these boards IMO. Not sure how easy that will be.
Ok, I see now. Let's leave the patch as it is then.
Thanks Arnd!
Cheers,
Michael
Quoted 34 lines · Arnd Bergmann
Quoted 23 lines · Michael Schmitz
Quoted 7 lines
drivers/scsi/gvp11.c static irqreturn_t gvp11_intr(int irq, void *data) · line 55
Quoted 3 lines
drivers/scsi/gvp11.c +3
-0+module_param (gvp11_xfer_mask, int,0444 ); +MODULE_PARM_DESC (gvp11_xfer_mask,"DMA mask (0xff000000 == 24 bit DMA)" ); +I think the comment is the wrong way round, it should be
0x00ffffffin this case, which also matches the default mask forZORRO_PROD_GVP_SERIES_II, in the match table:Quoted 10 lines
static struct zorro_device_id gvp11_zorro_tbl[] = { { ZORRO_PROD_GVP_COMBO_030_R3_SCSI, ~0x00ffffff }, { ZORRO_PROD_GVP_SERIES_II, ~0x00ffffff }, { ZORRO_PROD_GVP_GFORCE_030_SCSI, ~0x01ffffff }, { ZORRO_PROD_GVP_A530_SCSI, ~0x01ffffff }, { ZORRO_PROD_GVP_COMBO_030_R4_SCSI, ~0x01ffffff }, { ZORRO_PROD_GVP_A1291, ~0x07ffffff }, { ZORRO_PROD_GVP_GFORCE_040_SCSI_1, ~0x07ffffff }, { 0 } };
gvp11_xfer_maskworks inverse to what you'd expect (and inverse to what a DMA mask usually is defined as). DMA can _not_ be used if (address &gvp11_xfer_mask) isn't zero. See code indma_setup()for details.All those definitions have a '~' prefix, for that very reason.
Quoted 7 lines
I agree it isn't intuitive, and caused a little head scratching when preparing this patch. But I believe it is correct.
Now you could argue to shift the bit mask inversion to
gvp11_probe()or evendma_setup()instead to rule out such confusion in future, but that would be an actual code change and would benefit from testing on at least one of these boards IMO. Not sure how easy that will be.Ok, I see now. Let's leave the patch as it is then.
Acked-by Arnd Bergmann <[email protected]>
Hi Michael,
Quoted 7 lines · Michael Schmitz
Quoted 5 lines · Arnd Bergmann
Quoted 2 lines · Michael Schmitz
SCSI boards on Amiga. There now is no way to set a non-default DMA mask on these boards.
It might help to mention here in which cases the default mask is actually wrong.
Quoted 7 lines
All I have is:
Probably it's needed on A2000 with an accelerator card and GVP II SCSI, to prevent DMA to RAM banks that do not support fast DMA cycles.
from Geert's reply. I can add that. It just did sound a shade speculative...
Apparently gvp11_setup() became unused in 2.3.13pre2 (in 1999), when all *_setup() functions were removed from init/main.c, and some of them were reimplemented using __setup() in the driver sources where they belonged.
Quoted 7 lines
Quoted 7 lines
drivers/scsi/gvp11.c static irqreturn_t gvp11_intr(int irq, void *data) · line 55
Quoted 3 lines
drivers/scsi/gvp11.c +3
-0+module_param (gvp11_xfer_mask, int,0444 ); +MODULE_PARM_DESC (gvp11_xfer_mask,"DMA mask (0xff000000 == 24 bit DMA)" ); +I think the comment is the wrong way round, it should be
0x00ffffffin this case, which also matches the default mask forZORRO_PROD_GVP_SERIES_II, in the match table:
Quoted 10 lines
Quoted 10 lines
static struct zorro_device_id gvp11_zorro_tbl[] = { { ZORRO_PROD_GVP_COMBO_030_R3_SCSI, ~0x00ffffff }, { ZORRO_PROD_GVP_SERIES_II, ~0x00ffffff }, { ZORRO_PROD_GVP_GFORCE_030_SCSI, ~0x01ffffff }, { ZORRO_PROD_GVP_A530_SCSI, ~0x01ffffff }, { ZORRO_PROD_GVP_COMBO_030_R4_SCSI, ~0x01ffffff }, { ZORRO_PROD_GVP_A1291, ~0x07ffffff }, { ZORRO_PROD_GVP_GFORCE_040_SCSI_1, ~0x07ffffff }, { 0 } };
The default masks above were added (in some other form) in 2.1.91pre1 (in 1998). Before, people had to use gvp11_setup() to do that.
So I think it is safe to assume there is no longer a need to configure this manually.
Gr{oetje,eeting}s,
Geert
Signature · 5 lines
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected] In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds
Hi Geert
Am 30.08.2023 um 19:32 schrieb Geert Uytterhoeven:
Quoted 15 lines
Hi Michael,
Quoted 12 lines · Michael Schmitz
Quoted 5 lines · Arnd Bergmann
Quoted 2 lines · Michael Schmitz
SCSI boards on Amiga. There now is no way to set a non-default DMA mask on these boards.
It might help to mention here in which cases the default mask is actually wrong.
All I have is:
Probably it's needed on A2000 with an accelerator card and GVP II SCSI, to prevent DMA to RAM banks that do not support fast DMA cycles.
Quoted 6 lines
Quoted 2 lines
from Geert's reply. I can add that. It just did sound a shade speculative...
Apparently
gvp11_setup()became unused in 2.3.13pre2 (in 1999), when all*_setup()functions were removed frominit/main.c, and some of them were reimplemented using__setup()in the driver sources where they belonged.
But that wasn't done for gvp11_setup() ...
Quoted 18 lines
Quoted 7 lines
Quoted 7 lines
drivers/scsi/gvp11.c static irqreturn_t gvp11_intr(int irq, void *data) · line 55
Quoted 3 lines
drivers/scsi/gvp11.c +3
-0+module_param (gvp11_xfer_mask, int,0444 ); +MODULE_PARM_DESC (gvp11_xfer_mask,"DMA mask (0xff000000 == 24 bit DMA)" ); +I think the comment is the wrong way round, it should be
0x00ffffffin this case, which also matches the default mask forZORRO_PROD_GVP_SERIES_II, in the match table:Quoted 10 lines
Quoted 10 lines
static struct zorro_device_id gvp11_zorro_tbl[] = { { ZORRO_PROD_GVP_COMBO_030_R3_SCSI, ~0x00ffffff }, { ZORRO_PROD_GVP_SERIES_II, ~0x00ffffff }, { ZORRO_PROD_GVP_GFORCE_030_SCSI, ~0x01ffffff }, { ZORRO_PROD_GVP_A530_SCSI, ~0x01ffffff }, { ZORRO_PROD_GVP_COMBO_030_R4_SCSI, ~0x01ffffff }, { ZORRO_PROD_GVP_A1291, ~0x07ffffff }, { ZORRO_PROD_GVP_GFORCE_040_SCSI_1, ~0x07ffffff }, { 0 } };
Quoted 2 lines
The default masks above were added (in some other form) in 2.1.91pre1 (in 1998). Before, people had to use
gvp11_setup()to do that.
... because gvp11_setup() was already obsolete.
Thanks for dredging up that bit of history!
Quoted 2 lines
So I think it is safe to assume there is no longer a need to configure this manually.
I'll take your word for that. No need to apply this patch, then!
(Apologies for the noise, Arnd ...)
Cheers,
Michael
Quoted 4 lines
Gr{oetje,eeting}s,
Geert