[RFC] [PATCH] Fix up needless kmap:s

[RFC] [PATCH] Fix up needless kmap:s

[Pierre Ossman]

[-- Attachment #1: Type: text/plain, Size: 280 bytes --]

Hi Alex,

I'd like you to test and comment on the following patch.

Rgds
-- 
     -- Pierre Ossman

  Linux kernel, MMC maintainer        http://www.kernel.org
  PulseAudio, core developer          http://pulseaudio.org
  rdesktop, core developer          http://www.rdesktop.org

[-- Attachment #2: 0001-mmc-tifm-replace-kmap-with-page_address.txt --]
[-- Type: text/plain, Size: 2246 bytes --]

>From c8467c6b0ea0443f948c0ece281e1ca8e92f6ea7 Mon Sep 17 00:00:00 2001
From: Pierre Ossman <drzeus@drzeus.cx>
Date: Sat, 3 Feb 2007 13:36:41 +0100
Subject: [PATCH] mmc: tifm: replace kmap with page_address

Since we actively avoid highmem, calling kmap_atomic() instead
of page_address() is effectively only obfuscation.

Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
---
 drivers/mmc/tifm_sd.c |   15 ++++-----------
 1 files changed, 4 insertions(+), 11 deletions(-)

diff --git a/drivers/mmc/tifm_sd.c b/drivers/mmc/tifm_sd.c
index 7e607b7..e65f8a0 100644
--- a/drivers/mmc/tifm_sd.c
+++ b/drivers/mmc/tifm_sd.c
@@ -107,14 +107,9 @@ struct tifm_sd {
 
 };
 
-static char* tifm_sd_kmap_atomic(struct mmc_data *data)
+static char* tifm_sd_data_buffer(struct mmc_data *data)
 {
-	return kmap_atomic(data->sg->page, KM_BIO_SRC_IRQ) + data->sg->offset;
-}
-
-static void tifm_sd_kunmap_atomic(char *buffer, struct mmc_data *data)
-{
-	kunmap_atomic(buffer - data->sg->offset, KM_BIO_SRC_IRQ);
+	return page_address(data->sg->page) + data->sg->offset;
 }
 
 static int tifm_sd_transfer_data(struct tifm_dev *sock, struct tifm_sd *host,
@@ -127,18 +122,17 @@ static int tifm_sd_transfer_data(struct tifm_dev *sock, struct tifm_sd *host,
 	if (host_status & TIFM_MMCSD_BRS) {
 		/* in non-dma rx mode BRS fires when fifo is still not empty */
 		if (no_dma && (cmd->data->flags & MMC_DATA_READ)) {
-			buffer = tifm_sd_kmap_atomic(host->req->data);
+			buffer = tifm_sd_data_buffer(host->req->data);
 			while (host->buffer_size > host->buffer_pos) {
 				t_val = readl(sock->addr + SOCK_MMCSD_DATA);
 				buffer[host->buffer_pos++] = t_val & 0xff;
 				buffer[host->buffer_pos++] =
 							(t_val >> 8) & 0xff;
 			}
-			tifm_sd_kunmap_atomic(buffer, host->req->data);
 		}
 		return 1;
 	} else if (no_dma) {
-		buffer = tifm_sd_kmap_atomic(host->req->data);
+		buffer = tifm_sd_data_buffer(host->req->data);
 		if ((cmd->data->flags & MMC_DATA_READ) &&
 				(host_status & TIFM_MMCSD_AF)) {
 			for (cnt = 0; cnt < TIFM_MMCSD_FIFO_SIZE; cnt++) {
@@ -163,7 +157,6 @@ static int tifm_sd_transfer_data(struct tifm_dev *sock, struct tifm_sd *host,
 				}
 			}
 		}
-		tifm_sd_kunmap_atomic(buffer, host->req->data);
 	}
 	return 0;
 }
-- 
1.4.4.2

Re: [RFC] [PATCH] Fix up needless kmap:s

[Alex Dubov]

The patch looks ok.

However, due to certain peculiarities with memorystick and xd I have to emulate a scatter-gather
in software. Considering that this particular aspect of implementation is the same for all card
types on this TI chip, it can be shared by tifm_sd driver as well. In this case highmem won't be
needed to be avoided and kmap_atomic will come back (may be).

And, by the way, to what extent pagefault_enable/pagefault_disable calls are needed? They are
present in k(u)map_atomic even for non-highmem pages/architectures.



 
____________________________________________________________________________________
Expecting? Get great news right away with email Auto-Check. 
Try the Yahoo! Mail Beta.
http://advision.webevents.yahoo.com/mailbeta/newmail_tools.html 

Re: [RFC] [PATCH] Fix up needless kmap:s

[Pierre Ossman]

Alex Dubov wrote:
> The patch looks ok.
>
> However, due to certain peculiarities with memorystick and xd I have to emulate a scatter-gather
> in software. Considering that this particular aspect of implementation is the same for all card
> types on this TI chip, it can be shared by tifm_sd driver as well. In this case highmem won't be
> needed to be avoided and kmap_atomic will come back (may be).
>
>   

Fair enough. Just make sure you handle the big problem case (kmap only
maps a single page, not the entire sg entry).

> And, by the way, to what extent pagefault_enable/pagefault_disable calls are needed? They are
> present in k(u)map_atomic even for non-highmem pages/architectures.
>
>   

They tell the page fault handler that it may not sleep on a page fault.
Currently it only has the behaviour of disabling preemption though. I'd
guess it is present for non-highmem so that things behave somewhat
similar for both the highmem and non-highmem cases.

Rgds

-- 
     -- Pierre Ossman

  Linux kernel, MMC maintainer        http://www.kernel.org
  PulseAudio, core developer          http://pulseaudio.org
  rdesktop, core developer          http://www.rdesktop.org